PROCEEDINGS Volume 1 - ineag

PROCEEDINGS of the 8 th International Conference on Applied Financial
Economics - Volume 1
Samos Island, GREECE, 30 June - 02 July 2011
EDITED BY
Research and Training Institute of the East Aegean (INEAG), Greece
EDITOR
Chrysovaladis Prachalias
PUBLISHED BY
National and Kapodistrian University of Athens, Greece
ISBN: 978-960-466-085-8
ISBN [SET]: 978-960-466-087-2
ISSN: 1790-3912

ACKNOWLEDGMENTS
The editors would like to thank all contributing authors to this book for their effort to prepare their
submissions and presentations. We would like to thank the personnel of National and Kapodistrian
University of Athens for their kind offer to publish the proceedings of 8 th AFE Samos 2011.
The editors would like to thank the scientific committee and the reviewers who read carefully and
reviewed all contributions. The conference committee would like to thank for their support and kind
cooperation the Hellenic Telecommunications Organization.
2

MESSAGE FROM THE COORDINATOR OF THE SCIENTIFIC COMMITTEE
Dear Participants,
On behalf of the Scientific Committee I would like to warmly welcome you all to the 8th
International Applied Financial Economics (AFE) Conference. Despite their short history the
AFE International Conferences, co-organized this year by INEAG and the National and
Kapodistrian University of Athens, have already gained a worldwide reputation and have been
established as a forum in which academics, researchers and professional experts in the field of
finance from all over the world come together, interact, exchange ideas, and present their
research. A direct reflection of this success story is the number of submitted papers, which has
been increasing substantially year by year, as well as their high quality. There is little doubt that
scientifically, as well as in terms of participation, this year’s conference is indeed beyond
expectations, as the Scientific Committee has received more than one hundred and fifty very
interesting research papers, including one co-authored by a Nobel Prize laureate, while
presenters come from over twenty countries.
Over the last year there are signs that the world economy is recovering from the 2009 shock, but
at a much faster pace in the Far East than in the West. However, the international financial
system has not been stabilized yet and given that the world debt has increased up to about 80%
of the world GDP, further trouble to the financial system may be present in the medium to long
term. On the other hand Greece is in the midst of an acute crisis with an increased debt combined
with negative growth. It is noteworthy that several of the submitted manuscripts cope with all
that, clearly indicating the fast adaptation of this conference not only to current international
trends in financial research, but also to major real economic events and the developments in the
international financial markets. In that sense this conference offers a unique opportunity for firsthand
knowledge on the most updated financial research and I am certain that we will all take
advantage of this opportunity.
Wishing you all a very pleasant and fruitful stay on the beautiful island of Samos.
Alexandros E. Milionis
Assistant Professor
University of the Aegean
Department of Statistics and Actuarial - Financial Mathematics
3

Contents Page
PREFACE..............................................................................................................................................3
CONTENTS ...........................................................................................................................................4
SCIENTIFIC COMMITTEE...................................................................................................................10
STEERING COMMITTEE.....................................................................................................................10
CONFERENCE COORDINATOR AND SECRETARIAT............................................................................11
KEYNOTE SPEAKERS .........................................................................................................................11
KEYNOTE LECTURES ................................................................................................................ 12
PORTFOLIO SELECTION: A REVIEW..................................................................................................... 14
Jerome Detemple
ENHANCING FILTERED HISTORICAL SIMULATIONS ............................................................................. 29
Giovanni Barone-Adesi
INTERNATIONAL PROPAGATION OF THE CREDIT CRISIS....................................................................... 32
Ian Cooper
SHORT WAVELENGTH FINANCE: WHAT, WHO AND WHY.................................................................... 69
Nick Kondakis
ASSET ALLOCATION-PORTFOLIO MANAGEMENT.......................................................... 70
ON THE PERFORMANCE OF A HYBRID GENETIC ALGORITHM: APPLICATION ON THE PORTFOLIO
MANAGEMENT PROBLEM ................................................................................................................... 72
Vassilios Vassiliadis, Vassiliki Bafa, George Dounias
ASSET PRICING............................................................................................................................ 80
ASSET PRICING IN THE CYPRUS STOCK EXCHANGE: ARE STOCKS FAIRLY PRICED? ............................ 82
Haritini Tsangari, Maria Elfani
THE VALUATION OF EQUITIES WHEN SHAREHOLDERS ENJOY LIMITED LIABILITY.............................. 90
Jo Wells
PRICE PRESSURE RISK FACTOR IN CONVERTIBLE BONDS.................................................................. 100
Nikolay Ryabkov, Galyna Petrenko
THE PRICING OF EQUITY-LINKED CONTINGENT CLAIMS UNDER A LOGNORMAL SHORT RATE DYNAMICS
......................................................................................................................................................... 110
Rosa Cocozza, Antonio De Simone
4

LIQUIDITY AND EXPECTED RETURNS: NEW EVIDENCE FROM DAILY DATA 1926-2008...................... 120
M. Reza Baradaran, Maurice Peat
BANKING...................................................................................................................................... 130
ELECTRONIC BANKING IN JORDAN: A FRAMEWORK OF ADOPTION ................................................... 132
Muneer Abbad, Juma’h Abbad, Faten Jaber
PROCYCLICALITY OF BANKS’ CAPITAL BUFFER IN ASIAN COUNTRIES .............................................. 142
Elis Deriantino
DETERMINING EFFECTIVE INDICATORS FOR CUSTOMER PERFORMANCE IN REPAYING BANK LOAN (CASE
STUDY: IRANIAN BANKS) ................................................................................................................. 147
Fariba SeyedJafar Rangraz, Naser Shams
TWO-STAGE DEA APPROACH TO EVALUATE THE EFFICIENCY OF BANK BRANCHES......................... 155
Akram Bodaghi & Heidar Mostakhdemin Hosseini
MODELLING FUNCTIONAL INDICATORS FOR CUSTOMER PERFORMANCE IN REPAYING BANK LOAN (CASE
STUDY: IRANIAN BANKS) ................................................................................................................. 160
Fariba Seyed Jafar Rangraz, Jafar Pashami
MEASURING AGREEMENT WITH THE WEIGHTED KAPPA COEFFICIENT OF CREDIT RATING DECISIONS: A
CASE OF BANKING SECTOR .............................................................................................................. 165
Funda H. Sezgin, Ozlen Erkal
THE GLOBAL FINANCIAL CRISIS AND THE BANKING SECTOR IN SERBIA............................................ 173
Emilija Vuksanović, Violeta Todorović
COLLECTION MANAGEMENT AS CRUCIAL PART OF CREDIT RISK MANAGEMENT DURING THE CRISIS. 182
Lidija Barjaktarovic, Snezana Popovcic- Avric, Marina Djenic
HOW THE CROSS BORDER MERGERS AND ACQUISITIONS OF THE GREEK BANKS IN THE BALKAN AREA
IMPROVE THE COURSE OF PROFITABILITY, EFFICIENCY AND LIQUIDITY INDEXES OF THEM?............. 192
Kyriazopoulos George, Petropoulos Dimitrios
PERFORMANCE OF ISLAMIC BANKS ACROSS THE WORLD: AN EMPIRICAL ANALYSIS OVER THE PERIOD
2001-2008........................................................................................................................................ 202
Sandrine Kablan, Ouidad Yousfi
THE DETERMINANTS OF THE EAD .................................................................................................... 208
Yenni Redjah, Jean Roy, Inmaculada Buendía Martínez
A MACRO-BASED MODEL OF PD AND LGD IN STRESS TESTING LOAN LOSSES .................................. 215
Esa Jokivuolle, Matti Virén
XTREME CREDIT RISK MODELS: IMPLICATIONS FOR BANK CAPITAL BUFFERS.................................. 222
David E. Allen, Akhmad R. Kramadibrata, Robert J. Powell, Abhay K. Singh
5

COMMODITIES MARKETS...................................................................................................... 230
VOLATILITY DYNAMICS IN DUBAI GOLD FUTURES MARKET ............................................................ 232
Ramzi Nekhili, Michael Thorpe
STOCK RETURNS AND OIL PRICE BASED TRADING............................................................................ 241
Michael Soucek
OPTIMAL LEVERAGE AND STOP LOSS POLICIES FOR FUTURES INVESTMENTS.................................... 248
Rainer A. Schüssler
THE IMPACT OF INTERNATIONAL MARKET EFFECTS AND PURE POLITICAL RISK ON THE UK, EMU AND
USA OIL AND GAS STOCK MARKET SECTORS .................................................................................. 257
John Simpson
THE PROGRESS OF NATURAL GAS MARKET LIBERALISATION IN THE UK AND USA: DE-COUPLING OF OIL
AND GAS PRICES .............................................................................................................................. 264
John Simpson
QUANTITATIVE EASING ENGINEERED BY THE FED, AND PRICES OF INTERNATIONALLY TRADED AND
DOLLAR DENOMINATED COMMODITIES AND PRECIOUS METALS...................................................... 274
Gueorgui I. Kolev
AGRI-BUBBLES ABSORB CHEAP MONEY: THEORIES AND PRACTICE IN CONTEMPORARY WORLD ..... 281
Evdokimov Alexandre Ivanovich, Soboleva Olga Valerjevna
THE INFORMATION CONTENT OF IMPLIED VOLATILITY IN THE CRUDE OIL FUTURES MARKET.......... 290
Asyl Bakanova
LITHIUM POWER: IS THE FUTURE GREEN? ........................................................................................ 300
Satyarth Pandey, Veena Choudhary
CORPORATE FINANCE ............................................................................................................ 308
PREDICTING BUSINESS FAILURE USING DATA-MINING METHODS .................................................... 310
Sami BEN JABEUR, Youssef FAHMI
SIMULTANEOUS DETERMINATION OF CORPORATE DECISIONS: AN EMPIRICAL INVESTIGATION USING UK
PANEL DATA .................................................................................................................................... 318
Qingwei Meng
LEVERAGE ADJUSTMENT AND COST OF CAPITAL.............................................................................. 321
Sven Husmann, Michael Soucek, Antonina Waszczuk
DOES LEVERAGE AFFECT LABOUR PRODUCTIVITY? A COMPARATIVE STUDY OF LOCAL AND
MULTINATIONAL COMPANIES OF THE BALTIC COUNTRIES................................................................ 329
Mari Avarmaa, Aaro Hazak, Kadri Männasoo
MULTISTAGE INVESTMENT OPTIONS, TIME-TO-BUILD AND FINANCING CONSTRAINTS..................... 344
6

Elettra Agliardi, Nicos Koussis
RE-EXAMINING CAPITAL STRUCTURE TESTS:AN EMPIRICAL ANALYSIS IN THE AIRLINE INDUSTRY.. 355
Kruse Sebastian, Kalogeras Nikos, Semeijn Janjaap
DEUNDERSTANDING THE DIFFERENCE IN PREMIUM PAID IN ACQUISITION OF PUBLIC COMPANIES.... 367
Nahum Biger, Eli Ziskind
THE SPECIALITIES OF THE SMALL- AND MEDIUM-SIZE ENTERPRISES’ FINANCING AND THE
DETERMINANTS OF THEIR GROWTH IN HUNGARY.......................................................................... 377
Zsuzsanna Széles, Zoltán Szabó
CORPORATE GOVERNANCE .................................................................................................. 388
CORPORATE GOVERNANCE PRACTICES AND THEIR IMPACT ON FIRM’S CAPITAL STRUCTURE AND
PERFORMANCE; CASE OF PAKISTANI TEXTILE SECTOR ..................................................................... 390
Hayat M. Awan. Khuram Shahzad Bukahri, Rameez Mahmood Ansari
CORPORATE GOVERNANCE AND COMPLIANCE WITH IFRSS - MENA EVIDENCE................................ 401
Marwa Hassaan, Omneya Abdelsalam
BUSINESS PERFORMANCE EVALUATION MODELS AND DECISION SUPPORT SYSTEM FOR THE ELECTRONIC
INDUSTRY ........................................................................................................................................ 411
Wu Wen
APPLYING THE CONCEPT OF FAIR VALUE AT BALANCE SHEET ITEMS. THE CASE OF ROMANIA ........ 418
Marinela – Daniela Manea
THE IMPACT OF THE INTELLECTUAL CAPITAL ON THE COST OF CAPITAL: BRICS CASE .................... 429
Elvina R. Bayburina, Alexandra Brainis
BRAND VALUE DRIVERS OF THE LARGEST BRICS COMPANIES......................................................... 440
Elvina R. Bayburina, Nataliya Chernova
GOVERNANCE, BOARD INDEPENDENCE, SUB COMMITTEES AND FIRM PERFORMANCE: EVIDENCE FROM
AUSTRALIA ...................................................................................................................................... 451
Wanachan Singh, Robert T Evans, John P Evans
COST-BENEFIT ANALYSIS OF THE FINANCIAL STATEMENTS CONVERSION: A CASE STUDY FROM THE
CZECH REPUBLIC ............................................................................................................................. 464
David Procházka
NEGOTIATING SUCCESSION STRATEGY IN FAMILY RUN SME’S ......................................................... 473
Christopher Milner
CREDIT RATING MODEL FOR SMALL AND MEDIUM ENTERPRISES WITH ORDERED LOGIT REGRESSION: A
CASE OF TURKEY ............................................................................................................................. 483
Özlen ERKAL, Tugcen HATİPOĞLU
7

PRODUCTS WITH LONG AGING PERIOD IN THE AGRO-FOOD SYSTEM: ANALYSIS OF MEAT SECTOR .. 490
Mattia Iotti, Giuseppe Bonazzi, Vlassios Salatas
EMPLOYEE STOCK OPTIONS INCENTIVE EFFECTS: A CPT-BASED MODEL.......................................... 499
Hamza BAHAJI
EMERGING MARKETS.............................................................................................................. 510
COMOVEMENTS IN THE VOLATILITY OF EMERGING EUROPEAN STOCK MARKETS............................. 512
Radu Lupu, Iulia Lupu
OWNERSHIP STRUCTURE, CASH CONSTRAINTS AND INVESTMENT BEHAVIOUR IN RUSSIAN FIRMS.... 522
Tullio Buccellato, Gian Fazio, Yulia Rodionova
MORTGAGE HOUSING CREDITING IN RUSSIA: CRISIS OVERCOMING AND FURTHER DEVELOPMENT
PERSPECTIVES .................................................................................................................................. 523
Liudmila Guzikova
FOREIGN INVESTORS’ INFLUENCE TOWARDS SMALL STOCK EXCHANGES BOOM AND BUST:
MACEDONIAN STOCK EXCHANGE CASE............................................................................................ 532
Dimche Lazarevski
GENDER BIAS IN HIRING, ACCESS TO FINANCE AND FIRM PERFORMANCE: EVIDENCE FROM
INTERNATIONAL DATA..................................................................................................................... 540
Nigar Hashimzade, Yulia Rodionova
INTELLECTUAL CAPITAL DIMENSIONS AND QUOTED COMPANIES IN TEHRAN EXCHANGE, BASED ON
BOZBURA MODEL............................................................................................................................. 541
Rasoul Abdi, Nader Rezaei, Yagoub Amirdalire Bonab
EMPIRICAL FINANCE............................................................................................................... 552
ANALYZING THE LING BETWEEN U.S. CREDIT DEFAULT SWAP SPREADS AND MARKET RISK: A 3-D
COPULA FRAMEWORK ...................................................................................................................... 554
Hayette Gatfaoui
MODELLING OF LINKAGES BETWEEN STOCK MARKETS INCLUDING THE EXCHANGE RATE DYNAMICS 562
Malgorzata Doman, Ryszard Doman
PROPAGATION OF SHOCKS IN GLOBAL STOCK MARKET: IMPULSE RESPONSE ANALYSIS IN A COPULA
FRAMEWORK.................................................................................................................................... 572
Ryszard Doman, Malgorzata Doman
STOCK SPLITS AND HERDING............................................................................................................ 582
Maria Chiara Iannino
8

TIME-VARYING¬ BETA RISK FOR TRADING STOCKS OF TEHRAN STOCK EXCHANGE IN IRAN: (A
COMPARISON OF ALTERNATIVE MODELLING TECHNIQUES).............................................................. 592
Majid Mirzaee Ghazani
9

SCIENTIFIC COMMITTEE
Dr. Ian Cooper Professor, London Business School, UK
Dr. Jerome Detemple Professor, Boston University, School of Management, USA
Dr. Giovanni Barone -
Adesi
Dr. Tompaidis Stathis
Dr. Alireza Tourani-
Rad
Dr. Floros Christos
Dr. Gatfaoui Hayette
Dr. Milionis
Alexandros
Dr. Kondakis Nick
Dr. Chaker Aloui
STEERING COMMITTEE
Prachalias
Chrysovaladis
Dr. Papadimitriou
Athanasios
Professor, University of Lugano and Swiss Finance Institute, Switzerland
Associate Professor, University of Texas at Austin, McCombs School of
Business, Department of Finance, USA
Professor, Auckland University of Technology, Department of Finance,
Auckland, New Zealand
Senior Lecturer, University of Portsmouth, Business School, Department of
Economics, UK
Tenured Associate Professor University Paris I, Economics & Finance
Department, France
Coordinator of Scientific committee, Assistant Professor, University of
the Aegean and Bank of Greece
President and Managing Director of Kepler Asset Management LLC, New
York, USA
Professor, Faculty of Management and Economic Sciences of Tunis, Tunisia,
Director of the International Finance Group-Tunisia, Tunisia
Conference Director, PhD candidate, National & Kapodistrian University of
Athens, INEAG
Lecturer, Faculty of Business Administration University of Piraeus,
Giasla Evangelia MSc, National & Kapodistrian University of Athens, Greece
Dr. Papanagiotou
Evangelia
Dr. Christopoulos
Apostolos
University of the Aegean, Greece
Lecturer, National & Kapodistrian University of Athens, Faculty of
Economics
Mamzeridou Efi Mathematician, MSc from University of Sheffield
10

CONFERENCE COORDINATOR AND SECRETARIAT
Prachalias
Chrysovaladis
Conference Administrative Director , Research and Training Institute of East
Aegean
KEYNOTE SPEAKERS
Dr. Ian Cooper Professor, London Business School, UK
Dr. Jerome Detemple Professor, Boston University, School of Management, USA
Dr. Giovanni Barone -
Adesi
Dr. Kondakis Nick
Professor, University of Lugano and Swiss Finance Institute, Switzerland
President and Managing Director of Kepler Asset Management LLC, New
York, USA
11

SPECIAL KEYNOTE SPEECHES
12

PORTFOLIO SELECTION: A REVIEW
Jérôme Detemple
Boston University School of Management and CIRANO, Boston, USA
Email: rindisbmbu.edu
Keynote lecture: Conference on Applied Financial Economics
April 25, 2011
Abstract. This paper reviews portfolio selection models and provides perspective on some open issues. It starts with a review of the
classic Markowitz mean-variance framework. It then presents the intertemporal portfolio choice approach developed by Merton and
the fundamental notion of dynamic hedging. Martingale methods and resulting portfolio formulas are also reviewed. Their usefulness
for economic insights and numerical implementations is illustrated. Areas of future research are outlined.
Key words: Portfolio choice, Mean-variance model, Diffusion models, Complete markets, Monte Carlo simulation, Malliavin
derivative, Dynamic hedging, Bond numeraire.
1 Introduction
A question of long-standing interest in finance pertains to the optimal allocation of funds among various financial
assets available, in order to sustain lifetime consumption and bequest. The answer to this question is important for
practical purposes, both from an institutional and an individual point of view. Mutual funds, pension funds, hedge
funds and other institutions managing large portfolios are routinely confronted with this type of decision. Individuals
planning for retirement are also concerned about the implications of their choices. Quantitative portfolio models
help to address various issues of relevance to the parties involved.
Mean-variance analysis, introduced by Markowitz (1952), has long been a popular approach to determine the
structure and composition of an optimal portfolio. This type of analysis, unfortunately, suffers from several
shortcomings. It suggests, in particular, optimal portfolios that are independent of an investor’s horizon. A rigorous
dynamic analysis of the consumption-portfolio choice problem, as originally carried out by Merton (1969, 1971),
reveals some of the missing ingredients. It shows that optimal portfolios should include, in addition to meanvariance
terms, dynamic hedging components designed to insure against fluctuations in the opportunity set.
Merton’s analysis highlights the restrictive nature of mean-variance portfolios. Policies of this type are only optimal
under extreme circumstances, namely for investors with logarithmic utility (who display myopic behavior) or when
opportunity sets are deterministic (means and variances of asset returns do not vary stochastically). It also shows
that dynamic hedging terms depend on an investor’s horizon and modulate the portfolio composition as the
individual ages.
Merton’s portfolio formula is based on a partial differential equation (PDE) characterization of the value
function associated with the consumption-portfolio choice problem. This type of characterization, while leading to
interesting economic insights, presents challenges for implementation. PDEs are indeed notoriously difficult (if not
impossible) to solve numerically in the case of high-dimensional problems. This precludes implementations for large
scale investment models with many assets and state variables, and for investors with wealth-dependent relative risk
aversion.1
An alternative characterization of optimal portfolios is obtained by using probabilistic concepts and methods,
introduced with the advent of the martingale approach. Major contributions, by Pliska (1986), Karatzas, Lehoczky
and Shreve (1987) and Cox and Huang (1989), lead to the identification of explicit solutions for optimal
1 Brennan, Schwartz and Lagnado (1997) provide numerical results for a class of low dimentional problems when utilities are constant relative
risk averse.
14

consumption and bequest. Optimal portfolio formulas are derived by Ocone and Karatzas (1991) for Ito processes
and Detemple, Garcia and Rindisbacher (2003) for diffusions. These formulas take the form of conditional
expectations of random variables that are explicitly identified and involve auxiliary factors which, in diffusion
models, solve stochastic differential equations (SDEs). For implementation, Monte Carlo simulation is naturally
suggested by the structure of these expressions.
The martingale approach helps to pinpoint the motivation behind dynamic hedging, namely the stochastic
fluctuations in the interest rate and the market price of risk. Random changes in these variables give rise to an
interest rate hedge and a market price of risk hedge. Alternatively, following Sorensen (1999), Lioui and Poncet
(2001, 2003), Munk and Sorensen (2004) and Detemple and Rindisbacher (2010), it is also possible to rewrite the
hedges using long term bonds as units of account. From this perspective, the hedging behavior is seen to be
motivated by instantaneous fluctuations in a menu of long term bond prices and in the market price of risk expressed
in bond numeraire. The resulting formula highlights the intrinsic interest in long term bonds for asset allocation.
This paper reviews the different characterizations of optimal consumption-portfolio policies derived in the
literature. Section 2 describes the classic mean-variance Markowitz model. Section 3 reviews the dynamic model
developed by Merton. Characterizations of optimal policies and numerical implementations based on martingale
methods are presented in Section 4. Alternative formulas obtained by using long term bonds as numeraires are in
Section 5. Open issues and perspectives for future research are outlined in the concluding section.
2 The mean-variance model
The canonic static model goes back to Markowitz (1952). The model considers an investor who maximizes the
expected utility of terminal wealth
subject to the wealth constraint
Preferences in (1) can be written in terms of a utility function . The coefficient
is an investor specific preference parameter which determines risk tolerance. Initial wealth is . Terminal wealth in
(2) is the result of investing the vector of amounts in the risky assets available and initial wealth net of total risky
investments in the riskless asset. The vector is the vector of fractions of initial wealth in the risky
assets. The vector is the vector of random rates of returns on the risky assets and is the sure rate of return on the
riskless asset. The vector of means of the risky rates of returns is . The variance-covariance matrix is .
Unconstrained maximization with respect to gives the first order condition
Simple derivations lead to the solution,
Theorem 1. (Markowitz (1952)). The optimal portfolio solving the quadratic optimization problem (1)-(2) is
The optimal amount invested in the riskless asset is . The fraction of initial
wealth in the riskless asset is .
The optimal portfolio is linear in expected excess returns and inversely related to the variance-covariance matrix. It
reflects a trade-off between the means and variances/covariances of the excess returns on the risky assets.
15
(1)
(2)
(3)
(4)

Alternatively, one can also decompose the investor choice problem in two parts, the determination of the set of
efficient portfolio and the selection of the best efficient portfolio.
The first of these problems, , is the combination of two sub-problems. The first determines the efficient
set of risky assets. The second identifies the efficient set of risky and riskless assets. Sub-problem is
and has the solution
for . The set of efficient risky portfolios is . The efficient frontier is the
function where .
Sub-problem consists in
and has a solution which satisfies
At the maximum, the efficient frontier of risky assets is tangent to the line generated by combining the best portfolio
of risky assets with the riskless asset. The (overall) efficient frontier is the line
where
is the Sharpe ratio of the tangency portfolio of risky assets and is its variance. Any variance level can be
achieved by combining the tangency portfolio with the riskless asset.
The second problem, , is to maximize preferences subject to selecting portfolios in the efficient set
where is initial wealth. This leads to the condition
16
(5)
(6)

which corresponds to the original solution.
Figure 1 illustrates the principles at work. The tangent portfolio represents the optimal choice among risky assets.
This risky portfolio is then combined with the riskless asset to form the efficient line. The optimal allocation
between the risky portfolio and the riskless asset is determined by the tangency point with the indifference curve.
Mean
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
Optimal portfolio
Efficient frontier and optimal portfolio
Tangent portfolio
0.05
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Standard deviation
Figure 1: Efficient line, optimal indifference curve and optimal portfolio for mean-variance preferences. There are two risky assets with expected
rates of return , returns variances and returns covariance . The riskless rate of return is .
Although the mean-variance preferences (1) are intuitive and easy to work with, they also have undesirable
properties. One drawback is that additional wealth becomes undesirable beyond a certain level (marginal utility of
wealth becomes negative). Another issue is their inconsistency with second order stochastic dominance (Rothschild
and Stiglitz (1970)). A final element is the non-linear structure with respect to probabilities. This property limits
their applicability to dynamic choice settings (time-consistency problem).
3 Dynamic portfolio choice: dynamic programing
Another important limitation of the mean-variance model is the implicit assumption of a constant opportunity set
(the parameters are assumed to be constant). Yet, a wide body of empirical evidence points to stochastic
variations in means and variances of asset returns. The dynamic model developed by Merton (1969, 1971) is
designed to incorporate this important feature of the data.
The model is cast in continuous time. The financial market is comprised of risky assets and riskless asset. The
riskless asset is a money market account with instantaneous interest rate , where is a vector of state
variables. The vector of instantaneous rates of returns on the risky assets and the vector of state variables satisfy
where is the volatility matrix of asset returns, which is assumed to be invertible. The coefficient
is the vector of market prices of risk (the risk premia per unit risk) and is the
17
(7)
(8)

increment in a -dimensional Brownian motion. Invertibility of ensures that the market price of risk is
uniquely defined. It essentially implies that the market is complete, in the sense that all risks are hedgeable. The drift
and volatility of the process followed by state variables are assumed to satisfy the usual
conditions for existence of a unique strong solution (see Karatzas and Shreve (1991), p. 338).
The investor maximizes
subject to the constraints
for all . The preferences, in (9), are von Neumann-Morgenstern and indicate that the individual derives
utility from intermediate consumption (at time ) and from terminal wealth (bequest) . Equation (10) shows
the evolution of wealth when a consumption-portfolio policy is pursued. The inequalities in (11) capture the
physical requirement that consumption be non-negative and the financial requirement that terminal wealth be nonnegative.
The utility and bequest functions are assumed to be strictly increasing, strictly concave and
twice continuously differentiable, and to satisfy the Inada conditions at zero and infinity. Marginal utility and
bequest functions have the inverses . The time argument in these functions stands for a
subjective discount factor, which is assumed to be deterministic.
Merton’s characterization of the optimal policies is given next,
Theorem 2. (Merton (1971)). Let be the value function for the consumption-portfolio problem and let
be its first and second partial derivatives with respect to wealth and the
vector of state variables. Optimal consumption and bequest are
(12)
The optimal portfolio has two components, a mean-variance term and a dynamic hedging term . The
optimal amount invested is where
The value function solves the partial differential equation
where
and subject to the boundary conditions and .
In this dynamic environment, the marginal cost of funds is the derivative of the value function with respect to wealth
. At the optimum, the marginal utility of consumption must equal the marginal cost,
. Optimal consumption is then the inverse marginal utility evaluated at the marginal cost (the
first equation in (12)). The same rationale leads to the optimal bequest function (the second equation in (12)).
In contrast with the static analysis in Section 2, the optimal portfolio now has two terms. The first, (13), reflects the
18
(9)
(10)
(11)
(13)
(14)
(15)
(16)

instantaneous mean-variance trade-off in excess returns. The second, (14), is an intertemporal hedge related to
stochastic fluctuations in the state variables affecting means and variances. This term reflects a concern for the
future coefficients of asset returns. It is interesting to note that the intertemporal hedge vanishes if the interest rate
and the market prices of risk do not depend on the state variables, and . In this case
the value function does not depend on the state variables, , and the cross-partial derivatives
vanish, . Another instance where hedging does not matter is when the investor has logarithmic
utility and bequest functions. The value function becomes additively separable .
The optimal portfolio is then entirely motivated by the risk-return trade-off and the investor displays myopic
behavior.
Numerical implementation of the model requires the computation of the solution of the PDE characterizing the value
function. Except for low dimensional systems, this is a difficult task. Solutions, for realistic models with multiple
state variables and non-linear dynamics are typically difficult, if not impossible, to calculate.
Remark 1. Assume constant relative risk aversion and
where is the relative risk aversion coefficient and is a deterministic subjective discount factor. The value
function is multiplicatively separable , if the function satisfies certain
conditions. The optimal consumption-bequest policy is and . The portfolio
components become
where is the column vector of partial derivative of with respect to .
Under constant relative risk aversion, optimal policies are proportional to wealth. This reduces the complexity of the
PDE characterizing the solution of the model. Nevertheless, even in this case, the numerical resolution becomes a
challenge if the system of relevant state variables does not have a low dimensionality.
4 Dynamic portfolio choice: martingale method
A probabilistic approach can be used in order to overcome computational and other difficulties inherent in the PDE
(dynamic programming) method. This approach relies on martingale theory tools, which were introduced in finance
by Pliska (1986), Karatzas, Lehoczky and Shreve (1987), Cox and Huang (1989), Ocone and Karatzas (1991) and
Detemple, Garcia and Rindisbacher (2003). It also builds on the risk neutralization concept originally introduced by
Cox and Ross (1976) in the context of derivatives valuation and further developed by Harrison and Kreps (1981)
and Harrison and Pliska (1981).
4.1 1. Main results
The core of this approach rests on the identification of the state price density implied by the complete financial
market model postulated. This is the process
which is a path-dependent functional of the interest rate and the market prices of risk. The state price density is the
stochastic discount factor in the model at hand. It gives the value at time of a “dollar” received at time . The
conditional version of the state price density, , gives the value at time of a “dollar” received at time
.
The maximization problem of the investor can be restated in the static form (see Pliska (1986), Karatzas, Lehoczky
and Shreve (1987) and Cox and Huang (1989) for details)
19
(17)
(18)

subject to the static budget constraint
and the non-negativity constraints in (11). In the static problem, optimization is carried out over the consumptionbequest
policy. The portfolio choice is a residual decision. The choice is determined by the need to finance the
optimal consumption-bequest policy.
Theorem 3. (Ocone and Karatzas (1991), Detemple, Garcia and Rindisbacher (2003)). The optimal consumption
and bequest policies are
(20)
where the state price density in (17) and is the unique solution of the equation
The optimal portfolio is with
where , and are the absolute risk tolerance measures
and evaluated at optimal consumption and bequest. Moreover,
where is the Malliavin derivative process that satisfies the linear stochastic differential equation
and are the gradients of with respect
to .
The marginal cost of consumption is the state price density adjusted by a Lagrange multiplier so as to account for
the need to satisfy the static budget constraint. Thus, and the optimal consumption-bequest
policies (20) follow. The equation for is then dictated by the need to saturate the budget constraint. Optimal wealth
is the present value of future consumption and bequest
The optimal portfolio determines the volatility of wealth. To identify it, it suffices to calculate the volatility
coefficients on the left and right hand sides of (26). This can be done by applying the Clark-Ocone formula (see
Ocone and Karatzas (1991) and Detemple, Garcia and Rindisbacher (2003)) and the rules of Malliavin calculus (see
Nualart (1995) for a comprehensive treatment). The leads to (22)-(23).
The portfolio component (22) corresponds to the mean-variance term (13). The new element is that the leading term,
which was written as a ratio of derivatives of the value function in the previous formula, is now seen to represent the
cost of optimal risk tolerance. The portfolio component (23) is the dynamic hedge due to fluctuations in the state
variables. This dynamic hedge has two fundamental parts. The first is motivated by fluctuations in the interest rate.
This is the term depending on in (24). It represents an interest rate hedge. The second is due to fluctuations
in the market prices of risk. It comprises the terms in in (24) and represents a market price of risk hedge.
The two hedges can be written as
20
(19)
(21)
(22)
(23)
(24)
(25)
(26)

where
The Malliavin derivatives in these formulas capture the impact on of a perturbation in the Brownian motion
at time . Malliavin derivatives are similar to impulse response functions used in economics and other disciplines to
describe the reaction of endogenous variables to exogenous shocks.
The formulas in Theorem 3 open the way to implementations for high dimensional systems of state variables. All
expressions in the theorem are in explicit form as functions or functionals of the state price density and other
relevant variables. In the case of the portfolio, the components are expectations of random variables which depend
on the trajectories of the underlying Brownian motions. The Malliavin derivatives involved solve linear stochastic
differential equations (25). Computation of the portfolio components can be performed by simulating the solutions
of the relevant SDEs and calculating expectations by averaging over simulated values. 2 Monte Carlo simulation can
be implemented for arbitrary number of assets and state variables. It can also be carried out for utility functions with
wealth-dependent risk aversions and for non-linear state variable dynamics. 3
4.2 2. Numerical example: stochastic interest rate and market price of risk
A parametric model introduced in Detemple, Garcia and Rindisbacher (2003) is used for illustration of the method.
State variables satisfy the stochastic differential equations
(27)
with
The parameters are constants. The coefficients are
positive and . The Brownian motion has dimension .
The interest rate follows the NMRCEV process (27). This process has nonlinear mean-reversion (NMR) and
constant elasticity of variance (CEV). The speed of mean reversion is the nonlinear function . The
market price of risk follows the MRHEVID process (28). This process exhibits (linear) mean-reversion (MR) and
has a hyperbolic elasticity of variance (HEV), given by the function
Volatility converges to zero as approaches the points and . The mean also depends on the interest rate (ID)
through the function (29). At the points and the interest rate dependence vanishes and is pulled toward
2 An alternative is to apply a change of variables proposed by Doss (1977) in order to stabilize the volatility coefficients of the processes to be
simulated (see Detemple, Garcia and Rindisbacher (2003, 2005a)). The portfolio formula can be rewritten in terms of the transformed state
variables. Portfolio components can then be calculated by Monte Carlo simulation based on the transformed state variables.
3 Simulation-based methods have also been proposed by Cvitanic, Goukassian and Zapatero (2003), Brandt, Goyal, Santa-Clara and Stroud (2005)
and others. A review of simulation methods for portfolio choice appears in Detemple, Garcia and Rindisbacher (2008). A comparison of methods
is carried out in Detemple, Garcia and Rindisbacher (2005b).
21
(28)
(29)
(30)
(31)

. The MRHEVID process wanders stochastically within the band .
Parameter values are displayed in Table 1. With the exception of , they correspond to estimates reported in DGR.
Table 1: Parameter values
Interest rate Market price of risk
= 0.0027668 = 0.85576
= 0.0063138 × 12 = 0.30
= 37.008/(12 × 2 × 0.45432) = 3.0708
= 0.45432 = 1.50
= 0.154055 × 12 = 1.50
= 1.1741 = 2.9417
= 0.50
= 2.8313
Initial values are set at and . The stock volatility is . Implementation is carried out for
the model with HARA utility function over terminal wealth (no intermediate consumption) where and
(i.e., ). Monte Carlo estimates are calculated based on
trajectories and time discretization points per year.
Figure 2 shows the behavior of the optimal portfolio as initial wealth and the horizon vary. Figure 3 shows the
unconstrained portfolio behavior, when terminal wealth is allowed to become negative. Both figures plot the
portfolio and its components as fractions of initial wealth. The nonlinear effect of wealth is especially strong at low
levels of wealth. Nonlinear horizon effects are also important.
2
1
0
4000
3000
2000
Wealth 1000
0.5
0
-0.5
4000
3000
2000
Wealth 1000
Portfolio
10
10
30
20
Horizon
Interest rate hedge
30
20
Horizon
40
40
2
1
0
4000
3000
2000
Wealth 1000
0.5
0
-0.5
4000
3000
2000
Wealth 1000
Mean-variance demand
10
10
30
20
Horizon
Market price of risk hedge
30
20
Horizon
Figure 2: Constrained portfolio behavior for HARA utility with and . Wealth varies from to ,
horizon from to . Computations are based on trajectories and time points per year.
22
40
40

4
2
Unconstrained portfolio
0
0
4000
3000
2000
Wealth 1000 10
30
20
Horizon
40
4000
3000
2000
Wealth 1000
0.5
Unconstrained interest rate hedge
0
-0.5
4000
3000
2000
Wealth 1000
10
Unconstrained mean-variance demand
4
2
-0.1
-0.2
30
20
Horizon
40
-0.3
4000
3000
2000
Wealth 1000
10
10
20
30
40
Horizon
Unconstrained market price of risk hedge
20
30
40
Horizon
Figure 3: Unconstrained portfolio behavior for HARA utility with and . Wealth varies from to ,
horizon from to . Computations are based on trajectories and time points per year.
5 Optimal portfolios and bond numeraire
Additional perspective about the portfolio policy can be gained by using long term bonds as units of account. Such a
change of numeraire has proved useful for pricing fixed income derivatives (Jamshidian (1989), Geman (1989)).
The mathematical underpinnings and economic implications of numeraire changes are studied in Geman, El Karoui
and Rochet (1995). Applications to optimal portfolios are carried out in Sorensen (1999), Lioui and Poncet (2001,
2003), Munk and Sorensen (2004), Lioui (2007) and Detemple and Rindisbacher (2010).
Let be the price of a pure discount bond with maturity date . By standard valuation principles .
Using this bond as new unit of account entails dividing all prices by . The state price density in this bond
numeraire, which is called the forward density, is
The associated measure is called the forward measure. Pricing in the bond numeraire can be done directly, by using
as the stochastic discount factor for cash flows received at time . Thus, the value of a payoff is ,
which can also be written as where is the conditional expectation under the forward measure. It is also
useful to note that the original state price density is .
With this notation, optimal policies become
Theorem 4. (Detemple and Rindisbacher (2010)) Optimal consumption and bequest are and
. Intermediate wealth is
23

The optimal portfolio is with
The volatility of is , where is the volatility of the return on the pure discount
bond . The conditional expectation is under the forward -measure, .
By using long term bonds as units of account, the portfolio has been decomposed in three parts. The first component
is the mean-variance demand, which is now written so as to highlight the impact of the term structure of bond prices.
The second component is a term structure hedge. This hedge has a static flavor, as it is motivated by the
instantaneous fluctuations in the term structure of bond prices. The last component is a forward density hedge. This
hedge has an intertemporal flavor. It is motivated by stochastic fluctuations in the future volatilities of the forward
densities (equivalently, in the future market prices of risk in the bond numeraires).
The formula in Theorem 4 is useful in that it highlights the role of bonds. It immediately identifies a portion of the
dynamic hedging demand in Theorem 3 as a static hedge related to instantaneous fluctuations in bond prices. This
component motivates a fundamental need for bonds. 4 The formula also permits the integration of asset management
and term structure models. Sophisticated models, such as those in the Heath, Jarrow and Morton (1992) or Cox,
Ingersoll and Ross (1985) classes, can readily be used as backbones for asset allocation, by simply plugging the
relevant forward rates or bond prices in the portfolio formula.
6 Perspectives and future research
The models and approaches outlined above permit asset allocation in a variety of contexts taking account of realistic
features of returns. The Monte Carlo methods in Sections 4 and 5 are especially versatile and powerful. This section
discusses some limitations of the methods and focuses on a few areas of future research.
6.1 1. Incomplete markets and portfolio constraints
The assumption that all risks are hedgeable (complete markets) is perhaps the most restrictive aspect of the models
described above. Incomplete markets and portfolio constraints are financial market realities that are important for a
variety of investors, including pension plans, institutional investors and investment banks.
Constraints on portfolios can easily be incorporated in the dynamic programming framework by adding the relevant
inequalities as additional constraints to the optimization problem. In certain rare instances the resulting partial
differential equation for the value function can be solved. In most cases finding a closed form solution is elusive and
a numerical method must be used. As discussed above, the approach quickly runs into a curse of dimensionality.
Unless the model of interest has low dimensionality numerical resolution will prove difficult if not impossible.
Constraints on portfolios can also be incorporated in the static optimization problem. Various characterizations of
the solution have been proposed (see Karatzas, Lehoczky, Shreve and Xu (1991) and Cvitanic and Karatzas (1992)).
Monte Carlo implementation on the basis of these characterizations, however, remains elusive. The main difficulty
4 Models seeking to explain the demand for bonds include Wachter (2003) and Lioui (2007).
24
(32)
(33)
(34)
(35)

is that the characterization of the solution often involves forward-backward stochastic differential equations, which
are notoriously difficult to handle, both mathematically and computationally.
For illustration consider the following basic model with incomplete markets
where is a -dimensional Brownian motion of hedgeable risks and is a -dimensional Brownian motion of
unhedgeable risks. There are risky assets and the volatility matrix is invertible. The vector of market
prices of hedgeable risks is . The price of unhedgeable risks is a -
dimensional vector stochastic process to be identified in the optimization problem. This price is preferencespecific
and captures the shadow price of the incompleteness constraint. The state price density induced by
hedgeable risks is . The density associated with unhedgeable risks is . The stochastic discount factor for valuation
is .
The optimal portfolio in this case can be written as
with
where
In these expressions, represents the Malliavin derivative of the state variable vector with respect to hedgeable
risks . The portfolio has a new term, , with is a hedge against (hedgeable) fluctuations in the shadow price of
incompleteness. The latter is determined by the condition that it is not investable/hedgeable
where are given by (42)-(43) with in place of . This is a forward-backward SDE for .
Implementation of (38)-(44) for general models of the type (36)-(37) requires a numerical scheme for computing the
solution of (44). 5
6.2 2. Jumps
Discontinuous components in asset returns have also been amply documented in the empirical literature. The recent
credit crisis provides a stark reminder of the importance of jumps and the to insure against such events.
5 Solutions, with non-zero shadow price, have been found for special cases (e.g., Detemple and Rindisbacher (2005, 2010)).
25
(36)
(37)
(38)
(39)
(40)
(41)
(42)
(43)
(44)

Models of discontinuous returns are easily formulated. A general model is
where is a jump measure and is its compensator. Jump sizes are random and belong to some
subset of the reals. This structure includes a variety of settings of practical relevance. In particular, it permits
economy wide jumps which affect market returns as well as state variables.
In models with jumps one is immediately confronted with the issue of unhedgeable risks. Jump risk as well as jump
size risk may be unhedgeable. In the dynamic programming approach, the possibility of jumps introduces a nonlocal
component corresponding to the expected jump in the value function, in the Hamilton-Jacobi-Bellman
equation. Except for simple specifications, this term precludes closed form solutions. It also increases the
computational challenges faced by standard numerical methods for PDEs. The key for the martingale approach
remains the computation of the price of incompleteness.
6.3 3. Preferences
The vNM axioms place strong restrictions on individual behavior. The Allais (1953) and Ellsberg (1961) paradoxes
are well known examples of violations of the axioms underlying expected utility theory. In intertemporal settings,
expected utility imposes a direct link between attitudes toward risk and attitudes toward the passage of time. Such a
restriction is difficult to motivate from an economic point of view.
Various preference models have been proposed to address these and other issues. Recursive utility and its
continuous time adaptation, stochastic differential utility (SDU), have become especially popular during the past two
decades (see Epstein and Zin (1989) and Duffie and Epstein (1992)). Preferences in this class permit a separation
between risk and time preferences, leading to a more plausible model of individual behavior. The general SDU
formulation postulates
as a choice criterion. In this specification, is an instantaneous aggregator and is the value function
associated with the consumption-terminal wealth plan . The value function is the solution of the backward
SDE (47). The classic vNM criterion, as in (9), is recovered from the linear aggregator . For
general aggregators, it is not possible to write the value function in closed form, solely in terms of the consumptionbequest
plan.
First order conditions associated with (47) involve the gradient of the value process. Resulting characterizations
of the optimal consumption plan depend on the unknown value function (see Schroder and Skiadas (1999)). Optimal
portfolio formulas inherit this complexity and cannot be easily solved for or analyzed, as in the standard vNM case.
As a result, little is known about the behavior of optimal consumption-portfolio policies. Efficient numerical
methods for dealing with this problem are therefore of substantial theoretical and practical importance.
7 Conclusion
Portfolio selection has undergone a radical transformation in the last 6 decades. Modern methods provide economic
insights into a range of questions pertaining to optimal behavior. They also provide the tools needed for
implementation of optimal rules in contexts of economic and financial relevance. Nonlinear phenomena affecting
securities returns and utility functions with wealth-dependent risk aversions can readily be handled using Monte
Carlo methods. Large numbers of assets and state variables can also be incorporated in the analysis without
compromising on implementability. The flexibility of these methods is of considerable interest for pension plans and
26
(45)
(46)
(47)

institutions seeking to improve the control and performance of their asset management activities.
Yet, much remains to be done in certain areas. The issues discussed in the previous section constitute important
avenues for future research. Another aspect that deserves attention is the question of parameter estimation. This
question is critical for practical implementations. Econometric methods currently available often provide imprecise
estimates for model parameters. Estimation error is a type of risk that needs to be properly accounted for in the
allocation processs.
8 References
[1] M. Allais, “Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école
Américaine,” Econometrica, 21, 1953: 503-546.
[2] M.W. Brandt, A. Goyal, P. Santa-Clara and J.R. Stroud, “A simulation approach to dynamic portfolio choice
with an application to learning about return predictability,” Review of Financial Studies, 18, 2005: 831-873.
[3] M. Brennan, E. Schwartz and R. Lagnado, “Strategic asset allocation,” Journal of Economic Dynamics and
Control, 21, 1997: 1377-1403.
[4] J.C. Cox and C-f. Huang, “Optimal consumption and portfolio policies when asset prices follow a diffusion
process,” Journal of Economic Theory, 49, 1989: 33-83.
[5] J.C. Cox, J.E. Ingersoll and S.A. Ross, “A theory of the term structure of interest rates,” Econometrica, 53: 385–
407.
[6] J.C. Cox and S.A. Ross, “The valuation of options for alternative stochastic processes,” Journal of Financial
Economics, 3, 1976: 145–166.
[7] J. Cvitanic, L. Goukasian and F. Zapatero, “Monte Carlo computation of optimal portfolio in complete
markets,” Journal of Economic Dynamics and Control, 27, 2003: 971-986.
[8] J. Cvitanic and I. Karatzas, “Convex duality in constrained portfolio optimization,” Annals of Applied
Probability, 2, 1992: 767-718.
[9] J.B. Detemple, R. Garcia and M. Rindisbacher, “A Monte-Carlo method for optimal portfolios,” Journal of
Finance, 58, 2003: 401-446.
[10] J. Detemple, R. Garcia and M. Rindisbacher, “Representation formulas for Malliavin Derivatives of diffusion
processes,” Finance and Stochastics, 9, 2005a: 349-367.
[11] J. Detemple, R. Garcia and M. Rindisbacher, “Intertemporal asset allocation: a comparison of methods,”
Journal of Banking and Finance, 29, 2005b: 2821-2848.
[12] J. Detemple, R. Garcia and M. Rindisbacher, “Simulation methods for optimal portfolios,” in Handbooks in
Operations Research and Management Science, 15, Financial Engineering, J.R. Birge and V. Linetsky eds.,
Elsevier, Amsterdam, 2008: 867-923.
[13] J. Detemple and M. Rindisbacher, “Closed form solutions for optimal portfolio selection with stochastic
interest rate and investment constraints, Mathematical Finance, 15, 2005: 539-568.
[14] J. Detemple and M. Rindisbacher, “Dynamic asset allocation: portfolio decomposition formula and
applications,” Review of Financial Studies, 23, 2010: 25-100.
[15] H. Doss, “Liens entre équations différentielles stochastiques et ordinaires,” Annales de l’Institut H. Poincaré,
13, 1977: 99–125.
[16] D. Duffie and L. Epstein, “Stochastic differential utility,” Econometrica, 60, 1992: 353 394.
[17] D. Ellsberg, “Risk, ambiguity, and the Savage Axioms,” Quarterly Journal of Economics, 75, 1961: 643-669.
[18] L.G. Epstein and S.E. Zin, “Substitution, risk aversion, and the temporal behavior of consumption and asset
returns: a theoretical framework,” Econometrica, 57, 1989: 937-969.
[19] H. Geman, “The importance of the forward neutral probability in a stochastic approach of interest rates,”
Working Paper, ESSEC, 1989.
[20] H. Geman, N. El Karoui and J.C. Rochet, “Changes of numéraire, changes of probability measures and option
pricing,” Journal of Applied Probablity, 32, 1995: 443-458.
27

[21] M.J. Harrison and D.M. Kreps, “Martingales and arbitrage in multiperiod securities markets,” Journal of
Economic Theory, 20, 1979: 381-408.
[22] M.J. Harrison and S.R. Pliska, “Martingales and Stochastic integrals in the theory of continuous trading,”
Stochastic Processes and their Applications, 11, 2001: 215–260.
[23] D. Heath, R.A. Jarrow and A. Morton, “Bond pricing and the term structure of interest rates: a new
methodology for contingent claims valuation,” Econometrica, 60, 1992: 77-105.
[24] F. Jamshidian, “An exact bond option formula,” Journal of Finance, 44, 1989: 205-09.
[25] I. Karatzas, J.P. Lehoczky and S.E. Shreve, “Optimal portfolio and consumption decisions for a “small
investor” on a finite horizon,” SIAM Journal of Control and Optimization, 25, 1987: 1557-1586.
[26] I. Karatzas, J.P. Lehoczky, S.E. Shreve and G.L. Xu, “Martingale and duality methods for utility maximization
in an incomplete market,” SIAM Journal on Control & Optimization, 29, 1991: 702-730.
[27] I. Karatzas and S.E. Shreve, Brownian motion and stochastic calculus, 2nd Edition, Springer Verlag, New
York, 1991.
[28] A. Lioui and P. Poncet, “On optimal portfolio choice under stochastic interest rates,” Journal of Economic
Dynamics and Control, 25, 2001: 1141-1865.
[29] A. Lioui and P. Poncet, “International asset allocation: a new perspective,” Journal of Banking and Finance,
27, 2003: 2203–2230.
[30] A. Lioui, “The Asset Allocation Puzzle is still a puzzle” Journal of Economic Dynamics and Control, 31,
2007: 1185-1216.
[31] H. Markowitz, “Portfolio selection,” Journal of Finance, 7, 1952: 77-91.
[32] R.C. Merton, “Lifetime portfolio selection under uncertainty: the continuous time case,” Review of Economics
and Statistics, 51, 1969: 247-257.
[33] R.C. Merton, “Optimum consumption and portfolio rules in a continuous-time model,” Journal of Economic
Theory, 3, 1971: 273-413.
[34] C. Munk and C. Sorensen, “Optimal consumption and investment strategies with stochastic interest rates,”
Journal of Banking and Finance, 28, 2004: 1987-2013.
[35] D. Nualart, The Malliavin Calculus and Related Topics, Springer Verlag, New York, 1995.
[36] D. Ocone and I. Karatzas, “A generalized Clark representation formula, with application to optimal portfolios,”
Stochastics and Stochastics Reports, 34, 1991: 187-220.
[37] S. Pliska, “A stochastic calculus model of continuous trading: optimal portfolios,” Mathematics of Operations
Research, 11, 1986: 371-382.
[38] M. Rothschild and J.E. Stiglitz, “Increasing risk I: a definition,” Journal of Economic Theory, 2, 1970: 225-
243.
[39] M. Schroder and C. Skiadas, “Optimal consumption and portfolio selection with stochastic differential utility,”
Journal of Economic Theory, 89, 1999: 68-126.
[40] C. Sorensen, “Dynamic asset allocation and fixed income management,” Journal of Financial and Quantitative
Analysis, 34, 1999: 513-531.
[41] J. Wachter, “Risk Aversion and Allocation to Long-term Bonds,” Journal of Economic Theory, 112, 2003:
325–333.
28

ENHANCING FILTERED HISTORICAL SIMULATIONS
Giovanni Barone-Adesi
The Swiss Finance Institute at USI
AFE, Samos, July 1 2011
Abstract. Filtered historical simulation (FHS) is a powerful tool to simulate returns of complex
portfolios. Assets can follow different distributions. Time dependencies are accounted by filtering,
while cross dependencies are preserved by the choice of simultaneous standardized residuals. This last
procedure does not account for the stronger clustering of returns observed across assets in very
negative days. We discuss a new procedure to introduce this effect without hampering the nonparametric
nature of our simulation procedure.
The accurate simulation of security returns is a necessary step for the valuation and hedging
of securities. In fact security returns do not conform generally to the simplifying assumptions
of closed-form models, such as normal diffusion. Often the simulation task is complicated by
the need to simulate well the distribution of security returns consistent with current market
conditions, such as short-term volatility, while ensuring convergence to the long-term
distribution of security returns implied by long-dated contingent claims.
FHS. A method to simulate security returns accurately has been proposed by Barone-
Adesi in several papers with Bourgoin, Giannopoulos and Vosper (1998, 2000, 2001, 2002).
Their application was originally to risk-management, though later an empirical change of
measure allowed Barone-Adesi, Engle and Mancini (2008) to price index options.
The original filtered historical simulation (FHS) was based on the following assumptions:
1) Returns are generated by a time-series process of the GARCH family
2) The parameters of the time-series process can be estimated through the minimization
of squared errors with little loss of efficiency from the true likelihood function.
3) Correlations across assets follow a multivariate process with outcomes independent of
the scale of returns. As a special case, correlations may be constant.
Assumption (1) implies that standardized returns are stationary.
Assumption (2) implies that usual estimation procedures can be used.
Assumption (3) implies that scale changes do not prevent the use of parallel bootstrapping in
generating strips of simulated security returns across assets.
Our last assumption may be mis-specified because there is some empirical evidence (see
Embrechts for a summary) that clustering of negative returns may increase with scale. If that
is the case, our parallel bootstrapping is unable to adjust correlations to scale, making our
simulated distribution different from the original one. Note however that larger simulated
returns tend to be generated from larger standardized residuals and volatilities. These
variables show common clustering in time. Therefore our simulated returns preserve the
higher correlation found in larger observed returns, though they may fail to reflect the
relationship between size of returns and correlation accurately. The original FHS
29

extrapolation process may be considered a first order approximation of the return distribution
under these circumstances.
The FHS approach can be adapted to stress testing because it simulates the whole
distribution of security returns. It is not limited to the observed returns as regular
bootstrapping is. Therefore it is possible to sample from more extreme points in the tails of
the multivariate distribution by increasing the number of simulation runs. As an example,
applying FHS over a 10-day horizon using a database of 500 daily returns, the number of
different possible pathways is 500 10 for any given set of initial conditions. Therefore it is
possible to generate an almost arbitrarily large number of points of the simulated distribution.
This is especially important for portfolios of derivatives that may experience more stress at
points not on the tails of the distribution.
Finding the most stressful conditions for a given portfolio may be slow, because the
probability of each pathway is the inverse of the number of pathways. Of course the problem
of finding the most stressful conditions is easier for linear portfolios, for which the most
stressful conditions are given by the largest negative or positive returns, that can be preselected
in the simulation procedure.
Enhanced FHS. A limitation on the accuracy of FHS, that cannot be remedied by
increasing the number of simulation runs, stems from the neglect of the stronger clustering
often observed in empirical asset returns. Multivariate copulas of asset returns are not
Gaussian, even after they are standardized by dividing them by their time-changing standard
deviation.
Multivariate copulas are unfortunately cumbersome to calibrate, because of very low
probabilities associated with each cell far out in the tails. Moreover simulating returns
consistent with them would require multivariate random number generation that follows the
chosen copula, a cumbersome task of its own. Remember that FHS requires only selecting a
date within the historical data base at each step of the simulation.
Is it possible to improve the accuracy of FHS while retaining most of its simplicity? In
principle an obvious strategy could be to sample only standardized returns associated with the
current level of market volatility. This way strips of returns across assets in high volatility
days will reflect the clustering typical of those days, rather than the average correlation across
all the days in the historical data.
The above strategy may work only with extremely long series of historical data. The
reason is that the above scheme effectively partitions the historical data into volatility
buckets. Only one of them is available to draw simulated returns at each given day. However,
as Pritsker (2001) shows, FHS becomes fairly unreliable if only few historical returns are
available. With daily returns, ‘few’ is of the order of 500.
If we require that each volatility bucket contains 500 observations, we cannot use many
buckets of daily data. Otherwise our stationarity assumptions may be stretched. Empirical
investigations in progress (Barone-Adesi, Mira and Solgi, in progress) suggest that two
buckets may be the optimal compromise. Therefore we sample each day from the high or the
low volatility bucket of standardized returns, depending on our computed GARCH volatility.
30

This technique has been backtested on equally weighted portfolios of two stocks of DJIA
index (we have tested all possible pairs of components of this index), and daily VaR at 99%,
95%, and 90% confidence intervals are estimated. Our results show that the new proposed
technique (by using different joint distributions for the innovations in the low and high
volatility regimes) improves the VaR estimates.
References
Barone-Adesi G., F. Bourgoin and K. Giannopoulos ‘Don’t Look Back”, Risk, August 1998.
Barone-Adesi, G., K. Giannopoulos and L. Vosper ‘VaR without Correlations for Portfolios
of Derivative Securities’, Journal of Futures Markets, August 1999.
Barone-Adesi, G., K.Giannopoulos, and L.Vosper, 2001. ‘Backtesting derivative portfolios
with filtered historical simulation’, European Financial Management 8 (1), 31–58.
Barone-Adesi, G., Giannopoulos, K., 2002. ‘Non-parametric VaR techniques; myths and
realities’, Economic Notes 30 (July), 167–181.
Barone-Adesi, G., Engle R. F., Loriano M., 2008, ‘A GARCH Option Pricing Model with
Filtered Historical Simulation’, The Review of Financial Studies 21, 1224-258.
Embrechts, P., F. Lindskog, and A. McNeil (2003) ‘Modelling Dependence with Copulas
andApplications to Risk Management’. In: Rachev, S.T., ed., Handbook of Heavy Tailed
Distributions in Finance, Elsevier: North Holland, pp. 329-384.
Pritsker M., The Hidden Dangers of Historical Simulation’, Finance and Economics
Discussion Series 2001-27. Washington: Board of Governors of the Federal Reserve
System, 2001.
31

International propagation of the credit crisis*
Richard A. Brealey, Ian A. Cooper, and Evi Kaplanis**
Original version: November 2010
This version: April 2011
Abstract
In this study we examine the propagation of the recent crisis to banks outside the US.
We develop a framework combining stock market and structural variables that can be
used with both individual bank and country aggregate data. We find that the
differential incidence of the crisis measured by share price impact is explained by
prior correlation with the US banking sector, bank leverage and liability structure, the
foreign assets of banks, and the importance of banking in the economy. We find that a
simple measure of bank capital was a better predictor of crisis impact than the riskweighted
measure of Basel II, but do not find that banks were penalized for making
aggressive use of Basel II rules. These results are robust to various specifications and
whether we use country data or individual banks. Using this framework we test a
number of hypotheses which have been put forward in other studies. We find that
some results are sensitive to sample selection and test specification. We do not find
evidence that the incidence of the crisis was associated with mortgage holdings, stock
market returns prior to the crisis, or standards of governance. We do find that
countries with higher prior GDP growth suffered less in the crisis. We discuss the
implications of our results for bank regulation.
JEL Codes: F33, F36, G1, G15, G18, G21, G28, G32, G34.
Keywords: banking, financial institutions, capital structure, risk management,
international finance, monetary economics, credit crisis, international transmission,
contagion, financial regulation.
* We thank participants at seminars at CORE/University of Louvain, the University of
Luxembourg, and the University of Southern Denmark. This paper can be
downloaded from: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1712707.
** All authors at London Business School, Sussex Place, Regent’s Park, London
NW1 4SA, England, +44-207-000-7000, icooper@london.edu (corresponding
author), rbrealey@london.edu, ekaplanis@london.edu.
32

1. Introduction
The recent financial crisis has provided a natural experiment with which to test
hypotheses about crisis origination, propagation, and incidence. This has led to a
vigorous debate about the factors which caused some banks and countries to suffer
more than others. As well as having implications for modelling crises, this debate has
important practical implications for the design of regulation and economic
management. Some studies even appear to show that features of banks and economies
which are favorable in normal times led to worse outcomes during the crisis. Hence it
is important to understand as completely as possible the mechanism which caused
these relationships.
There have been two broad approaches to examining the impact of the crisis: using
country aggregates and using individual banks. At an aggregate level Rose and
Spiegel (2009, 2010) fail to find a relationship between cross-country variation in the
impact of the crisis and variables measuring cross-country trade and financial
linkages. Frankel and Savelos (2010), using more measures of crisis incidence and a
different measurement period, obtain somewhat stronger results. In line with previous
crises, they find that the level of central bank reserves and real exchange rate
overvaluation are significant indicators of the cross-country impact of the crisis. In
addition, lower past credit growth, larger current accounts and savings rates, and
lower external and short-term debt were associated with lower crisis incidence,
although these results are not robust across different crisis incidence measures and
specifications. Lane and Milesi-Ferretti (2010) focus on the impact of the crisis on
GDP and find that growth during the crisis was lower in countries with high GDP per
capita, high pre-crisis growth, and larger current account deficits.
At an individual bank level Acharya, Pedersen, Philippon, and Richardson (2010)
develop an approach based on pre-crisis stock price data. They propose a measure
equal to the return on a bank stock in the worst 5% of weeks for the index return
during the pre-crisis period. They combine this with a measure of leverage to give an
indicator which explains a significant amount of the share price impact of the crisis on
different US financial institutions. In contrast Beltratti and Stulz (2009) focus on
structural variables which measure characteristics of the banks and of regulatory and
governance regimes, rather than share price variables. They examine the factors which
explain the differential share price returns during the crisis for large banks from a
number of countries, including the United States. They conclude that “Overall, our
evidence shows that bank governance, regulation, and balance sheets before the crisis
are all helpful in understanding bank performance during the crisis.”
Other studies have examined specifically the role of multi-national banks in the
transmission of the recent banking crisis. Some of these (Popov and Udell (2010),
Navaretti et al. (2010), and Allen, Hryckiewicz, and Kowalewski (2010)) have looked
at the general issue of whether foreign-owned banks serve as a stabilising influence
and what causes them to adjust their activity in the host country. They show that the
activities of the bank in the host country are affected by characteristics of the parent
bank, such as its fragility, its losses on financial assets, and its reliance on interbank
borrowing.
33

The above studies focus on the recent crisis. There is in addition a more general
literature on the transmission of banking crises both domestically and cross-border.
Of most direct relevance here is the body of work that views transmission as a
consequence of linkages in financial institutions or investor portfolios. 1 For example,
Allen and Gale (2000) show how financial crises can spread as a result of the impact
on the interbank market of changing demands for liquidity. In this case, the degree to
which particular regions are affected by a crisis in one region depends on the
particular structure of the linkages between regions. Liquidity shocks can also work
directly through financial markets if an increase in demand for liquidity obliges
investors to reduce their exposure in a number of markets (e.g., Calvo (2005) and
Yuan (2005)).
Such linkages imply that the extent to which a shock is transmitted to another region
depends on the structure of the assets and liabilities of financial institutions or their
shareholders. Moreover, the resulting financial contagion is characterized by shifts in
the degree of comovement between bank values, so that the severity and pattern of a
crisis cannot be predicted simply from the comovement in non-crisis periods. 2
However, the role of international linkages in the recent crisis is unclear. Lane and
Milesi-Ferretti (2010) find that measures of international linkages such as trade
openness have little explanatory power with respect to differential crisis impact. Rose
and Spiegel (2010)) reach the counterintuitive conclusion that “if anything, countries
seem to have benefited slightly from American exposure.”
In this study we contribute to this literature in three main ways. First, we use a
combination of share price and structural variables to explain the impact of the crisis
for non-US countries and banks. We find that a stock market measure of international
linkages, the pre-crisis correlation of a foreign bank’s share return with the US bank
share return index, explains a significant amount of cross-country and cross-bank
differences in crisis impact. In common with contagion studies, we also find that the
relationship between share returns changed during the crisis and that the exposure of
banks to the crisis is related to structural variables. The variables we find to be
important measure the leverage, liability structure, international holdings, and size of
a country’s banking system. Hence both stock market and structural variables need to
be combined to give a more complete specification of the relationships that caused
differential crisis impacts. Omitting either could lead to misidentification of the
causes.
Second, we test our hypotheses using data for both individual banks and country
indexes. The framework we use is linear in the characteristics of individual banks, and
therefore gives an aggregate measure of systemic risk that is consistent with the
measures for individual banks. Hence it could be used to measure the contribution of
individual banks to country-wide systemic risk. We find that the results for our main
predictive variables are similar for both individual-bank and country-index data.
However, we find that results for some other variables are sensitive to data
availability and sample selection.
1 For a review of the literature on the transmission of financial crises, see Allen and Gale (2007).
2 There is a large empirical literature on the issue of whether correlations between markets increase
during crisis periods. See, for example, Bennett and Kelleher (1988); King and Wadhwani (1990);
Wolf (2000), Forbes and Rigobon (2002), and Corsetti et al (2002).
34

Third, we test a number of hypotheses that have been supported by other studies. For
example, it has been suggested that high exposure to the crisis was associated with:
1. aggressive use of the Basel rules (IMF (2008);
2. high pre-crisis levels of GDP growth (Lane and Milesi-Ferretti (2010));
3. high pre-crisis share returns and good governance (Beltratti and Stulz (2009));
4. economic development (Lane and Milesi-Ferretti (2010);
5. low international linkages (Lane and Milesi-Ferretti (2010), Rose and Spiegel
(2010)).
All these have important implications for policy but are in several cases
counterintuitive. The different empirical approach and different sample of our study
enables us to test the robustness of these propositions.
Our approach is most closely related to Beltratti and Stulz (2009) but it differs from
theirs in several important respects. Our framework uses the knowledge that the crisis
was transmitted from the US to other countries so we include a stock market measure
of the linkage between banks and the US banking sector, which we find to be highly
significant. Since our focus is on the performance of non-US banks, we collect a
much broader sample of these banks and use both individual-bank and country-
aggregate data. Our country sample consists of 50 non-US countries and our sample
of individual banks includes 381 non-US firms. The combination of individual-bank
and country data allows us to test whether the individual bank results are robust at the
aggregate level.
Our study is related to Acharya et al (2010), in that we use a measure based on
comovement between share prices as a primary variable. However, our interest is
international propagation rather than domestic US impact, and we use correlation
rather than their measure of extreme comovement. We examine which measure
performs better in the international context of our study and find that the correlation is
significantly better. We also combine the stock price measure with structural
variables, and find that these improve the cross-sectional prediction relative to the
stock market variable used alone.
2. Measuring the international propagation of the crisis
A full test of Allen and Gale (2000) or similar models would require detailed
information on the complex linkages between banks. Our interest is to develop a
parsimonious representation in which the incidence of the crisis can be explained by a
small number of observable variables. We derive our empirical approach from
knowledge of the mechanism that led to the development of the crisis. This concerns
the banking sector of a country and its links to the US banking sector. We do this
because:
1. The crisis was primarily a banking-sector crisis which then spread to the
remainder of the economy. So we look for transmission via the banking
sectors of different countries.
2. The crisis originated in the US, so its propagation should depend on
linkages with the US.
35

We consider a crisis emanating from country O concentrated in industry K. In this
paper the country of origination is the US and the industry is the banking industry. We
assume that the propagation of the crisis takes place via links between firms that are
members of industry K in different countries. Initially, we describe these links by the
relationship that exists in normal times between industry K in country j and industry
K in country O. We measure this by the regression of equity index returns of industry
K in country j on the industry stock index return in country O:
R = a + b R + e
(1)
j j j O j
t t t
Where: j = 1,..J, e ∼ N σ , R ∼ N σ ,
j
2
t (0, j )
O
2
t (0, O)
j
Rt is the return on the stock index of
industry K in country j in period t, and R Ot is the return on the stock index of industry
j
K in country O in period t. The key parameter in this regression is b , which
measures the responsiveness of industry K in country j to industry K in the country of
j
origin of the crisis. In normal times we have the standard expression for b :
j j j O
b = ρ σ / σ
(2)
j
where ρ is the correlation between R and O
R .
j
t
We assume that a crisis occurs in the period ( T, T + τ ) . We model the propagation of
the crisis by the relationship between the total return on the shares of industry K in
different countries during this crisis period. If the crisis period were simply a scaledup
version of a normal period, the cross-sectional relationship between returns would
R :
be the one resulting from equation (1) with a constant term substituted for O
t
Where:
R = a + Rb + e
(3)
j j j j
C C
j
C
t
R is the return on industry K in country j during the crisis period ( T, T + τ ) ,
R = R is the return on industry K in country O during the crisis period, and
j
C
O
C
(0,
2
j )
e ∼ N σ τ .
Equation (3) describes the cross-sectional relationship we would expect to hold in a
j
non-crisis period. It is heteroskedastic, so we scale by the standard error of e to give:
r = ψ + θρ + u
(4)
j
C
j j
j j j
O j
Where: rC = RC / σ , θ = R / σ , u ∼ N(0, τ ) . Equation (4) says that in non-crisis
times the normalised return in country j, conditional on the return in country O, is
proportional to its correlation with country O.
In a crisis period, however, the same relationship may not hold. We model the
difference between a crisis period and a non-crisis period by making the parameter θ
depend on other variables which measure the transmission mechanism of the crisis:
36

j j
θ = θ ( X )
(5)
j
Where X is a vector of variables which measure the vulnerability of country j to the
crisis. The model then becomes:
j j j j
r = ψ + θ ( X ) ρ + u
(6)
C
If the variables have an additive effect, equation (6) becomes:
r = ψ + ψ X + .. + ψ X + ψ ρ + u
(7)
j j j j j
C 0 1 1 n n n+
1
Equation (7) is the specification we use in the study.
3. Data
We test our hypothesis with two data sets. The first consists of aggregate banking
data for a sample of 50 countries. We then conduct a similar series of tests using data
for a sample of nearly 400 individual banks. This provides a substantial increase in
sample size but at the possible cost of more noisy data. The appendix lists the data
definitions and sources.
Banks differ considerably in the presentation of their accounts. Therefore, any
database of balance-sheet variables encounters an inevitable problem of consistency.
This problem is likely to be particularly severe in a cross-country study and adds to
the noise in the data for both our country-level tests and those for individual banks.
3.1 Country-Level Data
For the country-level tests we include all countries for which the following data are
available: (1) Datastream bank industry equity return indices for the period January
2005 to March 2009; (2) IMF International Financial Statistics data covering banking-
sector Total Assets, Foreign Claims, Demand Deposits, and Time Deposits for the end
of 2006; (3) IMF aggregate capital ratios for the banking sector; (4) IMF aggregate
Basel regulatory capital ratios for the banking sector. Our sample includes the 50
countries shown in Table 1. These cover 91% of world GDP excluding the US (using
IMF data for 2009).
j
We measure the impact of the crisis, R C , as the average weekly return on Datastream
bank industry equity indices in the period 21 May 2007 to 9 March 2009, a period in
which the U.S. index of bank returns fell by 79%. 3 We measure the independent
variables from the period before the crisis. We use 2 years of weekly data from
January 2005 to December 2006 to compute the correlation between a country’s bank
j
stock index and that of the US, ρ , and the standard deviation of each country’s bank
j
stock index return in non-crisis times, σ . We use the same data to calculate the
3 We also ran both the country and individual bank regressions with the cumulative return May 2007-
March 2009 as the dependent variable. The results were qualitatively similar.
37

extreme value measure suggested by Acharya et al. We leave a gap between this
period of data measurement and the crisis period to ensure that our independent
variables would have been known by May 2007 and to allow for uncertainty about the
exact dating of the crisis. To measure banking variables (which are included in the
j
vector of explanatory variables, X ) we use primarily IMF aggregate data for a
country’s banking sector at the end of 2006. 4 We use the data for “Other depository
corporations”, which are largely banks. This and other definitions, together with data
sources for the country-level data are given in the appendix.
The banks included in the Datastream Indices all have publicly listed stock, whereas
the IMF data include government-owned banks and cooperatives. If publicly owned
banks have different characteristics, this may affect our results. The sample size in
such a study is naturally limited. However, the excluded countries are generally tiny
and it is not clear that they would add much extra information to the study.
Given the limited number of observations, we take care to use only a few independent
variables and not to engage in data-mining. We select those variables that could play a
significant role in the transmission mechanism of the crisis from the US banking
sector to the banking sector of country j. Our hypothesised variables are:
• Leverage of the banking sector, which we measure by a capital ratio;
• Fragility of the banking sector, which we measure by the proportion of bank
liabilities that are not deposits. We hypothesise that these represent less
permanent sources of funding;
• Fragility as measured also by the use of derivatives;
• International linkages not captured by the correlation with the US, which we
measure by the proportion of foreign assets held by banks;
• The importance of the banking sector in the economy, which we measure by
the ratio of bank assets to GDP.
We augment these variables with four supplementary measures that have been
suggested in other studies as associated with crisis returns. These include two
governance measures, the prior return on bank stocks, and the prior growth in GDP.
Previous empirical studies suggest that these measures were negatively related to bank
performance during the crisis.
The variable that presents the most difficulty in measurement is the fragility of the
banking system. We estimate this in two ways. The first is the amount of less
permanent sources of funding. We hypothesize that time deposits and, to a lesser
extent, demand deposits are likely to represent more permanent sources of funding
and involve fewer linkages to other banks. Other liabilities are likely to represent
sources of funding that are more susceptible to early flight risk as a banking crisis
emerges. Our second measure of fragility is the use of derivatives. We hypothesize
that large derivatives positions are likely to make the banking system more fragile,
both because of the implicit leverage that derivatives contain and because derivatives
may serve to transmit risk internationally to other banks.
We were not able to locate a reliable country-level measure of the ratio of bank equity
to total assets. For our country study we therefore use two other measures of leverage
4 Banking data for Taiwan are taken from the Central Bank’s Website.
38

– (1) the ratio of capital to total assets, where capital includes both equity and
subordinated long-term debt, and (2) the Basel ratio of Tier 1 + 2 capital to riskweighted
assets. Both measures were taken from IMF Global Financial Stability
Report.
We include a measure of the size of the banking sector relative to GDP. A large
banking sector is more likely to have international linkages. In addition, it may cause
other problems in the economy as the crisis evolves. In this case the crisis could be
accentuated by feedback between the banking sector and the other parts of the
economy.
An OECD report argues that “the financial crisis can be to an important extent
attributed to failures and weaknesses in corporate governance arrangements”
(Kirkpatrick (2008)). Beltratti and Stulz (2009) test this assertion but in contrast to
the OECD they find that banks with more shareholder-friendly boards performed
worse during the crisis. We check whether this is also true for our sample. We use
two country-level measures of corporate governance, both described in Djankov et al
(2008) and reported in detail in Djankov et al (2005). The first is their anti-selfdealing
index, which measures the degree of protection that each country provides
against a specified tunnelling transaction. The second measure is their revised index
of anti-director rights, which updates and extends La Porta et al. (1998). Both
variables are available for 45 of our countries.
Finally, we examine two other measures that previous studies have found to be
associated with crisis returns. The first is the return on the country’s bank stocks
during 2006 and the second is the average rate of GDP growth in the five years to
2006.
In our regressions we transform those balance-sheet ratios that we expect to have a
positive association with returns by subtracting them from 1.0. Therefore, the
predicted sign on the coefficient is negative for each of these independent variables.
Table 1 provides some summary data for the dependent variable. Of the 50 countries
only China’s banking sector experienced a positive return from May 2007 to March
2009. There are, however, some regional patterns in the data. European banks
experienced unusually sharp falls in value, with Ireland, the most affected country,
experiencing a mean weekly return of -2.6%. Emerging and developing economies
generally fared better with a mean raw return of -0.6% per week, compared with a
mean return of -1.2% for advanced economies. 5 There is no relationship between the
severity of the declines in prices and the prior variability of returns; some of the
apparently most stable banking markets suffered the sharpest falls in value.
The distribution of the variables and their correlations are given in Tables 2 and 3.
The simple correlations between the standardized return and our main independent
variables (column 1) all have the predicted negative sign. The correlation between the
standardized return and each of the governance variables is close to zero. However,
the standardized return is quite strongly positively correlated with the prior growth in
GDP.
5 We use the definitions in the IMF’s World Economic Outlook.
39

3.2 Individual Bank Data
Our sample of individual banks consists of the components of the Datastream World
Bank Index in 2010. Since this list is subject to potential survivorship bias, we
supplemented it by merging it with the 200 largest banks by total assets in 2006,
based on The Banker’s 2007 listing of the top 1000 banks at the end of the previous
year. We exclude those banks whose stocks were first listed after the start of 2005.
The remaining sample includes companies that are not principally commercial banks.
For example, some are bancassurance companies, investment banks, or asset
managers. We exclude three cases where a bank also acts as the central bank, but
otherwise do not attempt to make what would be inevitably judgmental exclusions.
The result is a sample of 381 banks from 50 countries. 6 In contrast to our countrylevel
data, the sample does not include banks from Bulgaria, or Slovenia, but does
include banks from Bahrain and Iceland.
The returns and balance-sheet data for individual banks are taken from Datastream,
and are supplemented by data from Osiris and the banks’ annual reports. The
dependent variable is the average weekly return on the bank stock from May 2007 to
March 2009. We normalize this return by dividing by the standard deviation of
returns in the period 2005-2006. Where a bank was acquired for stock we include the
subsequent return on the stock of the acquiring company. Where a bank was
nationalized or acquired for cash, we include the cash payment and assume a zero
return in the subsequent weeks. 7
Definitions of the independent variables are shown in the Appendix. The first group
of variables in the table largely parallel those used in the country-level regressions.
For individual banks we consider three measures of leverage -- the ratio of equity to
total assets, the ratio of total capital to total assets, and the Basel Tier 1 + 2 capital
adequacy ratio. Where available, we collect separate measures of demand and time
deposits. However, many banks report only total deposits and therefore we use this
measure as an alternative independent variable.
The amount of foreign loans is available for only a small proportion of our sample.
Therefore, as in the country-level regressions, we use the country-average ratio of
foreign loans to assets. We also use the same measure of the ratio of banking assets to
GDP that we use in the country-level regressions.
In addition to our principal independent variables, we also examine four other balance
sheet variables that may provide information about the bank’s exposure to the crisis.
The first is the ratio of the bank’s short-term debt to total assets. Thus instead of
measuring fragility by the proportion of bank funding that is not provided by deposits,
we use the proportion that is funded by short-term debt. We predict that a bank that
relies on short-term wholesale funding will be more susceptible to the crisis. 8
6 Of these banks 360 were members of the Datastream indices and 21 were added from The Banker.
7 This assumption is largely immaterial. Payments for banks that were rescued by acquisition or
nationalization were generally either zero or very small.
8 Osiris provides a measure of money-market funding. This is available for a smaller sample and does
not offer any improvement over the Datastream measure.
40

Second, real-estate loans have been a common source of banking crises (Herring and
Wachter (1999) and Reinhart and Rogoff (2008)) and more specifically played a
leading role in the 2007 crisis. To the extent that banks were exposed to the US real-
estate market or that there was contagion across countries in real-estate markets, we
expect banks with a high real-estate exposure to be more sensitive to the crisis in the
US. Datastream provides data on the level of mortgage loans for a substantial
subsample of our banks, but data on holdings of mortgage-backed securities are
available for just a small subsample. We therefore look only at the ratio of mortgage
loans to total assets.
Finally, a bank’s involvement in the interbank market may serve as a proxy for its
international linkages. We therefore also collect data on interbank loans (an asset),
and loans due to other banks (a liability).
We again augment these measures with three supplementary variables that previous
studies suggest may be associated with bank performance during the crisis. These are
a measure of corporate governance, for which we use the Corporate Governance
Quotient (CGQ®), the average weekly return on the bank stock in 2006, and the
growth in country GDP over the five years ending in 2006.
As in the case of the country data, we restate the independent balance-sheet variables,
so that the predicted coefficient on each is negative. Tables 4 and 5 summarize the
distributions of the variables used in the individual bank regressions and their
pairwise correlations. The pervasive nature of the crisis is illustrated by the fact that
the mean return was negative for over 90% of the observations, with a mean weekly
return of -0.7%. Again there was a substantial difference between the performance of
banks in emerging and developing economies and those in advanced economies. In
the former case the mean weekly raw return was -.4% and in the latter case it was
-.9%.
The simple correlations between the standardized return and the main independent
variables all have the predicted negative sign. Both the governance variable and the
prior stock return are weakly negatively correlated with the standardized crisis return,
while the growth in GDP continues to be quite strongly positively correlated. In the
case of the independent variables, there is a high correlation between the three
measures of bank capital and, not surprisingly, between total deposits and time
deposits. There is a strong negative correlation between short-term debt and total
deposits.
The correlation between returns on the bank and the US banking index seeks to pick
up linkages between banks that are not captured by other independent variables such
as the relative amount of interbank activity or the size of the banking sector. The
weak association between these balance-sheet variables and the return correlation
suggests that the former may not be adequate proxies of bank linkages.
4. The role of correlation and leverage
4.1 Can pre-crisis correlation explain the propagation of the crisis?
41

Equation (4) says that the prior correlation with the US banking sector should help to
predict the impact of the crisis. Therefore, we first test whether the relationships that
hold in normal times can explain the cross-sectional impact of the crisis. The first
column of Table 6 shows the regression of the standardized crisis return variable on
the pre-crisis correlation with the US banking sector. Panel A is for the country
sample, Panel B for the full sample of individual banks, and Panel C for the restricted
sample of individual banks for which we have all three measures of leverage. The
adjusted R 2 ’s are .27, .23, and .21, suggesting that the impact of the US credit crisis
on other countries was related to the pre-existing correlation between their banking
sectors.
The fact that this is an incomplete explanation is not surprising. Even in a domestic
context in normal times this regression would not explain a large part of the crosssectional
dispersion of returns. 9 Moreover, if contagion depends on the particular
structure of interbank linkages, then the pattern of comovement during a crisis period
may differ from that in the pre-crisis period. We, therefore, tested whether the
correlations with the US bank index are stable between the two periods. In the case of
9 countries, or 18% of the sample, we can reject at the 5% level the hypothesis of no
change in the underlying correlations.
Figure 1 plots the normalized return against the correlation for the country sample. In
the raw data the average return on the bank indexes for emerging markets was higher
than for advanced economies. This difference is reduced, but not eliminated, when
we allow for the differing correlations with the US index. However, we show later
that it almost entirely disappears once we allow for differences in the structure of the
countries’ banking systems.
4.3 Bank capital and crisis propagation
Many regulatory recommendations focus on the role of bank capital in preventing
banking crises. Therefore, we examine the ability of bank capital measures to explain
the cross-country impact. We include the bank capital measure in a regression of the
form:
r = a + b ρ + b (1 − capitalmeasure) + u
(9)
j j j
C
1 2
The term (1 − capitalmeasure)
converts the capital measure to a measure of leverage
of the banking system, rather than its soundness.
As in the case of our country regressions we examine both the ratio of book capital
including subordinated and hybrid debt to total assets and the Basel capital adequacy
ratio. In addition, in the case of the individual banks we measure the ratio of book
equity capital (common stock plus preferred) to total assets.
9 The regression is similar to a cross-sectional regression of ex post returns on betas. The independent
variable is measured with error, which can lead to an errors-in-variables problem. This is the reason
that it is common to form portfolios of stocks before using betas in tests of other cross-sectional
relationships. In the case of the country-level regressions our correlation measure is based on a
portfolio rather than individual stocks, so our test should suffer less from errors-in-variables bias than
tests that use individual stocks.
42

The Basel measure of leverage uses risk-weighted assets as the denominator and
therefore captures the asset characteristics of the different banks. It seeks to take
account of the relative riskiness of the assets of the different banks, and in principle
should be a more accurate measure of the ability of the banks to withstand shocks.
Table 6 shows the results of regressing the standardized returns on both the
correlation with the index and the capital measure for three samples: countries,
individual banks, and those individual banks for which we have all three leverage
measures. The last of these provides a horse race between the three measures using an
identical sample.
All the leverage variables in Table 6 have the expected sign. The horse race in Panel
6C suggests that in the case of the individual banks, the equity ratio is somewhat more
informative, and, given the much larger sample size provided by this measure, we
employ it in subsequent regressions.
Panel 6C also shows, using a common sample, that the complex risk-weighting in the
asset calculation of the Basel II measure does not offer any improvement over the
other ratios in explaining the effect of the crisis on bank returns. This result is robust
in tests that include other variables in the regression. The finding lends force to
criticisms of the ability of the Basel measure to indicate the exposure of a country’s
banking system. Since the Basel measure must capture some of the risk
characteristics of different assets, its poor performance in explaining crisis exposure
suggests that banks have been adept at taking advantage of its loopholes, making it
less informative as a measure of crisis risk than a simple equity ratio.
A more extreme hypothesis has been suggested in IMF (2008). 10 This suggests that
banks which made aggressive use of the Basel rules were punished by the capital
markets in the crisis. The final column in Panels A and B tests this hypothesis. It
includes the difference between the Basel ratio and the simple leverage measure, as
well as the leverage measure itself. If banks were punished for making aggressive use
of the Basel rules, we expect this coefficient to be significantly negative, but it is not.
Thus we conclude that the Basel risk adjustments were not informative regarding
crisis risk, rather than that banks were actively punished for taking advantage of these
rules.
5. Country-level results
In this section we describe our country-level results with the expanded regression, and
then we look in the next section at how well these results are confirmed by the sample
of individual banks.
5.1 Banking sector fragility, balance sheet measures of international linkages,
and the importance of banking in the economy
We test whether adding other measures to the country-level regression improves our
ability to understand the international impact of the crisis. We include the fragility of
the banking system, measured by the proportions of bank liabilities financed with
10 See the discussion of Figure 1.17.
43

demand deposits and time deposits. Although the correlation measure explains part of
the international propagation of the crisis, it is possible that it does not capture all
dimensions of the international exposure of a country’s banking sector. So we also
include the ratio of total bank assets to GDP, as a measure of the importance of
banking in the economy. Finally, we include the proportion of foreign assets in the
aggregate balance sheet. Table 7 shows the results of including these variables in an
additive regression using the capital ratio as the measure of leverage. All the
variables, except ASSETS/GDP, are scaled to be in the range [0,1] and all are
constructed to have an expected coefficient less than zero.
In the first two columns of Table 7 all the coefficients are negative as predicted, and
the equation with all the added variables (column 2) explains 57% of the crosssectional
variation in the normalised returns. The most significant variables are the
prior correlation with the US, the capital variable, the level of time deposits, and the
relative size of the banking sector. By contrast, the demand deposit variable and the
proportion of foreign assets play little role. With the inclusion of the additional
explanatory variables the regional patterns in the residuals largely disappear. 11 Thus
the difference between the performance of banks in advanced and emerging
economies can be almost entirely explained by simple measures of their liability
structure and their importance in the economy.
The second column in Table 7 incorporates all those variables that we hypothesised
would be related to subsequent returns. For ease of reference we call it our principal
country-level regression.
The prior correlation with US bank stocks serves as a proxy for bank linkages that are
not captured by our other independent variables. In the third column of Table 7 we
omit this variable to see whether the remaining independent variables are able to take
up the slack. The removal of the correlation variable provides little help for our other
measures of bank linkages, namely the relative size of the banking sector and the level
2
of foreign claims. The adjusted R of this reduced regression is lower, but remains
respectable at .51.
5.2 The role of derivatives
The data available to measure the derivatives’ usage by different countries’ banking
sectors are of lower quality than the data for our other variables. BIS data on
aggregate derivatives usage measure only the market value of positions with positive
value for each of the countries in our sample. This could greatly underestimate the
exposure arising from derivative positions, which is likely to depend more on the
gross face value of positions both long and short, rather than on the net market value
of long positions.
We use instead data that measure the total amount of counterparty derivative
exposure, as defined by BIS, for the combined long positions held by banks in 24
11 The average residual standardized return is -.03 for advanced economies, and +.03 for emerging and
developing economies. The equivalent values for the standardized returns themselves were -.54 and
-.16.
44

major countries. The BIS reports how much of these aggregate derivative positions
are held against counterparties in each country. 12 We use these total amounts and
deflate them by the aggregate bank assets for each country.
The fourth column of Table 7 shows the result when the derivatives variable is
included. Although the variable has the correct sign in the bivariate regression, it no
longer does so when the other variables are included. The coefficient is insignificant
2
and the R is unchanged. Hence with the available data we are unable to find an
influence of derivative usage on the impact of the crisis. It may well be, however, that
such a relationship does exist and is masked by the poor quality of the available data
for this purpose.
5.3 Other Variables
The final two columns of Table 7 introduce the four additional variables that have
been suggested as associated with differential bank returns. The simple correlations
reported in Table 5 suggested that prior GDP growth was significantly and positively
associated with subsequent crisis returns. In the multiple regression this relationship
almost entirely disappears. Thus, once we allow for country differences in
correlations with US bank returns and balance-sheet structure, GDP growth and prior
stock returns have little to contribute.
In Table 7 the coefficients on the two governance variables have the opposite sign and
neither is significant. 13 Again, it remains possible that returns are truly related to
specific characteristics of corporate governance that are not picked up by our general
indexes. However, given the lack of any strong priors as to which characteristics
could matter, any exploration of this possibility would involve a substantial risk of
data mining.
6. Individual bank results
We now examine whether similar relationships hold at the level of individual banks.
The main results are contained in Table 8. 14
6.1 Measures of the comovement between individual bank returns and the US
bank index
The first column of Table 8 shows the relationship between the standardized return
during the crisis and the prior correlation with the US banking index. Despite the
significant measurement error in the independent variable, the coefficient is highly
significant and the variable explains almost a quarter of the variance in subsequent
returns. The ability of past correlation estimates to explain subsequent returns may be
reduced by possible instability in the correlations between the pre-crisis and crisis
periods. Therefore, we again test the hypothesis of no change in the underlying
12 This is reported in Table 9B of BIS, International Financial Statistics.
13 Since the two governance measures are quite highly correlated, we also added them separately into
the regression. The coefficients continued to have different signs and to be insignificant.
14 Given the varying sample sizes, one should be cautious when drawing comparisons between different
columns in Table 8.
45

correlations and reject the hypothesis for 16% of our sample at the 5% significance
level.
6.2 Regression of returns on the correlations and balance-sheet variables
The remaining columns of Table 8 summarize the results from progressively
introducing the balance-sheet variables. Since some items of data are available for
only a subset of banks, the samples vary between regressions. To facilitate
2
comparison between regressions, the final row in the table shows the R when
regression (3) is rerun using the same subset of data.
The second column of the table includes the ratio of equity to assets, the two deposit
variables and the two country-level variables. All the coefficients have the predicted
sign and all except the foreign claims measure are significant at the 5% level or better.
The coefficients on the two deposit variables are broadly similar, and this suggests
that we can usefully increase the sample size by replacing them with total deposits.
2
Regression (3) shows that, when we do this, the R increases to .46. All the
coefficients continue to have the predicted sign and all except foreign claims are
significant at the 1% level. Regression (3) corresponds most closely to our principal
country-level regression. For ease of reference, we term it the principal individualbank
regression.
As in the case of the country analysis, the standardized returns are considerably higher
for banks in emerging countries. However, almost all this difference is explained by
our independent variables. The relatively strong performance of emerging-country
banks is the result of their relative independence from the US banking market and
their more robust financial structure. 15
In line with our country regressions, we test whether omission of the correlation
variable results in more emphasis being placed on the remaining independent
variables. Column (4) of Table 8 shows that, with the exception of the foreign claims
variable, the coefficients are somewhat larger in magnitude and more significant.
2
However, these effects are relatively modest. The adjusted R of this reduced
regression is .40.
The simple correlation between standardized return and the short-term debt ratio was
strongly negative, suggesting that this variable should add to the explanatory power of
the equation. However, for most banks deposits and short-term debt together
constitute a high fraction of a bank’s funding. Including both variables would come
close to over-identifying the regression. Therefore, in column (5) of Table 8 we
substitute short-term debt for total deposits. The coefficient on short-term debt is
negative as predicted and strongly significant, but the coefficient on the equity ratio is
2
no longer significant and the R is reduced.
15 The average residual standardized return is -.01 for advanced economies, and +.02 for emerging and
developing economies. The equivalent values for the standardized returns themselves were -.28 and -
.11.
The average residual for advanced economies is -.01, and that for emerging and developing economies
is .02, which is insignificant.
46

The last three columns of Table 8 repeat our principal regression with the addition of
the remaining balance-sheet variables and the three measures suggested by previous
empirical research. Missing data are a problem in these regressions and therefore we
introduce the variables to maintain as far as possible the sample size. Regression (6)
includes the two interbank measures, the prior stock return, and the growth in GDP.
Only the last is significant, but it is quite highly correlated with the measure of bank
claims as a proportion of GDP, which now ceases to be significant.
Regression (7) omits the two interbank variables, but includes instead the level of
mortgages. The coefficient on mortgage loans is significant but, contrary to
predictions, positive.
The final column of Table 8 shows the results of adding to our principal regression the
Corporate Governance Quotient (CGQ®). The coefficients on the correlation with
the US banking index and on the balance-sheet variables remain negative and for the
most part significantly so. In contrast to the Beltratti and Stulz (2009) study, the
2
coefficient on the governance variable is positive though not significant. The R is
increased at .55. However, this is simply due to the changed sample. The final row
2
shows that an almost identical R is obtained when our principal regression is re-run
using the same sample.
7. A Measure of Extreme Comovement
If the comovement between banks differs during periods of turbulence, then simple
measures of correlation during normal periods may not be the best predictor of
comovement during the crisis. Acharya, Pedersen, Philippon, and Richardson (2010)
propose a measure equal to the return on a bank stock in the worst 5% of weeks for
the index return during the pre-crisis period. 16 Therefore, we check whether this stock
market measure of exposure to the crisis performs better than the correlation.
Translated to the current context, we measure this as the average standardized bank
return in the 5% of weeks that the US bank stock index gave the worst returns in the
period prior to the crisis (the “bad-weeks return”). Panel A of Table 9 compares the
effect of using this variable instead of the Pearson correlation coefficient in the
country regression. Columns 1 and 2 show the results for a simple regression of the
standardized return on the measures of comovement, Columns 3 and 4 incorporate
additional explanatory variables, while Column 5 includes both measures of
comovement in the one regression. Panel B provides a similar set of comparisons for
individual banks. 17
Regardless of whether we use country-level or individual-bank data, the simple
regression of standardized return on the bad-weeks return gives a lower adjusted R 2 ,
indicating that this variable captures less information relevant to the international
transmission of the crisis than the simple correlation measure. Columns 3 and 4 in
each panel shows that this relatively poor performance carries over when we add other
variables to the analysis, and Column 5 shows that tr of the correlation coefficient
16
For similar studies that have focused on extreme values to measure contagion, see Bae et al (2003)
and Gropp and Moerman (2003).
17
Note that a high value for the bad-weeks-return variable implies a low correlation with the US index.
47

continues to have greater explanatory power when both variables are included in the
regression..
The relatively poor performance of the bad weeks’ return variable differs from the
result that Acharya et al find in a test of the domestic US impact across different
financial institutions. 18 Their variable is derived from the worst days for the US bank
stock index in the period June 2006 to June 2007. We use weekly rather than daily
data because time-differences between stock exchanges make daily data unreliable in
international studies. In our context, the failure of the bad-weeks variable to predict
the cross-sectional impact of the crisis indicates that the bad weeks that happened
during generally good times did not contain useful information about the behaviour in
a crisis. So there must have been a difference between the international linkages that
operated during those “bad weeks in good times” and those that operated during the
crisis. The success of this variable in a domestic US context compared with its failure
internationally illustrates the potential danger of extrapolating US results to the
international context.
8. Robustness
The coefficients on our principal variables uniformly have the predicted sign and are
generally highly significant. The exception is the measure of the relative importance
of foreign claims, but even in this case the coefficient is consistently negative and for
the most part hovers on the borders of significance. The size of each coefficient
generally varies little with changes in model specification and sample size.
We perform two robustness checks. The first is to test for thin-trading bias and the
second is to check the sensitivity of our model to variations in the period over which
the returns are measured.
8.1 Thin trading
For some of our country indexes thin trading may have biased our estimates of the
prior standard deviation and the correlation with the US bank index. To test for the
possible effect of thin trading, we repeated our principal regressions using bi-weekly
2
returns to estimate the correlation. In the country regression, the R improved from
.57 to .60. All the coefficients had the predicted sign and all except the coefficients
on foreign assets and demand deposits remained significant at the 5% level or better.
Thin trading is equally a potential problem in our individual bank sample, where some
of the banks have a large majority shareholder. As a result, the free float is small and
the shares suffer from thin trading. As in the case of the country-level regressions, we
repeated our analysis using bi-weekly returns. The results were little changed. For the
main regression, all the coefficients were negative and all except the coefficient on
foreign assets were significant at the 5% level or better. The R 2 was reduced from .38
to .34. We also assessed the potential thin-trading bias by omitting the 11 banks
where there were (arbitrarily) 30 or more weekly returns of zero. The results were
almost identical.
8.2 Bank Returns during the rebound
48

By October 2010 the US banking index had rebounded by two-thirds, though it was
still nearly 60% below its 2007 high. We examined how far our principal variables
could explain the variation in country banking returns during the entire period May
2007 to October 2010 that included both the slump and partial rebound. The result for
the country variables is shown in Table 10.
This extension to our forecasting period places much higher demands on our data.
For example, by 2010 many large banks had been nationalized or acquired, so that the
components of the Datastream index were substantially different from four years
earlier. The regression now explains about one third of the cross-sectional dispersion
in bank index returns. All the coefficients have the predicted sign and the timedeposit
and bank-capital variables are significant at the 1% level. It is possible that the
decline in the performance of the regression is due to measurement problems or it
could be that once the crisis was over the variables which predict impact at the height
of the crisis had become less important. We leave this as an issue for further study.
9. Conclusions
We have shown that the cross-sectional incidence of the crisis was related to:
• The pre-existing correlation of the banking sector with the US;
• The equity ratio measured relative to an unadjusted balance sheet;
• The fragility of financing as measured primarily by the proportion of assets
funded by deposits;
• Banking assets as a proportion of GDP
These results were strongly significant and robust to changes in sample and model
specification. In addition, there was consistent but less significant evidence that bank
exposure to the crisis was related to the proportion of foreign claims.
We have shown that the significant leverage ratio was that measured relative to the
unadjusted balance sheet rather than Basel risk-weighted assets, but that banks were
not penalized for taking advantage of Basel rules. We have also shown that the most
informative measure of exposure derived from past returns was the correlation, not the
“bad weeks return” variable. We find that both stock market and structural variables
should be combined to give a more complete specification of the relationships which
caused differential crisis impacts. Omitting either could lead to misidentification of
the causes. Our results are robust to using individual bank and country index data. The
framework is linear in the characteristics of individual banks and therefore gives an
aggregate measure of systemic risk which is consistent with the measures for
individual banks. However, we find that results for some other variables are sensitive
to data availability and sample selection. We detect no evidence that crisis impact was
related to the quality of governance, or the prior share return. There is some evidence
of a connection with the prior growth in GDP, though this may well be a result of
multicollinearity, in particular the association between GDP growth and the relative
importance of the banking sector.
49

Our results are economically significant. This is most simply illustrated by the
bivariate analysis in Table 10, which shows the effect of our principal explanatory
variables on the returns of individual banks. The table groups the banks into quartiles
based on the magnitude of each variable and shows the mean weekly bank return for
each quartile. With one modest exception the average decline in the value of the first
quartile banks (those with the lowest ratios) is less than half that of banks in the fourth
quartile.
Our results illustrate the value of using the international impact of the crisis as a
natural experiment to test the robustness of empirical results found using US data.
Since recent international policy recommendations such as those in Acharya, Cooley,
Richardson, and Walter (2010), Financial Economists Roundtable (2010), Kane
(2010), Squam Lake Group (2010), and Scott (2010) are based either explicitly or
implicitly on assumptions about empirical relationships it is important that their
empirical foundations be robust to this type of analysis.
We also show the importance of deriving the empirical test of the propagation
mechanism from an understanding of the specific mechanism that operated. Our
results are not intended to be a model of the propagation of all international financial
crises. We derived the test from knowledge that the crisis was primarily a bankingsector
crisis which then spread to the remainder of the economy and that the crisis
originated in the US, so its propagation should depend on linkages with the US.
Future crises will not necessarily share these characteristics. However, regressions
that omitted the correlation variable continued to have strong explanatory power. 19
The implications of our results for controlling future risks depend on which features
of the recent crisis are likely to operate in a similar way in future crises. Policymakers
can influence four of the variables which we find to be important, -- capital ratios,
how banks are financed, international transactions between banks, and to a lesser
extent the size of the banking sector. They do not have direct control over the
correlation with other banking sectors, but they do have influence over the
international linkages between banks.
Our results show that the important balance sheet variables to regulate in order to
protect a country's banking system are the amount of the banking sector that is
financed with liabilities other than capital and deposits. This is potentially much
simpler than the Basel approach. We do not find that the refinement and
sophistication of the Basel risk-adjusted ratio helps to explain the cross-country
impact of the crisis in our test.
Since it was the differential impact of the banking crisis in different countries that led
to broader differences in their economic performance during the crisis, our findings
could be extended to attempt to measure the impact on other economic variables, such
as GDP. To do that it would be necessary to embed our model of banking sector
linkages in an extended model which includes the linkage between the banking sector
and aggregate economic activity.
19 The correct specification of market index is likely to be important. We reran our principal
regression for the individual banks using the correlation with the world stock market index rather than
the US banking index. The results were very similar to the regression with no correlation measure at
all.
50

References
Acharya, Viral V, Thomas F. Cooley, Matthew P. Richardson, Ingo Walter,
2010, Regulating Wall Street: The Dodd-Frank Act and the New Architecture of
Global Finance, New York University Stern School of Business.
Acharya, Viral V, Lasse H Pedersen, Thomas Philippon, and Matthew
Richardson, 2010, “Measuring Systemic Risk,” working paper, New York University
Stern School.
Allen Franklin, and Douglas Gale, D (2000), “Financial Contagion.” Journal
of Political Economy, 108(1), 1–33.
Allen Franklin, and Douglas Gale, D (2007), Understanding Financial Crises,
Oxford University Press.
Allen, Franklin, Aneta Hryckiewicz, Oskar Kowalewski, and Günseli Tümer-
Alkan, “Transmission of Bank Liquidity Shocks in Loan and Deposit Markets: The
Role of Interbank Borrowing and Market Monitoring,” University of Amsterdam,
November 2010
Bae, Kee-Hong, Karolyi, G. Andrew, René M. Stulz, R., 2003.” A New
Approach to Measuring Financial Contagion,” Review of Financial Studies 16, 717-
763.
Beltratti, Andrea, and René M. Stulz , 2009, “Why Did Some Banks Perform
Better during the Credit Crisis? A Cross-Country Study of the Impact of Governance
and Regulation,” ECGI Working Paper Series in Finance, Working Paper N°.
254/2009, July.
Bennett, Paul, and Jeanette Kelleher., 1988, “The international transmission of
stock price disruption in October 1987.” Quarterly Review of the Federal Reserve
Bank of New York, Summer, 17-33.
Calvo, Guillermo, 2005, Emerging Capital Markets in Turmoil: Bad Luck or
Bad Policy, Cambridge, MA, MIT Press.
Corsetti, Giancarlo, Marcello Pericoli, and Massimo Sbracia, , 2002, “Some
Contagion, Some Interdependence: More Pitfalls in Tests of Financial Contagion,”
CEPR Discussion Paper 3310.
Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei
Shleifer, 2008, “The Law and Economics of Self-dealing,” (December 2005).
Available at SSRN: http://ssrn.com/abstract=864645
Djankov, Simeon, Rafael La Porta, Florencio Lopez-de-Silanes, and Andrei
Shleifer, 2008, “The Law and Economics of Self-dealing,” Journal of Financial
Economics, 88 (3), 430-465.
Financial Economists Roundtable, 2010, “Reforming the OTC Derivatives
Market,” Journal of Applied Corporate Finance 22.3, 40-47.
Forbes, Kristin and Roberto Rigobon, 2002, “No Contagion, Only
Interdependence: Measuring Stock Market Comovements,” Journal of Finance 57 (5),
2223-2261.
Frankel, Jeffrey A. and George Saravelos, “Are Leading Indicators of
Financial Crises Useful for Assessing Country Vulnerability? Evidence from the
2008-09 Global Crisis,” NBER Working Paper No. 16047, June 2010.
Gropp Reint, and Gerard Moerman, 2003, “Measurement Of Contagion In
Banks´ Equity Prices,” European Central Bank, Working Paper No. 297, December.
Herring, Richard J., and Susan Wachter, 1999, “Real Estate Booms and
Banking Busts: An International Perspective,” Group of Thirty Occasional Papers.
IMF, 2008, Global Financial Stability Report, April.
51

Kane, Edward, 2010, “The Importance of Monitoring and Mitigating the
Safety-Net Consequences of Regulation-Induced Innovation,” working paper, Boston
College.
King, Mervyn and Sushil Wadhwani, 1990, “Transmission of Volatility
Between Stock Markets,” Review of Financial Studies, 3, 5-33
Kirkpatrick, Grant, 2008, “The Corporate Governance Lessons from the
Financial Crisis,” OECD report, Paris.
La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert
Vishny, 1998, “Law and Finance,” Journal of Political Economy 106, 1113-1155.
Lane, Philip R, and Gian Maria Milesi-Ferretti, 2010, The cross-country
incidence of the global crisis, forthcoming IMF Economic Review.
Navaretti, Giorgio B. Giacomo Calzolari, Alberto F.. Pozzolo, and Micol.
Levi, 2010, “Multinational Banking in Europe: Financial Stability and Regulatory
Implications Lessons from the Financial Crisis,” Centro Studi Luca d’Agliano
Working Paper No 292.
Popov, Alexander A., and Gregory. F. Udell, 2010, “Cross-border Banking
and the International Transmission of Financial Distress During the Crisis of 2007-
2008,” Working Paper Series 1203, European Central Bank
Reinhart, C. and K. Rogoff, 2008, “Is the 2007 U.S. Subprime Crisis So
Different? An International Historical Comparison”, American Economic Review 98,
339–344.
Rose, Andrew K, and Mark M Spiegel, 2009, “Cross-country Causes and
Consequences of the 2008 crisis: Early Warning,” working paper, Italian Ministry of
Economy and Finance.
Rose, Andrew K, and Mark M Spiegel, 2010, “Cross-country Causes and
Consequences of the 2008 crisis: International Linkages and American Exposure,”
Pacific Economic Review, forthcoming.
Squam Lake Group, 2010, The Squam Lake Report: Fixing the Financial
System, Princeton: Princeton University Press.
Scott, Kenneth E, 2010, “The Financial Crisis: Causes and Lessons,” Journal
of Applied Corporate Finance 22.3, 22-29.
Wolf, Holger C., 2000, “Regional Contagion Effects,” Unpublished working
paper. George Washington University, Washington, DC.
Yuan, Kathy, 2005, “Asymmetric Price Movements and Borrowing
Constraints: A Rational Expectations Equilibrium Model of Crisis, Contagion, and
Confusion,” Journal of Finance 60, 379-411.
52

Figure 1: The relationship between crisis returns and prior correlation with the
US banking sector
The figure shows the relationship between the normalised return during the crisis and the prior
correlation with the US banking sector for the country sample. Raw crisis returns are average
percentage weekly returns in the period May 2007 to March 2009. Standard deviations (percent per
week) are calculated using weekly data for the calendar years 2005-2006. The standardized return is
calculated as the ratio of the raw crisis return to the standard deviation. Correlation is the correlation of
the bank industry index with the bank industry index for the US using weekly data from January 2005
to December 2006.
53

Table 1
The country sample
The table shows the country sample. Raw crisis returns are average percentage weekly returns in the
period May 2007 to March 2009. Standard deviations (percent per week) are calculated using weekly
data for the calendar years 2005-2006. The standardized return is calculated as the ratio of the raw
crisis return to the standard deviation. All returns are for Datastream country banking sector stock
indices.
Country Mean Std. dev. Standar- Country Mean Std.dev. Standar-
crisis
return
2005-6
dized
return
crisis
return
2005-6
dized
return
Argentina -1.19 2.96 -0.403 Korea -0.91 3.41 -0.268
Australia -0.62 1.56 -0.396 Malaysia -0.39 1.49 -0.263
Austria -1.39 3.16 -0.441 Malta -0.61 3.31 -0.183
Belgium -2.27 2.09 -1.088 Mexico -0.27 3.52 -0.076
Brazil -0.18 3.56 -0.050 Netherlands -1.32 2.13 -0.620
Bulgaria -1.92 5.77 -0.333 Norway -0.95 2.78 -0.341
Canada -0.59 1.45 -0.410 Pakistan -0.99 5.33 -0.186
Chile -0.13 1.83 -0.071 Peru -0.49 3.90 -0.127
China 0.03 3.36 0.010 Philippines -0.64 2.77 -0.231
Colombia -0.11 4.73 -0.024 Poland -0.95 3.46 -0.273
Cyprus -1.65 3.60 -0.460 Portugal -1.28 1.59 -0.804
Czech -0.55 4.20 -0.130 Romania -1.56 5.94 -0.262
Denmark -1.65 1.97 -0.833 Russia -1.40 5.50 -0.254
Finland -0.86 3.11 -0.276 Singapore -0.86 1.95 -0.440
France -1.49 2.14 -0.697 Slovenia -1.05 2.18 -0.483
Germany -1.62 2.00 -0.811 S Africa -0.46 3.99 -0.116
Greece -1.49 3.17 -0.470 Spain -1.01 1.84 -0.547
Hong Kong -1.00 1.43 -0.698 Sri Lanka -0.34 2.90 -0.117
Hungary -1.54 5.56 -0.278 Sweden -1.01 2.47 -0.409
India -0.29 4.22 -0.068 Switzerland -1.28 2.29 -0.560
Indonesia -0.16 3.57 -0.043 Taiwan -0.28 3.36 -0.083
Ireland -2.60 2.31 -1.122 Thailand -0.37 3.36 -0.111
Israel -0.87 3.02 -0.289 Turkey -0.46 4.87 -0.095
Italy -1.44 1.71 -0.838 UK -1.36 1.66 -0.821
Japan -0.85 3.57 -0.237 Venezuela -0.35 2.95 -0.120
54

Table 2
Summary statistics for the country sample
The table shows summary statistics for the country sample. All returns are for Datastream country
banking sector stock indices. Raw crisis returns are average percentage weekly returns in the period
May 2007 to March 2009. Standard deviations (percent per week) are calculated using weekly data for
the calendar years 2005-2006. The standardized return is calculated as the ratio of the raw crisis return
to the standard deviation. Correlation is the correlation of the bank industry index with the US bank
industry index using weekly data from January 2005 to December 2006. Bad weeks return is the return
of the country index in the 5% of weeks in 2005-2006 during which the US bank index had the worst
returns. This is standardized by the standard deviation of returns. Equity ratio is (1-Book
Value(Equity))/Total Assets). Capital ratio is (1 - Book Value(Equity + Sub debt))/Total Assets).
Basel ratio is (1-Basel risk-weighted capital ratio). Demand deposits is (1-Demand deposits/Total
assets). Time deposits is (1-Time deposits/Total assets). Total deposits is (1-Total deposits/Total
assets). Bank claims/GDP is the ratio of total bank assets to GDP. Foreign claims is the ratio of total
foreign claims held by banks to total bank assets. Derivatives is the ratio of value of derivative
positions to total bank assets. Anti-director rights and self-dealing are Djankov et al’s indices. Return in
2006 is the average weekly return in 2006. GDP growth is the rate of growth in percent over the period
2001-2006. Unless otherwise indicated all independent variables are measured at the end of 2006. Data
sources are given in the Appendix.
Variable Number
of obs.
Mean Std. Dev. Median Maximum Minimum
Raw crisis return 50 -0.366 0.285 -0.278 0.010 -1.127
Standardized crisis return 50 -0.009 0.006 -0.009 0.000 -0.026
Correlation 50 0.269 0.163 0.273 0.572 -0.103
Standardized bad weeks
return 50 -0.403 0.427 -0.424 0.583 -1.187
1 - Equity ratio 45 0.898 0.046 0.902 0.999 0.778
1 - Capital ratio 49 0.921 0.028 0.927 0.970 0.850
1 - Basel ratio 50 0.866 0.031 0.874 0.951 0.780
1 - Demand deposits 48 0.814 0.118 0.847 0.979 0.335
1 - Time deposits 48 0.646 0.174 0.652 0.961 0.254
1 - Total deposits 48 0.460 0.180 0.434 0.858 0.039
Bank claims/GDP 48 1.824 1.494 1.407 6.828 0.362
Foreign claims 48 0.373 0.175 0.335 0.777 0.005
Derivatives 47 0.017 0.013 0.012 0.052 0.000
Antidirector rights 47 0.347 0.113 0.350 0.500 0.000
Self-dealing 47 0.501 0.232 0.460 1.000 0.090
Return in 2006 50 0.006 0.004 0.005 0.018 -0.002
GDP growth 50 9.495 6.845 7.837 34.678 0.385
55

Table 3
Correlation Matrix for the country sample
The table provides the matrix of Pearson correlation coefficients for our country sample. Variables are as defined in Table 2.
1 2 3 4 5 6 7 8 9 10
Standardized crisis return 1 1.00 0.83 -0.54 0.36 -0.41 -0.45 -0.26 -0.21 -0.42 -0.54
Raw crisis return 2 0.83 1.00 -0.31 0.27 -0.39 -0.37 -0.16 -0.14 -0.48 -0.56
Correlation 3 -0.54 -0.31 1.00 -0.55 0.36 0.40 0.22 0.35 0.11 0.34
Standardized bad weeks return 4 0.36 0.27 -0.55 1.00 -0.29 -0.17 -0.03 -0.26 0.02 -0.15
1 - Equity ratio 5 -0.41 -0.39 0.36 -0.29 1.00 0.57 0.38 0.05 0.16 0.18
1 – Capital ratio 6 -0.45 -0.37 0.40 -0.17 0.57 1.00 0.69 -0.09 0.13 0.07
1 - Basel ratio 7 -0.26 -0.16 0.22 -0.03 0.38 0.69 1.00 -0.21 0.10 -0.04
1 - Demand deposits 8 -0.21 -0.14 0.35 -0.26 0.05 -0.09 -0.21 1.00 -0.29 0.38
1 -Time deposits 9 -0.42 -0.48 0.11 0.02 0.16 0.13 0.10 -0.29 1.00 0.78
1 - Total deposits 10 -0.54 -0.56 0.34 -0.15 0.18 0.07 -0.04 0.38 0.78 1.00
Bank claims/GDP 11 -0.57 -0.44 0.22 0.01 0.49 0.21 0.07 0.37 0.28 0.51
Foreign claims 12 -0.29 -0.41 -0.15 0.20 0.07 -0.16 -0.18 0.12 0.31 0.38
Derivatives 13 -0.39 -0.24 0.42 -0.18 0.31 0.37 0.16 0.18 0.27 0.39
Anti-director rights 14 0.02 0.07 0.03 0.20 0.01 -0.11 -0.19 0.44 -0.27 0.03
Self-dealing 15 -0.07 0.11 -0.07 0.09 0.09 -0.01 0.11 0.33 -0.42 -0.19
Return in 2006 16 0.18 -0.04 -0.21 0.30 -0.06 -0.21 0.00 -0.24 0.06 -0.10
GDP growth 17 0.49 0.30 -0.41 0.22 -0.29 -0.45 -0.36 -0.33 -0.08 -0.29
Note: Sample sizes may differ among cells
56
25

Table 3 (Continued)
Correlation Matrix
The table provides the matrix of Pearson correlation coefficients for our country sample Variables are as defined in Table 2.
11 12 13 14 15 16 17
Standardized crisis return 1 -0.57 -0.29 -0.39 -0.02 0.07 0.18 0.49
Raw crisis return 2 -0.44 -0.41 -0.24 -0.07 -0.11 -0.04 0.30
Correlation 3 0.22 -0.15 0.42 -0.03 0.07 -0.21 -0.41
Standardized bad weeks return 4 0.01 0.20 -0.18 0.20 0.09 0.30 0.22
Equity ratio 5 0.49 0.07 0.31 -0.01 -0.09 -0.06 -0.29
Capital ratio 6 0.21 -0.16 0.37 0.11 0.01 -0.21 -0.45
Basel ratio 7 0.07 -0.18 0.16 0.19 -0.11 0.00 -0.36
Demand deposits 8 0.37 0.12 0.18 -0.44 -0.33 -0.24 -0.33
Time deposits 9 0.28 0.31 0.27 0.27 0.42 0.06 -0.08
Total deposits 10 0.51 0.38 0.39 -0.03 0.19 -0.10 -0.29
Bank claims/GDP 11 1.00 0.55 0.29 -0.15 -0.30 -0.15 -0.50
Foreign claims 12 0.55 1.00 0.18 -0.02 -0.11 0.29 0.01
Derivatives 13 0.29 0.18 1.00 -0.13 -0.14 -0.20 -0.41
Anti-director rights 14 0.15 0.02 0.13 1.00 0.54 -0.26 -0.16
Self-dealing 15 0.30 0.11 0.14 0.54 1.00 -0.20 -0.12
Return in 2006 16 -0.15 0.29 -0.20 0.26 0.20 1.00 0.48
GDP Growth 17 -0.50 0.01 -0.41 0.16 0.12 0.48 1.00
57
26

Table 4
Summary statistics for the individual bank sample
The table shows summary statistics for the country sample. All returns are from Datastream. Raw crisis
returns are average percentage weekly returns in the period May 2007 to March 2009. Standard
deviations (percent per week) are calculated using weekly data for the calendar years 2005-2006. The
standardized return is calculated as the ratio of the raw crisis return to the standard deviation.
Correlation is the correlation of the bank return with the return on the US bank industry index using
weekly data from January 2005 to December 2006. Bad weeks return is the return of the bank equity in
the 5% of weeks in 2005-2006 during which the US bank index had the worst returns. This is
standardized by the standard deviation of returns. Equity ratio is (1-Book Value of Equity))/Total
Assets). Capital ratio is (1-Book Value(Equity + Sub debt + Hybrid debt))/Total Assets). Basel ratio is
(1-Basel risk-weighted capital ratio). Demand deposits is (1-Demand deposits/Total assets). Time
deposits is (1-Time deposits/Total assets). Total deposits is (1-Total deposits/Total assets). Bank
claims/GDP is the ratio of total bank assets to GDP. Foreign claims is the ratio of total foreign claims
held by banks to total bank assets. Short-term debt is the ratio of short-term debt liabilities to total bank
assets. Interbank loans is the ratio of interbank assets to total bank assets. Due other banks is the ratio
of interbank funding to total bank assets. Mortgages is the ratio of real estate mortgages to total bank
assets. Governance is the CGQ index. Return in 2006 is the average weekly return in 2006. GDP
growth is the rate of growth in percent over the period 2001-2006. Unless otherwise indicated all
independent variables are measured at the end of 2006. Data sources are given in the Appendix.
Variable Number
of obs.
Mean Std. Dev. Median Maximum Minimum
Raw crisis return 381 -0.007 0.007 -0.006 0.007 -0.032
Standardized crisis return 381 -0.212 0.233 -0.156 0.271 -1.211
Correlation 381 0.160 0.136 0.163 0.580 -0.160
Standardized bad weeks
return 380 -0.978 1.948 -0.982 4.131 -5.525
1 - Equity ratio 361 0.921 0.074 0.937 0.988 0.198
1 - Capital ratio 273 0.904 0.067 0.913 0.982 0.194
1 - Basel ratio 289 0.872 0.041 0.881 0.963 0.644
1 - Demand deposits 256 0.815 0.141 0.853 1.000 0.389
1 - Time deposits 264 0.628 0.225 0.630 1.000 0.146
1 - Total deposits 358 0.365 0.213 0.335 1.000 0.059
Bank claims/GDP 368 1.921 1.272 1.784 6.815 0.361
Foreign claims 368 0.312 0.174 0.271 0.777 0.005
Short-term debt 355 0.108 0.895 0.076 0.592 0.000
Interbank loans 331 0.072 0.085 0.050 0.628 0.000
Due to other banks 322 0.093 0.079 0.080 0.550 0.000
Mortgages 234 0.175 0.193 0.102 0.815 0.000
Governance (CGQ) 118 52.040 28.956 51.365 99.950 0.790
Return in 2006 382 0.004 0.006 0.004 0.031 -0.012
GDP growth 385 8.113 7.847 6.315 34.678 0.385
58

Table 5
Correlation Matrix for the individual bank sample
The table provides the matrix of Pearson correlation coefficients for our individual bank sample. Variables are as defined in Table 4.
1 2 3 4 5 6 7 8 9 10
Standardized crisis return 1 1.00 0.90 -0.48 0.33 -0.19 -0.16 -0.15 -0.04 -0.35 -0.53
Raw crisis return 2 0.90 1.00 -0.38 0.30 -0.14 -0.12 -0.14 -0.08 -0.30 -0.48
Correlation 3 -0.48 -0.38 1.00 -0.59 0.18 0.12 0.10 0.09 0.22 0.31
1 – Standardized bad weeks 4
return
0.33 0.30 -0.59 1.00 -0.02 -0.01 0.08 -0.17 -0.11 -0.33
1 - Equity ratio 5 -0.19 -0.14 0.18 -0.02 1.00 0.96 0.63 0.00 -0.10 -0.20
1 - Capital ratio 6 -0.16 -0.12 0.12 -0.01 0.96 1.00 0.59 0.02 -0.02 -0.17
1 - Basel ratio 7 -0.15 -0.14 0.10 0.08 0.63 0.59 1.00 -0.20 0.25 -0.04
1 - Demand deposits 8 -0.04 -0.08 0.09 -0.17 0.00 0.02 -0.20 1.00 -0.40 0.17
1 - Time deposits 9 -0.35 -0.30 0.22 -0.11 -0.10 -0.02 0.25 -0.40 1.00 0.71
1 - Total deposits 10 -0.53 -0.48 0.31 -0.33 -0.20 -0.17 -0.04 0.17 0.71 1.00
Bank claims/GDP 11 -0.26 -0.16 0.16 0.08 0.29 0.22 0.23 0.03 0.07 0.02
Foreign claims 12 -0.19 -0.18 -0.07 -0.03 -0.01 -0.03 -0.14 -0.08 -0.08 0.23
Short-term debt 13 -0.45 -0.40 0.25 0.21 0.13 0.12 0.13 0.15 0.44 0.71
Interbank loans 14 -0.14 -0.11 0.01 -0.13 -0.14 -0.17 -0.19 -0.04 -0.07 0.10
Due to other banks 15 -0.09 -0.07 -0.02 -0.07 -0.18 -0.12 -0.21 -0.03 0.00 0.15
Mortgages 16 -0.16 -0.08 0.01 -0.04 0.16 0.13 0.15 -0.14 0.17 0.32
Governance (CGQ) 17 -0.09 -0.11 0.26 -0.21 0.27 0.21 0.10 0.06 0.08 0.17
Return in 2006 18 -0.05 -0.10 -0.08 0.06 -0.11 -0.11 -0.20 -0.28 -0.05 0.14
GDP growth 19 0.25 0.20 -0.22 -0.01 -0.25 -0.30 -0.30 -0.07 -0.13 0.01
Note: Sample sizes may differ among cells
59
28

Table 5 (Continued)
Correlation Matrix
The table provides the matrix of Pearson correlation coefficients for our individual bank sample. Variables are as defined in Table 4.
11 12 13 14 15 16 17 18 19
Standardized crisis return 1 -0.26 -0.19 -0.45 -0.14 -0.09 -0.16 -0.09 -0.05 0.25
Raw crisis return 2 -0.16 -0.18 -0.40 -0.11 -0.07 -0.08 -0.11 -0.10 0.20
Correlation 3 0.16 -0.07 0.25 0.01 -0.02 0.01 0.26 -0.08 -0.22
Standardized bad weeks return 4
0.08 -0.03 0.21 -0.13 -0.07 -0.04 -0.21 0.06 -0.01
1 - Equity ratio 5 0.29 -0.01 0.13 -0.14 -0.18 0.16 0.27 -0.11 -0.25
1 - Capital ratio 6 0.22 -0.03 0.12 -0.17 -0.12 0.13 0.21 -0.11 -0.30
1 - Basel ratio 7 0.23 -0.14 0.13 -0.19 -0.21 0.15 0.10 -0.20 -0.30
1 - Demand deposits 8 0.03 -0.08 0.15 -0.04 -0.03 -0.14 0.06 -0.28 -0.07
1 - Time deposits 9 0.07 -0.08 0.44 -0.07 0.00 0.17 0.08 -0.05 -0.13
1 - Total deposits 10 0.02 0.23 0.71 0.10 0.15 0.32 0.17 0.14 0.01
Bank claims/GDP 11 1.00 0.36 0.14 0.11 0.10 0.45 0.09 -0.20 -0.51
Foreign claims 12 0.36 1.00 0.21 0.43 0.35 0.39 0.07 0.44 0.26
Short-term debt 13 0.14 0.21 1.00 0.19 0.21 0.14 0.09 0.12 -0.05
Interbank loans 14 0.11 0.43 0.19 1.00 0.79 0.03 0.02 0.25 -0.02
Due other banks 15 0.10 0.35 0.21 0.79 1.00 -0.08 0.07 0.30 0.06
Mortgages 16 0.45 0.39 0.14 0.03 -0.08 1.00 0.11 0.06 -0.15
Governance 17 0.09 0.07 0.09 0.02 0.07 0.11 1.00 0.03 0.10
Return in 2006 18 -0.20 0.44 0.12 0.25 0.30 0.06 0.03 1.00 0.43
GDP growth 19 -0.51 0.26 -0.05 -0.02 0.06 -0.15 0.10 0.43 1.00
60
29

Table 6
Comparison of different leverage measures
This table presents regressions of the standardized crisis return using three different leverage measures.
Panel A is for the country bank indices, Panel B for individual banks, and Panel C for individual banks
using a common sample for all regressions. Equity ratio is the ratio of the book value of equity to bank
assets. Capital ratio is the ratio of the book value of (equity + subordinated debt) to bank assets. Basel
ratio is the Basel II risk-weighted capital ratio. Basel-Equity is the difference between the Basel and
equity ratios. All other variables are as defined in the Appendix. Standardized crisis return is the
average weekly return over the period May 2007 to March 2009 divided by the weekly standard
deviation in 2005-2006. The independent variables are measured at the end of 2006. Estimation is by
OLS. The table also reports the adjusted R-square and number of observations. T-statistics are given in
parentheses.
Panel 6A: Country regressions
Dependent variable: Standardized crisis return
Constant -.11*
(-1.69)
Correlation -.94***
(-4.41)
1 - Capital ratio
1 - Basel ratio
Basel-Capital
Ratio
Adjusted R
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
2
.27 .33 .28 .32
N 50 49 48 49
Panel 6B: Individual bank regressions
Dependent variable: Standardized crisis return
Constant -.08***
(-5.00)
Correlation -.82***
(-10.68)
1 - Equity ratio
1 - Capital ratio
1 - Basel ratio
Basel-Equity
ratio
(1) (2) (3) (4)
2.48**
(2.11)
-.74***
(-3.30)
-2.87**
(-2.20)
(1) (2) (3) (4) (5)
.22
(1.65)
-.80***
(-9.88)
-.33**
(-2.24)
1.04
(1.07)
-.88***
(-4.05)
-1.35
(-1.19)
.25
(1.45)
-.85***
(-9.16)
-.38**
(-1.97)
2.40*
(1.99)
-0.74***
(-3.23)
-2.75**
(-2.04)
-.60
(-.40)
0.44
(1.65)
-.82***
(-8.59)
-0.60**
(-1.97)
.87***
(2.81)
-.77***
(-7.97)
-1.04***
(-3.01)
-.04
(-.10)
Adjusted R 2
.23 .24 .25 .22 .23
N 381 361 273 289 287
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
61

Table 6 (continued)
Panel 6C: Individual bank regressions (common sample)
Dependent variable: Standardized crisis return
Constant -.09***
(-3.67)
Correlation -.84***
(-7.82)
1 - Equity ratio
1 - Capital ratio
1 - Basel ratio
(1) (2) (3) (4)
1.98***
(4.04)
-.70***
(-7.78)
-2.25***
(-7.03)
.61
(1.37)
-.79***
(-6.99)
-.79
(-1.58)
.47
(1.42)
-.81***
(-7.52)
-.65*
(-1.71)
Adjusted R 2
.21 .27 .22 .22
N 223 223 223 223
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
62

Table 7
Country-level determinants of crisis returns
This table presents regressions of the standardized crisis return for the country bank indices on countrylevel
variables, as defined in the Appendix. The standardized return is the average weekly return over
the period May 2007 to March 2009 divided by the weekly standard deviation in 2005-2006. All the
independent variables are measured at the end of 2006, with the exception of the correlation with the
US index, which is estimated from weekly data for 2005-2006, and the rate of GDP growth, which is
measured over the period 2001-2006. Estimation is by OLS. The table also reports the adjusted Rsquare
and number of observations. T-statistics are given in parentheses.
Dependent variable: Standardized crisis return
(1) (2) (3) (4) (5)
Constant 2.48** 2.99** 4.32*** 3.20** 1.83
(2.11) (2.54) (3.76) (2.56) (1.32)
Correlation coefficient -.74*** -.58***
-.57** -.63**
(-3.30) (-2.69)
(-2.55) (-2.32)
1 - Capital ratio
-2.87** -2.83** -4.04*** -3.12** -1.58
(-2.20) (-2.43) (-3.51) (-2.54) (-1.16)
1 - Demand deposits
-.18 -.57*
-.15 -.09
(-.57) (-1.87) (-.44) (-.26)
1 - Time deposits
-.39* -.54** -.37* -.46**
(-2.00) (-2.66) (-1.78) (-2.05)
Bank claims/GDP -.05* -.06* -.06** -.06
(-1.94) (-1.95) (-2.09) (-1.55)
Foreign claims -.27
-.14
-.29 -.31
(-1.29) (-.61) (-1.30) (-1.28)
Derivatives
1.37
(.52)
Return in 2006
.10
(.01)
GDP growth
.00
(.65)
Anti-director rights
.16
(.47)
Self-dealing -.16
(-.93)
Adjusted R 2
.33 .57 .51 .57 .60
N 49 47 47 46 45
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
63

Table 8
Firm-level determinants of crisis returns
This table presents regressions of the standardized crisis return for individual banks on firm-level
variables, as defined in Table 4. The standardized return is the average weekly return over the period
May 2007 to March 2009 divided by the weekly standard deviation in 2005-2006. All the independent
variables are measured at the end of 2006, with the exception of the correlation with the US index,
which is estimated from weekly data for 2005-2006, and the rate of GDP growth, which is measured
over the period 2001-2006. Estimation is by OLS. The table also reports the adjusted R-square and
number of observations. T-statistics are given in parentheses.
Dependent variable: Standardized crisis
return
Constant -.08***
(-5.00)
Correlation -.82***
(-10.68)
1 - Equity ratio
1 - Demand
deposits
1 - Time
deposits
1 - Total
deposits
Bank
claims/GDP
(1) (2) (3) (4) (5) (6) (7) (8)
.80***
(3.46)
-.57***
(-6.06)
-.45**
(-2.19)
-.27***
(-2.68)
-.34***
(-5.02)
-.03**
(-2.51)
.77***
(5.34)
-.49***
(-6.26)
-.70***
(-4.49)
-.49***
(-9.92)
-.02***
(-2.69)
.94***
(6.31)
-.92***
(-5.75)
-.62***
(-12.78)
-.04***
(-3.83)
.24*
(1.87)
-.66***
(-8.44)
-.20
(-1.40)
-.02**
(-2.16)
.99***
(4.32)
-.48***
(-5.35)
-.97***
(-4.00)
-.49***
(-8.57)
-.01
(-.86)
Foreign claims -.09 -.12* -.06 -.18*** -.20**
(-.90) (-1.95) (-.95) (-2.77) (-2.13)
Short-term debt -.67***
(-6.50)
Interbank loans .08
(.33)
Due other banks
Mortgages
-.05
(-.21)
.77***
(4.28)
-.45***
(-4.72)
-.79***
(-4.12)
-.56***
(-8.50)
.01
(.35)
-.17*
(-1.83)
1.53*
(1.91)
-.21
(-1.24)
-1.56*
(-1.79)
-.64***
(-6.45)
-.03
(-1.38)
-.15
(-1.07)
Return in 2006
-2.73
.15**
(2.10)
-2.32
(-1.13) (-.99)
GDP growth
.51** .59**
(2.49) (2.57)
Governance .00
(1.23)
Adjusted R 2
.23 .38 .46 .40 .39 .49 .49 .55
N 381 237 343 343 342 275 223 115
Comparable
adjusted R 2 for
regression (3)
-- .43 -- .46 .46 .49 .47 .55
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
64

Table 9 Regression of the standardized return during the crisis period on
alternative measures of comovement during the crisis
Panel A shows regressions of the standardized crisis return for the country bank indices on countrylevel
variables, as defined in the Appendix. Panel B shows similar regressions for standardized returns
for individual banks. In columns 1 and 3 the prior comovement is measured by the Pearson correlation
coefficient between bank returns and the US index, which is estimated from weekly data for 2005-
2006. In columns 2 and 4 it is measured by the average standardized return in the 5% of weeks where
the US bank stock index gave the worst returns in the period prior to the crisis (the “bad weeks
return”). Column 5 includes both measures of comovement in the same regression. The other
independent variables are measured at the end of 2006. Estimation is by OLS. The table also reports
the adjusted R-square and number of observations. T-statistics are given in parentheses.
Panel 9A: Country regressions
Dependent variable: Standardized crisis return
(1) (2) (3) (4) (5)
Constant -.11*
(-1.69)
Correlation
-.94***
coefficient
(-4.41)
Standardized bad
weeks return
-.27***
( -5.13)
.24**
(2.65)
2.99**
(2.54)
-.58***
(-2.69)
1 - Capital ratio -2.83*
(-2.43)
1 - Demand deposits -.18
(-.57)
1 - Time deposits -.39**
(-2.00)
Bank claims/GDP -.05*
(-1.94)
Foreign claims -.27
(-1.29)
3.67***
(3.31)
.19**
(2.53)
-3.52***
(-3.19)
-.29
(-.93)
-.46**
(-2.41)
-.06**
(-2.24)
-.22
(-1.04)
2.92**
(2.52)
-.42*
(-1.79)
.12
(1.59)
-2.81**
(-2.46)
-.10
(-.31)
-.38*
(-1.99)
-.06**
(-2.13)
-.29
(-1.39)
Adjusted R 2 .27 .11 .57 .57 .59
N 50 50 47 47 47
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
Panel 9B: Individual bank regressions
Dependent variable: Standardized crisis return
(1) (2) (3) (4) (5)
Constant -.08***
(5.00)
Correlation
-.82***
coefficient
(-10.68)
Standardized bad
weeks return
-.17***
(-13.76)
.04***
(6.76)
.77***
(5.34)
-.49***
(-6.26)
.88***
(5.96)
.02***
(3.78)
.77***
(5.33)
-.47***
(-4.88)
.00
(.38)
1 - Equity ratio -.70*** -.85*** -.70***
(-4.49) (-5.40) (-4.48)
1 - Total deposits -.49*** -.55*** -.49***
(-9.92) (-10.83) (-9.73)
Bank claims/GDP -.02*** -.04*** -.03***
(-2.69) (-4.13) (-2.70)*
Foreign claims -.12* -.06 -.12*
(-1.95) (-.89) (-1.88)
Adjusted R 2 .23 .11 .46 .42 .46
N 381 379 343 342 342
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
65

Table 10
Country-level determinants of crisis returns including the recovery
This table presents regressions of the standardized crisis return including recovery for the country bank
indices on country-level variables, as defined in Table 2. The standardized return including recovery is
the average weekly return over the period May 2007 to October 2010 divided by the weekly standard
deviation in 2005-2006. All the independent variables are measured at the end of 2006, with the
exception of the correlation with the US index, which is estimated from weekly data for 2005-2006.
Estimation is by OLS. The table also reports the adjusted R-square and number of observations. Tstatistics
are given in parentheses.
Dependent variable: Standardized crisis return including recovery
(1)
Constant 1.60**
(2.25)
Correlation -.01
(-.05)
Equity ratio
-1.46**
(-2.08)
Demand deposits
-.03
(-.13)
Time deposits
-.34***
(-2.86)
Bank claims/GDP -.00
(-.27)
Foreign claims -.04
(-.29)
Adjusted R 2
.27
N 47
*, **, and ***, significant at the 10, 5, and 1 percent level respectively.
Table 11
Economic significance of key measures
The table shows mean weekly returns 2007-2009 for individual banks grouped into quartiles by the
magnitude of their prior correlation with the US market and by key balance-sheet ratios. Correlation
with the US is the correlation of the bank return with the return on the bank industry index for the US
using weekly data from January 2005 to December 2006.
Correlation with (1 - equity (1 – total Short-term
US capital)/assets deposits)/assets debt/assets
Quartile 1 -.44% -.55% -.36% -.47%
Quartile 2 -.62 -.63 -.56 -.49
Quartile 3 -.64 -.62 -.66 -75
Quartile 4 -1.14 -1.02 -1.21 -1.12
66

Appendix
Variable definitions and data sources
The bank balance sheet data in IMF International Financial Statistics refer to Other Depository
Corporations (defined as resident financial corporations (except the central bank) and quasicorporations
that are mainly engaged in financial intermediation and that issue liabilities included in the
national definition of broad money).
Panel A: Country level variables
Variable Description Source of data
Raw crisis return Average percentage weekly return in the Datastream country banking
period May 2007 to March 2009 sector stock indices
Standardized crisis return Ratio of the raw crisis return to standard Datastream country banking
deviation. Standard deviations (percent
per week) calculated using weekly data
for calendar years 2005-2006
sector stock indices
Correlation Correlation of bank industry with bank Datastream country banking
industry index for the US using weekly
data from January 2005 to December
2006
sector stock indices
Bad weeks return Standardized return of the country index Datastream country banking
in the 5% of weeks in 2005-2006 during
which the US bank index had the worst
returns.
sector stock indices
Capital ratio 1-Book Value(Equity + Sub debt)/Total IMF Global Financial Stability
Assets
Report
Basel ratio 1-Basel risk-weighted capital ratio IMF Global Financial Stability
Report
Demand deposits 1-Demand deposits/Total assets IMF International Financial
Statistics, country tables
Time deposits 1-Time deposits/Total assets IMF International Financial
Statistics, country tables
Total deposits 1-Total deposits/Total assets IMF International Financial
Statistics, country tables
Bank claims/GDP Ratio of total bank assets to GDP IMF International Financial
Statistics, country tables
Foreign claims Ratio of total foreign claims held by IMF International Financial
banks to total bank assets
Statistics, country tables
Derivatives Ratio of value of derivative positions to BIS Consolidated Banking
total bank assets
Statistics, Table 9C
Anti-director rights Djankov et al’s index of self-dealing Djankov et al (2005), Table XII
Self-dealing Djankov et al’s revised index of antidirectors’
rights
Djankov et al (2005), Table III
Return in 2006 Average weekly share return in 2006 Datastream country banking
sector stock indices
GDP growth GDP growth in local currency 2001-2006 IMF International Financial
Statistics
67

Panel B: Individual bank variables
Variable Description Source of data
Raw crisis return As country variable measured for
individual bank
Datastream
Standardized crisis return As country variable measured for Datastream
Correlation
individual bank
As country variable measured for
individual bank
Datastream
Bad weeks return As country variable measured for
individual bank
Datastream
Equity ratio As country variable measured for
individual bank
Datastream
Capital ratio As country variable measured for
individual bank
Osiris
Basel ratio As country variable measured for Datastream, The Banker,
individual bank
company annual reports, and
Osiris
Demand deposits As country variable measured for
individual bank
Datastream
Time deposits As country variable measured for
individual bank
Datastream
Total deposits As country variable measured for
individual bank
Datastream
Bank claims/GDP As country variable IMF International Financial
Statistics, country tables
Foreign claims As country variable IMF International Financial
Statistics, country tables
Short-term debt Short-Term Debt/Total Assets for
individual bank
Datastream
Interbank loans Interbank Loans/Total Assets for
individual bank
Datastream
Due other banks Interbank Liabilities/Total Assets for
individual bank
Osiris
Mortgages Mortgage Loans/Total Assets for
individual bank
Datastream
Governance CGQ governance index for individual
bank
Bloomberg
Return in 2006 As country variable measured for
individual bank
Datastream
GDP growth As country variable IMF International Financial
Statistics
68

SHORT WAVELENGTH FINANCE: WHAT, WHO AND WHY
Dr. Nick Kondakis
Kepler Asset Management, 100 Wall Street, New York, NY 10005
nick@kondakis.com
Despite the noble efforts on both sides of the Atlantic we are still under the dark spell of the
Global Financial Crisis. One year since our last discussion here, we still have not solved any of
the major problems that have caused the worst post-war crisis: The bad loans are still there,
simply hiding by staying away from the mark-to-market accounting, waiting to surface again in
the wake of further declining real estate markets. The too-big-to-fail financial institutions are
growing both in size and numbers. The greatly anticipated financial services industry reform has
been heavily diluted. The derivatives market is still growing happily unregulated. And the
European credit crisis is getting markedly worse with the Greek situation deteriorating further
and more Eurozone countries taking or preparing to take the same road as Greece.
The extraordinary amounts of liquidity pumped into the financial system in order to prevent
the next Great Depression from happening again have managed to support the world stock
markets and create fresh asset bubbles in most commodity markets. In this environment, classic
long term quantitative models geared to identify value/growth opportunities have lost a lot of
their predictive ability and, in addition, the risks have grown significantly due to the expected
increase in correlation between the various asset classes during the inevitable corrections.
It is increasingly obvious that the field of quantitative investing is becoming more and more
commoditized especially in the “wavelengths” longer than a few hours. Rigorous research has
led us to identify predictable regimes in increasingly shorter scales. As the time base gets
smaller, the factors entering the quantitative prediction models move away from the
fundamentals and the long term price correlations to more technical ones like the market
microstructure and the short-term correlations.
These short wavelength regimes, mostly within minutes in an intraday session, are ideally
suited for trading strategies by smaller firms and proprietary quantitative groups especially in the
face of a possible escalation in crisis-related volatility. Their exploitation is becoming more
possible as technology is following Moore’s Law: Ultra-low latency order-based data feeds can
drive data-hungry quantitative models trading millions of shares a day with relatively small
amounts of capital. A typical day can yield a few hundred Gigabytes of exchange feed data and
the research capabilities have to be able to match these sizable samples. The processing power
available has reduced decision times well below the one microsecond barrier and exchange order
round trip times are routinely below a couple of hundred microseconds.
For sure, the new environment has created new challenges in model development, technology
management and human resources and is rapidly transforming the world of Quantitative Finance.
69

ASSET ALLOCATION-PORTFOLIO
MANAGEMENT
70

ON THE PERFORMANCE OF A HYBRID GENETIC ALGORITHM: APPLICATION ON THE
PORTFOLIO MANAGEMENT PROBLEM
Vassilios Vassiliadis, Vassiliki Bafa, George Dounias, Management and Decision Engineering Laboratory, Department of
Financial & Management Engineering, University of the Aegean, Greece
Email: v.vassiliadis@fme.aegean.gr, v.bafa@yahoo.gr g.dounias@aegean.gr , http://fidelity.fme.aegean.gr/decision/
Abstract. In this study, a hybrid intelligent scheme which combines a genetic algorithm with a numerical optimization technique is
applied to a cardinality-constrained portfolio management problem. Specifically, the objective function aims at maximizing the
Sortino Ratio with a constraint on tracking error volatility. What is more, results from the proposed algorithm are compared with other
financial and intelligent heuristics, such as financial rule-of-thumbs and simulated annealing. In order to obtain a better insight on the
hybrid’s behavior, out-of-sample results are shown. The contribution of this work is twofold. Firstly, some useful conclusions
regarding the performance of the proposed hybrid algorithm are drawn, based on experimental simulations. Secondly, some basic
points, based on the comparison between the proposed algorithm and the benchmark heuristics, are highlighted. Finally, concerning
the cardinality-constrained optimization problem, financial implications are discussed in some extent.
Keywords: Genetic Algorithm, Portfolio Optimization, financial heuristics, hybrid algorithm, evolutionary mechanisms
1. Introduction
Nowadays, the portfolio optimization problem is of crucial important, for many reasons. Firstly, there is a portion of
investors who seek good investment opportunities such as investing to a number of stocks from a certain market,
rather than investing to a single stock or the market as a whole. Investing to a single stock usually incurs high risk.
On the other hand, investing to the stock market, considering as the portfolio defined by the market, incurs high
transaction costs. Secondly, the portfolio optimization problem becomes even more complex, if multiple, and
sometimes conflicting, objectives or many real-world constraints are considered. As a result, many investors aim at
finding high-quality, near-optimum combinations of stocks, which satisfy certain objectives. In this point, it is worth
mentioning that both the concept and notion of portfolio management problem were first introduced by Markowitz
in his nominal work (Markowitz, 1952). However, in Markowitz’s portfolio selection problem, the objective is to
minimize the portfolio’s risk, with a constraint to the portfolio’s expected return. Recently, other objectives and
constraints, which correspond to the current investment needs, have been of great interest.
A number of various methodologies have been applied to the portfolio optimization problem, which can be
divided into two discrete phases. In phase one, a combination of assets has to be determined. In doing so, several
heuristics techniques such as financial rules-of-thumb, metaheuristics algorithms such as Tabu Search and other
intelligent techniques have been applied. After the combination of assets has been determined, the amount of capital
invested in each of these assets (portfolio’s weights) has to be computed. Traditional methodologies from statistics
and mathematics (e.g. non-linear programming algorithms) have been implemented in order to calculate the
portfolio’s weights. However, one drawback of these techniques is that the possibility of getting trapped in local,
low-quality, optimum areas is considerable. In order to overcome this obstacle, intelligent metaheuristics from the
field of Artificial Intelligence (AI) may be used. AI comprises of several methodologies whose main characteristic is
that each of them employs certain intelligent heuristics, mostly based on the way natural systems work and evolve,
in order to solve problems from various domains such as industrial application, financial problems etc.
In this study, a hybrid genetic algorithm is applied in order to solve a complex portfolio optimization problem.
More specifically, a certain type genetic algorithm is used in order to find good combination of assets, and the LM
algorithm is applied so as to find optimal weights for the portfolios. The objective is to maximize the financial ratio
Sortino, which takes into account both the concept of expected return and risk. Moreover, there's a constraint, which
refers to the tracking error volatility of the constructed portfolios. The objective function of the problem, as well as
the constraint on the tracking error volatility, is non-linear. The main aim of this study is to highlight the
effectiveness of the proposed hybrid scheme both in terms of solutions’ quality and computational effort required. In
order to provide some useful insights regarding the performance of the genetic algorithm, some benchmark
methodologies are applied to the problem at hand.
72

This paper is organized as follows. In section 1, some introductory comments regarding both the methodology
and the application domain are presented. In section 2, some basic studies, regarding the application of genetic
algorithms in the portfolio optimization problem, are shown. In section 3, several methodological issues are
discussed in brief. In section 4, the mathematical formulation of the problem is presented. In section 5, results from
experimental simulations are provided. Also, a detailed analysis on these results is given. Finally, in section 6, some
concluding remarks and future work directions are provided.
2. Literature Review
Several studies have been conducted regarding the application of genetic algorithms in the portfolio optimization
problem. In this section, a selection of standard works in the field is presented in brief. In any case, this analysis is
not considered to be exhaustive, but only representative of the application of genetic algorithms in portfolio
optimization.
In their study, [Branke, Scheckenbach, Stein, Deb & Schmeck 2009] proposed a hybrid scheme comprising of
principles from evolutionary algorithms and the critical line algorithm (for parameter quadratic programming) with
the aim of solving complex portfolio optimization problems with nonlinear constraints. The task of the MOEA is to
define a set of subset in the solution space. Then, for each subset found, the critical line algorithm generates a set of
optimal portfolios. Authors deal with the classical mean-variance optimization problem and their dataset comprised
of Hang Seng 31, S&P 98, Nikkei 225 and S&P’s 500 markets. Cardinality constraint was set to 4 and 8 assets.
Parameters for the evolutionary algorithms were: population size of 250 and 30 generations. Results are promising
regarding the superiority of the proposed methodology.
In their paper, [Maringer & Kelleler 2003] proposed a hybrid algorithm consisting of the Simulated Annealing
algorithm and several mechanisms from Evolutionary Programming. The optimization problem was the classical
mean-variance Markowitz’s approach. Cardinality constraint was set to 9 and 39, for two datasets, namely the
FTSE100 and DAX30 indices. In order to find the optimum solution, the population size was set to 100 and the
number of generations to 750. First tests led to promising results and supported the findings for the algorithm
presented in their study.
In another work, [Streichert, Ulmer & Zell 2003] dealt with the classical Markowitz portfolio optimization
problem, using principles of evolutionary strategies. Specifically, two extensions to evolutionary algorithms are
introduced. First, a problem specific representation of evolutionary algorithms is introduced. Second, in the previous
algorithm, a local search mechanism is introduced in order to enhance the performance of the scheme. As far as the
application domain is concerned, data from the Hang Seng 31 stock market are used. Cardinality was set to 2, 4 and
6 assets, while the parameters of the evolutionary scheme were: population size of 500, tournament group size 8,
crossover probability 1 and mutation probability 0,01.
In their paper, [Chen & Hou 2006] applied a modified genetic algorithm, which has the ability to efficiently
solve combinatorial optimization problems. The main characteristic of the algorithm is the specific combination
encoding scheme and genetic operators, both of which are designed for solving combination optimization problems.
This method is applied to the portfolio optimization problem with the aim of maximizing the portfolio’s expected
return with a built-in constraint on the portfolio’s risk. Regarding the application domain, data from the Taiwan
Stock Exchange were used. The parameters of the genetic algorithm were: population size of 200, crossover
probability 1 and mutation 0,05. Cardinality constraint was set to 20 assets. The experimental results demonstrate
the feasibility and effectiveness of the combination GA for the integer portfolio optimization problem.
[Chen, Hou, Wu & Chang 2009] applied a specific type of genetic algorithms to the investment portfolio
problem. The objective of the problem is to maximize portfolio’s expected return, under a constraint in the
portfolio’s risk. Dataset comprised of stocks from the Dow Jones Industrial Average. The portfolios were formed
using 9 months of historical data and they were adjusted every 3 months. The parameters of the genetic algorithm
were: population size of 200, crossover probability 1, mutation probability 0,05 and number of generations 500.
All in all, it can be seen that several types of genetic algorithms, from standard to more advanced, have been
applied to the portfolio optimization problem. In most cases, the formulation of the problem refers to the classical
73

Markowitz’s mean-variance optimization. However, there have been studies, which incorporated non-linear realworld
constraints. The applicability and effectiveness of genetic algorithms in the portfolio optimization problem is
quite obvious.
3. Methodology
In this study, a variant of the genetic algorithm is proposed, combined with a mathematical optimization tool, for
solving the portfolio management problem. The standard genetic algorithm was firstly proposed by Holland
[Holland 1992]. Their main characteristics lie in the concept of evolutionary process. More particularly, as in the
real world, genetic algorithms apply the mechanisms of selection, crossover and mutation in order to evolve the
members of a population, through a number of generations. The ultimate goal is to reach a population of goodquality.
In order to assess the quality of each member of the population, the concept of fitness value is introduced.
All in all, genetic algorithms are computationally simple procedures, though powerful in their search for the optimal
solution. Moreover, they are not limited by any assumptions regarding the search space, thus enabling them to apply
quite capable strategies.
In practice, the genetic algorithm operates upon a number of possible solutions called population. Each member
in the population is called chromosome, representing a solution to the problem. In this study, each combination of
assets (portfolio), along with their corresponding weights, comprises a chromosome. Therefore, the population is a
collection of the ‘best-so-far’ portfolios. At first, the population is randomly initialized. Then, for each generation
(epoch) of the algorithm, the chromosomes evolve by combining with each other, using the crossover and mutation
operators. As the number of generations increases, better quality solutions are kept in the population. However, due
to the fact that the size of the solution space is vast, there is a considerable probability of stagnation, meaning that
after a certain number of generations the evolution process may halt, i.e. the genetic algorithm is stuck to a local
optima region. This could be a potential problem for the performance of the genetic algorithm. It is important to
note, in this point, that the evaluation of each chromosome is based on the value of the objective function of the
optimization problem at-hand (fitness value).
In the previous paragraph, the concepts of evolutionary operators, i.e. selection, crossover and mutation, were
mentioned. Crossover and mutation are applied to already selected (good) members of the population so as to
produce ‘children’ with better solution-characteristics (assets/weights) and therefore better fitness value. However,
the main point of focus in this study lies in the mechanism of selection, which refers to the procedure of picking
members of the existing population in order to produce descendants (new portfolios). Particularly, the basic aim is
the assessment of how the choice of a specific selection strategy affects the performance of the generic algorithm.
Three alternative selection operators are implemented: a) selection of N-best members from the existing population,
b) application of roulette-wheel process for selection, c) application of tournament selection process. According to
the first method, only the best solutions from the existing population are considered for the evolution of the process.
The second method (roulette wheel) can be described as follows: a probability of selection, based on the fitness
value, is assigned to each member of the existing population. The mathematical formula which calculates this
probability is:
Finally, the third approach (tournament selection) works as follows: firstly, all members of the existing
population are split into n-groups, randomly. Then, from each group, the best member is chosen.
In order to determine the weights of portfolio’s assets found by the genetic algorithm, a local search non-linear
programming technique, namely the Levenberg-Marquardt method, is applied. In this case, the solution space is
continuous and has upper and lower limits, defined by the floor and ceiling constraints. For a given combination of
assets, the aim of the Levenberg-Marquardt method is to find a vector of weights which minimizes the given
objective function under certain constraints.
74
(1)

4. Portfolio Optimization Problem
The portfolio optimization problem deals with finding the optimal combination of assets and their corresponding
weights, as well. Harry M. Markowitz, with his seminal paper [Markowitz 1952], established a new framework for
the study of portfolio optimization. In his classical problem, the objective of the investor is to minimize the
portfolio’s risk, whereas imposing a tight constraint in portfolio’s expected return.
Nowadays, more complex formulations of the portfolio optimization problem are tackled. Non-linear objectives
and constraints are introduced to the classical formulation, in a way that real-world situations are reflected. The
objective of the portfolio optimization problem is to maximize a financial ratio, namely the Sortino ratio [Kuhn
2006]. Sortino ratio is based on the preliminary work of Sharpe (Sharpe ratio) [Sharpe 1994], who developed a
reward-to-variability ratio. The main concept was to create a criterion that takes into consideration both assets’s
expected return and volatility (risk). However, in recent years, investors started to adopt the concept of “bad
volatility”, which considers returns below a certain threshold. Sortino ratio considers only the volatility of returns,
which fall below a defined threshold.
Also, in this work, there is a constraint on tracking error volatility, i.e. a measure of the deviation between the
portfolio’s and benchmark’s returns, is imposed. This restriction refers to a passive portfolio management, namely
index tracking, which aims at constructing a portfolio using assets from a (stock) market in a way that attempts to
reproduce the performance of the market itself. Passive portfolio management, as a concept, is adopted by investors
who believe that financial markets are efficient, i.e. it is impossible to consistently beat the market.
The formulation of the financial optimization problem is presented below:
s.t.
where,
E(rP), is the portfolio’s expected return
rf, is the risk-free return
θ0(rP), is the volatility of returns which fall below a certain threshold and equals
wi, is the percentage of capital invested in the ith asset
K, is the maximum number of assets contained in a portfolio (cardinality constraint)
rB, is the benchmark’s daily return
H, is the upper threshold for the tracking error volatility 2
1 The constraint on tracking error volatility is incorporated into the objective function, using a penalty term (0.8)
2 H equals 0,0080
75
(2)
(3)
(4)
(5)
(6) 1
(7)

, is the probability density function of the portfolio’s returns. Assuming that portfolio’s returns follow a normal
distribution, the probability density function can be defined as:
3. Experimental results
In order to extract some useful conclusions regarding the performance of the hybrid scheme, a number of
independent simulations have been conducted. In each simulation, the configuration settings of both the algorithm
and the problem have been properly adjusted.
The dataset comprised of 93 daily returns, corresponding to the period 04/01/2010 – 29/05/2010, of 49 stocks of
the FTSE/ASE40 Index. In this point, it has to be mentioned that all stocks of the Index have been taken into
consideration (even those stocks corresponding to firms which have been excluded the Index). The reason for doing
this is to eliminate the effect of survivorship bias 3 .
In what follows, the configuration settings of both the hybrid algorithm and the portfolio management problem
are presented (Table 1). As it can be seen, a range of values for the number of generations and the cardinality of the
portfolio are used.
Parameters for Genetic Algorithm
Population 200
Generations 20/30
Crossover Probability 0,90
Mutation Probability 0,35
Number of best members for selection (for n-best members selection) 20
Number of groups (for tournament selection) 20
Number of members in each group (for tournament selection) 10
Parameters for optimization problem
[Floor Ceiling] – Constraint [-1 1]
Cardinality 10/20
Table 1. Configuration settings
In table 2, the main findings of the simulations are presented. Specifically, percentiles of the distribution of the
independent runs are presented. If the percentile of variable X is a in 0.95 confidence level, then there is a
probability of 95% that X will get values larger than a. So, it is preferable for the percentiles to have large values,
indicating a distribution that is shrunk as far right as possible. What is more, benchmark results from other heuristics
are presented, in order to provide some means of comparison. The following financial heuristics have been used:
constructing portfolios with a) maximum Sortino ratio, b) maximum cumulative return, c) maximum Sharpe ratio, d)
maximum expected return with a penalty term in tracking error volatility and e) random choice of assets. The
weights are calculated applying the LMA method.
Based on the results, the following basic points can be pointed out. First of all, the roulette wheel process yields
the worst results in all cases. This mechanism lies in the random selection process in a great extend. Thus, somebody
could state that the selection of members from the existing population is implemented through a ‘pure’ random
process. On the other hand, it seems that the selection of N-best members for re-production is the best selection
operator. Preliminary results indicate that although this mechanism is biased to pick good members from the
population, it does not stuck to local optima regions, compared to the alternative mechanisms. Distribution of
simulation results is denser and is located more to the right. Also, we could remark that the hybrid scheme outperforms
the financial rules. As far as the financial implications are considered, it is quite obvious that portfolios
with many assets yielded better results, compared to low cardinality portfolios.
3 Tendency for failed companies to be excluded from performance indices mainly because they no longer exist. This effect often causes the
results of the studies to skew higher because only companies which were successful enough to survive until the end of the time period of the
study are included.
76

Percentiles 0,05 0,50 0,95
Hybrid Algorithm
Cardinality: 10
Generations: 20
N-best 1,9424 2,2061 2,5134
Roulette Wheel 1,3118 1,6167 1,9677
Tournament Selection 1,8482 2,1742 2,5935
Generations: 30
N-best 1,9685 2,4310 2,6940
Roulette Wheel 1,5755 1,7760 2,1270
Tournament Selection 1,9720 2,3090 2,7317
Cardinality: 20
Generations: 20
N-best 2,2554 2,6970 3,3285
Roulette Wheel 1,8660 2,1488 2,4879
Tournament Selection
Generations: 30
2,1896 2,6218 3,0311
N-best 2,4437 2,8810 3,4704
Roulette Wheel 2,0992 2,3512 2,7197
Tournament Selection 2,3770 2,7617 3,3117
Financial Rules Best solution
Cardinality: 10 Cardinality: 20
Sortino ratio 0,6272 0,8068
Cumulative return 0,5909 0,7795
Sharpe ratio 0,5515 0,4784
Expected return 4
0,7136 0,7747
Random asset selection 0,3024 1,6182
Table 2. Simulation results
In table 3, bootstrapping is applied to the original dataset with the aim of producing a number of different
scenarios. The reason for that is to examine the performance of both the hybrid algorithm and the financial heuristics
in ‘unknown data’, by providing statistics of the distribution of results 5 [Gilli & Winker 2008].
In order to apply the specific sampling method, we considered that the cardinality constraint defines, in a way, a
unique optimization problem, i.e. results from the 10-asset problem cannot be compared directly to the 20-asset
problem (where the distribution of results is more acceptable). Also, for each cardinality the ‘global’-best portfolio
found in the hybrid scheme is used for implementing the resampling technique. In essence, 500-scenarios for the
stocks returns-series were produced based on the original dataset. By doing this, the unique characteristics of the
original dataset’s distribution are kept. Then, for each scenario, the best portfolio was applied and a value for the
objective function was calculated. As a final result, the percentiles of the objective function’s distribution are
presented in each case. For the 10-asset portfolio optimization problem, the ‘global’-best portfolio was found when
the number of generations was set to 30 and tournament selection was applied. For the 20-asset portfolio
optimization problem, the hybrid scheme yielded the ‘best’ results in 20 generations and when selection of N-best
members from population was applied. Results indicate that the hybrid scheme out-performs the financial rules in
‘unknown’ data.
Percentiles 0,05 0,50 0,95
Cardinality 10 20 10 20 10 20
Hybrid Algorithm 2,1161 2,3176 2,9547 3,7487 4,0191 6,7839
Sortino ratio 0,1754 0,3885 0,6201 0,8068 1,0506 1,2000
Cumulative return 0,1970 0,4338 0,5923 0,7820 1,0365 1,3816
Sharpe ratio 0,1840 0,2719 0,5425 0,4784 0,9809 0,7187
Expected return 0,3673 0,4086 0,7050 0,7747 1,1259 1,2258
Random asset selection 0,0152 1,0740 0,3105 1,6182 0,6277 2,1613
Table 3. Results from bootstrapping
4 Tracking error volatility constraint included
5 By applying bootstrapping in the original dataset, the produced scenarios of returns retain the properties of the original dataset.
This is a more ‘fair’ approach than applying the portfolios found to another dataset (there is not any forecasting component in the
proposed methodology).
77

4. Conclusions
In this study, the performance of a hybrid intelligent scheme was analyzed. More specifically, the proposed
technique comprised a genetic algorithm and the LMA algorithm. This hybrid scheme was applied to a portfolio
optimization problem. The GA component aimed at finding high-quality combination of assets, whereas the LMA
algorithm computed the optimal weights. This paper focused on the application of alternative mechanisms for
selection, which is a main component of the genetic algorithm. Three different mechanisms were applied: N-best
members, roulette wheel and tournament selection. The selection operator plays a vital role in the process of
portfolio construction.
Results from experimental simulations indicate that the application of the N-best mechanism is acceptable,
compared to the other two strategies. In order to apply the specific mechanism, the only thing to be considered is the
N-best portfolios in each generation. The important matter is that no random selection component is implemented.
So, there is a ‘clear’ bias towards the best solutions. However, this is not always the case, because in several studies
the issue of getting stuck in local optima regions, when considering only the best-so-far solutions, hinders the
optimization process. Another point of this study is that the hybrid intelligent scheme out-performed the financial
rules implemented, in all cases. Nevertheless, in order to assess the performance of the proposed techniques in
‘unknown’ data, we applied a method of resampling from the original dataset so as to produce new scenarios of the
stocks’ returns-series. Afterwards, the best portfolios were applied to the produced data series, and the distributions
of results were calculated. The hybrid scheme yielded better results, compared to the financial heuristics, in this case
also.
However, in order to obtain some better insight regarding the overall performance of the hybrid scheme, the
following basic future research directions are proposed. First of all, the proposed hybrid intelligent scheme should
be compared with other hybrid intelligent algorithms, whose main characteristic (and advantage) is the application
of good searching strategies. Complex search strategies could be able to capture any trends and patterns in the
dataset. Another interesting direction is the implementation of an intelligent trading system, whose main
components could be: a) a hybrid nature-inspired algorithm for portfolio optimization, b) a set of intelligent trading
rules (for application in the validation phase) and c) re-balancing of the portfolio in specific (maybe pre-determined)
time intervals.
5. References
Branke, J., Scheckenbach, B., Stein, M., Deb, K. & Schmeck, H., Schmeck, H. 2009, ‘Portfolio Optimization with
an Envelope-based Multi-objective Evolutionary Algorithm’, European Journal of Operational Research, pp.
684-693
Chen, J. S. & Hou, J. L. 2006, ‘A Combination Genetic Algorithm with Applications on Portfolio Optimization’, In
Advances in Applied Artificial Intelligence, Berlin, Germany, vol. 4031, pp. 197-206
Chen, J. S., Hou, J. L., Wu, S. M. & Chang, C. Y. W. 2009, ’Constructing investment strategy portfolio by
combination genetic algorithms’, Expert Systems with Applications, vol. 36, pp. 3824-3828
Gilli, M. & Winker, P. 2008, ‘A review of heuristic optimization methods in econometrics, Swiss Finance Institute
Research Paper, no. 08-12,
Holland, J. H. 1992, ‘Genetic Algorithms’, Scientific American, pp. 66-72
Jeurissen, R. 2005, ‘A hybrid genetic algorithm to track the Dutch AEX-Index’, Bachelor thesis, Informatics &
Economics, Faculty of Economics, Erasmus University of Rotterdam
Kuhn, J. 2006, ‘Optimal risk-return tradeoffs of commercial banks and the suitability of probability measures for
loan portfolios’, Springer Berlin
Maringer, D. & Kellerer, H. 2003, ‘Optimization of Cardinality Constrained Portfolios with a Hybrid Local Search
Algorithm’, OR Spectrum, vol. 25, no. 4, pp. 481-495
Markowitz, H. 1952, ‘Portfolio Selection’, The Journal of Finance, vol. 7, no. 1, pp. 77-91
Sharpe, W. F. 1994, ‘The Sharpe ratio’, Journal of Portfolio Management, pp. 49-58
Streichert, F., Ulmer, H. & Zell, A. 2003, ‘Evolutionary Algorithms and Cardinality Constrained Portfolio
Optimization Problem’, Selected papers of the International Conference on Operations Research, Heidelberg,
pp. 3-5
78

ASSET PRICING
80

ASSET PRICING IN THE CYPRUS STOCK EXCHANGE: ARE STOCKS FAIRLY PRICED?
Haritini Tsangari & Maria Elfani, University of Nicosia, Cyprus
Email: tsangari.h@unic.ac.cy
Abstract. The purpose of the current research is to examine the efficiency of the Cyprus Stock Exchange (CSE) during the years
2002-2007. This is the time period after the big crash of the CSE, when the index rose more than 700% from 1 st January 1999 to 1 st
December 1999 and fell more than 80% on 30 th September 2001. It is also the period that marks the beginning of the liberalization of
interest rates in Cyprus, which started in 2002. Asset pricing is performed for 30 stocks listed on the CSE, using a two-pass regression
methodology. The FTSE/CySE 20 Index is used as a market proxy and 13-week Cyprus T-bills are used as the risk-free asset. Results
from the first-pass regression showed that 27% of the stocks had a beta coefficient higher than 1, and 62.5% out of these stocks were
from the financial industry. Results from the second-pass regression showed that, overall, the stocks are overvalued. Moreover, the
regression coefficient for beta is not equal to the average market risk premium and there is no relationship between stock risk premium
and beta. An extension of the second-pass regression tested the hypothesis that non-systematic risk should not explain the risk
premium. The results indicated that the hypothesis was not supported. Conclusions and recommendations are made accordingly.
Keywords: Cyprus Stock Exchange; Two-step regression; Asset pricing; Efficiency.
JEL classification: G12
1 Introduction
The Cyprus Stock Exchange (CSE) is an example of an emerging market whose short but turbulent history makes it
interesting to study. The CSE started its operation officially in 1996. Although the conditions appeared to be
promising, CSE experienced a big boom in 1999 and a huge crash in the next two years (see Tsangari, 2011). Many
people, mostly small investors, saw their lifetime savings vanish into the air, while the crash was accompanied by a
‘scandal’ and a number of violations and breaches. The current study will examine the efficiency of the CSE for the
years 2002-2007. This period was intentionally selected for many reasons, the main reason being that it is the time
period right after the big crash of the CSE and thus marks a new era for the market.
The traditional Capital Asset Pricing Model (Sharpe, 1964; Lintner, 1965; Mossin, 1966) will be used, in
a first attempt to perform the necessary tests for the Cyprus stock market during the period of interest. More
than four decades after its development, the CAPM is still widely used both by academics and practitioners, with its
attraction being that it offers powerful and intuitively pleasing predictions about how to measure risk and its relation
with expected return (Fama & French, 2004). A two-stage regression methodology will be employed, where in the
first-pass time series regression the betas of the stocks will be estimated, and in the second-pass regression the betas
will be used to test the risk-return relationship (Miller & Scholes, 1972). If the relevant market proves to be
efficient, investors will benefit from the fair stock prices with all available information fully reflected in them.
2 Asset Pricing in the Cyprus Stock Exchange
The Capital Asset Pricing Model (CAPM) provides a testable prediction about the relation between risk and
expected return by identifying a portfolio that must be efficient if asset prices are to clear the market of all assets
(Fama & French, 2004). In an efficient market all available information is fully reflected on the stock price,
providing its fair price, ensuring that it is neither underpriced nor overpriced.
2.1 The Capital Asset Pricing Model
CAPM builds on the portfolio theory developed by Harry Markowitz (Markowitz, 1952). The model assumes that
all investors are risk averse and when making investment decisions they only care about the mean and variance of
their one-period investment return, minimizing the variance given the expected return and maximizing the expected
return given the investment variance. Sharpe (1964) and Lintner (1965) add two key assumptions to the Markowitz
model to identify a portfolio that must be mean-variance-efficient. The first assumption is complete agreement:
82

given market clearing asset prices at time t-1, investors agree on the joint distribution of asset returns from time t-1
to time t, which is the distribution from which the returns we use to test the model are drawn. The second
assumption is that there is borrowing and lending at a risk-free rate, which is the same for all investors and does not
depend on the amount borrowed or lent. The familiar Sharpe-Lintner CAPM equation is given by:
E(Ri)�Rf�[E(RM)�Rf)]βi, i=1,…,N. (1)
In equation (1), βi is the market beta of asset i, is the asset’s systematic risk, given by the covariance of its return
with the market return divided by the variance of the market return, RM,
Cov(
Ri
, R
� i �
2
� ( R )
The beta of the fully diversified market is equal to 1. Therefore if the beta of the stock is more than 1, it means
that the stock is more volatile than the market, while if beta is less than 1 it means that the stock has lower risk. In
words, the CAPM in equation (1) states that the expected return on any asset i is the risk-free interest rate, Rf, plus a
risk premium, which is the asset’s market beta, ßi, times the premium per unit of beta risk or expected excess return
on the market, E(RM) - Rf. Similarly, the model predicts that the assets plot along a straight line, with an intercept
equal to Rf and a slope equal to E(RM) - Rf.
Unrestricted riskfree borrowing and lending is considered an unrealistic assumption. Black (1972) develops a
version of the CAPM without riskfree borrowing or lending. He shows that the CAPM’s key result – that the market
portfolio is mean-variance-efficient - can be obtained by instead allowing unrestricted short sales of risky assets
(which, however, can also be considered to be an unrealistic assumption). The Black version says only that E(RzM)
must be less than the expected market return, so the premium for beta is positive. In contrast, in the Sharpe-Lintner
version of the model, E(RzM) must be the riskfree interest rate, Rf, and the premium per unit of beta risk is E(RM) -
Rf. The success of the Black version of the CAPM in early tests produced a consensus that the model is a good
description of expected returns. These early results, coupled with the model’s simplicity and intuitive appeal, pushed
the CAPM to the forefront of finance (Fama & French, 2004).
The Sharpe-Lintner and Black versions of the CAPM share the prediction that the market portfolio is meanvariance-efficient.
This implies that differences in expected return across securities are entirely explained by
differences in market beta; other variables should add nothing to the explanation of expected return. It can be argued
that the efficiency of the market portfolio is based on many unrealistic assumptions, but, as Fama & French (2004)
stress, all interesting models involve unrealistic simplifications, which is why they must be tested against data. The
hypothesis that market betas completely explain expected returns can be tested using cross-section regressions (e.g.
Fama & MacBeth, 1973), or using time-series regressions (e.g. Gibbons, 1982; Stambaugh, 1982). Fama & French
(2004) look at both time-series and cross-section regressions from a different point of view: not as a strict test of the
CAPM, but as test of whether a specific proxy for the market portfolio is efficient in the set of portfolios that can be
constructed from it and the left-hand-side assets used in the test, so they can be considered as tests of efficiency.
Some of the recent cases of empirical tests which have shown consistency with the CAPM model, finding
significant relationships between beta and returns include Fletcher (1997) (UK stock market for the period 1975-
1994), Hodoshima, Garza-Gomez, & Kunimura (2000) (Tokyo Stock Exchange for the period 1956-1995), Shakrani
& Ismail (2001) (Malaysian Islamic unit trust for the period 1999-2001), Sandoval & Saens (2004) (Argentina,
Brazil, Chile and Mexico for the period 1995-2002), Tang & Shum (2004) (Singapore market), and Gursoy &
Rejepova (2007) (Turkish Stock Exchange).
2.2 The two-step regression methodology
The current study will follow the two-step regression methodology used in the early CAPM tests (e.g. Miller &
Scholes, 1972). The first-pass regression is given by the equation:
Rit�Rft�ai�bi(RMt�Rft)�eit
83
M
M
)
(2)

where Rit – Rft is the monthly risk premium of stock i, bi is the sample estimate of the beta coefficient of each stock,
RMt – Rft is the monthly risk premium of the market index over the sample period (t=1,…T, where T=total number of
months under examination) and eit is the monthly residual for stock i. Therefore, in the first pass regression the beta
of the stock is estimated using time series regression, where each of the risk premiums is regressed on the market
risk premium.
The Sharpe-Lintner CAPM says that the average value of an asset’s excess return (Rit - Rft) is completely
explained by its average realized CAPM risk premium (its beta times the average value of (RMt - Rft)). This implies
that “Jensen’s alpha,” the intercept term in the time-series regression of equation (2) is zero for each asset. This can
be seen by rearranging equation (1), by moving Rf to the left side; the CAPM assumes that the intercept alpha is
equal to zero. If alpha is zero, then the asset is exactly on the Security Market Line (SML), the line which measures
the linear relationship between risk (beta) and expected return (E(Ri)), or that the asset is fairly priced. Similarly, if
alpha is positive then the asset is plotted above the SML or the asset is undervalued and if alpha is negative, then the
asset is plotted under the SML or it is overvalued.
The output of the first-pass regression, the estimated betas, will then be used as input into the second-pass
regression. The second-pass regression is given by:
i
f
R � R = γ0 + γ1bi , i=1,………n. (3)
i
f
R R � are the sample averages (over all months) of the excess return on each of the n stocks, which are regressed
on bi, the estimated beta coefficient of each asset, in order to determine the relationship between risk (beta) and
return. Therefore, if the CAPM is valid, then the intercept, γ0 (alpha) should be zero and the coefficient γ1 should
R R � , which is the sample average of the excess return of the market index.
equal M f
In the current article, the second-pass regression model will, be further expanded, with the variances of the n
residuals, σ²(ei), as an additional independent variable. The purpose of this expansion is to test if the expected excess
return on assets is determined only by the systematic risk, as measured by beta, and is independent of the
nonsystematic risk, as measured by the variance of the residuals. Notice that the variances of the residuals, σ²(ei), are
estimated from the first-pass regression. The extended second-pass model is, thus, as follows:
R R � = γ0 + γ1bi + γ2σ²(ei) (4)
i
f
The extended regression equation (4) is estimated with the hypotheses γ0=0, γ1= M f R R � , γ2=0. The
hypothesis that the coefficient γ2 should be zero is consistent with CAPM, which assumes that the risk premium
depends only on beta and there is no other variable in the right-hand-side of the equation that is significant for the
regression model. In other words, γ2=0 agrees with the notion that nonsystematic risk should not be “priced”, that is
that there is no risk premium earned for bearing nonsystematic risk.
2.3 The Cyprus Stock Exchange
The Cyprus Stock Exchange (CSE) has been officially operating since 29 March 1996, and it currently has 122
listed companies (Cyprus Stock Exchange, 2011). CSE is supervised by the Ministry of Finance. During its early
years of operation CSE experienced a big boom followed by a crash: the market rose strongly in 1999, when its
main index had an up to 700% increase, from 97 points on 1 January 1999 to 852 points on 1 December 1999, but
fell sharply after that, by more than 80%, to 103 points on 30 September 2001 (see Tsangari, 2011). The crash was
accompanied by a ‘scandal’ and a number of violations and breaches.
The current study examines the market efficiency for the years 2002-2007, the time period right after
the big crash. This period marks a new era for CSE. First, a fully automated online settlement and
clearing system was launched in 2002, providing high liquidity and marketability, thus reducing the risk of
84

the market being manipulated by few players. In 2002 the Ministry of Finance in its efforts to address the
problem of CSE decided to redesign the institutional and legal framework with the view to address the
challenge of integration into the unified financial market of the European Union (EU) (Tsangari, 2011). It is
also the period that marks the beginning of the liberalization of interest rates in Cyprus, which started in 2002. In
addition, the CSE, within its strategic plans for 2004–2006, moved to an agreement with the Athens Stock
Exchange for the development of a common platform between the two stock markets. When Cyprus
joined EU on May 1 2004, the adjustments to the financial system and the operations of the Cyprus financial
market began, making CSE an integral part of European financial system and the European Financial Market,
making it one the of the most interesting emerging markets nowadays.
2.3.1 Selection of CSE companies for the efficiency tests
The companies listed in CSE are divided into different economic sectors, according to the FTSE Industry
classification benchmark: Financials, Government Bonds, Corporate Bonds, Consumer Services, Industrials,
Consumer Goods, Telecommunications, Technology and Basic Materials. Among these sectors, the financial sector
represents 92.7% of the total volume of the all sectors (e.g. Cyprus Stock Exchange, 2009). More specifically, in the
financial sector, Bank of Cyprus Public Company Ltd and Marfin Popular Bank Public Company Ltd comprise the
largest market capitalization with approximately 64% of the total market capitalization (Cyprus Stock Exchange,
2009). By the end of the period of interest there were about 140 listed companies, whereas the market capitalization
of shares (excluding investment companies) reached 16.45 billion euros (Cyprus Stock Exchange, 2007).
In the current study 30 companies were selected for inclusion in the efficiency tests. The companies were
selected so that they represent all sectors of the Cyprus economy and give a representative picture of the market,
especially in terms of market capitalization. The selected companies have been listed in the Cyprus Stock Exchange
since its official opening in March 1996. All the historical data have been collected from the official website of the
Cyprus Stock Exchange. The stock return (Ri) will be calculated as (Pt – Pt-1)/Pt-1 where Pt is the closing price of the
stock on day t. The monthly return will be calculated using the average of the daily returns in each month.
2.3.2 Selection of market index used as market proxy for CAPM tests
The CSE began to cooperate with the London Stock Exchange and Financial Times to establish the FTSE/CySE 20
index in November 2000. The purpose of FTSE/CySE 20 Index is “to provide a real time measure of the Cyprus
Stock Market on which index-linked derivatives could be traded” (FTSE, 2009, p.3). The FTSE/CySE 20 Index is
managed by the FTSE International Limited, the Cyprus Stock Exchange (CSE), and the FTSE/CySE Index
Advisory Committee. The eligible securities to include in FTSE/CySE 20 Index are the 20 largest securities valued
by their market capitalization after passing the investibility screens. According to the Ground Rules, the investibility
screens consist of four criteria: first, 20% of the stock’s shares must be publicly available for investment (free float),
in other words they should not be in hand of a single party or parties; second, the securities must be accurately and
reliably priced so that the market value of the company should be determined satisfactorily; third, the securities
should be traded at least 50% of the business day during the six calendar months; fourth, the securities must have
traded for at least 20 trading days before the review collection date takes place. Based on these eligibility criteria,
the authors have decided to use the FTSE/CySE 20 Index as the market proxy to test the CAPM in CSE. According
to the Ground Rules, the FTSE/CySE 20 index is calculated by using the formula:
n
�
i�1
X W A
where Xi is the latest trading price of the i th component security, n is the total number of securities in the index, Wi is
the weight of the i th component security (or number of ordinary shares issued by the company), Ai is the free float
percentage of capitalization available to all investors, and the divisor, d, is the total issued share capital of the index.
The performance of the FTSE/CySE 20 Index for the years 2002 to 2007 is shown in Figure 1.
i
85
d
i
i
,

Figure 1. The performance of the market index for the time period under investigation (2002-2007).
In the CAPM tests, the market return (RM), will be calculated using the market proxy, FTSE/CySE 20 index.
More specifically, RM will be calculated as (Mt – Mt-1)/Mt-1 where Mt is the value of the FTSE/CySE 20 index in
month t. All the historical data have been collected from the official website of the Cyprus Stock Exchange.
2.3.3 Selection of risk-free instruments for CAPM tests
The Central Bank of Cyprus issues short-term risk-free instruments, namely Treasury Bills (T-bills), 13-week and
52-week. Both types of T-bills are issued to the public via bid-price auction. The 13-week T-bills were first
auctioned publicly in 1996, while the 52-week T-bills were publicly auctioned in 1998. Only the 52-week T-bills are
listed in the CSE so that the prices are determined at the stock exchange floor. The 13-week T-bills which are not
listed in the CSE are issued either by auction or at fixed prices that meet the government investment needs of the
Social Security Fund (Stephanou & Vittas, 2006).
The authors have decided to use the 13-week Cyprus T-bills to represent the risk free rate, Rf, in the efficiency
tests. The reasons that 13-week T-bills were chosen over 52-week T-bills were, first, Eurostat reports the rate of 13week
T-bills for Cyprus, and, second, the shorter-term nature of 13-week T-bills gives it the representative role of a
risk free asset; similar to the use of 90-day T-bills in tests for the US market, shorter-term T-bills yields are more
consistent with the CAPM as originally derived and reflect truly risk-free returns in the sense that T-bill investors
avoid material loss in value from interest rate movements (Bruner, Eades, & Schill, 2010). Monthly T-bill rates are
collected for the years 2002 to 2007 for inclusion in our tests. It should be stressed that before 2002 the rate was
constant at 9%, and, therefore, data are collected after the liberalization of interest rates. All the historical data have
been collected from the official websites of the Central Bank of Cyprus.
3 Results
3.1 First-Pass Regression
The sample included 30 stocks from the Cyprus Stock Exchange. Results from the first-pass regression of equation
(2) showed that 29 companies have a significant beta coefficient (p-value

overall results, and especially the fact that both the lowest and highest beta belong to companies in the Financial
services, show that there is no specific pattern in terms of the level of systematic risk in relation to the industry
associated with it. The first-pass regression also produced the residuals, the variance of which will be later used as
input in the extended second-pass regression.
3.2 Second-Pass Regression
In the second-pass regression of equation (3) the sample averages of the excess return of the stocks (averages are
over the 72 monthly observations, for the years 2002-2007) were regressed on the estimated beta coefficients of the
29 stocks, taken from the first-pass regression. The results appear in Table 1.
Independent Variable (coefficient) B Standard Error t p-value
Intercept (γ0) -0.038 0.001 -35.651

eta, and high-beta stocks have (on average) yielded lower returns than they should have on the basis of their beta.
Although the observed premium per unit of beta is lower than the Sharpe-Lintner model predicts, the relation
between average return and beta is roughly linear, consistent with the Black version of the CAPM, which predicts
only that the beta premium is positive. Third, equation (4) assumes that the nonsystematic risk, represented by the
coefficient for variance of residuals (γ2), should equal zero, i.e. it should not explain the risk premium of the stock.
However, in our findings γ2 is equal to 3.027, and it is significantly different from zero (p-value=0.020

Bruner, R. F, Eades, K. M., & Schill, M. J. (2010). Case studies in Finance: managing for corporate value creation.
6 th edition, New York: McGraw Hill.
Cyprus Stock Exchange (2007). CSE Bulletin, March 2007, Issue 122.
Cyprus Stock Exchange (2009). CSE Bulletin, May 2009, Issue 148.
Cyprus Stock Exchange (2011). CSE Bulletin, March 2011, Issue 170.
Daniel, K., & Titman, S. (1997). Evidence on the characteristics of cross sectional variation in stock returns. Journal
of Finance, 52, 1-33.
Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial
Economics, 33, 3-56.
Fama, E. F., & French, K. R. (2004). The Capital Asset Pricing Model: theory and evidence. Journal of Economic
Perspectives, 18(3), 25-46.
Fama, E. F. ,& MacBeth, J. D. (1973). Risk, return and equilibrium: empirical tests. Journal of Political Economy,
81(3), 607-636.
Fletcher, J. (1977). An examination of the cross-sectional relationship of beta and return: UK evidence. Journal of
Economics and Business, 49, 211-221.
FTSE (2009). Ground Rules for the Management of the FTSE/CySE 20 Index, Version 2.0, November 2009.
Gibbons, M. R. (1982). Multivariate tests of financial models: a new approach. Journal of Financial Economics,
10(1), 3-27.
Gursoy, C. T., & Rejepova, G. (2007). Test of Capital Asset Pricing Model in Turkey. Dogus University Journal, 8,
47-58.
Hodoshima, J., Garza-Gomez, X., & Kunimura, M. (2000). Cross-sectional regression analysis of return and beta in
Japan. Journal of Economics and Business, 52, 515-533.
Lewellen, J. (2002). Momentum and autocorrelation in stock returns. Review of Financial Studies, 15, 533-563.
Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital
budgets. Review of Economics and Statistics, 47(1), 13-37.
Markowitz, H. M. (1952). Portfolio Selection, Journal of Finance, 7(1), 77-91.
Miller, M. & Scholes, M. (1972). Rate of return in relation to risk: a reexamination of some recent findings. In M. C.
Jensen (Ed.), Studies in the Theory of Capital Markets (pp. 47-78). New York: Praeger.
Mossin, J. (1966). Equilibrium in a capital asset market. Econometrica, 34(4), 768-783.
Sakhrani, M.S., & Ismail, A.G. (2001). The conditional CAPM and cross-sectional evidence of return and beta for
Islamic Unit Trust in Malaysia. Bangkel Economi, Universiti Kebangsaan Malaysia, Bangi.
Sandoval, E.A., & Saens, R. N. (2004). The conditional relationship between portfolio beta and return: evidence
from Latin America. Cuadernos de Economia, 41, 65-89.
Sharpe, W. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of
Finance, 19(3), 425-442.
Stambaugh, R. F. (1982). On the exclusion of assets from tests of the two-parameter model: a sensitivity analysis.
Journal of Financial Economics, 10(3), 237-268.
Stephanou, C., & Vittas, D. (2006). Public debt management and debt market development in Cyprus: Evolution,
current challenges and policy option. Economic Policy Papers, No. 12-06.
Tang, G.Y.N., & Shum, W. C. (2004). The risk-return relations in the Singapore Stock Market, Pacific Basin
Finance Journal, 12, 179-195.
Tsangari, H. (2011). Emerging markets investment opportunities: Cypriot investors and Russian mutual
funds. International Journal of Business and Emerging Markets, 3(1), 89-106.
Zhang, L. (2005). The value premium. Journal of Finance, 60, 67-103.
89

THE VALUATION OF EQUITIES WHEN SHAREHOLDERS ENJOY LIMITED LIABILITY
Jo Wells, Bangor Business School, UK
Email: j.wells@bangor.ac.uk
Abstract. This paper introduces an alternative approach to the valuation of shares where shareholders enjoy limited liability. Limited
liability for shareholders is a relatively recent phenomenon, and insights into appropriate valuation methods can be drawn by first
considering the case where shares carry unlimited liability. This leads to the argument that rather than the commonly adopted
approach of directly modelling the share price using geometric Brownian motion, it is more appropriate for the underlying
shareholders' equity to be considered as the stochastic variable in the modelling exercise, with a birth and death model chosen as the
underlying stochastic process. The limited liability condition is imposed at a later stage in the valuation framework. These arguments
lead to the derivation of the value of the market value of equity for a non dividend-paying firm as a perpetual American-style call
option on shareholders' equity with a strike price of zero. The pricing of such an option is considered and potential empirical
applications discussed.
Keywords: equity valuation, perpetual options
JEL classification: G12, G13
1 Introduction
Traditional approaches to equity valuation involve the calculation of the present value of future earnings or cash
flows using an appropriate discount rate which takes into account the risk associated with those earnings or cash
flows (Gordon, 1959; Basu, 1977). This paper considers the fundamental nature of equity investment in the presence
of limited liability. The introduction of limited liability changes the nature of a share from a linear investment to a
non-linear option-type instrument. Within this framework, a share can be viewed as a perpetual American-style call
option on shareholders' equity with a strike price of zero. A consideration of the pricing of such an option leads to a
number of empirically testable hypotheses.
The paper proceeds as follows: Section 2 explains how in the presence of limited liability, a share can be
viewed as a perpetual call option on shareholders' equity with strike price of zero. Section 3 discusses the pricing of
such an option for an all-equity firm which pays no dividends. Section 4 presents the qualitative results produced by
this approach for the case of the non dividend-paying all-equity firm. Section 5 goes on to consider the impact of
leverage and dividends on the results presented in Section 4. Finally, Section 6 concludes.
2 Shares as Call Options
The key observation underpinning this paper is that limited liability changes the nature of the equity in a firm
from a linear investment to an option-type contract. This observation has previously been made, but to the author's
knowledge, the full description of the option-like nature of share investment arising from limited liability is new to
the literature. Merton (1974), for example, identifies a share as a call option on the assets of a firm with strike price
equal to the current level of debt, but the Merton model assumes a fixed maturity and models the assets of the firm
as a geometric Brownian motion. These two assumptions are questioned in the current paper.
In general, we can think of a very simple stylized balance sheet comprising total assets A financed by debt D
and shareholders' equity E. In the simple initial case, the firm pays no dividends and has no debt, so E = A. 1 Firstly,
it is instructive to consider the nature of equity investment in the absence of limited liability. Such concepts are not
as strange as they might at first seem. Limited liability is a relatively recent concept in many markets. In the UK, for
example, limited liability was not freely available to firms until the Limited Liability Act of 1855 (see Hunt, 1936
for a full discussion). Before that, only a small number of firms, typically those set up to undertake large utility
projects offering a high degree of perceived public good, had been granted limited liability by Royal Charter or
1 Throughout the paper, shareholders' equity is used to refer to the continuous variable measured by the book value of equity, which will be used
to refer to the accounting value published annually in the balance sheet.
90

Special Act of Parliament. Prior to the 1855 Act, therefore, stock market investment in the UK was a highly risky
undertaking which could easily result in personal ruin.
In the absence of limited liability and corporate insolvency, a share gives its owner a simple linear participation
in the shareholders' equity on the firm's balance sheet. The value of shareholders' equity is unconstrained and can
become negative. In such circumstances, shareholders may be called upon to inject further capital to support the firm
(essentially a negative dividend).
Shareholders in a firm have a fundamental right that is not generally highlighted in the literature. At any point in
time, the shareholders can collectively choose to continue with the firm's operations or to close the firm, in which
case they walk away with the current value of the shareholders' equity. The shareholders in a non dividend-paying
firm receive no cash flows from the firm but the value of their shares is expected to grow over time (and they can
sell shares to generate income as required, in the spirit of Miller & Modigliani, 1961).
Without limited liability, shareholders have this right but when shareholders' equity is negative, they would be
in a position of having the choice of paying into the company to make good the entire deficit in order to close it, or
continuing with operations (and possibly having to pay in a certain amount to support them). The shareholders'
choice of whether to close or continue with the investment does not, therefore, constitute an option contract.
The introduction of limited liability changes the nature of equity investment completely. Now, losses are limited
to the value of the share. Shareholders will not have to pay additional funds in to close the company should
shareholder' equity become negative. Instead of holding a direct participation in shareholders' equity itself, investors
now hold an asset that is akin to an option on shareholders' equity. In a traded stock market, the asset that is actually
being traded is the option contract, and the underlying shareholders' equity is not now directly traded 2 .
In order to model stock prices, therefore, we need to model shareholders' equity and hence the value of the
option held by shareholders in the presence of limited liability. We can think of a share as a call option with strike of
zero on the shareholders' equity of the firm (the strike price being zero since there is no cost associated with the
decision to close the firm and walk away with the shareholders' equity). The option is perpetual (it has no fixed
expiry date), and as the shareholders can exercise it at any time, it is American-style. Since the option price cannot
become negative, this in turn means that share prices cannot now become negative, which is consistent with realworld
experience. Corporate insolvency provisions, which in their simplest form mean that once shareholders’
equity hits zero the firm ceases to trade and the shareholders receive no payoff, can be viewed as a knock-out feature
at zero. The introduction of limited liability thus changes the nature of a share considerably from the perspective of
the investor, from a linear contract to a non-linear option-type contract.
Section 3 considers the pricing of such an option for the most simple case of an all-equity firm which pays no
dividends.
3 Pricing the Perpetual Call Option on Shareholders' Equity
The pricing of perpetual options has been considered in both the financial options and real options literatures.
Samuelson (1965) considers the pricing of warrants, including perpetual warrants, where the underlying is a stock
following a geometric Brownian Motion. More recent papers extend the approach to incorporate jump-diffusion
(see, for example, Gerber & Shiu, 1998). In the real options literature, perpetual call options are relevant to the
valuation of options to invest (see, for example, McDonald and Siegel, 1986, and Dixit and Pindyck, 1994).
The modelling of the perpetual American-style call option with strike of zero described in Section 2 requires
both a stochastic variable and an associated stochastic process to be specified. Shareholders’ equity is chosen to be
the stochastic variable. In the financial options literature, it is common practice to model stock prices as a geometric
Brownian motion. This approach assumes that logarithmic returns are normally distributed and prices therefore
lognormally distributed, in line with empirical evidence. The share price is therefore directly constrained to be
2 the collusion required by individual shareholders in the decision to exercise the option and close the firm is not explicitly considered in this
paper. Rather, it is assumed that shareholders will choose collectively to exercise the option only at a point where it is optimal to do so.
91

positive by the distributional assumptions implicit in the stochastic process. The same approach cannot be taken
with shareholders' equity as the stochastic variable, since it would then be impossible for shareholders' equity to
reach zero and hence impossible for limited liability to have an impact on share valuation in the way that is proposed
in this paper.
The approach taken in this paper is to model shareholders' equity as a birth and death process. Birth and death
processes, also known as square root processes, have been used widely in the modelling of populations (see, for
example, Moran, 1958) and in finance in stochastic volatility models of asset prices following Heston (1993). Cox
and Ross (1976) propose the square root process as an alternative to geometric Brownian motion for the modelling
of stock prices. The continuous limit of the birth and death process for total assets is given by
dA � �A dt � � Adz
(1)
In the case of an all-equity firm (A = E), shareholders' equity E follows the same process as total assets A:
dE � �E dt � � Edz
(2)
In this framework, shareholders' equity can reach zero with positive probability. The birth and death process
implicitly carries an absorbing barrier at zero, which in the context of the current problem is useful in that it prevents
shareholders’ equity from becoming positive again once it has reached zero (i.e. it implicitly takes corporate
insolvency into account). Observed share prices are positive not because the underlying stochastic variable is
constrained in the sense of a geometric Brownian motion but because they reflect call option prices, which by
definition can only be positive.
This observation underpins a further important difference between this paper and some of the prior literature on
perpetual option pricing. By the very nature of the problem, shareholders' equity (the underlying stochastic variable)
is not directly tradeable. As discussed in Section 2, when buying and selling shares, investors are in fact buying and
selling options on shareholders' equity. Since it is not possible to hedge perfectly the risk associated with the option
in the way that is assumed by standard Black-Scholes type pricing derivations (which would assume the ability to
directly trade in shareholders' equity), the differential equation governing the option price must take into account
this residual risk. The option is valued in a risk-adjusted setting with the drift term adjusted to compensate the option
writer for the unhedgeable risk. Letting C be the value of the perpetual American-style call option on shareholders'
equity E with strike price of zero where E follows equation (2), applying Itô's Lemma 3 gives
2 � 1 2 d C dC �
dC � �
� � E � �E
dt � �
dE dE �
�
2
� 2
�
dC
E
dE
dz
After adjusting the drift for the unhedgeable residual risk, the differential equation to be satisfied by the option
value function is
1
2
d C
dE
dC
��E � E � � � 0
2
2
� E � 2 �� rC
(4)
dE
where λ is the market price of risk, which is assumed to be constant. For the drift μ to exceed the risk-free return
r by an amount which compensates exactly for the risk σE 0.5 would require μ = r + λσ/E 0.5 . The birth and death
process implies reduced risk in percentage terms at higher levels of shareholders' equity, which translates into lower
required values of μ to compensate for that risk, as shown in Figure 1.
3 noting that
� C(
E)
�t
=0 since the option is perpetual
92
(3)

35%
30%
25%
20%
15%
10%
5%
0%
0 1 2 3 4 5 6 7 8 9 10
Figure 1. Required μ such that μ = r+λσ/E 0.5 (r = 0.03, σ = 0.10, λ = 2)
Where this holds, the solution of (4) follows the standard result that the perpetual call option will never be
exercised and its value is simply its intrinsic value E at any point in time. If, on the other hand, the possibility is
allowed that the rate of growth of a firm's assets need not satisfy this criteria in an incomplete market where the
underlying stochastic variable (shareholders' equity) is not traded, then very different results are obtained. Firstly,
the coefficient of the first derivative term in equation (4) is modified to (r+α)E where α represents the excess of μ
over the required rate shown in Figure 1. This gives
1
2
d C
dE
dC
dE
2
2
� E � �r � � �E � rC � 0
(5)
2
The solutions of the equation will therefore be price curves corresponding to particular values of α. The value of
α relevant to any particular firm at a particular point in time is an empirical issue to be determined separately. It is
not the case that firm values can be expected to move along a single price curve. From Figure 1, it is apparent that
the hurdle required (in terms of asset growth) to attain any particular value of α is much higher for a small firm than
a large firm, and we might therefore expect i) small firms to have lower α on average than larger firms and ii) α to
increase, other things equal, as a firm grows.
Equation (5) has no known closed-form solution. Instead, a Frobenius series solution is proposed following the
approach of Pinto et al (2009). The proposed solution takes the form
� � � �
� � an
n�0
n�
p
(6)
C �
where the values taken by p and the series an are determined by the particular valuation problem in the form of
the differential equation (5) and appropriate boundary conditions. Shareholders' equity is replaced by the scaled
variable ξ such that σ 2 ξ=E in order to ensure that the resulting Frobenius series converges to zero and hence can be
easily evaluated numerically. The Pinto et al method uses the Frobenius series to solve the equation analytically at a
point arbitrarily close to zero, with the solution then expanded numerically to higher values of E. From (6),
differentiating as required and then adjusting the counters so as to consider a power of ξ equal to n+p throughout,
dC
�
d�
� �
n��1
�n � p �1�
a
n�1
�
dC
n�
p
� �
n�
p
� � an�
(7b)
d�
n�0
so �n � p�
93
(7a)

2
d C
2
d�
�
� �
n��2
�n � p �1��n
� p � 2�
d
C
a
n�2
�
n�
p
� �
2
n�
p
� a
2
n�1�
(7d)
d�
n��1
so � �n � p��n
� p �1�
Substituting equations (6) and (7) into (5),
�
2
�
�
n�
p
n�
p
n�
p
�n p��n
� p �1�a
� ��r
� � � �n � p�a
� � r a � � 0
� � �
n�1
n
2 n��1
n�0
n�0
�
In order for equation (8) to hold, the total coefficient on each of the powers of ξ must be zero. Considering first
the case n=-1, the coefficient of ξ p-1 is (from the first term only)
� 1� 0 0 � � pa
p (9)
which has solutions p=0 and p=1. In general, the required solution will be a linear combination of the two
solutions
� � � �
1 � � an
n�0
n�1
, and (10a)
� � � � � � � �
2 � � ln � C1
� � bn
n�0
n
(10b)
C �
C �
where the first solution C1 corresponds to the larger of the two roots of (9) and the second solution C2 is the
Frobenius series corresponding to the root p=0 plus the natural logarithm of the first solution. A boundary condition
C(0)=0 is now imposed (whereby the option value is required to reach zero when shareholders’ equity reaches the
absorbing barrier of zero). This is satisfied by C1 but not C2, since C2(0)=b0. The required solution must therefore be
a multiple of C1 alone. Substituting the solution p=1 into (8) and considering this time the coefficients of ξ n+p for the
case n≥0 gives
�
2
2
a
a
n
n�1
�n 1��n
� 2�a
� �r � � ��n �1�a
� ra � 0
� n�1
n n
Changing the counter and simplifying,
2
�
�r � n�r
� � ��
2
� n�n
�1�
Once the initial term a0 has been found, the remaining terms in the series solution can be determined
analytically by equation (12) and the option value hence obtained from equation (6). To find the value of a0
corresponding to the particular solution of the valuation problem, further boundary conditions must be imposed on
the solution to the differential equation. The standard "high contact" or "smooth pasting" conditions commonly
applied to the valuation of American-style options (see, for example, Samuelson, 1965 and Merton, 1973), whereby
the level and first derivative of the option value are matched to those of the option payoff at the optimal exercise
point, cannot be applied here since the optimal exercise point is either infinity (the option will never be exercised) or
zero (the option will be immediately exercised) depending on whether α is positive or negative.
94
�
n
(7c)
(8)
(11)
(12)

Where α is positive, the pricing approach produces values of the option which are in excess of the current level
of shareholders’ equity. Ingersoll (1987, p373) refers to this as the “problem of infinities”, whereby the unbounded
potential payoff from future exercise combines with the low probability of exercise to produce a finite present value.
Conversely, where the drift term is insufficient to compensate for risk (i.e. α is negative), the pricing method
produces values which are below the current level of shareholders’ equity, in which case the optimal action by
shareholders it to exercise their option to “take the money and run”.
The optimal exercise point where α is positive is at infinity. The option payoff (for the unscaled variable E) has
a constant slope of 1 and the solution curves obtained tend to straight lines at high levels of E. The smooth pasting
condition that the option value and the first derivative of the option value match those of the payoff at infinity can
therefore be approximated by setting the first derivative of the solution equal to one at the highest level of E
considered. The solution curve is first calculated using a value a0=1, and then the required value of a0 in equation (6)
calculated so as to equate the derivatives of the option value and the payoff at the highest level of E considered.
Equation (6) is then used to calculate option values at a range of levels of shareholders' equity. Section 4 presents
the qualitative results and Section 5 briefly discusses how these results might be extended to incorporate leverage
and dividends.
4 Results
Figure 2 shows the values obtained from the pricing approach described in the previous section with parameters
r = 0.03, σ = 0.1 and λ = 2, and α ranging from 0% to 3%. The x-axis value E is the value of shareholders' equity (≈
book value of equity) and the y-axis value is the value of the perpetual call option (≈ market value of equity).
C
25
20
15
10
5
0
0 2 4 6 8 10
E
Figure 2. Market Value of the All-Equity Firm (r = 0.03, σ = 0.1, λ = 2)
From Figure 2, it is seen that the market value of the all-equity firm is increasing with firm size. Where α = 0, the
drift is exactly sufficient to compensate for risk at all levels of E, the option will never be exercised, and it is priced
at its intrinsic value consistent with the findings of prior research pricing perpetual options using a complete markets
approach. Higher values of α, i.e. a higher growth rate of assets and hence shareholders' equity, produce higher
option values. Negative values of α (not shown in Figure 2) produce option values below the option's intrinsic value.
Here, investors should immediately exercise the option ("taking the money and running"). Figure 2 suggests that the
market values of small firms are more sensitive to changes in the value of α than those of large firms in percentage
terms.
It is important to note that the playing field is not level for all firms. As shown in Figure 1, small firms need to
generate a much higher drift term μ than large firms in order to enjoy the same α as a result of the choice of the
square root (birth and death) process. Changes in the parameters σ and λ impact upon the required drift rate μ a firm
must generate in order to achieve a given level of α and hence plot on a particular curve in Figure 2.
95
α=0%
α=1%
α=2%
α=3%

Considering first the volatility parameter σ, Figure 3 shows how the drift μ corresponding to α=0 (i.e. the
minimum drift for the option to have value in excess of its intrinsic value) changes with σ. The solid black line
corresponds to the same parameters as the solid black line in Figure 1 (r = 0.03, σ = 0.1 and λ = 2). The impact of
changes in σ on required drift and hence α is more pronounced for small firms (represented by low E for the allequity
firm) than large firms (high E). At high levels of σ, it becomes increasingly difficult for small firms to
achieve the required drift to compensate for risk, and hence increasingly difficult for the option represented by
traded equity to price at above intrinsic value.
35%
30%
25%
20%
15%
10%
5%
0%
0 1 2 3 4 5 6 7 8 9 10
Figure 3. Required μ such that μ = r+λσ/E 0.5 (r = 0.03, λ = 2)
Figure 4 goes on to consider the impact of the market price of risk λ on required values of μ and hence option
values. The market price of risk is positively related to required drift, hence increases in λ will, ceteris paribus,
reduce the level of α achieved by individual firms and hence reduce option values. The impact of changes in the
market price of risk on required μ and hence α is more equal across firm size than that of changes in σ.
35%
30%
25%
20%
15%
10%
5%
0%
0 1 2 3 4 5 6 7 8 9 10
Figure 4. Required μ such that μ = r+λσ/E 0.5 (r = 0.03, σ = 0.10)
Figures 3 and 4 show how the levels of asset growth required to achieve a given excess growth rate α vary with σ
and λ. At high levels of volatility and market price of risk, firms (an in particular small firms) can be expected,
ceteris paribus, to achieve low levels of α and hence market values of equity close to book value (low market to
book ratios). When risk and the market price of risk are low, any given rate of asset growth μ corresponds to a
higher excess return α and hence higher option value (higher market to book ratios). Figure 5 shows the market to
book values corresponding to the option values in Figure 2.
96
σ=0.05
σ=0.10
σ=0.15
λ=1.5
λ=2.0
λ=2.5

M
T
B
V
20
18
16
14
12
10
8
6
4
2
0
0 1 2 3 4 5 6 7 8 9 10
E
Figure 5. Market to Book Values (r = 0.03, σ = 0.10, λ=2)
Figure 5 suggests that small firms will trade at higher book to market values than large firms to the extent that they
are able to generate the same rate of excess asset growth above that required to compensate for risk. Considering the
spread of values in Figure 5, one would expect to see a much higher degree of variation in market to book values
across small firms than large firms. These results are also relevant to the consideration of well-known anomalies in
stock market returns including size effects (Banz, 1981), value effects (Fama and French, 1992), the equity premium
puzzle (Mehra and Prescott) and the equity volatility puzzle (Shiller, 1981) and provide the basis on which specific
testable hypotheses relating to these observed anomalies can be derived.
5 Leverage and Dividends
Sections 3 and 4 considered the value of an all-equity firm which pays no dividends when the asset held by
shareholders is valued as a call option on shareholders' equity. In section 3, a mixed analytic-numerical method was
used to value the perpetual American-style option with strike zero where the underlying stochastic variable follows a
birth and death process. This section considers how the approach might be extended to take into account leverage
and dividends.
5.1 Leverage
In the stylized leveraged firm, total assets are financed by a combination of debt and shareholders' equity. Assuming
that the notional outstanding amount of debt is fixed, shareholders' equity follows the stochastic process
� � D ��E dt � � D Edz
dE � 1 1 �
(13)
E
E
Figure 5 shows the relationship between required μ and E for three different levels of D/E. As previously, the
solid black line corresponds to an all-equity firm (D/E=0) with r = 0.03, σ = 0.1 and λ = 2. Increases in leverage
(D/E) reduce the hurdle rate of required μ for all firms, with the effect similar across firm sizes and reducing
marginal benefits from increased leverage. Ceteris paribus, increasing leverage increases α and hence option
(market) value. The marginal impact diminishes as leverage increases, and given that the values of small firms
appear to be more sensitive to changes in α than those of large firms (Figure 2), the same percentage changes in
leverage can be expected to have greater percentage impact on the values of small firms than large firms.
97
α=0%
α=1%
α=2%
α=3%

35%
30%
25%
20%
15%
10%
5%
0%
0 1 2 3 4 5 6 7 8 9 10
Figure 5. Required μ such that μ = r+λσ/E 0.5 (r = 0.03, σ = 0.10, λ = 2)
D/E=0
D/E=0.5
Incorporating limited liability to shares therefore introduces implications for the impact of leverage on firm value
beyond the standard "debt shield" view of capital structure.
5.2 Dividends
Dividends essentially represent a partial distribution of shareholders' equity to shareholders. Shareholders (option
holders) receive these dividends whilst the option remains unexercised. The usual approach to incorporating
dividends into option pricing models is to adjust the drift of the underlying, but this approach is not appropriate here
as it implies that the underlying asset (shareholders' equity) is separately traded and pays dividends to its holders,
rather than to the holders of the option contract. In the current context, dividends can be viewed as a partial exercise
of the option, with a proportion of shareholders' equity distributed to shareholders (option holders) and the market
value of equity (the residual unexercised option value) reduced by an amount which depends on the size of the firm,
the amount of the dividend (the amount exercised) and the excess asset growth rate α (Figure 2).
Incorporating limited liability to shares therefore introduces implications for the adjustment of share values in
response to dividend announcements which differ from standard taxation-based arguments.
6 Summary
This paper explicitly considers the impact of limited liability on the valuation of shares. A share (market value) can
be viewed as a perpetual American-style call option with strike zero on shareholders' equity (book value). The
qualitiative results presented in Section 4 and the extensions discussed in Section 5 present a number of areas in
which the normative approach taken in this paper provides a foundation for the derivation of testable empirical
hypotheses.
7 Acknowledgements
I would like to thank Dr Helena Pinto for providing the Mathematica code used in Pinto et al (2009) and which
forms the basis of the approach used in solving the particular valuation problem considered in this paper. Any errors
remain my responsibility.
98
D/E=1

8 References
Banz, R. (1981). The Relationship between Return and Market Value of Common Stock. Journal of Financial
Economics, Vol. 9, 3-18.
Basu, S. (1977). Investment Performance of Common Stock in Relation to their Price Earnings Ratios: A test of the
Efficient Market Hypothesis. Journal of Finance, Vol 32(2), 663-678.
Cox, J.C. & Ross, S.A. (1976). The valuation of options for alternative stochastic processes. Journal of Financial
Economics, Vol. 3, 145-166.
Dixit, A.K., & Pindyck, R.S. (1994). Investment Under Uncertainty. Princeton University Press.
Fama, E.F. and K.R. French (1992). The Cross-Section of Expected Stock Returns. Journal of Finance, Vol. 47(2),
427-265.
Gerber, H.U., & E.S. Shiu (1998) Pricing Perpetual Options for Jump Processes, North American Actuarial Journal,
Vol. 2 (3),101-112.
Gordon, M.J. (1959). Dividends, Earnings and Stock Prices. Review of Economics and Statistics, Vol. 41 (2), 99–
105.
Heston, S.L. (1993). A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and
Currency Options. The Review of Financial Studies, Vol. 6 (2), 327–343.
Hunt, B.C. (1936). The Development of the Business Corporation in England, 1800–1867. Harvard University
Press.
Ingersoll, J.E. (1987). Theory of Financial Decision-Making. Rowman & Littlefield Publishers, Inc.
McDonald, R., Siegel, D. (1986). The value of waiting to invest. Quarterly Journal of Economics, Vol. 101, 707-
728.
Mehra, R. and E.C. Prescott (1985). The Equity Premium: A Puzzle, Journal of Monetary Economics, Vol. 15, 145-
161.
Merton, R.C. (1973). Theory of rational option pricing, Bell Journal of Economics and Management Science, Vol. 4,
141-183.
Merton, R.C. (1974). On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance, Vol.
29, 449-470.
Miller, M. & F. Modigliani (1961). Dividend policy, growth, and the valuation of shares. Journal of Business, Vol.
34, 411-433.
Moran, P.A.P. (1958). Random processes in genetics, Proceedings of the Cambridge Philosophical Society, Vol.54,
60–71.
Samuelson, P.A. (1965). Rational Theory of Warrant Pricing. Industrial Management Review, 6(2), 13-32.
Shiller, R.J. (1981). Do stock prices move too much to be justified by subsequent changes in dividends? American
Economic Review, Vol 71, 421-436
99

PRICE PRESSURE RISK FACTOR IN CONVERTIBLE BONDS
Nikolay Ryabkov, Swiss Finance Institute, University of Zurich, Switzerland
Galyna Petrenko, Universidad Carlos III de Madrid, Spain
Email: nkryabkov@gmail.com
Abstract This paper sheds the light on the previously unnoticed effect of hedge funds on the asset pricing of their investment
instruments. Demand fluctuations from the convertible arbitrage hedge funds generate a price pressure risk on convertible bonds
pricing. This risk is materialized into forced selling risk factor during market turbulence and liquidity. We aim at constructing the
price pressure risk factor based on the sensitivity of individual convertible bonds to the demand of hedge funds through long/short
factor mimicking portfolio and showing its impact on the convertible pricing and on a risk models for a convertible bond portfolio. In
construction of the factor we rely on information about the total assets under management of convertible arbitrage hedge funds. We
test the significance of new factor in the multi-factor risk modeling framework for a convertible bond portfolio and as a missing risk
factor for pricing models of convertible arbitrage hedge funds.
Keywords: convertible bonds, hedge funds, asset pricing
JEL classification: G20
1 Introduction
Convertible bond is a fixed income security that gives their holders an option to convert it into a predetermined
number of shares of an issuing firm. Traditionally, the convertibles were used for the outright investment through
the bottom-up approach, however since 90s along with growth of hedge funds, the focus of investment in
convertibles had been shifting towards arbitrage. Plain vanilla convertible arbitrage consists of long position in
convertible bond and short in the underlying stock. This strategy requires high leverage to obtain sound performance
results in spite of seemingly rich arbitrage opportunities due to potential mispricing of convertible bonds. Therefore
convertible arbitrage hedge funds might be quite vulnerable during liquidity crisis.
Over the past twelve years the market for convertible bonds experienced periods of significant troughs. First,
convertible bonds negatively reacted to the 1998 crisis after the LTCM collapse. Second example is a 2005 period
when convertible arbitrage hedge funds faced large redemptions from investors after poor past performance. Finally
the 2008 credit crunch crisis brought another drop in prices of convertibles in October 2008. In spite of different
reasons of those tail events, there is one common factor, namely significant price pressure on the convertible bonds
valuation from the demand side of hedge funds. The convertible bonds remain a niche product and the recent growth
in trading volumes and new issues were primarily driven by the demand of hedge funds. When a hedge fund
requires liquidity for any exogenous reason, it is forced to sell convertible bonds positions driving the convertibles
prices down and distracting the fundamental valuation of convertible bonds. The fluctuations in the hedge funds
demand creates a price pressure on the convertible bonds which should constitute a "forced selling" risk factor in the
convertible pricing models. Moreover, price pressure hypothesis would contribute to the explanation of systematic
convertible bonds underpricing in so that they carry a premium for the unexpected demand fluctuations.
Hedge funds are often called a liquidity provider and thus the price pressure risk factor is essentially the
liquidity risk factor. Therefore it should have the same features as other liquidity factors. Assets with high liquidity
sensitivity should be traded at a price premium relative to the securities with low sensitivity. The liquidity breadth
increases during market downturn. A high correlation in selling activity of hedge funds can lead to large trade
imbalance and can lower overall liquidity. The excess sell imbalance on the illiquid securities such as convertible
bond, is growing up. The financial crisis has more pronounced effect on the liquidity of small, volatility and high exante
liquidity beta stocks. Hedge funds choose to sell less liquidity-sensitive (low liquidity beta) securities because
selling illiquid securities during a crisis is very expensive. Therefore convertible arbitrage hedge funds typically
carry a liquidity risk premium through the mismatch of the liquidity of long and short positions which cannot be
hedged.
100

On the other hand, increase in risk aversion induces flight to quality that require increasing overall liquidity of
portfolio holding and thus portfolio flees to securities with higher liquidity during recession. Moreover higher risk
aversion can increase selling interest and decrease buying interest in convertibles during a market crash and/or
recessions. Hedge funds sell securities whose liquidity is less sensitive to market decline and they buy securities
with greater liquidity (flight to quality effect). On the top of the price pressure liquidity factor generated by hedge
funds, converts as well as corporate bonds are usually more illiquid than equities. This fact magnifies the impact of
liquidity concerns on the convertible bond pricing. To sum up, price pressure risk factor as well ex-ante liquidity
risk factor of convertible securities appears as a risk factor for an investor fund holding long position in convertibles
both mutual funds and arbitrage hedge funds alike. The convertibles holding carries the risk premium for price
pressure factor which explodes during market crashes and may or may not have been mitigated by flight-to-quality
effect during longer periods of recessions.
The objective of the paper is to test the hypothesis of price pressure of hedge funds on the convertible bonds
market and identify the price pressure as distinct risk factor in a convertible bond pricing model. It is important to
control for another risk factors when assessing the impact of hedge funds on convertible bonds. Therefore we first
develop the multi-factor risk model for convertible bond embedding the equity risk, bond-type risk factors and
volatility risk factors to capture the option premium. The purpose of the multi-factor model is to filter out other
potential risk factors that influence the convertible bonds pricing. We keep the model as parsimonious as possible
because it is only a benchmark model for our study. We call it conventional risk factor model and estimate it for the
individual convertible bonds, constituents of the UBS Global Convertible Bond Index in the period from October
2002 to September 2009 on monthly data. Second, we measure the hedge fund demand for convertible bonds based
on the estimated assets under management of the convertible arbitrage hedge funds taken from the TASS Lipper
hedge funds database. We understand that hedge funds are not the only investor in convertible bonds. There are
mutual funds or institutional flows however their demand would be more stable in comparison with hedge funds as
there is no evidence on the growing assets base of other convertible bond investors. We focus on convertible
arbitrage hedge funds because their primary investment strategy involves long position in convertible bonds. We do
not exclude possibilities that other hedge funds may invest in converts but there is no way to know it because
holding information of hedge funds is unavailable. Therefore we assume that most of demand was created by
convertible arbitrage hedge funds. Any change in the demand should be caused by change in the assets under
management and thus assets size reflects the demand dynamics. We normalize the total assets of convertible
arbitrage hedge funds by total convertible bond market capitalization to construct the excess demand value.
The methodology of construction and testing the new risk factor on convertible bond pricing is the following.
First we establish Granger causality relationship between hedge fund demand and convertible bond index returns.
There are two approaches: first directly testing for Granger causality between convertible bond index return and
hedge fund demand; second testing for Granger causality between residuals from estimated conventional convertible
bond multi-factor model and hedge fund demand. The second approach adds value as it clears out other potential
systematic risk factors influencing the convertible bond pricing. Moreover, as we expect difference between
financial crisis and normal time as well as between recession and expansion periods, we can control by adding time
dummy variables directly in the factor risk model regressions and thus amending the residuals. We provide three
types of output: for model without dummy variables, with dummy on the NBER recession period from January 2007
through June 2009; and with dummy on "post-Lehman" period from August 2008 through March 2009. Moreover
dummy variables would mitigate effect of structural break in convertible bond regressions. Second we estimate
sensitivities of individual convertible bonds and aggregated index towards hedge fund demand by running the
regression analysis controlling for other risk factors. The null hypothesis of the price pressure existence is significant
negative coefficient on the hedge fund demand measure. Convertibles are undervalued during recessions and
therefore we expect that total effect of recession will have positive effect on the expected bonds returns. Third we
construct the risk factor based on the sensitivity of convertible bonds estimated by 12- and 24-months rolling
regressions. We rank universe of convertible bonds by its sensitivity. The risk factor is a out-of-sample return of
long/short portfolio of convertible bonds where we take long position in the convertible bonds from top decile and
short position from low decile. The cumulative and monthly risk-adjusted return should produce out-performance
results in the backtesting relative to the index as in this way we would hedge against the demand stochastic
fluctuations. Finally we try to estimate the risk factor premium following Fama-Macbeth procedure.
The structure of the paper is the following Section 2 discusses relevant existing literature on the convertible
bonds. Section 3 describes data which are used for our empirical analysis. Section 4 discusses empirical
101

methodology and states the testable hypotheses. Section 5 shows empirical results of Granger causality tests
between hedge fund demand and convertible bonds returns. Section 6 summarizes our results .
2 Literature Review
Our paper aims at contributing to the academic research literature devoted to convertible bonds which is
relatively disperse in terms of topics and applications. One of the most popular themes in convertible bond research
is the existence of perceived mispricing anomaly in the convertible bond valuation. Our paper contributes to this
stream of literature by creating a new priced risk factor as a state variable in the pricing modeling of convertible
bonds and argues that ignoring this risk factor would add to mispricing differential. The fact of systematic
mispricing of convertibles has been well documented in the existing academic literature. A good survey of possible
explanations is provided in the paper by M. Ammann et al. (JBF 2003). For example, Calamos (2003) advocated the
underpricing anomality to the underestimation of the stock volatility. On the other hand, Lhabitant (2002) argued
that mispricing is driven by the complexity of convertible bonds valuation. The paper by Agarwal et al. (2007)
suggests that convertible bonds are mispriced due to illiquidity, small size issue, and again complexities of pricing.
Special attention in the academic literature was given to the performance of new issues, as convertibles are
more likely to be mispriced at issue than other asset classes due to its complexity. Both short-term and long-term
performance of new issues was of interest by researchers and practitioners. New issuance performance was studied
in the paper by Henderson (2005) who reported that convertibles are underpriced at issuance and found consequent
positive excess risk-adjusted returns. Association of increase in short selling of equity underlying near the
convertible bond issuance time with liquidity and efficiency was discussed in the paper by Choi et al (2005).
Mitchell et al. (2007) are first to mention the existence of forced selling risk in the convertible bonds valuation
by comparing mispricing of convertible bonds (relative to theoretical model) to returns of hedge funds with large
holdings of convertible bonds. They also conducted an analysis on the merger arbitrageurs activity and merger
targets during market crashes. This is the first step in estimating the hedge fund demand risk for convertible bonds
which leads to forced selling risk at market turning points. However relying on theoretical model might be
misleading due to the complexity of convertible valuation and possible model misspecification risk. Therefore we
prefer relying on the empirical estimation of sensitivities of convertible bonds towards hedge funds demand in
deriving the asset pricing implications.
Risk and return characteristics of convertible arbitrage hedge funds were empirically studied by Agarwal et al.
(2007). The authors constructed the following factor portfolios mimicking convertible arbitrage strategy: positive
carry, volatility arbitrage and credit arbitrage. They explained abnormal returns of convertible arbitrage hedge funds
as a liquidity premium. In addition, they constructed the supply demand misbalance factor which found to be
significant in explaining the cross-section of convertible arbitrage hedge funds returns. In our paper, the price
pressure factor is also based on the demand/supply misbalance however our objective is to show that this is the
important factor first and foremost for the convertible bond pricing and only as a consequence a convertible bond
investor will pay a price premium. This is the channel how the factor appears to be significant in risk models for a
fund holding long position in convertible bonds.
As price pressure factor can be thought as a liquidity factor, many papers on the liquidity measure and how liquidity
is important in portfolios risk models and funds' performance measurement models are relevant for our purposes. In
terms of liquidity risk factor, paper by Brunnermeier (2009) predicts that investors choose not to sell illiquid assets
and instead choose to sell liquid assets . The paper by Acharya & Pedersen (2005) predicts that investors are
unlikely to sell high liquidity beta assets and instead sells other assets in downturn. We also observe this behavior
while liquidity beta is substituted by sensitivity of convertible bonds to hedge fund demand and study how effect of
price pressure varies during financial crisis and recession periods.
Finally, the only study of convertible bond mutual funds, to the best of our knowledge, was conducted by Amman et
al. (2007) who constructed a set of factors driving the convertible bond fund performance. It includes stock factors,
bond factors, option factors and fund specific factors. The authors addressed the problem of time-varying betas in
their multi-factor regression models by employing rolling regressions and models with latent variables using
Kalman filtering. This type of models constitutes the performance measurement methodology for the convertible
102

ond fund. We use the setup of multi-factor models for convertible bonds in the most parsimonious case to create
control variables for alternative risk sources.
3 Data
The sample of convertible bonds represents constituents of the UBS Global Convertible Index (UCBIGLBL Index).
The data are at monthly frequency from October 2002 to September 2009. It is a broad-based index representing the
convertible market and is often used as a benchmark for global convertible funds. It is market capitalization
weighted total return index which was launched on 30 September 1998. However historical data for all its
constituents are available from 30 September 2002 onwards. The constituents of the index contain equity-linked
convertible bonds issues such as convertibles, exchangeables, mandatory issues, bonds with warrants and similar
products which must be convertible into a listed share. The index is devised by UBS and independently maintained
by MACE advisers which is supplier of convertible analysis tools and data.
Another empirical source comes from demand side represented by convertible arbitrage hedge funds. The data
source of the convertible arbitrage hedge funds is TASS Lipper hedge funds database. We combine both live and
graveyard hedge funds into our sample in order to reduce survivorship bias. Moreover we filtered hedge funds
selecting only those that reports in USD at monthly frequency. Convertible arbitrage is primary investment strategy
for such hedge funds. There is a total of 84 convertible arbitrage funds in the sample. For the aggregate
computations, we take official CSFB/Tremont Convertible Arbitrage hedge funds index.
For additional factors, we use various data sources. The equity factor is the market return in excess of risk free
rate where that the market includes all NYSE, AMEX, and NASDAQ stocks and risk free rate which is 1-month Tbill
rate. The data are from the K. French data library\footnote{Data are publicly available and regularly updated on
the following web site. Risk free rate is originally from Ibbotson Associates. Furthermore, the data on the term
spread and default spread are collected from the FRED database of the Federal Reserve Bank of St. Louis. We take
Moody's Seasoned Aaa and Baa Corporate Bond Yield, 1-Year GS1 and 10-year GS10 Treasury Constant Maturity
Rate. The credit risk or the default spread is measured by the difference between Baa and Aaa yields. The interest
rate risk or term spread is a difference between GS10 and GS1. The term spread captures variation in slope of the
yield curve. Variations in level of yield curve is captured by risk-free rate or short term T-bill yield approximated
by 3-months treasury TB3MS. Finally, month-end closing prices for the CBOE Volatility Index VIX index is
downloaded directly from the Chicago Board Option Exchange web page (www.cboe.com) and used as a measure
of the volatility factor. Finally, the trend-following hedge funds factors as in the paper Fung & Hsieh (2004) are
available from the D. Hsieh's data library.
For robustness check section, we will look at the short interest data of equity underlying. Unfortunately we are
not able to collect the short interest data for each individual stocks underlying the convertible bond issue. Therefore
we construct approximate measure on the short interest index available in Bloomberg. We consider the New York
Stock Exchange US Short Interest Mid-Month Index ( NYSINYSE Index ). The index represents the short interest
data as reported around the 15th of the month for the current month and is later revised around the 6th of the next
month. This index is the mid month short interest value for the New York Stock exchange, reported between the
18th and 25th of the month. We consider level and monthly percentage growth of the index as a potential variable
of interest in our study.
4 Testable hypotheses and methodology
In this section, we present the empirical methodology of the paper step by step. First we assess the impact of hedge
funds demand on the convertible bond expected return and formulate testable hypotheses. Second we estimate
sensitivities of convertible bonds towards hedge funds demand and construct portfolios of deciles based on the
sensitivities ranking. Thirdly, we create the risk factor as long/short portfolio of convertible bonds from extreme
deciles and test its importance for the pricing implications. Finally, we test this risk factor in the framework of risk
modeling for a fund that invests in convertible bonds.
103

First, we construct the multi-factor pricing model for convertible bonds in order to identify the common
sources of return and risk. The general form of multi-factor model for each security:
M
RCB
� � �i
Fi
i�1
where βi are factors loadings and Fi are the factors returns, and RCB is convertible bond return in excess of risk-free
rate. As a hybrid security, a convertible bond has variety of risk exposures coming from the equity risk, bond
structure and embedded option. Our goal is to keep the multi-factor model as parsimonious as possible and at the
same time to control for main systematic risk exposures. The equity risk factor is the stock market return in excess
of risk free rate. There are two main bond risk factors such as factors capturing credit risk (default spread between
Corporate bond yields) and interest rate risk (term spread between long-term and short-term treasury rates).
Moreover, option factor is measured by volatility risk captured by growth in first difference of CBOE VIX index:
Rt � � � �1MKTt
� � 2TERM
� � 3DEF
� � 4 fdVIX � ut
Three types of regression estimations are conducted: without dummy variable; with dummy variable on
recession period from January 2007 to June 2009; with dummy variable on post-Lehman period from August 2008
to March 2009. The OLS regression method is applied to the regressions for the monthly convertible index returns.
In addition, we run the analysis for individual convertible bonds using the fixed-effect type of panel data estimation
method.
Second, we construct the hedge funds demand measure as a ratio of total assets under management of
convertible arbitrage hedge funds normalized by the market capitalization of convertible bonds index. Price pressure
is primarily generated by hedge funds who invest in convertibles. We assume constant demand from other
convertible market participants. This fact can be implicitly confirmed by looking at the growth in assets under
management for convertible arbitrage hedge funds from our hedge funds sample versus market capitalization of
convertible bonds. We further assume that convertible arbitrage hedge funds are main participants in trading of
convertible bonds. Total assets under management are proportional to the hedge fund demand. Normalization to
market capitalization of convertible bonds is required to insure consistency of the demand with total convertibles
issues. Normalization by convertible bond index is also related to supply/demand misbalance factor for convertible
arbitrage hedge funds used in the paper by Agarwal & Naik (2007). We would like to show that this is not the
misbalance between supply and demand but rather exogenous shocks influencing the demand equation and not the
supply. The drawback of total assets under management as a demand measure is that it excludes leverage
considerations. However the TASS database does not have monthly leverage indicator for the hedge fund. Leverage
is only constant fund characteristic and therefore we cannot use it properly to track the dynamics of this variable.
For robustness check, we can add constant leverage ratio to AUM measure (average leverage as defined in the TASS
database) to enhance the demand level.
Furthermore, we analyze the causality relationship between convertible bond returns and hedge fund demand.
The Granger-test is conducted for both returns of convertible bonds index and for residuals from the conventional
multi-factor model. On the one hand, the reason for the causality testing is to avoid potential endogeneity bias in all
consecutive regressions. On the other hand, the causality relationship is not trivial and impossible to detect without
formal testing procedure. Under the null hypothesis, hedge fund demand Granger-causes convertible bond returns
such that convertible arbitrage hedge funds is the driving force in the convertible bonds market. We expect positive
sign of the causality. However the magnitude of causality might differ across bull and bear period of market or
during financial crisis periods and even sign might change during market turning points. We assume that this is
likely to happen either during recession or during financial crisis. Therefore we test the hypothesis including two
dummies on 2007-2009 recession and also for post-Lehman months to see how convertible reacted to those market
events. Alternative hypothesis is based on the fact that assets are chasing returns. Growth in hedge funds demand is
explained by attractive investment opportunities in convertible universe and even arbitrage hedge funds earns most
of their performance on the long positions (in our on convertibles rather than on shorting stock underlying).
Therefore convertible bond return should Granger-cause hedge fund demand. Demand is lower whenever historical
returns of convertibles are lower.
Null Hypothesis 1: Convertible arbitrage hedge funds demand Granger-causes convertible bonds return.
Alternative Hypothesis 1: Convertible bonds return Granger-causes convertible arbitrage Hedge funds demand.
104

Thirdly, we estimate sensitivities of individual convertible bonds towards hedge fund demand using
conventional factors as control variables. We expect on average negative values of estimated sensitivities
highlighting the price pressure effect from hedge funds.
Null Hypothesis 2: Hedge funds demand negatively and statistically significant affect monthly returns of convertible
bonds.
Alternative Hypothesis 2: There is no statistically significant relationship between hedge fund demand convertible
bond monthly returns.
If it is determined that demand side has negative impact on the expected convertible bond returns, an investor
in convertible bond should hedge against demand fluctuation in his optimal convertible bond portfolio. Sensitivity
is a liquidity indicator for which investor is expected to pay a premium. Whether it is priced or not is the goal of the
next step in our empirical study. We estimate the rolling regression for sensitivities on a fixed-size sample window
of 24 months with 1 month rebalancing in order to construct the time-series of sensitivities of convertible bonds. We
rank the universe of convertible bonds into deciles according to values of sensitivities. Afterwards, we form the
long/short portfolio by buying bonds with high sensitivity (top decile) and selling those with low sensitivity (low
decile). We compute the out-of-sample returns of this portfolio which is effectively the factor mimicking portfolio.
We further test whether this is priced risk factor for convertible bond universe estimating the risk premium. We
conduct the regression analysis using recession and post-Lehman dummy variables to track the impact of financial
crisis on the risk premium. During crisis, price pressure factor becomes forced selling risk factor and the factor
return differential is more pronounced during crisis.
Null Hypothesis 3: Price Pressure risk factor is priced risk factor (significant risk premium) for a convertible bonds
universe.
Alternative Hypothesis 3: There is no statistically significant risk premium between hedge fund demand risk factor
and a convertible bond fund.
If the price pressure is a priced risk factor for convertible bond, it should appear as a portion of risk modeling
for any convertible bond investor including convertible arbitrage hedge funds. Absent of pure convertible bond
mutual fund database, we test whether the factor appears to be significant for convertible arbitrage hedge funds
however we admit this testing procedure should be better applied for long only convertible bond mutual funds as
they are investors who can perceive the hedge funds price pressure as a systematic risk factor. Adding new
systematic risk factor can also contribute to improvement of multi-factor performance measurement model for a
convertible bond fund. We formulate the following hypotheses:
Null Hypothesis 4: A convertible bond fund has significant risk exposure to price pressure risk factor (betas) and
has lower alpha when accounting for missing risk factor.
Alternative Hypothesis 4: There is no statistically significant relationship between convertible bond fund returns and
price pressure risk factor.
In order to evaluate the performance of the convertible arbitrage hedge fund, we add the price pressure risk
factor into existing benchmark factor model for a . We run the analysis across different types of benchmark model.
First we use one-factor model where the only systematic risk factor is the market which is the Credit Suisse/Tremont
convertible arbitrage hedge fund index monthly return in this case. Second, we apply the six-factor model by
Hasanhodzic & Lo (2004) where factors correspond to different asset classes. Namely, the following factors are
employed: S&P 500 total return, the USD Index return, the Goldman Sachs Commodity Index (GSCI) total return,
the Moody's Corporate Aaa Bond Index return, the spread between the moody US Aggregate Long Credit BAA
Bond Index and the Lehman Treasury Long Index (default spread), and the first-difference of the CBOE Volatility
Index (VIX). Finally, we use original eight-factor risk model as in the paper by Fung & Hsieh (2004) that capture
the risk of well-diversified hedge fund portfolios. The first three factors are trend-following risk factors which were
constructed in the paper by Fung & Hsieh (2004): PTFSBD - Bond Trend-Following Factor, Return of PTFS Bond
lookback straddle; PTFSFX - Currency Trend-Following Factor, Return of PTFS Currency Lookback straddle; and
PTFSCOM - Commodity Trend-Following Factor, Return of PTFS Commodity Lookback Straddle. The next two
factors are equity-oriented risk factors: S&P500 - Equity Market Factor: the SP500 index monthly total return; and
Size - The Size Spread Factor: Russell 2000 index monthly total return minus SP500 monthly total return. The last
two factors are bond-oriented risk factors: Bond - The Bond Market Factor: the monthly change in the 10-year
105

treasury constant maturity yield (month end-to-month end), available at the St.Loius FED Economic Data database
(FRED); and Credit - The Credit Spread Factor, the monthly change in the Moody's Baa yield less 10-year treasury
constant maturity yield (month end-to-month end), available at the St.Loius FED Economic Data database (FRED).
5 Empirical Results
5.1 Conventional Factor Model
Table 1 reports time-series estimation of multi-factor model for the convertible bond index returns. OLS coefficients
are displayed in the table along with their t-statistics. Three types of regression estimations are conducted: without
dummy variable (Model I); with dummy variable on recession period from January 2007 to June 2009 (Model II);
with dummy variable on post-Lehman period from August 2008 to March 2009 (Model III). First column shows that
all factors except term spread are significant. Convertible bond index has high equity market beta of 0.91 which can
be explained by high average delta of convertibles. Default risk has negative effect on expected bond returns.
Moreover, as with any option, high volatility leads to higher expected return of the option embedded in convertible
structure. Adjusted R-squared coefficient of the conventional model is about 0.60. Adding the recession dummy
does not change dramatically coefficients but the dummy variable is statistically significant with positive estimated
coefficient value. That means that convertible bonds have positive expected return and investment attractiveness
during recession which can be explained by downside protection because of the bond floor. This is consistent with
idea that convertible bonds are providing balanced investment opportunities hedging downside risk and still bearing
equity upside. Risk aversion of investors is rising during recession and they are likely to switch to bond-type
investment. Recession dummy even increased the adjusted R-squared coefficient. However post-Lehman dummy
variable, though being defined as in the middle of recession period, has significantly negative coefficient. We
explain this negative impact by the forced-selling risk of convertible that indeed happened in October 2008 and
partially in March 2009. This was the period of liquidity shortage and investor panicking. Back that time
convertibles were sold off massively which had strong impact on the convertible bond returns.
Model I Model II Model III
const 4,85 6,85 1,35
(4,95) (6,72) (1,46)
emkt 0,91 1,03 0,39
(7,29) (8,77) (3,11)
term -0,15 -0,17 -0,24
(-0,44) (-0,52) (-0,87)
def -3,48 -6,29 0,73
(-4,72) (-6,54) (0,87)
vix gr 7,51 8,50 3,96
(2,19) (2,71) (1,44)
Recession
NaN
5,83
(4,08)
NaN
Post-Lehmann
NaN NaN
-16,51
(-6,96)
Adj R2 0,59 0,66 0,75
Table 1: Conventional Model for a convertible bond index. OLS estimates.
Table 2 reports similar regression analysis for individual bonds by applying the fixed-effect panel data
estimation technique. For individual bonds, VIX factor is insignificant and on the other hand term spread happened
to be statistically significant. The equity market beta has lower value of about 0.65. The dummy variables are
statistically significant however signs of estimated coefficients are opposite. During recession, expected returns on
individual bonds are negative. This is probably due many present high delta convertibles which carry high equity
market risk. Coefficient on post-Lehman dummy variable is highly negative that shows convertible bonds were
quite undervalued and under price pressure during this period.
106

Model I Model II Model III
emkt 0,65 0,64 0,63
(39,55) (37,63) (33,71)
term -0,35 -0,27 -0,37
(-4,86) (-3,60) (-5,08)
def 1,10 1,51 1,52
(7,78) (9,17) (7,62)
vix gr -1,19 -1,06 -1,02
(-3,24) (-2,68) (-2,56)
Recession
NaN
-1,28
(-5,04)
NaN
Post-Lehmann
NaN NaN
-1,19
(-3,02)
Adj R2 0,14 0,14 0,14
Table 2: Conventional Model for the individual convertible bonds. Fixed-effect panel data estimation method.
Before starting adding the hedge fund demand into regression analysis, we investigate the Granger-causality
relationship between hedge fund demand and convertible bond returns in order to avoid endogeneity bias issues and
detect true causality relationship. The results of the Granger test are reported in the Table 3. We carry four model
specifications. First method is directly to test the causality between convertible bond return index and hedge fund
demand. This is the Model I in the Table. Second method is to establish causality between hedge fund demand and
residuals from the conventional multi-factor model for convertible bond index. This is the Model II in the Table.
Model III is the Grange-cause test of OLS residual from the conventional model with recession dummy of
convertible bond index against hedge fund demand. Model IV is the Grange-cause test of OLS residual from the
conventional model with post-Lehman dummy of convertible bond index against hedge fund demand. The
consideration of residual instead of bond return is important to derive the causality relationship free of other
systematic risk factors. F,G are values of F-statistic. Fc and Gc are the critical value from the F-distribution. The
rule is such that if F>Fc then we cannot reject null hypothesis that convertible bonds causes hedge funds demand;
if G>Gc then we cannot reject null hypothesis that hedge fund demand causes convertible bond returns. The test is
conducted for 5 lags and at 5 percent level of significance.
Model I Model II Model III Model IV
F 1,71 2,42 0,25 1,55
F crit value 3,96 3,11 3,96 3,96
G 0,90 5,26 6,18 4,52
G crit value 3,96 3,11 3,96 3,96
Table 3: Granger Causality test between Convertible bond Index and hedge fund demand generated by assets under management of convertible
arbitrage hedge funds normalized by the convertible bond index market capitalization.
In no model specification, we found convertible bonds cause the hedge fund demand fluctuation. In case of
Model I, there is causality in other direction either which can be explained by interaction with other factors.
However whenever we control for the additional factors, we cannot reject the null hypothesis that hedge fund
demand causes convertible bond returns. This test assures us that endogeneity bias should not be an issue whenever
we study the effect of hedge fund demand on the convertible bond returns.
5.2 Price Pressure Factor
As we establish some impact of convertible arbitrage hedge fund on the convertible bond returns, we would like to
find if the price premium exists for convertible bonds with different sensitivities towards demand and moreover
construct the risk factor as a top minus bottom decile portfolio of convertible bonds sorted by their sensitivity.
First for each individual convertible bond issue, we estimate its sensitivity to hedge fund demand on the sample
window of either 12 months or 24 months. Sensitivity is estimated within different model misspecification. First,
107

we consider univariate framework. Second, we add convertible bond index capturing the systematic risk factor.
Finally, as model III, we use our conventional model. Results are reported in Table 4.
Model Model II Model III Model IV Model V
Cbind 0,33
(28,61)
NaN NaN NaN
emkt 0,65 0,64 0,64
(40,83) (39,06) (37,62)
def -0,29 -0,19 -0,32
(-3,72) (-2,38) (-4,01)
term 1,04 1,43 1,35
(7,04) (8,42) (6,10)
vix -1,03 -0,77 -0,79
(-3,35) (-2,25) (-2,12)
HF demand -0,41 -0,15 -0,07 -0,09 -0,05
(-12,20) (-4,38) (-2,10) (-2,64) (-1,44)
D1recession -1,24
(-4,96)
NaN
D2Lehmann
NaN NaN NaN
-0,78
(-1,97)
AdjR2 0,01 0,06 0,14 0,14 0,14
Table 4: OLS estimates of the sensitivities of individual convertible bonds toward hedge fund demand.
As a portfolio formation, we use next month out-of-sample returns. We rank our convertible bonds according
to their sensitivities and form the equally-weighted deciles portfolios. Then we compute the next month out-ofsample
return of this portfolio. Differential between returns of such portfolio will prove the existence of price
premium for convertible bond factor and that it is priced in equilibrium.
We finally select bonds with top 10 and bottom 10 statistically significant value of estimated sensitivity (at 5
percent level). The portfolio consists of equally-weighted long position in top 10 and short position in bottom 10
bonds. We compute the out-of-sample return of this portfolio at next month. We also report alphas of top and low
decile portfolio and show the alphas of factor mimicking portfolio as a difference between top and bottom alphas
(for the estimation window of 12 month). Positive alphas indicates the existence of risk premium.
Window 12 Model I Top Model I Low AlphaTop-Low
Alpha 0,13
-0,19
0,31
(0,20)
(-0,19)
Emkt 0,57
0,75
(6,59)
(5,63)
Term 0,07
-0,42
(0,29)
(-1,12)
Def -0,02
0,50
(-0,04)
(0,67)
Vix -1,53
-5,21
(-0,65)
(-1,45)
AdjR2 0,56 0,53
Window 12 Model II Top Model II Low Alpha Top-Low
Alpha 0,17
-0,86
1,03
(0,23)
(-0,75)
emkt 0,57
0,71
(6,37)
(5,18)
term 0,07
-0,36
(0,27)
(-0,95)
def -0,08
1,44
(-0,12)
(1,33)
108

vix -1,50
-5,53
(-0,64)
(-1,54)
Dummy 0,14
-2,02
(0,13)
(-1,20)
AdjR2 0,55 0,53
Window12 Model III Top Model III Low Alpha Top-Low
Alpha 0,37
-0,80
1,17
(0,48)
(-0,67)
emkt 0,61
0,64
(5,39)
(3,72)
term 0,07
-0,43
(0,30)
(-1,12)
def -0,86
2,19
(-1,17)
(1,45)
vix -4,07
-7,91
(-1,81)
(-1,72)
Dummy 1,76
-3,70
(0,95)
(-0,97)
AdjR2 0,71 0,48
Table 5: Performance of the decile portfolios sorted by the sensitivities.
For future research and estimation steps, we estimate the value of risk premium. and test the price pressure risk
factors as a missing risk factor for convertible arbitrage hedge funds. Moreover we investigate differential of price
pressure for different types of convertibles with different greeks.
6 Summary
This paper sheds the light on the previously unnoticed effect of hedge funds on the asset pricing of their
investment instruments. Demand fluctuations from the convertible arbitrage hedge funds generate a price
pressure risk on convertible bonds pricing. This risk is materialized into forced selling risk factor during market
turbulence and liquidity. We aim at constructing the price pressure risk factor based on the sensitivity of
individual convertible bonds to the demand of hedge funds through long/short factor mimicking portfolio and
showing its impact on the convertible pricing and on a risk models for a convertible bond portfolio. In
construction of the factor we rely on information about the total assets under management of convertible
arbitrage hedge funds. We test the significance of price pressure as a priced risk factor for convertible bonds.
Further analysis is required to estimate corresponding risk premium and identify price pressure as a missing risk
factor for pricing models of convertible arbitrage hedge funds.
7 References
Ammann, M., Kind, A. & Wilde, C. (2003). Are convertible bonds underpriced? An analysis of the French market.
Journal of Banking and Finance, 27,635-653.
Ammann, M.,Kind, A. & Seiz, R. (2007). What Drives the Performance of Convertible-Bond Funds. Working Paper
Agarwal, V., & Naik, N. (2004). Risk and Portfolio Decisions Involving Hedge Funds. The Review of Financial
Studies, 17, 63-98.
Fung, W. & Hsieh, D. (2004). Hedge Fund Benchmarks: A Risk Based Approach. Financial Analyst Journal, 60,
65-80.
Hasanhodzic, J.& Lo, A. (2004). Can hedge-fund returns be replicated? The linear case. Journal of Investment
Management, 2, 5-41
109

THE PRICING OF EQUITY-LINKED CONTINGENT CLAIMS UNDER A LOGNORMAL
SHORT RATE DYNAMICS
Rosa Cocozza and Antonio De Simone,
University of Napoli “Federico II”, Italy
Email: rosa.cocozza@unina.it, a.desimone@unina.it,
www.docenti.unina.it/rosa.cocozza
Abstract. We propose a numerical procedure for the pricing of financial contracts whose contingent claims are exposed to two sources of
risk: the stock price and the short interest rate. More precisely, in our pricing framework we assume that the stock price dynamics is
described by the Cox, Ross Rubinstein (CRR, 1979) binomial model under a stochastic risk free rate, whose dynamics evolves over time
accordingly to the Black, Derman and Toy (BDT, 1990) one-factor model. To this aim, we set the hypothesis that the instantaneous
correlation between the trajectories of the future stock price (conditional on the current value of the short rate) and of the future short rate is
zero. We then apply the resulting stock price dynamics to evaluate the price of a simple contract, i.e. of a stock option. Finally, we compare
the derived price to the price of the same option under different pricing models, as the traditional Black and Scholes (1973) model. We
expect that, the difference in the two prices is not sensibly large. We conclude showing in which cases it should be helpful to adopt the
described model for pricing purposes.
Keywords: option pricing, stochastic short rate model, binomial tree
JEL classification: C63, C65, G13
1 Introduction
In the modern option pricing theory many attempts have been accomplished in order to release some of the
traditional assumptions of the Black and Scholes (1973) model and, in general, to develop a pricing framework
depending not only on the solely underlying dynamics . Very distinguished in this field are the models allowing for
stochastic interest rate, as suggested for the first time by Merton (1973). Afterward, Brennan and Schwartz (1980)
proposed a stochastic interest rate model to evaluate the price of convertible bonds. Within this context, if the
conversion date coincides with the bond maturity, the future value of the bond is certain and no assumption on the
interest rate dynamics is necessary. In this case the issues faced in evaluating the convertible are not sensibly
different from those arising in determining the market value of an equity-linked endowment policy. On the contrary,
if the conversion occurs before the maturity, some assumptions about the future dynamics of the term structure are
necessary because, in this case, the future value of the bond is a random variable depending on the unknown level of
the interest rates at the conversion date.
Even in the insurance segment the specification of the interest rate dynamics is very helpful. Recently, life
insurance companies issued policies that typically combine a guaranteed return with a contingent spread on a
reference asset return paid out to the policyholder under particular conditions. This is the case of participating life
insurance policies allowing policyholders to gain a defined amount and at the same time to participate to the
eventual additional profit margin of the insurance company according to a predefined participation rate. As showed
elsewhere (Cocozza and Orlando, 2007; Cocozza et al. 2011), these contracts embed an option that exposes the
issuer to two sources of financial risk: the return on the reference asset and the appropriate discount factor. On the
contrary, if we consider an equity-linked endowment policy, the equilibrium price, as Brennan and Schwartz (1976)
showed in their seminal paper, is equal to the sum of the present value of a zero coupon bond and those of an
immediately exercisable call option on the reference asset or, alternatively, to the present value of the reference asset
plus that of an immediately exercisable put option on the same reference asset. As a matter of fact, such a kind of
contracts is not an insurance policy but a proper financial product issued not only by the life insurance companies
but also by other financial institutions.
In this article we propose an innovative numerical procedure for the pricing of financial contracts whose
contingent claims are exposed to two risk sources: the stock price and the interest rate. More precisely, in our pricing
framework we assume that the stock price dynamics is described by the Cox, Ross and Rubinstein (1979) binomial
model (CRR) under a stochastic risk free rate, whose dynamics evolves over time accordingly to the Black, Derman
and Toy (1990) one-factor model (BDT). The BDT model avoids some drawbacks that typically affect equilibrium
model of the term structure, such as negative spot interest rates. At the same time it offers the relevant opportunity to
efficiently calibrate the risk factor trajectories, preventing from the adoption of parameters not allowing the
endogenous reproduction the observed term structure. Last but not least, the conjoint adoption of CRR and BDT
110

models may sensibly reduce the computational efforts in the estimation of the parameters by adopting implied
volatility measures.
The paper is structured as follows. In section 2 main contributions to option pricing theory in a stochastic
interest rate framework are reported together with a brief illustration of the BDT and the CRR models. Section 3
reports the main assumptions and the mechanics of the numerical procedure adopted to determine the price of a
plain vanilla call option. Section 4 shows some numerical examples while section 5 concludes the paper with main
final remarks.
2 Option pricing models with stochastic interest rate
As stated, a stochastic interest rate model for option pricing was proposed for the first time by Merton (1973),
where the Gaussian process was adopted to describe the continuous-time short rate dynamics. The adoption of a
Gaussian process was very common in the ’80s and in the early ‘90s before the advent of the lognormal term
structure models. A discrete time dynamics for short rate process equivalent to those adopted by Merton was
subsequently discussed by Ho and Lee (1986), while other option pricing formulae under Gaussian interest rate were
introduced by Rabinovitch (1989) and Amin and Jarrow (1992).
The success of the Gaussian models of the term structure relies on the mathematical tractability and thus on the
possibility of obtaining closed formulas and solutions for the price of stock and bond options. In fact, the Gaussian
process was for the first time adopted to derive the price of bond options by Vasicek (1977). Furthermore, the
calibration of the Gaussian models does not require particularly demanding computational effort.
Although the Gaussian models have been very successful for research purposes, some relevant inner drawbacks
prevented their diffusion among the practitioners, as for example the possibility for the interest rate trajectories to
assume negative values. In response, other equilibrium models for the interest rate term structure have been
developed. One of this is the well-known Cox, Ingersoll and Ross (1985) model (CIR), where the interest rate
dynamics is described by a square root mean reversion process that, under the Feller condition (Feller, 1951), does
not allow the interest rate to become negative. CIR dynamics has subsequently been adopted also to describe the
stochastic short rate framework for pricing stock option (Kunitomo and Kim, 1999 and 2001) and for pricing
endowment policies, with an asset value guarantee, where the benefit is linked to fixed income securities (Bacinello
and Ortu, 1996).
However, as showed elsewhere (De Simone, 2010), equilibrium models are in general not able to ensure an
efficient calibration of the interest rate dynamics, because they are based on a limited number of parameters, in
general unable to guarantee an acceptable fitting of the model prices to market prices. Moreover, a satisfactory
calibration of the model is sometimes a hard task, because many equilibrium models rely on an instantaneous
interest rate (spot or forward) that, in general, is not directly observable on the market. The relevance of this
problem increased over time, especially after the diffusion of standard market practices of pricing derivatives within
the Black and Scholes environment (Black and Scholes, 1973; Black, 1976).
As mentioned, limitations of the equilibrium models may be overcome by market models, as for example the
BDT model and the Libor Market Model (Brace et al. 1997). A particular characteristic of both models is the
assumption of lognormal interest rate dynamics, even if this hypothesis applies only asymptotically for the BDT
model. Such feature allows for a satisfactory calibration by adopting implied volatility measures according to the
standard market practices. However, opposite to many equilibrium models, market models do not in general allow to
obtain closed price formulas so that the price of interest rates derivatives has to be evaluated numerically. Between
the two mentioned models, we choose the BDT because of its simplified approach in pricing derivatives. The BDT
model allows to obtain a binomial tree for the dynamics of the Libor rate by adopting, as input data, the term
structure of interest rates and of the corresponding volatilities. An exhaustive explanation of the procedure adopted
for the construction of the tree is reported in Neftci (2008). Figure 1 shows an hypothetical BDT tree for the 12monts
spot Libor rate L(t, s), where s – t = 12 months for each t, s.
At time t = 0 the 12-months spot Libor rate L(0,1) is directly observable and therefore not stochastic. After one
year, at time t = 1, the following 12-months Libor rate L(1,2) can go up to the level L(1,2)u or down to the level
L(1,2)d. Similarly, at time t = 2 the Libor rate L(1,2)j, in the state of the world j=u, d, may go up or down with equal
risk neutral probability. We finally notice that since L(t, s)ud=L(t, s)du the BDT tree is recombining.
In the next section we show how the information from the BDT tree is employed to describe the dynamics of
the interest rate adopted as risk free rate in the CRR model. We choose such models for two main reasons: (1) both
the models are based on a binomial tree and (2) both the risk factors are lognormal in the limit.
111

Figure 1. The BDT tree of the 12-months Libor rate L(t, s).
An important property of the implied stock tree is that the rate at which the stock price increases/decreases (u/d)
j
is constant over time, so that ST � St
j , j=u,d. As shown by Cox et al. (1979), the risk neutral probability p is:
m � d
p �
u � d
where m=1+L(t, T), and where u∙d=1 because also the CRR tree is recombining. Moreover, we remark that the
discrete time dynamics approximates the Black and Scholes dynamics of the stock price, according to the following
stochastic differential equation (SDE):
St � �St � ��
CRR St
��
t
where St is the stock price at time t, μ is the drift of the process Δ is the time distance between two observations,
� CRR is the instantaneous volatility coefficient and � t can be interpreted as the outcome of a binomial random
variable under the natural probability. Notice that as Δ→0, � t tends in distribution to a standard normal random
� (1)
variable so that t � � can be interpreted as a Wiener increment. As a result, the final stock price tends in distribution
to a lognormal random variable. This property entails that the adoption of implied volatility measures for pricing
equity linked contingent claims ensures a satisfactory calibration of the model to the market data. Of course, as the
drift change, the expected value of the stock price changes too because of the subsequently modification of the
probability space. On the contrary, the levels of the future stock prices in each state of the world only depend on the
current stock price and on the volatility parameter � CRR . In the next section we adopt the Libor rate as risk free rate
in evaluating equity linked contingent claims. Despite the Libor rate refers to banks AA or AA- rated, we consider it
as a proxy of the risk free rate. In section 5 we study how to release this assumption and a possible solution to this
issue is proposed.
3 The procedure
In the majority of cases, when there is a two risk factor pricing model we have to ascertain:
1. the dynamics according to which the two factors evolve;
2. the measure of the correlation between the two dynamics;
3. the estimate of relevant parameters under risk neutral environment.
The three tasks can be extremely difficult to perform properly. As a consequence, a certain level of approximation is
often required. Even the simulation based on copula function, not always available with reference to the distribution
under observation, can be very questioning with reference to the choice of an appropriate correlation measure. In
this order of ideas, we developed a numerical procedure to get the arbitrage free price of a European call option in a
stochastic short rate framework. To begin with, we notice that the numerical procedure adopts the results from the
BDT and from the CRR models, whose assumptions hold also for our model.
112

3.1 Main assumptions
According to the stated environment (section 2), the arbitrage free dynamics of the stock price is described by
equation 1. Therefore the risk neutral dynamics of the stock price is defined as follows:
p
� St � r�t
�S
t � ��
CRR St
��
t (2)
where r(t) is the instantaneous risk free interest rate at time t under the corresponding appropriate risk neutral
probability p. The crucial point in our approach is that the instantaneous risk free rate is piecewise constant and
evolves over time according to the BDT dynamics, since the spot rate r(t) is not a random variable unless the tenor
(δ) of the rate matures.
In other words, once we choose a term structure (in our examples the Libor rate term structure), the variability
of the rate can be observed in practice according to the ‘nodes’ of the term structure. Therefore, if the rate is, as in
our case, the Libor, the variability time interval goes from 1 week up to 12 months and coincides with the tenor the
chosen rate. Therefore, once we adopted the 12 months rate, the dynamics over time on a discrete time interval is
equal to 12 months. This accounts for a piecewise constant dynamics. The length of the time interval of the process
describing the rate dynamics is therefore set according to the relevant node of the term structure. Theoretically, any
node could be used, but if we decide to go for a BDT application we need also volatility data. Since not all the nodes
have the same liquidity, the significance of the corresponding implied volatility is not the same across the term
structure. We are therefore forced towards those nodes showing the maximum liquidity, since this guarantees the
more efficient measure of implied volatility. If we assume (or better observe) that the most liquid node is 12 months,
we will adopt a BDT model on a year basis. As a consequence on a certain time horizon (longer than one year in the
case under estimation), we will have an array of one year rates defining a corresponding set of probability spaces
which can be used for evaluating the stock price dynamics. As a consequence, the evaluating numeraire is piecewise
constant and in a sense “rolls over time” according to the term structure tenor.Therefore what can be regarded as a
random variable at the evaluation date is the δ-rate. As a consequence, the instantaneous risk free rate r(t) and the δ-
Libor rate L(t, T), with δ = T – t, are connected as follows:
1
exp��
r�t
�n��
� , t

Figure 2. The joint PDF of the short rate L(t, T) and of the stock price St in the case of zero correlation.
Now, let us consider the case in which the short rate increases to L(1,2)u and also the stock price increases to
u
S1 . Since the two events are independent, the probability that they occur contemporaneously is:
u
u
�S L�1,
2�u
| S , L�0,
1��
� Pr ob S | S , L�0,
1�
� �* Pr ob L�1,
2�u
| S , L�0,
1�
Pr ob 1 � 0
1 0
� 0
1
� � p1
* . This procedure may be repeated
2
for each possible couple of the short rate and stock price at time t=1.
Now, let us consider the state of the world where stock price is u
S1 and the Libor rate is L(1,2)u. At the
successive time step, t=2, the stock price may again increase to
2
S 2 S0u
uu � with risk neutral marginal probability
p2 or decrease to S 2 S0ud
S0
ud
u
m2
� d
� � with probability 1� p2
, with p2
� and m L u
u � d
u
2 � 1�
( 1,
2)
. At the same
time, the Libor rate may increase to L(2,3)uu or decrease to L(2,3)ud with equal probability. The probability that
both the stock price and the Libor rate increase for two consecutive times is therefore:
uu
u
1 1
Pr ob�S
2 � L�2,
3�uu
| S1
, L�1,
2�u
��
p1
* p2
* * . We repeat this procedure until, at the end of the second time
2 2
interval, all the 16 final states of the world and their respective probabilities are available, as described in table 1.
State of the
world
Libor rate Stock price Probability
1 L(2,3)uu uu
S 2
2 L(2,3)ud uu
S 2
3 L(2,3)du uu
S 2
4 L(2,3)dd uu
S 2
5 L(2,3)uu ud
S 2
6 L(2,3)ud ud
S 2
7 L(2,3)du ud
S 2
8 L(2,3)dd ud
S 2
9 L(2,3)uu du
S 2
10 L(2,3)ud du
S 2
11 L(2,3)du du
S 2
114
h1
h2
h3
h4
h5
h6
h7
h8
h9
h10
h11

12 L(2,3)dd du
S 2
13 L(2,3)uu dd
S 2
14 L(2,3)ud dd
S 2
15 L(2,3)du dd
S 2
16 L(2,3)dd dd
S 2
Table 1: An example of the discrete joint probability density function of the stock price and of the Libor rate
The current stock price S0 and the future stock price S 2 are thus linked as follows:
h � S 2 � 1
S0
� E0
� �
�1�
L�1,
2��
1�
L�0,
1�
(4)
where h is the joint probability measure associated to the future values of the stock price and of the Libor rate and
h
E0 denotes the expected value under the probability measure p conditional on the information available at the
valuation date t=0.
Equation (4) states that the current stock price S0 is the expected value of the future stock price S 2 discounted
at an appropriate stochastic interest rate. We notice that the discount factor is only partly included in the expectation
operator because, at time t=0, the current Libor rate L(0,1) is known while the Libor rate that will run during the
period from 1 to 2 is a random variable. Since the discount factor and the stock price are independent random
variables, it can be shown that:
� �
� � ��2
p
E0
S 2
1�
L 0,
2
p q � 1 � 1
S0
� E0
�S 2 �E0
� � �
(5)
�1�
L�1,
2��
1�
L�0,
1�
Equation (5) implies a very important property: the probability measure h, defined as the joint probability
associated to the future values of the stock price and of the Libor rate, is such that the stock price is a martingale.
Once the joint PDF of the final stock price and of the Libor rate is obtained, it is quite a simple task to determine the
price C0 of a European call option at time t=0, with strike price X:
�
h
C0
� E0
�
��
� �S 2 � X � � 1
�
1�
L�1,
2�
1�
L�0,
1�
��
4 Numerical examples
(6)
Equation 6 implies that the price of a call option is approximately equal to the price of the same option
calculated by means of the CRR model. In other words, the model here developed can be regarded as an extension
of the implied stock tree model on a roll-over basis. However, this could not be the case if we release the hypothesis
of zero correlation between interest rate and stock price, because in the original CRR model the interest rate is not
stochastic and therefore there is no reason for any relationship between the rate and the stock price. On the contrary,
a change of the term structure of interest rates produces a difference in the price of the call option, since that change
can be regarded as the adoption of a different parameter. Therefore, there is a certain interest in evaluating pricing
differences emerging from the implementation of one model against the other. To this aim it is necessary to get the
joint PDF of the stock price and of the interest rate.
Let us consider the simple framework of the previous section where n=2, δ=1. We set the stock price equal to
100 and its implied annual volatility to 20%. Finally, we assume that the spot Libor rate is 1% for the first year and
2% for the second year and 3% for the third year, while the term structure of the volatility is equal to 20% and 30%
for the second and the third year respectively. All the parameters are reported in table 2.
115
h12
h13
h14
h15
h16

Parameters
Time (years)
0 1 2
S0 100 - -
� T �
L 0, 1% 2% 3%
� CRR - 20% 20%
� BDT - 20% 30%
Table 2: Parameters adopted for the calibration of the model.
Two remarks are necessary. Firstly, we notice that the Libor rate for maturities over 12 months is not directly
observable on the market. Those values can however be obtained from the swap curve on the Libor rate by means of
bootstrap technique (details in Hull, 2009). Secondly, we notice that to the aim of pricing a plain vanilla call option
it is not necessary to specify the 3-year interest rate. We decide however to consider it to show the final joint PDF of
the interest rate and stock price. Such PDF may be adopted to compute the price of financial products whose value
depends contemporaneously on the level of stock price and interest rate at the maturity, as for example convertible
bonds.
Figure 3a and 3b show respectively the BDT interest rate tree and the binomial stock tree (with marginal
probabilities) calculated adopting the parameters shown in table 2.
Figure 3a. Example of the BDT tree Figure 3b. Example of the binomial stock tree
Given the values in figure 3a and 3b, it is a simple task to determine the joint PDF of the interest rate and of the
stock tree. For example, the probability that the stock price will be 100 at the end of the 2 year jointly with an
interest rate level of 9.3% is 0.25*0.4998.
We remark that, if we set the hypothesis of zero correlation, the price of the call option calculated by means of
the procedure exposed in section 3 coincide with the price of the same option calculated by means of the CRR
model in the case where the term structure of the interest rate is not flat. More precisely, the risk free rate for the first
year coincides with the corresponding spot rate while the risk free rate for the second year coincides with the
corresponding forward rate. In this way we are able to incorporate market expectations (at the valuation date) on the
future interest rates in the pricing of option.
To show how interest rate expectations affect the stock option pricing, we compare the price of a call option,
derived in our framework, to the Black and Scholes price. We decide to adopt the Black and Scholes model for the
comparison firstly, because the CRR price tends to the Black and Scholes price as Δ→0, and secondly, because the
Black and Scholes price is not able to capture the expectations on the future interest rates.
116

Table 3 reports the differences (in percentage) between the price of a two year ATM vanilla call option
calculated by means of our procedure (P) and the price of the same option calculated by means of the Black and
Scholes (1973) model (Bls), according to the following formula:
P � Bls
* 100 .
Bls
Such differences are calculated for different term structures of the interest rates. The other parameters adopted
to evaluate the price differences are those reported in table 2 but this time we consider, for each year, a higher
number of steps for the stock tree. More precisely, since the stock exchange is open about 254 days per year, we
thus set n=508 (Δ=1/254).
We thus notice that if the term structure is flat (see the numbers on the diagonal of table 3), the percentage
difference with the Black and Scholes formula is quite negligible, from .03% to .05%. However, as expected, such
differences tend to increase as the difference between the interest rates for the two maturities increases. If, on the
contrary, L(0,1)L(0,2)) the term structure is upward (downward) sloping and the price differences
are positive (negative).
5 Final remarks
1% 2% 3% 4% 5%
1% 0.05% 6.96% 13.10% 18.58% 23.48%
2% -7.41% 0.04% 6.68% 12.60% 17.90%
3% -15.12% -7.10% 0.04% 6.41% 12.11%
4% -23.06% -14.45% -6.80% 0.04% 6.15%
5% -31.22% -22.01% -13.82% -6.51% 0.03%
Table 3. Price differences with respect to the Black and Scholes (1973) formula for different interest rate term structures.
This paper shows a procedure to determine the price of financial contracts that are exposed to two sources of
risk: the stock price and the interest rate. In particular, we assume that each risk factor evolves over time according
to a binomial tree so that the final distribution is, in the limit for both risk factors, lognormal. To this aim, we set
some hypotheses and in particular we assume that the correlation between the interest rate and the stock price is zero
and that the Libor rate proxies for the risk free rate. We showed that under this assumptions, the stock price is a
martingale under a particular (joint) probability measure that results from the product of the risk neutral marginal
probabilities of the two considered risk factors. Even if this assumptions appear to be clearly unrealistic, we
therefore set them to simplify the pricing approach. In this section, some techniques are proposed in order to release
such hypotheses.
In particular, the assumption of zero correlation between the interest rate and the stock price may be released by
redistributing, in a different way, the joint probabilities calculated in the case of independency that are exposed in
figure 2, among the possible states of the world. For example, we can set the hypothesis that the stock price and the
interest rate show perfectly negative correlation by equally distributing the probabilities (in case of independence) of
contemporaneously up movements and down movements of stock price and rate to the other two states of the world,
according to figure 4. The terminal stock price dynamics will thus result in a binomial random variable and, in the
limit, is lognormal.
117

Figure 4. The joint PDF of the short rate L(t, T) and of the stock price St in the case of perfectly negative correlation.
We notice that, if this is the case, the instantaneous correlation will be -1 but the terminal correlation may
depend not only on the way the probabilities are redistributed, but also on the variances of the two risk factors. For
these reasons, more sophisticated techniques must be applied in order to calibrate the model to the empirical
correlation between interest rate and stock price. Finally we point out that, if a similar technique can be applied to
impose perfectly positive correlation between the two risk factors, it is a harder task to fit the correlation parameter
to the observed market data. In this case the problem is twofold. Firstly, the correlation between interest rate and
stock price must be estimated. Secondly, we have to choose a redistribution rule (different from the above
mentioned rule) such that the correlation of the model is equal to the estimated correlation. This second point can be
solved in a very simple way. In the case of perfectly negative correlation, we set the probability of contemporaneous
up and down movement of the two risk factor equal to zero. If on the contrary we decide to equally redistribute to
the other two states of the world only a percentage of the probability of contemporaneous up and down movement,
it will result in a correlation equal to -γ. If we thus decide to equally redistribute only a 50% of the probability of
contemporaneously up and down movement to the other states of the world, it will result in a correlation of -0.5. On
the contrary, if we decide to equally distribute a percentage γ of the probability of opposite movements of the
considered risk factors to the other two states of the world, it will result in a correlation equal to γ.
We notice however that by relaxing the hypothesis of zero correlation, equation 5 does not hold and the stock
price dynamics does not expose the martingale properties. However, even if the current stock price were still an
unbiased estimate of the future stock price, it can be noticed that relaxing the hypothesis of zero correlation may
produce a price for a call option different from those that can be obtained by adopting the Cox, Ross and Rubinstein
(1979) model.
The second strong hypothesis of our model is in the adoption of the Libor rate as risk free rate. Since the Libor
rate is in general higher than a AAA interest rate, its adoption will result in higher (lower) price for a call (put)
option. To solve this problem, we can however consider the Libor rate as formed by the sum of a basic risk free rate
R(t, T) and a spread φ(t, T). In this case the risk free rate R(t, T) will be a random variable composed as follows:
Assuming that the spread is not a random variable and that it is constant over time at a certain level, it will only
be function of the tenor of the interest rate δ, and the risk free rate R(t, T) will be a random variable with the same
distribution of L(t, T), the same variance but a lower average.
6 Aknowledgments
The authots thank Dr. D. Curcio for valuable research assistance. Although the paper is the result of a joint
effort of the authors, sections 1 and 3.1 are due to R. Cocozza, whilst sections 2, 3.2, 4 and 5 are due to A. De
Simone.
118

7 References
Amin, K. I., Jarrow, R. A. (1992). Pricing options on risky assets in a stochastic interest rate economy.
Mathematical Finance, 2(4), 217-237.
Bacinello, A. R., Ortu, F. (1996). Fixed income linked life insurance policies with minimum guarantees: Pricing
models and numerical results. European Journal of Operational Research, 91(2), 235-249.
Black, F. (1976). The pricing of commodity contracts. The Journal of Political Economy, 3(1-2), 167-179.
Black, F., Derman, E., Toy, W. (1990). The pricing of options and corporate liabilities. The Journal of Political
Economy, 46(1), 33-39.
Black, F., Scholes, M. (1973). The pricing of options and corporate liabilities. The Journal of Political Economy,
81(3), 637-654.
Brace, A., Gątarek, D., Musiela, M. (1997). The market model of interest rate dynamics. Mathematical Finance,
7(2), 127-155.
Brennan, M. J., Schwartz, E. S. (1976). The pricing of equity-linked life insurance policies with an asset value
guarantee. Journal of Financial Economics, 3(3), 159-213.
Brennan, M. J., Schwartz, E. S. (1980). Analyzing convertible bonds. Journal of Financial and Quantitative
Analysis, 15(4), 907-929.
Cox, J. C., Ross, S. A., Rubinstein, M. (1979). Option pricing: A simplified approach. Journal of Financial
Economics, 7(3), 229-263.
Cox, J. C., Ingersoll, J. E., Ross, S. A. (1985). A theory of the term structure of interest rates. Econometrica, 53(2),
385-407.
Cocozza, R., Orlando, A. (2009). Managing structured bonds: An analysis using RAROC and EVA. Journal of Risk
Management in Financial Institutions, 2(4), 409-426.
Cocozza, R., De Simone, A., Di Lorenzo, E., Sibillo, M. (2011). Participating policies: Risk and value drivers in a
financial management perspectives. Forthcoming in the 14th Conference of the ASMDA International Society,
7-10 June 2011.
De Simone, A. (2010). Pricing interest rate derivatives under different interest rate modeling: A critical and
empirical analysis. Investment Management and Financial Innovations, 7(2), 40-49.
Feller, W. (1951). Two singular diffusion problems. Annals of Mathematics, 54(1), 173-182.
Ho, T. S. Y., Lee, S.-B. (1986). Term structure movements and pricing interest rate contingent claims. The Journal
of Finance, XLI(5), 1011-1029.
Hull, J. C. (2009). Options, futures and other derivatives. Prentice Hall.
Kim, Y.-J., Kunitomo, N. (1999). Pricing options under stochastic interest rates: A new approach. Asia-Pacific
Financial Markets, 6(1), 49-70.
Kunitomo, N., Kim, Y.-J. (2001). Effects of stochastic interest rates and volatility on contingent claims. CIRJE-F-
129 (Extended Version of CIRJE-F-67, 2000), University of Tokyo.
Merton, R. C. (1973). Theory of rational option pricing. The Bell Journal of Economics and Management Science,
4(1), 141-183.
Neftci, S. (2008). Principles of financial engineering. Academic Press, Advanced Finance Series.
Rabinovitch, R. (1989). Pricing stock and bond options when the default-free rate is stochastic. Journal of Financial
and Quantitative Analysis, 24(4), 447-457.
Vasicek, O. (1977). An equilibrium characterization of the term structure. Journal of Financial Economics, 5(2),
177-188.
119

LIQUIDITY AND EXPECTED RETURNS: NEW EVIDENCE FROM DAILY DATA 1926-2008
M. Reza Baradaran & Maurice Peat, The University of Sydney, Australia
Email: rezab@econ.usyd.edu.au
Abstract: This study analyzes the liquidity effect on stock expected returns in NYSE by using a new low-frequency proxy for
transaction costs over 1926-2008 which is a longest available data. Past research has concentrated on post-1963 because of lack of
data to construct liquidity measure. The results from the entire sample of 1926-2008 show that expected returns are increasing with the
stock level illiquidity. However, liquidity level has explanatory power in cross-sectional variation of stock expected returns only over
the post-1963 period and it is, both economically and statistically, insignificant for the whole sample and the pre-1963 period. These
findings are robust after taking into account various characteristics such as size and risk controls including CAPM beta and Fama-
French factors and also the systematic liquidity risk factor. Also the results for liquidity effect appear to be distinct from the size
effect. On the other hand, evidence from the entire sample and pre-1963 suggest that systematic liquidity risk is significant in
association with cross-sectional stock expected returns. These results are in favor of sample specific explanation for the illiquidity
level premium documented in many studies for post-1963 period. Nevertheless, analysis over the whole period 1926-2008 shows that
systematic liquidity risk plays significant role in cross-sectional variation of stock expected returns.
Keywords: Liquidity, Asset pricing, Transaction costs, Effective spread
JEL classification: G120
1 Introduction
Liquidity is important in many financial markets for both investors and policy makers and a large and growing body
of work has considered identifying liquidity costs and their impacts on asset pricing (e.g. Amihud and Mendelson,
1986; Gottesman and Jacoby, 2005; Korajczyk and Sadka, 2008 and Hasbrouck, 2009). A major problem in
investigating the role of liquidity in asset prices is that while it is suggested to use long time-series in asset pricing
studies 1 , the intra-day data that enable the estimation of transaction costs from the actual sequences of trades and
quotes are not available prior to 1983 (in the US markets). Even the data related to the common low-frequency
measures (such as quoted spreads) are not available prior to 1963. Therefore, current literature in liquidity-asset
pricing concentrates mostly on the post-1963 period (in the US markets).
This paper uses a new low-frequency proxy and examine the association of stock expected returns and liquidity
from 1926 onwards. The proxy is Effective Tick4, henceforth EFFT, developed by Holden (2009) and intends to be
a proxy for the intra-day effective spreads. Liquidity has many facets and it is important to note that EFFT, as a
proxy for intra-day effective spread, captures only execution cost dimension of liquidity and do not include the total
price impact of a trade. With this notation, we refer to EFFT as a proxy for liquidity and use the transactions costs
(or trading costs) and liquidity interchangeably in this study.
This examines the role of liquidity in explaining cross-sectional variation of stock expected returns in pre-1963,
for which there is a lack of research, in post-1963, and over the entire period of 1926-2008. Using all common
shares of NYSE over 1926-2008 and employing Fama-Macbeth approach in portfolios, we find that expected returns
are increasing with the stock level illiquidity. However, this positive relation is not significant statistically at the
conventional levels. Nonetheless, we find that a one percent (standard deviation) of liquidity level translates into a
monthly expected premium of about 0.12 percent or 12 basis points. This result is robust after taking into account
various characteristics such as size and risk controls including Fama-French three factors and the systematic
liquidity factor. Results from the subsamples show that liquidity level has explanatory power in cross-sectional
variation of stock expected returns only over the post-1963 period. This effect is both economically and statistically
insignificant for pre-1963 period. On the other hand, evidence from the entire sample and pre-1963 suggest that
market liquidity risk is marginally significant in association with cross-sectional stock expected returns. These
findings suggest that liquidity statistically affect the cross-sectional variation of expected returns over the entire
sample as well as the sub samples, though the channel for this effect seems to be different over the periods. For post-
1963 the premium with respect to the liquidity level is more prevalent than that of the systematic liquidity risk. The
1 In asset pricing studies, realised returns are usually used as the proxy for expected returns. Since the variance of realised returns around the
expected returns is high, long time-series data provide a large amount of data that increase the power of asset pricing tests (Amihud et al., 2005).
Accordingly, many asset pricing tests make use of US equity returns from 1926 onward (Hasbrouck, 2009).
120

opposite is true over the pre-1963 period and the entire sample. These results provide evidence for the sample
specific explanation for the illiquidity level premium documented in many studies for post-1963 period.
Nevertheless, analysis over the whole period 1926-2008 shows that systematic liquidity risk plays significant role, as
the liquidity effect, in stock expected returns.
The paper proceeds as follows. Section 2 reviews the EFFT and its construction. Section 3 presents data and
methodology that includes data description, variable and portfolio constructions, and asset pricing tests. Results are
provided and discussed in section 4. Section 5 offers the concluding remarks.
2 EFFT: a low-frequency proxy for the effective spreads
EFFT, developed by Holden (2009) is the daily proxy for the effective spread and picks up on two attributes of the
daily data: price clustering on trading days and reported quoted spreads for no-trade days. The proxy has two
components corresponding to each of these attributes. The first component, effective tick, based on the observable
price clustering is a proxy for the effective spread. The second component is the average quoted spread from any notrade
days that exist and enriches effective tick by incorporating the information related to no-trade days. First we
review the effective tick and then conclude the EFFT estimator. Effective tick is based on the idea that the effective
spread on a particular day equals the increment of the price cluster on that particular day. For example on a $1/8
fractional price grid, if the spread is $1/4, their model assumes that prices end on even-eights, or quarters. Thus, if
odd-eight transaction prices are observed, one must infer that the spread must be $1/8. This implies that the simple
frequency with which closing prices occur in particular price clusters (in a time interval) can be used to estimate the
corresponding spread probabilities and ,hence, infer the effective spread for that interval. For example on a $1/8
fractional price grid, the frequency with which trades occur in four, mutually exclusive price cluster sets (odd $1/8s,
odd $1/4s, odd $1/2s, and whole dollars), can be used to estimate the probability of a $1/8 spread, $1/4 spread, $1/2
spread, and a $1 spread, respectively. There are similar clusters of special prices on a decimal price grid (off
pennies, off nickels, off dimes, off quarters, and whole dollars) that can be used to estimate the probability of a
penny spread, nickel spread, dime spread, quarter spread and whole dollar spread, respectively. In order to construct
the effective tick proxy for a time interval, the first step is to compute the frequency of each price cluster within that
time interval. Take St as the realisation of the effective spread at the closing trade of day t and assume that St is
randomly drawn from a set of possible spreads Sj (for example in $1/8 fractional price grid, S1 = $1/8 spread, S2 =
$1/4 spread, S3 = $1/2 spread and S4 = $1 spread) with corresponding probabilities γj , where j=1,2,…,J and S1< S2

Then, the effective tick proxy is calculated as the probability-weighted average of each effective spread size
divided by the average price ( p i ) in time interval i :
J
�
j s j
j�1
EffectiveTicki
�
pi
ˆ �
(4)
Holden (2009) incorporates the average of the quoted spreads in no-trade days into the effective tick estimator
and concludes the EFFT. EFFT for the time interval i is the probability weighted average of the effective estimator
and the average of the quoted spreads from no-trade days:
� 1 NTD
�(
1 � ˆ �)
� NQSt
when NTD�0,
� NTD t�1
�
�0
when NTD � 0
EFFTi
� ˆ � � ( EffectiveTicki
) �
(5)
Pi
where NQSt is the quoted spread computed by using reported bid and ask quoted prices on no-trade days, and �ˆ
is the estimated probability of a trading day given by
TD
�ˆ �
(6)
TD � NTD
where TD and NTD are the number of trading days and no-trade days over the time interval, respectively.
Holden (2009) and Goyenko et al. (2009) reports high correlation between this proxy and high-frequency and lowfrequency
benchmarks. They also show that this measure performs better against the high-frequency benchmarks
than other available low-frequency proxies for liquidity and/or transaction costs.
3 Data and Methodology
3.1 Data
Daily transaction data from CRSP daily file from December 31st, 1925 until December 31st, 2008 for all the
common stocks (CRSP codes 10 and 11) listed in NYSE are employed to estimate the monthly EFFT. Monthly
returns and other required data to compute characteristics for the stocks are downloaded from CRSP monthly file.
Data for Fama and French three factors (market, size and value) are downloaded from French’s website (French,
2010). To be included in the monthly cross-sectional analysis a stock must be traded at the beginning and end of the
year. It also needs to have monthly data on return and market capitalisation at the start and end of the year. In
addition, as suggested by Fama and French (1992), we exclude financial firms because they usually have high
leverage. This screening process yields on average 3035 stocks. NYSE introduced decimal pricing regime by
applying the new regime to some pilot firms from 28 August, 2000 and then completely switched to decimal grids
on 29 January, 2001. We eliminate the pilot firms from our sample which started to be quoted and traded based on
decimal pricing system at various times after August 28, 2000. Since estimation of EFFT is based on the tick sizes,
this elimination makes the computation of EFFTs across the stocks, consistent. This filtering removes 88 pilot firms.
In order to prepare data for the cross-sectional analysis, for each stock-month, EFFT and size are computed.
EFFT is calculated at the end of the month using daily trade data and employing equations 1 to 6. From January
1926 to January 2001, during which NYSE was using fractional price grid, price increments as small as $1/64 are
used. From February 2001 to December 2008, during which NYSE has had decimal pricing system, tick sizes have
been $0.01, $0.05, $0.10, $0.25 and $1. Size is the market capitalisation of the equity at the end of the month.
Table 1 reports estimated statistics of time-series averages of EFFT, size and returns across all stocks (in each
month) for the whole period of 1926-2008.
Variable Mean Std Dev Median
Return
EFFT
Size (in $ Billion)
0.0117
0.0124
1.5108
0.072
0.007
2.3508
0.014
0.011
0.4380
Table 1: Summary statistics for the individual stocks
The average of monthly effective spread (EFFT) is approximately 1.24%. Size variable displays considerable
skewness and its distribution is skewed to the right. Therefore, in our empirical analysis we use relative measure for
122

size (the log market capitalisation relative to the median), as suggested by Hasbrouck (2009) to ensure the
stationarity of the Size time-series. This relative measure is constructed at the end of the month as follows. If mjt
indicates the natural logarithm of the equity market capitalisation of the stock j at the end of the month, the log
relative market capitalisation is the difference between mjt and the cross-sectional median of the natural logarithm of
the equity market capitalisation of all the stocks computed in the month. The log size relative to the median captures
the cross-sectional variation while removing the non-stationary long-run components.
3.2 Methodology
We employ the procedure proposed by Fama-MacBeth (1973) and use portfolios to test whether EFFT, as a
security characteristic, has incremental explanatory power for returns relative to common risk factors and after
controlling for other well known stock characteristics. Portfolios are formed annually based on the information
available at the start of the year: EFFT and beta estimated over the prior period. Using the data over the prior 3-4
years, pre-ranking (rolling) betas of individual stocks are estimated employing market model regressions of monthly
excess returns (over the one-month Treasury bill return). The market index is the CRSP equally-weighted
AMEX/NYSE/Nasdaq index. At the end of a year all stocks in the sample are ranked in five equal groups by the
pre-ranking beta estimates. Each of these five beta groups is then divided into another five equal subgroups by
ranking stocks based on their average monthly EFFT computed over the prior year. This concludes twenty five
portfolios, as asset representatives, with almost equal numbers of stocks which are rebalanced every year. Then we
compute the portfolio variables (returns, EFFT, size) and unconditional betas (after ranking) as the equally weighted
averages of stock variables. We applied the equally-weighted average rather than the value-weighted because as
Hasbrouck (2009) points out value-weighted averages are likely to suppress variation in liquidity due to the inverse
association of liquidity and size.
EFFT
Group
Lowest
2
3
4
Highest
Mean
EFFT, [Beta], , (Size), Number of firms
Beta Group
Lowest 2 3 4 Highest Mean
0.31
[0.4804]
(7.3334)
27
0.46
[0.5255]
(3.0663)
27
0.59
[0.5292]
(2.5494)
27
0.76
[0.5763]
(1.0996)
27
1.52
[0.719]
(0.3827)
29
0.728
[0.5661]
(2.8863)
28
0.34
[0.6387]
(7.2161)
27
0.51
[0.6304]
(2.7027)
27
0.67
[0.7328]
(1.3179)
27
0.89
[0.7637]
(0.6008)
26
1.85
[0.8957]
(0.2289)
29
0.85
[0.7323]
(2.4133)
28
0.38
[0.7863]
(4.2774)
27
0.56
[0.7991]
(1.8587)
27
0.75
[0.8537]
(0.9112)
27
1.02
[0.9871]
(0.4061)
27
2.29
[1.0715]
(0.1966)
29
1.00
[0.8995]
(1.53)
28
0.45
[0.9187]
(2.8952)
27
0.67
[1.105]
(1.1772)
0.61
[1.1232]
(1.8086)
28
0.94
[1.2892]
(0.7556)
0.42
[0.7895]
(4.7061)
27
0.63
[0.8709]
(1.9121)
Table 2: Summary Statistics for the 25 annually re-balanced portfolios of NYSE firms
Table 2 reports the summary statistics of the portfolio values over 1930-2008 (948 test-month period). Portfolio
EFFTs vary between 0.31 percent to 4.55 percent, while betas change between 0.48 and 1.42, and returns range from
0.76 percent to 1.72 percent. The values for beta and EFFT in higher ranked portfolios imply that low liquid stocks
123
27
0.91
[1.0143]
(0.6184)
27
1.27
[1.1664]
(0.3029)
27
2.99
[1.1945]
(0.1568)
29
1.26
[1.0809]
(1.0301)
28
28
1.3
[1.3077]
(0.4029)
28
1.9
[1.338]
(0.1796)
28
4.55
[1.4242]
(0.0934)
30
1.86
[1.2965]
(0.6480)
28
27
0.84
[0.8875]
(1.1590)
27
1.17
[0.9663]
(0.5178)
27
2.64
[1.0610]
(0.2117)
29
1.14
[0.9150]
(1.7015)
28

tend to be high beta stocks. Portfolio market capitalisation varies between $93.4 million and $9.33 billion. Within
each beta group, return generally increases when EFFT increases and market capitalisation decreases. This suggests
that illiquid stocks tend to have smaller market value and higher returns. Within each EFFT group, market
capitalisation decreases when portfolio betas increase, while return increases. This implies that smaller stocks are
riskier and associated with higher returns.
3.2.1 Asset pricing tests
In order to explore the relation between stock expected returns and EFFT, and based on Fama-Macbeth
approach, a cross-section model is estimated across all portfolios for each month t, where monthly portfolio excess
returns are a function of asset characteristics and risk factor loadings:
K
J
rit
� � 0 � � � � � � � Z � �
(7)
kt ki jt ji,
t it
k � 1 j � 1
where rit is the excess returns on portfolio i in month t of the test year and Zji,t is characteristic j of portfolio i,
estimated at the end of the preceding year from data in that year and known to investors at the beginning of the test
year (during which they make their investment decisions). βki is the unconditional risk variables for portfolio i
computed over the sample. The slopes δkt and λkt are the effects of characteristics on the expected excess returns and
the risk factor premia in cross-section month t, respectively, and εit are the residuals.
The overall coefficient estimates is the standard Fama-Macbeth estimator which is the arithmetic average of the
time-series of the OLS estimates of coefficients in equation 7. The t-statistics are adjusted with respect to
heterscedasticity and autocorrelations (HAC) by using Newy-West (1987) method with three lags. Following
characteristics and risk variables are used in model 7.
3.2.2 Characteristics variables
3.2.2.1 Transaction cost proxy
EFFT: the transaction costs proxy for portfolio i in month t of the teat year y is the average of the portfolio
monthly EFFTs in the preceding year.
3.2.2.2 Control variables
Relative Size: to account for size, we use relative measure for size (the log market capitalisation relative to the
cross-sectional median) to ensure the stationarity of the size time-series. The log relative market capitalisation for
portfolio i in month t of the test year y is the log relative market capitalisation of the portfolio at the end of the
preceding year.
Momentum: in our empirical analysis, the momentum variable for portfolio i in month t is the portfolio
monthly cumulative returns over the first half the preceding year. Novy-Marx (2008) documents that a stock’s
intermediate horizon past performance, the first six months of the prior year, drives the momentum.
3.2.3 Systematic risk variables
The risk variables are the projection coefficients (loading factors) in the K-factor return generating process
rit K
� �0 � �
k �
� f
ki kt
j
� eit
where rit is the excess returns on portfolio i in month t of the test year and fkt is factor realisations in month t . βki
is factor loadings for the respective risk factors and eit is the residuals. The (unconditional) betas of each portfolio
are obtained via OLS time-series regression of equation 8 over the sample. In our asset pricing analysis we use
various sets of factors. The basic set includes only excess market return as the systematic factor. So equation 8
becomes a classical CAPM model. The market return is CRSP equally weighted AMEX/NYSE/Nasdaq index and
the risk-free rate is one-month Treasury bill rate. The second set is the three Fama-French factors i.e. i.e. Mkt, the
market portfolio excess returns, SMB which is the mimicked performance of a portfolio that is long in small firms
and short in large firms, and HML which is the mimicked performance of a portfolio that is long in high book-tomarket
equity firms and short in low book-to-market equity firms. The third set consists of excess market returns
and the liquidity factor, making model 8 a liquidity-augmented CAPM. The forth set contains three Fama-French
three factors and the liquidity factor, suggesting model 8 as a liquidity-augmented Fama-French model. The liquidity
124
(8)

factor, LIQ, is the mimicked performance of a portfolio that is long in low liquid firms and short in high liquid firms
after controlling for their market risk beta. The construction of our liquidity factor, LIQ, is similar to the
construction of SMB in Fama and Frecnh (1993). At the start of each year we rank all NYSE common stocks based
on their beta computed using previous 3 to 4 years. We form three portfolios based on 30th and 70th NYSE
percentiles; low beta, neutral beta and high beta. Then within each beta portfolios we sort the stocks based on their
previous year monthly average EFFT and construct two portfolios: low liquid and high liquid portfolios. The
breakpoint is the median value for liquidity within each beta portfolio. As a result, we have six portfolios at the start
of each year. The liquidity factor, LIQ, is the monthly average return on the three (equally weighted) low-liquid
portfolios minus the monthly average return on the three (equally weighted) high-liquid portfolios.
4 Pricing results
We conduct the cross-sectional analysis using equation 7 over four sample periods: 1- the entire sample: from
January 1926 to December 2008, 2- pre-1963: from January 1926 to December 1962, 3- post-1963: from January
1963 to January 2001 and 4- from January 1963 to December 2008. The reason we have analysed post-1963 data
into two samples is that NYSE has introduced decimal regime in February 2001 that may affect the variation of
liquidity and consequently affect our asset pricing results. Table 3 presents the regression results (employing
equations 7 and 8) when we use the CAPM beta as the risk control for four periods.
Specification
Variable 1 2 3 4 5 6
Panel A: 01/ 1926 -12/2008
Intercept 0.002 (1.08) 0.0033 (1.91) 0.0074 (3.7) 0.0014 (0.74) 0.0057 (2.26) 0.005 (1.96)
Beta 0.007 (2.27) 0.003 (1.03) 0.0021 (0.69) 0.0076 (2.29) 0.002 (0.67) 0.0025 (0.87)
EFFT 0.1955 (1.86) 0.1151 (1.05) 0.1156 (1.21)
Relative Size -0.0015 (-2.89) -0.0006 (-1.17) -0.0005 (-0.91)
Momentum -0.0069 (-1.6) -0.0044 (-1.19)
Panel B: 01/ 1926 -12/1962
Intercept 0.0005 (0.17) 0.0017 (0.59) 0.0067 (1.51) 0.001 (0.31) 0.0062 (1.33) 0.0073 (1.55)
Beta 0.0107 (1.92) 0.0081 (1.51) 0.0064 (1.17) 0.0103 (1.8) 0.0061 (1.13) 0.0044 (0.87)
EFFT 0.0883 (1.24) 0.0314 (0.44) 0.0417 (0.57)
Relative Size -0.0014 (-1.61) -0.0011 (-1.19) -0.0014 (-1.47)
Momentum -0.0053 (-0.83) -0.0014 (-0.22)
Panel C: 01/1963-01/ 2001
Intercept 0.0032 (1.45) 0.0041 (1.88) 0.0066 (2.67) 0.0014 (063) 0.0039 (1.61) 0.0024 (1.01)
Beta 0.0038 (1.14) -0.0013 (-0.28) -0.0003 (-0.001) 0.0064 (1.57) -0.0009 (-0.25) 0.0019 (0.45)
EFFT 0.2782 (2.66) 0.306 (3.11) 0.236 (2.48)
Relative Size -0.0012 (-1.91) 0.0002 (0.38) 0.0006 (1.04)
Momentum -0.0089 (-1.25) -0.0127 (-2.4)
Panel D: 01/1963-12/ 2008
Intercept 0.0029 (1.49) 0.0043 (2.21) 0.007 (3.09) 0.0018 (0.86) 0.0047 (2.08) 0.0034 (1.53)
Beta 0.0034 (1.1) -0.0018 (-0.57) -0.0011 (-0.34) 0.005 (1.33) -0.0019 (-0.59) 0.0001 (0.002)
EFFT 0.3081 (1.61) 0.2687 (1.66) 0.2132 (1.41)
Relative Size -0.0014 (-2.45) -0.0001 (-0.15) 0.0003 (0.62)
Momentum -0.0088 (-1.4) -0.0121 (-2.68)
Table 3: Liquidity estimates based on CAPM. Numbers in parentheses are t-statistics, adjusted by Newey-West method
Panel A reports the cross sectional results for the entire sample. Specification 1 contains CAPM beta and the
intercept. The estimated market price of risk is 0.7 percent with a t-statistic of 2.27. Specification 2 includes EFFT
as the only stock variable. The point estimate of EFFT is positive and marginally significant, but it becomes
statistically insignificant after controlling for size and momentum (specification 6). Nonetheless, the coefficient
associated with EFFT is economically interpretable. The point estimate is 0.12 (with a t-statistic of 1.21). This
estimate suggests that if (the standard deviation of) the transaction costs is one percent, the associated monthly
expected premium is about 0.12 percent or 12 basis points. This value is associated with a premium equal to about
1.4 percent on annual basis. This value is not large. The average EFFT over the entire sample is about 1.2 percent
(table 1). This value implies that average EFFT, as the proxy for transaction costs, has had a premium equal to about
1.7 percent on an annual basis throughout our sample.
The point estimate of 0.12 for EFFT using more than 80 years data can be justified by discussing its equivalent
turnover rate for a round-trip trader. A representative trader who makes a round-trip trade over a specific period
incurs 2EFFT transaction costs (once when she buys and once when she sells). A point estimate of 0.12 for EFFT
125

suggests that the expected gross returns over about 17 (2/0.12) months impound 2EFFT for this trader. 17 months
holding period is equal to 70 percent turnover per year. This rate of turnover seems to be reasonable as an average
for NYSE from 1926 to 2008. As Hasbrouck (2009) mentions the average annual turnover rate for NYSE has been
around one in 2008 and are well under one for most of the 20 th century. The historical data section of the NYSE
website reports 103 percent turnover rate for 2005 and 12 percent rate for 1960. So our result for the coefficient of
EFFT is consistent with the trading stories.
The premium commanded by EFFT is comparable with that of Hasbrouck’s Gibbs estimate, as reported in
Hasbrouck (2009). Hasbrouck’s Gibbs estimate, as a low-frequency measure for transaction costs, is a modification
of Roll’s (Roll, 1984) estimator and measures deviations of transaction prices from efficient values, i.e., effective
execution costs. Our results can be compared with that of Hasbrouck (2009) since both Gibbs estimate and EFFT
measures transaction costs and Hasbrouck has reported his regression results for the period of 1927-2006.
Hasbrouck (2009) finds a coefficient of 0.93 for his transaction costs proxy, Gibbs estimate. However, the point
estimate is marginally significant at 5 percent level (t-statistics is 2.07) only when he controls for CAPM beta and
does not take into account the characteristics controls. In other specifications, there is no significant result for the
regression coefficient of transaction costs. So our finding provides further support that transaction costs does not
have significant explanatory power in explaining cross-sectional variation of returns. From economical point of
view, Hasbrouck’s finding implies that one percent transaction costs leads to 11.2 percent annual premium which is
very large. Moreover, as he explains this large magnitude suggests about a six times turnover rate per year for NYSE
which is unrealistic. It seems that the premium commanded by EFFT, compared to Gibbs estimate accords more
with the straightforward trading stories.
Specification 3 contains relative size in lieu of EFFT, as the only stock characteristics. The coefficient of
relative size is negative, as we expected, and statistically significant at 1 percent level (panel A table 3). However, it
becomes insignificant when we include EFFT in specification 5 while EFFT is still insignificant. This pattern can be
seen when we compare specifications 2,3 and 5 in pre-1963 and post-1963 subsamples too; the t-statistics of the
coefficient associated with the relative size is attenuated in specification 5 while that of EFFT is barley altered.
These findings imply that effects usually attributed to size are more precisely ascribed to transaction costs. Also,
our finding shows that inclusion of momentum barely modifies the economical and statistical significance of EFFT.
Now we focus on the results from subsamples. In all panels and specifications the proxy for liquidity, EFFT,
has a positive sign as we expected. Interestingly, while EFFT is not significant using more than 80 years data (panel
A), it is significant at the 1 percent level over the period of 1963-2001 (Panel C). This significant positive result
accords with that of previous research that analyses post-1963 data. However, when we extend the sample to 2008
(panel D) EFFT becomes insignificant at least at the conventional 1 and 5 percent significance levels. A likely
reason is that decimalisation after 2001 has made the cross sectional variations in spreads, and so EFFTs, decreased.
This can attenuate the significance level. However, the EFFT coefficient does not change considerably in each
specification when we move from panel C to D. EFFT does not show significant explanatory power in pre-1963
period (panel B). The point estimates (in all specifications) for EFFT over pre-1963 period are less than those
reported for post-1963 periods. While the EFFT coefficient in post-1963 sample is at least 0.21 (specification 6 in
panel D), its maximum value is 0.09 for pre-1963 (specification 2 in panel B). Furthermore, the p-values to reject the
null hypothesis are also higher for pre-1963 when we compare the corresponding specifications in four sub-samples.
This low magnitude and t-statistics suggest that liquidity (level) effect has been less prevalent statistically AND
economically for pre-1963.
Table 4 presents the regression results when we control for Fama-French three risk factors. Specification 1 in
all panels reports only the market price of three factors and the intercept. The coefficients corresponding to each risk
factors and their t-statistics varies in all panels, but generally the price of size factor (SMB) is positive and
significant whereas those of market and book-to-market risk factors are insignificant and close to zero.
The findings are barely modified when we move from the CAPM-based results (table 3) to the FF three factorbased
findings (table 4). The sign of coefficient associated with EFFT is again positive in all specifications and for
all samples. Also, EFFT point estimate is not statistically significant over the period of 1926-2008 (panel A) and
pre-1963 (panel B) whereas it is significant for post-1963 (panels C and D). Furthermore, the magnitude of the
coefficient associated with EFFT is virtually unaltered for the entire sample (panel A specification D) and it is about
0.12.
Nevertheless, for the period of 1963 to 2008 (panel D) the regression coefficient associated with EFFT is
significant at 5 percent level whereas it was insignificant when we used only CAPM beta as the risk control (panel D
table 3). Moreover, the point estimates for EFFT for the post-1963 analysis (Panel C and D table 4) almost double
those counterpart estimates reported in table 3. Also their corresponding statistical significance is also improved
when we control for FF three risk factors. On the other hand, the point estimate of EFFT is halved in pre-1963
126

sample (panel B table 4) compared to that of reported in table 3 and their corresponding t-statistics are slightly
attenuated. These slight changes in magnitude and significance levels of EFFT for the sub-samples of pre-1963 and
post-1963 confirm our previous key findings in table 3; idiosyncratic liquidity does not have explanatory power in
cross-sectional variation of stock returns before 1963, but it matters in expected returns after 1963.
Specification
Variable 1 2 3 4 5 6
Panel A: 01/ 1926 -12/2008
Intercept 0.008 (3.72) 0.0056 (2.36) 0.009 (3.5) 0.0074 (3.35) 0.0066 (2.35) 0.0056 (1.93)
Mkt_beta -0.0018 (-0.77) 0.0001 (0.05) -0.0014 (-0.6) -0.0013 (-0.53) 0.0003 (0.12) 0.0005 (0.23)
SMB_beta 0.006 (3.46) 0.0041 (1.69) 0.005 (2.56) 0.0058 (3.21) 0.003 (1.46) 0.0033 (1.51)
HML_beta 0.0003 (0.15) 0.0011 (0.63) -0.0006 (-0.29) -0.0002 (-0.11) 0.0006 (0.33) 0.0008 (0.39)
EFFT 0.0983 (0.62) 0.1263 (0.79) 0.1210 (1.13)
Relative Size -0.0006 (-1.02) -0.0004 (-0.8) -0.0001 (-0.23)
Momentum -0.0074 (-1.93) -0.0089 (-2.27)
Panel B: 01/ 1926 -12/1962
Intercept 0.0038 (1.05) 0.0015 (0.42) 0.0058 (1.19) 0.0034 (0.91) 0.0051 (1.01) 0.0057 (1.12)
Mkt_beta 0.004 (0.95) 0.0058 (1.37) 0.0048 (1.13) 0.0047 (1.12) 0.0053 (1.2) 0.0045 (1.1)
SMB_beta 0.0066 (2.2) 0.0051 (1.55) 0.0049 (1.68) 0.0067 (2.18) 0.0047 (1.56) 0.0046 (1.46)
HML_beta 0.0014 (0.52) 0.0009 (0.3) -0.0012 (-0.46) 0.0004 (0.15) -0.0007 (-0.25) -0.001 (-0.46)
EFFT 0.0746 (0.99) 0.0142 (0.16) 0.0101 (-0.12)
Relative Size -0.0011 (-1.28) -0.001 (-1.14) -0.001 (-1.3)
Momentum -0.0045 (-0.79) -0.0027 (-0.45)
Panel C: 01/1963-01/ 2001
Intercept 0.0073 (1.82) -0.006 (-1.16) 0.0083 (2.01) 0.004 (0.91) -0.0048 (-1.01) -0.0052 (-1.06)
Mkt_beta -0.003 (-0.8) 0.0075 (1.7) -0.0058 (-1.39) 0.0015 (0.32) 0.0058 (1.34) 0.0074 (1.56)
SMB_beta 0.0033 (1.72) -0.0061 (-2.00) 0.0068 (2.35) 0.0032 (1.52) -0.0043 (-1.17) -0.0029 (-0.85)
HML_beta 0.0017 (0.65) 0.0038 (1.46) 0.0024 (0.91) 0.0004 (0.14) 0.0033 (1.21) 0.0017 (0.61)
EFFT 0.5263 (3.01) 0.5358 (2.9) 0.4805 (2.78)
Relative Size 0.001 (1.14) 0.0002 (0.26) 0.0004 (0.49)
Momentum -0.0112 (-1.83) -0.0108 (-1.97)
Panel D: 01/1963-12/ 2008
Intercept 0.0057 (1.67) -0.0041 (-0.99) 0.0064 (1.82) 0.0044 (1.2) -0.0034 (-0.87) -0.0029 (-0.73)
Mkt_beta -0.0028 (-0.86) 0.0054 (1.47) -0.0042 (-1.22) -0.0004 (-0.12) 0.0048 (1.34) 0.005 (1.29)
SMB_beta 0.0028 (1.45) -0.0057 (-2.08) 0.005 (1.79) 0.0026 (1.29) -0.005 (-1.46) -0.0035 (-1.05)
HML_beta 0.0036 (1.4) 0.0056 (2.21) 0.0032 (1.34) 0.0018 (0.76) 0.0047 (2.02) 0.0029 (1.28)
EFFT 0.4639 (2.15) 0.446 (2.00) 0.4336 (2.17)
Relative Size 0.0005 (0.71) 0.0001 (-0.06) 0.0003 (0.38)
Momentum -0.0103 (-2.07) -0.0104 (-2.2)
Table 4: Liquidity pricing estimates based on FF3 model. Numbers in parentheses are t-statistics, adjusted by Newey-West method
In addition, comparing the specifications 2,3 and 5 in all panels of table 4 confirms our previous finding that the
effect ascribed to the size is more precisely explained by trading costs. When both EFFT and relative size are
included into the model, it is observed that the significance of relative size attenuated while that of the EFFT is
slightly modified.
The coefficient associated with the momentum variable is significant over the entire sample (panel A table 4),
though its sign is negative which suggest a negative feedback effect in NYSE. The explanatory power of momentum
variable beside Fama-French three factors is consistent with the Carhart’s (1997) finding that suggests previous
returns mimics the momentum risk factor to augment Fama-French three factors. Nonetheless, inclusion of
momentum barely changes the economical and statistical significance of EFFT over entire sample and all
subsamples.
In summary, findings from tables 3 and 4 suggests that while there is a positive and significant relation between
idiosyncratic transaction costs and returns over post-1963 sample, this relation is not significant when we use the
entire sample and also pre-1963 subsample. Since pre-1963 period is associated with the higher level of aggregate
illiquidity than post-1963 and there is a possibility that systematic risk factors that we have controlled so far have
not impounded the market level liquidity risk, we control for the market liquidity risk and re-run our asset pricing
analysis. Table 5 report the expected returns specifications when we control for CAPM beta and liquidity beta,
calculated based on LIQ liquidity factor. The specification results show that the general findings are the same as
those we obtained in table 3. The sign of the EFFT regression coefficient is positive in all samples, as we expected.
Also, the liquidity effect is statistically significant only for post-1963 and the point estimate of EFFT for the whole
period (panel A) is barely changed.
However, the statistical significance of EFFT for the whole period (panel A) and both its statistical and
economical significance for pre-1963 (panel B) are attenuated when we control for the market liquidity risk.
127

Interestingly, this is vice versa for post-1963 (panels C and D); the regression coefficients associated with EFFT are
doubled and the statistical significance generally increases.
Specification
Variable 1 2 3 4 5 6
Panel A: 01/ 1926 -12/2008
Intercept 0.0048 (2.64) 0.0037 (1.89) 0.0059 (2.24) 0.0045 (2.38) 0.0048 (1.83) 0.0039 (1.44)
Market Beta 0.0045 (1.54) 0.0044 (1.38) 0.0039 (1.28) 0.0043 (1.42) 0.0034 (1.11) 0.0039 (1.26)
Liquidity Beta -0.0046 (-3.7) -0.0029 (-1.35) -0.0041 (-2.81) -0.0042 (-3.37) -0.0018 (-0.93) -0.0021 (-1.1)
EFFT 0.0985 (0.56) 0.1113 (0.66) 0.1754 (1.04)
Relative Size -0.0003 (-0.53) -0.0004 (-0.68) -0.0001 (-0.27)
Momentum -0.0058 (-1.49) -0.005 (-1.28)
Panel B: 01/ 1926 -12/1962
Intercept 0.0026 (0.87) 0.0016 (0.54) 0.0061 (1.26) 0.003 (0.99) 0.0055 (1.17) 0.0065 (1.37)
Market Beta 0.009 (1.67) 0.0097 (1.72) 0.0067 (1.24) 0.0082 (1.54) 0.0071 (1.37) 0.0055 (1.07)
Liquidity Beta -0.0049 (-2.23) -0.004 (-1.3) -0.0026 (-1.13) -0.0048 (-2.1) -0.0019 (-0.72) -0.0024 (-0.89)
EFFT 0.0282 (0.26) 0.0059 (0.05) 0.034 (-0.3)
Relative Size -0.0011 (-1.07) -0.001 (-1.06) -0.0012 (-1.2)
Momentum -0.0034 (-0.58) -0.0021 (-0.34)
Panel C: 01/1963-01/ 2001
Intercept 0.0063 (2.7) 0.0017 (0.61) 0.0061 (2.49) 0.0042 (1.73) 0.0012 (0.43) -0.0005 (-0.17)
Market Beta 0.0011 (0.32) -0.0011 (-0.29) 0.002 (0.59) 0.0038 (0.92) -0.0014 (-0.35) 0.0015 (0.34)
Liquidity Beta -0.0037 (-2.57) 0.0028 (0.94) -0.0063 (-3.24) -0.0031 (-2.22) 0.0025 (0.76) 0.0021 (0.65)
EFFT 0.5327 (2.8) 0.6078 (2.88) 0.5489 (2.77)
Relative Size 0.0008 (1.09) 0.0002 (0.31) 0.0005 (0.75)
Momentum -0.0096 (-1.52) -0.0103 (-1.78)
Panel D: 01/1963-12/ 2008
Intercept 0.0062 (2.98) 0.0041 (1.77) 0.0057 (2.51) 0.0048 (2.3) 0.0038 (1.58) 0.0021 (0.84)
Market Beta 0.0006 (0.19) -0.0014 (-0.43) 0.0014 (0.44) 0.0019 (0.54) -0.002 (-0.52) -0.0001 (-0.02)
Liquidity Beta -0.0042 (-2.94) 0.0002 (0.1) -0.0058 (-3.28) -0.0033 (-2.53) 0.0011 (0.37) 0.0011 (0.36)
EFFT 0.3075 (1.12) 0.3632 (1.36) 0.4108 (1.61)
Relative Size 0.0005 (0.8) -0.0001 (-0.11) 0.0003 (0.53)
Momentum -0.0101 (-2.00) -0.0114 (-2.28)
Table 5: Liquidity pricing estimates based on augmented-CAPM. Numbers in parentheses are t-statistics, adjusted by Newey-West method
Therefore, the evidence from the entire sample and also pre-1963 show that a part of the liquidity effect associated
with the liquidity level is attributed to the market liquidity risk, though it is economically small. Nonetheless, the
liquidity risk is less important for the post-1963 period.
Specifications 1,3 and 4 in all panels of table 5 show that market liquidity risk, as the only variable representing
liquidity effect, is statistically significant, though economically insignificant. However, when liquidity level is
included into the model (specifications 2,5 and 6) the liquidity beta becomes insignificant. The liquidity level is also
insignificant for the entire sample and pre-1963 (panels A and B) but significant for post-1963. This again suggests
that both the liquidity level and the liquidity risk are responsible for the liquidity effect. Nevertheless, market
liquidity risk is less important for post-1963.
What’s more, comparing specifications 2,3 and 5 in table 5 confirms our previous finding that the effect
associated with size is more attributed to the liquidity level. Furthermore, table 4 shows that Momentum is generally
significant, but it does affect neither point estimates nor the significance level of the EFFT and liquidity beta.
We further control for Fama-French three factor loadings besides the loading associated to the liquidity state
factor, but the results barley modify our previous findings (corresponding table is not tabulated).
5 Concluding remarks
There is a big issue to study pricing implications of liquidity; on the one hand, we require a long time-series of
data to learn about asset expected returns. On the other hand, high frequency data that are usually used to estimate
liquidity are not available for a long period (they are not available prior to 1983 in US market). There are two
solutions for this issue. The first is to limit the sample to the post-1983 period and use a high frequency measure to
estimate the liquidity and apply it to asset pricing study with the hope that they tell us about the pricing implications
of liquidity. The second approach is to use low-frequency data (that are available for much longer periods) and to
employ low-frequency measures to estimate liquidity and use those estimates in asset pricing tests. This paper
adopts the second approach and uses EFFT as a low-frequency proxy. EFFT is computed using daily data and
intends to be a proxy for the intra-day effective spreads. This paper examines the cross-sectional association of stock
128

expected returns and trading costs over 1926-2008 in NYSE by using EFFT as a proxy for liquidity. We conduct the
analysis over the entire sample and pre-1963 (over which there is a lack of research in liquidity asset pricing and
post-1963 (that has been covered by almost all the research in this area). Our findings from the entire sample shows
that expected returns are increasing with illiquidity of the stock. This is consistent with the usual view that stocks are
priced so that their gross returns include compensation for trading costs. However, this positive relation is not
significant statistically at the conventional levels. Nonetheless, it is economically significant and interpretable. We
find that a one percent (standard deviation) of liquidity level translates into a monthly expected premium of about
0.12 percent or 12 basis points. This result is robust after taking into account various characteristics such as size and
risk controls including systematic liquidity factor. Our results from the subsamples show that liquidity level has
explanatory power in cross-sectional variation of stock expected returns only over the post-1963 period. This effect
is both economically and statistically insignificant for pre-1963 period. Although controlling systematic liquidity
risk barely changes these findings, new interesting finding emerges. Evidence from the entire sample and pre-1963
suggest that market liquidity risk is marginally significant in association with cross-sectional stock expected returns,
though this relation is economically insignificant.
These findings suggest that liquidity statistically affect the cross-sectional variation of expected returns over the
entire sample as well as the sub samples, though the channel for this effect seems to be different over the periods. It
seems that over the post-1963 the premium with respect to the liquidity level is more prevalent than that of the
systematic liquidity risk. This is opposite over pre-1963 period and the entire sample.
6 References
Amihud, Y. and H. Mendelson, 1986. Asset pricing and the Bid-Ask Spread. Journal of Financial Economics. 17:
223-249.
Amihud, Y., H. Mendelson, L.H. Pedersen, 2005. Liquidity and asset prices. Foundations and Trends in Finance.
1,4: 269-364.
Carhart, M., 1997. On persistence of mutual fund performance. Journal of Finance. 52: 57-82.
French K.R., 2010. Tuck School of Business at Dartmouth, accessed 9 March 2010.
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/index.html.
Fama, E.F., K.R. French, 1992. The cross section of expected stock returns. Journal of Finance. 47: 427-466.
Fama, E.F., K.R. French, 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial
Economics: 33: 3–56.
Fama E., J. MacBeth, 1973. Risk, Return, and Equilibrium: Empirical Test. Journal of Political Economy: 81 (3)
607-636.
Gottesman, A. and G. Jacoby, 2005. Payout policy, taxes, and the relation between returns and the bid-ask spread.
Journal of Banking and Finance. 30: 37-58.
Goyenko, R. Y., C. W. Holden, C.A. Trzcinka, 2009. Do liquidity measures measure liquidity? Journal of Financial
Economics. 92: 153–181.
Hasbrouck, J. (2009). “Trading Costs and Returns for U.S. Equities: Estimating Effective Costs from Daily Data.”
The Journal of Finance LXIV (3): 1445-1477.
Holden, C. W. (2009). "New low-frequency spread measures." Journal of Financial Markets 12: 778–813.
Korajczyk, R. A. and R. Sadka (2008). “Pricing the commonality across alternative measures of liquidity.” Journal
of Financial Economics 87(1): 45–72.
Newey, W., West, K., 1987. A simple, positive-definite, heteroscedasticity and autocorrelation consistent covariance
matrix. Econometrica 55, 703–708.
Novy-Marx, R. (2008). “Momentum is not Momentum.” Working paper, University of Chicago.
129

BANKING
130

131

ELECTRONIC BANKING IN JORDAN: A FRAMEWORK OF ADOPTION
Muneer Abbad, Prince Mohammad Bin Fahd University, Saudi Arabia
Juma’h Abbad, Al Al Bayte University, Jordan
Faten Jaber, Oxford Brookes University, UK
Abstract: Why some customers use e-banking systems whereas others do not is the problem that motivated this study. This study
examines the factors of customers’ technology adoption based on the TAM. E-banking adoption is studied from the information
systems acceptance point of view referring to idea that customers must use the information system to make bank transactions and
hence more knowledge on the factors that affect IT adoption is needed in order to better understand and facilitate the acceptance.
Perceived ease of use, perceived usefulness, subjective norm, security and trust, Internet experience, and enjoyment are the important
factors that affect customers’ adoption of e-banking in Jordan.
Keywords: Electronic banking, Structural Equation Modeling, Technology Acceptance Model
1- Introduction
The world is changing at a staggering rate and information technology (IT) is considered to be the key driver for
these changes. Based on numbers published by Internet World Stats (2008), there are approximately 1.5 billion
Internet users around the world and approximately 42 million Internet users in the Middle East. Internet and
advances in information and communication technologies (ICT) have had a profound effect on the banking industry.
Porter and Millar (1985) have found that banking is one of the most information-intensive sectors. Many of the
current studies in the banking industry have been conducted in developed countries. Hence, the current study
attempts to contribute by focusing on the banking industry of a developing economy, Jordan.
2- E-banking Definitions and Benefits
E-banking is “an umbrella term for the process by which a customer may perform banking transactions
electronically without visiting a brick-and-mortar institution. The following terms all refer to one form or another of
e-banking: personal computer (PC) banking, Internet banking, virtual banking, online banking, home banking,
remote electronic banking, and phone banking. PC banking and Internet or online banking is the most frequently
used designation” (www.bankersonline.com). E-banking technology could represent a variety of different services,
ranging from common automatic teller machine (ATM) services, direct deposit to automatic bill payment (ABP),
electronic transfer of funds (EFT), and computer banking (PC banking) (Kolodinsky et al., 2001). Gopalakrishnan et
al. (2003) defined e-banking as “the use of the Internet as a remote delivery channel for banking services”. In this
research, e-banking is defined as “Using Internet and websites for doing transactions without physical presence of
customers on the physical environment of a bank”.
The main benefits of e-banking for customers are: speed, convenient, available, accessible, time saving, and cost
saving. The main benefits of e-banking for banks are: more efficient, reducing costs, eliminating location
constraints, expansion of reach, and dependence on information technology. However, these benefits of e-banking
cannot be achieved unless customers use the e-banking services.
3- E-banking in Jordan
The banking sector is very dynamic and liberal in Jordan. It comprises 23 banks, eight of which are subsidiaries of
foreign banks and two are Islamic banks (CBJ, 2008). A new banking law, which aims at improving the industry's
efficiency, came into force in 2000, the new law protects depositors’ interests, diminishes money market risk, guards
against the concentration of lending and includes articles on new banking practices (e-commerce, e-banking) and
money laundering. Some banks have started adopting modern banking practices such as automated check clearing
and the use of magnetic check processors, unified reporting forms and electronic data-transmission networks. Most
of the banks in Jordan have started e-banking services. Table 1 summarizes the names of the main e-banking
services that provided by Jordanian banks (Source: Banks’ Websites, 2008).
132

4- Theoretical Background
Four main research models have been developed in the area of information technology acceptance studies. The first
model was the theory of reasoned action (TRA) (Fishbein and Ajzen, 1975) and its extended model, the theory of
planned behavior (TPB) (Ajzen, 1985) which viewed a person’s intention to perform or not to perform a behavior as
the immediate determinant of the action. The second model was adapted theoretically from TRA is TAM (Davis,
1986) that developed and validated scales for two main variables, perceived usefulness and perceived ease of use.
The two variables have been hypothesized to be fundamental factors of user acceptance of information technology
(see Figure 1). The third model was a theoretical extension of TAM, known as TAM2, (Venkatesh and Davis, 2000).
This model explained perceived usefulness and usage intentions in terms of social influence and cognitive
instrumental process. The fourth model was the innovation diffusion theory (IDT) perspective (Rogers, 1995). This
decision process theory proposed that there exist five distinct stages to the process of diffusion. These are
knowledge, persuasion, decision, implementation and confirmation. These models addressed all direct or indirect
influences in actual usage behavior.
In this study, an extended version of the technology acceptance model (TAM) will be developed to investigate the
factors that affect customers’ adoption of e-banking in Jordan.
5- Methodology
This study will use a theoretical model to examine and predict Jordanian customers’ acceptance of the e-banking
system. Technology Acceptance Model (TAM) was used to construct the proposed model of customers’ acceptance
of e-banking. The first step is to construct the proposed model. Then, the data will be collected and analyzed in the
second step. In this research, the first step will be completed and to be continuo in the future. A majority of studies
using TAM have relied on survey methodology. The survey method will be similar to that used in previous TAM
studies, to maintain continuity in the research regarding perceived usefulness and perceived ease of use. The
addition of new constructs will require supplementary questions that follow previous work in the field. In the future,
data will be gathered through the use of questionnaire-style survey given to banks’ customers in Jordan.
A survey will be used to test the hypotheses regarding the structure of the proposed model. To analyze the data, the
structural equation modeling (SEM) techniques will be applied. According to SEM techniques, t-test of significance
will be conducted to test the hypothesized relationships of factors in the proposed model. Then, path model will be
utilized to analyze relationships between the factors in the research model to explain customers’ adoption of ebanking.
6- Literature Review: Factors Affecting Customers’ Adoption of E-Banking
6.1 Subjective Norm
Subjective norm defined as “the person’s perception that most people who were important to him think he should or
should not perform the behavior in question” (Fishbein and Ajzen, 1975). Davis (1986) omitted subjective norm in
the original TAM. Several studies found no significant effect of subjective norm on behavioral intention. For
example, Mathieson (1991) found no significant effect of the subjective norm on intention to use. On the contrary,
some studies found significant effect of subjective norm. For example, Venkatesh and Davis’ (2000) study found
that the effect of the subjective norm had a direct effect on the intention to use but the subjective norm did not have
a significant effect on intention to use after a three month post implementation of the information technology in a
voluntary setting. Taylor and Todd (1995) did find a significant direct effect of subjective norm on intention to use
with potential information technology users. Thus, subjective norm contributed to the explanation of intention to use
information technology with subjective. Accordingly, SN was hypothesized to affect the customers’ intention to use
e-banking:
H1: Subjective norm has a significant effect on intention to use e-banking.
The relationship between SN and PU was supported by TAM2 as a significant relationship (Venkatesh and Davis
2000). Research on SN concluded that this construct was the weakest among the TPB set of constructs (Hu and
Chau, 1999). Even though Davis (1989) concluded that SN is not a determinant of behavioral intention (BI), other
studies established a significant relationship between SN and both PU (Venkatesh and Davis, 2000) and BI (Ajzen,
133

1991; Riemenschneider et al., 2003; Taylor and Todd, 1995). Thus this relationship has been proposed, supported
and tested in more than one model (TAM, TRA, TPB, and DTPB). Accordingly, SN was hypothesized to affect the
PU:
H2: Subjective norm has a significant effect on perceived usefulness.
6.2 Security and Trust
Security has been widely recognized as one of the main obstacles to the adoption of e-banking and it is considered
an important aspect in the debate over challenges facing e-banking. Some studies (e.g. Khalfan et al., 2006; Al-
Sabbagh and Molla, 2004) conducted in the Omani banking industry, reported that security concerns have been one
of the major issues in the e-banking adoption. Mattila and Mattila (2005) also claimed that security has been widely
recognized as one of the main barriers to the adoption of Internet innovation.
Trust is construct that is closely related in the literature. The work of Pavlou (2003) supported the direct effect
of trust on intention. Other work emphasized the role of trust in predicting usefulness, intention and usage (Suh and
Han 2003; Gefen et al., 2000). Also, Warkentin et al. (2002) concluded that trust can influence intention.Trust is
“The willingness of a party to be vulnerable to the action of another party based on the expectation that the other
will perform a particular action important to the trustor irrespective of the ability to monitor or control that other
party” (Mayer et al., 1995). In other words, it implies that something of importance could potentially be lost as a
result of engaging in the trusting relationship.
In e-banking context, the customer willing to be dependent upon the Internet based on the expectation that it
will perform what the customers expect it to do. Because e-banking is highly uncertain (Clarke, 1997), it is expected
that trust is one of the main factors influencing the acceptance of e-banking. Gefen (2003), Palvou (2003) and others
extended the TAM model to include trust in online based business setting. Trust was hypothesized to be an
antecedent of PU and PEOU. Researchers (e.g. Gefen, 2003; Newell et al., 1998) have empirically verified that
customer trust has an impact on store loyalty. According to the literature and TAM model, it is expected that
security and trust will have a significant effect on behavioral intention through PU and PEOU:
H3: Security and trust have a significant effect on perceived usefulness.
H4: Security and trust have a significant effect on perceived ease of use.
6.3 Internet Experience
Experience of users was a significant factor in many studies in the technology acceptance research. Additionally, it
has been utilized in several studies as a predictor or moderator in the technology acceptance research (e.g. Speier
and Venkatesh, 2002; Venkatesh and Davis, 2000; Igbaria et al., 1995; Harrison and Rainer, 1992).
Igbaria et al. (1995) and Zmud (1979) stated that prior experiences has long been regarded a factor identifying
individual differences in technology acceptance research. Prior experience, such as computer or Internet use
experience was supported in several studies as strongly influencing intention to use of a specific system through
perceived usefulness and perceived ease of use (Chau, 1996; Agarwal and Prasad, 1999). This was because as
people get more experiences about the system and learn necessary skills, they are likely to develop more favorable
perception of its ease of use (Hackbarth et al., 2003). People tend to adopt information systems are that compatible
to those previously adopted and used (Dearing and Meyer 1994). This is consistent with the study of Eastin (2002),
where the experience in telephone usage positively influenced the adoption of on-line shopping, banking and
investing. Similarly, in the study of Liao and Cheung (2001) about Internet-based e-shopping behavior in Singapore,
IT education and Internet experience were significant antecedents in the development of people’s intention to eshopping.
Anandarajan et al. (2000) found that perceived usefulness was related to time spent on the Internet. Ease
of use correlated positively with use of the Internet for business activity.
Based on evidence in prior researches, customers’ experience with Internet was conceptualized as an external
variable. In addition, to explain user beliefs concerning usefulness and ease of use toward e-banking, prior
experience on the Internet has to be considered.
H5: Internet experience has a significant effect on perceived usefulness.
H6: Internet experience has a significant effect on perceived ease of use.
134

6.4 Enjoyment
Enjoyment refers to the extent to which the activity of using a computer is perceived to be enjoyable in its own right
(Davis et al., 1992). This is contrasting to the PU, which can be seen as an extrinsic motivation whereas perceived
enjoyment (PE) as an intrinsic motivation to use information systems.
Davis et al. (1992) findings suggest that increasing the enjoyability of a system would enhance the acceptability
of IT. Later, other studies have proved this relationship in different contexts like: the Internet Usage (Teo et al.,
1999; Agarwal et al., 2000; Moom and Kim, 2001), the use of the personal computer (Igbaria et al., 1997),
enterprise applications (Venkatesh and Davis, 1994) and a high school e-learning system (Lee, 2006). In e-banking
context, Pikkarainen et al. (2004) added perceived enjoyment, information on online banking, security and privacy
and quality of Internet connection to the consumers’ acceptance of e-banking model. Yi and Hwang (2003) again
studied self-efficacy, enjoyment, and learning goal orientation in the context of TAM with university students. Selfefficacy
appeared to directly influence the use, whereas enjoyment and learning goal orientation mediated through
self-efficacy, usefulness and ease of use. Usefulness and ease of use in turn influenced the decision to use through
behavioral intention. Baierova et al. (2003) stated that people using the web for entertainment valued enjoyability.
The enjoyability of Internet as a channel to make transactions online and based upon the preceding research and
TAM model, the following hypothesis are proposed:
H7: Enjoyment has a significant effect on perceived usefulness.
H8: Enjoyment has a significant effect on perceived ease of use.
6.5 Behavioral Intention, Perceived Ease of Use, and Perceived Usefulness
In Davis et al’s (1989) study, the TAM model and TRA model were compared, added behavioral intention in their
model, and found a strong relationship between perceived usefulness (PU) and intention (perceived usefulness
accounted for 57% of behavioral intention’s variance). According to the technology acceptance model, the theory of
reasoned action, and the theory of planned behavior, behavioral intention was the most appropriate predictor of
actual use (Ajzen, 1985; Ajzen and Fishbein, 1980; Davis et al., 1989; Fishbein and Ajzen, 1975). Perceived ease of
use and perceived usefulness were fundamental determinants of user acceptance in the technology acceptance model
(Davis, 1989). Computer with perceived ease of use, perceived usefulness has been a strong determinant of usage
intentions. The understanding of determinant factors of perceived usefulness as well as perceived ease of use was
important. Venkatesh and Davis (2000) extended the technology acceptance model, including key determinants of
the technology acceptance model’s perceived usefulness, to understand how the effects of these determinants were
changing with information technology. They found subjective norm, voluntariness, image, job relevance, output
quality, and result demonstrability significantly influenced a user’s technology adoption.
Based on the studies reviewed above, the proposed study model tested proposed hypotheses of modified technology
acceptance model with e-banking domain:
H9: Perceived usefulness has a significant effect on intention to use e-banking.
H10: Perceived ease of use has a significant effect on intention to use e-banking.
H11: Perceived ease of use has a significant effect on perceived usefulness.
The proposed model is shown in Figure 2 with arrows representing direction of relationships. The structural model
is shown in Figure 3.
7- Conclusions and Discussions
The primary objective of the study was to study customers’ acceptance of e-banking in the context of the technology
acceptance model (TAM) added with new variables derived from e-banking acceptance literature on one hand and
from an interviews with e-banking specialists on the other. Because of that, this research has extended the TAM in a
previously unexplored direction. The proposed model showed that, as hypothesized, customers’ intent to use the ebanking
is affected by perceived ease of use, perceived usefulness, subjective norm, Internet experience, security
and trust, and enjoyment.
135

The empirical examination of the adoption of e-banking using a structural model based on the extension of
TAM will be tested and validated in the second step. This research will provide further evidence of the
appropriateness of applying TAM to measure the acceptance of e-banking. The proposed model identified the major
factors affecting customers’ adoption of e-banking. These factors were perceived ease of use, perceived usefulness,
subjective norm, Internet experience, security and trust, and enjoyment.
8- Implications and Contributions
The study makes significant contribution across all area of IT adoption and usage research and practice. A summary
of the main contribution are:
1. The development of a conceptual model that explains and predicts the factors that influence the adoption of
information technology in general and e-banking in specific.
2. The empirical support for proposed hypotheses based on the integrative research framework and literature.
3. The proposed model can be used to predict successive usage behavior of customers of e-banking systems.
4. The proposed model extends the domain of the TAM to customers’ adoption of e-banking.
5. The proposed model expanded the technology acceptance model by including subjective norm, Internet
experience, security and trust, and enjoyment as external factors.
6. The results of this research provide managers with information about the planning of e-banking Websites
and service selection in future.
9- Limitations and Future Research
The findings reported in this research may be limited to some issues. For example, e-banking (using Internet) is only
one of the various types or forms of online banking products available. Thus, the relationships found in this research
may or may not differ according to whether consumers are using telephone banking, electronic money transfer,
direct bill payment or other forms. The results may generalize to e-banking (using Internet) as a service. In order to
prove the proposed model explanatory ability, the present model has yet to be tested on different technologies in the
banking industry (for instance, mobile banking) or technological innovations in general (for instance, video on
mobile phones).
Behavioral intention is the closest construct that can be used as a surrogate for e-banking usage. Using
behavioral intention is rich, but does not replace exploring actual usage of a system. In this study, a self-reported
measure provided by respondents was used to measure their intention to use e-banking, which might not be the best
measure. A more accurate measure would be to explore users’ usage in a longitudinal study and control for actual
usage of the system. It is suggested that future studies should focus on refining the proposed variables to include
other factors (e.g. cost and quality of Internet connection, information on banks’ websites) with large samples and
investigate different types of technology innovations. This research provides the foundation for additional research
in the developing countries related to the technology acceptance domain such as adoption of e-government, ecommerce,
electronic customer relationship management (e-CRM), and e-learning.
This research used the TAM model as baseline model to find the main factors that affect customers’ adoption of
e-banking; other models could be used as baseline model such as TRA, IDT, TPB, and TAM2. Finally, banks
employees’ adoptions of e-banking, needs to be investigated since the employees are the other users of the system.
10- References
Agarwal, R. and Prasad, J. (1999) ‘Are individual differences germane to the acceptance of new information
technologies?’ Decision Sciences 30, (2) 361-91.
Agarwal, R., Sambamurthy, V. and Stair, R. (2000) ‘Research report: the evolving relationship between general and
specific computer self-efficacy: an empirical assessment.’ Information Systems Research 11, (4) 418-430.
Ajzen, I. (1985) From intention to actions: A theory of planned behavior. New York: Springer Verlag.
136

Ajzen, I. (1991) ‘The theory of planned behavior, Organization Behavior and Human.’ Decision Process 50, (1) 179-
211.
Ajzen, I. and Fishbein, M. (1980) Understanding attitudes and predicting social behavior. Prentice-Hall, Englewood
Cliffs, NJ.
Al-Sabbagh, I. and Molla, A. (2004) ‘Adoption and Use of Internet Banking in the Sultanate of Oman: An
Exploratory Study.’ Journal of Internet Banking and Commerce 9, (2) 1-7.
Alwin, D. and Hauser, R. (1975) ‘Decomposition of effects in path analysis.’ American Sociological Review 40, (1)
37-47.
Anandarajan, M., Simmers, C. and Igbaria, M. (2000) ‘An exploratory investigation of the antecedents and impact
of Internet usage: An individual perspective.’ Behavior & Information Technology 19, (1) 69-85.
Baierova, P., Tate, M. and Hope, B. (2003) ‘The impact of purpose for Web use on user preferences for Web design
features.’ 7th Pacific Asia Conference on Information Systems, 10-13.
Bollen, P. (1998) Structural equations with latent variables. New York: Wiley.
Byrne, K. (1998) Structural equations modeling with LISREL, PRELIS, and SIMPLIS: Basic concepts, applications,
and programming. Lawrence Erlbaum Associates, Mahwah, NJ.
CBJ (2008) Central Bank of Jordan. [online] available from [20
December 2008].
Chau, P. (1996) ‘An empirical assessment of a modified technology acceptance model.’ Journal of Management
Information Systems 13, (2) 185–204.
Clarke, A. (1997) ‘The Structural and Spatial Impacts of Bank Mergers.’ Abstracts of the Association of American
Geographers. Ft. Worth, TX, 44-45.
Cohen, J. (1988) Statistical power analysis for the behavioral sciences. Hillsdale, NJ: Lawrance Erlbaum Associates.
Davis, F. (1986) A technology acceptance model for empirically testing new end-user information systems: Theory
and results. (Doctoral dissertation, Sloan School of Management, Massachusetts Institute of Technology).
Davis, F., Bagozzi, R. and Warshaw, P. (1992) ‘Extrinsic and intrinsic motivation to use computers in the
workplace.’ Journal of Applied social Psychology 22, (14) 1111-1122.
Davis, F. (1989) ‘Perceived usefulness, perceived ease of use, and user acceptance of information technology.’ MIS
Quarterly 13, (3) 319-340.
Davis, F., Bagozzi, R. and Warshaw, P. (1989) ‘User acceptance of computer technology: a comparison of two
theoretical models.’ Management Science 35, (8) 982-1003.
Dearing, J. and Meyer, G. (1994) ‘An Exploratory Tool for Predicting Adoption Decisions.’ Science
Communication 16, (1) 43-57.
Eastin, M. (2002) ‘Diffusion of e-commerce: An analysis of the adoption of four e-commerce activities.’ Telematics
and Informatics 19, 251-267.
Fishbein, M. and Ajzen, I. (1975) Belief, attitude, intention and behavior: An introduction to theory and research.
Reading, MA: Addison-Wesley.
Fornell, C. and Larcker, D. (1981) ‘Evaluating structural equation models with unobservable variables and
measurement error.’ Journal of Marketing Research 18, (1) 39-50.
Gefen D., Straub D. and Boudreau, M. (2000) ‘Structural equation modeling and regression: Guidelines for research
practice.’ Communications of the Association for Information Systems 4, (7).
Gefen, D. (2003) ‘TAM or just plain habit: A look at experienced online shoppers.’ Journal of End User
Computing15, (3) 1-13.
Hackbarth, G., Grover, V. and Yi, M. (2003) ‘Computer playfulness and anxiety: Positive and negative mediators of
the system experience effect on perceived ease of use.’ Information and Management 40, (3) 221-232.
Hair, J., Black, B., Babin, B., Anderson, R. and Tatham, R. (2006) Multivariate data analysis. 6th ed, Englewood
Cliffs: Prentice Hall.
Harrison, A. and Rainer, R. (1992) ‘The influence of individual differences on skill in end-user computing.’ Journal
of Management Information Systems 9, (1) 93-111.
137

Hu, P. and Cahu, P. (1999) ‘Physician acceptance of telemedicine technology: an empirical investigation.’ Topics in
Health Information Management 19, (4) 20-35.
Hu, P., Chau, P., Sheng, O. and Tam, K. (1999) ‘Examining the technology acceptance model using physician
acceptance of telemedicine technology.’ Journal of Management Information Systems 16, (2) 91-112.
Igbaria, M., Gamers, T. and Davis, G. (1995) ‘Testing the determinants of micro-computer usage via a structural
equation model.’ Journal of Management Information Systems 11, (4) 87-114.
Igbaria, M., Zinatelli, P., Cragg, N. and Cavaye, L. (1997) ‘Personal computing acceptance factors in small firms: a
structure equation model.’ MIS Quarterly 21, (2), 279–305.
Internet World Stats (2008) Usage and population statistics. [online] available from
[10 August 2008].
Khalfan, A., AlRefaei, Y. and Al-Hajry, M. (2006) ‘Factors influencing the adoption of Internet banking in Oman: a
descriptive case study analysis.’ Int. J. Financial Services Management 1, (2) 155-172
Kline, R. (1998) Principles and practice of structural equation modeling. New York, NY: The Guilford Press.
Kolodinsky, J. and Hogarth, J. (2001) ‘The adoption of electronic banking technologies by American consumers.’
Consumer Interests Annual 47, (1) 1-9.
Lee, Y. (2006) ‘An empirical investigation into factors influencing the adoption of an e-learning system.’ Online
Information Review 30, (5) 517-541.
Liao, Z. and Cheung, Z. (2001) ‘Internet-based e-shopping and consumer attitudes: an empirical study.’ Information
and Management 38, (5) 299–306.
Mathieson, K. (1991) ‘Predicting user intentions: comparing the Technology Acceptance Model with the theory of
planned behavior.’ Information Systems Research 2, (3) 173-191.
Mattila, A. and Mattila, M. (2005) ‘How perceived security appears in the commercialization of internet banking.’
International Journal of Financial Services Management 8, (3) 206-217.
Mayer, R., Davis, J. and Schoorman, F. (1995) ‘An integrative model of organizational trust.’ Academy of
Management Review 20, (3) 709-734.
Moon, J. and Kim, Y. (2001) ‘Extending the TAM for a World Wide Web Context.’ Information and Management
38, (4) 217-231.
Newell, Stephen J., Goldsmith, Ronald E. and Banzhaf, J. (1998) ‘The Effect of Misleading Environmental Claims
on Consumer Perceptions of Advertisements.’ Journal of Marketing Theory and Practice 6, (2) 48-60.
Pavlou, P. (2003) ‘Consumer acceptance of electronic commerce: integrating trust and risk in the technology
acceptance model.’ International Journal of Electronic Commerce 7, (3) 101-134.
Pikkarainen, T., Pikkarainen, K., Karjaluoto, H. and Pahnila, S. (2004) ‘Consumer acceptance of online-banking: an
extension of the technology acceptance model.’ Internet Research 14, (3) 224–235.
Porter, M. and Millar, V. (1985) ‘How information gives you competitive advantage.’ Harvard Business Review 63,
(4) 149-160.
Riemenshneider, C., Harrison, C., Harrison, D. and Mykytyn, P. (2003) ‘Understanding IT adoption in small
business: Integrating current theories.’ Information and Management 40, (1) 269-285.
Rogers, E. (1995) Diffusion of innovations.4th ed, New York, Free Press.
Ross, D. (1975) ‘Direct indirect and spurious effects: comment on casual analysis of inters organizational relations.’
Administrative Science Quarterly 20, (1) 295-307.
Segars, A. (1997) ‘Assessing the un-dimensionality of measurement: A paradigm and illustration within the context
of information systems research.’ Omega 25, (1) 107-121.
Sekaran, U. (2003) Research Methods for Business: A Skill Building Approach. USA: John and Wiley Sons.
Speier, C. and Venkatesh V. (2002) ‘The hidden minefields in the adoption of sales force automation technologies.’
Journal of Marketing 65, (1) 98-111.
Suh, B. and Han, I. (2003) ‘The impact of customer trust and perception of security control on the acceptance of
electronic commerce.’ International Journal of Electronic Commerce 7, (3) 135-161.
138

Tabachnick , B. and Fidell, L. (2001) Using multivariate statistics. 4th ed, London: Allyn and Bacon.
Taylor, S. and Todd, P. (1995) ‘Understanding information technology usage: a test of competing models.’
Information Systems Research 6, (2) 144-176.
Teo, T., Lim, V. and Lai, R. (1999) ‘Intrinsic and extrinsic motivation in Internet usage.’ Omega International
Journal of Management 27, (1) 25-37.
Venkatesh, V. and Davis, F. (2000) ‘A theoretical extension of the technology acceptance model: four longitudinal
field studies.’ Management Science 46, (2) 186-204.
Venkatesh, V. and Davis, F. (1994) ‘Modeling the Determinants of Perceived Ease of Use.’ Proceedings of the
International Conference on Information Systems, Vancouver, BC, 213-228.
Warkentin, M. Gefen, D., Pavlou, P. and Rose, G. (2002) ‘Encouraging citizen adoption of e-government by
building trust.’ Electronic Markets 12, (3) 157-162.
Yi, M. and Hwang, Y. (2003) ‘Predicting the use of web-based information systems: self-efficacy, enjoyment,
learning goal orientation, and the technology acceptance model.’ Int. J. Human-Computer Studies 59, 431-449.
Zmud, R. (1979) ‘Individual Differences and MIS Success: A Review of the Empirical Literature.’ Management
Science 25, (10) 966-979.
Subjective
Norms
Security
and Trust
Internet
Experiences
Enjoyment
Figure 1: Technology Acceptance Model (TAM)
Perceived
Usefulnes
s
Perceived
Ease of
Use
139
Behavioral
Intention

e12
e13
e14
e16
e15
e19
e17
e18
e22
e20
e21
SN1
SN2
SN3
IE2
IE3
ST1
ST2
ST3
ENJ2
ENJ3
ENJ6
SN
IE
ST
ENJ
Figure 2: The Proposed Model
Structural Proposed Model
e5
PU1
e6
PU2
PU
PEU
e7
PU3
e1
IU1
PEU1 PEU4 PEU5 PEU6
e8
e9
Figure 3: Structural Proposed Model
e10
e11
Table 1: E-banking Services Provided by Jordanian Banks
Bank Name of E-banking Services (December 2008)
Arab bank Arabi online, Hala Arabi (Phone banking services), and SMS banking
(Notification services)
The Housing Bank for Trade and SMS, e-banking, and shopping via the Internet (UBU)
Finance
Arab Jordan Investment Bank Online banking / Jordan (for customers and employees)
HSBC Internet banking (Personal and Business)
Bank of Jordan The phone bank, the Internet banking, the mobile bank, and SMS
Ahli Bank Online banking
Islamic International Arab Bank Phone banking services, Internet shopping and services, SMS, and VISA
electronic card services
Jordan Kuwait Bank SMS, Internet banking, phone banking, cyber branch, “Western Union”
Money transfer, Zain pre-paid Mobile
Union Bank Union online, SMS services, and money gram transfers
JIF Bank Internet banking, phone banking, and mobile banking
Jordan Commercial Bank SMS, phone banking, and e-statement
Egyptian Arab land Bank Not available
Cairo Amman Bank E-channels (phone banking, Internet banking, and SMS banking)
National bank of Kuwait (NBK) Online banking service (Watani online)
Standard Chartered Bank E-Banking
ABC Bank Phone banking, Mobile banking, ABC online, and ABC SMS banking
Capital Bank of Jordan Bank E, e-banking, and SMS-banking
Societe Generale De Banque-
Jordanian
Internet banking
Rafidain Bank Not available
140
e2
IU2
IU
e3
IU3
e4
IU4

Audi Bank Dial Audi, Audi online, and mobile phone
BLOM Bank eBlom (Internet banking) and AlloBLOM phone
141

PROCYCLICALITY OF BANKS’ CAPITAL BUFFER IN ASEAN COUNTRIES
Elis Deriantino, Bank Indonesia
Email: elis_deriantino@bi.go.id
Abstract. By developing two models to estimate the effect of business cycle on bank’s capital buffer and the effect of Capital buffer
on bank’s loan supply on annual panel data (1997-2009) of 63 commercial banks in ASEAN countries, we find strong evidence of
procyclicality pattern of capital buffer among banks in ASEAN countries. Nevertheless, this procyclicality effect is somewhat small,
given that a decrease of 1 percentage point in GDP growth will reduce loan growth by around 0.4 percentage points due to the rise of
capital buffer. As Basel Committee for Banking Supervision (2010) proposes a new capital requirement regime to address issue of
procyclicality of capital requirement, this empirical finding may become input for country’s bank regulator in determining optimal
capital buffer level in such a way that it will effective to prevent volume of credit from being excessive during the upturn sides of
business cycle while provide banks greater resilience that enable them to continue reasonable lending activities during the downturn
sides of business cycle.
Keywords: Capital buffer, Procyclicality, Business cycle
JEL Classification: E32, G21
1 Introduction
The risk sensitivity of capital requirement as proposed by Basel (1988 Basel Standard and Basel II) framework is
considered leading to a certain degree of cyclicality in capital requirement that potentially will amplify business
cycle fluctuation through decreasing bank’s lending activity during the downturns side of business cycle hence pose
a threat to the stability of macroeconomic and financial system, in a so-called procyclicality of capital requirement.
Many previous studies, among those are the works by Bikker & Metzmeker (2004), Ayuso et al. (2002), Chiuri et al.
(2001) and Drumond (2009) confirm the procyclicality of capital requirement by pointing out to negative comovements
between business cycle and banks’ capital. However, in practice, most banks hold more capital (capital
buffer) than the regulatory minimum. Stronger supervision for market discipline, lesson learnt from the past crises
and the need to adopt a sound risk management to anticipate increasing probability of default during economic
downturns are some factors that motivate banks to hold more capital despite the fact it may more costly for banks to
hold more capital (Borio, et al (2001) and Ayuso et al. (2002)). This additional buffer should provide banks with
greater resilience that also enable them to maintain a reasonable volume of lending activities during economic
downturns. Study by Jokipii & Milne (2006) finds that capital buffers of RAM (10 countries that joined the
European Union (EU) in May 2004) banks correlate positively with the business cycle and banks in these countries
tend to hold more capital buffer than those of banks in other EU regions. This confirms well capitalized bank may
have countercyclical prudent behavior. Nevertheless, the above studies focus to identify the existence of
procyclicality of capital requirement but lack in detail assessment on effect of the pattern on banks’ lending activity.
Thus, even though it is generally considered that the capital requirement is procyclical, it still needs to assess
whether this pattern will have substantial impact to the volume of credit to economy. This implies two policy
questions, ie. (i) do bank’s capital buffer actually exhibit a significant procyclical pattern and (ii) does this pattern of
capital buffer constraint loan supply of banks substantially?
The aim of this paper is to answer the two policy questions. We employ annual bank-level panel data of 63
listed banks with coverage period 1997-2009 in five Association of South East Asian Nations (ASEAN) countries:
Indonesia, Singapore, Malaysia, Thailand and Philippine. To the best of our knowledge, none of existing studies has
explored the evidence from ASEAN countries using data period that covers two major crises that hit the region: the
1997/98 Asian Financial crisis and the 2008/09 Global Financial crisis. By examining the ASEAN data, the present
study will contribute to the literature on this area.
By developing two models to estimate the effect of business cycle on banks’ capital buffer and the effect of
Capital buffer on bank’s loan supply we find strong evidence of procyclicality pattern of capital buffer among banks
in ASEAN countries. Banks are found to reduce their loan growth during economic downturns due to a rise in
capital buffer as a result of impaired loan quality (rising NPL). Nevertheless, this procyclicality effect is somewhat
small, given that a decrease of 1 percentage point in GDP growth will reduce loan growth by around 0.4 percentage
142

points due to the rise of capital buffer while during the observed period banks’ loan growth average in those ASEAN
countries is around 11%. Moreover, the results also indicate that risk proxy NPL has significant and positive
relationship with capital buffer, meaning that banks with a relatively risky credit portfolio tend to hold more capital
buffer. This evidence which is also supported by the fact of ASEAN banks tendency to hold sizeable buffer above
minimum requirement (the average of banks’ capital buffer is around 13.5% above the country’s minimum
regulatory capital during the observed period) shows these ASEAN banks are adopting relatively sound risk
management that has contributed to moderate the effect of procyclicality of capital requirement. These prudently
capitalized banks come up after years of strengthened prudential regulation and supervisory framework as a result of
lesson learnt from the region own financial crisis in 1997/98.
2 Data, Methodology & Empirical Models
2.1 Data
We employ annual unbalanced bank-level panel data of 63 listed banks with coverage period 1997-2009 in five
ASEAN countries: Indonesia, Singapore, Malaysia, Thailand and Philippine. Due to the data availability, we select
banks that minimum have data coverage period from 2004-2009. The data period covers full business cycle in
respective countries and two major crises that hit the region: the 1997/98 Asian Financial crisis and the 2008/09
Global Financial crisis.
Banks indicators data are obtained from Bankscope while macro economy data of each country are from CEIC.
2.2 Methodology & Empirical Models
To address the two policy questions, we adopt the strategy of Wong et al (2010) by developing two models, ie:
1. Estimating the effect of business cycle on banks’ capital buffer
In this model, capital buffer (Buffer) is modeled as a function of business cycle (proxied by real GDP growth) with
control variables as prescribed in previous empirical studies including Return on Equity (ROE) as proxy for cost of
holding capital and ratio of Non Performing Loan (NPL) as proxy for banks’ risk profile:
Bufferi,t = α0 + α1Bufferi,t-1 + α2GDPj,t + α3NPLi,t + α4ROEi,t + μi + εi,t …………………………..(1)
Where i= individual bank index, 1,2…,N; j=country index, 1,2…,M; t=year index, 1,2…,T; μi captures
individual bank time invariant idiosyncrasies effect and εi,t is an error term.
The dependent variable Buffer is defined as additional capital that is set by individual bank in a country above
country’s minimum regulatory capital 1 , ie. bank’s Capital Adequacy Ratio - country’s regulatory minimum
requirement.
The inclusion of the first lag of Buffer as one of independent variable is intended to capture bank’s adjusting
cost (Ayuso et al, 2002).
Real GDP growth (GDP) is a proxy for business cycle indicator. A negative co-movement between GDP and
Buffer indicates the procyclicality pattern of capital buffer. On contrast, a positive relationship between capital
buffer and business cycle indicates banks adopt prudent capital behavior by increasing its capital during the upturns
side of business cycle in order to have adequate buffer to cover loss that most likely to increase as economic enter
downturn phase (Borio et al, 2001).
The cost of holding capital is proxied by Return on Equity (ROE), and its effect on Buffer is expected to be
negative.
Banks’ risk profile is proxied by Non Performing Loan ratio (NPL) and its impact on capital buffer is also
expected to be negative.
1 Minimum regulatory capital ratio for Indonesia and Malaysia: 8%, Thailand:8.5%, Philipine:10% and Singapore: 10% since May 2004 (12%
before May 2004).
143

Given the dynamic nature of this model due to the inclusion of lagged Buffer as an independent variable, we
estimate the model by employing two steps System Generalized Method of Moments system (GMM Sys) method.
We choose GMM Sys over differenced GMM because when T is small and series persistency is high (α1 close to 1),
the differenced GMM estimator has poor finite sample bias and low precision since lagged levels of the series
provide weak instruments for subsequent first differences while utilizing GMM Sys will reduce this finite sample
bias and increase the precision of estimator due to the exploitation of additional moment conditions (Blundell et al,
2000, Bond et al, 2001). By utilizing Monte Carlo simulation, Soto (2009) provides evidence that GMM Sys
generates lower bias and higher efficiency than other estimators for panel data with small number of cross section N
and T and high persistency series. We also estimate the model using Least Square Dummy Variable (LSDV) and
Ordinary Least Square (OLS) methods to check whether or not the coefficient of lagged capital buffer of GMM Sys
estimation is biased. An unbiased coefficient of lagged capital buffer should lie between those estimated by FELS
and OLS, given the fact that LSDV estimators are downward biased due to the negative correlation between the
transformed lagged dependent variable and the transformed error term whilst OLS estimators are upward biased due
to the positive correlation between the lagged dependent variable and the individual effects.
2. Estimating the effect of Capital buffer on bank’s loan supply.
In this stage, we model loan growth (Loan) as a function of banks’ capital buffer (Buffer), business cycle (GDP) and
interbank market interest rate (IR):
Loani,t = β0 + β1GDPj,t + β2IRj,t + β3Bufferi,t + ѵi + ζi,t ……………………………...(2)
Where i= individual bank index, 1,2…,N; j=country index, 1,2…,M; t=year index, 1,2…,T; νi captures
individual bank time invariant idiosyncrasies effect and ζi,t is an error term.
The procyclicality impact of capital buffer on bank loan growth will be indicated by negative co-movement
between Buffer and Loan. Both business cycle and interest rate are proxies for demand side factors of loan growth.
Business cycle is expected to have positive correlation with loan growth while interest rate will have negative
impact on loan growth.
We estimate model (2) using LSDV method. Finally, the procyclicality effect of capital buffer on lending
activity is then calculated as a multiplication of sensitivity of banks’ capital buffer to GDP growth in model (1) and
sensitivity of loan growth to capital buffer in model (2):
3 Empirical finding
3.1 Descriptive statistics
(α2*(mean GDP/mean Buffer) / (1-α1)) *( β3*(mean Buffer/mean Loan))
Table 1 below shows that in general banks in ASEAN hold a sizeable capital buffer. The resilience and risk profile
of banking system during the global financial crisis 2008/09 is much improved than a decade ago when Asia
financial crisis hit the region in 1997/98 as indicated by almost double Buffer for period of 2008/09 than that of
1997/98 as well as lower NPL in 2008/09 than that of 1997/98. The ASEAN region has also been considered to have
high potential to emerge as another economic force in the world. Recent economic development as indicated by
GDP growth has provided evidence of greater resilience of ASEAN countries in weathering the economic downturn
caused by the 2008/09 global financial crisis, particularly Indonesia and Philippine that despite lower economic
growth than those of previous years, still experience positive economy growth during the 2008/09 global crisis.
144

3.2 Regression Result
Mean Min Max St.Dev
Period: 1997-2009
Buffer 13.46 -8.5 133.3 18.10
Loan 10.53 -237.53 194.40 35.74
NPL 9.52 0.13 89.98 11.30
ROE 13.69 0.00 59.55 8.72
GDP 3.82 -13.13 9.24 4.24
IR
Period: 1997-1998 Asian Financial Crisis
7.55 0.44 51.06 5.77
Buffer 6.58 -8.00 37.20 9.95
Loan -21.18 -135.05 68.81 46.49
NPL 12.74 0.14 57.07 14.08
ROE 12.51 0.00 53.76 11.37
GDP -0.93 -13.13 8.55 7.53
IR
Period: 2008-2009 Global Financial Crisis
17.89 1.50 51.06 15.21
Buffer 11.46 1.88 121.00 15.54
Loan 11.82 -83.50 193.42 23.03
NPL 4.01 0.17 15.43 2.99
ROE 11.59 0.23 37.39 6.41
GDP 1.93 -2.30 6.01 3.95
IR 2.87 0.44 11.24 3.35
Table 1. Descriptive Statistics
The results on the table below suggest that strong evidence of procyclicality pattern of capital buffer among banks in
ASEAN countries. Banks are found to reduce their loan growth during economic downturns due to a rise in capital
buffer as a result of impaired loan quality (rising NPL). Nevertheless, this procyclicality effect is somewhat small,
given that a decrease of 1 percentage point in GDP growth will reduce loan growth by around 0.4 percentage points
due to the rise of capital buffer, while during the observed period banks’ loan growth average in those ASEAN
countries is around 11%. The estimation result of model (2) also indicates that credit rationing during economy
downturn in ASEAN banks is driven more by demand side factors rather than by the supply-driven capital buffer.
Dependent Variable
Buffer
Loan
Estimation Method
Independent variable
OLS LSDV GMM Sys LSDV
c -0.38 2.57 0.51*** 21.31***
[-0.26] [1.04] [2.96] [4.17]
Buffer(i,t-1) 0.88*** 0.65*** 0.80***
[26.08] [9.57] [293.56]
GDP(j,t) -0.27** -0.18** -0.45*** 1.88**
[-2.26] [-2.03] [-20.40] [2.30]
NPL(i,t) 0.28** 0.33* 0.49***
[2.38] [1.65] [41.87]
ROE(i,t) 0.08 0.04 0.02
[1.61] [0.78] [1.61]
IR(j,t) -1.30***
[-4.30]
Buffer(i,t) -0.51***
[-4.75]
Adj-R -sqr 0.79 0.81 0.19
DW 1.92 1.93 1.70
AR (1) (p-val) -1.92 (0.06)*
AR (2) (p-val) 0.48 (0.63)
Sargan test (p-val) 55.71 (0.37)
Table 2. Regression Result
Note: *,**,*** indicate a level of confidence of 90%, 95% and 99%, respectively.
Moreover, the results of model (1) also indicate that risk proxy NPL has significant and positive relationship
with capital buffer, meaning that banks with a relatively risky credit portfolio tend to hold more capital buffer. This
evidence shows that banks in ASEAN region are adopting relatively sound risk management that has contributed to
moderate the effect of procyclicality of capital requirement. This relatively sound risk management is also
145

supported by the evidence of the tendency of ASEAN banks to hold sizeable buffer above minimum requirement
(the average of banks’ capital buffer is around 13.5% above the country’s minimum regulatory requirement during
the observed period). These prudently capitalized banks come up after years of stronger supervision as a result of
lesson learnt from the region own financial crisis in 1997/98.
4 Conclusion and Policy Implication
By developing two models to estimate the effect of business cycle on banks’ capital buffer and the effect of Capital
buffer on bank’s loan supply on annual panel data (1997-2009) of 63 commercial banks in ASEAN countries, we
find strong evidence of procyclicality pattern of capital buffer among banks in ASEAN countries. Banks are found
to reduce their loan growth during economic downturns due to a rise in capital buffer as a result of impaired loan
quality (rising NPL). Nevertheless, this procyclicality effect is somewhat small, given that a decrease of 1
percentage point in GDP growth will reduce loan growth by around 0.4 percentage points due to the rise of capital
buffer.
As Basel Committee for Banking Supervision (2010) proposes a new capital requirement regime to address
issue on procyclicality of capital requirement, of which banks are required to set conservative capital and
countercyclical capital buffer, this empirical finding may become input for country’s bank regulator when
considering implementing this new capital regime. Taking into account the nature of their banks procyclicality
effect, banks’ regulator may determining optimal capital buffer level in such a way that it will effective to prevent
volume of credit from being excessive during the upturn sides of business cycle while provide banks greater
resilience that enable them to continue reasonable lending activities during the downturn sides of business cycle.
5 References
Ayuso, J., A. Gonzales, J. Saurina. 2004. Are capital buffers pro-cyclical?: Evidence from Spanish panel data.
Journal of Financial Intermediation 13,249-264.
Basel Committee on Banking Supervision. 2010. Countercyclical capital buffer proposal. Consultative document.
Bank for International Settlements.
Bikker, J., P. Metzemakers. 2004. Is bank capital procyclical? A cross country analysis. DNB Working Paper
No.009/2004, De Nederlandsche Bank NV.
Blundell, R., Bond, S., and Windmeijer, F. 2000. Estimation in dynamic panel data models: improving on the
performance of the standard GMM estimators. The Institute of Fiscal Studies Working Paper, No. 00/12.
Bond. S., Leblebiciouglu, A., and Schiantarelli, F., 2004, GMM estimation of empirical growth models, mimeo,
September 2001.
Borio,C.,C.Furfine, and P. Lowe. 2001. Pro-cyclicality of the Financial System and Financial Stability: Issues and
Policy Option, BIS Papers, No. 1.
Chiuri M C, Ferri G and Majnoni, G, 2001. The macroeconomic impact of bank capital requirements in emerging
economies; past evidence to assess the future, mimeo, World Bank.
Financial Stability Forum, 2009. Addressing Procyclicality in the Financial System.
Soto, M. 2009. System GMM estimation with a small sample. Institut d’Analisi Economica, Barcelona.
Wong, E., Fong, T., and Choi, H. 2010. An empirical assessment on procyclicality of loan-loss provisions of banks
in EMEAP Economies. Presentation delivered at 11 th Annual Bank of Finland/CEPR conference, Helsinki 7-8
October 2010.
Drumond, I. (2009). Bank Capital Requirements, Business Cycle Fluctuations and the Basel Accords: A Synthesis.
Journal of Economic Surveys, Vol. 23, Issue 5, pp. 798-830.
Jokipii, T and Milne, A. (2006). The cyclical behavior of European bank capital buffer. Research Report. Swedish
Institute for Financial Research.
146

DETERMINING EFFECTIVE INDICATORS FOR CUSTOMER PERFORMANCE IN REPAYING BANK
LOAN (CASE STUDY: IRANIAN BANKS)
Fariba SeyedJafar Rangraz, Amirkabir University of Technology
Dr. Naser Shams, Amirkabir University of Technology
EMAIL: RANGRAZFARIBa6384@gmail.com
Abstract Nowadays, decision making in the field of investment and banking has become an important issue. This is due to unstable
economic conditions and unknown factors leading to people’s success. Developing countries are faced with limited capital. Hence,
there is less attention to bank customers and a framework has not been developed to scientifically evaluate the project process.
Therefore, there are lots of unfinished or lost projects in those countries, including Iran.
This paper aims to evaluate the money lending process based on Iranian bank’s customers. The 5C’s model will be used to measure
the relationship between each element of the model by taking into consideration the factors leading to customer satisfaction. A
quantitative method will be used to measure the findings. From an Iranian banks expert’s point of view, character is the most
important index among 5C’s in order to lending and it has the highest value.
Key words: Banking, 5C’s, lending, repay, scoring
1. Introduction
Decreasing risks and the descending possibility of not repaying are ways used by banks referred to as Computation
technique. However, the unstable economy and external environmental factors force banks to review their method of
decision making.
K. Bryant has presented a model for evaluating agronomic loans (Bryant, k.,2001). He has used Duchessi
research for designing his model. According to Bryant, it is necessary the lender knows 5 factors when consumer
credit is being evaluated (Duchessi, P., 1995). These factors are: credit, capital, capacity, collateral and character.
Bryant introduced an expert system for decision making (ALEES). In this system, quality factors such as skill,
experience and the intelligence of the loan expert are keys while quantity factors are mixed.
Rosman and others have studied different strategies that lenders use for confirming borrowers. They suggest
that the structure of the lending process is in 2 parts: financial and nonfinancial. There are some features in each part
(Rosman Andrea., and Bedard, Jean, 1999)
Tansel and Yardakhl presented a model for evaluating decisions. This model is based on determining the credit
of manufacturing firms that want to borrow from the Turkish banks (Yardakhl, M. , and Tansel, Y., 2004). In this
model, all quality and quantity indexes related to the profitability of a company have been recognized. The obtained
results can give a special score to the company using the AHP technique.
Chan and Mak presented a method showing the benefits of selecting an improved manufacturing technique
consisting of the following three approaches (Chan, F.T.S., and Mak, K.L. , 2000):
� Strategic approach: achieving common goals
� Economic approach: economic benefits
� Analytical approach: economical and non-economic benefits and risks
Reitan introduced efficient factors when evaluating loans. He estimated the ability of experts in determining the
value and measuring factors of the project. He compared them based on the importance of each factor. He also
concluded why certain evaluating processes are incorrect. He categorized 42 indexes into 6 groups. These are the
Manager’s character and experience, services offered, market conditions, different pertinent aspects of the
organization and financial matters (Reitan, Bjornar., 1998).
In this paper, the factors affecting the lending decision making process has been studied. Different results due to
various situations, features and abilities of the borrower have been attained.
147

This information is categorized in 5 groups. The effective factors in bank decision making are named the 5C’s.
They are (Tim Hill of COCC):
� Character
� Capacity
� Condition
� Capital
� Collateral
1.1. Character
Character is the customer’s impact on the lender. The lender decides whether the customer is reliable to repay
Therefore, the customer’s education and experience in that project would be surveyed. Also, an employee’s quality
of performance and experience should be considered (Tim Hill of COCC).
1.2. Capacity
The capacity of the loan applicant refers to their imbursement ability and it is evaluated using expected cash flow.
For those customers requesting short term loans, banks experts would pay attention to the method of payment such
as cash. Other factors considered are the customer’s assets which can be transformed to cash immediately. Future
cash flow and the company’s potential in honoring short time commitments are considered too. In contrast, clients,
who extend long term loans, usually focus on long term assets. Other factors considered are the difference between
the assets and debts, permanent activities, profitability and the company’s potential (Accounting organization of
Iran, 2007).
1.3. Capital
Capital is equal to the net worth or net asset. Moreover, it is the reflection of the company’s success and value.
Although this criterion can be sinister with considering variability of asset worth and official worth, official accounts
may not show asset market worth. However inflation and increase in price level is the reason for this and sometimes
account based asset value is less than market worth. But, considering worth based on asset official worth is not
critical for the lender. Capital is the money which a client invests personally on a project and it is explanatory of
corporation in loss if the firm is broken (SBA, North Catalina). Perhaps it’s best to use the word ‘bankrupt’ or
‘folds’ instead of ‘broken’.
1.4. Collateral
Collateral is studied from a lending aspect and is one of the basic factors in the lending decision making process.
The other notable role of collateral is being one of the effective elements in evaluating the lender and nominating his
credit condition. Regardless of how much the employees of a bank are careful and cautious in lending; there
sometimes would be problems in repayment. In this situation, the alternative solution is collateral. The Bank must
decide what type of collateral is needed to recoup the loan. Property, a house or land, equipment or personal
assurance may be required (Edward Poll,J.D.,M.B.A.,CMC,2003).
2. Research method
The method of research for this paper has been based on Hypothesis testing, but research goals have not been
limited in edge of Hypothesis testing Other testing methods besides Hypothesis testing have also been used. This
paper is heuristic kind, from environmental studies aspect it is survey kind. Due to conceptual base, this paper is
descriptive because it studies what it is. The technique used for gathering information was a questionnaire.
Some of Iranian banks experts were interviewed and finally a questionnaire was selected as the best means for
gathering information for this research. For evaluating indexes, a questionnaire was designed consisting 26
questions. These questions have been categorized in 5 groups like the 5C’s indexes.
148

It should be mentioned that the Likert scale used for scoring factors 1,3,5,7,9 were used to show more variance
among groups instead of 1,2,3,4,5. So, the accuracy rose.1 was defined as the less important item and 9 as the most.
The description of scoring can be seen in the appendix.
The statistical society for this research consists of the Iranian banks experts. Selected banks were from
government owned banks and informal banks in order to gather a comprehensive aspect of Iranian banks instead of a
specific kind.
The sampling method in this study was random.70 of questionnaires were gathered completely. You can see the
questionnaire that is used in this research in the appendix.
2.1. Research hypothesize
The main hypothesis was: “customer financial indexes are notable in bank expert decision making.”
The secondary hypothesis was: “how loan applicant character, capacity, capital, condition and collateral are
effective in bank expert decision making.”
2.2. Method of scoring indexes
The Friedman Test was used for credit scoring but in this method, the weight of indexes, which could be achieved
from number of questions, has not been attended. For solving this problem, this method was used:
� Firstly, experts score questions using the scale that was mentioned.
� For the grading indexes, the means of each question scores was estimated and then mean of each index
scores was calculated.
� Then, it was considered every question has 10 points. So the number of questions for each index was
multiplied by 10. Afterwards, the result was divided by the number of all questions (26). The final result
was the coefficient. The total of these coefficients important should be equal to 100.
� In next step, the means of each factor was multiplied by coefficient.
� This number was multiplied by sum of mean*coefficient of each index, and finally the score of each index
was obtained.
3. Tests and results
For evaluating reliability of the questionnaire, Cronbach's Alpha test was applied using SPSS software, the result
showed that Cronbach's Alpha was higher than 0.7 which proved the reliability of the answers.
Before studying the results of the different tests, it should be mentioned that these tests were conducted using
SPSS software.
3.1. BOXPLOT
This graph was used for diagnosing outliers and controlling symmetric of data distribution. In graph below
BOXPLOT of data can be seen:
149

Figure1-BOXPLOT of data
The graph illustrates Skewness to left in character, condition and collateral can be seen obviously. Also Skewness to
left in capacity is clear. Data distribution in capital is symmetric. In collateral, outlier can be seen.
3.2. Histogram method
Histograms of indexes were extracted for checking normalization. The histogram of character data is shown in
figure2 as an example.
.
Figure2 - Histogram of character data
Figure 2 and other Histograms show distribution of none of the societies is normal.
3.3. QQPLOT
QQPLOT was used for studying normalization. QQPLOT Graph for Character is shown in figure3 as an example:
150

Figure3-QQPLOT for character data
Figure 3 and other QQPLOT graphs show that none of the societies are normal.
3.4. Kolmogorov-Smirnov Test
Kolmogorov-Smirnov Test for studying normalization was used, results are below:
Character Capacity Capital condition collateral
N 576 560 140 280 140
Normal Parameters a
Mean 6.8889 6.2643 6.9429 5.8143 8.2857
Std. Deviation 2.29665 2.08276 1.69541 2.07247 1.22495
Most Extreme Differences Absolute .255 .220 .228 .216 .420
Positive .179 .146 .187 .153 .280
Negative -.255 -.220 -.228 -.216 -.420
Kolmogorov-Smirnov Z 6.121 5.211 2.695 3.621 4.971
Asymp. Sig. (2-tailed) .000 .000 .000 .000 .000
a. Test distribution is Normal.
Table1-results of One-Sample Kolmogorov-Smirnov Test
Results of Histogram part, QQPLOT part, and Kolmogorov-Smirnov Test part show that societies are not
normal and T-test or Z-test can’t be used.
3.5. Coherency test of samples
Non-parametric Friedman Test was used for comparing coherent coherence samples. These hypothesize were
considered:
H 0: There is no coherence among 5 indexes.
H 1: 5 indexes are independent.
N 140
Chi-Square 527.79
Df 4
Asymp. Sig. .000
Table2- results of Friedman Test
151

Table 2 shows that indexes are important in this order: collateral, capital, condition, character and capacity.
(This result does not consider coefficient). Also results show that in the significant level of � = 0.05,
5 indexes are
interdependent.
4. RESULT OF INDEXES SCORING
The previous mentioned method was used for scoring indexes, results of that could be seen in table3:
Index Sum(Mean of each factor* coefficient)/
(Mean of each factor* coefficient)*100
Character 39.95
Capacity 28.98
Capital 8.03
Condition 13.45
Collateral 9.58
Table3-results of indexes scoring
Table 3 shows that indexes scoring ends to this order: Character, Capacity, Condition, Collateral, and Capital.
As it is seen, from Iranian banks expert’s point of view, character is the most important index among 5C’s and it
has the highest value.
5. Conclusion
According to the final results, Iranian banks experts believe that character is the most important factor among 5C’s
and they pay attention to this more. Character refers to tendency of costumer to honor his commitments well. It can
be said that character means commitment. Furthermore, people’s effort is as their character in order to develop a
successful project. Moreover, ability to repay on time, the loyalty of employees, their educational background, and
duration of staying in a stable place are among these.
Besides, capacity is seen as the second important factor in Iranian banking. The main factors in decision
making are evaluating financial power and profitability, forecasting future activities, information about cash flow
and the method of achieving and consuming cash. The latter are also the basic parameters for viewing capacity.
Capacity or surveying cost ratio is achieved from product cost and capital cost. Increasing costs and
decreasing income that id related to a low sale can affect loan repayment and have a negative impact on repayment
costs.
The applicant’s power of loan repayment is highly related to their ability to use resources correctly and gain
profit from economic unit facilities. Then lending should be performed after estimating real necessity of the
company. It should be proportionate to the company’s ability in using cash.
Financial analysis techniques could be used when surveying risks and other effective factors in evaluating
company’s capacity.
Furthermore, among these 5 indexes, customer capital has little score. Capital represents the customer’s
share in the project, so this concept is an important factor in showing the customer’s commitment to do the project
as well as he can.
While a firm uses bank loan in order function commercially, the proportion of capital sources for the firm
when acceptance of financial share is high, the risk would transfer from bank to the firm. This is due to the unclear
financial position of firm.
152

At the end, the lowest score is given to Collateral. Collateral provides a mechanism to repay a specific debt
when delinquent, at this point; the banks experts are sure about repayment.
Finally, collateral is identified as having the least value among 5C’s from Iranian banks experts’ point of view.
As discussed above, all the items are important in decision making and none can be ignored. Recognition of the
importance of these aspects helps banks to correct their method of decision making. It should be noted that ignoring
items like collateral and condition could become a risk for bank capital.
6. Refrences
Bryant, k.;"An agricultural loan evaluation expert system", Expert system with application, 2001, pp75-85.
Duchessi, P.;"A knowledge engineered system for commercial loan decisions", Financial management, 1995, (17-3),
pp 57-65.
Rosman Andrea., and Bedard, Jean."Lenders Decision Strategies and loan structure decisions", Journal of Business
Research, 1999, pp.83-94.
Yardakhl, M. , and Tansel, Y. ;"AHP approach in the credit evaluation of manufacturing firms in Turkey'', Int. J. of
production Economics, 2004(88),pp. 269-289.
Chan, F.T.S., and Mak, K.L. ;"An Integrated Approach to Investment Appraisal for Advanced manufacturing
technology". ,Human factors and Ergonomics in manufacturing,2000(19)
Reitan, Bjornar.;" Criteria Used by Private Bank Officers to Evaluate Nea Ventures: An Analysis of Gaps and
Shortcomings" http://www.sbaer.uca.edu/Research/1998/ICSB/n015.htm
Tim Hill of COCC, "The 5C's getting money from a bank"
Accounting organization, "Iranian accounting based theories", publishing committee of standards of Iran accounting,
Tehran 2007
"The 5C's of credit-small business resource", SBA(US Small Business administration)-North Catalina
Edward Poll,J.D.,M.B.A.,CMC, " Understanding the four C's", Lawyers and bank loans ,law practice TODAY,
september2003
7. APPENDIXES
Questionnaire:
How much does each of these factors effect in tour decision to lend?
For scoring factors, consider that:
1= means that this item is the less important item from banks experts point of view for lending.
3= means that this item is a few important item from banks experts point of view for lending.
5= means that this item is a neural item from banks experts point of view for lending.
7= means that this item is an important item from banks experts point of view for lending.
9= means that this item is the most important item from banks experts point of view for lending.
Character questions:
Customer education: …………..
Company activity in related industry: …………..
Customer credit in industry: …………..
Customer background in past loans and his repayment method in the past: …………..
Commercial fame: …………..
Loan repayment: …………..
Loan application times: …………..
Customer credit background in other banks: …………..
Company management constancy: …………..
Profitability: …………..
Capacity questions:
Company financial ability in repaying: …………..
153

Company financial statements: …………..
Fixed asset value: …………..
Liquidity asset and cash flow: …………..
Obtaining financial resource method: …………..
Product variety: …………..
Competitors: …………..
Target market: …………..
Capital questions:
Account based asset value: …………..
Company management ability and his background: …………..
Condition questions
Encountering national economical fluctuation: …………..
Encountering international economical fluctuation: …………..
Economic and politic society condition: …………..
Place of related industry in society: …………..
Collateral questions:
Amount of customer collateral: …………..
154

TWO-STAGE DEA APPROACH TO EVALUATE THE EFFICIENCY OF BANK BRANCHES
Akram Bodaghi & Heidar Mostakhdemin Hosseini
Mazandaran University of Science and Technolog & Tosse-Eh Farda Bank, Iran.
Email: akram.bodaghi@gmail.com, m_hosseini@cid.ir, www.cid.ir
Abstract. In emerging market economies, performance analyses in the services industries, specially banking sector, take more and
more attention. This paper illustrates an evaluation of 50 pilot branches of one Iranian private bank, Tosse-Eh Farda Bank (previously
Tosse-Eh Credit Institute), using Data Envelopment Analysis (DEA). Additionally, we implement an extension to this model with the
aim of ranking the efficient Decision Making Units (DMUs) applying Ordinary Linear Square (OLS). Special emphasis was placed on
how to present the DEA results to management in order to provide more guidance on what to manage and how to accomplish the
changes. Eventually, the potential management uses of the DEA results were presented.
Keywords: Data Envelopment Analysis (DEA), Linear programming, Banking, Efficiency, Ordinary Linear Square, Ranking
JEL Classification: C52, C82, D61, G21
1 Introduction
In retrospect, banks have focused on various profitability measures to evaluate their performance. Usually multiple
ratios are selected to focus on the different aspects of the operations. However, ratio analysis provides relatively
insignificant amount of information when considering the effects of economies of scale, the identification of
benchmarking policies, and the estimation of overall performance measures of firms. As alternatives to traditional
bank management tools, frontier efficiency analyses allow management to objectively identify best practices in
complex operational environments. Compared to other approaches, Data Envelopment Analysis (DEA) is a better
way to organize and analyze data since it allows efficiency to change over time and requires no prior assumption on
the specification of the best practice frontier. In addition, it permits the inclusion of random errors if necessary.
Since the introduction of DEA technology, a considerable number of researchers have applied it in financial service
industry. Cook et al. (2000) investigated the use of quantitative variable in bank branch evaluation using DEA.
Paradi and Schaffnit (2004) evaluated the performance of the commercial branches of a large Canadian bank. They
introduce non-discretionary factors to reflect specific aspects of the environment a branch is operating in. Asmild et
al. (2004) evaluate the performance of Canadian banking industry over time. Bala and Cook (2003) incorporate
expert knowledge within the DEA framework.
They first apply a discriminate or classification tool to quantify the functional relation that best captures the
expert's mental model for performance. The outcome of this first phase is an orientation of variables to aid in the
definition of inputs and outputs. The resulting orientation then defines the DEA model that makes up the second
phase of the model. Camanho and Dyson (2005) investigated the bank branch performance under price uncertainty.
Halkos and Salamouris (2004) measured the Greek bank performance using DEA. Isik and Kabir (2003) utilize a
DEA-type Malmquist Total Factor Productivity Change Index to examine productivity growth, efficiency change,
and technical progress in Turkish commercial banks during the deregulation of financial markets in Turkey.
Additionally, Guan and Dipinder (2005), Athanassopoulos and Giokas (2000), Devaney and Weber (2004), Pille and
Paradi (2002), Mercan al at. (2003), Penny (2004), studied the use of DEA in financial institutions, to mention a
few.
This paper presents an evaluation of 50 pilot branches of the first Credit Institute in Iran using DEA. The rest of
the paper is organized as follows. Section 2 gives a brief review of DEA. Section 3 provides the models,
methodology utilized in this paper and then DEA results and the potential use of the DEA results were presented.
Finally, the conclusions are given in section 4.
155

2 The Principles of Data Envelopment Analysis
Production process can be defined as a process that can turn a set of resources into desirable outcomes by production
units. During this process, efficiency is used to measure how well a production unit is performing in utilizing its
resources to generate the derived outcomes. Each of the various DEA models seeks to determine which of the n
decision making units (DMUs) define an envelopment surface that represents the best practice, referred to as the
empirical production function or the efficient frontier. Units that lie on the surface are deemed efficient in DEA
while those units that do not, are termed inefficient. DEA provides a comprehensive analysis of relative efficiencies
for multiple input-multiple output situations by evaluating each DMU and measuring its performance relative to an
envelopment surface composed of other DMUs. Those DMUs forming the efficiency reference set are known as the
peer group for the inefficient units. As the inefficient units are projected onto the envelopment surface, the efficient
units closest to the projection and whose linear combination comprises this virtual unit form the peer group for that
particular DMU. The targets defined by the efficient projections give an indication of how this DMU can improve to
be efficient. Consider n DMUs to be evaluated, DMU j (j=1,2…n) consumes amounts j X ={ x ij } of inputs (i=1,
2, …, m) and produces amounts j Y ={ rj y }of outputs (r=1 ,…, s). The efficiency of a particular DMU 0 can be
obtained from the following linear programs (input-oriented BCC model (1984)).
(1)
min
s.
t.
Y�
� s
�X
�
0
� �
� , �,
s , s
�
1 � � 1
�
�,
s , s
� Y
0
� 0
0
� X�
� s
�
z
� � � �.
1 s
�
� 0
�
�
�
� �.
1 s
�
Performing a DEA analysis actually requires the solution of n linear programming problems of the above form,
one for each DMU. The optimal variable θ is the proportional reduction to be applied to all inputs of DMU0 to move
it onto the frontier. A DMU is termed efficient if and only if the optimal value θ* is equal to 1 and all the slack
variables are zero. This model allows variable returns to scale. The dual program of the above formulation is
illustrated by:
(2)
min
s.
t.
v
T
� Y � v
� v
u
T
0
X
T
� , v
0
free
w
T
0
� 1
� �Y
� u
X � u
�
T
� � � ��.
1
�
� ��.
1
0
0
�
0
1 � 0
�
If the convexity constraint ( 1 � �1
) in (1) and the variable u0 in (2) are removed, the feasible region is
enlarged, which results in the reduction in the number of efficient DMUs, and all DMUs are operating at constant
156

eturns to scale. The resulting model is referred to as the CCR model. The reader is advised to consult the textbook
by Cooper, Seiford and Tone (2000) for a comprehensive treatment of DEA theory and application methodology.
3 Specifications and Data
A number of different approaches can be used for modeling the banking processes. Each of which is used to obtain a
different aspect of efficiency measures. The most important two approaches are the production approach and the
financial intermediation approach. Under the production approach, banks are viewed as institutions making use of
various labor and capital resources to provide different products and services to customers. Thus, the resources
being consumed such as labor and operating cost are deemed as inputs while the products and the services such as
loans and deposits are considered outputs of the banks. This model measures the cost efficiency of the banks. Under
the financial intermediation approach, banks are viewed as financial intermediaries collecting deposits and other
loanable funds from depositors and lend them as loans or other assets to others for profit. The different forms of
funds that can be borrowed and the cost associated with performing the process of intermediation are considered as
inputs. The forms in which the funds can be lent are outputs of the model. This model measures the economic
viability of the banks. There were two inputs (official cost and wage cost) and six outputs (Different kind of
Deposits, Loans and different type of facilities, Delaying Claims, Operational profit, Other Services and Branches’
superficial quality) in the DEA model. Input orientation (the LP is oriented to minimize inputs) was selected for the
DEA models in this research. I point out that management was more interested in minimizing the consumption of
inputs subject to attaining the desired output levels. BCC model is utilized so as to consider the size effect.
3.1 The Proposed Method
In this analysis, 50 branches of a private bank in Tehran, Iran are evaluated. The degree of correlation between
inputs and outputs is an important issue that has great impact on the robustness of the DEA model. Thus, a
correlation analysis is imperative to establish appropriate inputs and outputs. On the one hand, if very high
correlations are found between an input variable and any other input variable (or between an output variable and any
of the other output variables), this input or output variable may be thought of as a proxy of the other variables.
Therefore, this input (or output) could be excluded from the model. On the other hand, if an input variable has very
low correlation with all the output variables (or an output variable has very low correlation with all the input
variables), it may indicate that this variable does not fit the model. Correlation analyses were done for each pair of
variables and the following table presents the details.
Wage Cost Official
Cost
Deposits
Loan &
Facilities
Delaying
Claims
Operational
Profit
Other
Services
Superficial
Quality
Wage Cost 1.00 0.81 0.79 0.70 0.80 0.44 0.86 0.32
Official Cost 0.81 1.00 0.51 0.53 0.48 0.20 0.58 0.57
Deposits 0.79 0.51 1.00 0.51 0.71 0.80 0.94 0.31
Loan & Facilities 0.70 0.53 0.51 1.00 0.86 0.08 0.58 0.20
Delaying Claims 0.80 0.48 0.71 0.86 1.00 0.23 0.76 0.28
Operational Profit 0.44 0.20 0.80 0.08 0.23 1.00 0.69 0.25
Other Services 0.86 0.58 0.94 0.58 0.76 0.69 1.00 0.26
Superficial Quality 0.32 0.57 0.31 0.20 0.28 0.25 0.26 1.00
Table 1: Correlation Coefficients between the Inputs and Outputs
I did not find any evidence of very high correlation between any one input variable and any other (nor between
output variables) and any one input variable having very low correlation with any of the output variables (nor
between output variable and input variables) in the above five tables. This is a reasonable validation of my DEA
models. Otherwise, the sensitivity analysis on the impact of including and excluding different variables on the
efficiency should be performed. The input oriented BCC model is run and Table 2 summarizes the results for the
model.
157

3.2 The Efficient DMUs Ranking
Topics Result
Average Score 0.86
Standard Deviation 0.19
Maximum Efficiency Score 1.00
Minimum Efficiency Score 0.23
Number of DMUs 50
Number of Efficient DMUs 23
Number of Efficient DMUs
exhibiting IRS
3
Number of Efficient DMUs
13
exhibiting CRS
Number of Efficient DMUs
exhibiting DRS
Table 2: DEA Results BCC
We employ Data Envelopment Analysis (DEA)-BCC model to evaluate decision-making units (DMUs) into
efficient and inefficient ones base upon the multiple inputs and output performance indices. Afterwards, we consider
that there is a centralized decision maker (DM) who ‘owns’ or ‘supervises’ all the DMUs. The DM has an interest in
discriminating the efficient DMUs (eDMUs). We present a new method that determines the most compromising set
of weights for the indices’. The total of the new efficiency scores of eDMUs with the most compromising set of
indices’ weights has the least total gaps to the compromised data. The eDMUs that have efficiency score equal to
one are located on the data. The other eDMUs are either located above or below the data. The approach is analog to
the ordinary linear square method (OLS) of the residuals in statistical regression analysis. Finally, by this we could
rank all the branches.
4 Conclusions
This paper uses DEA to assess the branch performance in a private bank. The branches operate fairly efficiently on
the whole although there is still room for improvement. Special emphasis was placed on how to present the DEA
results to management so as to provide more guidance to them on what to manage and how to accomplish the
changes. Eventually, recommendations to management’s use of DEA results were given.
5 Acknowledgements
I owe special thanks to Tosse-Eh Farda Bank for providing the relevant data set and also Professor Heidar
Mostakhdemin Hosseini of Raja University for his valuable comments.
158
7

6 References
Asmild Mette, Paradi J. C., Aggarwall Vanita, and Schaffnit Claire (2004), “Combining DEA Window Analysis
with the Malmquist Index Approach in a Study of the Canadian Banking Industry”, Journal of Productivity
Analysis 21, No. 1, 67-89.
Athanassopoulos, A.D. and Giokas, D. (2000), “The Use of Data Envelopment Analysis in Banking Institutions:
Evidence from the Commercial Bank of Greece, Interfaces, Vol 30, no 2, 81-95.
Bala, Kamel and Cook, Wade D. ( 2003), “Performance Measurement with Classification Information: an Enhanced
Additive DEA Model”, Omega 31, No. 6, pp439-450.
Banker, R., D., Charnes, A., and W. W Cooper (1984), "Models for Estimating Technical and Scale Efficiencies in
DEA", Management Science 30(9): pp.1078-1092.
Casu, B., Girardone, C., and Molyneux, P. (2004), “Productivity Change in European Banking: a Comparison of
Parametric and Non-Parametric Approaches”, Journal of Banking and Finance 28, No. 10, 2521-2540.
Camanho A.S., Dyson and R.G. (2005), “Cost Efficiency Measurement with Price Uncertainty: a DEA Application
to Bank Branch Assessments”, European Journal of Operational Research 161, No. 3, 432-446.
Cook, W.D., Hababou, M. and Tuenter, H.J. (2000), “Multicomponen Efficiency Measurement and Shared Inputs in
Data Envelopment Analysis: an Application to Sales and Service Performance in Bank Branches”, Journal of
Productivity Analysis 14, 209-224.
Cooper, W.W., L.M. Seiford and K. Tone (2000), “Data Envelopment Analysis: A Comprehensive Text with
Models, Applications, References”, Kluwer Academic Publishers, Boston.
Guan, H.L. and Dipinder, S.R. (2005), “Competition, Liberalization and Efficiency: Evidence from a Two-Stage
Banking Model on Banks in Hong Kong and Singapore”, Managerial Finance 31, No. 1, 52-77.
Halkos, G.E. and Salamouris, D.S. (2004), “Efficiency Measurement of the Greek Commercial Banks with the Use
of Financial Ratios: a Data Envelopment Analysis Approach”, Management accounting Research 15, No. 2,
201-224.
Isik, Ihsan and Kabir Hassan, M. (2003), “Financial Deregulation and Total Factor Productivity Change: An
Empirical Study of Turkish Commercial Banks”, Journal of Banking and Finance 27, No. 8, 1455-1485.
Mercan, M., Reisman A., Yolalan, R., and Emel, A.B. (2003), “The effect of Scale and Mode of Ownership on the
Financial Performance of the Turkish Banking Sector: Results of a DEA-Based Analysis”, Socio-economic
Planning Sciences 37, No. 3, 185-202.
Neal, Penny (2004), “X-Efficiency and Productivity Change in Australian Banking”, Australian Economic Papers,
Vol 43, Issue 2, 174-191.
Paradi, J.C. and Schaffnit Claire (2004), “Commercial Branch Performance Evaluation and Results Communication
in a Canadian Bank - a DEA Approach”, European Journal of Operational Research 156, No. 3, 719-735.
Pille. P., and Paradi, J.C. (2002), “Financial Performance Analysis of Ontario (Canada) Credit Unions: An
Application of DEA in the Regulatory Environment”, European Journal of Operational Research 139, No. 2,
339-350.
159

MODELING FUNCTIONAL INDICATORS FOR CUSTOMER PERFORMANCE IN REPAYING BANK
LOAN (CASE STUDY: IRANIAN BANKS)
Fariba Seyed Jafar Rangraz, Amirkabir University of Technology
Jafar Pashami, Sharif University of Technology
Email: RANGRAZFARIBa6384@gmail.com
Abstract Nowadays, decision making has become an important issue in many subjects and areas including investment and banking.
This is due to the unstable economic conditions and the unknown factors leading to people’s success. Developing countries face with
limited capital. Hence, there is less attention to bank customers on the basis of scientific methods in order to develop a framework of
project evaluation process. Therefore, there are lots of unfinished and untouched research projects in those countries including Iran.
This paper aims to evaluate the money lending process based on Iranian banks customers. The 5C’s model is used to measure the
relationship between each element of the model by taking into consider the factors leading to customer satisfaction. A quantitative
method is applied to measure the findings. MLP using cross-validation and SVM are used in order to estimate the percentage of
customer loan request that banks should agree and also predict customer reaction in repaying the loan. These methods show down to
be more successful in estimating the first output (acceptable percentage of customer loan request) but for having better results for
second output(predicting the costumer performance in repaying), the input feature must be changed.
Key words: Banking, lending, repaying, 5C’s, MLP, SVM
1. Introduction
Decreasing risks and the descending possibility of not repaying are ways used by banks referred to as Computation
technique. However, the unstable economy and external environmental factors force banks to review their method of
decision making.
K. Bryant has presented a model for evaluating agronomic loans (Bryant, k.,2001). He has used Duchessi
research for designing his model. According to Bryant, it is necessary the lender knows 5 factors when consumer
credit is being evaluated (Duchessi, P., 1995). These factors are: credit, capital, capacity, collateral and character.
Bryant introduced an expert system for decision making (ALEES). In this system, quality factors such as skill,
experience and the intelligence of the loan expert are keys while quantity factors are mixed.
Rosman and others have studied different strategies that lenders use for confirming borrowers. They
suggest that the structure of the lending process is in 2 parts: financial and nonfinancial. There are some features in
each part (Rosman Andrea., and Bedard, Jean, 1999)
Tansel and Yardakhl presented a model for evaluating decisions. This model is based on determining the
credit of manufacturing firms that want to borrow from the Turkish banks (Yardakhl, M. , and Tansel, Y., 2004). In
this model, all quality and quantity indexes related to the profitability of a company have been recognized. The
obtained results can give a special score to the company using the AHP technique.
Chan and Mak presented a method showing the benefits of selecting an improved manufacturing technique
consisting of the following three approaches (Chan, F.T.S., and Mak, K.L. , 2000):
� Strategic approach: achieving common goals
� Economic approach: economic benefits
� Analytical approach: economical and non-economic benefits and risks
Reitan introduced efficient factors when evaluating loans. He estimated the ability of experts in determining the
value and measuring factors of the project. He compared them based on the importance of each factor. He also
concluded why certain evaluating processes are incorrect. He categorized 42 indexes into 6 groups. These are the
Manager’s character and experience, services offered, market conditions, different pertinent aspects of the
organization and financial matters (Reitan, Bjornar., 1998).
160

In this paper, the factors affecting the lending decision making process has been studied. Different results due to
various situations, features and abilities of the borrower have been attained.
This information is categorized in 5 groups. The effective factors in bank decision making are named the 5C’s.
They are (Tim Hill of COCC):
� Character
� Capacity
� Condition
� Capital
� Collateral
1.1. Character
Character is the customer’s impact on the lender. The lender decides whether the customer is reliable to repay
Therefore, the customer’s education and experience in that project would be surveyed. Also, an employee’s quality
of performance and experience should be considered (Tim Hill of COCC).
1.2. Capacity
The capacity of the loan applicant refers to their imbursement ability and it is evaluated using expected cash flow.
For those customers requesting short term loans, banks experts would pay attention to the method of payment such
as cash. Other factors considered are the customer’s assets which can be transformed to cash immediately. Future
cash flow and the company’s potential in honoring short time commitments are considered too. In contrast, clients,
who extend long term loans, usually focus on long term assets. Other factors considered are the difference between
the assets and debts, permanent activities, profitability and the company’s potential (Accounting organization of
Iran, 2007).
1.3. Capital
Capital is equal to the net worth or net asset. Moreover, it is the reflection of the company’s success and value.
Although this criterion can be sinister with considering variability of asset worth and official worth, official accounts
may not show asset market worth. However inflation and increase in price level is the reason for this and sometimes
account based asset value is less than market worth. But, considering worth based on asset official worth is not
critical for the lender. Capital is the money which a client invests personally on a project and it is explanatory of
corporation in loss if the firm is broken (SBA, North Catalina). Perhaps it’s best to use the word ‘bankrupt’ or
‘folds’ instead of ‘broken’.
1.4. Collateral
Collateral is studied from a lending aspect and is one of the basic factors in the lending decision making process.
The other notable role of collateral is being one of the effective elements in evaluating the lender and nominating his
credit condition. Regardless of how much the employees of a bank are careful and cautious in lending; there
sometimes would be problems in repayment. In this situation, the alternative solution is collateral. The Bank must
decide what type of collateral is needed to recoup the loan. Property, a house or land, equipment or personal
assurance may be required (Edward Poll,J.D.,M.B.A.,CMC,2003).
2. Research method
For evaluating indexes, a questionnaire was designed consisting 26 questions and distributed among 100 bank
experts. Questions categorized in 5 groups like 5C’s indexes. In addition, some of the Iranian bank's experts were
interviewed to support the information collected from questionnaires.
The Likert scale was used to measure attitudes, preferences, and subjective reactions of respondents.
Statistical sample size in this research consists of Iranian banks’ experts. Selected banks were some public and
private banks. This selection helped to collect more comprehensive data sets.
MLP using cross-validation and SVM were used in order to estimate the percentage of customer loan request that
banks should agree and predict customer reaction in repaying the loan. MLP with one hidden layer and 3 neurons in
the hidden layer and one neuron in the output layer was used. Then vectors had been normalized.
161

26 features were defined base on 5C’s so the amount of weights is equal to (26*3)+3=81 .At least 10 times of data
must be available in order to training and testing the network well. As there were 100 questions as input data, so
Cross-Validation method was used in order to conquest deficiency of data. This method is used when enough data is
not available and there is no possibility to allocate enough data for training and testing. Kfold method was used,
each time 10 data were used for testing and 80 data were used for training. This activity was done for 90 data and 10
last data were not given to network and they were used at the end for checking answers.
2.1. Research hypothesize
The main hypotheses of this research were:
� Customer financial indexes are notable in bank expert decision making.”
� Loan applicant character, capacity, capital, condition and collateral are effective in bank expert decision
making.”
3. Research result
It is supposed that Y1 isthe percentage that bank accepts with costumer request amount for loan and Y2 is the
category that costumers belong to. Likert scale was used for defining Y2. In this paper, in order to show more
variance among groups 1, 3,5,7,9 were applied instead of 1, 2,3,4,5 numbers. So the accuracy became higher. Y2
was defined like this:
Y2=1 means that costumer do very badly in repaying.
Y2=3 means that costumer do badly in repaying.
Y2=5 means that costumer do neural in repaying.
Y2=7 means that costumer do well in repaying.
Y2=9 means that costumer do very well in repaying.
The graph of training error related to first output (acceptable percentage of customer loan request) using MLP is
shown in figure1:
figure1: the graph of training error
The testing error for the first output was ev1 =0.002.
For the second output, the same network was used and the error was about 0.01. From this notable error, it can be
concluded that all the features used for predicting the amount of loan that bank accepts are not appropriate to find
the repayment trend. It is thought when a lender lends the loan, he should be sure of repaying trend and these are not
independent. But in this research, data were incoherence and this ends to error. SVM network was used for
classification too but there was an error about the previous error or some percentage lower. Then in order to find
162

which features are more important and with which compound error can be decreased, the network was implemented
using 1 to 7compounds of 26. But the error didn’t become better in any compound.
As implementing program needed lots of times it was used for some of features, but other compounds of features
may reduce error and end to best selection of features or even decide to have new features.
Then an unsupervised networks, Kmeans was used for classification in order to show that these features end to label
different with the second output data. This result was another confirm to previous discussion.
4. Conclusion
Attending MLP network, input features can estimate the first output (the percentage that banks accept to amount of
customers’ request of loan) well and the error of training is equal to the error of test which is 0.002. Using SVM and
MLP shows that these features for defining the second output (predicting the costumer performance in repaying)
have a test error about 0.01 which is notable and in order to decreasing error, features should be changed and current
features are not reliable for predicting the repaying trend.
5. References
Bryant, k.;"An agricultural loan evaluation expert system", Expert system with application, 2001, pp75-85.
Duchessi, P.;"A knowledge engineered system for commercial loan decisions", Financial management, 1995, (17-3),
pp 57-65.
Rosman Andrea., and Bedard, Jean."Lenders Decision Strategies and loan structure decisions", Journal of Business
Research, 1999, pp.83-94.
Yardakhl, M. , and Tansel, Y. ;"AHP approach in the credit evaluation of manufacturing firms in Turkey'', Int. J. of
production Economics, 2004(88), pp. 269-289.
Chan, F.T.S., and Mak, K.L. ;"An Integrated Approach to Investment Appraisal for Advanced manufacturing
technology". ,Human factors and Ergonomics in manufacturing,2000(19)
Reitan, Bjornar.;" Criteria Used by Private Bank Officers to Evaluate Nea Ventures: An Analysis of Gaps and
Shortcomings" http://www.sbaer.uca.edu/Research/1998/ICSB/n015.htm
Tim Hill of COCC, "The 5C's getting money from a bank"
Accounting organization, "Iranian accounting based theories", publishing committee of standards of Iran accounting,
Tehran 2007
"The 5C's of credit-small business resource", SBA(US Small Business administration)-North Catalina
Edward Poll,J.D.,M.B.A.,CMC, " Understanding the four C's", Lawyers and bank loans ,law practice TODAY,
september2003
6. APPENDIXES
Questionnaire:
How much does each of these factors effect in tour decision to lend?
For scoring factors, consider that:
1= means that this item is the less important item from banks experts point of view for lending.
3= means that this item is a few important item from banks experts point of view for lending.
5= means that this item is a neural item from banks experts point of view for lending.
7= means that this item is an important item from banks experts point of view for lending.
9= means that this item is the most important item from banks experts point of view for lending.
Character questions:
Customer education: …………..
Company activity in related industry: …………..
Customer credit in industry: …………..
Customer background in past loans and his repayment method in the past: …………..
163

Commercial fame: …………..
Loan repayment: …………..
Loan application times: …………..
Customer credit background in other banks: …………..
Company management constancy: …………..
Profitability: …………..
Capacity questions:
Company financial ability in repaying: …………..
Company financial statements: …………..
Fixed asset value: …………..
Liquidity asset and cash flow: …………..
Obtaining financial resource method: …………..
Product variety: …………..
Competitors: …………..
Target market: …………..
Capital questions:
Account based asset value: …………..
Company management ability and his background: …………..
Condition questions
Encountering national economical fluctuation: …………..
Encountering international economical fluctuation: …………..
Economic and politic society condition: …………..
Place of related industry in society: …………..
Collateral questions:
Amount of customer collateral: …………..
164

MEASURING AGREEMENT WITH THE WEIGHTED KAPPA COEFFICIENT OF CREDIT RATING
DECISIONS: A CASE OF BANKING SECTOR
Funda H. Sezgin and Ozlen Erkal, University of İstanbul, Turkey
Email:hfundasezgin@yahoo.com; ozlenerkal@istanbul.edu.tr
Abstract. An important reason for the global economic crisis affecting the whole world is large amount of loss caused by the banks
which gone bankrupt although they have high level of credit scores by credit rating agencies. Now a days it is very important to
measure agreement among decisions they made with each other since we distrust these agencies. In this study the weighted kappa
coefficient is measured for agreement of credit rating decisions made for banks in member countries of the European Union. This
coefficient is effective when number of level is large and especially significance level of disagreement among levels is not equal to
each other. Credit rating agencies evaluate the same status independently. Weighted kappa coefficient is an extended kind of kappa
coefficient measuring common classification of agreement between them.
Keywords: Credit Rating, Weighted Kappa Coefficient, Banking Sector
JEL classification: G21, C10, G32
1 Introduction
Together with the globalization process, the increasing complexity of financial markets and the variety of debtors,
both investors and official authorities have begun to attach more importance to the concept of rating. An important
reason that contributed to the importance of rating in international markets is that rating has started to be made use
of in supervising and regulating financial markets. The decision of The Basel Committee on Banking Supervision
under Bank for International Settlements (BIS) to use rating as a tool in regulating banks’ capital adequacy has been
the most important evidence of the fact that rating is an integral part in markets and that it will continue to play an
ever more important role.
Rating as an expression of the credibility of a debt is becoming increasingly important for banks and their
customers. In addition to political crises, financial crises that have a significant negative effect on a state’s stability
and banking crises as a sub-branch have long brought to the agenda the idea of setting standards in banking and
financial markets. The Basel Committee on Banking Supervision, an embodiment of this idea, aims to measure
banks’ capital adequacy and their capacity to meet minimum agreed standards (Longstaff & Schwartz, 1995).
The credit rating approaches for banks have brought about significant improvements to credit demand
assessments following Basel II, which consisted of a framework of greater durability, stability and contribution
compared to Basel II. Rating, playing an important role in the implementation of Basel II credit risk measurement
approaches, has paved the way for a transition from a subjective crediting process based on expert opinion to an
objective crediting process in which risk is quantified (Nickell et al., 2000).
As an international standard, Basel II, which led to radical changes in banking activities and in their regulation
and supervision with its scope and technical specifications, has inevitable deep impacts on banks, authorities on
banking supervision and all parties related with banks (customers, rating firms, data validation firms, etc.) as well as
on states’ economies.
In this study, the agreement of decisions made by world’s leading credit rating agencies in rating EU banks will
be measured using weighted kappa coefficient. The weighted kappa coefficient is a generalized version of the kappa
coefficient which measures the classification agreement between different decision-makers who independently rate a
same situation. Hence, the classification agreement between credit rating agencies will be measured. The reason for
studying this field is the fact that the top-rated banks declared bankruptcy following the global economic crisis,
which subsequently led to a decline in the credibility of and trust in credit rating agencies.
165

2 The Concept of Rating
Rating is a tool that makes life easier for markets and their beneficiaries by translating information acquired through
costly and time-consuming hard work into simple symbols. Credit rating agencies use certain categories which are
illustrated by means of easily comprehensible symbols such as letters, numbers or both. Rating is applicable not only
for a country but also for a single bank or enterprise. Rating has been used in different areas, such as securities,
commercial firms, financial institutions and banks. A summary of the benefits rating bring to economic life are
1- It provides the economy with financial markets which show reliability and stability in their development,
2- Provides outside sources for the economy and helps markets integrate with international markets,
3- Classifies the general risk level in the economy, increases the efficiency of financial operations and finances
development more effectively (Jarrow & Turnbull ,2000).
The aim of establishing a rating system is to assess a firm’s risks more objectively. This not only helps to create
a common language but also gets the banks to make very similar pricings for a firm.
Rating, which is a standard and objective approach developed by professionals, helps define a debtor’s
credibility and capacity to make regular payments, and assesses and affects his/her role in monetary and capital
markets. Rating is made up of strong and useful symbols which help institutions, credit demanders/suppliers,
investors, intermediaries and regulators in monetary and capital markets in their assessments (Nye and Eke, 2004).
There are two types of ratings:
External Ratings: External rating is an indicative credit rating given by rating agencies to relatively large-scale
firms to allow them to borrow from capital markets.
Internal Ratings: Internal ratings are the ratings given to debt demanders by banks in accordance with the banks’
own internal rating criteria. In internal ratings, banks request their customers’ financial statements, official
documents and all qualitative and quantitative credentials. The basic condition for such a system to run smoothly is
an easy and standard access to chronologically-stored workable data (Lando, 1999).
It was in the mid-1980s when developments in ratings gained impetus. The debt crisis of 1970s and the
Mexican crisis of 1984 brought to the agenda the need for a faster and more effective form of rating in the
international monetary system. Both the disappointing performances of the IMF and the World Bank, and a rapid
development and diversification of the financial system brought the concept of rating to the fore. Especially,
� Acceleration of capital movement and the formation of new investment opportunities,
� Capital flows becoming an object of trade in financial markets,
� An increased interest of a wide audience including small to large scale banks and bond holders in financial
markets raised the importance of countrywide risk ratings.
There has been an increased need for rating agencies in recent years. Classification of countries according to
their risk levels provide an insight for investors as to what kind of risks they might encounter in the countries they
plan to invest in.
2.1 Credit Rating Agencies and Rating Methodologies
Credit rating agencies are companies that assign credit ratings for borrowers’ capacity and willingness to fulfil their
debt obligations. Ratings represent an independent opinion and cannot be qualified as advice for investment. Credit
rating agencies can assign ratings for states, financial institutions, companies and various capital market issues. Such
agencies are supervised by relevant state regulatory and supervisory bodies. They are subjected to supervision by
Securities and Exchange Commission (SEC) in the USA and Capital Markets Board in Turkey. Agencies that meet
certain criteria have been registered as “Nationally Recognized Statistical Rating Organization” (NRSRO) by the
Securities and Exchange Commission (White, 2002).
166

In 1975 SEC gave life to NRSRO in order to standardize the operations of rating agencies and their entries to
the market. Currently, there are ten NRSROs:
� A.M. Best Company, Inc
� DBRS Ltd.
� Egan-Jones Rating Company
� Fitch, Inc.
� Japan Credit Rating Agency, Ltd.
� LACE Financial Corp.
� Moody’s Investors Service, Inc.
� Rating and Investment Information, Inc.
� Realpoint LLC
� Standard & Poor’s Ratings Services
Credit score or rating score is the measurement of creditworthiness by a rating agency. In a broad definition,
credit rating is an independent opinion about whether an organization can fulfil its financial obligations in time or
not. Results are translated into symbols in order to make this opinion easily comprehensible. The symbols are
universally applicable. A corporation or a state with the strongest capacity to fulfil its obligations is rated AAA.
Grades from AAA to BBB are investment grades, while those lower than BBB are non-investment grades indicating
riskiness. Below are the rating tables (Table 1, Table 2 and Table 3) used by credit rating agencies, such as S&P,
Fitch and Moody’s.
Credit ratings can be categorized as long- or short-term, and foreign or local currency. The data used in the
study are based on banks’ long-term payment performances and their USD credit scores. These are taken from
Reuters’ data on banks from December 2010. Banks rated by Fitch, S&P and Moody’s are used as observations
units and the agencies’ agreement is measured in pairs with weighted kappa coefficient.
2.1.1 Standard and Poor’s Rating Definitions
Standard and Poor’s (S&P)’s grades fall into two broad categories: “Investment merit” and “Speculative”. There is
a total of ten rating grades, four of which are under “Investment merit” with the remaining six under “Speculative”.
AAA is the highest grade given by S&P to corporations with a capacity to pay their debt at ease. BBB and BB stand
on a critical line. While BBB is investment merit, BB is speculative. The lowest grade D means that it is impossible
for a corporation to pay off its debt and implies bankruptcy. S&P’s grading scale is in Table1.
S&P Long-term Credit Ratings Scale
Investment AAA Extremely strong capacity to meet financial commitments.
AA Very strong capacity to meet financial commitments.
A Strong capacity to meet financial commitments, but somewhat susceptible to adverse economic conditions and changes in
circumstances.
BBB Adequate capacity to meet financial commitments, but more subject to adverse economic conditions.
Speculative BB Less vulnerable in the near-term compared to other speculative ratings but faces major ongoing uncertainties to adverse
business, financial and economic conditions.
B More vulnerable to adverse business, financial and economic conditions but currently has the capacity to meet financial
commitments.
CCC Currently vulnerable and dependent on favorable business, financial and economic conditions to meet financial
commitments.
CC Currently highly vulnerable.
C Currently highly vulnerable obligations and other defined circumstances.
D Payment default on financial commitments.
2.1.2 Moody’s Investors Service Rating Definitions
Table 1: S&P Credit Ratings Scale
Long-term credit ratings of Moody’s Investors Service are opinions about relative credit risk of obligations of
one year or more. Like S&P, Moody’s grades fall into two broad categories: “Investment merit” and “Speculative”.
There is a total of nine rating grades, four of which are under “Investment merit” with the remaining five under
“Speculative”. Moody’s Investors Service’s highest grade is Aaa, followed by Aa, A and Baa under “Investment
167

merit”. Ba, B, Caa, Ca and C fall under “Speculative”. The lowest grade C indicates no prospect of debt
reimbursement. Moody’s Investors Service’s grading scale is in Table 2.
Moody’s Long-term Credit Ratings Scale
Investment Aaa The highest capacity to meet obligations, with minimal credit risk.
Aa High capacity to meet obligations and are subject to very low credit risk.
A Upper-medium grade to meet obligations and are subject to low credit risk.
Baa Moderate capacity to meet obligations, moderate credit risk, may possess certain speculative characteristics.
Speculative Ba Speculative characteristics and substantial credit risk.
B Speculative and high credit risk.
Caa Poor standing, very high credit risk.
Ca Highly speculative and likely in, or very near, default, with some prospect of recovery of principal and interest.
C In default, with little prospect for recovery of principal or interest.
2.1.3 Fitch Ratings’ Rating Definitions
Table 2: Moody’s Investors Service Credit Ratings Scale
In Fitch Ratings’ rating scale, there are four types of rating grades under “Investment merit” and eight under
“Speculative”. The highest grade is AAA in the “Investment merit” category, while speculative DDD, DD and D
grades indicate non-payment prospects. Compared to DD and D, DDD indicates a higher probability (90% - 100%)
of payment and accrued interest collection. D indicates a low probability of payment. Fitch Ratings’ grading scale is
in Table 3.
Fitch Ratings Long-term Credit Ratings Scale
Investment AAA Highest credit quality. ‘AAA’ ratings denote the lowest expectation of default
risk. They are assigned only in cases of exceptionally strong capacity for
payment of financial commitments. This capacity is highly unlikely to be
adversely affected by foreseeable events.
AA Very high credit quality. ‘AA’ ratings denote expectations of very low default
risk. They indicate very strong capacity for payment of financial commitments.
This capacity is not significantly vulnerable to foreseeable events.
A High credit quality. ‘A’ ratings denote expectations of low default risk. The
capacity for payment of financial commitments is considered strong. This
capacity may, nevertheless, be more vulnerable to adverse business or economic
conditions. Possesses risky characteristics.
BBB Good credit quality. ‘BBB’ ratings indicate that expectations of default risk are
currently low. The capacity for payment of financial commitments is considered
adequate but adverse business or economic conditions are more likely to impair
this capacity. ‘BBB’ is the lowest investment category.
Speculative BB Speculative. ‘BB’ ratings indicate an elevated vulnerability to default risk,
particularly in the event of adverse changes in business or economic conditions
over time; however, business or financial flexibility exists which supports the
servicing of financial commitments. BB rated bonds and bills are not considered
investment merit.
B Highly speculative. ‘B’ ratings indicate that material default risk is present, but a
limited margin of safety remains. Financial commitments are currently being
met; however, capacity for continued payment is vulnerable to deterioration in
the business and economic environment.
CCC, CC, C Default is highly probable. May not be able to meet financial obligations.
Capacity to fulfil obligations depends on economic conditions over time. ‘CC’
indicates default of some kind appears probable, while ‘C’ indicates default is
imminent or inevitable.
DDD, DD, D Default. The ratings of obligations in this category are based on their prospects
for achieving partial or full recovery in a reorganization or liquidation of the
obligor. While expected recovery values are highly speculative and cannot be
estimated with any precision, the following serve as general guidelines. ‘DDD’
obligations have the highest potential for recovery, around 90% - 100% of
outstanding amounts and accrued interest. ‘DD’ indicates potential recoveries in
the range of 50% - 90% and 'D' the lowest recovery potential below 50%.
Table 3: Moody’s Investors Service Credit Ratings Scale
168

3 Weighted Kappa Coefficient
Often some disagreements between the two raters can be considered as more important than others. For example,
disagreement on two distant categories should be considered more important than on neighbouring categories on an
ordinal scale. For this reason, Cohen (1968) introduced the weighted kappa coefficient. Agreement (ωjk ) or
disagreement (νjk ) weights are a priori distributed in the K 2 cells of the K x K contingency table (Vanbelle & Albert,
2009). The weighted kappa coefficient is defined in terms of agreement weights;
With
K K
^
K
p � p
� � �
ow jk kj
j �1 k �1
It can also be defined with disagreement weights,
With q v p
^
K
w
w
p � p
�
1�
p
and
q
� 1 �
q
ow ew
ew
K K
p � p p
� � �
ew jk j . . k
j �1 k �1
K K
ow � �� jk jk and
K K
qew � �� v jk p j. p.
k
j�1 k �1
j�1 k �1
o w
e w
(0≤ νjk≤1 and νjj=1).
(0≤ ωjk≤1 and ωjj=1)
Although weights can be arbitrarily defined, two weighting schemes are most commonly used. These are the
“linear” weights introduced by Cicchetti and Allison (1971),
j � k
� jk �1 �
K �1
and the quadratic weights introduced by Fleiss & Cohen (1973),
�
jk
� j � k �
�1 � � �
� K �1
�
Note that the disagreement weights νjk = (j - k) 2 are also commonly used (Ludbrook, 2002; Agresti, 1992) and
that Cohen's kappa coefficient is a particular case of the weighted kappa coefficient where ωjk=1 when j=k and
ωjk=0 otherwise (Ben-David, 2008).
Weighted kappa coefficient’s value is between -1 and +1. Where weighted kappa coefficient =1, two decision
makers are in “perfect agreement”. When observed agreement equals expected agreement, then weighted kappa
coefficient=0 and the decision makers are in “random agreement”. When weighted kappa coefficient =-1, the two
decision makers are in “perfect disagreement” (Fleiss et. al., 2003).
Randomly selected ratings used in interpreting k statistical value are similarly used in interpreting weighted
kappa coefficient statistical value that corresponds to the level of agreement between decision makers. Accordingly,
values higher than 0.75 are interpreted as “perfect agreement” beyond expected agreement, values less than 0.40 as
“slight agreement” beyond expected agreement, and values between 0.40 and 0.75 as “slight to good agreement”
beyond expected agreement (Landis & Koch, 1977).
2
169

4 Empirical Analysis
In the study, the agreement among common assessment decisions made by credit rating agencies for the top-ten
banks (in assets size) in EU member states is measured by the weighted kappa coefficient. A total of six categories
were selected (AAA, AA, A, BBB, BB and B). Grade units lower than B were not taken into account. For top-rated
banks declared bankruptcy which resulted in loss of credibility and trust for credit rating agencies following the
financial crisis, banks rated B and higher were given priority in this study, which takes into account the credit scores
given in 2009-2010. The reason here is to explore the recent agreement between credit rating agencies who adopted
stricter policies following the post-crisis regulations in the banking sector in many countries. Banks with lower
grades were not included in the sampling. The weighted kappa coefficient to be used in all applications in order to
measure the agreement among rating agencies was calculated using both linear and quadratic weighting schemes.
The statistical values for kappa, weighted linear kappa and weighted quadratic kappa were determined before the
analysis. For kappa coefficient was low (0.689), it was decided that weighted kappa coefficient (linear=0.925 and
quadratic=0.891) would be suitable for use.
Initially, credit rating agencies “S&P” and “Moody’s” were employed in the study as the first pair of decision
makers. The categorical distribution for the 200 banks rated by both of the agencies is shown in Table 4.
Moody's
AAA AA A BBB BB B Total
AAA 2 0 0 2 0 0 4
S&P AA 2 15 10 10 0 0 37
A 3 10 33 1 0 0 47
BBB 0 0 1 20 10 0 31
BB 0 0 0 2 45 2 49
B 0 0 0 0 0 32 32
Total 7 25 44 35 55 34 200
Table 4: S&P and Moody’s Results
Linear and quadratic weights are shown in Table 5 and Table 6.
Moody's
AAA AA A BBB BB B
AAA 1 0.75 0.56 0.32 0.3 0
S&P AA 0.75 1 0.75 0.56 0.3 0.28
A 0.56 0.75 1 0.75 0.6 0.32
BBB 0.32 0.56 0.75 1 0.8 0.56
BB 0.28 0.32 0.56 0.75 1 0.75
B 0 0.28 0.32 0.56 0.8 1
Table 5: Lineer Weights Results
Moody's
AAA AA A BBB BB B
AAA 1 0.89 0.85 0.72 0.4 0
S&P AA 0.89 1 0.89 0.85 0.7 0.43
A 0.85 0.89 1 0.89 0.9 0.72
BBB 0.72 0.85 0.89 1 0.9 0.85
BB 0.43 0.72 0.85 0.89 1 0.89
B 0 0.43 0.72 0.85 0.9 1
Table 6: Quadratic Weights Results
170

The observed and expected ratios calculated with linear weighting schemes are 0.965 and 0.923 respectively.
With quadratic weighting schemes these are 0.991 and 0.973 respectively. The weighted agreement exceeding the
expected weighted agreement among the relevant credit rating agencies is calculated 82.78 % with linear weights
and 92.76 % with quadratic weights.
Credit rating agencies “Moody’s” and “Fitch” were employed in the study as the second pair of decision
makers. The categorical distribution for the 185 banks rated by both of the agencies is shown in Table 7.
Moody's
AAA AA A BBB BB B Total
AAA 5 1 0 0 0 0 6
Fitch AA 1 15 5 6 0 0 27
A 2 10 16 1 0 0 29
BBB 0 0 1 38 8 0 47
BB 0 0 0 5 33 5 43
B 0 0 0 0 0 33 33
Total 8 26 22 50 41 38 185
Table 7: Fitch and Moody’s Results
The observed and expected ratios calculated with linear weighting schemes are 0.927 and 0.875 respectively.
With quadratic weighting schemes these are 0.972 and 0.916 respectively. The weighted agreement exceeding the
expected weighted agreement among the relevant credit rating agencies is calculated 79.32 % with linear weights
and 88.01 % with quadratic weights.
Credit rating agencies “S&P” and “Fitch” were employed in the study as the third pair of decision makers. The
categorical distribution for the 194 banks rated by both of the agencies is shown in Table 8.
S&P
AAA AA A BBB BB B Total
AAA 2 1 0 0 0 0 3
Fitch AA 2 9 3 3 0 0 17
A 2 8 14 2 0 0 26
BBB 0 0 2 32 4 0 38
BB 0 0 0 3 31 4 38
B 0 0 0 0 15 57 72
Total 6 18 19 40 50 61 194
Table 8: S&P and Fitch Results
The observed and expected ratios calculated with linear weighting schemes are 0.940 and 0.896 respectively.
With quadratic weighting schemes these are 0.982 and 0.950 respectively. The weighted agreement exceeding the
expected weighted agreement among the relevant credit rating agencies is calculated 88.21% with linear weights and
91.01 % with quadratic weights.
As a result of the abovementioned comparisons S&P and Fitch were identified as the two agencies with the
greatest agreement according to the values of the weighted kappa coefficient calculated using both linear and
quadratic weighting. Using the same method, the study concludes that S&P and Moody’s are the two rating agencies
with the lowest agreement. In light of the degrees of agreement used as agreement measure and considered fit in the
application, degrees of agreement exceeding the expected level of agreement is quite high among the three leading
credit rating agencies in the sector. Hence, it wouldn’t be wrong to count on the credit scores given by any of these
three agencies.
171

5 Conclusion
Despite various criticisms regarding rating agencies’ increased importance for national and international markets,
the importance of rating agencies for market players is great. As a result of globalization together with the
integration of financial markets, the influence of rating on markets is rising day by day. Markets acknowledge that
rating agencies are successful in rating risks. On the other hand, economic crises and unexpected companies
declaring bankruptcy pose questions about performances of rating agencies. Several studies revealing very different
results were conducted to see whether rating agencies have any information value for markets or not.
During the Basel II process, credit rating agencies gained ever more importance. Basel II’ requirement for
internal and external audit in banks indicates that the need for credit rating agencies will increase in the future.
For classification of units is a complicated process, discovering which decision maker is more reliable than the
other poses a more important concern. Hence, classifying a unit sample from two or more decision makers
independently and identifying the agreement and meaningfulness of this operation become a matter of interest.
When the number of levels increase and the significance of disagreement among levels is not equal, it is more
appropriate to use the expanded weighted kappa coefficient statistic. Looking at the global economic crisis and
banks declaring bankruptcy despite being given the top credit scores by credit rating agencies, the study aims to
measure the agreement among credit rating agencies.
Data from EU banks rated by credit rating agencies were compared in pairs. The subsequent levels of agreement
correspond to high agreement values reached in the application. Hence, as a result of the application, it was
concluded that the credit scores given by credit rating agencies are applicable. In other words, these agencies share
an agreement. While there are high levels of agreement between all credit rating agencies, agreement is calculated
the highest between S&P and Fitch, and the lowest between S&P and Moody’s.
6 References
Agresti, A. (1992). Modelling Patterns of Agreement and Disagreement. Statistical Methods in Medical Research,
1(2), 201-218.
Ben-David, A. (2008). Comparison of Classification Accuracy Using Cohen’s Weighted Kappa, Expert Systems
With Applications, 34(3), 825-832.
Cicchetti, D. V.& Allison, T. (1971). A New Procedure for Assessing Reliability of Scoring EEG Sleep Recordings.
American Journal of EEG Technology, 11(2), 101-109.
Cohen, J. (1968). Weighted Kappa: Nominal Scale Agreement with Provision for Scaled Disagreement or Partial
Credit. Psychological Bulletin, 70(1), 213-220.
Fleiss, J. L. & Cohen, J.(1973). The Equivalence of Weighted Kappa and the Intraclass Correlation Coefficient as
Measure of Reliability. Educational and Psychological Measurement, 33(2), 613-619.
Fleiss, J. L., Levin, B., & Paik, M. C. (2003). The Measurement of Interrater Agreement. Statistical Methods for
Rates and Proportions 3rd ed., John Wiley & Sons. Inc., New-Jersey.
Jarrow, R. & Turnbull, S. (2000). The Intersection of Market and Credit Risk. Journal of Banking &Finance 24(2),
271-299.
Landis, J. R., & Koch, G. G. (1977). An Application of Hierarchical Kappa-type Statistics in the Assessment of
Majority Agreement Among Multiple Observers. Biometrics, 33(3), 363-374.
Lando, D. (1999). Some Elements of Rating-Based Credit Risk Modeling. Working Paper, No:123 Department of
Operations Research, University of Copenhagen.
Ludbrook, J., (2002). Statistical Techniques for Comparing Measurers and Methods of Measurement: a Critical
Review,.Clinical and Experimental Pharmacology and Physiology, 29(1), 527-536.
Longstaff, F. A., & Schwartz, E. S. (1995). Valuing Credit Derivatives. Journal of Fixed Income, 5(1), 6-12.
Nickell, P., Perraudin, W., & Varotto, S. (2000). Stability of Rating Transitions. Journal of Banking & Finance ,
24(1), 205-229.
Nye, R. P., Eke S. (2004). Türkiye’de Kredi Derecelendirmesi, Activeline, 30-46.
White, L. J. (2002). The Credit Rating Industry: an Industrial Organization Analysis. Ratings Rating Agencies and
the Global Financial System 1st ed., Kluwer Academic Publishers, New-York.
Vanbelle, S. & Albert, A. (2009). A Note on the Linearly Weighted Kappa Coefficient for Ordinal Scales. Statistical
Methodology, 6(3), 157-163.
172

THE GLOBAL FINANCIAL CRISIS AND THE BANKING SECTOR IN SERBIA�
Emilija Vuksanović, PhD
Violeta Todorović, M.A.
Abstract. Financial crises in general are a common part of the development of the banking systems in the modern times. The modern
times and environment can be defined as the complex, dynamic, heterogeneous and unpredictable. The biggest crisis of the 21 st
century (the "Subprime" crisis) spread very quickly to include the real economy and encompassed the whole world. Although it is not
yet possible to evaluate completely the scope of the crisis which started in June 2007 on the real estate market in the USA, it is
obvious that the global connections influenced its spreading to the financial markets and economy of other continents. In reality the
market globalization stipulated the globalization of the financial crisis and the recession of the economy. The Serbian banking sector
was not immune to the global financial crisis either. The first blow of the crisis was felt in the banks in October 2008, when there was
a large-scale withdrawal of the deposits and the growth of the banks` interest rates. However, if compared with the neighbouring
countries, the Serbian banking sector met the crisis adequately capitalized and highly solvent, owing to the anti-cyclic monetary policy
and the undertaken prudential measures by the National Bank of Serbia. The aim of this paper is the analysis of the state of the
banking system in Serbia in the circumstances of the Global financial crisis and its future prospective in the circumstances of the
increased instability and turbulence.
Key words: financial crisis, banking sector, regulatory measures, capital adequacy
JEL classification: G2
1. Introduction
The first major financial crisis in 21st century that includes "esoteric instruments, unaware regulators, and skittish
investors“ (Reinhart and Rogoff, 2009, p. 291), has spread rapidly to real economy and grasped the whole world.
Presence of global connection caused the far-reaching consequences of current financial crisis on world economy
and finance. This caused global recession followed by the drop in living standard, increase of unemployment and
inflation and growing budgetary and foreign trade deficit. Naturally, Serbian banking sector was not immune on the
effects of global financial crisis.
The subject of this analysis will be the intensity and the character of world trends’ influence on banking system
in Serbia, on one hand and on the other the effectiveness of national financial regulations as a response to crisis
effects. Since the banks are the bearers of financial system in Serbia, it is logical to expect the key effects of the
financial crisis to reflect primarily on banking sector, and the fact that this sector is slightly more developed in
comparison to other sectors in Serbia, leads to the presumption that precisely this sector will be crucial in
stabilization and further growth of Serbian economy. The response of national financial regulations to the effects of
global financial crisis was based precisely on these facts.
In order to establish to what extent and in which way banks were exposed to the effects of global financial crisis
this paper will first of all, analyse the characteristics of Serbia's banking sector in pre-crisis period. Further analysis
deals with determining concrete effects of banking crisis and effectiveness of the regulations’ response. The analysis
will be rounded up by determined open questions and estimates of the perspective of further development.
2. Characteristics of Serbian Banking Sector in Pre-crisis Period
During the 90s of the last century, Serbian banking sector was practically ruined, First of all, because of the
conditions under which banks operated (political instability, high inflation rate, economic isolation, the loss of
foreign exchange savings and the fact that citizens completely lost trust in banking sector). Such state of the
banking sector was in no way sustainable, which called for starting mechanisms for its reconstruction.
After the reform of banking system was conducted, analyses showed that transformation and reconstruction
processes initiated in 2001 brought about significant positive effects in the banking sector:
- the structure of the banking sector was enhanced (on one hand, overall number of banks was reduced from
86 in 2001, to 49 by the end of the year; also the number of state-owned banks reduced, while on the other hand, the
number of private banks owned by either foreign or domestic private persons increased)
� The paper is the result of the work on project funded by the Ministry of Science and Technology of the Serbia Republic, No. 47004 (2011-
2014) titled: Upgrading the public policy in Serbia in function of improvement of the social security of citizens and sustainable economic growth.
173

- the losses of entire banking sector have been reduced (by the beginning of 2001, the losses were 48 billion
RSD, and already by the middle of 2002 they dropped to 11 billion RSD);
- the trust of citizens was regained, viewed from the aspect of savings, especially foreign exchange savings;
- the share of capital in total liabilities has increased (by the end of 2000 the share of capital was only 3,4%
of total liabilities, while by the end of 2001 it amounted to 46185,7 million dinars, in other words, its share in total
liabilities increased to 15,9%);
- improved compliance of banks’ business operations from the aspect of relative business indicators, as
shown in table 1.
Indicators 31. 12. 2001. 31. 12. 2000.
Capital adequacy ratio (min. 8%) 21,9 0,7
Short-term placement to resources ratio (>=100%) 138,4 138,0
Foreign currency liabilities to foreign currency assets ratio (95-105%) 94,3 96
Large and largest possible exposure ratio (max 80%) 233,3 3928,9
Share in company capital ratio (max 15%) 6,0 17,9
Ratio of the share in capital of banks and other financial institutions (max
51%)
5,2 33,3
Ratio of the placements in capital assets (max 20%) 33,1 149,8
Table 1 – Relative Business Indicators
Source: Annual Report 2001, National Bank of Yugoslavia, p. 90. http://www.nbs.rs/internet/latinica/90/index.html
Based on the presented table, a conclusion can be drawn that in comparison to 2000, compliance of banks’
operation was significantly improved in terms of capital adequacy, large and largest possible exposures, share in
company capital and placement in capital assets.
The process of privatization of banks in which state had major or minor share started in 2004. Existing
ownership structure in Serbian banks, the dominant part being taken by foreign ownership, represents a significant
indicator of internalization of banking sector. The process of internalization was especially intensified in the period
2004-2007, when the number of domestic (private and state-owned) banks reduced markedly in comparison to the
foreign banks, from the aspect of number of banks, capital and total assets. By end-2009, total number of banks
operating in Serbia reduced to 34, in 20 of which foreign stakeholders have majority shares, four are owned by
domestic natural or legal persons, while the Republic of Serbia has the majority of shares in ten banks. Based on the
current state, it can be concluded that majority of Serbian banking sector is in foreign ownership. However, when
compared to other countries in transition process (countries of South-East Europe, which includes Serbia as well),
the level of internationalization of Serbian banking sector is much lower.
In terms of market share, Serbian banking sector is still quite fragmented, that is to say, insufficiently focused,
which is obvious from the fact that even 18 banks have a market share less than 2% (measured share of the assets of
a bank in relation to total assets of entire banking sector). By the end of Q4 of 2009 only one bank - Intesa had
market share above 10%, to be precise 14,25%. 1 Therefore, market structure, as a relevant factor in the development
of banking market, in combination with competition, represents the factor of efficiency and stability growth of
banking system and economic development of every country. That is why it can be expected that the
internationalization of Serbian banking system will continue through mergers and affiliation, for the purpose of
boosting the capital, improving competitiveness and cutting down borrowing costs.
Clearing the balance and privatization of state-owned banks created prerequisites for a more efficient and more
profitable business operation of banking sector.
ROE and ROA indicators (traditionally the most relevant indicators for measuring profitability) have increased
exponentially in only few years. The highest value of ROA was achieved in 2008, despite the fact that global
financial crisis had already shaken the banking sector. However, it should be noted that in due to the crisis, in Q3 of
2009, ROA marked a drop of 0,66% in comparison to the previous quarter. What is concerning is the drop of ROA
by the end of 2009 was 1,02% (by the end of 2008 ROA was 2,08%). Return on equity (ROE) has also recorded a
continual rise in the initial years of transition, while it peaked in 2006, only to suffer a slight drop the following year
due to recapitalization of banks. 2 After that ROE went up again (by the end of 2008 ROE was 9,28%), but global
1 Bank Supervision - Fourth Quarter Report 2009, National Bank of Serbia’s Bank Supervision Department,
http://www.nbs.rs/export/internet/cirilica/55/55_4/index.html
2 Banks have performed a recapitalization due to the NBS directive on gross placement ratio for the citizens according to the total capital which
could not exceed the maximum of 150% at the time. With maximum ratio achieved, banks reacted by increasing the capital in order to maintain
and increase the lending volume for citizens. This way a space for additional lending of the citizens was created. When the crisis set in, NBS
increased margin of this ratio to 200% as a part of simulative measures for overcoming the crisis, which created even more space for lending.
174

financial crisis caused it to drop (by the end of 2009, it dropped to 4,62%). The drop in values of both indicators is
the result of the decrease of profit before tax by 42,3% in comparison to the same period in 2008. 3
From the viewpoint of indicators of financial position and strength, Serbian banking system recorded a slight
nominal and real interannual increase in 2009 (table 2). In spite of unfavourable macroeconomic trends caused by
global financial crisis and weaknesses of internal financial system (pressure of inflation, Dinar depreciation), it is
evident that the stability of Serbian banking system was maintained in 2009. Capital adequacy ratio is still high
(21,44% by the end of 2009), which is a result of good risk management in banks and conservative regulatory policy
of National Bank of Serbia. Therefore, trend of reducing total liabilities in liability structure was continued.
On the other hand, Serbian banking system has a stable funding source structure, since the share of domestic
deposits in total liabilities is approximately 70%. In combination with high amount of the capital, it is clear that
balance structure of banks in Serbia is much more favourable than that of most banks in Europe. Also banking sector
in Serbia has adequate currency structure of deposits (75,4% of total deposits are in foreign currencies). However,
maturity structure is not satisfactory, due to dominant share of short-term deposits (95,7%) in total deposits. That is
why there is a big maturity incoherence between the source and placement (long-term placements are up to 5 times
higher than long-term sources). 4
Indicator 31
31
31 Growth index for 2008 3/2 Growth index for 20094/3
.12.2007. .12.2008. .12.2009.
1 2 3 4 5 6
Balance sheet total 15
17
21
114 122
61,8 76,9 60,4
Total capital 32
41
44
128 107
8,5 9.9 7,5
Lending activity 76
0,9
Deposit activity 96
0,2
10
27,6
10
24,7
12
78,3
13
01,2
135 124
107 127
Table 2 – Indicators of financial position and strength (in billions RSD)
Source: Bank Supervision – Report for Fourth Quarter 2009, National Bank of Serbia’s Supervision Department,
http://www.nbs.rs/export/internet/cirilica/55/55_4/index.html
In addition, structural imbalances between banks in terms of financial strength and business results represent a
challenge for further development of banking sector. There is a small number of leading banks on Serbian banking
market and a comparatively large number of banks with modest business results. Apart from that, there are also
banks with high losses in their operation. These differences are expected to become more visible as the competition
grows stronger. 5
Based on the previous analysis it can be concluded that foreign direct investments in Serbian banking system
and appearance of foreign banks brought about a more competitive approach from the banks and increased available
funding sources. In the period 2001-2006 over 20 billion dollars flowed in Serbia through remittances of emigrants,
foreign investments, donations, and loans, including privatization and borrowings of the companies and banks
themselves. (Kovačić, 2006, p. 59) This created conditions for a rapid, dynamic and stable growth of banking sector
in an economic environment characterised by slow growth, liquidation of numerous companies and poverty.
However, when compared to other countries in the region, especially to new EU members and candidates for
ascension, development of Serbian banking system is still pretty modest. Share of total liabilities in domestic
product is lower than in mentioned countries. Such situation is only natural, since the transition process in Serbia
starts in 2001, that is to say, over ten years later than in other European countries that are in the transition process.
Still, general state of banking system in Serbia is much better than economy sector, which is clearly seen in the
results of business operation, growth of balance assets, changes in legal regulations, compliance with international
standards in calculation of business results and risk evaluation.
3 Bank Supervision – Fourth Quarter Reportl 2009, National Bank of Serbia’s Bank Supervision Department,
http://www.nbs.rs/export/internet/cirilica/55/55_4/index.html
4 Ibid.
5 Serbian banking sector 2009 – Analysis of fiunancial position and financial results, Association of Serbian Banks, p. 13, http://www.ubs-
asb.com/Portals/0/vesti/76/Analiza2009.pdf
175

Generally speaking, restructuring of banks, conducted with active role of the state, has created a pretty stable
banking sector, which even in crisis situation maintained a good level of profitability, adequate capitalization and
corporative performance. However, the problem yet unsolved are high interest margins, due to lack of competition.
This makes it more difficult for the economy sector to finance the investments through borrowing. This is why a
more intense and more subtle competitive battle on banking market can be expected to the purpose of keeping
existing and attracting new clients. Basic strategic goal of future development of banking sector in Serbia is
improved efficiency in business, that is to say, reduction of financial intermediation costs. 6
3. The impact of the crisis on Serbian bank sector and the regulations' response
The first impact of economic crisis Serbian banks felt in October 2008 when in just a month and a half, in panic
attack, the foreign currency savings were withdrawn in the amount of 960 million Euros. At the same time, the
domestic banks with foreign ownership started prematurely to return credit lines to their parent banks, which
represented additional factor for reducing the foreign currency offer in Interbank Market, and consequently,
depreciation of dinar followed.
To increase the foreign currency offer, the National Bank of Serbia took the following measurements:
1)introduced subsidies on new borrowings abroad, through changes of regulations regarding the required reserves,
and provided that new borrowing is not burdened by allocation of required reserve, 2) two times increased the
amount of foreign currency required reserve that banks calculate on foreign currency base (but keep in dinars on
bank account)- at first with 10% to 20%, and then 20% to 40% of allocated liabilities, 7 and 3) „froze" the absolute
amount of allocated required reserve based on September 2008 in order to prevent prepayment in foreign banks.
This was a disincentive for banks not to return loans prematurely to their parent banks, because their allocated
reserves are not returnable according to the basis of required reserve. It is significant that this measurement didn't
explicitly forbid banks to return the loans to their parent banks, but it was an indirect disincentive. Also, this
measurement was only temporary. (Kokotović, 2008, p. 71)
All these measurements, with everyday interventions of the National Bank of Serbia on Interbank Market and
with 17.75 % increased key policy rate didn’t stop depreciation of dinar.
Stopping cross-border loans, saving retreats and high interest rates on indexed loans, made a big impact on loan
declination in Serbia in the last quarter of 2008However, during 2009 the increase of bank lending followed. The
total amount of placed loans at the end of 2009 was 1.278,3 billion of dinars which is 24.4% more than at the and of
2008, when the total amount of placed loans was 1027,6 billion RSD. It was approved 750,4 billions RSD to the
corporate lending, to the household 395 billion RSD, public sector 118 billion RSD, of which 106.7 billion RSD is
state debt. 8
At the same time, there was the increase in non-performing loans. Based on data of non-performing loans that
banks submit to the National Bank of Serbia, at the end of 2009, their percentage in total of approved loans, on net
level, was 8.53% or 99 billion RSD. Otherwise, total credit indebtedness of companies and citizens in Serbia at the
end of 2009 was 1.400 billion RSD, which is 13.3% more than at the end of 2008. 9
Thanks to the required reserves high rates policy towards external indebtedness and to domestic and foreign
currency deposits, the Serbian bank sector in advanced stages of economic crisis showed the coverage of the foreign
currency reserve deposit up to 86%, which is much higher than the indicators in other countries in the region, where
it was 35%. 10 However, comparing to the neighbouring countries, Serbia has the highest allocation rates of required
reserves (table 3), which led to lower efficiency and profitability of Serbian banks.
state Required reserve rate
Serbia During 2010- 5% on dinar savings, 40% on citizens’ foreign currency savings, and 45% on
6
Basic parameters of the efficiency of banks in Serbia have high negative values. Costs of intermediation (net interest margin) are much higher
than in developed countries. According to: Boško Živković, Financial Sector - Banking, within the study: Reforms in Serbia: Achievements and
Challenges, Center for liberal-democratic studies, Belgrade, June 2008, p. 41.
7
This is a temporary measure, adopted with the aim to help surplus of foreign liquidity remains in Serbian banking sector, without using it for
settlements with banks abroad or for reduction of foreign deposits.
8
Bank Supervision – Report for Fourth Quarter 2009, National Bank of Serbia’s Supervision Department,
http://www.nbs.rs/export/internet/cirilica/55/55_4/index.html
9
Banking Sector in Serbia 2009-2010, Business Magazine, 31.01. 2010, http://www.naslovi.net/2010-01-31/poslovni-magazin/bankarski-sektoru-srbiji-2009-2010/1521472
10
Financial Stability report, National Bank of Serbia, September 2008, p. 43. http://www.nbs.rs/internet/latinica/90/index.html
176

Croatia
corporate foreign currency deposits
2011 plans- unique 25% foreign currency required reserve and 5% dinar required reserve
13%
Bosnia and
Herzegovina
14% on short-term sources and 7% on long-term sources
Bulgaria 10% on domestic sources, 5% on foreign sources and 0% on state sources
Romania 15%
Hungary 2%
Slovakia 2%
Slovenia 2%
Albania 10%
Table 3- Rate of required reserves in Serbia and neighbouring countries
Source: We are all speculators, Politika online, May 20th 2010, http://www.politika.rs/rubrike/Ekonomija/Svi-smo-mishpekulanti.sr.html
Although the Serbian bank sector finished business 2009 with positive financial result (14 banks showed
operating loss, which makes around 17% of total bank sector), the efficiency of use of resources is not enough,
which was reflected on height of return on assets employed (ROA) (1.02%) and on height of return on proper
capital- ROE (4.62%) at the end of 2009. Though in 2009 both rates recorded the mild growth, the Serbian bank
sector still takes lower position in profitability compared to the neighbouring countries. It is cleared that on
efficiency and economy of use of assets employed the big impact had high rates of immobilisation of banks'
financial potentials based on required reserve.
Looking at the current situation, it can be concluded that Serbian bank sector has well endured the main impact
of world’s economy crisis. This situation is, primarily, the result of National Bank of Serbia’s conservative
regulations, with excessive disincentives for corporate and citizens’ lending. In fact, the National bank has in time
recognized the warning signals that pointed to danger of excessive growth in bank lending. During 2007, macro
indicators pointed that loan growth was already in alarming zone and that it was rapidly going towards critical point.
At the same time, the citizens’ loans were growing faster than citizens’ deposits, there were major concentration of
lending in long-term indexed loans, while maturity structure was considerably incompatible (the ratio of loans over
12 months and term deposit with the same maturity was 4:1).
Therefore, the fact that the crisis led to 2.8% fall of economic activity in the country during 2009 cannot be
bypassed. However, none of 34 banks, that are currently operating on the market, had no major problems of
liquidity.
No matter that almost all commercial banks had profit fall during 2009 compared to the previous year, stability
was maintained, even the size of foreign currency savings was increased comparing to the period of before the crisis.
At the end of 2009 the citizen savings was 6.2 billion of Euros or 577 billion RSD, which means that in the last ten
years the total savings were increased for 4500% or 45 times. (Dugalić, 2010, p. 4) This success is that much bigger
considering that at the end of 2008, in the atmosphere of uncertainty due to large number of bank bankruptcy
abroad, almost one billion of Euros of foreign currency savings were withdrawn out of the banks. The reasons for
positive changes regarding the citizen savings should be, primarily, found in: Successful restructuring of bank
sector; restoring public confidence in banks; decision taken to increase the amount of insured deposits from 3000 to
50000 Euros, whose payment in case of liquidation or bankruptcy of bank is guaranteed by state, and the attractive
bank offers, reflected in high interest rates in domestic and foreign currency savings. (Dugalić, 2010, p. 4)
As a confirmation of previous states the 2010 IMF report can be used, in which was estimated that in Serbia in
2009, by the Government and the National Bank of Serbia’s measures, negative consequences of economic crisis
were mitigated and that the country ended that year with better macroeconomic indicators than other Eastern
European countries
The negative consequences of the crisis in Serbia would be more serious if banks, whose founders are from
abroad, did not keep the same size of the loans as in 2008In fact, in March of 2009, on the recommendation of IMF,
ten banks (later that number was increased to 27) signed Financial Sector Support Program in Vienna, so called
Vienna Agreement, committing themselves to reprogram private debt during 2009 that comes to due in amount of
4.5 billion Euros. In that way it was provided to the companies in Serbia to postpone the payments of earlier loans
177

and at the same time take the new, subsidized loans for settlement of regular duties towards the state and suppliers.
Considering that the goal of the National Bank of Serbia is to draw capital and new loans in a market manner, the
Vienna Agreement represented temporary and emergency measurement, whose validity ceased at January 1st 2011.
In accordance with the Vienna Agreement, the National Bank of Serbia prepared special measurements to
support financial stability of the country with goal to preserve the public trust in banking system and to maintain
financial and macroeconomic stability. In addition, those special measurements were available only to those banks
that took the obligation to maintain the exposure to Serbia till the end of 2010, based on the 2008 December level
and maintaining indicators of capital adequacy and liquidity indicators on given level.
For the banks that meet these conditions, the National Bank of Serbia provided the following facilities: new
liquidity sources, i.e. dinar loans with a repayment period of no longer than 12 months, and short-term foreign
currency swap transactions, including the abolishment of reserve requirements for deposits and loans received from
abroad from October 2008 to December 2010 until their maturity date. Also, banks were allowed to include in their
capital, for regulatory purposes, subordinated liabilities up to 75% of their core capital, in the calculation of arrears
on loans whose repayment terms were rescheduled under the framework defined, and for the purposes of their
classification requirements, banks are permitted to apply the subsequently agreed maturity date, as well as to raise
foreign exchange risk ratio from 10% to 20% of their capital. 11
From today’s perspective, the banks signatory not only kept their exposure to Serbia, but during 2009 increased
their investments for 2%. At the same time, in the conditions when the number of non-performing loans increased at
the end of 2009 comparing to the same period in 2008, the banking system was adequately capitalized, and
capitalization of 21.44% is for 9.44% above legal minimum.
In order to assess the ability of the banking system to maintain minimal prescribed capital rate of 12% and in
case of worsening business conditions, the National Bank of Serbia, has conducted three stress tests. During the last
testing, whose results were published in October 2009, the National bank of Serbia used regression model that was
supposed to show the influence of independent variables, such as: The change of GDP, production gap, depreciation
and the change of interest rates to deterioration of loan portfolio and losses according to non-performing loans,
during 2010 and 2011. (Jelašić and Erić Jović, 2009)
16 banks were tested, with the involvement of over 80% of the balance sheet of the total banking system.
Starting from the pessimistic scenario, by which the increase of non-performed loans would be up to 13.9%, the
indicator of capital adequacy would still stay above the legal minimum of 12%Therefore, the biggest impact on
increase of non-performing loans had the fall of economic activity seen through the change of GDP and production
gap. Based on that, it can be concluded that even if the pessimistic scenario would be realized, the banking sector
would be capable of covering all the potential losses with the existing capital and reserves. That means that banking
sector in Serbia has no need for emergency and preventive financing, because the indicator of the capital adequacy
would still be above the statutory minimum, i.e. it would be lower from the initial 22.36% to 16.04% in 2010, i.e.
15.16% in 2011. 12
Comparing to the neighbouring countries, only the National Bank of Serbia took all the relevant prudential
measurements in order to preserve macroeconomic and financial stability and safety of banking sector in the crisis
period. Recognizing the importance of banks for the stability of the financial system in Serbia, the National Bank led
anti-cyclical policy, and took the measurements regarding the minimizing risks of loan growth. In that way strong
absorbers to cover unexpected losses were created and banking sector was capitalized adequately and met the crisis
highly liquid.
e
a
s
u
r
e
Prudential measurements
11 Vienna’s Initiative, National Banks of Serbia, http://www.nbs.rs/export/internet/cirilica/18/18_6/index.html.
12 Banking sector highly capitalized and resistant to macroeconomic shocks, National Bank of Serbia, 13. 01. 2010,
http://www.nbs.rs/internet/latinica/scripts/showContent.html?id=3876&konverzija=yes
178

High level of required
reserve
The exposure limits
and risk concentration
Measurements with the
goal of slowing the loan growth
Strengthening the
foreign currency liquidity
Memos on cooperation
with international institutions*
Programs aimed at
maintaining the exposure of
banks
Serbia √ √ √ √ √ √ √ √
Croatia � √ √ √ √ � √ √
Hungary � √ � √ √ √ √ √
Romania √ √ √ � √ � √ √
Bulgaria � √ √ � √ � √ √
* On programs of consolidated supervisions of bank groups or "home host" supervisions for more comprehensive risk view
** Mainly on relation: Central bank-supervisor-Ministry of Finance
Table 4- Measurements taken by supervised institutions
Relaxation of the
regulatory framework
Memos on cooperation
with domestic institutions
regarding the stability
preservation**
Source: Mira Erić Jović, Serbian banking sector in the global financial crisis, presentation, Palic, May 2009, p. 14.
It can be concluded that the National Bank of Serbia, establishing the set of prudential measurements, 13 wanted
to mitigate the negative effects of crisis through improving liquidity, solvency preserving and strengthening trust in
the banking system.
4. Open questions and prospects
Evidently, global crisis didn’t have a major impact on Serbian banking system, which is still comparatively stable,
with high liquidity and adequate capitalization. First of all, there were no direct risks in terms of investments in
mortgage loans securitizations and other high-risk financial instruments that are the basis of global financial crisis.
Furthermore, restrictive measures of NBS, that caused high liquidity, adequate capitalization and overestimated
reservations for uncollectible loans, turned out to be the advantage of national banking system in comparison to
other countries from the region. At the same time, these measures pretty quickly mitigated negative psychologically
induced factors that lead to widespread withdrawal of citizens’ funds on the very brink of the crisis. Having in mind
that the banks have successfully met citizens' demands for withdrawal of funds, already in December 2008, not only
was the outflow stopped but there was also new deposit inflow. This, practically, preserved the credibility of the
banking sector.
However the crisis had its impact on other sectors, which calls for highly coordinated actions of the
Government, National Bank of Serbia and all other relevant institutions in order to maintain the credibility of
financial sector, enable attraction of foreign capital and improve Serbia's reputation. In accordance with afore
mentioned, by the beginning of 2011, Serbian Government adopted a programme for subsidized loans for the
economic subjects and citizens, with the aim to ease the negative effects of the crisis. Seven billion dinars have been
allocated for liquidity loans, financing of working assets and loans for export businesses with subsidized interest
rates. Even though this kind of lending can stimulate various frauds, at this point it is a necessary action that will
boost economic activity and mitigate negative effects of recessive trends of real sector.
A while later, more precisely, in April 2011, the Government has adopted measures for citizens’ debt
rescheduling, that will reduce monthly instalments of long-term loans for the period of two years. Additionally,
grace period of two years will apply for all the loans with maturity date longer than 12 months, except for subsidized
consumer and cash loans and loans with already arranged grace period and rescheduling. At the moment of adopting
the measures, 11 leading banks accepted the debt rescheduling programme. At the same time, National bank of
Serbia has adopted the decision that ensures greater flexibility of banks in support for citizens and enterprises.
Actually the limit on monthly instalment deduced from salary was abolished, so now total monthly debt obligations
13 Required reserves for foreign currency loans have been abolished, the maturity date was extended by one year, conversion of indexed credits
into dinars was enabled, as well as premature payment, ratio citizens’ loans/capital was increased from 150% to 200% etc.
179

is in ratio with regular net monthly income reduced by the minimum consumer basket for the first adult household
member. Furthermore, the clients no longer have to put down a deposit or down payment of 30% while applying for
a foreign currency loan. National Bank of Serbia has also reduced the period in which the debtor should discharge
obligations towards the bank for the purpose of a more favourable classification according to delay criteria, from six
to three months, and the regularity in discharge of obligations from 60 to 30 days within the period.
Even though by the middle of 2010, uniform rates of allocation to required reserves (as shown in table 3), by the
beginning of 2011 a decision on introduction of differentiated rates depending on maturity of dinar and foreign
funding sources. Hence, for dinar fund sources with maturity up to two years required reserve rate will remain 5%,
while required reserves will not be calculated for dinar sources with maturity over two years. For foreign currency
funding sources with the maturity up to two years the rate is 30%, while it remains 25% for the maturity over two
years. At the same time the allocation of dinar portion of the foreign currency required reserves was calculated at
different rates, 15% for sources with up to two years maturity and 105 for sources over two years maturity.
The basic objective of this decision related to controlling the inflation, i.e. preventing large amounts of dinar on
the market in the state of high inflation. Moreover, the aim was to achieve more balanced rates of required reserves
that ranged from 9-43% in the previous period. Starting from different effects the new decision will have on banks,
depending on their activities in the previous period, compliance of calculation and allocation of required reserves
will be conducted in three instalments starting from February 17, and ending with the calculation on April 17 2011.
The biggest dilemma about this new decision concerning required reserves rates addresses the increase of dinar
portion in foreign currency sources. Actually, the banks are afraid that higher dinar required reserves will reduce the
offer of dinar loans and increase the financing costs for banks. This measure is somewhat in conflict with
dinarization on which the Government insists on. Virtually, a measure is introduced to limit dinar liquidity, which
leads to a question: how will the liquidity be provided for dinar loans? Having that in mind, a rise of loan prices is
expected, which will eventually depend on competition and individual bank policy,
In order to maintain the credibility of overall system of financial intermediation, it is necessary to pay more
attention to protection of financial services consumers, as it can minimise the examples of bad practice. To that
purpose, by the end of 2010, a Draft of The Law on Financial Services Consumer Protection was adopted, which
conflicted the opinions of National bank and Association of Serbian banks. The aim of this new Law is regulation of
protection of rights and interests of financial services consumers, better position and informing of the clients. It
should provide better information flow and rights of clients in their cooperation with business banks. Even though
the draft was designed in accordance with European directives, banks are completely right when they point out
certain controversial regulations that do not meet interests neither of banks nor the clients. If such a draft was to be
adopted, certain types of loans would definitely have to be abolished or their price should have to be increased. This,
primarily, applies to credit cards and dinar loans with fixed interest.
Finally, it should be noted that even though Serbian banking sector recorded a slight rise of the indicators of
financial strength and results by the end of 2009, caution should not yet be abandoned in favour of too much
optimism. Economy facing recession in 2009 and the fact that there is a high functional dependency between
economy and banking sector, represent the key challenges for stability of banks in Serbia. This analysis emphasises
that the main risks of banking sector are caused by the prolonged recession and the growing lending risk, due to
foreign currency borrowings of companies and citizens (in the structure of lending portfolio of banks, even 75% of
the loans are indexed in foreign currency) In the light of these facts, an increase of uncollectible loans is expected in
the upcoming years. Having in mind that granted loans are tied to Euro or Swiss Franc, stability will depend on
monetary policy of NBS. Amidst the expected challenges in the country and abroad, priorities of banking sector in
Serbia must be based on efficient risk management and quality of assets, in order to continue the uptrend of capital
adequacy.
5. References
Annual Report 2001. National Bank of Yugoslavia. http://www.nbs.rs/internet/latinica/90/index.html.
Banking Sector in Serbia 2009-2010. Business Magazine. 31.01. 2010. http://www.naslovi.net/2010-01-31/poslovnimagazin/bankarski-sektor-u-srbiji-2009-2010/1521472.
180

Banking Supervision – Report for Fourth Quarter 2009. National Bank of Serbia’s Supervision Department.
http://www.nbs.rs/export/internet/cirilica/55/55_4/index.html.
Dugalić, V. (2010). From Banking Point of View – Introduction. Banking. Association of Serbian Banks. Belgrade.
No. 1-2.
Erić Jović, M. (2009). Serbian Banking Sector in the Global Financial Crisis. presentation. Palic. May.
Financial Stability Report, National Bank of Serbia. September 2008.
http://www.nbs.rs/internet/latinica/90/index.html.
Jelašić, R., & Erić-Jović, M. (2009). Current Monetary and Macroeconomic Trends – Results of Stress Tests of
Serbian Banking System. National Bank of Serbia. Belgrade. October 8.
http://www.nbs.rs/export/internet/latinica/15/konferencije_guvernera/prilozi/20091008_amk_prezentacija.pdf .
Kokotović, S. (2008). Effects of NBS Measures on Banking System – Review 3, Reviews: Global Financial Crisis
and Serbia. Quarter monitor No. 15. October-December.
Kovačić, I. (2006). Changes in Serbian Banking 2003-2006. Survey – Republic of Serbia, No 2. Belgrade.
National Bank of Serbia. Banking Sector Highly Capitalized and Resistant to Macroeconomic Shocks. 13. 01. 2010.
http://www.nbs.rs/internet/latinica/scripts/showContent.html?id=3876&konverzija=yes.
Reinhart, R. C., Rogoff, S. K. (2009). Is the 2007 US Sub-prime Financial Crisis So Different? A international
historical comparison. Panoeconomicus. Vol. 3.
Serbian Banking Sector 2009 – Analysis of Financial Position and Financial Results. Association of Serbian Banks.
http://www.ubs-asb.com/Portals/0/vesti/76/Analiza2009.pdf.
Vienna's Initiative, National Bank of Serbia, http://www.nbs.rs/export/internet/cirilica/18/18_6/index.html
We are all speculator., Politika online. May 20th 2010. http://www.politika.rs/rubrike/Ekonomija/Svi-smo-mishpekulanti.sr.html.
Živković, B. (2008). Financial Sector – Banking. Within the study: Reforms in Serbia: Achievements and
Challenges. Center for liberal-democratic studies. Belgrade. June.
181

COLLECTION MANAGEMENT AS CRUTIAL PART OF CREDIT RISK MANAGEMENT DURING
THE CRISIS
Lidija Barjaktarovic, Snezana Popovcic- Avric, Marina Djenic,
Faculty of Economics, Finance and Administration, FEFA, Serbia
Email: lbarjaktarovic@fefa.edu.rs, www.fefa.edu.rs
Abstract: Collection management is one of the major areas of today's modern banking. With the appearance of the economic crisis
and coming into general illiquidity in the economy itself, banks increasingly have to manage portfolios with constant non-perform
loans increase (from 2% in average up to 20% in the area of Europe) which has direct impact on the quality of the portfolio (in terms
of classification, reservation and profitability). Influence is reflected both in profitability through the income statement of commercial
banks (in its early stage) and the capital of the banks themselves (in the late stage). For that reason management have to manage
collection management in a way to create the best model that can retain or improve the quality of credit portfolio. This issue is one of
the biggest challenges for the banking division.
The paper points to the complexity of the collection model and proposes a new model for the collection for corporate loans.
Developed banking markets, have fully developed mechanisms, procedures, and organized places where trade their bad debts and,
accordingly, advanced collection model. Domestic practices will just follow the practice of developed countries.
Key words: Collection management, credit risk, NPL, LGD, specific provisioning
JEL classification: G21
1. Introduction
Collection management has impact on bank’s profitability and all risk parameters (Loss Given Default, Standard
Risk Cost, and Risk Adjusted Price), decrease of new non-performing loan (NPL) portfolio and improvement of the
quality of credit portfolio. During the crisis relation between NPL and total credit portfolio is increasing. Before the
crisis this relation in Serbia was between 3% and 5%, now it is between 10% and 15%, and in some banks up to
20% (at the end of third quartile of 2010 in accordance with the data of National bank of Serbia). This bad structure
directly increases the quality of portfolio and increases its cost. On this way additional exposure is obligatory but
limited, because it is hard to cover the costs of existing portfolio.
In order to improve the quality and profitability of credit portfolio, we propose collection model for corporate
loans (on the basis of analysis of current credit portfolio data in Erste bank Serbia). This model has practical use in
Erste bank Serbia from October 2010. The real benefit of it we will now during this year.
2. Collection management – definition and importance
Collection management is the process of planning, organizing, control and monitoring of credit customers of the
banks with the purpose to collect credit receivables. Also, it includes all activities within the bank and in relation
with the customer in order to execute efficiently collection of its credit receivables. The bank should be aware that
only partnership and close relationship with the customer can build healthy portfolio without delay and problems in
collection of due debts.
Collection management is the complex part of credit process which involves different organizational parts of the
bank and employees with different competence. It is important to achieve synergy in their interaction and
cooperation. Also, it is important to emphasize that successful collection management depends on proper monitoring
of external factors and information.
182

In accordance with Basel II 1 credit portfolio of the bank consists of: 1) Performing loans - PL (live credits
which are not declared due and payable or it is the credit where principal amount and interest in repayment do not
have delays longer than 90 days) and 2) Non-Performing Loan – NPL (loans which are declared due and payable or
the credit is not in the accordance with the credit contract terms and conditions).
The most important benefits which come with the successful collection management are: 1) increase of the
banks’ profitability through usage of the reservation or through other incomes, 2) decrease of effects of new NPL
and improvement in the portfolio quality, 3) impacts on the main risk parameters: LDG (Loss Given Default), SRC
(Standard Risk Cost) and RAP (Risk Adjusted Price).
The most important influence of collection management is the decrease of reservation (which can be booked in
the profit and loss account or on the position of the equity) which has impact on increase of profitability or decrease
the equity. In the case of the group of customers which have defined level of the reservation on their group level,
collection management has small impact because the level of their reservation is low. But in the case of the small
companies or the companies which are not the part of any group, the level of reservation in relation with the gross
level of receivables is relatively high (30%, 50%, 70%,…), the successful collection management releases that part
of the cost in the profit and loss account or in the equity.
For example 2 the credit customer has matured obligation toward the bank in the amount of EUR 1.2 million,
where 1) EUR 1 million representing a principal amount with regular interest rate is matured and it is the balanced
part of the receivable. 2) EUR 0.2 million represents penalty interest rate and fees, which is booked off-balance
within the bank. It means that the total matured receivables of the credit customer toward bank are EUR 1.2 million.
Balanced part of the receivable is reserved with EUR 0.6 million, and off-balance part of reservation is EUR 0.2
million, which means that the total reservation is EUR 0.8 million.
In the case that the bank, sells the collateral at EUR 1.2 million, the bank will decrease directly the cost in the
profit and loss account through the decrease of the reservation in the amount of EUR 0.6 million, and income on the
basis of other income in the amount of EUR 0.2 million (because the bank managed to collect off-balance part too).
This way the collection management impact on the profitability of the bank represents an increase of the profit in the
amount of EUR 0.8 million.
In the practice this is a very important benefit, especially for the banks whose portfolios show stagnation or
decreasing trends.
On the other hand, controlled increased of NPL portfolio is one of the most important tasks which bank has
during the year. This task has become very demanding, especially with the appearance of the world economic crisis,
because in case of many banks their credit portfolios had stagnation or decreased trend, while NPL had natural
growth. At the same time the economic crisis had impact on NPL growth, which resulted in increase of portion of
NPL in the total credit portfolio. If this percentage was between 3% and 5% before the crisis, now it is between 10%
and 15%, and very often it goes up to 20%. This bad structure directly determines the quality of portfolio and it’s
pricing (increasing). This way, in order to increase the volume, additional exposure is obligatory but limited,
because it is impossible to cover the costs of the existing portfolio. 3
At the end it is important to mention that the collection management and management of NPL is the most
important determinant of PD (Probability of Default) and LGD (Loss Given Default).
3. Proposal of the collection management model within the credit risk management
Credit risk is the probability that the bank will not be in position to collect total receivables from the customers, i.e.
principal amount and all connected interest rates and fees. Having in mind the cause of the credit transaction there
1 Relevant regulation for banking risk management is determined by the Basel Rules. Also, important regulation for collection
management is local regulation of each country connected to the performing business of legal entities, process of liquidation or
bankruptcy of the company and laws connected to the collaterals (mortgages, pledge, etc.).
2 The example is based on the current law regulation in Serbia.
3 The examples are based on the average of the biggest bank on Serbian market. Source: National Bank of Serbia.
183

are 3 types of credit risk: 1) default risk (exists at the moment of loan approval), 2) premium credit risk (exists at the
moment of usage of the loan and has impact on problems in repayment of the loan) and 3) risk of the worthiness of
the credit rating (exists in the moment of the loan repayment). It is important to emphasize that the credit risk can be
monitored in the banking book (critical threat on the credit portfolio which has impact on the banks in liquidity) and
trading book of the bank.
The bank is successful if it fulfils the following criteria: 1) volume of new approved loans and increase of the
credit portfolio in accordance with the defined targets, 2) profitability of the bank, and 3) the level of nonperforming
loans. Therefore, we may conclude that the successful collection management is one of the most
important tasks for the bank.
Collection management model should cover the following areas: 1) aims of the model, 2) organization of the
model, 3) instruments of the model, 4) control and monitoring within the model (PL and NPL).
3.1 Aims of the collection management model
The basic aims of collection management of the bank’s credit portfolio are: 1) regular servicing of the credit
commitments of the customers, 2) minimizing the delay in servicing the credit commitments toward the bank, 3)
minimizing the number of NPL in credit portfolio.
The advanced aims of collection management of credit portfolio of the bank are: 1) synchronizing additional
costs on problem loans with collection management of receivables of problematical customers, in order to have the
lowest volatility of the costs on profit and loss account; 2) predicting the potential problems and future allocation of
the costs. Advanced collection management means that the costs of one year grow progressively, from 0 to the
defined value for particular year. Only through the constant cost it is possible to predict more precisely the total
result of the bank. Volatility of those costs is not desirable.
3.2 Organization of the collection management model
The organization of the model includes: 1) classification of the debtors in the model, 2) participants and their tasks
within the model.
There is a wide range of criteria for dividing the credit customers of the portfolio, such as: exposure, industry,
maturity, rating of the customer, diversification in accordance with the delay basket (in the practice: delay up to 30
days, delays between 30 and 60 days, delays between 60 and 90 days and delays above 90 days). Banks try through
credit policy to define aims which represents the optimal diversification of the portfolio, including own needs,
market parameters and regulation frame.
In the context of collection management there are two critical determinants of the portfolio quality and
diversification: 1) rating of the customer (represents the probability that the customer will not fulfil commitment on
due date. The rating of the customer consists of two parameters: quantitative and qualitative criteria. After the
defined rating of the customer, there are two important groups of customers within the portfolio: liquid and nonliquid
customers, i.e. PL and NPL) and 2) delay in fulfilling the commitments toward the bank (i.e. delay basket).
For final segmentation of the customer in the portfolio, expert opinion is relevant (based on individual assessment of
risk management division of the bank). The main subjects in creditor-debtor relation are creditors (banks) and
debtors (companies).
Banks are the most important players on the financial markets. Central banks through business banks execute
monetary and credit policy. In terms of collection management and assessment of credit risk within the bank the
following divisions are relevant: corporate division, risk management division (corporate credit risk management,
collateral management and collection management), credit committee, legal division and back office (credit
administration). It is important that they have good cooperation. All of those divisions and departments have precise
instructions for executing a loan application. Corporate division is important in daily cooperation with the customer,
but also at the time of approving and monitoring the loan. Good communication and partnership relation is essential
in collection management process. Risk management division is responsible for individual credit risk assessment in
the following segments: initial crediting, collection management and collateral management. Tasks of risk
184

management division are: 1) restructuring: business renewal or refinancing, monitoring of the problematic loons,
watch lists, stress renewal or refinancing, 2) liquidation: collection from the bankruptcy, collection from the
collateral, sale of business, sale of receivables, 3) reporting and analysis: delay reports, work-out reports,
provisioning management, help in budgeting process.
Debtor (company) represents a crucial part of the collection management. At the same time the company is the
heart of the financial restructuring. The reasons why the company can have the problems in repayment of the loan
are different, and it is very hard to identify them at the beginning. Some of the causes of the problem can be: 1)
management (35%), market – competition (30%), financials (20%) and other (15%). 4 It is evident that the
management is the most responsible for the new problematic situation. When entering the restructuring process, the
change of the management can be a very hard and painful issue for the shareholders of the company. If the
diagnostic of the new problems clearly showed it, it means that this unpopular measure should be implemented as
soon as possible.
Also, it should be mentioned that in Serbia, in many cases, the owners of the companies are at the same time the
managers of the company. In that situation, it is very difficult to talk with the owners and persuade them that they
are bad managers.
Banks are usually the biggest creditors of the company. It should be noted that the approval of the loan is the
main business of the bank, comparing to the other creditors of the company. So, when the company is faced with the
financial problems banks should be ready to take proactive approach in solving the new situation. It means that the
bank will: 1) start negotiation with other banks creditors of the same company (or group of connected companies)
and 2) take action toward the company.
3.3 Architecture of the collection management model
Every house stands on its own foundations. There is no successful model without clear instruction. Organization of
the process is the pillar and basis of further development and improvement of this process. Organization of the
model is established on bilateral basis between the creditors (bank) and debtors (companies). But also, it should be
established widely, including the market and current regulation. Important parts of organization are collection
management process for PL portfolio and NPL portfolio.
Collection management of PL portfolio consists of two areas: 1) processes and actions within the bank which
has final action toward the client, 2) processes and actions toward the client. Those processes are in the function to
present bank’s position and attitude toward the credit customer. The first process within the bank is fluctuation, but
the second process should be always consistent and linear (it should be standardized; defined by the procedures of
the bank adopted by the supervisory body).
Collection management of NPL portfolio is very hard to be standardized. In the practice, there are a huge
number of the credit files which have discrepancy and its solutions are not directly connected with the standards.
The possible ways for solving the problem are: 1) moratorium (subject of analysis and discussing the strategy), 2)
implementation of the adopted decision in terms of collection of receivables from the core business of the company
(restructuring, refinancing, restructuring of the business, etc.), 3) implementation of the decision that the collection
will be done from collection of collateral.
3.4 Instruments of the collection management model
The moment of the loan approval is crucial for making a final decision about the possible way of collecting the
receivable. At the time of analyzing and approving the loan, the bank considers two possibilities for credit
repayment: 1) source of repayment can be cash flow from regular (core) business, i.e. primary source of repayment
(CF1) and 2) source of repayment can be collateral of the loan i.e. secondary source of repayment (CF2).
4 This statistic represents a specificity of the Serbian market (based on the experience of risk management divisions of Serbian
banks). Source: Bibliography 7.
185

Core business is the main source of income and possibility for repayment of the loan. When granting the loan,
banks put the main focus on the core business, i.e. estimation whether the regular activity of the company will be
enough to cover the commitments. The credit criteria are defined by credit policy of the bank. Common parts of
those standards are: 1) general requirements (minimal qualitative criteria, additional qualitative preconditions,
minimal quantitative criteria, prohibit type of transactions), 2) financial requirements in accordance with the
segmentation of the customers (turnover – t/o is criteria for segmentation and the type of financing, for example
project financing has different criteria), such as: rating (which determine the ratios which follows; for better ranked
credit customers, lower ratios are required), ratio ST debt toward t/o (10%-40%), collateral coverage (min 20%,
ratio equity toward LT debt (60-100%), ratio collateral toward total exposure (20%-75%). Cash covered loans
(including bank’s guarantee issued by the first rank bank), overdrafts and loans with low risk are excluded from
those criteria. 5
The financial criteria should be revised in case of problematic situation, i.e. at the moment of difficulties in
repayment of the matured obligations. The most important tasks of collection management are to synchronize the
new situation with the customer’s business. This is the approach which makes the biggest benefit for the creditor and
debtor. At the moment of restructuring of the loan initially defined financial standards are decreasing, up to breakeven
margin. Under this level, business restructuring does not make sense. Also, it should be stressed that any
business restructuring is not good restructuring. Thus, it is necessary, before any restructuring, to answer the
question whether we as a creditor believe in this business? Before we answer this question, we should provide a
detailed and serious analysis.
The bank uses collateral, as secondary source of repayment of the loan, when the core business of the company
stopped. All instruments which bank use as secondary security can be classified as follows: mortgages, pledge,
corporate guarantees, assignment of receivables, and other collaterals. The crucial thing is to determine the proper
value at the time of approving the loan, and to reconsider the value of the collateral at the moment of collecting the
receivables.
3.5 Control and monitoring within the collection management model
PL and NPL clients are subject of control and monitoring within the collection management model. It means that it
should be recognizable when PL customer has problems in repayment and which proactive measures the bank will
take toward the customer in order to avoid delay. Also, the model identifies in NPL portfolio which problems may
appear and which analysis should be done in order to minimize the negative effect of new problems in NPL. 6
As we have already mentioned, the model consists of two analyses: quantitative analysis and qualitative
analysis. Quantitative analysis represents changes which already appeared in customer’s business and may affect the
regular repayment of the loan and core business of the customer. All early warning signs can be categorized as
follows (big changes in customer’s behaviour, market data, problems in daily business and signs of fraud).
Qualitative analysis represents a portfolio analysis on the basis of historical data. Historical data can be internal
(internal data base) and external (market data). Internal indicators are: days of delay, rating, industry and t/o through
the account within the bank. External indicators are: blockade of the account, financial data (t/o, gross profit margin,
EBIT, net profit, total assets, equity, short term loans, long term loans, receivables, payables, inventories).
On the basis of quantitative and qualitative criteria, all credit customers of the bank are divided in 3 zones in
accordance with the risk level: 1) red zone – the most risky customers of the portfolio, 2) yellow zone – the zone of
medium risk, 3) green zone – zone of the low risk. After that corporate credit risk manager sends information to the
responsible account manager which customers are potentially problematic in the future in order to organize the
meeting with such customers and prepare proper strategy for the customers i.e. collection of the receivables.
Practically it means preparing of review application. Generally, the risk management division monitors credit
portfolio on permanent basis, but if they assessed that it is necessary they can monitor the credit risk on particular
customer i.e. loan.
5 The conclusion on the basis of the experience of Serbian banks in the previous period. Source: Bibliography 7.
6 Erste bank a.d. Novi Sad implemented presented model in the second part of 2010.
186

4. Collection management impact on credit portfolio of the bank and its importance for credit risk
management as whole
Collection management has direct impact on the bank’s profitability through the cost of reservation. Cost of
reservation is the result of bank’s assessment that determined part of credit portfolio that will not be collected in the
future. It means that the bank will book this reservation in the future as loss in financial reports. So the estimation of
the level of reservation, especially for the NPL part of portfolio is crucial for the bank. On this way, risk
management division manages the total pull of reservations which considers as relevant to cover future loss of credit
portfolio.
Calculation of individual change of value represents the total effect of the executed collection from the core
business of the customer or by selling the collateral or by bankruptcy of the company in the following period, by
using net present value with effective interest rate as discounting factor. Estimated non-collected part of the
principal amount represents a particular change of the value. Total expected value of the collection has two sources:
core business of the company (cash flow 1 – CF1) and collaterals or bankruptcy of the company (cash flow 2 –
CF2).
Core business represents additional assessment of the creditworthiness of the customer. Bank on the basis of
current company’s situation and projections of company’s business makes decision in which period the company
will be able to serve the commitments toward the bank and how long it will last. It is evident that this assessment is
specific and more restrictive comparing to the initial approval of the loan. The new analysis has following crucial
questions: 1) Is the management capable to manage the company? 2) Is the company blocked by other creditors in
the amount which is hardly achievable for the debtor? 3) Does the company have the market and in which way is the
position changed? 4) Are the financial projections minimally on break-even point? 5) What are the structure and the
value of the collateral? After analysing those questions, the bank is in the position to estimate the perspective of the
business and possibilities of credit repayment. New restructuring of the credit commitments can be executed through
official or silent restructuring. Both ways of restructuring are monitored on monthly basis. If the customer does not
have delay in repayment, we can say that the restructuring is successful.
Collaterals or bankruptcy represents assessment of collection of receivables on the basis of collaterals which
bank has on disposal (mortgages, pledges, and contracts on goods or receivables). Estimation of cash flow is based
on the bank’s expectation when and for how much the collateral will be sold in the future. The bank uses a
conservative approach for collection of receivables.
The bank uses the model which helps in adequate estimation of two segments of collection from collateral:
amount and term. Further sub-criteria of the model allow better estimation of the value of the collateral such as:
quality, the level of regional development, the level of market development and collateral conditions. We can say
that the estimation of the collection of receivable from bankruptcy and core business, through the restructuring of
the loan or through the restructuring of whole business is subject of individual assessment and it is very hard for
quantification.
At the end, sum of present value of expected collection from core business (CF1) and present value of expected
collection from the collateral or bankruptcy (CF2) represents the value which bank expects to collect on the basis of
maturity receivables. Different from the value of total receivables are reservations i.e. individual change of the value
of particular credit.
5. Non-performing loans in Serbia
The Serbian banking sector recorded credit growth in the previous period. At the end of the third quartile of 2010
credit portfolio achieved level of EUR 14.8 billion (table 1 and figure 1).
Years 2001 2005 2007 2008 2009 2010
Credits 6,326 5,082 9,602 11,598 13,331 14,854
Table 1: Volume of credit activity of Serbian banking sector (in mil EUR)
187

Figure 1: Volume of credit activity of Serbian banking sector (in mil EUR)
The majority of portfolio represents corporate segment (56%), then retail segment (31%), public sector (11%),
and other (2%). The most problematic issue is growth of the exposure toward the public sector in the previous
period (table 2 and figure 2), because it means increase of consumption and problem of repayment for young
generation.
Type of the customer 2007 2008 2009 2010
Corporate 5,520 7,135 7,826 8,298
Retail 3,818 4,112 4,119 4,587
Public sector 175 196 1,230 1,639
Others 89 155 156 330
Total 9,602 12,598 13,331 14,854
Table 2: Structure of Serbian banking credit portfolio in the period 2007 -2010 (in mil EUR)
Figure 2: Structure of Serbian banking credit portfolio in the period of 2007-2010 (in mil EUR)
Non-performing loans (NPL) of the Serbian banking sector represents 17.8% of the total loan portfolio i.e. EUR
2.6 billion (table 3 and figure 3). If we analyse the structure of NPL we can notice that NPL corporate segment
represents 72% of NPL (it is in accordance with the total portfolio structure), NPL retail segment represents 14%,
and NPL other 13%.
IX 2009 XII 2009 III 2010 VI 2010 IX 10
NPL in total
credit portfolio
Structure of NPL
17.7 15.7 16.5 17.5 17.8
Corporate 78 76 77 78 72
Retail 15 16 15 15 14
Others 7 8 8 8 13
Table 3: NPL of Serbian banking sector including the structure (in %)
188

Figure 3: NPL of Serbian banking sector including the structure (in %)
Having in mind the importance of corporate segment for the growth of one economy we will continue to
analyse the structure of NPL corporate segment (figure 4). Processing industry represents 36% of NPL corporate
segment, 31% trade, 10% construction, 9% real estate, 8% agriculture, 6% hotel and restaurants, 0% education and
0% electricity.
Figure 4: Structure of NPL corporate segment of Serbian banking sector (%)
At the end of 2008 Austrian banks in Serbia established a restrictive credit policy (known as traffic light)
toward corporate segment: 1) red light (decrease of exposure toward defined segment, without new approvals):
construction, production and sale of cars, energetic, textile industry, processing industry and tourism; 2) yellow light
(increased watch): furniture industry and exporters of fruits and vegetables; 3) green light (standard cooperation):
telecommunication, IT and media, agriculture, pharmaceuticals, processing and sale of food and beverages,
transport, services, municipalities and public sector. If we compare this with the current NPL we can notice that this
restrictive credit policy was in place.
Also, it is important to analyse the level of reservation in the financial reports of the banks. Unfortunately, their
level during the crisis had impact on lower profit than it was projected (table 4 and figure 5). Thus the proposed
collection management model (which Erste bank Serbia started to use in the last quartile of 2010) will have positive
effect in the following period. We can consider the results of the implemented model at the end of the third quartile
of this year.
189

2008 2009 2010
Profit 26.08 36.11 24.28
Net income 34.74 20.03 20.70
Cost of reservation 19.56 65 28.41
Table 4: Relevant financial indicators of Serbian banking sector (in bill RSD)
6. Summary
Figure 5: Relevant financial indicators of Serbian banking sector (in mil EUR)
Collection management has impact on the profitability of the bank and equity, through the price of the product, cost
of reservation and deducted item of the equity. The economic crisis, which had impact on the financial and real
sector also stressed the importance of the collection management and necessary changes of collection management
model. The collection management of credit receivables is possible in two ways: 1) restructuring (credit or business)
and 2) selling the collaterals. These two ways should be considered and implemented during the entire credit life
(not only after 90 days of delay in fulfiling credit obligations). Overall and proactive approch is necessary for the
sucessful collection management. The proposed collection management model is aimed to predict problems in
repayment of the loans which are not presently problematic. The idea is to stop further growth of NPL and to
improve the quality of the existing credit portfolio. Arhitecture of the model covers all credit customers – PL and
NPL. The proceses in the model are divided in following way (from the perspective of the creditor – bank): 1)
process and actions within the bank which have impact on the approach toward the customer, 2) processes and
actions toward the customers. Those processes send proper message to the customer regarding the strategy toward it.
The work showed that there is room for further improvement of the collection management. Also, it is evident
that this process is complex and important part of the credit risk management process in the bank. The model helps
the bank to establish adequate collection management process and relation between the bank and debtor (company).
Collection management has a big impact on the level of reservation – general and specific provisioning. The
adequate assesment of collection and successful collection of receivables are the guarantee for reliable and precise
quality of own reservations, which have impact on the profit and loss account of the bank and the equity. The
assesment of specific provisioning is especially important, having in mind their impact on the total pool of
reservation. At the end it is important to emphasize that the collection management represents the mirror of the
overall credit policy of the bank. It is not the product of well-determined input parametars of risk and credit
standards, but alo it is the mirror of the internal functioning of different divisions involved in the credit process.
Therefore, we can say that the successful collection management is the equivalent to the total credit business – from
the moment of opening the credit file until closing of the credit file, and represents the value of the bank and its
credit portfolio on the market.
7. Acknowledgments
This Research Paper was the part of the project “Advancing Serbia’s Competitiveness in the Process of EU
Accession”, no. 47028, in the period 2001-2015, financed by Serbian Ministry of Science and Technological
Development.
190

8. References
Altman, E. I., Hotchkiss, E. (2005). Corporate Financial Distress and Bankruptcy: Predict and Avoid Bankruptcy,
Analyze and Invest in Distressed Debt, 3rd Edition, Wiley.
Barjaktarovic, L. (2009). Risk management, University of Singidunum, Belgrade.
Basel Committee on Banking Supervision (2001). The Internal Ratings- Based Approach – Supporting Document to
the new Basel Capital Accord, (available at www.bis.org).
Basel Committee on Banking Supervision, Internal Convergence of Capital Measurement and Capital Standards
(2006), A Revised Framework Comprehensive Version, Bank for International Settlements, Basel, (available
at http:/www.bis.org).
Caouette, J., Altman, E., and Narayanan, P. (1998). Managing Credit Risk – The Next Great Financial Challenge,
Wiley.
Committee of European Banking Supervision (CEBS): http://www.c-ebs.org / [available data on March 2011]
Crosby, P. and Bohn, J. (2009). Modeling Default Risk, (Moody's KMV Document), Revised 2001, (available at
www.kmv.com).
Institutional Investor Journals (2010). Turnaround Management II: a Guide to Corporate Restructuring, Euro
money International Investor.
Gilson, S. C. (Author), Altman, E. I. (Foreword) (2010). Creating Value through Corporate Restructuring: Case
Studies in Bankruptcies, Buyouts, and Breakups, Wiley.
Grieser, S. and Wulfken, J. (2009). Performing and Non-Performing Loans across the World: a Practical Guide,
Euro money Institutional Investor.
European Commission: http://ec.europa.eu/internal_market/bank/regcapital/index_en.htm / [available data on
March 2011]
Larkin, B. (2008). Restructuring and Workouts: Strategies for Maximising Value, Globe Law and Business.
National bank of Serbia: http:/www.nbs.rs/ [available data on March 2011]
Rating Manuel (2009), Erste Group.
Vasilski, A. (2010). Collection management from the borrower’s perspective during the world’s economic crisis,
The Faculty of Economics, Finance and Administration, Belgrade.
Work-out procedures – Holding Level (2010), Erste Group.
191

HOW THE CROSS BORDER MERGERS AND ACQUISITIONS OF THE GREEK BANKS IN THE
BALKAN AREA IMPROVE THE COURSE OF PROFITABILITY, EFFICIENCY AND LIQUIDITY
INDEXES OF THEM?
Kyriazopoulos George, Applications Professor at Finance Department, TEI of Western Macedonia
Kozani GR-50100, GREECE, Email: kyriazog@teikoz.gr, kyriazopoulosg@yahoo.com
Dr. Petropoulos Dimitrios, Assistant Professor of Agricultural Economy at TEI Kalamatas
Kalamata, GREECE, Email: d.petro@teikal.gr
Abstract. In this paper we try to give an explanation about how the cross border mergers and acquisitions of Greek bank my affect the
profitability the liquidity and the efficiency of them. In the 1st chapter of the paper we describe the Balkan macroeconomic
environment as the favorable conditions and the opportunities that were created, so that the Greek and foreign banks to be pushed to
expand in the particular environment. In the second chapter of the present work we present with details the expansions of the Greek
banks and other European banks in each Balkan country separately. In the present work a list is made of all the mergers and
acquisitions made by the Greek banks in the Balkan countries during the decade 2000-2009. At the end we talk about the positive
influence of the cross border mergers and acquisitions on deposits and loans and therefore on the ROAA and ROAE indexes of the
Greek Banks are shown. The analysis of profitability, liquidity and efficiency with financial indexes takes place for the time period of
2003-2007 and is not expanding on the years 2008 and 2009 (period of the appearance of the financial crisis). The above distinction is
made in order to show greater homogeneity and in order to obtain the estimation of the magnitudes under conditions of economic
stability and progress. The behavior and the values of the financial indexes during the phase of the financial crisis it could become the
object of further research in another paper.
Keywords: bank, banking sector, productivity, efficiency, liquidity, cross-border mergers and acquisitions.
JEL Classification Codes: G2-Financial Institutions and Services
G21 – banks; Other Depository Institutions; Micro Finance, Institutions; Mortgages
1. Introduction
In the past twenty years, cross-border mergers and acquisitions have had an ever increasing role in the process of
bank internationalization. The Greek Banks with the accession of the Greek Economy to the Economic and
Monetary Union, but mainly with the introduction of the Euro into Greece, began to seek new sources of revenues in
new markets, where costs were lower and the interest rate of the loans higher. The Greek banks since the year 2000
adopted aggressive policies of expansion with cross border merges and acquisitions mainly of local banks in the
Balkan countries increasing the magnitudes of their balance sheets, especially with respect to the consumer and
mortgage loans, in an effort to increase by any means the total profitability liquidity and efficiency of their groups.
The aggressive policy of cross border and acquisitions in the Balkan area of the Greek banks was influenced
substantially by the introduction of euro into the Greek economy, which acquired a strong currency and at the same
time a reduction of the foreign exchange risk. This way, actually it was left for the banks only the management of
the credit risk, since the political risk of the Balkan countries had been eliminated with the restoration of democracy
in these countries. The banks offer differentiated products to their clients, which can be adapted to the particular
conditions of life, preferences, risk characteristics and the needs of the population and the businesses of the Balkan
countries. The wave of cross border mergers and acquisitions in the Balkans caused some fears, mainly with regard
the stability of the system, the supply of adequate amount of capital through the granting of loans to small and
medium size firms and to consumers, just as the loan pricing practices towards them. However, with «Basil II»
there are already complete ways of control, while the granting and pricing of loans, shows not to be finally affected
from the buyouts and mergers.
The trend for cross border mergers and acquisitions according to the banking theory is attributed to different
reasons, as for example the optimization of cost, the increase of power in the market, the capability of greater
dispersion of risk, and reasons that have to do with corporate governance and the “creation of an empire” by the
chief executive officers of each organization.
192

2. The Macroeconomic Environment in the Balkan Countries and the opportunities for the Greek
Banks
After the fall of the communist regimes in 1989, much of the banking sector in the SEE countries was still
underdeveloped and centrally planned, while money had only a limited role as a medium of exchange. In general the
banking sector in SEE countries has developed significantly in the past 10 years (1998-2007). However, there are
still many challenges ahead for the banking sector in the SEE countries, while it is less developed relative to the
Central Eastern European transition countries. 1
Turkey starts to have EBRD Index from the year 2008. In the table1 exhibits the improvement of the South
Eastern European (SEE) countries banking sector over the past 6 years, according to the EBRD Index of Banking
System Reform:
Table 1: SEE Countries' Banking System Reform Index 1998-2003
Country Year EBRD Index of Banking System
Reform from 1 to 4
Albania 1998-2003 (2004-2007) 2,0-2,3 (2,67)
Bosnia/Herzegovina 1998-2003 (2004-2007) 2,3 (2,67)
Bulgaria 1998-2003 (2004-2007) 2,7-3,3 (3,33-3,67)
Serbia/Montenegro 1998-2003 (2004-2007) 1,0-2,3 (2,3-2,67)
Fyrom 1998-2003 (2004-2007) 3,0 (2,67)
Romania 1998-2003 (2004-2007) 2,3-2,7 (3,0-3,33)
Source: European banking for Reconstruction and Development (EBRD)
EBRDIndex1-4
According to the EBRD index of the banking system reform, all SEE countries are classified around 2,3.
3,5 4
2,5 3
1,5 2
0,5
0
1
Albania
EBRD INDEX OF BANKING SYSTEM YEAR 2003 and 2007
Bosnia/Herzegovina
Bulgaria
Serbia/Montenegro
Countries
As you can see in the above diagram all countries have improved their EBRD Index except for FYROM that has a
little decrease of 0,3 from the year 2003 to 2007.
According to the EBRD index of the banking system reform, all SEE countries are classified around 3. This
classification means: “there has been progress in establishment of bank solvency and of a framework for prudential
supervision and regulation, while there is significant lending to private enterprises and significant presence of
private banks”. We observe that all the countries have an improvement in the time period 2004-2007 from the time
period 1998-2003.
The macroeconomic progress in the region of the Balkans is strong and economic activity increases steadily (2006:
4,5%-5% on the average) mainly due to the strong domestic demand. In most of the Balkan countries the increased
borrowing of the households has accelerated the rate at which the imports of consumer goods is rising, thus
worsening significantly the trade balance. Something equivalent started in Greece towards the end of the decade of
the 90’s and lasted up to the middle of 2007 the fact that contributed to the present financial crisis. 2
The financial sector in the Balkans is generally healthy. The banks become active in a generally favorable
entrepreneurial environment, with total macroeconomic activity on the rise and under conditions and perspectives of
increasing profitability. The credit expansion, especially towards the households, is strong in all the countries of this
1 Stubos G., Tsikripis I., Banking Sector Developments in South-eastern Europe May 2004
2 Kyriazopoulos G., Petropoulos D., “Does the cross border merger and acquisitions of the Greek Bank in the Balkan Area affect on the course of
profitability efficiency and liquidity index of them?” May 2011
193
Fyrom
Romania

egion, with rates of increase higher than the European average. The quality of the loans made by the banks is
satisfactory with acceptable indexes of loans that are not being paid back to total loans. However, the possibility of
future deterioration of these indexes is visible, as the loan load assumed by the households is increasing at a fast
rate. The percentage of loans in foreign currency (mainly USA dollars and Euros) to total loans is quite high and in
some instances exceeds the usual limits of safety (e.g. in Albania for 2005: 73%).
2.1 The Expansion of Greek Banks in the Balkans 34
The expansion of the large Greek banking groups in the Balkans began at the beginning of the decade of 1990 and
increased in the time period of 1998-2007, and it came about as a result of the gradual convergence of the Greek
banking system and the Geek economy to the standards of the more developed European countries. The showing
signs of coming to maturity Greek banking system created the need of expansion outside of the borders the Balkans,
due to the cultural and geographical proximity were creating conditions of competitive advantages, and thus defined
the new field of Greek bank activity. The gradual adaption of the Balkan countries to free market conditions, the
continuation of structural changes, the massive programs of privatizations and the perspectives of economic
development constituted essential factors for the attraction of the Greek banks and the banks of other European
countries to the region. Initially the expansion of the Greek banks in the Balkans aimed mainly at serving the Greek
businesses that became active in the region seeking new opportunities before the banks wake up. The Greek banks
being familiar with the activity and the credit risk of the Greek enterprises accompanied them in their expansion in
the Balkans. Later on, the Greek banks enlarged gradually their activities in the Balkan area taking advantage of the
rising market in the sector of retail banking and by capitalizing on investment opportunities in sectors such as the
buying of land and the development of commercial and private home zones.
In the time period 2004-2007, the presence of large Greek banks in the Balkans does not constitute a
circumstantial financial exploitation, but is a long term plan of strategic investment, which expresses itself either
through autonomous development or as development through buyouts of local banks. In 2005 the five largest Greek
banking groups had created a network of 958 branches in total, they employed more than 15.000 people, they had
acquired a share of 17% of the granting of loans market and their profits exceeded the amount of €138 million. Also
in the year 2007 from this region came the 20% of their total profits. In some of the Balkan countries the Greek
banks have at the present time a leading position. The Greek banks, through their expansion in the Balkans, have as
a goal to improve their profitability as the retail banking in Greece gradually approaches the levels of saturation of
most of the countries in the Eurozone. Indicatively is reported that in 2005 the granting of loans to the households
and businesses as a percentage of the Gross National Product (GNP) in the Balkan countries was fluctuating from
14,5% (Albania) to 44% (Bulgaria), as opposed to 76% in Greece and 104%in the Eurozone.
3. Cross border Mergers and Acquisitions of Greek Banks in the Balkan Countries.
At the time period of 2003-2007 the Balkans offer significant margins of business development to the banks of
European countries, because of the high rate of credit expansion and of the still low level of development of the
financial markets. However, for as long as the upgrading of the systems and products being offered by the European
banks expands, the Greek banks will have to adapt their strategies in an environment of increasing competition. The
expansion of the Greek banks to the developing neighboring markets of the Balkans and the Mediterranean
remained the central axis of their strategy in the time period 2003-2007 just before the crisis began. However, at the
present time more and more the valuations of the financial institutions in the region are getting larger in such a
degree, that in some cases the total price of the buyout is prohibitive for the Greek banks.
3.1. Greek Banking activity in the Balkan Area 5
At the time period 2003-2007 the Greek banking activity in the regions of the Balkan countries refers to the
countries of Serbia, Fyrom, Romania, Bulgaria and Turkey. In these countries the Greek banks that expanded are
3 The source for numbers and percentages is IMF and ECB
4 Kyriazopoulos G., Petropoulos D., “Does the cross border merger and acquisitions of the Greek Bank in the Balkan Area affect on the course of
profitability efficiency and liquidity index of them?” May 2011
5 Kyriazopoulos G., Petropoulos D “Cross Border Mergers and Acquisitions in the Balkan Countries after the introduction of Euro in Greece”
Esdo Kavala June 2010
194

among the largest banks of Greece, regardless if their origin and the majority of their shareholders are from outside
Greece. These banks are the National Bank of Greece, Piraeus Bank, Alpha Bank, Efg Eurobank, ATE Bank,
Emporiki Bank, and Egnatia Bank, now after a big merger that took place in Greece it’s called MarfinEgnatia Bank.
The country with the greater penetration in the Albanian banking market is Greece. The five big Greek banks have a
strong presence and they concentrate 35,7% of the granting of loans to customers, the USA follows with 16,7%
through the American Bank of Albania and Austria with 13,9%, through the Raiffeisen Bank.
The year 2007 was a milestone for Bulgaria since at the beginning of that year she became a member of the
European Union. In Bulgaria, banking activities coming from the side of Greece and other foreign countries did not
present considerable interest. The total amount of banking mergers and acquisitions in Bulgaria from 2004 to 2006
was $265,97 for the Greek banks. More specifically the Greek banks that played a leading part in the cross border
mergers and acquisitions were the EfgEurobank and the Piraeus Bank, which bought out the DZI Bank and the
Eurobank (Bulgarian) respectively each one. The percentage of participation and the amounts of the buyouts are
shown in table 2 below.
Table 2 The Greek banks that merged and bought out banks of Bulgaria
Greek Bank Buyer Bulgarian Bank Target Participation (%) Buyout Amount $ millions
EfgEurobank DZI Bank 74,26% 200,20
Piraeus Eurobank 99,70% 65,77
Total 265,97
Source: Mergerstat Intellinet Bloomberg Athens Stock Exchange
Moreover, the merger of the premises of the Piraeus bank and its subsidiary Piraeus Eurobank AD during the first
half of 2006 had as a result the creation of a larger bank, that holds the 8 th position in the classification of banks,
outdistancing the Economic Investment Bank, SG-Express Bank and DZI Bank, which were holding the 8 th , 9 th and
10 th position in the classification of banks respectively in Bulgaria in the year 2005.
Predominant position in the banking market of FYROM holds the National Bank of Greece after she bought out
the 68,4% of the second largest bank of the country (Stopanska Banka Skopje) in April 2000 for €60 millions.
Through the Stopanska Banka Skopje the National Bank of Greece has a 23,8% share of the market based on assets
and a 27% market share based on the granting of loans. The Alpha Bank entered the banking market of FYROM in
1999 after buying out the Kredinta Banka, which later was renamed Alpha Bank Skopje. In 2005, the Alpha Bank
Skopje had a market share based on the granting of loans 2,6%. The EFG-Eurobank-Ergasias, the Emporiki Bank
and the Piraeus Bank did not have a presence in FYROM until the year 2007.
Romania became a full member of the European Union along with Bulgaria in January of 2007. The banking
sector of Romania consisted of 39 banks of which 30 are foreign and 1 is state owned. The concentration of the
banking sector is not very high as the 5 of the largest banks control only 60% of the sector’s total capital. In
Romania at the beginning the banking activities on the part of Greece did not exhibit enough interest compared with
the interest shown by other countries. The total amount of mergers and acquisitions in Romania for the time period
2004 -2006 was $81,07 millions for the Greek banks. In 2005, the Greek banks were concentrating in total 13,6% of
the granting of loans to customers. The percentages of participation and the amount of acquisitions from the Greek
banks are shown in the following table 3.
Table 3: The Greek banks that were merged and bought out banks of Romania
Greek Bank Buyer Romania Bank Target Participation (%) Amount of Buyout ($ millions)
ΑΤΕ Mindbank 57,13% 41,07
40,00
NBG Banka Romaneasca 81,65%
Total 81,07
Source: Mergerstat. Intellinet. Bloomberg. Athens Stock Exchange
In Serbia the banking activity on the part of Greece and on the part of the other countries exhibited sufficient
interest. The total amount of mergers and acquisitions in Serbia during the period 2004 -2006 reached the amount of
$973,68 millions for the Greek banks. More specifically the Greek banks that played a leading part in the mergers
and acquisitions in Serbia are the Alpha Bank, the EfgEurobank, the ATE Bank, the National Bank of Greece and
the Piraeus Bank which bought out the banks Jubanka, National Savings Bank, AIK Banka, Vojvodjanska Banka
195

and Atlas Banka respectively each one. The percentage of participation and the amount of buyout are shown in the
following table 4.
Table 4: The Greek banks that merged and acquired banks of Serbia
Greek Bank Buyer Serbian Bank Target Participation (%) Buyout Amount ($ Millions)
Alpha Bank Jubanka 100,00% 206,69
EfgEurobank National Savings Bank 90,20% 93,28
ΑΤΕ Bank ΑΙΚ Banka 20,00% 101,78
NBG Vojvodjanska Banka 100,00% 489,57
Piraeus Bank Atlas Banka 88,23% 34,45
MarfinLaiki Bank Centrobanka 76,00% 47,91
Total 973,67
Source: Mergerstat. Intellinet. Bloomberg. Athens Stock Exchange
The countries with the greatest penetration in the Serbian market are Italy (2005: 18% share on assets), Austria and
Greece (2005: 17% share on assets each one). In 2005, the large banks of Greece exhibited limited capability to
penetrate the Serbian market. At the present however the total presence of the large Greek banks has been
strengthened after the buyout of the 6 th largest bank of the country by the NBG. In 2005, the 5 large Greek banks
were concentrating a market share based on the granting of loans 12,86%, from 8,66% before the buyout of
Vojvodjanska Banka.
In Turkey the banking activities on the part of Greece did not exhibit adequate interest as opposed to other
countries. The total amount of banking mergers and buyouts in Turkey during the period 2004-2006 reached the
amount of $2.770,00 and it had to do with the buyout of a 46,00% shares in the Finansbank by the National Bank of
Greece at the above mentioned amount. The percentage of participation and the amount of buyout are shown in the
following table 5.
Greek Bank Buyer Turkish Bank Target Participation (%) Amount of Buyout ($ Millions)
NBG Finansbank 46,00% 2.770,00
Alpha Altemnatibank 47,00% 246,30
EfgEurobank Tekfenbank 70,00% 182,00
Total 3.198,30
Source: Mergerstat. Intellinet. Bloomberg. Athens Stock Exchange
Indicative of the importance the Greek banks attribute to their activities abroad is the magnitude of investment in the
region of Southeastern Europe. Up to date they have invested more than €6 billion and they operate 1.900 premises
(that represents 1/3 of the total number of premises) 6
Thus the «domestic banks attain by far the first position with respect the number of deals that took place in the
region» 7 , as in a total of 40 across the border deals that were made in the last 4 years in Turkey, Bulgaria, Serbia,
Albania, Romania and Fyrom, the 13 were made by Greek financial institutions and mainly by the NBG, the
EfgEurobank, the Piraeus Bank and the Alpha Bank. Already in a four year time period (2002-2006) of dynamic
development the Greek banks have won significant shares in the markets of the countries where they have engaged,
while they have as a target in the next 2-3 years these countries «to produce» 20-30 %of their revenues and about
10-15% of their profitability. However, it should be noted that the Greek banks were very generous in the evaluation
of the banks –targets that they chose to buyout in the past not so much with concern the absolute magnitudes in
Euros that they paid out as with mainly the indexes of evaluation. This was so either because they were late to
decide, which resulted to increased evaluations, or because the number of acquisitions targets had decreased, which
6 Bank of Greece and Union of Greek Banks «The International Credit System: Challenges and Perspectives», Scientific Marketing, volume of September 2007
7 «With advantage the Greek banks in the Balkans», Georgas V., Eleftherotypia, April 6, 2009
196

esulted to a higher price completion of the deals. One of the most expensive buyouts that took place, as it is
mentioned in the article, was the one of the Bulgarian bank DZI in September of 2006.Eurobank paid $200,2
millions for 74% of the bank, evaluating 6,8 times its accounting value, when other buyouts in the country by
foreign banks were completed at 2,5 times their equity value. Equally expensive was the price paid by EfgEurobank
for the National Savings Bank of Serbia, where it paid 6 times its accounting value, while among the most expensive
were the two large buyouts made by the National Bank of Greece. The first one of the Vojvodjanska Banka for $490
million, an amount equivalent to 5,3 times its accounting value and 150 times of the before taxes earnings and the
second one of the Turkish Finansbank for $2,77 billions an amount 4 times its accounting value.
The Greek banks completed 16 buyouts of banking institutions in the Balkans region during the period 2004-
2006 as it is shown in the following table 6.
Table 6: Mergers and Acquisitions between 2002-2006 in the SEE
Greek Bank Buying out Foreign Bank Country Buyout Percentage (%)
NBG Vojvodjanska Banka Serbia 100,00
Finansbank Turkey 46,00
Banca Romaneasca Romania 98,88
United Bulgarian Bank Bulgaria 99,91
EfgEurobank Dominet Bank Poland 100,00
DZI Bank and Postbanka Bulgaria 74,26 and 100
Universal Bank Ukraine 99,34
Tekfenbank Turkey 70,00
National Savings Bank Serbia 100,00
Bancpost Romania 78,00
Alpha Bank Alternatifbank Turkey 47,00
Jubanka Serbia 99,99
Piraeus Bank Atlas Banka Serbia 100,00
Eurobank Bulgaria 99,85
Egyptian Commercial Bank Egypt 93,47
ΑΤΕ Bank AIK Banka Serbia 20,00
Midbank Romania 74,00
Source: Deloitte
Table 7: Results of the indexes ROAA and ROAE from the cross border mergers and acquisitions in 2006
BANK Equity / Assets (%)
31/12/2006
Loans / Deposits
(%) 31/12/2006
Return on Average
Assets (after tax)
(%) ROAA
31/12/2006
Return on Average
Equity (after tax)
(%) ROAE
31/12/2006
1 National 8,7 77,6 1,6 22,3
2 Eurobank 5,3 140,2 1,4 24,7
3 Alpha 4,7 132,8 1,4 28,3
4 Piraeus 5,0 128,2 1,7 32,4
5 Emporiki 5,1 107,2 0,6 11,3
6 ATE 5,7 67,3 0,7 12,8
7 Postal Savings 6,5 43,0 1,0 14,2
8 Marfin 12,7 87,4 3,0 16,3
9 Geniki 4,4 112,5 -1,3 -25,5
10 Egnatia 7,2 96,2 0,5 6,9
11 Attica 5,3 91,3 0,1 1,5
12 Aspis 5,5 101,5 0,5 8,8
13 Proton 24,6 83,8 3,9 12,3
14 Bank of Cyprus 6,3 68,9 1,3 21,2
15 Laiki Bank 5,5 69,6 1,1 20,6
Source: Calculate from Bank Financial Statements
In the below diagram we can see the course of ROAA and ROAE of the Greek Banks in the year 2006. From the top
5 Greek Banks as we observe Piraeus Bank has the biggest index of ROAE so it has the biggest benefit form its
cross border merges and acquisitions in the Balkans and icresed the wealth of its shareholders more than the others.
197

ROAAandROAE
40
30
20
10
0
-10
-20
-30
National
Eurobank
Alpha
The course of Greek Bank ROAA and ROAE in the Year 2006
Piraeus
Emporiki
ROAA (%) 2006 ROAE (%) 2006
ATE
Postal Savings
Marfin
Banks
The elements that we needed to draw the above diagram are taken from the table 7.
4. The Profitability, efficiency and liquidity of the Greek Banks [8] in the years 2003-2007
Geniki
Egnatia
Attica
Aspis
Proton
Bank of Cyprus
MarfinLaiki Bank
The profitability of the SEE countries banking sector, has improved significantly over the last six years. This was a
result of the general reform of the banking system and the high intermediation spread in these countries. In the
future, the financial institutions in the SEE countries should find different sources of profitability, since the
intermediation spread is expected to fall as the economies stabilize, the interest rates fall and the competition among
banks increase. The financial institutions should seek for these new sources of profitability in the retail banking and
the asset management in the Balkan clients, while they should try to increase their market share. On the other hand
they should control their operating expenses and expand their activities quite careful, in order to minimize their
losses from bad loans. The table 8 below presents two indexes of the SEE countries’ banks profitability, the Return
on Assets and the Return on Equity. [9]
Table 8: SEE Countries' Bank Profitability
Country Year ROA ROE
Albania 1998 and 2002 -1,8% and 1,2% -82,3% and 19,1%
Bosnia/Herzegovina 1998 and 2002 n/a and n/a n/a and n/a
Bulgaria 1998 and 2002 1,7% and 2,1% 15,8% and 16,2%
Serbia/Montenegro 1998 and 2002 n/a and n/a n/a and n/a
Fyrom 1998 and 2002 2,0% and 1,5% 8,2% and 6,9%
Romania 1998 and 2002 0,1% and 2,6% 1,0% and 18,3%
Greece 2003 0,9% 12,8%
EU Large Banks 2003S 0,4% 11,4%
Source: National Central Banks, Bank of Greece, ECB
The analysis of the financial statements of a business includes besides the selection of the appropriate index and
the comparison, without which the resulting conclusions do not have any meaning and most probably they do not
lead to the correct explanation. The comparison makes sense when it is done in relation to time and in relation to the
similar businesses or the sector. This double comparison gives the capability of a more correct explanation of the
indexes and consequently of the business condition (Papoulias, 2000). In order to achieve this we will use the
following indexes [10] :
4.1. Profitability – Efficiency Indexes in Greek Banking Sector
a. Return on Assets (ROA): ROA = Net Profits / Total Assets
This index reflects the administrations capability to use efficiently the financial resources (assets) that has at its
disposition, in order to create profits.
[8] Petropoulos D., Kyriazopoulos G., “Profitability, efficiency and liquidity of the co-operative banks in Greece” ICOAE Athens August 2010
[9] Stubos G., Tsikripis I., Banking Sector Developments in South-eastern Europe May 2004
[10] The above indexes are recorded in the bibliography as the most appropriate indexes for measuring economic magnitudes and characteristics
(profitability, return, liquidity) of the banks.
198

Return on Investment (ROE): ROE = Net Profits / Total Equity [11]
This index reflects the efficiency with which the bank uses the capital of its owners, as it shows the size of profits
that were created by the capital that was invested by the shareholders (owners) of the bank- enterprise.
4.2. Liquidity Indexes in Greek Banking Sector
a Loans / Deposits This index shows the banks’ needs in relation to loans and deposits.
b Total Assets / Total Loans This index shows the proportion of loans that the bank retains. A high value means low
efficiency and low risk.
The comparative analysis of the above indexes to the banking sector as a whole for Greece, in essence includes
the comparative analysis with the commercial banks that operate in Greece. The participation of the co-operative
banks to the totals of the economic magnitudes of the banking sector in Greece is very limited. Consequently, the
corresponding indexes referring to the whole banking sector, essentially are the indexes of the commercial banks.
4.3. Results in Greek Banking Sector
Here are the results of the Greek Banking System for profitability and efficiency.
4.3.1. Profitability – Efficiency in Greek Banking Sector
a. ROA = Net Profits / Total Assets
Analyzing the specific index we could:
� Compare the efficiency among the co-operative banks.
� Observe the efficiency through time.
� Compare efficiency of co-operative banks with the efficiency of the banking sector as a whole.
� Investigate the reasons of the changes through time.
From the analysis of Table 9 we find out, that the efficiency increases 63,7%. Moreover the average of 5 year period
(2003-2007) is 0,90
Table 9 Return on Total Assets (ROA) – Index
Bank 2003 2004 2005 2006 2007
Percentage
change
Average of five year
period of time
Total of banks 0,69 0,65 0,96 1,07 1,13 63,77 0,90
Source: Union of co-operative banks of Greece (ESTE) / Bank of Greece / self processing
ROARates
1,2
1
0,8
0,6
0,4
0,2
0
ROA
2003 2004 2005 2006 2007 Average
Year
b ROE = Net Profits / Total Equity
Analyzing the specific index of efficiency of the shareholders total equity, we can find out if the purpose of
achieving a satisfactory result has succeeded. From the index in the table 10 we suppose that the banks achieved
their purpose, because we can see that the index has upper course. Finally, we observe that the change of the specific
index for the total of the banking sector has an increase of 74,77%. As reasons for the above behavior of the
efficiency of total shareholder equity index could be mentioned the following:
� The increase of the total banks premises, it was followed by improvement, except year 2004, of the way
they are managed.
� Good management of total banks capital.
� High productivity of total banks.
[11] Vasiliou D., Financial Management Patra 1999.
199

Table 10 Return of Total Equity (ROE) – Index
Bank 2003 2004 2005 2006 2007
Percentage change
2007-2003
Average of five year
time period
Total of banks 8,52 7,70 11,80 13,47 14,89 74,77 11,28
Source: Union of co-operative banks of Greece (ESTE) / Bank of Greece / self processing
ROERates
16
14
12
10
8
6
4
2
0
4.3.2. Liquidity Results in Greek Banking Sector
ROE
2003 2004 2005 2006 2007 Average
Year
a Loans / Deposits
Analyzing the specific index of liquidity, we can find out the capability of a bank to fulfill its obligations. From
Table 11, we find out that the deposits more than cover the loans, for the total of the banking sector. We, also,
observe that the average for the time period under investigation for the total of the banks is 80,55%.
Table 11 Liquidity (Loans / Deposits) – Index
Bank 2003 2004 2005 2006 2007
Percentage
change
2007-2003
Average of five year
time period
Total of banks 74,16 77,42 79,77 84,88 86,54 16,69 80,55
Source: Union of co-operative banks of Greece (ESTE) / Bank of Greece / self processing
Rates
90
85
80
75
70
65
1. Liquidity=Loans / Deposits
2003 2004 2005 2006 2007 Average
Year
b Total Assets / Total Loans
Analyzing Table 12 we find out that the specific index it shows large differentiations and variations. We still
observe that the percentage of the specific index for the total of the banks is negative. This negative result effect the
average of 5 year time period.
Table 12 Liquidity (Total Assets / Total Loans)*100 – Index
Bank 2003 2004 2005 2006 2007
Percentage
change
2007-2003
Average of five year
time period
Total banks 205,27 186,22 187,83 175,87 178,20 -13,19 186,68
Source: Union of co-operative banks of Greece (ESTE) / Bank of Greece / self processing
200

Rates
210
200
190
180
170
160
5. Conclusions
2. Liquidity=Total Assets / Total Loans
2003 2004 2005 2006 2007 Average
Year
The Greek banks recognizing the opportunities for increasing their earnings, that arise in the market of the Balkan
countries, after the entrance of some of them in the European Union, just as the gradual accession of the rest, they
took care of expanding their magnitudes with bank buyouts in these countries. From the tables and the data of the
present paper it is understood that the Greek banks have substantially benefited from the cross border mergers and
acquisitions, since they have increased significantly the amount of their deposits and their loans portfolio as well and
have an adequately high ROAE index and of course it remains unknown whether the wave of cross-border mergers
and acquisitions in the banking sector of the Balkan countries will continue at the same rate, given the present
economic crisis.
We believe that in a few years from today the banking map of the Balkans will include a much lesser number of
banks, though of significantly larger size. The Greek banks will need to expand quickly and to acquire the
appropriate size, so as to have a leading part at the next stage of mergers and buyouts in the region after the crisis.
One basic result that comes out from this paper is that the banks in spite of the relatively short period of their
business activity in the Balkans have achieved significant magnitudes and they tried well to covered basic needs of
the local societies.
The estimation of the profitability and efficiency of the banks reaches satisfactory levels. However the course of
liquidity indexes for the banks is going down. The challenges of the Balkans have been recognized by the Greek
Banks and they expand in the Balkan countries. The growing economic strength in the Balkan area, it considered
that will allow the Greek Banks to improve their performance and expand their activities.
6. References
Brakman S., Garretsen H., Marrewijk van C., “Cross-border Mergers and Acquisitions on revealed comparative advantage and
merger waves” January 2008 TI 2008-013/2 Tinbergen Institute Discussion Paper
Kyriazopoulos G., Zissopoulos D., Sariannidis N., «The advantages and the disadvantages of mergers and acquisitions in the
international and Greek economy» ESDO Florina Sept. 2009.
Kyriazopoulos G., Petropoulos D., “Cross Border Mergers and Acquisitions in the Balkan Countries after the introduction of
Euro in Greece” Esdo Kavala June 2010
Kyriazopoulos G., Petropoulos D., “What are the advantages and disadvantages from the banks mergers and acquisitions? Does
Altman’s z-score model for bankruptcy motivate banks for mergers and acquisitions?
Evidence from the Greek banking system”. ICOAE Athens August 2010
Kyriazopoulos G., Petropoulos D., “Does the cross border mergers and acquisitions of the Greek banks in the Balkan area affect
on the course of profitability efficiency and liquidity indexes of them?” EBEEC 2011 Pitesti Romania
Pasiouras F., Tanna S., Zopounidis C., (2007), “The identification of acquisition targets in the EU banking industry:
An application of multicriteria approaches”, International Review of Financial Analysis, Volume 16, Issue 3, Pages 262-281.
Petropoulos D., Kyriazopoulos G., “Profitability, efficiency and liquidity of the co-operative banks in Greece” ICOAE Athens
August 2010
Rhoades Stephen A., 1986, "Bank Operating Performance of Acquired Firms in Banking Before and After Acquisition",
Washington Federal Reserve System, Staff Studies, No 149, (May).
Stubos G., Tsikripis I., Banking Sector Developments in South-eastern Europe May 2004
Topaloglou L., University of Thessaly Department of Planning and Regional Development Interaction, Perceptions and Policies
in the Boundaries of European Union: The Case of the Northern Greek Cross Border Zone Economic Alternatives, issue 1,
2008.
Vasiliou D., Financial Management Patra 1999
201

PERFORMANCE OF ISLAMIC BANKS ACROSS THE WORLD: AN EMPIRICAL ANALYSIS OVER
THE PERIOD 2001-2008
Sandrine Kablan, ERUDITE, University of Paris Est Créteil, France.
Ouidad Yousfi, MRM-CR2M, University of Montpellier 2, France.
Email: ouidad.yousfi@univ-montp2.fr
Abstract Our study aims at analyzing Islamic bank efficiency over the period 2001-2008. We use stochastic analysis frontier (SFA) to
estimate efficient frontier. Our results show that Islamic banks in our sample were efficient at 92%. The level of efficiency could
however vary according to the region where they operate. Asia displays the highest score with 96%. Indeed, country like Malaysia
made reforms in order to allow these banks to better cope with the existing financial system, display the highest scores. On the
contrary countries with Islamic banking system do not necessarily display efficiency scores superior to the average. The subprime
crisis seems to have impacted those banks indirectly. And market power and profitability have a positive impact on Islamic banks
efficiency, while it is the contrary for their size. The latter implies that they do not benefit from scale economy, may be because of the
specificity of Islamic financial products.
Keywords: Islamic Finance, Islamic Banks, performance, efficiency, stochastic frontier analysis.
JEL classification: G21, G24, G15.
1. Introduction
The market share of Islamic banks increases by 15% per annum (Moody's, 2008), this last decade. The
emergence and boom of Islamic finance, lead several economists to write on this topic. Many studies discussed in
depth about the rationale behind the prohibition of interest (Chapra, 2000), but also the policy implications of
eliminating the interest payments (see among others Khan, 1986, Khan and Mirakhor, 1987 and Dar 2003).
However, most of the existing literature on Islamic banking unleashes various studies made on the measurement of
performance in Islamic banks: they examine the relationship between profitability and banking characteristics.
A first group of studies are interested in the performance of Islamic banks in a specific country, through
financial ratios. Those ratios capture (a) profitability, (b) liquidity, (c) risk and solvency and (d) efficiency. For
instance, Saleh and Rami (2006) focus on the performance of the first and the second Islamic banks in Jordan:
Jordan Islamic Bank for Finance and Investment (JIBFI) and Islamic International Arab Bank (IIAB). They notice
that they play a major role in financing ventures in Jordan, particularly short-term investment, and both banks have
increased their activities and expanded their investment but, the JIBFI still has higher profitability. They conclude
that Islamic banks have high growth in the credit facilities and in profitability. Samad (2004) focused on the post
Gulf War period of 1990-2001 in Bahrain, and examined the performance of the interest-free Islamic banks and the
interest-based conventional commercial banks. His study shows that there is no major difference between the two
sets of banks in terms of profitability and liquidity performances but there is a significant difference in credit
performance. Kader and Asarpota (2007) evaluate the performance of the UAE Islamic banks by comparing the
Islamic and conventional banks. They examine the balance sheets and income statements of 3 Islamic banks and 5
conventional banks between 2000 and 2004. Their results show that Islamic banks are more profitable, less liquid,
less risky and more efficient than conventional ones. They conclude that the SPL principle (see annex) is the main
reason for the rapid growth of Islamic banks and suggest that they should be regulated and controlled in a different
way as the two kinds of banks have different characteristics in practice.
Again, Samad and Hassan (2000) performed an intertemporal study in which they compared the performance of
the Bank Islamic Malaysia Berhad (BIMB) between two periods of time 1984-1989 and 1990-1997. Then they
evaluate the interbank performance by comparing the BIMB's performance with 2 conventional banks (one smaller
and another larger than the BIMB) as well as 8 conventional banks. The results show that there is a significant
improvement of the BIMB performance between 1984 and 1997 but this improvement is less important than in the
conventional banks. Moreover, Islamic banks are less profitable and less risky but more liquid than conventional
banks. Moin (2008) compared the performance of Islamic banks relatively to conventional banks in Pakistan. The
study makes comparison of Meezan Bank Limited (MBL) which is the oldest Islamic bank in Pakistan and a group
202

of 5 conventional banks for the period of 2003-2007. He adopted an inter-bank analysis of the income statements
and the balance sheets of the two groups. The study found that there is no difference in terms of liquidity between
the two sets of banks. Besides, the MBL is less profitable, more solvent (less risky), and also less efficient
comparing to the average of the conventional banks but it is improving considerably between 2003 and 2007. This is
explained by the fact that the latter banks have a dominating position in the financial market with a longer history
and experience than the Islamic banks in Pakistan which have started their business few years back. Sarkar (1999)
studies the case of Islamic banks in Bangladesh. He finds that Islamic products have different risk characteristics
and concludes that prudential regulation should be modified. Those studies related each to one country and using
financial ratios tend to converge towards one conclusion. Islamic banks may be as efficient as conventional ones;
however there is a necessity of reforms, regulation and control for each banking system where they operate.
A second group of studies are interested in Islamic banks across several countries. Bashir (1999) and Bashir
(2001) examined the balance sheets and the income statements of a sample of 14 Islamic banks in 8 Middle Eastern
countries between 1993 and 1998. He analyzed the determinants of Islamic Banks' performance, specifically the
relationship between the profitability and the banks' characteristics. He found that the measure of profitability is an
increasing function of the capital and loan ratios. Besides, the study highlights the empirical role that adequate
capital ratios and loan portfolios play in explaining the performance of Islamic banks. Factors such as non-interest
earning assets and customer and short-term financing, etc contribute to the increase of the Islamic banks' profit.
Hassan and Bashir (2003) 1 , confirm the results of Bashir (2001) in the sense that the performance of Islamic banks is
affected not only by the bank's characteristics but also by the financial environment. Their results indicate that
controlling for macroeconomic environment 2 , financial market structure, and taxation; the high capital and loan-toasset
ratios improve the banks' performance. The study also provides an interesting but surprising results such as the
positive correlation between profitability and overhead; and the negative impact of the size of the banking system on
the profitability except net on interest margin.
Lastly, the third group of studies is interested in using efficiency frontier methods. Yudistira (2003) analyzed
the impact of financial crises on the efficiency of 18 Islamic banks over 1997-2000. This study is based on a nonparametric
approach Data Envelopment Analysis (DEA). It assesses a technical frontier of efficiency composed of
best practice banks. The efficiency score provided indicates how well a bank transform its inputs in an optimal set of
outputs. He highlighted the small inefficiency scores of 18 Islamic banks as compared to conventional banks. Sufian
(2007) adopted the same approach as Yudistira (2003) to examine the efficiency in domestic and foreign Islamic
banks in Malaysia between 2001 and 2004. He provided evidence that these banks improve their efficiency slightly
in 2003 and 2004. However, domestic Islamic banks are found marginally more efficient than foreign Islamic banks.
Besides Islamic banks profitability is significantly and positively correlated to three different types of efficiency:
technical, pure technical and scale efficiencies.
To measure and analyze technical and cost efficiency of Islamic Malaysian banks, Mokhtar, Abdullah and Al-
Habsh (2006) used the Stochastic Frontier Approach (SFA). Their findings show that, on average, the efficiency of
the overall Islamic banking industry (full-fledged Islamic banks and Islamic windows) has increased during the
period of study while that of conventional banks remained stable over time. However, the efficiency level of Islamic
banks is still lower than that of conventional banks. The study also reveals that full-fledged Islamic banks are more
efficient than Islamic windows 3 for local banks, while Islamic window of foreign banks tend to be more efficient
than those of domestic banks.
All those studies focus on one or a few countries. Our study aims at examining and evaluating the performance
of Islamic banks operating in 17 countries in Middle East, Asia and Africa, but also in United Kingdom. This scope
of analysis will allow us to compare Islamic bank efficiency through the differences characterizing those countries.
We used a stochastic Frontier Approach (SFA) over the period 2001-2008 to estimate a cost-efficiency frontier and
derived scores of cost efficiency, while taking into account explanatory variables.
The current paper is structured in the following way: Section (2) presents the model and the sample. The results
are discussed in Section (3). We conclude in Section (4).
1 They consider a larger sample in 21 countries between 1994 and 2001 and use cross-country bank level data.
2 The Islamic banks seem to have higher profit margins in favorable macroeconomic environment.
3 It refers to conventional banks that offer Islamic financial services, as part of their activity.
203

2. Model specification and Data
In the current study, we retain the parametric approach and use more specifically a stochastic frontier analysis
(SFA). It has the advantage of being more accurate than the nonparametric approach like DEA. As we explained it
before, it allows separating the error term of the inefficiency term. It is therefore less sensitive to measurement
errors and outliers 4 . As objective function, we choose a cost function. It allows taking into account the constraints of
banks as financial companies, seeking to optimize their financial performance. Thereby minimizing the costs
induced by the efficiency frontier, we will take into account this constraint. As functional form, we choose the
Translog, as it best suits the multi-products characteristic of banking technology, involving multiple inputs and
outputs, cf. Mester (1997), Bauer et al. (1998), Roger (1998) and Isik and Hasan (2002). We will assume a seminormal
truncated distribution for the inefficiency term, while the random error follows a normal distribution. We use
the maximum likelihood method for the estimate. Panel data allow us to gain in estimation accuracy by increasing
the number of data. However, our data are unbalanced as some banks are not observed at certain points in time. We
will use the intermediation approach as it assesses bank efficiency as a whole. Besides, the principle of Islamic
banking is participation in the company that is using the funds on the basis of PLS principle. Therefore, the
intermediation approach emphasizes intermediation function carried out by Islamic banks. This led us to determine
as inputs labour, physical capital and deposits. The prices of those inputs are measured respectively by personnel
expenses/total assets (PERSONEXP), other expenses/ total assets (OTHEREXP) and income for deposits/total
deposits (INTERESTEXP). For outputs, we have net loans (LOANS), net liquid assets (LA) and total earning assets
(SECURITIES). This classification is justified by the fact that Islamic banks engage in other types of profitable
activities, since they do not charge interest on loans and deposits (see table 3).
In studies of efficiency measurement for production units, economists usually allow for a second step. It’s set up
to explain the determinants of efficiency. Battese and Coelli (1996) showed that the two-step estimate biases the
efficiency scores. Indeed, the elements used in the second stage to explain efficiency influence its determination in
the first step. Thus, by excluding them from the expression of the efficiency frontier function, the result is a
measurement bias. Battese and Coelli advised therefore to introduce in the frontier function a vector of explanatory
variables. Thus, the computed model will be as equation 1 that follows:
n
t
n m 1
LnCTijt
� � 0 ��
� m ln pm,
ijt ��
� s ln ys,
ijt � ���
m,
n ln pm,
ijt ln pn,
ijt
m
s
2 m n
(1)
t s
n t
1
� �� � s,
t ln ys,
ijt ln yt
, ijt ���
� m,
s ln pm
ln ys,
ijt � zijt
� vijt
� uijt
2
s
t
m
where pm and pn are input prices and ys and yt are outputs quantities. Because of the specific form of the cost
frontier function, we impose constraints on symmetry, αm,n= αn,m and βs,t = βt,s homogeneity in prices Σmαm =1 and
adding-up Σmαm,n =Σnαn,m = Σmδm,s =0 5 . We also consider homogeneity constraints by normalizing total cost, the
labor price and physical capital price by financial capital price. The composite error term also takes a specific
functional form. The random components, vijt are independently and identically distributed according to standard
normal distribution, N (0;σv 2 ) while the bank inefficiency components, uijt > 0 are independently but not identically
distributed according to a truncated-normal distribution. The Stochastic Frontier Analysis assumes that the
inefficiency component of the error term is positive; that is, higher bank inefficiency is associated with higher cost.
The inefficiency of bank i in country j at time t is defined as exp (ûijt) where ûijt is the estimated value of uijt.
However, only the composite error term εijt = vijt - uijt can be observed from estimation of the cost function. The best
predictor of uijt is therefore the conditional expectation of uijt given εijt = vijt - uijt. To retrieve the inefficiency
component from the composite error for each bank from the cost function estimation, we use the method of Jondrow
et al. (1982) to calculate the conditional expectation. To investigate factors that are correlated with bank
inefficiencies, we use the so called conditional mean model of Battese and Coelli (1993, 1995), which permits in a
single-step estimation of the cost function and identification of the correlates of bank inefficiencies. In particular, the
estimation procedure allows for bank inefficiencies to have a truncated-normal distribution that is independently but
4 We use SFA instead of TFA or DFA, because the first one although easy to implement, leads to poor information. DFA requires the assumption
that cost efficiency is time invariant. Besides, when the time period of the panel is short, the random noise terms may not average 0, and
substantial amounts of random noise will appear in the cost inefficiency error component.
5 Homogeneity constraints are imposed by normalizing total costs and costs of two of the three outputs by the price of the third one.
s
204

not identically distributed over different banks. The mean of the inefficiency term is modelled as a linear function of
a set of bank-level variables. Specifically, the inefficiency terms, uijt are assumed to be a function of a set of
explanatory bank-specific variables, zijt and a vector of coefficients to be estimated, θ:
u � z � � w
(2)
ijt
ijt
where the random variable, wijt has a truncated-normal distribution with zero mean and variance, σu 2 . The point
of truncation is -zijt θ so that wijt > -zijt and uijt > 0. The inefficiency component of the composite error term therefore
has a truncated normal distribution, whose point of truncation depends on the bank-specific characteristics so that
the inefficiency terms are non-negative. To estimate the stochastic efficiency frontier, measures of bank inefficiency
and correlates of bank inefficiencies given by Equations (1) and (2), we use the Frontier econometric program
developed by Coelli (1996).
The variables influencing efficiency and therefore enabling to explain it, are related to characteristics of the
banking firm and its production process, as well as the environment in which banks operate. The size of the bank has
often been used in the literature as determinants of efficiency. Allen and Rai (1996) showed that large banks can
take advantage on economies of scale by sharing costs in the production process. It is measured by the logarithm of
total assets. The same authors and more specifically, authors that have worked on Islamic banks such as Yudistira
(2003) take into account regulatory and competitive conditions under which banks operate. Thus a variable used to
catch profitability of banks is measured by net income/total assets (ROA) (or net income/equity (ROE)); and for
Risk Taking Propensity we used the ratio equity/total assets. Indeed, Islamic banks refrain from charging interests
on loans and deposits to devote themselves to the principle of PLS. This redefinition of the banking practices lead to
new risks that conventional banks do not incur. Hence, there is a double interest here in our study to assess the
impact of their risk taking propensity on efficiency (see table 3).
Another variable that could have an impact on efficiency is the market share. It is measured by the ratio of total
deposits of the bank/total deposits in the whole banking system. It can increase costs for the banking system in
general because it results in slacks and therefore inefficiency that can not be solved. However, it can have a positive
impact on efficiency, if it is the result of consolidation and market selection of the largest and most efficient banks.
It appears therefore through lower costs, providing the market is contestable. The GDP per capita is a proxy of the
level of development. It influences many factors related to demand and supply of banking services, mainly deposits
and loans. Therefore, countries with a higher level of development are supposed to have more developed banking
system, with more competitive interest rates and profit margins. Demand density for banking products (measured by
deposits per square kilometre), has a negative impact on costs. In countries with high demand density, banks support
lower costs in the distribution of banking products. Again, the provision of banking services may be affected by
population density. In countries where this variable is low, banking costs are higher and banks are not encouraged to
increase their efficiency. We will test wether those variables are significant or not, according to their relative
correlation. We used data from balance sheets and income statements in their standard universal version of Database
Bankscope. The values of the variables are expressed in current dollars and have been deflated by the consumer
price index of the current year in order to reflect macroeconomic differences among countries. The macroeconomic
variables come from International Financial Statistics, from the IMF, available through DataStream. Total deposits
in each country for the calculation of market power were converted into dollars using market exchange rate end of
period.
3. Summary
Our paper provides the following results: Islamic banks are efficient at 92,72% on average over the period 2001-
2008. This efficiency differs depending on the region with maximum efficiency displayed by Islamic banks
operating in Asia (96, 21%). This level reflects the strong performance of Malaysian and Pakistani banks that
constitute most of our Asian sample. Malaysia in particular is emerging as one of the most developed centers in
Islamic finance, after Iran and Saudi Arabia. Since 1975 the government has reformed the financial system, so that it
promotes the development of Islamic banking alongside conventional finance. This was especially possible through
the Malaysian Islamic Banking Act of 1983. For the Pakistani case, since 1978, the government has fostered the
transformation of the banking system through the constitution of Commission for Transformation of Financial
System (CTFS), and the establishments of Islamic Banking Department by the State Bank of Pakistan. Thus the
government accompanied and framed this transformation through an appropriate regulatory system. Besides, Islamic
205
ijt

anks operating in Africa displayed an average efficiency score of 93,34%, with 92,75% for Sudan whose banking
system is essentially Islamic (government legislation). On the other hand, Islamic banks operating in United
Kingdom have an efficiency of 93,25%. This country has made efforts to include in its banking regulation, specific
rules for Islamic banks to better exercise in the British environment. Therefore, those banks attract capital from
Muslim immigrants and also petrodollars from the Middle East seeking investment opportunities. Finally, the
Middle East region has an efficiency score of about 92,49%. Especially, Iran which banking system is essentially
Islamic (government legislation) displays an average efficiency of 94,38%. As a whole, Islamic banks efficiency has
a decreasing trend over the period of analysis with a peak in 2005. It may be associated with the war in Iraq that
began in 2004, which gave rise to an oil shock. Petrodollars’ inflows into Islamic banks could explain this peak.
Besides, the lowest levels of efficiency appear in 2007 and 2008. This period corresponds to the subprime crisis.
Therefore we wanted to check its impact by using a dummy variable (D_subprime). However, it was not significant
(table 5, regression iv). Despite the decrease of Islamic banks efficiency during this period, we cannot assert that
this is due to the crisis.
To conclude, Islamic banks have expanded significantly in recent years because of increasing petrodollars
inflows, following the oil shocks. These banks are growing at a rate of 15% per year since the early 2000s. And
wherever they settle, the authorities try to implement adequate regulation in order to enable them to integrate the
banking system of these countries. It is within this context that our study is inserted to measure and understand what
explains the efficiency of These Islamic banks. At this purpose, we use the method of stochastic frontier in one step
(Battese and Coelli, 1996). This allows us to integrate in the cost frontier explanatory variables of efficiency. Thus,
this study shows that size has a negative impact on efficiency indicating that because they distribute Islamic
financial services, they may not benefit from economies of scale.
Profitability has a positive impact on efficiency, which is consistent with the literature. Finally, market power
has a positive impact on efficiency, the more clients Islamic banks have, the more efficient they are. Furthermore,
our study shows that in general Islamic banks are efficient with an average of 92,72%. However there are
differences across regions. Banks operating in countries with an Islamic banking system are not necessarily the most
efficient. The most efficient region is Asia (96%), with Pakistani and Malaysian banks. However operating in a
country where Islamic banking is government legislation or in Middle East is less costly for Islamic banks. Another
result observed is the decrease in efficiency at end of period (2007-2008). Although this period corresponds to the
subprime crisis we found no evidence that this decrease was due the crisis. A deepening of this study is to measure
the efficiency of Islamic banks in relation to conventional banks.
4. References
Al-Jarrah, I. and Molyneux, P. (2003), “Cost Efficiency, Scale Elasticity and Scale Economies in Arabian Banking”
paper presented in 10 th Annual Conference of Economic Research Form For the Arab Countries, Iran and
Turkey in Morroco 16-18 December 2003.
Allen, L. and Rai A., 1996 “Operational Efficiency in banking: An international comparison”, Journal of banking
and finance 20, 655-672.
Battese, G.E. and Coelli, T.J. (1993), “A stochastic frontier production incorporating a model for technical
inefficiency effects”, Working Papers in Applied Statistics, No. 65, Department of Economics, University of
New England, NSW, Australia.
Battese, G.E. and Coelli, T.J. (1995), "A Model for Technical Inefficiency Effects in a Stochastic Frontier
Production Function for Panel Data," Empirical Economics 20(2), 325-32.
Battese, G.E. and Coelli, T.J. (1996), "Identification of Factors Which Influence The Technical Inefficiency Of
Indian Farmers," Australian Journal of Agricultural Economics, Australian Agricultural and Resource
Economics Society 40(02).
Bashir, A.H. M., (1999), “Risk and Profitability Measures in Islamic Banks: The Case of Two Sudanese Banks,”
Islamic Economic Studies 6, 1–24.
Bauer, P.W., Berger, A.N., Ferrier, G.D. and Humphrey, D.B. (1998), "Consistency Conditions for Regulatory
Analysis of Financial Institutions: A Comparison of Frontier Efficiency Methods," Journal of Economics and
Business 50(2), 85-114.
Berger, A.N. and Humphrey D.B. (1997), “Efficiency of financial institutions: International Survey and Directions
for Future Research”, European Journal of Operational Research 98, 175-212.
206

Berger, A. N. and Mester L. (1997), “Inside the black box: What explains differences in the efficiencies of financial
institutions?”, Journal of Banking and Finance 21, 895-947.
Berger, A., Hunter W., Timme S. (1993), “The Efficiency of Financial Institutions: A Review and Preview of
Research Past, Present, and Future”, Journal of Banking and Finance 17(2-3), 221 249.
Chapra, M. U. (2000), “Why Has Islam Prohibited Interest? Rationale Behind the Prohibition of Interest”, Review of
Islamic Economics, 9, 5–20.
Coelli, T. (1996), “A guide to Frontier Version 4.1: A computer program for stochastic Frontier production and cost
function estimation”, CEPA Working paper.
Cebenoyan, S., Cooperman, E. and Charles, R. (1993). "Firm Efficiency and the Regulatory Closure of S & Ls: An
Empirical Investigation", The Review of Economics and Statistics 75, 540-545.
Cihak, M. and Hesse, H. (2008), "Islamic Banks and Financial Stability: An Empirical Analysis," IMF Working
Papers 08/16, International Monetary Fund.
Dar, H. (2003), Handbook of International Banking, Edward Elgar, chap. 8.
Hasan, Z. (2005), “Islamic Banking at the Crossroads: Theory versus Practice”, in “Islamic Perspectives on Wealth
Creation” edited by M. Iqbal M. and R. Wilson, Edinburgh University Press, UK pp. 3-20.
Hasan, I., Lozano-vivas A. and Pastor J., 2000, “Cross-border performance in European Banking”, Bank of Finland
Discussion Papers 24.
Hassan, M. K. (2003), Textbook on Islamic Banking, Dhaka, Bangladesh: Islamic Economic Research Bureau.
Isik, I. and M.K. Hassan, (2002), “Technical, scale and allocative efficiencies of Turkish banking industry”, Journal
of Banking and Finance 26, 719-766.
Kader, J.M., A.J. Asaporta, and Al-Maghaireh, A. (2007), “Comparative Financial Performance of Islamic Banks
vis-à-vis Conventional Banks in the UAE.”, Proceeding on Annual Student Research Symposium and the
Chancellor’s Undergraduate Research Award, http://sra.uaeu.ac.ae/CURA/Proceedings.
Mester, L.J., (2007), "Some thoughts on the evolution of the banking system and the process of financial
intermediation," Economic Review 1(2), 67 – 75.
Moktar, H.S., N. Abdullah and S.M. Al-Habshi (2006), “Efficiency of Islamic Banks in Malaysia: A Stochastic
Frontier Approach,” Journal of Economic Cooperation among Islamic Countries 27 (2), 37–70.
Rogers, K.E. (1998), "The non traditional activities and the efficiency of US commercial banks", Journal of banking
and Finance 22, 467-482.
Samad, A. (2004), “Bahrain Commercial Bank’s Performance during 1994-2001”, Credit and Financial
Management Review 10 (1), 33-40.
Samad, A. and M.K. Hassan (2000), “Performance of Islamic Bank during 1984-1997: An exploratory study.”
Thought on Economics 10 (1), 7-26.
Samad, A. (1999), “Relative Performance of Conventional banking vis-à-vis Islamic Bank in Malaysia.” IIUM
Journal of Economics and Management 7 (1), 1-25.
Sarker, M.A. (1999), “Islamic Banking in Bangladesh: Performance, Problems, and Prospects,” International Journal
of Islamic Financial Services, 1.
Yudistra, D. (2004), “Efficiency in Islamic Banking: An Empirical Analysis of Eighteen Banks,” Islamic Economic
Studies 12 (1), 1–19.
207

THE DETERMINANTS OF THE EAD
Yenni Redjah & Jean Roy, HEC Montréal
Inmaculada Buendía Martínez, Universidad Castilla-La Mancha & HEC Montréal
Email: jean.roy@hec.ca
Abstract. Defaults by individuals were at the source of the recent financial crisis, thus the need to fully understand credit risk from
personal borrowers. Expected loss from credit is usually decomposed in probability of default, loss given default and exposure at
default (EAD), the latter factor being yet the least investigated. This research seeks to contribute by identifying the determinants of
EAD. This empirical study took place at a large Canadian financial cooperative having some 150B$ in assets. It is worth noting that
Canada in particular and financial cooperatives in general, exhibited great resiliency during the crisis. The sample consisted of more
than 11000 cases of default occurring between 2003 and 2008 on revolving lines of credit granted to individuals. The independent
variables consisted of 11 idiosyncratic and 10 macroeconomic variables. The methodology followed basically that of Jacobs (2008),
which was applied to corporate loans. The results show that several factors are significant, namely the borrower’s age, the exposure
limit, the amount drawn, the interest rate applied on the line of credit and the utilization behavior. Moreover, the relationship of EAD
to macroeconomic factors points to its procyclicality. Overall, more than 50% of the variance of EAD can be explained. In sum, the
research sheds light on a credit factor, EAD on credits to individuals, which has remained rather obscure up to now. The improved
understanding of EAD can lead to better risk modeling, better credit management and, potentially, improve financial stability.
Keywords: Exposure at default, Credit risk, Provision for loan losses, Bank capital
JEL classification: G21
1 Introduction
Credit exposure is the most important risk assumed by financial institutions with regards to their traditional
activities. The recent crisis is an illustration if we consider how subprime loans were granted and managed. During
the past decades credit risk has inspired many researches led by practitioners, academics as well as regulators. These
latest, through the Basel Committee, promote sound credit risk management practices to insure the stability and
solvency of the financial system. In the second Basel Accord a formula emerged for the loss provision calculation
which put in relation the probability of default (PD), the loss given default (LGD) and the exposure at default
(EAD). Given that the latter factor is the least investigated, this research seeks to contribute by identifying the
determinants of EAD for household accounts.
The focus will be made here on the revolving credit lines granted to individuals contrasting with past empirical
studies on EAD for corporate loans. The data set includes 11278 defaults between 2003 and 2008 that have been
recorded by a large Canadian financial institution. The exposure at default will be tested on macroeconomic and
idiosyncratic variables using a stepwise approach. Several variables turned out to be significant in explaining the
EAD such as the utilization ratio before default, the authorized and the utilized amounts, the age, the interest rate
charged on the credit line, the 3 month risk free rate and the TSE300 index. These last two variables will support the
hypothesis that the EAD is procyclical.
The rest of the paper is divided as follow. The first section summarizes the literature dedicated to the EAD
factor. The second section presents the data base and the descriptive statistics. The third section presents the
methodology, the results and the residual analysis. A conclusion will close the paper.
2 Literature review
The literature dedicated to the EAD can be divided into three segments. Firstly, there is the documentation provided
by the BIS, through the Basel committee, which gives the qualitative and mathematical framework regarding the
EAD estimation. Secondly, several theoretical papers, published mainly before 2006, were used by the committee as
consultative papers to build the regulatory framework. Finally, we can find a few empirical studies on EAD,
published for the most after 2006, which can be used to compare the results obtained here. A brief summary of this
literature is presented in the two following sub-sections.
208

2.1 The Basel II Accord and the EAD
Under Basle II, the EAD is a key factor for the estimation of both the loss provision that appears in the balance sheet
and the regulatory capital that the financial institutions have to maintain. Together, they provide a confidence level
of 99.9% for the bank solvency. Thus, they cover the quasi-totality of the loss distribution. Two formulas have then
been developed to estimate respectively the expected loss for which a provision is estimated (equation 1) and the
regulatory capital (equation 2). These two equations show the importance of having an accurate estimation of the
EAD with regard to their linear impact on these two measures.
The EAD can be estimated in two ways using the Loan Equivalent Factor (equation 3) or the Credit Conversion
Factor (equation 4). The LEF represents the portion of the unutilized line that is expected to be drawn before default
(Sthephanou & Mendoza, 2005). The CCF is defined by Jiminez, Lopez and Saurina (2006) as the fraction of the
total commitment at time t that will have been drawn when the borrower reaches the default time. The CCF can be
seen as a utilization ratio of the credit line. The regulators recommend using the CCF instead of the LEF to estimate
the EAD which is why we choose the first as our proxy. This factor will then be our dependent variable. Given that
the EAD is usually expressed in monetary terms, we only have to apply the equations 5 or 6 to obtain its estimation.
E(
L)
� PD*
LGD*
EAD
K
�
2.2 Empirical studies of the EAD
�1� �1
�1
�LGD �����
( PD)
� � ��
( �)
��
1�
� ��
PD � LGD��
EAD�2�
LEF ( � ) �
i,
t
CCF ( � ) �
i,
t
EAD ( � ) �
i,
t
EAD ( � ) �
i,
t
�Drawni , t ��
� Drawni
, t �� Unutili
, t �3� �Drawni , t � LEFi
, t ( � ) �Unutili
, t �� Total _ Commitment � EADi
, t ( � ) � �Drawni , t �Unutili
, t ��4� Drawni
, t � LEFi
, t ( � ) �Unutili
, t �5� CCF ( � ) �(
Drawn �Unutil
) �6� i,
t
i,
t
i,
t
Even if studies on EAD are scarce, we can still point out some interesting results. Jiminez, Lopez and Saurina
(2006) found a clear difference between non-defaulted and defaulted loans regarding the utilization ratio. In fact, for
a given date of default, the non-defaulted borrowers have maintained a ratio of 50% during the five previous years.
The second group, composed by the defaulted loans, showed a ratio of 60% five year before and raised monotically
to reach the 90% mark at the year of default. In another paper, Jacobs (2008) implemented a simultaneous
regressions model with three different dependent variables (LEF, CCF and EADF) for which the same set of
independent variables was applied. The CCF turned out to be the more efficient model with regard to the R². The
key findings were that the EAD risk decreases with the PD, the credit quality and with higher levels of leverage and
liquidity. On the other side, the EAD risk increases with the company size and for unsecured loans. Finally,
subordinated debt induces more hazardous EAD than senior debt. Concerning the constitution of an EAD database,
Gruber and Parchert (2006) argue that unlike the PD which requires considering the entire portfolio of loans, the
EAD needs a data base including only the defaulted loans.
Regarding the procyclicality of the EAD, Asarnow and Marker (1995) found that for a sudden deterioration of
the economic condition or the credit quality, the utilization of the normally unutilized portion of the loan increases
faster for borrowers with a good credit rating. They explain it by the fact that less secure companies are facing more
monitoring by banks. Thus, the procyclicality in the EAD is driven primarily by the obligors classified among the
credit grades. Stephanou and Mendoza (2005) adopted the same point of view and assumed that the obligors that
have a good credit rating make default following a sudden deterioration of their financial situation which could be
possibly due to an unexpected economic downturn. On the other hand, more risky borrowers make default after a
gradual worsening of their solvency. In a broader perspective, Allen and Saunders (2003) explain that banks reduce
the volume of loans during economic slowdown at a moment where companies need it the most to overcome
liquidity problems. Other empirical studies such as Jacobs (2008), Kim (2008), Jiminez and Mancia (2007), Jiminez
and Al (2006) all find several evidences that the EAD follows a cyclical pattern. The number of default increases
during an economic crisis. The values of the CCF and LEF also reach their peak at these same moments. The
variables that explain the most this phenomena are the GDP, the three month risk free rate, the growth rate of real
estate prices and finally the S&P 500 index. However, we have to be cautious when comparing our results to these
past studies since the populations studied are not of the same type. In fact, we concentrate here on loans granted to
individuals while the past literature focuses on corporate loans.
209

3 Database and descriptive statistics
Before analyzing the data, let us see how the defaults are recorded. Thus, defaults are identified for individuals who
received a tracking code, borrowers for whom the interest rate was reduced and those who are 90 days late on their
payments. Also, defaults include restructured loans, non performing loans and loans for which the guarantee was
taken over by the institution. The following sections present the data and variables, and describe them statistically.
3.1 The database
The database includes 11278 cases of default between 2003 and 2008 on revolving credit lines granted to individuals
in the province of Quebec. A large set of variables have been taken into account. The idiosyncratic variables include
the age of the borrower, the utilization ratio twelve months before default (12 variables), the interest rate charged on
the credit line, the exposure limit, the drawn amount, the credit score and a large set of dummies including the types
of credit line (7), collateral (6), interest rate (3), credit rating (9) and years (6). The macroeconomic variables (8)
include indicators on interest rates (the 3 months Canadian Treasury Bill and the bank rate), the real estate market
(the 5 year Canadian mortgage rate, the REIT index and the Canadian New Housing Starts), consumption (spending
on durable goods) and on the overall performance of the economy (GDP and the TSE300 index). In addition of their
growth rate, the face value of the 3 month T-bill and the 5 year mortgage rate were also added.
3.2 Descriptive statistics
We can now take a closer look at some selected variables. Table 1 presents the annualized observations of the CCF,
calculated by equation 4, and the averages of the idiosyncratic variables. Firstly, we can see that the exposure limit
and the drawn amount have grown monotonically which amplify theoretically the EAD risk, but instead the CCF
reach his minimum during the crisis (2007, 2008) which could imply a bigger monitoring of the institution. The
credit rating have also decline, which could be a sign that the credit portfolio managers wanted to deal with more
secure borrowers, or, in accordance with Stephanou & Mendoza (2005), credit grade obligors are responsive to a
deterioration of the economy. Globally, the CCF is 0,7887 with relatively fine parameters of skewness and kurtosis.
However, the bimodal aspect of the CCF distribution, with peaks at 0 and 1, indicates that it has not a Normal shape.
2003 2004 2005 2006 2007 2008 TOTAL
Obs.¹ 479 899 1394 2089 2362 4055 11278
Proportions 0,0425 0,0797 0,1236 0,1852 0,2094 0,3595 -
CCF ² 0,8035 0,8496 0,7938 0,8050 0,7866 0,7647 0,7887
StdDev 0,3489 0,3104 0,3644 0,3495 0,3520 0,3543 0,3515
Skewness -1,5769 -2,0319 -1,4986 -1,5636 -1,4323 -1,2853 -1,4529
Kurtosis 3,8371 5,5830 3,5052 3,7980 3,4351 3,0721 3,4903
Exposure Limit ($) 5 945,15 6 444,53 6 777,98 10 176,74 13 512,38 14 714,06 11 609,40
StdDev ($) 8 710,56 11 508,61 14 236,32 29 359,39 31 475,24 36 625,92 29 987,94
Drawn Amount ($) 4 733,30 5 147,52 5 132,89 7 564,09 9 335,13 10 267,03 8 293,48
StdDev ($) 8 329,50 8 603,45 12 338,06 25 613,64 23 159,89 28 695,96 23 704,05
Interest Rate 0,0948 0,0874 0,0913 0,1012 0,1057 0,0937 0,0969
StdDev 0,0285 0,0293 0,0283 0,0307 0,0327 0,0326 0,0319
Credit Rating 5,42 5,59 5,46 5,38 5,05 5,25 5,29
Mode 6 7 7 6 6 7 7
Age 36,84 37,00 41,10 37,11 38,46 39,31 38,66
1 - Defaults on individuals credit lines between 2003 and 2008
2 - CCF = Sum[ Drawn at the default time ÷ Exposure Limit] ÷ Observations
Table 1: Annualized descriptive statistics: idiosyncratic variables
Table 2 shows that the resulting CCF is in line with the literature since it is constantly increasing as the
borrowers reach the default moment. The fact that the CCF is smaller at the time of default is probably a second
evidence of a stronger monitoring by the financial institution who tries to recover as much as it can to minimize the
default size. Logically, we obtain a utilized amount that grows linearly while the exposure limit remains stable.
210

Months to Default Obs. CCF Std Dev Skew Kurt Exp. Lim. Std Dev Drawn Std Dev
0 11278 0,7887 0,3515 -1,4529 3,4903 11 609 29 988 8 293 23 704
-1 11232 0,8829 0,2361 -2,1959 6,9024 11 726 30 431 9 178 24 456
-2 11177 0,8743 0,2435 -2,1025 6,4484 11 773 30 596 9 081 24 191
-3 11119 0,8681 0,2491 -2,0370 6,1357 11 947 31 226 9 046 24 360
-4 10884 0,8527 0,2628 -1,9429 5,7570 12 140 31 811 8 895 24 313
-5 10622 0,8397 0,2738 -1,8264 5,2737 12 133 31 834 8 706 24 113
-6 10338 0,8268 0,2857 -1,7502 4,9873 12 164 32 223 8 465 24 081
-7 10046 0,8122 0,2975 -1,6109 4,4311 12 130 32 227 8 241 23 782
-8 9790 0,8011 0,3076 -1,5447 4,1680 12 123 32 381 8 082 23 615
-9 9495 0,7931 0,3147 -1,4876 3,9548 12 120 32 458 7 920 23 434
-10 9205 0,7816 0,3241 -1,4117 3,6844 12 109 32 617 7 687 22 998
-11 8901 0,7764 0,3275 -1,3867 3,5999 12 085 32 676 7 523 22 737
-12 8611 0,7693 0,3315 -1,3434 3,4726 12 063 32 969 7 361 21 998
Table 2: Descriptive statistics: Evolution of the CCF 12 months before default
Since the credit score is a key element in measuring the credit quality and the probability of default, Table 3
presents the descriptive statistics regarding this variable. First of all, we see that the traditional relationships that we
can expect are holding between the CCF and the credit rating. The majority of the observations were rated between
6 and 8 which are the less secure borrowers. The worst score here is in fact 8 since 9 is an unclassified category. The
exposure limit could be a technique for this financial institution to manage its credit risk. In fact, the total
commitment size decreases with the credit quality. The age follows the same trend, where good ratings are assigned
to much older borrowers. High rated obligors are also benefiting from low interest rates that increase linearly until
the fifth credit score. From there, interest rates start to decrease, potentially for monitoring purposes.
1 2 3 4 5 6 7 8 9
Observations 387 847 1422 1506 1362 1933 2032 1524 264
Proportions 0,0343 0,0751 0,1261 0,1335 0,1208 0,1714 0,1802 0,1351 0,0234
CCF 0,5579 0,7122 0,7679 0,8001 0,7993 0,8018 0,8142 0,8376 0,7907
StdDev 0,3977 0,3834 0,3664 0,3446 0,3474 0,3479 0,3378 0,3062 0,3533
Skewness -0,2221 -0,9579 -1,3174 -1,5611 -1,5416 -1,5568 -1,6552 -1,8812 -1,4514
Kurtosis 1,4198 2,2927 3,0577 3,8254 3,7366 3,7967 4,1238 5,1395 3,5047
Age 50,2 46,7 42,5 39,9 37,6 35,8 34,6 33,8 32,1
Exposure Limit ($) 32 414 24 554 16 227 12 161 8 393 10 123 7 324 6 641 4 700
Drawn Amoount ($) 16 019 15 136 11 255 8 988 6 311 7 748 5 799 5 711 3 402
Interest Rate 7,9555 8,8141 9,5839 10,1124 10,3359 10,1992 10,2339 9,9325 9,2356
Table 3: Descriptive statistics: Distribution of the CCF by credit rating
Finally, Table 4 shows the information regarding macroeconomic variables. The CCF is negatively correlated
with all the variables, specially the TSE300 index, GDP and expenses on durable goods, meaning that the EAD risk
increases in bad economic environment. Here are the first signs of procyclicality in the exposure at default.
Growth Rate Std Dev Correl. vs CCF P-Value
3 months T-Bill -0,0087 0,1045 -0,1957 0,0995
Bank Rate -0,0053 0,0645 -0,3097 0,0081
5 year Mortgage Rate 0,0001 0,0265 -0,3753 0,0012
Gross Domestic Product 0,0039 0,0045 -0,4613 0,0000
Expenses on Durable Goods 0,0020 0,0065 -0,4939 0,0000
New Housing Starts 0,0183 0,1870 -0,1019 0,3945
S&P/TSE 300 Index -0,0012 0,0485 -0,4165 0,0003
S&P/TSX REIT Index 0,0053 0,0418 -0,2464 0,0369
Table 4: Descriptive statistics: Monthly Canadian macroeconomic variables from 2003 to 2008
211

4 Methodology and results
First of all, this chapter explains the methodology used to test empirically the EAD, which was inspired by previous
studies such as Jacobs (2008) and Jiminez et al. (2006). The results of the models follow in the nest section. Finally,
the residual analysis ends the section.
4.1 Methodology
The model developed here consists of a linear combination of idiosyncratic and macroeconomic variables to explain
the CCF which is our proxy for the EAD. Similarly to the studies mentioned above, we will also assume that the
CCF is normally distributed, which could appear as a strong hypothesis with regard to the bimodal aspect of the
distribution. Thus, we have opted for a Stepwise approach which is an appropriate choice for two main reasons.
Firstly, it admits overidentified models. Secondly, it is appropriate when no theoretical basis exists in the literature
that determines specific models to implement.
The stepwise procedure uses an iterative method that adds or removes variables to build a final model that
minimizes the estimation error or maximizes the coefficient of determination by choosing the optimal set of
independent variables. At each iteration (corresponding to a potential model), the program calculates the residual
sum of squares. For each model, the stepwise application computes the F-statistic which is obtained by comparing
the residual sum of squares of the initial model and the model after removing a variable. If the statistical significance
is not improved, the variable is maintained only if it is sufficiently significant (significance level of 10%). At the
end, only the variables that are significant at the 5% level will be included in the final model.
The observed CCF is our dependent variable. As Jiminez et al. (2006), we applied a logarithmic transformation
of the dependent variable to resolve a problem of heteroskedasticity. The fifty eight independent variables (thirtyone
dummies) included in the regression model are those listed in the section 3.1. The general equation that is
th
implemented is as follow, where CCF is the percentage exposure at time of default� for the i observation.
Log( CCFi
( � )) � � � � j * IdiosyncVarij
� �k
* MacroVark
� �t
( 7)
The stepwise method implies that the variables interact with each other creating a number of iterations that
grows exponentially with the number of variables. Also, depending on the order of the variables in the initial model
the stepwise application can lead to different optimal solutions. Thus, running this program leads to a solution that is
locally optimal. In this paper, tests have been made to insure that the solutions that are presented are stable and
unique.
4.2 Results
For brevity sake, Table 5 gives the outcome of the final model resulting from the Stepwise application. Thus, among
the 58 variables that were included in the initial model, one in four reached the 5% significance level mark. At first
sight, one notices that the number of observations dropped to 11140 due to the logarithmic transformation of the
dependent variable. First we see that there are 14 variables that enter the final model, combining for a R² of 51.4%.
The F-Statistic shows that the model is reliable. The age is negatively related to the CCF meaning that younger
borrowers carry more risk. The utilization ratio one month before the time of default is certainly the variable that
explains the most the CCF with regard to its significance level. So even if default is primarily due to a progressive
degradation of the CCF (Table 2), the true value of the EAD can be known only one month before default with some
confidence. The two variables UR-2 and UR-4 have a much lower importance due to the small value of their
coefficient.
In line with the findings of Jiminez et al. (2006) we found that the exposure limit and the drawn amounts are
good predictors of the EAD. In fact, the more the financial institution authorizes a substantial limit, the smaller
would be the exposure at default in percentage. However, this result must not be that surprising as we have seen that
the highest exposure limits were granted to the good quality creditors.
212

The interest rate charged on credit lines has also shown a significant and positive impact on the CCF. Thus,
when the interest rate increases, the EAD increases since payments on the credit line become bigger. In the other
hand, dummy variables are not vital here. For the type of products for example, if we combine the number of
observations of the three variables that turned out to be significant we obtain only 15% of the population, which
does not contribute notably to the global explanation of the EAD. Same goes for the collateral for which we have
only 22 data points.
Coefficient Std Dev T-Stat P-Value
Age -0,0010 0,0002 -5,9130 0,0000
Dum Credit Line Type: Personnal 0,0318 0,0123 2,5928 0,0095
Dum Credit Line Type: Student 0,4494 0,0606 7,4168 0,0000
Dum Credit Line Type: Protected 0,1165 0,0585 1,9903 0,0466
Utilisation Ratio -1 month 2,0976 0,0734 28,5746 0,0000
Utilisation Ratio -2 months -0,1849 0,0787 -2,3477 0,0189
Utililisation Ratio -4 months -0,1350 0,0355 -3,8028 0,0001
Dum Collateral: Specul. Res. Mrtg. -0,7138 0,1223 -5,8381 0,0000
Dum Interest Rate: Floatting -0,1317 0,0330 -3,9922 0,0001
Interest Rate Charged 0,0051 0,0018 2,8590 0,0043
Exposure Limit -1,15E-05 5,81E-07 -19,7620 0,0000
Drawn Amount 1,34E-05 6,34E-07 21,0641 0,0000
Delta 3 months T-Bill 0,1047 0,0380 2,7549 0,0059
Delta S&P/TSE300 -0,3003 0,0957 -3,1399 0,0017
Model Statistics
R² 0,5140 Model F-Stat 641,56
Estimator Std Dev 0,4317 Observations 11140
Table 5: Final model results. Independent variable = Log(CCF) for individual credit lines from 2003 to 2008
Finally, we can see that the CCF is driven by two macroeconomic variables, the 3 month T-Bill and the
S&P/TSE 300 index. As we have seen earlier, interest rates are positively related to the CCF, meaning that a
growing interest rate increases the credit risk. In fact, financial charges becoming more important, liquidity issues
could emerge, which might result in a default. The TSE 300 index shows that the CCF is correlated negatively with
the economic cycle. Thus, the EAD risk is more important during economic downturns.
These findings confirm the results of the previous studies on the EAD even if they focused on corporate loans.
In fact, utilization behavior of the credit line, the exposure limit, reaction to the economic cycle were all confirmed
and moreover with the same coefficient signs. In addition, we found for the first time that the interest rate charged
on credit line has a role to play on the EAD fluctuation. The age also explains the EAD with the traditional
relationship that we expect from this variable, which is negative meaning that younger borrowers carry more risk.
4.3 Residual analysis
The model implemented here respects the Gauss-Markov theorem, and then we can state that it is the best linear
unbiased estimator (BLUE). In fact, the 1.8250 mark that we have obtained for the Durbin-Watson test tells us that
there is no dependence among the variables. Also, the Breusch-Pagan test runed under Matlab shows that there is no
problem of heteroskedasticity. Finally, the error vector has an expected value of zero. We complete the Gauss-
Markov theorem by assuming that the database is randomly constituted and by attesting that the model is linear.
However, this model is not consistent due to the fact that the error terms do not follow a normal distribution, which
is not sufficient to invalidate the significance levels and the reliability of the final model. The hypothesis made on
the normality of the CCF explains the inconsistency of the model, which in reality follow a bimodal distribution
with peaks reached at the two extremities (0 and 1).
5 Conclusion
The recent financial crisis reminds us that managing credit portfolios is still more than ever a big issue in modern
finance. Among the different topics we choose here to focus on the EAD which has not retained the interest of many
213

esearches since the implementation of the new Basel capital accord. The other contributions of this paper came
from the quality of the database which was granted by a large Canadian financial institution and also by the fact that
EAD is tested on individual lines of credit, as opposed to corporate loans that were considered in past empirical
studies.
An econometric approach based on the Stepwise application has allowed us to determine that the age, the
exposure limit, the drawn amount, the interest rate, the utilization ratio a month before default, the 3 month
Canadian T-Bill and the S&P/TSE300 index had all an explanatory power on the EAD. These results are in line with
those listed in the literature review, meaning that there are little differences between the EAD of corporate and
individual loans.
Future research can try to formalize the dependence that is hardly suspected between the EAD and the two
others parameters of the expected loss (PD and LGD). Also, given that there are few empirical studies on the EAD,
future research could improve the literature by developing new methodologies, adding different kinds of credit line,
etc. Finally, this paper, we hope, will contribute to enrich the literature on the EAD, thanks to significant results that
were obtained. Improving the understanding of measures like the EAD will help financial institutions improve their
solvency, which contributes to a better financial system.
6 References
Allen, L. & Saunders, A. (2003). A survey of cyclical effects in credit risk measurement models, BIS Working
Papers, 126, Monetary and Economic Department, January.
Araten, M. & Jacobs, M. Jr. (2001). Loan equivalents for revolving credits and advised lines, The RMA Journal,
May, 34-39.
Asarnow, E. & Marker, J. (1995). Historical Performance of the US Corporate Loan Market 1988-1993, Journal of
Commercial Lending, 10(2), Spring, 13-32.
Basel Committee on Banking Supervision (2004). International Convergence of Capital Measurement and Capital
Standards: A Revised Framework, Bank for International Settlements, June.
Gruber, W. & Parchert, R. (2006). Overview of EAD Estimation Concepts. In B. Engelmann & R. Rauhmeier
(Eds.), The Basel II Risk Parameters: Estimation, Validation and Stress Testing (pp. 177-196). Springer Berlin
Heidelberg, Business and Economics.
JACOBS, M. Jr (2008). An empirical study of Exposure at Default, Washington, D.C., Office of the Comptroller of
the currency, OCC Working Paper, June.
Jiminez, G. & Mancia, J. (2009). Modelling the distribution of credit losses with observable and latent factors,
Journal of Empirical Finance, 16(2), March, 235-253.
Jimenez, G., Lopez, J.A. & Saurina, J. (2006). What do one million credit line observations tell us about Exposure
at Default? A study of credit line usage by Spanish firms, Banco De España, June.
Kim, M.-J. (2008) Stress EAD: Experience of 2003 Korea Credit Card Distress, Journal of Economic Research, 13,
73-102.
Moral, G. (2006). EAD estimates for facilities with explicit limits. In B. Engelmann & R. Rauhmeier (Eds.), The
Basel II Risk Parameters: Estimation, Validation and Stress Testing (pp. 197-242). Springer Berlin Heidelberg,
Business and Economics.
Ross, T., Minh, H. & Jarrad, H. (2007). Modeling exposure at default, credit conversion factors and the Basel II
accord, Journal of Credit Risk, 3(2), Summer, 75-84.
Stephanou, C. & Mendoza, J.C. (2005). Credit Risk Measurement Under Basel II: An Overview and Implementation
Issues for Developing Countries, Working Paper 3556, World Bank Policy Research, April.
214

A MACRO-BASED MODEL OF PD AND LGD IN STRESS TESTING LOAN LOSSES
Esa Jokivuolle, Aalto School of Economics, Bank of Finland
Email: esa.jokivuolle@bof.fi
Matti Virén, University of Turku, Bank of Finland
We present a macro variable-based empirical model for corporate bank loans’ credit risk, which captures the well-known positive
relationship between probability of default (PD) and loss given default (LGD; ie, the inverse of recovery) and their counter-cyclical
movement with the business cycle. In the absence of proper micro data on LGD, we use a random-sampling method to estimate the
annual average LGD. We specify a two equation model for PD and LGD which is estimated with Finnish time-series data from 1989-
2008. We also use a system of time-series models for the exogenous macro variables to derive the main macroeconomic shocks which
are then used in stress testing aggregate loan losses. We show that the endogenous LGD makes a considerable difference in stress tests
compared to a constant LGD assumption.
Keywords: stress testing, loan losses, default rate
JEL classification: G28
1. Introduction
Credit risk portfolio models for bank loans may be divided in two categories: 1) models based on portfolio theory,
and 2) empirical models based on macro variables and their interactions. Interest in the second group, the macrobased
models has been growing because regulators are increasingly requiring stress tests in which loan loss
scenarios are linked to a stressed state of the macro economy. In the macro-based models, default rates (which are
then used as estimates of probability of default (PD)) are modeled as functions of macro variables such as the output
gap and interest rates. However, the second central element of credit risk, loss given default (LGD) is typically not
modeled within the system of macro-based equations although evidence from corporate bonds clearly suggests that
PDs and LGD are positively correlated; apparently via the common exposure to the business cycle.
The absence of an “endogenous” LGD in the macro-based models is understandable because of the scarcity of
time-series data on bank loan LGDs. As a result, LGD is often taken as a fixed parameter (see e.g. Sorge and
Virolainen 2006) or its cyclical variation is modeled in a somewhat ad hoc manner. For prudential or stress testing
purposes, the fixed LGD parameter is supposed to be chosen in a conservative manner to represent a “downturn”
LGD. Clearly, if LGD data is available, more accurate estimates of the risk related to LGD and its joint dynamics
with PDs can be provided (see Miu and Ozdemir 2010). First, the positively correlated PD and LGD can produce a
much larger combined effect on credit risk in the case of severe recessions. Second, a non-constant LGD is
apparently needed to explain the observation that although even in “normal times” there are quite a few business
failures, loan losses remain small (see Figure 1 which illustrates the Finnish case). This may be partly explained by
the fact that new companies, which almost by definition are small, dominate the number of failures in good times.
However, the most obvious explanation is the contra-cyclical behavior of the LGD.
In this paper we present a model for the aggregate credit risk of banks’ corporate loans, in which the loans’
(average) probability of default and the (average) loss given default are jointly determined by a group of macro
variables and their joint dynamics. We extend the previous macro-based models of bank loans’ credit risk in two
ways. First, we use a recently suggested estimation method to obtain a time-series for the average LGD across
defaulted corporate loans (see Jokivuolle, Virén and Vähämaa 2010). Second, we incorporate the well-known
common exposure of PD and LGD to the business cycle.
As regards the first extension, the idea in the LGD estimation method suggested by Jokivuolle et al. (2010) is
the following. In the absence of data on actual LGDs, the method provides a way to back out an estimate of the LGD
from auxiliary data which reflect the unobservable LGD. It is first assumed that the LGD is a constant in the cross
section of defaulted companies but can vary from one period to another. The method is based on the simple identity
that in a given period losses from banks’ corporate loans are equal to the outstanding loans of the defaulted
215

companies in that period times the (constant) LGD. We have data on aggregate bank loan losses, the total number of
companies and defaulted companies, and the distribution of debt among all companies. However, as we do not know
which companies actually defaulted in a given period (year), the method uses random sampling to form the
distribution of possible LGDs which satisfy the above mentioned identity. The mean of this distribution (or in
principal any other central statistic, such as the median) can then be used as the final LGD estimate. Overall, we
believe that the method of Jokivuolle et al. (2010) may be helpful in estimating the PD-LGD correlation in cases
where no actual data on LGDs are available.
As to the second extension, we refine the preliminary approach taken in Jokivuolle et al. (2010) to incorporate
the endogenous LGD in a macro-based model of PD which in turn is based on the model of Sorge and Virolainen
(2006). We first find that the output gap, unlike as a factor explaining the PD, does not explain sufficiently well the
time variation in our estimated LGD series. Hence, we replace it by the aggregate “gross profit rate” which more
directly reflects the financial state of the corporate sector. The gross profit rate explains about 60% of the annual
time-series variation in LGD and helps establish in our PD-LGD model a relatively high positive correlation
between the PD and the LGD.
Following Sorge and Virolainen (2006), and using data from Finland, we then simulate the distribution of loan
losses with a two-step procedure. In the first step random shocks from the estimated model are simulated with the
Monte Carlo method in order to produce scenario paths for PD and LGD over a three-year horizon. In the second
step the PD and LGD scenarios are used to simulate cumulative aggregate loan losses for the three-year horizon for
a representative portfolio of corporate loans. We show that the extended model with the “endogenous” LGD
produces considerably higher loss risk estimates than a benchmark model with a constant LGD. Following Miu and
Ozdemir (2010), a constant downturnLGD would have to be 15-33% higher than the unconditional expected LGD in
order to match the unexpected loss estimates obtained from the endogenous LGD model.
The paper is organized as follows. In section 2 we discuss the LGD estimation method introduced in Jokivuolle
et al. (2010) and estimate a time-series for the average LGD in Finland over 1988-2008 which covers two very
pronounced recessions. In section 3 the joint model of PD and LGD, along with estimation results is presented,
using the Finnish data. Section 4 presents simulation results on the loan loss distribution, comparing the joint PD-
LGD model with a benchmark model with a constant LGD. Section 5 concludes.
2. Estimation of the average LGD
In this section we recap and extend the description of the bank loan LGD estimation procedure, suggested in
Jokivuolle et al. (2010). Their method starts by assuming that the LGD is the same for all defaulted corporate loans
in any given year (or any period, depending on data availability), but LGD can vary from year to year. The data
needed to estimate the LGD for a given year include banks' aggregate loan losses from corporate loans, the number
of companies in the beginning of the year and the number of defaulted companies during the year, and the loan size
distribution across companies.
The estimation method proceeds as follows. Consider a population of firms i, i=1,..., Nt, where t is the time index.
Let li.t denote the amount of bank loans that firm i holds in year t, and recall that LGDt is assumed constant across
firms in a given year. We indicate default by Di,t which obtains value 1 if firm i defaults, 0 otherwise. We can now
write total loan losses in year t (TLLt) as follows:
N
t
t
TLLt = � Di,
tli
, t LGDt
� LGDt
�
i�1
N
i�1
In other words, total loan losses in year t simply equal the sum of loans of defaulted firms, multiplied by the
common LGD.
As in the case of data available to us from Finland, we do not know which firms defaulted in a given year, but we do
know the total number of defaulted firms each year, which is denoted by kt. We also know the total loan losses from
banks’ corporate clients (TLLt) and we have information of the bank loan size distribution across firms so that in the
estimation procedure li,ts are taken as given.
216
D
l
i,
t i,
t
(1)

As one can either observe or measure everything else than LGD in equation (1), LGD can be estimated on the
basis of equation (1). This is done by drawing random samples of size kt out of the annual population of Nt firms.
For each random sample of kt defaulted firms with loans li,t, LGDt is solved from equation (1). Denote this by
LGD * t. In addition, the method imposes a feasibility criterion on LGD * t that it has to lie between 0 and 100%. After a
sufficient number of random samples (we used 100 000), which meet the feasibility criterion, LGD * ts are organized
into a frequency distribution. The expected LGD * t of the distribution is then used as the final estimate of the annual
LGDt. To better understand the intuition of the estimation method, note that if one actually knows which firms
defaulted in a given year, then the procedure would simply reduce to using equation (1) to directly solve the LGDt.
Alternatively, if all loans were of equal size (or could be approximated to be so), also then LGDt could be directly
solved from (1). This follows from the fact that, if li,t = lt for each i, then the term � t N
Nt
to lt
� Di,
t � lt
kt
i�1
of loans.
1
D
i t i t l , ,
would simply reduce
regardless of which companies actually defaulted because all companies had the same amount
Figure 2 provides the entire time series of estimated LGDts over the sample period 1988-2008. In 1991, in the
beginning of the Finnish 1990s crisis, LGD reached its peak at 73% in our estimations. The average LGDt over time
in our sample is about 39% which is broadly in line with earlier literature (see e.g. Schuermann, 2004) as well as
with the assumption of 45%, central in the Basel capital adequacy framework.
Schuermann (2004) surveys factors which are systematically related to LGD. He finds the business cycle to be
of primary importance. As seen from Figure 2, our estimated LGD series for Finland also appears to be strongly
related to the business cycle, measured by output gap. In the following section we turn to modeling this relationship.
Secondly, one could try to extend the LGD estimation procedure to industry effects by allowing a separate LGD for
each industry. However, this would extend the estimation procedure to a vector of LGDs rather than a single LGD
and would hence add complexity, along with additional data requirements, to the procedure and might substantially
increase the computational burden of the procedure.
3. Modeling PD and LGD
We define PD as the aggregate corporate default rate and model it by output gap, real interest rates and indebtedness.
While the output gap reflects firms’ general demand conditions and hence drives the cyclical behavior of the default
rate, interest rates and indebtedness affect the sensitivity of firms’ financial health to demand shocks.
Turning to LGD, Figure 2 shows that the annual average LGD, which we have estimated in section 2, varies
much over time and that it has a strong positive relationship with the PD. A natural way to take into account the
cyclical features of LGD would seem to be to model LGD, just like the PD, as a function of the output gap.
However, such an LGD equation fits the Finnish data quite poorly; the R 2 would be only about 20% (see Jokivuolle
et al. 2010). Therefore, we abstain from using the output gap as the main determinant of LGD. Instead we use the
aggregate gross profit rate as the explanatory variable of LGD. This choice can be motivated by the following
arguments. First, the gross profit rate reflects cyclical changes in the economy. Second, it reflects the financial
position of the firms in that higher gross profit rate allows firms to finance their projects with lower leverage (e.g.,
using retained earnings instead) and hence the debt burden is likely to be smaller. The recovery rate from less
indebted firms tends to be higher. Third, a more profitable corporate sector, reflected by a higher gross profit rate,
can better absorb forced asset sales by defaulted companies (cf. the argumentation above based on Schleifer and
Vishny 1992). The model with the gross profit rate regressor turns out to fit the data quite well; as we will see below
the R 2 is typically above 60% in sample periods.
Following Sorge and Virolainen (2006), PD is expressed in a logistic functional form to ensure that in the stress
test simulations PD always lies between zero and one. The same restriction applies to LGD, so we apply the logistic
transformation also to the LGD. Starting with the PD, we define
PD = (1+exp(yPD)) -1 , (2)
where yPD is the macroeconomic index whose parameters must be estimated. A higher value for yPD implies a better
state of the economy with a lower default probability p, and vice versa. yPD is given by the logit transformation
217

L(PD) = ln((1-PD)/PD) = yPD
(3)
Analogous notation applies to LGD so that have the corresponding index yLGD.
The macroeconomic indexes for PD and LGD are determined by the exogenous macroeconomic factors (already
discussed above) in a linear model which in the time domain take the form:
yjt = ∑ βjixit + υjt, (4)
in which j=PD, LGD, �j is the corresponding set of regression coefficients for explanatory variables xi,t, for i = 1...4,
and �j,t is the error term. Equations (2) - (4) form a multifactor model of PD and LGD. The systematic risk
components are captured by the macroeconomic variables xi,t and the idiosyncratic PD and LGD shocks are captured
by the error terms �j,t.
As in Sorge and Virolainen (2006), individual macroeconomic variables are modelled as a system of univariate
AR(2) processes:
xit = αi0 + αi1xi,t-1 + αi2xi,t-2 + εit
The system of equations (2) - (5) govern the joint evolution of the PD, the LGD and the associated macroeconomic
factors with a 6 � 1 vector of error terms, E, and a 6 � 6 variance-covariance matrix of errors, �:
��
�
� ��
��
, � �
E � �� �� ~ N(
0,
�)
, � � � � .
��
�
���
, � ��
�
The PD-LGD system with AR(2) processes for the independent macro variables is estimated with Finnish
quarterly data which cover the period 1989Q1-2008Q4. The results make sense both in terms of signs and
magnitudes of coefficients and overall predictive power. Note that because of the logistic transformation, the signs
of coefficients both in the PD and the LGD equation must be reversed to arrive at the correct economic
interpretation. Hence, for example, after the positive sign of the output gap coefficient in the PD equation is
switched to negative, we obtain the expected relationship that PD decreases as the output gap increases.
The system was estimated with the SUR estimator to take into account the correlation between all residuals,
both of the PD and LGD equation and the AR(2) equations. Residual correlation is the channel via which the model
captures the cyclical co-movement of PD and LGD because in our model specification the PD and LGD equations
do not share common observable factors. In the data, the correlation between the PD and LGD series is as high as
ca. 90%. The PD-LGD correlation is captured via (1) correlation between residuals of the LGD and the PD equation
and via (2) correlation between the gross profit rate (which is the sole exogenous variable in the LGD equation) and
the exogenous variables of the PD equation (ie, output gap, interest rate and indebtedness). Of primary interest is the
correlation between output gap and gross profit rate because these present the cyclical factors in the PD and LGD
equation, respectively. Their correlation is ca. 24% which is not very high. By contrast, the correlation between
gross profit rate and the two other exogenous variables of the PD equation, interest rate and indebtedness, are quite
high; -0.87 in the case of interest rate and -0.95 in the case of indebtedness. These correlations suggest that debtrelated
variables and debt capacity (reflected in gross profit rate) are significant drivers of the PD-LGD comovement.
Finally, the correlation between the residuals of the PD and the LGD equations is 0.57. This indicates
that a large share of the PD-LGD correlation observed in the data cannot be attributed to the set of the observable
exogenous variables present in the model. PD and LGD simulations with the model in section 4 were able to
generate a PD-LGD correlation of 0.44 which falls short of the ca. 0.90 PD-LGD correlation in the data.
Nonetheless, we conclude that the model captures the PD-LGD cyclical co-movement, if not its full quantitative
scale, and is hence useful in considering its implications for credit risk. These implications are the subject of the next
section.
4. Simulation results
Given the estimated model for PD and LGD and following Sorge and Virolainen (2006) we simulate the banking
sector loan losses from corporate loans over a three-year horizon. This is done with Monte Carlo simulations where,
in the first place, we generate a set of values, say, 10 000 random scenario paths, for PD and LGD. We use a sample
of 3000 Finnish firms and their bank loans, which is available to us only for year 2002. In doing so, we assign the
218
(5)

aggregate PD as the probability of default for each company in the sample portfolio. Binomial random draws
according to the common PD are used to determine defaulted firms in each simulation round. Given the simulated
value of LGD for each period, which is similarly assumed to be the same for all defaulted companies, we obtain the
loan loss for each individual defaulted company. Individual losses are then aggregated to portfolio level loan losses.
From the resulting 10 000 aggregate loan loss scenarios we can then compute the expected value of loan losses plus
various distributional measures like 99% and 99.9% value-at-risk figures of the loan loss distribution (see Figure 3
for an example of our simulated loan loss distributions). The “raw” value-at-risk figures are further transformed into
“unexpected loss” figures by subtracting from them the expected loss.
Our key result is to compare the unexpected loan losses between the case of (1) endogenous LGD, based on our
PD-LGD model and the case of (2) “exogenous” LGD where the LGD equation is “shut down” and the LGD in
simulating loan losses is fixed at a constant which is equal to the expected LGD obtained from the PD-LGD model.
Table 1 shows that the 99% unexpected loss increases by 19% and the 99.9% unexpected loss by 30% when the
endogenous LGD is used instead of the exogenously given constant LGD. Figure 3 illustrate the higher skewness of
the loan loss distribution based on the endogenous LGD model. These results clearly show that taking into account
the systematic “risk” in the LGD by endogenizing it, makes a sizable difference in credit portfolio risk estimates.
Miu and Ozdemir (2010) consider the following interesting question. How large a constant “downturn” LGD
estimate should be that it compensates for the fact that LGD varies in time and that PD and LGD are correlated? We
can use our PD-LGD model to ask the same question. How big should the constant LGD parameter be so that the
exogenous LGD model would produce the same unexpected loss (either at 99% or 99.9% level) as the endogenous
LGD model? We find that approximately a 45% constant LGD (instead of the 39% expected LGD used in Table 1
baseline case) is required to match the 99% unexpected loss level of the endogenous LGD model. Similarly, a 52%
constant LGD is required to match the 99.9% unexpected loss of the endogenous LGD model. Adjustments of this
size are somewhat lower than in the model by Miu and Ozdemir (2010).
So far we have used the model for simulating the loan loss distribution by using sample averages as starting
values for the exogenous macro variables. As in Sorge and Virolainen (2006), the framework also offers the
possibility condition simulations on different macro shock scenarios. For example, one may consider a scenario in
which everything goes “wrong” as it often does before large financial disasters. Typically, output falls, indebtedness
increases (perhaps as a consequence of a collapse of the exchange rate regime) and interest rates increase as a result
of loss of policy credibility (or, as a result of defending the exchange rate regime). An example of this kind of rather
extreme events is the Finnish depression in the early 1990s (see Conesa et al. 2007). However, the recent global
financial crisis of 2008-2010 showed up in Finland quite differently from the 1991-1993 depression. Although
output fell by almost the same amount, this time there were no major debt or interest rate shocks. The gross profit
rate fell again but now its initial level was considerably higher. As a result, loan losses remained rather limited,
given the drop in economic activity (see Figure 1).
In Table 1, we consider the effects of different shocks by introducing a permanent one standard deviation shock
to the exogenous macro variables for the three-year horizon. Conditional on such initial shocks, the stochastic
simulation is carried out as above. We focus on the conditional expected loss, given the chosen initial shocks. The
shocks are introduced either one by one or all simultaneously (see row “All shocks” in Table 1). In addition, we
include a simultaneous shock scenario which roughly captures the macro state that prevailed in Finland in 1991-
1993.The expected loan losses as a share of the aggregate loan stock, conditional on such a macro scenario, are 6.5%
cumulatively over the three-year period. This model based scenario falls short of reality; the roughly 10% loan loss
rate in Finland over 1991-1993. However, the fit is much closer than what we would obtain with the constant,
exogenous LGD model. An even higher fit is obtained by introducing a very strong “sudden stop” shock to the
system. If a one-period shock which is of the same magnitude as the sum of the 12 periodical shocks is introduced in
the first period, loan losses exceed the above mentioned 10% level (11.5%). This reflects the persistence of shocks in
the system: a sudden stop shock has a larger impact than a series of smaller shocks that amount to the same total
magnitude.
In addition, the model allows us also to consider idiosyncratic shocks to LGD. As already discussed in section
3.1, these could result from, say, a sudden change in asset prices (not explicitly modeled in the current framework)
due to a change in expectations. The results of such a shock are also provided in Table 1; again we consider a shock
that corresponds to one standard deviation of the error term of the LGD equation. Such an idiosyncratic LGD shock
219

produces a similar set of responses as the different macro shocks in Table 1. It is interesting to ask what are the
relative contributions of the different types of shocks; how much can be attributed to shocks in exogenous variables
and how much to shocks in the respective variable itself. To further investigate this question we estimated a simple
VAR model for all variables. The variance decomposition of the VAR showed that the share of idiosyncratic LGD
shocks is about 52% while the corresponding number for the PD would be “only” 36%. This additional analysis
suggests that we should not only focus on the exogenous variables as potential sources of banks’ loan losses but
should also be aware of the potential influence of institutional, structural and expectations-related shocks, the
potential magnitude of which in the current framework is captured by the idiosyncratic LGD error term.
5. Conclusions
In the light of empirical data from Finland as well as our simulations of unexpected bank loan losses it is quite
obvious that an assumption of a constant (exogenous) LGD is not correct. Simulations show that credit risks
assessed with the help of the endogenous LGD model are considerably higher than those obtained from the
benchmark exogenous LGD version of the model. Another way to illustrate this is that a constant “downturn” LGD
would have to be 15 to 33% higher than the time-series average LGD in order to match the unexpected loss levels
obtained with the endogenous LGD model (cf. Miu and Ozdemir 2010).
The aim of endogenizing the LGD has incurred a certain cost in that we have had to abandon the industryspecific
PDs of the original Sorge and Virolainen (2006) model and replace them with the aggregate PD of the
corporate sector. A natural extension would be to try to bring back the industry-specific structure, should additional
data requirements be met in order to obtain industry-specific LGDs. This could perhaps be achieved with a
somewhat condensed industry categorization structure. Also the introduction of asset prices to the operational
structure of the model could be considered.
6. Acknowledgements
We would like to thank Kimmo Virolainen who kindly provided the original computer code, Mauri Kotamäki for
excellent research assistance, and Anni-Mari Karvinen and Liisa Väisänen for valuable help with the LGD
estimation. The usual disclaimer applies.
7. References
Basel Committee on Banking Supervision (2006) Basel II: International Convergence of Capital Measurement and
Capital Standards: A Revised Framework - Comprehensive Version, June 2006.
Conesa, J., Kehoe, T. and Ruhl, K. (2007) Modeling great depressions: the depression in Finland in the 1990s. In by
T. Kehoe and E.Prescott (Editors) Great Depressions of the Twentieth Century (Paperback), Federal Reserve
Bank of Minneapolis.
Jokivuolle, E., M. Virén, O. Vähämaa (2010) Transmission of macro shocks to loan losses in a deep crisis: The case
of Finland in Model Risk in Financial Crises - Challenges and Solutions for Financial Risk Models, Rösch, D.
and Sceule, H. (eds), Risk Books, London, 2010.
Miu, P. and Ozdemir, B. (2010) Basel requirement of downturn LGD: modeling and estimating PD & LGD
correlations. Journal of Credit Risk, forthcoming. Available at SSRN: http://ssrn.com/abstract=907047
Schleifer, A., R. Vishny (1992) Liquidation values and debt capacity: a market equilibrium approach. Journal of
Finance 47, 4: 1343–66.
Sorge, M., Virolainen, K. (2006) A comparative analysis of macro stress-testing methodologies with application to
Finland. Journal of Financial Stability 2(2), 113–151.
220

Simulation PD LGD Expected loss rate
Base: endogenous LGD 0.0044 0.39 0.021
GDP shock 0.0057 0.39 0.027
Interest rate shock 0.0058 0.39 0.027
Debt shock 0.0052 0.39 0.025
Gross profit rate shock 0.0044 0.53 0.028
LGD shock 0.0044 0.45 0.024
All shocks (excl. LGD shock) 0.0085 0.52 0.054
Finnish crisis 1991-93
0.0087
0.60
0.065
Base: exogenous LGD
0.0044
0.39
0.020
All shocks (excl. LGD shock) 0.0086 0.39 0.039
In all cases, a shock is defined as one standard deviation of the error term in the respective AR(2) models, or of the error term of the LGD
equation in the case of the LGD shock.
Table 1 Expected PD and LGD and expected loss in the conditional and unconditional simulations
5
4
3
2
1
0
Bankruptcy rate p.a.
loan losses in 2009 prices
86 88 90 92 94 96 98 00 02 04 06 08
Figure 1 Aggregate corporate default rate (PD) and loan loss rate in Finland
.08
.06
.04
.02
.00
-.02
-.04
-.06
gap
LGD
88 90 92 94 96 98 00 02 04 06 08
Figure 2 Estimated quarterly average LGD (right scale) and output gap (left scale) in Finland
0
0 1 2 3 4 5 6 7 8 9
Loss in % of total credit exposure
The figure illustrates the unconditional simulation with endogenous LGD.
Frequency (%)
2.5
2
1.5
1
0.5
Figure 3 Example of the simulated loan loss distributions
221
.8
.7
.6
.5
.4
.3
.2
.1

XTREME CREDIT RISK MODELS: IMPLICATIONS FOR BANK CAPITAL BUFFERS
David E. Allen, Edith Cowan University
Akhmad R. Kramadibrata, Edith Cowan University
Robert J. Powell, Edith Cowan University
Abhay K. Singh, Edith Cowan University
Abstract. The Global Financial Crisis (GFC) highlighted the importance of measuring and understanding extreme credit risk. This
paper applies Conditional Value at Risk (CVaR) techniques, traditionally used in the insurance industry to measure risk beyond a
predetermined threshold, to four credit models. For each of the models we use both Historical and Monte Carlo Simulation
methodology to create CVaR measurements. The four extreme models are derived from modifications to the Merton structural model
(which we term Xtreme-S), the CreditMetrics Transition model (Xtreme-T), Quantile regression (Xtreme-Q), and the author’s own
unique iTransition model (Xtreme-i) which incorporates industry factors into transition matrices. For all models, CVaR is found to be
significantly higher than VaR, and there are also found to be significant differences between the models in terms of correlation with
actual bank losses and CDS spreads. The paper also shows how extreme measures can be used by banks to determine capital buffer
requirements.
Keywords: credit risk, conditional value at risk, conditional probability of default, historical simulation, Monte Carlo simulation.
JEL Classification: G01, G21, G28
Acknowledgements: Allen and Powell thank the Australian Research Council for funding support.
1. Introduction
The Global Financial Crisis (GFC) has raised widespread spread concern about the ability of banks to accurately
measure and provide for credit risk during extreme downturns. Prevailing widely used credit models were generally
designed to predict credit risk on the basis of ‘average’ credit risks, or in the case of Value at Risk (VaR) models on
the basis of risks falling below a pre-determined threshold at a selected level of confidence, such as 95 percent or 99
percent. The problem with these models is that they are not designed to measure the most extreme losses. It is
precisely during these extreme circumstances when firms are most likely to fail, and it is exactly these situations that
the models in this study are designed to capture. Although the use of VaR is widespread, particularly since it’s
adaptation as a primary market risk measure in the Basel Accords, it is not without criticism. Critics include
Standard and Poor’s analysts (Samanta, Azarchs, and Hill, 2005) due to inconsistency of VaR application across
institutions and lack of tail risk assessment. VaR has also been criticised by Artzner, Delbaen, Eber and Heath
(1999; 1997) as it does not satisfy mathematical properties such as subadditivity. Conditional Value at Risk (CVaR)
on the other hand, measures extreme returns (those beyond VaR). The metric has been shown by Pflug (2000) to be
a coherent risk measure without the undesirable properties exhibited by VaR. CVaR has been applied to portfolio
optimization problems by Uryasev and Rockafellar (2000), Rockafeller and Uryasev (2002), Andersson et.al (2000),
Alexander et al (2003), Alexander and Baptista (2003), Rockafellar et al (2006), Birbil, Frenk, Kaynar, & Noyan
(2009) and Menoncin (2009). CVaR has also been explored as a measure of sectoral market and credit risk by Allen
and Powell (2009a, 2009b), but compared to VaR, CVaR studies in a credit context are still in their infancy.
Given the importance of understanding and measuring extreme credit risk, the aims of this study include
demonstrating how CVaR techniques can be applied to prevailing models to measure tail risk, to investigates to
what extent these CVaR measures are significantly differently from VaR measures, to show how asset value
fluctuations can impact on bank capital buffer requirements, and to ascertain which of the models most highly
correlate with actual measures of credit risk (including Credit Default Swap spreads, delinquent loans and chargeoffs).
To ensure a thorough examination of CVaR metrics we use a range of models (four in total as explained in
Section 2), as well as apply two techniques (Historical and Monte Carlo Simulation) to each model.
222

2. Data and Methodology
2.1. Data
Data is divided into Pre-GFC (January 2000 to December 2006, this 7 year timeframe aligns with Basel Accord
advanced model credit risk requirements) and GFC (2007 to 2009). We also generate an annual measure for each
model for each of the 10 years in the dataset. For our Merton / KMV based models (Xtreme-S and Extreme Q)
which require equity prices, we obtain daily prices from Datastream, together with required balance sheet data,
including asset and debt values. To ensure a mix of investment and speculative entities we use entities from the S&P
500 as well as entities from Moody’s Speculative Grade Liquidity Ratings list (Moody's Investor Services, 2010a).
In both cases we only include rated entities, for which equity prices and Worldscope balance sheet data are available
in Datastream. Entities with less than 12 months data in either of the 2 periods are excluded. This results in 378
entities consisting of 208 S&P 500 companies and 170 speculative companies. Credit ratings required for the
transition based models are obtained from Moody’s (Moody's Investor Services, 2010b). We use Standard and
Poor’s (2009) US transition probability matrices for each year in our study. For the Pre-GFC vs GFC periods, we
average the matrices for the relevant years in the dataset. Annual delinquent loans and charge-off rates were
obtained from the U.S. Federal Reserve Bank (2010). Annual CDS figures for US Corporates were obtained from
Datastream. These CDS figures were extracted by credit rating, and weighted according to the dollar value of debt
for each credit rating category in our data sample.
2.2. Methodology Model 1: Xtreme-S
We use the Merton / KMV approach to estimating default, and then modify this calculation to incorporate a CVaR
component (which we term CPD as the model uses probability of default as opposed to VaR). The Merton/KMV
model is well documented (for example, Crosbie & Bohn, 2003). In summary the point of default is where the firm’s
debt exceeds asset values. Distance to default (DD) and probability of default (PD) are measured as
2
ln( V / F)
� ( � � 0.
5�
V ) T
DD �
� T
PD � N(
�DD)
(2)
where V = market value of firm’s assets, F = face value of firm’s debt (in line with KMV, this is defined as
current liabilities plus one half of long term debt), µ = an estimate of the annual return (drift) of the firm’s assets, N
= cumulative standard normal distribution function.
We define conditional distance to default (CDD) as being DD on the condition that standard deviation of asset
returns exceeds standard deviation at the 95 percent confidence level, i.e. the worst 5 percent of asset returns. We
term the standard deviation of the worst 5 percent of returns as CStdev, which we substitute into equation 1 to obtain
a conditional DD:
2
ln( V / F)
� ( � � 0.
5�
V
) T
CDD � (3)
CStdev
V
T
Our Monte Carlo approach generates 20,000 simulated asset returns for every company in our dataset. We
generate 20,000 random numbers based on the standard deviation and mean obtained using the Historical approach.
We then follow the same approach as for the Historical model, applying the standard deviation of all simulated
returns to equation 1 to measure DD and the standard deviation of the worst 5 percent of simulated returns to
equation 3 to measure CDD.
2.3. Methodology Model 2: Xtreme-Q
Quantile regression per Koenker & Basset (1978) and Koenker and Hallock (2001) is a technique for dividing a
dataset into parts. Minimising the sum of symmetrically weighted absolute residuals yields the median where 50
223
V
(1)

percent of observations fall either side. Similarly, other quantile functions are yielded by minimising the sum of
asymmetrically weighted residuals, where the weights are functions of the quantile in question per equation 3. This
makes quantile regression robust to the presence of outliers.
where pτ(.) is the absolute value function, providing the τth sample quantile with its solution.
We measure Betas for fluctuating assets (as measured by the Merton Model) across time and across quantiles,
and the corresponding impact of these quantile measurements on DD. Our y axis depicts the asset returns for the
quantile being measured (we measure the 50 percent quantile which corresponds roughly to the standard Merton
model, and the 95 percent quantile to give us our CStdev). The x axis represents the returns for all the asset returns
(all quantiles) in the dataset. The Historical approach is based on the actual historical asset fluctuations. The Monte
Carlo approach uses 20,000 simulated asset returns generated in the same manner as for Xtreme-S.
2.4. Methodology Model 3: Xtreme-T
We base this on CreditMertics transition methodololgy (Gupton, Finger, & Bhatia, 1997). In summary, this model is
based upon obtaining the probability (ρ) of a bank customer transitioning from one grade to another as shown for the
following BBB example:
BBB ρAAA ρAA ρA ρBBB ρBB ρB ρCCC/C ρD
External raters such as Moody’s and Standard & Poor’s (S&P) provide transition probabilities for each grading
and we use the S&P US transition probabilities. .Based on the probability table, Historical VaR is obtained by
calculating the probability weighted portfolio variance and standard deviation (σ), and then calculating Historical
VaR using a normal distribution (for example 1.645σ for a 95 percent confidence level). We extend this VaR
methodology (Gupton et al., 1997) to calculate Historical CVaR by using the lowest 5 percent of ratings for each
industry.
CreditMetrics (see also Allen & Powell, 2009b) use Monte Carlo modelling as an alternate approach to
estimating VaR, and we follow this approach for our Monte Carlo CVaR. Transition probabilities and a normal
distribution assumption are used to calculate asset thresholds (Z) for each rating category as follows:
(3)
Pr(Default)= Φ(ZDef/σ)
Pr(CCC) = Φ(ZCCC/σ) - Φ(ZDef/σ) (5)
and so on, where Φ denotes the cumulative normal distribution, and
ZDef = Φ -1 σ (6)
Scenarios of asset returns are generated using a normal distribution assumption. These returns are mapped to
ratings using the asset thresholds, with a return falling between thresholds corresponding to the rating above it. In
line with this methodology we generate 20,000 returns for each firm from which portfolio distribution and VaR are
calculated. We extend this methodology to calculate Monte Carlo CVaR by obtaining the worst 5 percent of the
20,000 returns.
2.5. Methodology Model 4: Xtreme-i
CreditPortfolioView (Wilson, 1998) adjusts transition probabilities based on industry and country factors calculated
from macroeconomic variables, recognising that customers of equal credit rating may transition differently
depending on their industry risk. However, a study by APRA (1999) showed that banks did not favour using
macroeconomic factors in their modelling due to complexities involved. Our own iTransition model (Allen &
Powell, 2009b) uses the same framework as CreditPortfolioView, but (on the basis that differences in industry risk
will be captured in share prices), incorporates market VaR (fluctuations in the share prices of industries) instead of
macroeconomic variables to derive industry adjustments. This is done by calculating market VaR for each industry,
then calculating the relationship between market VaR and credit risk for each industry, using the Merton model to
224

calculate the credit risk component. We classify data into sectors using Global Industry Codes (GICS), which are
Energy, Materials, Industrials, Consumer Discretionary, Consumer Staples, Financials, Health Care, Retail,
Information Technologies, Telecommunications and Utilities. These factors are used to adjust the Xtreme-T model
as follows using a BBB rated loan example:
BBB ρAAAi ρAAi ρAi ρBBBi ρBBi ρBi ρCCC/Ci ρDi
Other than the industry adjustments, our Xtreme-i Historical and Monte Carlo VaR and CVaR calculations
follow the same process as Xtreme-T.
3. Results and implications for capital
Historical
Model Metric Pre GFC GFC
Xtreme-S DD 8.64 4.07
CDD 2.53 1.26
Xtreme-Q DD 8.10 4.04
CDD 2.31 1.96
Xtreme-T VaR 0.0190 0.0453
CVaR 0.0433 0.0908
Xtreme-i VaR 0.0182 0.0570
CVaR 0.0453 0.1052
Monte Carlo
Model Metric Pre GFC GFC
Xtreme-S DD 8.63 4.06
CDD 3.03 1.08
Xtreme-Q DD 8.09 3.81
CDD 2.81 1.84
Xtreme-T VaR 0.0162 0.0507
CVaR 0.0527 0.0865
Xtreme-i VaR 0.0175 0.0543
CVaR 0.0524 0.0859
Table 1. Results Summary. DD (measured by number of standard deviations) is calculated using equation 1. CDD is based on the worst 5 percent
of asset returns using equation 3. VaR (95 percent confidence level) and CVaR (average of losses beyond VaR) are daily figures and can be
annualised by multiplying by the square root of 250, being the approximate number of annual trading days. The pre-GFC period is the 7 years
from 2000 – 2006 whereas the GFC period is the 3 years from 2007 – 2009.
Table 1 shows large differences between VaR and CVaR, or DD and CDD. For example, all Historical models
show CDD being approximately 3 times higher than CVaR during the pre-GFC period, increasing to approximately
5 times higher for the transition based models (Xtreme-T and Xtreme-i) over the GFC period. The Monte Carlo
models show similar trends to their corresponding Historical models, although Historical VaR and Monte Carlo VaR
are slightly closer than Historical CVaR and Monte Carlo CVaR. The reason VaR is closer is because there are a
large number of Historical VaR observations (95 percent of historical observations) to compare to the extremely
large number of Monte Carlo VaR observations, whereas the Historical model generates only a small number of
CVaR observations (5 percent of historical observations) compared to the large number of Monte Carlo CVaR
observations (5 percent of 20,000 observations), Although there are differences between the models in the extent of
the variation between the quantiles, the difference between VaR and CVaR (or DD and CDD) is nonetheless
significant for all models at the 99 percent level using F tests for changes in volatility. Implications for banks are
225

that provisions and capital calculated on below threshold measurements will clearly not be adequate during periods
of extreme downturn as illustrated in Figure 1.
Q u a n t ile A s s e t V a lu e F lu c t u a t io n s
DD 1.84 β 3.26
σ 0.0258
DD 2.81 β 2.13
σ 0.0168
DD 3.81 β 1.57
σ 0.0124
All Asset Value Fluctuations (10 years)
DD 5.98 β 1.0
σ 0.00789
DD 8.09 β 0.74
σ 0.00584
Figure 1. Illustration of Fluctuating Risk. The figure shows the results of the Quantile Regression (Xtreme-Q) Model for the 50 percent and 95
percent quantiles for pre- GFC and GFC periods. The pre-GFC period is the 7 years from 2000 – 2006 whereas the GFC period is the 3 years
between 2007 – 2009. The y axis is calculated on the asset fluctuations (σ), using the Merton model, for the quantile in question. The x axis is the
median σ for the entire 10 year period. Thus the Beta( β) for the 50 percent Quantile for the 10 year period is one. Where σ for a particular
quantile is less (greater) than the median for the 10 year period, β)1, and DD increases (reduces) accordingly.
As asset value σ is the denominator of the DD equation (equation 1), as σ increases (reduces) from one level to
another (i.e from σ1 to , σ2) DD reduces (increases) by the same proportion. Thus the numerator of the equation (a
measure of capital – the distance between as assets and liabilities) needs to increase to restore DD back to the same
level (i.e., as per the BOE observation in Section 1, capital (K) will need to increase by the same proportion to
restore market confidence in the banks’ capital). Thus the required change in capital (K*) is;
K* = K x σ2 / σ1 (7)
Based on Figure 1, during the extreme fluctuations of the GFC (as measured by the 95 percent quantile) US
banks needed 3 times more capital than during ‘median’ circumstances (as measured by the 10 year median). Whilst
we have used the Xtreme-Q model to illustrate this, the same principle applies to all the models - a trebling of VaR
or DD requires treble capital (100 percent buffer) to deal with it.
When comparing different models, it is important to consider how well their relative outcomes compare to
actual credit risk volatility. Using 10 years of annual data, we correlate our measures for the four models to three
measures of actual credit risk. The first of these measures is Credit Default Swap (CDS) spreads, a measure of the
premium the market is prepared to pay for increased credit risk. The second and third are Delinquent Loans and
Charge-off rates as reported by the US Federal Reserve. These correlations are reported in Table 2. Various lags
were tested, with most correlations being most significant with no lag and some correlations (the shaded areas of
Table 2) being most significant with a 1 year lag (e.g. a 2009 measurement for actual risk compared to a 2008
measurement model). To avoid over-reporting of figures, we show only the results of the Histotrical model (the
Monte Carlo models produce very similar outcomes). The structural based models (Xtreme-S and Xtreme Q) show a
higher correlation with CDS spreads than the other models. This is because CDS spreads change daily with market
226

conditions, as does the asset value component of the structural model. The transition based models (Xtreme-T and
Xtreme-i) which depend on ratings (more sluggish than CDS spreads as ratings are often updated only annually)
show no significant correlation in the same year, but a higher correlation when using a one year lag.
Model Metric CDS Spreads Delinquent Loans Charge-off rates
Xtreme-S DD 0.915 ** 0.786 ** 0.606
CDD 0.906 ** 0.826 ** 0.647 *
Xtreme-Q DD 0.914 ** 0.789 ** 0.608
CDD 0.885 ** 0.865 ** 0.683 *
Xtreme-T VaR 0.573 0.929 ** 0.936 **
CVaR 0.741 * 0.893 ** 0.908 **
Xtreme-i VaR 0.831 ** 0.952 ** 0.926 **
CVaR 0.768 ** 0.920 ** 0.920 **
Table 2. The table correlates the Historical model metrics produced by each of our four models for each of the ten years in our data sample with
three measures of actual credit risk of US banks, being CDS Spreads, Delinquent Loans, and Charge-off rates. Level of significance is measured
by a t-test, with * denoting 95 percent significance and ** denoting 99 percent significance. Non shaded areas are where highest correlation is
experienced with no lag, and the shaded areas with a 1 year lag.
All four models show highly significant (99 percent confidence) correlation with delinquent loans, meaning that
the metrics of all the models are a good indicator of actual defaults. There is very high significance shown by the
transition based models’ correlations with charge-off rates. The timeline in Figure 3 shows how both CDS spreads
and the structural model respond quickly to market events, resulting in high correlation between these two items,
whereas ratings (and thus transition models) react slower to market events and thus have a higher correlation with
actual write-offs which usually occur sometime after initial market deterioration.
Of note is that there is very little difference in the correlation significance levels for VaR (DD) as compared to
CVaR (CDD). This means that, although CVaR (CDD) are at much higher levels than VaR as previously discussed,
the trend (percentage increase or decrease from year to year) is similar for both VaR (DD) and CVaR (CDD) .
4. Conclusions
Figure 2. Timeline and Correlations
This paper has shown how CVaR type metrics can be applied to credit models to measure extreme risk. All four
models show highly significant differences between VaR (DD) and CVaR (CDD) measures. Increased volatility
227

equires capital buffers to deal with the increased risk and the paper demonstrated how this buffer requirement can
be measured. Significant differences were observed between the structural based models (Xtreme-S and Extreme-
Q) as compared to the transition based models (Xtreme-T and Xtreme-i). The changes in risk measured by the
structural models are more consistent with changes experienced in CDS Spreads than those shown by the transition
based models, because both structural models and CDS spreads respond rapidly to market conditions. The opposite
is true of charge-offs where the transition based models show much greater correlation than the structural based
models, as there is generally a delay between defaults and charge-offs, and credit ratings also often respond slower
(often annually) to market conditions than structural models. All models show significant correlation with
delinquent loans.
5. References
Alexander, G. J., & Baptista, A. M. (2003). CVaR as a Measure of Risk: Implications for Portfolio Selection:
Working Paper, School of Management, University of Minnesota.
Alexander, S., Coleman, T. F., & Li, Y. (2003). Derivative Portfolio Hedging Based on CVaR. In G. Szego (Ed.),
New Risk Measures in Investment and Regulation: Wiley.
Allen, D. E., & Powell, R. (2009a). Structural Credit Modelling and its Relationship to Market Value at Risk: An
Australian Sectoral Perspective. In G. N. Gregoriou (Ed.), The VaR Implementation Handbook (pp. 403-414).
New York: McGraw Hill.
Allen, D. E., & Powell, R. (2009b). Transitional Credit Modelling and its Relationship to Market at Value at Risk:
An Australian Sectoral Perspective. Accounting and Finance, 49(3), 425-444.
Andersson, F., Uryasev, S., Mausser, H., & Rosen, D. (2000). Credit Risk Optimization with Conditional Value-at
Risk Criterion. Mathematical Programming, 89(2), 273-291.
Artzner, P., Delbaen, F., Eber, J., & Heath, D. (1999). Coherent Measures of Risk. Mathematical Finance, 9, 203-
228.
Artzner, P., Delbaen, F., Eber, J. M., & Heath, D. (1997). Thinking Coherently. Risk, 10, 68-71.
Australian Prudential Regulation Authority. (1999). Submission to the Basel Committee on Banking Supervision -
Credit Risk Modelling: Current Practices and Applications.
Birbil, S., Frenk, H., Kaynar, B., & Noyan, N. (2009). Risk Measures and Their Applications in Asset Management.
In G. Gregoriou (Ed.), The VaR Implementation Handbook. New York: McGraw-Hill.
Crosbie, P., & Bohn, J. (2003). Modelling Default Risk: Moody's KMV Company.
Federal Reserve Bank. (2010). Federal Reserve Statistical Release: Charge-off and Delinquency Rates.
Gupton, G. M., Finger, C. C., & Bhatia, M. (1997). CreditMetrics - Technical Document. New York: J.P. Morgan &
Co. Incorporated.
Koenker, R., & Bassett, G., Jr. (1978). Regression Quantiles. Econometrica, 46(1), 33-50.
Koenker, R., & Hallock, K. (2001). Quantile Regression. Journal of Economic Perspectives, 15(4), 143 - 156.
Menoncin, F. (2009). Using CVaR to Optimize and Hedge Portfolios. In G. N. Gregoriou (Ed.), The VaR Modeling
Handbook. New York: McGraw Hill.
Moody's Investor Services. (2010a). Moody's Speculative Grade Liquidity Ratings. Retrieved 31/10/10. Available at
www.moody's.com
Moody's Investor Services. (2010b). Research & Ratings. Retrieved 31/10/2010. Available at www.moodys.com
Pflug, G. (2000). Some Remarks on Value-at-Risk and Conditional-Value-at-Risk. In R. Uryasev (Ed.),
Probabilistic Constrained Optimisation: Methodology and Applications. Dordrecht, Boston: Kluwer Academic
Publishers.
Rockafellar, R. T., & Uryasev, S. (2002). Conditional Value-at-Risk for General Loss Distributions. Journal of
Banking and Finance, 26(7), 1443-1471.
Rockafellar, R. T., Uryasev, S., & Zabarankin, M. (2006). Master Funds in Portfolio Analysis with General
Deviation Measures. Journal of Banking and Finance, 30(2), 743-776.
Samanta, P., Azarchs, T., & Hill, N. (2005). Chasing Their Tails: Banks Look Beyond Value-At-Risk. ,
RatingsDirect.
Standard and Poor's. (2009). Default, Transition, and Recovery: Annual Global Corporate Default Study and Rating
Transitions.
Uryasev, S., & Rockafellar, R. T. (2000). Optimisation of Conditional Value-at-Risk. Journal of Risk, 2(3), 21-41.
Wilson, T. C. (1998). Portfolio Credit Risk. Economic Policy Review October, 4(3).
228

229

COMMODITIES MARKETS
230

231

VOLATILITY DYNAMICS IN DUBAI GOLD FUTURES MARKET
Ramzi Nekhili Faculty of Finance and Accounting, University of Wollongong in Dubai, UAE
Michael Thorpe, School of Economics and Finance, Curtin University of Technology, Perth, Australia
Email; m.thorpe@curtin.edu.au
Abstract. This study explores the volatility dynamics of gold futures traded on the Dubai Gold and Commodities Exchange. We test
the effffiects of margin trading reform implemented by the Emirates Securities and Commodities Authority on the dynamic relationship
between the daily gold futures volatility and volume, open interest, and futures returns. We find that volatility dynamics with respect
to volume and return are consistent with other futures markets patterns but not with the open interest, especially after the reform.
Moreover, the reform has decreased trading volume and open interest and increased gold futures volatility.
Keywords: Volatility; Dubai gold futures; Margin trading
JEL classification: C30; G10
1 Introduction
The Dubai Gold and Commodities Exchange (DGCX) was established in 2005 and is the first international online
derivatives market in the Middle East. An electronic trading platform allows members around the world to direct
market assess. DGCX is jointly owned by the Dubai government’s Dubai Multi Commodities Center (DMCC),
Financial Technologies (India) Limited, and the Multi Commodity Exchange of India Limited (MCX). DGCX
trades futures contracts on gold, silver, fuel oil, steel, freight rates, cotton and three major currencies. Futures
options contracts are traded for gold only.
Dubai, the “City of Gold”, has historically been a major trading centre for spot gold, with the Dubai Multi
Commodities Center (DMCC) estimating that in 2006, Dubai’s import and export of gold amounted to 489 and 274
tonnes, respectively. The gold futures contract began trading on November 22, 2005, and the gold options on futures
were introduced on April 30, 2007. The trading volume of the contract has been rising steadily with a total of 71,316
contracts (representing USD
1.5 billion in value) traded in March of 2010 (representing an average of around 4,000 contracts a day). The
contracts are traded on the DGCX’s electronic platform and continuously from Monday through Friday between
8:30 am and 11:30 pm Dubai time, corresponding to 12.30 am to 4:30 pm New York time, 4:30 am to 7:30 pm
London time and 12:30 pm to 3:30 am Singapore time. Hence the operating hours of the market in Dubai overlap
exchanges in other major global centres. The size of the futures contract is 32 troy ounces (1 kg) of 0.995 purity
according to the Dubai Good Delivery Standard. Delivery is made with Dubai Gold Receipt. The contract matures in
bi-monthly intervals, i.e., February, April, June, August, October and December, and the prices are quoted in USD
(per troy ounce).
DGCX is regulated by the Emirates Securities and Commodities Authority (ESCA) which is the regulatory
authority for both the Dubai Financial Market and the Abu Dhabi Securities Market. ESCA has implemented several
regulatory reforms that have impacted the operations of DGCX. These have been aimed at improving the efficiency
of the market, to protect investors from unfair and incorrect practices and to provide regularity and stability for
market trading with a view to ensuring smooth and prompt liquidation of positions. Such reforms will be taken into
consideration to test their effects on the volatility dynamics of the Dubai gold futures market.
In the literature little attention has been paid to emerging markets with most attention paid to the effffiects of
general regulations in the US market (see for example Ma et al., 1993, and Yang et al., 2001). An exception to this
is a study by Chan et al. (2004) which has addressed the futures markets in China. There is a large body of literature
that has looked at the determinants of the volatility of futures prices (see for example Najand and Yung, 1991,
Foster, 1995, and Fung and Patterson, 2001). One area of this research focuses on the volatility of commodity
futures. An early study by Garcia et al. (1986) investigated the impact of lagged volume in five commodity futures
232

contracts on volatility and found significant positive relationships. Bessembinder and Seguin(1993) examined the
link between volatility, volume and open interest of contracts. Their results suggested that trading volume had a
significant positive effffiect on volatility, while open interest had a significant negative effffiect. The study by Chan et
al. (2004) examined the daily volatility of four futures contracts on Chinese futures exchanges and found diffffierent
patterns of volatility under diffffierent government regulatory reforms. Their results for volume and open interest
effffiects are consistent with the literature, with positive and negative relationships respectively. Regulation is also
shown to amplify the effffiects of these factors. The study also reports that both positive and negative returns are
positively related to volatility, with negative returns associated with a more significant impact.
In this paper, we examine the volatility dynamics of Dubai gold futures with respect to changes in variables
such as volume, open interest, and futures returns. The study also seeks to shed light on the impact of margin trading
reform implemented by ESCA on the volatility dynamics of Dubai gold futures. The study will be of practical
benefit for the evolving finance industry in the United Arab Emirates (UAE) and the region generally. Relatively,
little analysis of financial markets in the Gulf region has been undertaken to date and no study has been conducted of
Dubai futures markets. It is expected that this investigation will provide a platform for further on-going research in
other derivatives markets which have been recently established in the UAE. Moreover, the relevance of this study
stems from the importance policymakers and regulators place on improving the efffficiency of financial markets and
from the need for market participants to improve their understanding of emerging future markets.
The rest of the paper proceeds as follows. Section 2 presents the framework within which we conduct our
empirical estimation. Section 3 describes our data and presents our results and Section 4 checks the robustness and
consistency of our results. We conclude in Section 5.
2 Methodology
To investigate the dynamics of Dubai gold futures volatility, we first measure the daily volatility of futures prices
using two approaches of extreme-value method such as Parkinson (1980) and Rogers and Satchell (1991). The
Parkinson measure uses daily high and low futures prices and Rogers-Satchell measure incorporates daily opening
and closing futures prices in addition to Parkinson’s instruments. Respectively, they go as follows:
2
1 � High � t
V � log( ) , (1)
t � �
4 log 2 � Lowt
�
� High High � � Low Low �
t t t t
V � log( ) log( ) log( ) log( ) (2
t � � � � �
)
� Open Close Open Close
t t � � t t �
Next, we examine how gold futures volatility relates to volume, open interest, and positive and negative returns.
We envisage using lagged volume as an indicator of flow of information, to avoid simultaneity relationships with
volatility, and open interest as an indicator of market depth. Chan et al. (2004) use open interest as level of hedging
activities that could mitigate futures volatility. In addition, to test whether there is evidence of asymmetric effffiects of
returns on volatility, we include the positive and negative returns in the volatility specification. The regression
specification is as follows:
V � � �� X � � � X � e , (3)
4
t 0 1 1t �1
it it t
i�2
where X1t−1 is the log trading volume of the futures contract at time t − 1, X2t is the log open interest, X3t is
the positive future returns at time t equivalent to max[0, Rt]; and X4t is the negative future returns at time t
equivalent to min[0, Rt], with Rt being the daily futures returns measured as the logarithmic diffffierence between two
consecutive futures prices.
Specification (3) tests a number of hypotheses. First, we can see whether the effffiect of volume on volatility is
positive, α1 > 0. Second, we test the market depth effffiect on volatility, α2 < 0. Finally, we test whether good news
and bad news have effffiects on volatility by checking respectively the coefffficient signs as α3 > 0and α4 < 0.
We also directly addresses the regulatory reform concerning margin trading undertaken by the Emirates
233

Securities and Commodities Authority during the study period, since the inception of the futures contract. We
conduct our analysis over the entire sample period as well as over two sub-periods that are pre and post reform. The
reform concerns changes in the regulations on margin trading that took effffiect in June 2008. Among many other
decisions, ESCA has set an initial margin of not less than 50% of the market value of the securities traded on
margin, as well as a maintenance margin of not less than 25% of the same traded market securities. In addition,
DGCX imposes an extra margin call on all open positions when volatility is high.
Our proposition is that the margin trading reform reduces the trading volume and open interest and has an
impact on gold futures price volatility as through changes in market liquidity and depth. We expect that the volatility
dynamics represented by Eq.(3) will display diffffierent results before and after the reform. Such proposition has been
highlighted by Tesler (1981) who showed that an increase in cost of trading may lower the volume and open interest
and hence liquidity, which may, in turn, increases future price volatility.
The regression technique adopted for the analysis is the generalized method of moments (GMM) of Hansen
(1982). This approach has been widely used in the literature to study the determinants of futures volatility. For a
recent example, see Holmes and Tomsett (2004). This technique addresses the issue of time-varying conditional
heteroskedasticity as well as the presence of any unconditional distributional properties. It also handles
contemporaneous relationships between the variables of interest and provides autocorrelation consistent estimates.
In considering both heteroskedasticity and autocorrelation, the Newey-West (1987) method for selecting the
bandwidth is employed. We also take instruments from the independent variables and test whether specification (3)
is exact using the J-statistic to test for overidentifying restrictions. Finally, both volatility measures are used to check
the robustness of the results.
3 Data and Results
The data consists of daily data of gold futures contracts traded on the Dubai Gold and Commodities Exchange.
These contracts are the most active ones in Dubai. The sample period covers the contracts traded from May 2007 till
June 2010. The data is collected from the DGCX and includes daily high future price, daily low price, opening price,
closing price, trading volume and opening interest.
Table 1 provides descriptive statistics for the full sample and the two sub-periods. A preliminary investigation
of the raw data reveals that the average daily return is positive as seen by the diffffierence between the mean of
positive returns and the absolute mean of negative returns. Post-reform trading volume and open interest have
decreased comparing to pre-reform figures. The volatility, measured by both Parkinson and Rogers and Satchell, has
increased in magnitude (around 0.01%) after the margin trading reform. This finding goes in line with Tesler’s
(1981) argument that an increase in margin requirements may serve to increase futures price volatility if market
liquidity is reduced.
Table 2 displays the GMM estimation results using both Parkinson (Panel A) and Rogers-Satchell (Panel B)
volatility measures. Over the full period of the study the results indicate a negative relation between negative futures
returns and volatility and a positive relation between positive returns and volatility. This is in line with the findings
in the literature. There is an asymmetric effffiect observed, however, when considering the significance of the
coefffficients of positive and negative returns. The absolute magnitude of positive returns is higher than the negative
returns, which does not conform to the evidence from developed markets, but is similar to the results found for the
Chinese market by Chan et al. (2004). Unexpectedly, the opening interest coefffficient is significant with a positive
sign but this effffiect may be due to influences due to the selection of the full period, masking sub-period effffiects.
Finally, and in line with the expected results from the literature, the volume of trades is positively related to gold
futures volatility. These results are similar for both measures of volatility. Nevertheless, the adjusted R 2
is higher
using the Parkinson volatility measure which provides support for this approach in modelling the daily volatility of
gold futures in this study. Moreover, the J-statistic is low enough to reject the hypothesis of overidentifying
restrictions imposed in the regression.
In looking at the two sub-periods, the asymmetric effffiect of returns is again seen to be significant with respect to
volatility. As observed for the full period, the magnitude in absolute terms is higher for the positive returns than for
234

the negative returns. It seems that market participants react more to good news than bad news. The effffiect of volume
on volatility, although positive, becomes insignificant after the reform. This is most likely due to the fact that the
flow of information was improved and that speculation by day traders had lost momentum.
Open interest is seen to be negatively related to volatility pre-reform, as expected, but exhibits a positive
relationship post-reform. It would appear that the level of hedging activity mitigated the volatility during the earlier
period, but that the regulation on margin trading in 2008 resulted in the positive relationship between open interest
and gold futures price volatility. This signal of a decrease in the market depth is also equivalent to a decrease in
liquidity and consequently has increased price volatility. Given that DGCX imposes an extra margin call on all open
positions in time of high volatility, the resultant increase in cost seems to make hedgers not hold their positions for
long, with the result that their investments become speculative positions rather than hedge positions. This in turn
could not contribute to stabilizing Dubai gold futures prices.
Overall, it appears that gold futures volatility has increased significantly due to the implementation of the
margin trading regulatory reform. Market dynamics with respect to volume and return are consistent with other
futures markets patterns but not with the open interest, especially after the reform took effffiect. This can be a feature
of an emerging future market such as the one of Dubai.
4 Additional Robustness Checks
Having used unconditional volatility estimates, we could think of presenting the results with conditional volatility
such as the GARCH-type to see how the basic investigation of the analysis could be altered. We first assume that the
returns follow a martingale with drift and GARCH(1,1) volatility specification, then we extract the conditional
GARCH-type volatility and run a regression on the variables of interest. The following model highlights this
specification:
2
R � � � u , and u � iid(
0,
� ) (4)
t t t t
2 2 2
� t � a � b� t�1 � cut�1.
The assumption of i.i.d. innovations is almost certain to be violated but may not limit the purposes of the
analysis. Nevertheless, and being aware of the non-normality of the innovations, we assume Student-t distribution of
the return innovations. Table 3, Panel A, displays the estimation results and shows significant ARCH and GARCH
effffiects. This tells that there is volatility persistence in the Gold Furures returns indicating that large volatility
increases do last at least the following day. Panel B of Table 3 shows the results of the estimation of Eq. (3) using
the conditional volatility extracted from specification (4). The results confirm the inferences obtained from the
analysis conducted with the unconditional volatilities with one exception that is the significant negative relationship
between the lagged trading volume and the volatility.
Furthermore, we undertake an additional robustness check by separating out specific gold futures contracts
traded within the study period to highlight the effffiect of margin trading switching regime. To make sure accounting
for the margin trading reform effffiect, we choose the contracts that are maturing in August 2008 and August 2009.
The starting trading dates of these contracts are, respectively, August 8, 2007, and August 8, 2009. In the DGCX,
the last trading day is the business day six days prior to delivery; therefore the last trading days for the contracts are
respectively, July 31, 2008, and July 31, 2009.
Table 5 displays the GMM estimation results using only Parkinson measure of volatility. In every future
contract, we find that the asymmetries in the return effect on volatility are statistically significant. We also find no
effect of open interest on the volatility dynamics in August 2009 contract, similar to previous post-reform results. In
addition, and consistent with the previous results, the patterns found in the volatility with respect to volume, open
interest, and returns are similar to the ones found using the complete time series.
235

5 Conclusion
This paper investigates factors influencing the volatility of the gold futures contracts traded on the Dubai Gold and
Commodities Exchange (DGCX). The study looks at the period May 2007 to June 2010. The effffiect on market
volatility of the margin trading reform introduced in June 2008 is also considered.
In line with expectations, the volume of trading, which can be considered a proxy for speculative market
activity, is observed to be positively linked to volatility. The effffiect of open interest, a measure of market depth or
hedging activity, is shown to vary over the two sub-periods considered. Pre-reform, the results indicate a negative
relationship with volatility in line with expected findings. However, a positive relationship is evident in sub-period
after the reform. The results also suggest that the regulation of margin trading has the effffiect of raising market
volatility.
Overall, the study also found, in line with the literature, that there was an asymmetric effffiect of returns on
volatility. Negative returns were associated with lower volatility while positive returns were positively related to
volatility. However, an unexpected result was that positive returns appear to have a greater impact on volatility than
negative returns. There appears to be more reaction to good news than bad.
Future research avenues can be addressed such as testing the predictive power and information content of gold
future volatility relative to other measures such as option implied volatility in explaining the future realized
volatility.
6 References
Bessembinder, H. & Seguin, P.J., 1993. Price Volatility, Trading Volume, and Market Depth: Evidence from
Futures Markets. Journal of Financial and Quantitative Analysis, 28, 21-39.
Chan, K.C., Fung, H.G., & Leung, W.K., 2004. Daily volatility behavior in Chinese futures markets. Journal of
International Financial Markets, Institutions, and Money, 14, 491-505.
Foster, A.J., 1995. Volume-volatility relationships for crude oil futures markets. Journal of Futures Markets, 15,
929-951.
Fung, H.G. & Patterson, G.A., 2001. Volatility, global information, and market conditions: a study in futures
markets. Journal of Futures Markets, 21, 173-196.
Garcia, P., Leuthold, R.M., & Zapata, H., 1986. Lead-lag relationships between trading volume and price variability:
new evidence. Journal of Futures Markets, 6, 1-10.
Holmes, P. & Tomsett, M., 2004. Information and noise in U.K. futures markets. Journal of Futures Markets, 24,
711-731.
Ma, C.K., Wenchi Kao, G., & Frohlich, C.J., 1993. Margin requirements and the behavior of silver futures prices.
Journal of Business Finance and Accounting, 20 (1), 41-60.
Najand, M. & Yung, K., 1991. A GARCH examination of the relationship between volume and price variability in
futures markets. Journal of Futures Markets, 11, 613-621.
Newey, W. & West, K., 1987. A simple positive semi-definite heteroskedasticity and autocorrelation consistent
covariance matrix. Econometrica, 55, 703-708.
Parkinson, M., 1980. The Extreme Value Method for Estimating the Variance of the Rate of Return. Journal of
Business, 53, 61-65.
Rogers, L.C.G. & Satchell, S.E., 1991. Estimating variance from high, low, and closing prices. Annals of Applied
Probability, 1, 500-512.
Tesler, L.g., 1981. Margins and futures contracts. Journal of Futures Markets, 1, 225-253.
Yang, J., Bessler, D.A., & Leatham, D.J., 2001. Asset storability and price discovery of commodity futures markets:
a new look. Journal of Futures Markets, 21, 279-300.
236

237

238

239

240

STOCK RETURNS AND OIL PRICE BASED TRADING
Michael Soucek, European University Viadrina,Germany
Email: soucek@europa-uni.de,
Abstract. While the linkage between the energy prices and stock prices has been widely investigated, the literature discussing
practical implications of this relationship is rather scarce. The aim of this paper is to prove, if the relationship between these two
markets is economically exploitable. As a proxy for the equity returns I use MSCI reinvestment index for five developed countries as
well as MSCI emerging markets index. For oil prices the Brent type oil is considered. Since the analyzed return series are stationary
and Granger causality between oil and stock returns is observable, trading strategy based on the bivariate VAR(p) model will be
employed. The trading rule provides significant abnormal returns measured by Sharpe ratio and Jensen’s alpha for weekly and
monthly data over the last twenty years for developed markets. The standard deviation of the oil based trading strategy is significantly
lower than that for the buy and hold strategy. The results are robust to sub-period analysis and variation in the in-sample period. Using
the information from the oil market to predict stock returns, the results proved that the markets does not fulfill the requirements of
semi-strong market efficiency in the sense of Jensen (1978).
Keywords: Stock Returns, Oil Prices, VAR, Market Efficiency, Trading strategy
JEL classification: G11, G14, Q40
1 Introduction
Oil prices and their effect on the stock markets is an object of ongoing research in academic literature. From the
asset pricing point of view, identifying what drives equity prices is the key topic. According to the theory, the value
of a firm equals the discounted sum of the expected future cash flows. The cash flows themselves are assumed to be
influenced by a company's relative economic condition. Possible drivers of equity prices are for instance, interest
rates, exchange rates, income, revenue or production costs and diverse macroeconomic factors such as GDP or
prices of resources. Because oil is one of the important inputs for almost all kinds of industries, the hypothesis about
the link between the oil and stock market seems reasonable. The aim of this paper is to draw possible implications
for market participants. Most studies on the relationship between oil and stock prices conclude by making a
statement about the connection and causality between the time series data. While the majority of recent research
studies agree on the existence of a link between oil price and stock price (Driesprong, Jacobsen, & Maat, 2008;
Sadorsky, 1999; Park & Ratti 2008; Basher & Sadorsky 2006; Nadha & Faff 2008), it is still possible that even
though the results are statistically significant, they are not economically exploitable. This paper introduces a trading
strategy based on the underlying VAR( ) model containing oil returns as a predictor of future stock returns. It could
be shown, that for monthly and weekly returns, a trading technique based on a bivariate VAR( ) model can provide
investors with abnormal returns and has comparably favorable risk-return characteristics relative to the buy-and-hold
policy. This trading strategy outperforms the market return measured by Jensen's alpha and Sharpe ratio, and
exhibits significantly lower standard deviation of its returns when compared to the market index. The results hold
especially for developed markets. The emerging market index does not seem to be strongly affected by the oil price
changes. To confirm the validity of our results, the study contains a necessary robustness check of the trading rule
itself. This test has been frequently omitted in the literature.
The results of the study provide not only valuable information for investors, hedgers and other market
participants but also contribute to the ongoing research on market efficiency. Fama (1970) provides the textbook
definition of an efficient market, which has been broadly cited in the literature: "A market in which prices always
fully reflect available information is called efficient." A more detailed definition has been developed by Jensen
(1976): A market is efficient with respect to an information set, if it is impossible to make economic profits by
trading on the basis of this information set. Because economic profits are defined as risk-adjusted returns after
deducting transaction costs, Jensen's definition implies that the efficient market hypothesis (EMH) can be tested by
considering the net profits and risk of trading strategies based on available information. Jensen introduced three
types of market efficiency: Weak form efficiency, in which the information set is limited to the information
contained in the past price history of the market; Semi-strong form efficiency, where the information set is all
information that is publicly available at certain time; and strong form efficiency, where the information set is all
public and private information available at certain time. A good survey of the results of a technical analysis used for
241

testing the efficiency market hypothesis, as well as a discussion on the theoretical concepts related to technical
trading is provided by Park (2007). In this paper, information from the oil market was used to trade equity, and the
obtained results support the hypothesis that the developed markets does not fulfill the requirements of a semi-strong
efficient market. This fact could not be proved for the emerging markets index. The paper is structured as follows. In
section 2, the empirical methodology for investigating the relationship between the series as well as the dataset is
introduced. Section 2 also features the design and implementation of a trading technique based on the bivariate
VAR( ) model. The out-of-sample results are described in section 3. Section 4 concludes the paper.
2 Model and Data
2.1 Model
Based on Sims’s findings (Sims, 1980) the relationship between two stationary time series can be empirically
investigated by the vector autoregressive model. In fact, this approach has often been used in academic literature to
evaluate the relationship between stock and oil prices. In the most basic bivariate case, a VAR consists of two
equations and can be written as follows:
t
P
1 � �
P
1p
yt
� p � ��1
p xt
� p
p=
1
p=
1
y = c � � �
(1)
t
P
P
2 � � 1p
xt
� p � ��
1p
yt
� p
p=
1
p=
1
x = c � � �
(2)
Where c1 and c1 denote the constants and et is white noise error term. yt and xt denote the observations of two
stationary time series. p represents the used lag length. In the following study, one of the equations describes the oil
price dynamics, while the other represents the analyzed equity index movements. One of the advantages of using a
VAR is that in doing so, it is not necessary to provide prior assumptions about which variables are exogenous and
which are endogenous. In this model, all variables are treated as endogenous. Each variable depends on the lagged
values of all other variables in the system. Because the model is sensitive to the choice lag-length, the appropriate
number of lags is usually estimated using the Akaike Information Criterion (AIC) or Schwarz Bayesian Criterion
(SBC). In the case of non-stationary time series, which are integrated of order one, i.e. (I(1)), the VAR in first
differences can be estimated, and conventional asymptotic theory can be used for hypothesis testing.
2.2 Data
This study employs monthly data from January 1990 to March 2011. The Morgan Stanley Capital International
(MSCI) European reinvestment index has been used to track the development of the equity market. I consider MSCI
indices for the USA, European Union, Germany, France, Canada and emerging markets MSCI. The log returns have
been calculated as a difference in log prices. I consider monthly prediction intervals calculated at the end of the
month and weekly prediction intervals calculated on Wednesday of each week, to avoid the impact of higher trading
volumes on Monday and Friday. The one-month Eurodollar London Interbank Offering Rate (LIBOR) was used for
the approximation of the risk free interest rate. The Brent type crude oil in US dollar is used as a proxy for the oil
price. All prices has been converted to US dollar.
2.3 Unit Root Test
The first step is to investigate the stationarity of the observed univariate time series, the existence of which is the key
assumption of the OLS based VAR( ) estimation. I adopt the Augmented Dickey-Fuller (ADF) test for unit roots to
evaluate the property. For all time series, I run following regression:
� �
p=
1
P
t
t
yt = � � �t
� �yt
�1 � � �yt
� p 1 � � t ,
(3)
242

where the optimal lag p is based on the AIC. The null hypothesis of the ADF-test is g = 0. If the variable is
integrated of order one yt-1, it provides no relevant information for predicting yt outside of that already obtained in
Dyt-p. In such a case, the null hypothesis cannot be rejected; i.e., the time series y has a unit root. The t-values are
compared with the Dickey-Fuller (DF) t-values and summarised in table 1. As the results indicate, all log price
series are integrated of order one, which implies that the log return time series are stationary. The result has been
verified using Phillips-Perron univariate test and is in line with the general accepted fact that the stock return time
series are stationary.
2.4 Granger causality
Analysis of the estimated coefficients themselves sometimes reveals uninteresting results. Bivariate VAR models
are in most cases used to test Granger causality and to determine whether lags of one variable help to explain the
current value of some other variable. The results provide information about the dynamics of the data. Because the
appropriate model for the relationship between log prices of stocks and oil seems to be VAR( ) in first differences,
which is nothing but VAR( ) in a stationary log return series, we can analyze Granger causality between the returns
of Brent oil and analyzed equity indices. The null hypothesis of the test is that the variable does not Granger cause
variable x and vice versa. To verify causality, I estimate unrestricted equation (1) and equation (2) under the
restriction and then use the F-Statistics to test for the equality of both models. If the explanatory power
differs, it could be concluded that Granger causes . Because relationship is not symmetric, Granger causality in
one direction does not imply causality in the other. The test results for the causality relationships are summarized in
table 1. The column 3 and 4 in table 1contain the p-values of the Granger causality test. Additionally, the data have
also been tested for instantaneous causality. If, in period , adding to the information set improves the forecast
of , there is instantaneous causality between the two variable (Lütkepohl, 2005). Because this concept of
causality is symmetric, we speak about instantaneous causality between and , and deem the direction
irrelevant. Last column summarizes the p-values for the test for instantaneous causality. The obtained statistically
significant link for the stock indices is also in line with common findings.
MSCI Index Lag ADF Oil -> Market Market -> Oil Inst., Causality
Weekly Returns
United States 7 -12.447 0.01091 0.04831 0.0778
European Union 3 -16.169 0.02116 0.00951 0.0001
Gerrnany 3 -16.661 0.01026 0.01449 0.0309
France 3 -16.565 0.01301 0.00638 0.0079
Canada 8 -9.506 0.29070 0.00004 0.0001
Emerging Markets 8 -9.172 0.01650 0.00001 0.0090
Monthly Returns
United States 1 -10.878 0.0775 0.3208 0.3766
European Union 4 -6.2122 0.2194 0.0143 0.3270
Gerrnany 1 -11.555 0.3450 0.0390 0.8450
France 3 -7.028 0.3460 0.0740 0.3340
Canada 1 -10.052 0.3430 0.9178 0.0001
Emerging Markets 1 -9.348 0.0825 0.1399 0.0169
2.5 Trading Strategy
Table 1: ADF and Granger causality
Studies on the link between oil prices and stock markets typically end with a statement concerning the causal
relationship between the two time series, concentrating on the strength, direction and significance of the
relationship. Literature on the implications for investor decisions or portfolio diversification is rather scarce. Geman
& Kharoubi (2008), for instance, discuss a copula approach for optimal maturity choice via diversification with oil
futures, and in addition provide a valuable survey of recent literature on the topic. Arouri & Nquyen (2010) shows
that diversification with oil spot price-based products can improve the risk-return profile of stock portfolios. While
the majority of recent research studies agree about the existence of the link between oil price and stock price, it is
possible that while the results are statistically significant, they are not economically exploitable. Existing literature
243

pays little attention to this fact. In a recent paper (Driesprong, Jacobsen & Maat, 2008) has been shown, that a
simple trading rule based on one-month lagged oil returns yields a better risk-return profile than a buy-and-hold
strategy, and shows that the economic significance of the oil price shocks (Hong, Torous & Valkanov, 2007) is
stronger than for other possible predictors, such as interest rates or dividend yields. This paper expands upon the
idea of exploiting the relationship between the two markets and shows that a bivariate-VAR-based trading rule
strongly decreases the risk of an investment and yields abnormal returns for the investor in the case of developed
markets. The results are robust and valid for weekly and monthly returns over the last twenty years. Furthermore, the
paper extends the literature on market efficiency based on Jensen (1976). The market is efficient with respect to an
information set if, it is impossible to make economic profits by trading on the basis of this information set. The oil
price changes belong to what is classified as publicly available information and, from this point of view, places this
study as a test of the semi-strong market efficiency hypothesis. The trading strategy is to exploit one step ahead
forecasts of the stock returns based on the movement in the oil spot prices. To obtain the forecasts bivariate rolling
VAR(p) model was employed:
�
P�1
�
P�1
�
p=
0
p=
0
~
y = ˆ ˆ
ˆ
t 1 c1
� � p yt
� p � � p xt
� p .
(4)
yt-p stands for the index stock returns and xt-p describes past changes in the prices of Brent type oil.
~
yt�1 is the one
period ahead predicted stock index return. Parameters �ˆ and �ˆ are estimated from the 48 month rolling in-sample
window. Optimal lag length p has been estimated for every single rolling window using AIC.The trading rule is
designed to generate a signal for the investor to invest in the equity market or LIBOR bonds each month. If the
forecasted equity market return
~
yt�1 is greater than the current risk-free rate, it generates a signal to invest in the
equity market; otherwise, it signals that the investor should invest in bonds. Similar trading rules have been used in
various studies on technical trading (Park, 2007). This strategy is easily implemented using index futures or another
instrument tracking the index such as an exchange traded fund (ETF). Transaction costs are assumed to be 0.1%
which is consistent with the literature on technical trading. The returns were compared with the returns acquired by
holding the market index.
To evaluate the results and to show that they are not spurious, diverse tests and robustness checks have been
implemented to measure the applicability of the trading strategy. I discuss the risk and return characteristics of
strategy using Sharpe ratio and standard deviation comparison. Furthermore, CAPM-based Jenson's alpha was
estimated running following regression:
OS
t
Index
f = i t f t
r � r � � � ( r � r ) � � ,
(5)
Where rt OS describes the return of the oil strategy in time t and rt Index the return of the corresponding market index;
this is nothing but the return of the of the buy-and-hold strategy. The estimate bi is the systematic risk of the trading
strategy compared to the market return and ai is the Jenson's alpha. To account for possible heteroscedasticity and
autocorrelation (HAC) in the explanatory variables, the test statistics are calculated using HAC standard errors.
Descriptive statistics for the trading rule and the results of the Jensen’s alpha regression are summarized in the table
2 for weekly data, and the results obtained for monthly observations are provided in table 3.
3 Summary
3.1 Results
For all analyzed indices for developed markets, VAR-based trading strategies generate higher mean returns and
smaller standard deviations than the buy-and-hold policy. Using the F-Test for a comparing of variances implies that
the variance of the buy and hold strategy is significantly larger that of the trading rule. Remarkably large Sharpe
ratios also support the hypothesis of a more favorable risk-return relationship of the oil-based strategy compared to
the buy-and-hold policy. Tables 2 and 3 describe more precisely the economic significance of the trading rule. The
tables provide comparism of descriptive statistics of the buy-and-hold (BH) and oil based trading strategy. Mean is
244

given in % per month. SD stands for standard deviation, SR for Sharpe ratio in %. The table also contain the
estimates of the Jenson's alpha regression, as well as the tests for reasonable trading signals. The t-values are
calcultated using (HAC) standard errors. Jenson's alpha is also given in % per anum. For analyzed indices, the
Jensen's alpha is significantly positive. Implementation of this strategy for the USA MSCI equity index would yield
on average risk adjusted 6.4% per anum more than the buy-and-hold strategy. Furthermore, beta coefficients, which
measure the systematic risk of the trading strategy, are significantly smaller than one, confirming that the oil strategy
is substantially less risky than investing in the market index. A good trading strategy should provide reasonable
signals for going in or out of the stock market. If the rule implies holding the market index related product, one
should expect significant positive returns of the market index (H1). A good trading strategy should be capable of
indicating the bullish market. On the contrary, if the rule implies the optimal step is to invest in bonds, one might
expect a downward trend in the market index and negative returns on the equity market. If the return of the market
index in this period is significantly positive, the trading technique is missing potential sources of additional gains. In
this sense, one might expect the market returns to be less than or equal to zero in the periods following the trading
strategy's sell signal (H2). Columns H1 and H2 in tables 2 and 3 contain the t-values for coresponding tests. I
observe that for all indices, the trading strategy yields positive returns after a buy signal, and after a sell signal, the
market is bearish but not significant for the majority of indices. It is generally remarkable, that even if we unable to
prove Granger causality for all analyzed pairs, especially for monthly observations, the trading technique provides
good results for all equity indices. A possible explanation for this is that while the causality itself was tested for the
whole sample, only the relationship in the short in-sample is relevant for the trading rule. The profitability of the
trading strategy indicates that the power of causality relationships could vary over different subsamples. The
exploitability of the trading rule gives further evidence of the significance of the relationship.
MSCI Index Mean BH SD BH SR BH Mean SD SR
United States 0.118 2.411 1.972 0.231 1.852 8.658
European Union 0.096 2.819 0.877 0.236 1.911 8.661
Gerrnany 0.107 3.367 1.085 0.288 2.388 9.110
France 0.105 3.167 1.083 0.199 1.083 5.581
Canada 0.192 3.021 4.027 0.285 2.220 9.647
Emerging Markets 0.091 3.192 0. 622 0. 066 2.064 -0.229
H1 H1 � t(� ) � t( � )
United States 3.193 -1.535 0.132 2.993 0.594 11.258
European Union 3.203 -1.569 0.154 2.969 0.462 8.162
Gerrnany 3.235 -1.905 0.199 3.288 0.507 8.811
France 2.200 -0.870 0.110 2.060 0.527 9.494
Canada 3.565 -1.040 0.148 2.575 0.542 7.967
Emerging Markets -0.068 0.573 -0.013 -0.216 0.418 4.745
3.2 Robustnes checks
Table 2: Weekly Returns Trading Strategy
The trading decision is based on the comparison of the forecast generated by VAR(p) and risk-free rate represented
by LIBOR. It is theoretically possible that the signal exhibits predictive power because of expected return estimates
or because LIBOR itself has significant predictive power for the future market development. In the latter case, it
would not be possible to conclude that the oil price based strategy really outperforms the market. Even if it is
essential to analyze the rule when evaluating the results of the trading technique, verifying the validity of the trading
rule has been frequently omitted in the literature. In this study, to prove whether or not there is predictability power
of interest rates on the stock returns, I run a simple regression model where for the sample period from 1990 to
2011:
I
t
r = � � �
(6)
i � � i LIBORt
�1
where ai is a constant and ei is the error term for the sample period from 1990 to 2011. For instance for the US stock
market HAC standard errors based t-statistics for monthly data is -0.178 and for weekly returns and weekly risk free
rates -0.091. This implies a lack of significant predictive power of the US term structure. The regression results are
245
t

insignificant for analyzed indices. To conclude, this trading technique is suitable for exploiting the oil price-based
stock return forecasts. Furthermore, to check for the robustness of the empirical results, the additional following
changes were made to the underlying strategy. First, I divided the dataset into two sub-samples such that both
samples have ten year out-of-sample periods (i.e., 1990-2004 and 1996 and 2011). In both cases, the trading rule
provided higher returns and a significant Jenson's alpha. Refining the in-sample period to three or five years
similarly leads to no change in the results.
MSCI Index Mean SD SR Mean SD SR
United States 0.516 4.593 4.490 0.834 3.085 17.010
European Union 0.414 5.387 1.934 0.769 3.363 13.670
Gerrnany 0.472 6.808 2.384 0.827 3.418 15.127
France 0.455 6.031 2.419 0.490 4.497 7.536
Canada 0.819 6.255 8.146 0.638 4.581 7.172
Emerging Markets 0.374 7.264 0.886 0.399 4.894 1.822
Buy Sell � t(� ) � t( � )
United States 3.494 -0.965 0.430 2.518 0.455 5.013
European Union 2.712 -0.740 0.418 2.244 0.398 3.854
Gerrnany 2.735 -0.407 0.475 2.250 0.254 4.208
France 1.612 -0.171 0.258 1.296 0.554 5.506
Canada 1.699 0.937 0.057 0.201 0.531 4.530
Emerging Markets 0.689 0.372 0.059 0.216 0.455 3.974
4 Conclusion
Table 3: Monthly Returns Trading Strategy
This paper contributes to the literature on the oil and stock price co-movements. The impact of the oil price changes
is generally significant but varies for different indices. I observed strong relationship for developed countries.
Studies typically end at this stage of analysis with a statement about significance, power and direction of the
relationship between the commodity and stock time series. This paper extends the literature illustrating how to
exploit the obtained information for trading strategies. The trading rule based on the simple bivariate VAR( ) for
forecasting future stock returns significantly outperforms the buy-and-hold strategy in term of expected return and
risk. The systematic risk of the trading strategy is smaller than one and yields large Sharpe ratios. The standard
deviation of the strategy is significantly lower than the standard deviation of the market return, a fact which supports
the favorable properties of the strategy in comparison to the buy-and-hold policy. The oil-based trading technique
provides an additional significant positive Jenson's alpha, as well as significant positive returns in the periods after a
buy signal for all indices. The emerging market index does not seem to react on the changes in oil price and is not
economically exploitable with a VAR based trading rule. The results are robust to variation in the out-of-sample and
in-sample periods. Additionally, the robustness of the trading technique construction has been verified, which is a
necessary step usually omitted in the literature. In addition to the obvious implication for the market participants, the
results contribute to the discussion on the market efficiency. From the EMH point of view, these results prove that
developed stock markets are not efficient (Fama, 1970). Because publicly available information from other markets
were employed, the semi-strong market efficiency hypothesis (Jensen, 1976) can be rejected.
5 References
Basher, S. A. & Sadorsky P. (2006). Oil price risk and emerging stock markets. Global Finance Journal, 17(2), 224-
251.
Driesprong, G., Jacobsen, B. & Maat, B. (2008). Striking oil: Another puzzle? Journal of Financial Economics,
89(2), 307-327.
Fama, E. (1970). Efficient capital markets: a review of theory and empirical work. Journal of Finance, 25(2), 383-
417.
246

Geman, H. & Kharoubi C., (2008). WTI crude oil futures in portfolio diversification: The time-to-maturity effect.
Journal of Banking & Finance, 32(12), 2553-2559.
Hong, H., Torous, W. & Valkanov, R., (2007). Do industries lead stock markets? Journal of Financial Economics,
83(2), 367-396.
Jensen, M. (1976). Some anomalous evidence regarding market efficiency. Journal of Financial Economics, 6(2),
95-101.
Jones, C. & Kaul G. (1996). Oil and the stock markets. Journal of Finance, 51(2), 463-491.
Lütkepohl, H. (2005). New Introduction to Multiple Time Series Analysis. Berlin: Springer Verlag.
Nandha, M. & Faff R. (2008). Does oil move equity prices? a global view. Energy Economics, 30(3), 986-997.
Newey, W. K. & West K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation
consistent covariance matrix. Econometrica, 55(1), 703--708.
Park, C.-H. & Irwin S. H. (2007). What do we know about the profitability of technical analysis? Journal of
Economic Surveys, 21(4), 786-826.
Park, J. & R. Ratti (2008). Oil price shocks and stock markets in the U.S. and 13 European countries. Energy
Economics, 30(5), 2587-2608.
Sims, C. A. (1980). Macroeconomics and reality. Econometrica, 48(1), 1-48.
Sadorsky, P. (1999). Oil price shocks and stock market activity. Energy Economics, 21(5), 449-469.
Sadorsky, P. (2001). Risk factors in stock returns of Canadian oil and gas companies. Energy Economics, 23(1), 17-
28.
247

OPTIMAL LEVERAGE AND STOP LOSS POLICIES FOR FUTURES INVESTMENTS
Rainer A. Schüssler, Westphalian Wilhelms-University of Munster, Germany
Email: 05rasc@wiwi.uni-muenster.de
Abstract. Our paper presents a framework to assess the portfolio-wide effects of simultaneously applying stop loss limits and
leverage to dynamic mechanical investment strategies in futures markets. We systematically separate the management of portfoliowide
risk and the assumed risk for individual assets. We show the steps needed to formulate and solve a resulting finite horizon
Markov decision problem. The impact of optimally interfacing stop loss rules and leverage is explored using daily data from 1995 to
2010 across 34 different commodity futures markets. Empirical results indicate advantageous wealth distributions for different futures
investment strategies and risk preferences.
Keywords: Risk Management, Dynamic Stochastic Programming, Commodity Futures Trading, Stop loss, Leverage
JEL classification: C6, G11
1 Introduction
We introduce a general framework to assess the portfolio-wide effects of simultaneously applying stop loss limits
and leverage to mechanical dynamic investment strategies in futures markets. We systematically separate the
management of portfolio-wide risk and the risk for individual assets. We show how to formulate the investment
problem as a finite horizon Markov decision problem. The key concept underlying our analysis is the decoupling of
risk management on the portfolio level and the risk taken for individual positions. We consider the underlying
investment strategy as entirely exogenous, that is, portfolio constituents, their direction of exposure (long or short)
and their respective portfolio weights are determined by a mechanical dynamic selection rule for portfolio for
portfolio allocation. Our objective is to provide a method to systematically scan the optimal kind of implementation
for a given strategy.
We apply the method to investment strategies in commodity futures markets for the time from 1995 until 2010.
There are several reasons why we chose commodity futures: first, recent empirical studies in this field show that
active investment strategies for commodity futures based on price-based signals would have been rewarding in the
past. Price-based signals, such as momentum of returns and the shape of the term structure, appear to be able to
detect time-varying risk premia in commodity futures markets (Gorton & Rouwenhorst, 2007; Basu & Miffre,
2009). Secondly, commodity futures markets are an excellent platform for active investment strategies for several
reasons: commodity futures markets are subject to rather small transaction costs ranging from 0.0004% to 0.033%
(Locke & Venkantesh, 1997). Besides, short positions can be taken in commodity futures as straightforwardly as
long positions. Further, commodity futures markets are highly liquid, at least for the nearby and second nearby
contract, and offer the possibility to enhance returns by high leverage levels. The most important reason to apply the
framework to commodity futures data are the stylized facts of commodity futures prices: one stylized fact of
commodity futures prices is their trending behavior characterized by sharp spikes for many markets. This finding
gives rise to the conjecture that there is some time-series dependence for the returns of individual commodity
futures, which can be exploited by applying stop loss rules to individual futures investments. The low correlations
between returns of different (sectors of) commodity futures is a further stylized fact. Therefore, a huge amount of
idiosyncratic risk can be diversified and risk is reduced on the portfolio level.
We suggest exploiting temporary trends of individual futures by increasing leverage. At the same time, portfolio
risk is limited by dynamically optimizing the superposition policy (i.e. the joint decision about stop loss and
leverage level) that is to be applied to individual futures for the next period. The choice of the superposition strategy
depends on the investor’s utility function, current wealth, remaining periods until planning horizon, and the expected
return distribution for available investment opportunities. Our approach differs from the conventional use of stop
loss policies. Stop losses are usually applied to determine when to exit a risky investment in favor of a riskless
alternative investment. In our setting, we also use stop loss limits to determine conditions to close individual
investment positions. In contrast to common use, we do not employ stop loss rules to determine when the overall
strategy is to be replaced by a riskless investment. Stop loss rules for an overall investment strategy are expected to
248

e rewarding only if there is substantial time-series dependence present of the overall strategy returns. However, for
many settings, it is not obvious why past performance of an overall strategy should indicate future performance,
particularly in the case of low data frequencies. Rather, we treat risk management for futures investment strategies
on an overall level analogously to a dynamic asset allocation problem and the risk management of a futures
investment strategy consists in the different decision variables: in a (basic) dynamic asset allocation problem, an
investor decides on the weights of risky and a riskless asset. In contrast, for risk management, the decision variables
are superposition policies that are applied to the basic investment strategies and thus generate new return profiles. In
our empirical application, we will explore if our approach succeeds in improving an investor’s utility by applying
superposition rules to basic investment strategies identified by mechanical trading rules.
The rest of the paper is organized as follows. Section 2 establishes the general framework for analyzing the
portfolio-wide impact of our superposition policies on mechanical investment strategies in futures markets. Chapter
3 provides empirical results for the framework applied to investment strategies in commodity futures markets.
Section 4 concludes.
2 General framework
2.1 Selection rules for basic futures investment strategies
Consider a futures investment opportunity set � : � { 1,...,
S}.
A generic mechanical trading rule, that maps
signals � for the S futures into a vector of portfolio weights g* : � � w is applied at equally spaced points in
time belonging to a finite set � � { 0,...,
t �1}
. Decisions are indexed with the set � � { 1,...,
J}
. That is, the jth
t � . The length of each trading period is � t . At each review period t � � ,
decision, j � � , is made at time �
j
a mechanical rule g * �� �t
� identifies the market exposure (long, neutral or short) of the futures of the opportunity
set � . The portfolio constituents are chosen along with the vector of respective portfolio weights. g * �� � �
exploits a vector of mechanically generated signals � based on information available up to period t , that is the
information set � t . The portfolio is held up until the next review period for trading, � � � � � t 1 j . Applying the
dynamic decision rule, the resulting set of futures positions is denoted basic strategy (BS). Long and short positions
are denominated active positions and are indexed by the set � : � { 1,...,
M } � � . �t is partitioned into a set of
D equally spaced points in time � : � { 1,...,
D}.
The step size between two subsequent points in time is � � .
Without loss of generality, assume that t represents months and � denotes days. Hence, one month is partitioned
into t� � t ... t .
1 � �
2 � Before we proceed with a description of the considered stop loss and leverage policies,
D
some assumptions are necessary.
Assumption 1
g * � � is strictly mechanical, that is, no discretionary decisions are taken and thus, the possibility for
Rule � t �
backtesting strategies that are produced applying g �� � �
Assumption 2
* is ensured.
The returns on the basic strategy, generated by application of g �� � �
t
* , are i.i.d at monthly data frequency.
Assumption 3
Each investment opportunity can be bought or sold at the designated amount, i.e., there are no divisibility problems.
249
t
j
t

2.2 Superposition policies
2.2.1 Stop loss policies
We restrict the set of stop loss rules to trailing stops that automatically adjust upwards and trigger after cumulative
: � � � � � � proceeds as follows: Close
losses exceed a critical value. The rule for creating the set � � min � � max ��
,...,
t , t
position m at time t� �1
for the remaining time of the current trading interval � � �1
D �
performance for position m falls short of a fixed threshold � relative to the maximum within the period �t , t �
if the cumulated
since inception of investment position m at the previous review period t . Thus, � is a set of rules for the entire
range of considered stop loss limits, � � ���
�� � � �� ��.
2.2.2 Leverage policies
min
,..., max
The leverage level is a matter of how much risk an investor wants to take. Futures trading requires a relatively small
amount of margin and an investor is not very constrained by the amount of initial capital committed to trading. If the
leverage level is one contracts are fully collateralized and the investor takes an unlevered position, e.g., a futures
contract with value $100,000 is backed by a $100,000 cash deposit. For a leverage level of 2, futures contracts are
collateralized only with half the futures contract value. And a leverage of 0.5 means that futures contracts are
collateralized with twice the underlying contract value. We consider the discrete set
leverage levels.
L : � { lmin
,..., lmax
} of
2.3 Dynamic programming formulation
2.3.1 Action space
The stop loss policy and the choice of leverage have to be determined simultaneously. After applying a
superposition policy, i.e. a combination of stop loss limits and leverage level to the basic strategy , the resulting
strategy is labeled portfolio strategy. An action set �: � { 1,...,
A}
for superposition policies that can be applied to
the basic strategy is generated as Cartesian product of the sets � and L , that is � � L � ��� , l�
| � � � � l � L�
addition, a set of alternative investments, { ret }
AI : � rf , with the single element rf
: . In
ret as risk-free investment
opportunity with constant guaranteed return, is considered. Thus, � consists of | � | � | L | �1
elements. To keep
computational complexity at a manageable level we have to restrict the number of feasible superposition strategies.
F
Therefore, the feasible action space � , with � � �
F
F
, is equipped with a subset of � . Pruning � to � , both
the number of feasible actions and the feasible actions themselves have to be determined. While the number is
mainly an issue of computer capacity, the actions that are chosen to be decision variables within the dynamic
programming have to be identified according to problem-specific economic criteria. This issue will be discussed in
more detail in chapter 3. Backtesting the (hypothetical) returns of the whole set of portfolio strategies, with data
F
available up to the current period, provides decision support for the choice of � . Returns of the portfolio strategy,
that is, (feasible) superposition policies are applied to the basic strategy (BS),
SP BS F F
F F
ret : � ret �� � L : � ��� , l�
| � � � � l � L ��.
Hence, the feasible action set comprises : ret AI.
SP
F
� � �
All referred returns are defined as discrete returns. For commodity futures markets we refer to discrete excess
returns.
2.3.2 State space
Wealth W constitutes the single state variable in the problem formulation. Thus, an adequate description for the
stochastic evolution of { , t � 0}
is to be identified. The transition equation equation for wealth is represented as
W t
F �W a , �
�1 , �1
� Wt g t t �t in the most general notation. Hence, the wealth in the following period is a function of
250
1
�
,

current wealth W t , the action
according to �
t�1
F
t
a �
F
� , and t�1
SP F
W � � ret �a , ��
t � 1 t�1
t t�1
� representing randomness. More concretely, wealth evolves
W � . The returns of the portfolio strategy for the next period,
SP
F
rett� 1 , depend on a t , that is the simultaneous choice of stop loss rule and leverage level. Of course, these
returns are random. To be more precise about � , we resort to a resampling scheme to simulate the data generating
process for
SP
ret . The procedure is applied in order to avoid errors due to possible misspecifications using
parametric estimations for the stochastic process for
SP
ret . It is crucially important to notice that the resampled
monthly historical returns are (value-weighted) overall portfolio returns. Therefore, it does not matter which
g * � � at any review period. Instead, the
commodity futures had been chosen by the mechanical selection rule � �
individual portfolio constituents are regarded as entirely interchangeable and returns of the portfolio strategy are
g * � � . Hence, the consequences of
defined as the outcome of applying the mechanical trading rule � t �
Assumption 1 and Assumption 2 become evident: returns generated applying g �� � �
t
* have to be i.i.d on a
monthly data frequency to carry out the analysis with wealth as single state variable because, for this case, there is
no time-series dependence of returns on the portfolio level. The portfolio-level view constitutes an enormous
simplification for our analysis because time series characteristics for returns of individual futures (and correlations
with other futures) become irrelevant.
2.3.3 Value Function
The terminal value function at the end of the planning horizon T is stated as
V
T
�W �
T
�
�
: � �
�
�W
T
1��
WT
1�
�
� � � ( protection _ level �W
T
, W
T
, W
T
�
�
protection _ level
protection _ level
The perceived risk by the investor is directly linked to wealth. Above a specified protection level, the terminal value
function is of the CRRA type and the parameter � � 0 determines the level of risk aversion. Below the protection
� � protection _ level �W
, � � 0 to control for
level, missing the target is penalized by the linear term � �
downside risk. � and � for all levels of wealth have to be chosen such that
1��
WT
1�
�
� WT
� � � ( protection _ level �WT
) . Our modeling choice to account for downside risk to account
for downside risk is an alternative formulation of a chance-constrained investment problem. The capital should be
invested so that a minimum return is exceeded with a probability to be specified, say 95% or 99%, and the expected
return should be as large as possible. We will explore empirically whether downside risk protection can be achieved
by utility maximization using the linear penalty term for a given value for � .
function can be stated according to the Bellman equation:
For periods t � � , the value
2.3.4 Backward recursion
V
t
F
�W � f ( W , a ) � E [ V �W � | W ]}
t
max{ F F
at
��t
t t
t t�1
t�1
t
� � .
The dynamic optimization problem can be solved using backward dynamic recursion, conditioning on the level of
wealth. According to the specified state transition for wealth, we obtain for t � T �1,...,
1
251
T
t

V
F
SP F
�W � � f ( W , a ) � � E [ V ( W � ( 1�
ret ( a , � ))) | W ]} .
t
t
max{ F F
at
��t
t t
t t�1
t
t�1
t t�1
t
SP
ret t
*,
For each t � � , two different sets of monthly returns,
, are resampled: one in-sample set
in
St of size
in
| St | and one out-of-sample set
out
St out
of size | St in in
| and � � St
out out
and � � St
. Both the in-sample and
the out-of-sample sets are generated on the basis of the same historical daily portfolio returns. Thus, in-sample and
out-of sample refer to different (due to resampling) monthly returns from the same pool of daily returns. The
sample S includes the in-sample returns used for generating the single-period utility maximization problems and
in
t
the sample S entails the monthly out-of-sample returns employed for evaluating the obtained solution. Using two
out
t
different samples, one for optimizing and the other for evaluating, prevents from optimization bias. In the following,
* *
�
*
V � refers to the out-of-sample estimate of the value function.
V � � refers to the in-sample estimate, whereas � �
t
t
i
WT 1
in
T
Starting at period T �1,
wealth is parameterized into I discrete wealth levels � . We solve the problem in
period T �1
for each level of wealth, i.e. for each specified grid point, using sample S � 1 , and obtain the optimal
F*
superposition policies, aT � 1 . The optimal superposition policy tells us which combination of leverage level and stop
loss of the feasible action space maximizes the expected utility of the next month. The obtained solutions are
evaluated using Monte Carlo samples with the out-of-sample returns to estimate the expected utility of the single-
i
period utility maximization problem of each period. For each level of wealth, WT � 1 , a corresponding value for the
i
out-of-sample value function, T � T � W V *
�1 �1
is obtained. The value function of period T �1
is the induced value
function for the T � 2 single-period optimization, and the procedure is repeated until all optimizations in period 1
are done. Finally, in period 0, the initial wealth is known and the final optimization using the period 1 value
function as implied utility function is conducted. As a final result, we receive a sequence of optimal policies
* *, F , i i *, F , i i
depending on each possible state for each time period (and level of wealth), � � { a0 �W 0 �,..., aT
�1
�W T �1
�}. In
each period of the backward recursion, a different independent sample of large size is used for evaluation to avoid
* *
V � would suffer from an optimization
optimization bias. Depending on the sample size, the in-sample estimate t � �
bias that would be carried forward between stages, whereas the out-of-sample estimate
* ��� V of the portfolio
decision represents an independent evaluation without any optimization bias. Thus, the sequential optimizationevaluation
procedure is an important part of the solution algorithm. We simulate a large number of paths of the
controlled state variable (wealth) to evaluate out-of-sample paths for the evolution of wealth from the inception of
the investment in period 0 until the end of the planning horizon in period T . In the following section, we apply our
method for investment strategies in commodity futures markets and discuss the results.
3 Empirical application
3.1 Mechanical selection rules for commodity futures
Gorton & Rouwenhorst (2006) introduce an active commodity futures strategy that seeks to exploit the term
structure of commodity futures prices by taking long positions in backwardated contracts and short positions in
contangoed ones. They exploit the term-structure signals of 12 commodities and implement a simple long-short
strategy that buys the 6 most backwardated commodities and shorts the 6 most contangoed commodities. Miffre &
Rallis (2007) employ momentum signals and tactically allocate wealth towards the best performing commodities
and away from the worst performing ones. At the end of each month, they sort futures contracts into quintiles based
on their average return over the previous R months (ranking period). The futures contracts in each quintile are
equally weighted. The performance of both the top (winner) and bottom (loser) quintiles is monitored over the
subsequent H months (holding period). The resulting R-H momentum strategy buys the winner portfolio, shorts the
loser portfolio and holds the long-short position for H months earning considerable excess returns. Shen, Szakmary
& Sharma (2010) provide further empirical evidence for abnormal returns of momentum strategies in commodity
252
t

futures markets for intermediate holding periods. Fuertes, Miffre & Rallis (2010) provide an empirical study
combining both momentum and term structure signals. They proceed as follows: first, they compute the degree of
backwardation and contango at the end of each month and the 1/3 breakpoints to split the cross section of futures
contracts into 3 portfolios, labeled low, medium and high. They then sort the commodities in the high portfolio into
2 sub-portfolios (high-winner and high-loser) based on the mean return of the commodities over the past R months.
In effect, the high-winner and high-loser portfolios contain 50% of the cross-section that is selected with the first
term-structure sort or 50% �33. 3%
of the initial cross-section that is available at the end of a given month.
Intuitively, high-winner is thus made of the commodities that have both the highest degree of backwardation at the
time of portfolio construction and the best past performance. Similarly, they sort the commodities in the low
portfolio into 2 sub-portfolios (low-winner and low-loser) based on their mean return over the past R months. Lowloser
contains therefore commodities that have both the highest contango at the time of portfolio construction and
the worst past performance. The combined strategy buys the high-winner portfolio, shorts the low-loser portfolio
and holds this position for one month. The choice of one-month holding period (H=1) and monthly rebalancing is
due to the fact that the momentum strategies with H=1 and R=1 proved to be the most profitable ones in Miffre &
Rallis (2007). This finding is supported by the empirical study of Shen, Szakmary & Sharma (2007). The doublesort
strategies could generate returns (net of transaction costs) clearly above the passive benchmark and also
performed much better risk-adjusted. The robustness analysis of Shen et al. (2007) suggests that the superior profits
of the double-sort strategies are not an artifact of lack of liquidity, transaction costs or data mining. Also, the success
of their approach proves to be robust to alternative choices for ranking and holding periods and different
specifications of the risk-return relationship. Despite these encouraging results, the returns of the active strategy
turned out to be substantially more volatile and exhibited more downside risk. Bottom line, the term structure of
commodity prices and momentum signals seem to have been valuable tools for allocation across individual
commodity futures in the past.
We consider four basic mechanical investment strategies based on term structure or momentum signals. For all
strategies we apply one month as lookback period and one month as holding period. The choice of a one month
investment period and a one month lookback period is arbitrary, as are all investment and lookback periods for
momentum strategies. Strategy (a) High goes long one third of all current portfolio constituents whose term structure
is most backwardated. Strategies (b)-(d) also select long or short positions for one third of the number of current
portfolio constituents. Strategy (b) High-Winner double-sorts commodity futures according to the degree of
backwardation and momentum and takes long positions for the chosen commodity futures. Strategy (c) Winner is
only based on a momentum signal and takes long positions in the futures with the highest futures return in the
previous month. Strategy (d) High-Winner-Low-Loser goes long in the same manner the High-Winner (as in
strategy (b), but only half of them), the other part consists of shorted Low-Losers which are ranked on contango and
lowest previous returns. They are expected to underperform, therefore they are shorted. For each strategy, all
included commodity futures are weighted equally
3.2 Parameterization
To determine the action space � we have to define a suitable range of stop loss limits and leverage levels
which are to be applied to the returns of the basic strategies (a) -(d) to obtain the portfolio strategy. The set for stop
: � � � � � � is scanned for the range between � 0 and 20 . 0 � . For the extreme
� min max ,...,
loss limits � � � ��
case of min 0 � � , investment position m is closed at the next trading day for the remaining time of the trading period,
that is the end of month in our setting if the performance index (starting at value 1 at the first day of the month) falls
short 0% in reference to the maximum value of the performance index since beginning of the month. Thus, a stop
loss threshold � min � 0 triggers closing of position m if the performance index does not outperform the maximum
reached so far in the current month. Using the stop loss limit max 0.
20 � � at the other end of the spectrum, accepts a
20% maximum drawdown before liquidation of the position is triggered. Note that the suffered loss can exceed the
specified stop loss limit γ. Consider, for example, γ=10%, and the performance index is 1.01 at trading day δ, while
the maximum value of the index equals 1.10 for this month. Thus, the position is not closed at the following trading
day δ+1 because 1.01/1.10-1=-0.0818>-0.10. At the next trading day, δ+2, the index drops to 0.96. At this, point
closing the future is triggered but the investor has already suffered 0.96/1.10-1=-0.1273

We scan the grid for stepwidth 0.01 from � min � 0 to � max � 0.
20 , given a certain leverage level. For the set of
attainable leverage levels, l �{
0.
5,
1,
2,
3}
is considered. The action set� thus comprises |Ψ|×|L|+1 possible actions,
that is |Ψ|×|L| superposition policies and the risk free investment opportunity. To keep computation feasible, the
actions considered so far are divided into feasible und unfeasible actions. This procedure is done for each possible
level of wealth, l �{
0.
5,
1,
2,
3}
.To proceed, we backtest the suggested superposition policies with strategies (a)-(d)
and backtest results are used to preselect potential superpolicies.. Table 1 shows the superposition policies that are
chosen to be inputs, that is decision variables, for the dynamic programming setup. Thus, the feasible action space
F
� consists of 13 elements presenting a computationally tractable dimension.
SP1
SP4
SP7
SP10
Leverage: 3
Stop: not applied
Leverage: 2
Stop: not applied
Leverage: 1
Stop: not applied
Leverage: 0.5
Stop: not applied
SP2
SP5
SP8
SP11
Leverage: 3
Stop: 12.5%
Leverage: 2
Stop: 7.5%
Leverage: 1
Stop: 5.0%
Leverage: 0.5
Stop: 5.0%
SP3
SP6
SP9
SP12
Leveragev:3
Stop: 7.5%
Leverage:2
Stop: 5.0%
Leverage:1
Stop: 2.5%
Leverage:0.5
Stop: 2.5%
Table 1: The table shows the feasible combinations of stop loss limits and leverage levels. Note that, for simplicity, we do not select different
feasible superposition strategies for each of the considered basic investment strategies. Instead, three different stop loss limits are chosen. Two
stop loss limits are selected in the range for a high ratio of average annual geometric return to the average annual standard deviation of returns,
the third alternative is that no stop loss mechanism is applied. These pre-selected superposition strategies serve as the feasible actions in the
dynamic programming setting.
3.3 Results
We evaluate the four investment strategies we proposed in the previous section with all associated superposition
policies. The discrete-time finite-horizon MDP is solved for a planning horizon of T=12 months and monthly review
periods. Starting with an initial wealth W 0 =$100,000, we set the protection level at $95,000. We use a step size of
250 and an attainable range between $0 and $400,000. Thus the number of grid points equals 1601 and W ∈
{$0,...$,400,000}. Such a dynamic programming framework requires a maximum level of wealth to be considered.
All other levels of wealth exceeding this one are assumed to have the same utility. This causes a conservative bias as
this level is approached. We assume that the amount of starting capital W₀ is $100,000, optimization results become
unreliable as we approach $400,000 because exceeding this amount is not of any additional use in the model. The
problem is aggravated the more periods are still to go. Therefore we have to be cautious interpreting optimization
results depending on the starting capital and the time horizon when the state of wealth is above a certain region.
However, a maximum value of $400,000 means wealth would have to quadruplicate within a year. Thus, our choice
is not very restrictive. An analogous problem arises when wealth approaches $0. We set the risk aversion parameters
to � � 0.
01for
the upper part and � � 100 for wealth levels below the protection level. To quantify how much upside
potential we give up in favor of the downside protection we contrast our results with the parameter combination
� � 0 and � � 0 (i.e. an entirely-risk neutral investor). In order to get evidence of the added value of the
superposition strategies, we compare the resulting distributions when all superposition policies are considered, that
is { 1,...,
A}
a F � with � 13.
F
A The base case only allows to select unleveraged positions without stop loss, i.e. the
controls a �{
7,
10,
13}
are available. Superposition 7 is Leverage 1 and no stop loss, superposition 10 is Leverage
0.5 and no stop loss, 13 represents the risk free rate with annualized return of 3%. Tables 2-5 report the results for
each of the four basic investment strategies after applying the superposition policies. 99% (95%) means the level of
wealth that is exceeded in 99% (95%) of (5,000) out-of-sample simulation runs.
254

High-Strategy
99% Wealth 95% Wealth Expected Wealth
Superposition
� � 0 , � � 0
$64,500 $80,500 $138,410
� � 100 , � � 0.
01
$ 94,750 $95,500 $119,950
Base Case
� � 0 , � � 0
$72,500 $81,500 $107,180
� � 100 , � � 0.
01
$96,250 $97,750 $103,640
Table 2: Out-of-sample simulation results for High-Strategy
High-Winner-Strategy
99% Wealth 95% Wealth Expected Wealth
Superposition
� � 0 , � � 0
$78,500 $92,750 $150,260
� � 100 , � � 0.
01
$ 94,250 $95,250 $136,440
Base Case
� � 0 , � � 0
$70,500 $81,250 $113,660
� � 100 , � � 0.
01
$95,250 $96,750 $106,790
Table 3: Out-of-sample simulation results for High-Winner-Strategy
Winner-Strategy
99% Wealth 95% Wealth Expected Wealth
Superposition
� � 0 , � � 0
$68,500 $82,500 $135,690
� � 100 , � � 0.
01
$ 94,750 $95,500 $121,510
Base Case
� � 0 , � � 0
$81,375 $89,250 $116,660
� � 100 , � � 0.
01
$94,750 $96,250 $110,640
Table 4: Out-of-sample simulation results for Winner-strategy
High-Winner-Low-Loser-Strategy
99% Wealth 95% Wealth Expected Wealth
Superposition
� � 0 , � � 0
$70,250 $81,500 $117,190
� � 100 , � � 0.
01
$ 94,875 $96,500 $107,380
Base Case
� � 0 , � � 0
$75,000 $82,250 $106,690
� � 100 , � � 0.
01
$96,250 $98,250 $103,750
Table 5: Out-of-sample simulation results for High-Winner-Low-Loser-Strategy
255

The results show strong evidence in favor of the superposition policies. For each of the investment strategies, we
receive a considerable increase for the expected wealth, both for the risk controlled case and for the risk-neutral
case. As expected, we find that downside protection is not for free. Protecting the downside, we give up a fairly high
portion of upside chance.
4 Conclusion
To the best of our knowledge, the impact of applying stop-loss rules and leverage for futures investments has not yet
been subject to analysis. Our framework can be thought of as a kind of meta-strategy for determining optimal risk
policies for futures investment strategies. Decisions are identified for both individual contracts and the portfoliowide
risk management. The suggested procedure is designed to capture temporary trends by employing leverage, to
stop losses by making use of time-series dependence to limit risk and to diversify idiosyncratic risk of individual
futures contracts. Although our framework is rather simple, it allows for flexible utility functions and any return
distributions. The portfolio-level view for risk management facilitates analysis and allows considering an arbitrarily
large investment opportunity set. Empirical results for investment strategies for commodity futures markets are
encouraging and show the possible value of the framework. In particular, the proposed framework limits the
downside risk. Extending the framework or applying more elaborated trading signals are promising areas of future
research.
5 References
Basu, D. & Miffre, J. (2009). Capturing the risk premium of commodity futures. Working paper, EDHEC Business
School.
Erb, C., & Harvey, C. (2006). The strategic and tactical value of commodity futures. Financial Analysts Journal 62,
69-97.
Fuertes, A.-M., Miffre, J. & Rallis, G. (2010). Tactical Allocation in Commodity Futures Markets: Combining
Momentum and Term Structure Signals. Journal of Banking and Finance 34, 2530-2548.
Gorton, G. & Rouwenhorst, K. (2006). Facts and fantasies about commodity futures. Financial Analysts Journal 62,
86-93.
Gorton, G., Hayashi F. & Rouwenhorst, K.-G. (2007). The Fundamentals of Commodity Futures Returns. Yale ICF
Working Paper No. 07-08.
Locke, P. & Venkatesh, P. (1997). Futures market transaction costs. Journal of Futures Markets 17, 229-245.
Miffre, J. & Rallis, G. (2007). Momentum strategies in commodity futures markets. Journal of Banking and Finance
31, 1863-1886.
Kaminski, K.M. & Lo, A.W. (2008). When do stop- loss rules stop-loss? Working paper, Swedish Institute for
Financial Research.
Schneeweis, T., Kazemi, H. & Spurgin, R. (2008). Momentum in Asset Returns: Are Commodity Returns a Special
Case? The Journal of Alternative Investments 10, 23-46.
Pirrong, C. (2005). Momentum in futures markets. Working paper. University of Houston.
Shen, Q., Szakmary, A. & Sharma, S. (2007). An examination of momentum strategies in commodity futures
markets. Journal of Futures Markets 27, 227-249.
Shen, Q., Szakmary, A. & Sharma, S. (2010). Trend-following trading strategies in commodity futures: A reexamination.
Journal of Banking and Finance 34, 409-426.
256

THE IMPACT OF INTERNATIONAL MARKET EFFECTS AND PURE POLITICAL RISK ON THE UK,
EMU AND USA OIL AND GAS STOCK MARKET SECTORS
John Simpson, School of Economics and Finance, Curtin University, Perth, Western Australia.
Email: simpsonj@cbs.curtin.edu.au www.cbs.curtin.edu.au
Abstract. A key question in European integration is whether or not the global oil and gas market sector (over the period which
includes the oil price hikes of 2001 to late 2007 and the global financial crisis from mid 2008 to early 2009) has been impacted to a
greater or lesser extent by major Western country oil and gas stock market sectors. An examination is undertaken of data from those
sectors in the USA, the EMU and the UK as well as political, social and legal data (embodied in political risk ratings) for the USA, the
UK and the key economies of the EMU (that is, France, Germany and Italy). This short study uses analysis of both unlagged and
lagged data in multivariate models to test these relationships and finds a closer relationship politically and financially between the
USA and the UK than between the EMU and the UK markets. The EMU and the major EMU countries considered separately in oil
and gas market sector movements and political risk ratings movements have very little financial and political influence on the
movement of the global oil and gas market sector over the period.
Keywords: oil and gas, market sectors, country risk, political risk, integration, USA, UK, EMU, multivariate models.
JEL Classification: F15, F59
1. Introduction
According to theory and the Law of One Price as revisited by Asche et al, (2000), in an integrated market, prices on
homogenous goods from different producers and suppliers should move together. Price differentials should only
indicate differences in transportation costs and quality. These price differentials for energy should flow through to
oil and gas stock markets sectors and be reflected in oil and gas share returns. They should also be reflective of the
degree of oil and gas stock market integration of the major Western stock markets.
There is a connection between market integration and political risk in energy markets. There is also a
relationship between energy prices, political risk and prices in energy stock market sectors. For example, Asche et al
(2000) find that cointegration tests showed different border prices for gas to Germany moved proportionally over
time. This indicated integration of the German gas market. They also studied whether or not there were large price
differences between gas from Norwegian, Dutch and Russian exporters. They find differences in mean prices
(Russian gas was cheaper) and the reasons for the price differences were ascribed to differences in volume flexibility
and perceived political risk. The question in the case of EMU countries is whether or not energy market integration
in the leading EMU countries energy sectors and political risk changes in EMU countries have flowed over to global
oil and gas stock market sector integration.
It is logical that oil and gas listed companies need to be combined as a market sector. There is a proven strong
connection between oil prices and gas prices and this translates to a need to include energy companies in one
sectorial stock price index. Most researchers agree that the price of oil has something to do with the price of gas (For
example, Okugu, 2002; Mazighi, 2005 and Eng, 2006). Some also agree that other energy prices, such as coal are
related to oil and gas prices (For example, Bachmeier, 2006; Pindyck, 1999). If this is the case in North America as
well as in Continental Europe, then an interim global gas pricing model remains possible, with oil and coal prices
partial forecasters of gas prices. This direction of energy markets should also be indicated in the degree of
integration of American and European oil and gas stock market sectors.
The issues are in respect to the integration of global oil and gas stock market sectors 1. Are the factors driving
the major Western oil and gas stock market sectors over the period of study mainly political or economic and
financial (that is, stock market related) in nature? 2. Is the EMU oil and gas market a greater political and legal force
impacting the main global oil and gas markets over the period of study compared to the UK? 3. Is the UK oil and
gas stock market sector more economically and politically integrated with the US and global markets than it is with
the EMU? In respect to the last issue it needs to be remembered that whilst the UK is a member of the European
Union it has not fully committed to membership of the EMU.
257

A definition of political risk is required for this study. Political risk is the willingness of countries to service
their external commitments (For example, Bourke & Shanmugam, 1990; Cantor & Packer, 1996). This is influenced
by human, cultural, social and legal factors that provide a subjective quantification of influences such as, the degree
of corruption, the history of law and order and the quality of the bureaucracy (ICRG, 2009). It is suggested that
political risk ratings in Western economies have become more volatile over the period since “9/11” terrorist attacks,
the Iraqi war, the corporate governance issues in the USA related to for example, Worldcom and Enron, and more
recently the finance and banking governance issues that led to the global financial crisis in the USA, the UK and the
major Western European countries (For an indication of the volatility in political risk ratings over the full period of
the study, see Appendix 2).
Political risk ratings are used by financial economists as a management tool for assessing economic, financial
and political riskiness in doing business with different countries either at a macro or a microeconomic level (For
example they assist banks lending international to ascribe credit risk premia to arrive at a market interest rate).
Political risk ratings are part of country or sovereign risk ratings, with the other two components of economic risk
and financial risk remaining a measure of a country’s ability to meet external obligations.
Country stock market indices, sectorial or otherwise, have proven to be reliable indicators of economic and
financial conditions. Major Western stock markets sectors in oil and gas have also be quite volatile over the period
from 2001 to 2008 for similar reasons relating to oil prices, US corporate governance issues and the global financial
crisis. Oil and gas market data also reflect global economic and financial conditions and may also be impacted by
political risk factors due to the importance of oil and gas as an essential commodity and because the OPEC cartel is
comprised of developing countries with high levels of country and political risk.
2. Data
Global and country oil and gas market monthly indexed data are collected from the DataStream database covering
the period June 2001 to December 2008 for the world, the USA the UK and the EMU. Over a similar period
monthly political risk ratings are gathered from the ICRG (2009) for the UK, The USA and the major EMU
countries in France, Germany and Italy.
The world oil and gas stock market price index values are reported by Datastream who base their series on stock
exchange oil and gas price indices that commonly use a representative sample of publicly listed oil and gas related
companies in each country, with the stock prices reflected in the index converted into US Dollars at current
exchange rates. The companies included in the index generally represent around 85% of the volumes traded in the
country oil and gas markets. The index is regularly re-assessed (at least every quarter) to identify changes in the
trading volumes of each represented company share. Then a new portfolio is compiled, with new weightings based
on the changes in trading activity in each share.
The companies represented in the index commonly represent around 70% of the total oil and gas stock market
capitalisation of listed companies in each market. The indices generally reflect information that has been updated
daily for the morning following the reference day and may be regarded as an important global economic indicator,
reflective in part of global supply and demand conditions.
3. The Model
Changes in price indices and political risk ratings are studied in a single period (lags excluded) ordinary least
squares regression format as follows:
�P
OGW
T
��7(
�SR
Where:
��
OGW
t
GERMt
��1(
�P
OGUK
t
) ��8(
�SR
) ��2(
�P
ITALt
) �e
OGW
t
OGUSA
t
) ��3(
�P
OGEMU
t
) ��4(
�SR
USAt
) ��5(
�SR
) ��6(
�SR
FRAt
.......... .......... .......... .......... .......... .......... .......... .......... .......... .......... .......... ... 1)
258
UKt
)

�POG 's
Represent the changes in oil and gas price index values for the world (the dependent variable), the UK, the
USA and the EMU indices respectively.
�SR' s represent the changes in political risk ratings for the USA, the UK, France, Germany and Italy respectively.
� is the regression intercept for the world oil and gas regression at time t.
OGEMU
t
� ’s represent the regression coefficients for each of the above independent variables.
It is also useful, in respect to issue number 3 to provide a basic study on UK political risk relationships with the
USA and the key EMU countries. Another single period ordinary least squared multiple linear regression model is
tested as follows.
�SR � � � �1(
�SR
� � 2(
�SR
) � � ( �SR
) � e
UKt
4. Findings
UKt
FRAt
GERM t
USAt
UKt
.......... ... 2)
An indication of the volatility in level series oil and gas stock market sectors for the world, the USA, the UK, and
the EMU is provided in Appendix 1. The study moves to first differences to remove serial correlation problems in
the regression errors. The findings of this study of Equation 1) in first differences (price changes and political risk
ratings changes) are reported in Table 1.
Regression statistics Value
Adjusted R Square 0.9187
Durbin Watson test statistic 2.2092
t-statistic UK oil and gas market index changes 13.9038
t-statistic US oil and gas market index changes 19.8146
t-statistic political risk changes UK -2.4572*
Note: Variable are significant at the 1% level except for *, which is significant at the 5% level.
Table 1: Regression results for the world oil and gas market index changes
The results show that there is not a problem with serial correlation in this regression as the Durbin Watson
(DW) test statistic is significantly greater than 2. The results of the model may be relied upon. The explanatory
power of the model is strong with an adjusted R Square value of 0.9187 (91.87%). The t-statistics show that the
EMU oil and gas market does not possess a significant relationship with global oil and gas markets, but the US
market and then the UK market are significant parts of the explanatory power of the model (t statistics at 19.8146
and 13.9038, which are significant at the 1% level).
The t-statistics also indicate that political risk changes in the UK are an explanatory variable in this model
(where the t-statistic at -2.4572 is significant at the 5% level), but its contribution to the explanatory power is
substantially less than the changes in the oil and gas market indices in the UK and the USA. Changes in political risk
ratings in Germany, France and Italy are not significant explanatory variables.
Logically there is a positive relationship between the UK and the USA oil and gas price changes and those of
the world oil and gas market. As returns increase in one market so they do in the other. There is a negative
relationship between changes in political risk in the UK and the changes in the world oil and gas market index. That
is, as political risk ratings decrease in the UK (that is, as political risk reduces) the world oil and gas prices and
returns increase. This is not in accordance with the risk/return relationships in financial economics theory but it may
indicate risk aversion on the part of the UK political environment when oil and gas markets are considered. It may
be that when prices in the volatile oil and gas sector increase, the UK risk analysts perceive an improving domestic
political environment.
259

Table 2 shows the results of a system that incorporates the interaction of political risk changes in the UK with
political risk changes in the major EMU countries (France, Germany and Italy) and the USA.
Statistic Value
Adjusted R Square 0.0923
Durbin Watson test statistic 1.9684
t-statistic USA political risk changes 3.8126
Standard error of regression 0.7788
Note: All statistics are significant at the 1% level.
Table 2: Regression results for UK political risk ratings changes
These results indicate a model that is clearly not fully specified, but nevertheless has an explanatory power of
around 9.23% (adjusted R Square value is 0.0923, which is significant at the 1% level). This number may be relied
upon as DW test statistic is significantly greater than 1.5 and serial correlation in the regression errors is not a
problem. Around 9% of the variance of the changes in the UK political risk is explained predominantly by the
changes in political risk in the USA (the t-statistic for USA political risk changes at 3.8126 are significant at the 1%
level). France, Germany and Italy political risk changes are not significant explanatory variables in the system
studied at any level of significance. The positive relationship between the changes in USA political risk ratings and
those of the UK indicates that as political risk in the USA increases, political risk in the UK also increases.
Equation 1) is respecified (in non stationary level series) into a VAR with all variables in that equation
optimally lagged. On a VAR Stability Condition test it the findings are that the VAR satisfies the test with no root
lying outside the unit circle. The VAR lag order selection criteria indicates, through the Likelihood Ratio and
Schwartz Criteria that the optimal lag is 1-2 months. The VAR in the specified form with the world oil and gas
market treated endogenously has strong explanatory power with an adjusted R Square value of 0.8720 significant at
the 1% level. VAR based Granger causality test Chi Square statistic over a 2 month lag verify that there is no
significant evidence at the 10% level of one-way causality running from the independent variables to the world oil
and gas market.
There is evidence at the 1% level that all variables in the model influence the UK oil and gas sector (when the
UK is treated endogenously) and at the 10% level all variables influence the US oil and gas market (when the US oil
and gas market is treated endogenously). When the EMU oil and gas market is treated endogenously there is also
Granger causality running from the other variables at the 10% level of significance. When each country political risk
variable is treated endogenously all of the other variables in each case are shown to influence that variable at levels
of significance ranging from 1% to 10%.
Pair-wise Granger causality tests on two month lags were also run on changes in the oil and gas market indices
and changes in the political risk ratings. Significance levels of F statistics up to the 10% level were observed.
Granger causality runs from the world oil and gas market changes to the UK market changes (at 1%), from the US
market to the UK market (at 1%). There is dual causality between political risk changes in the UK and changes in
the US oil and gas market. The changes in political risk from the UK having slightly greater statistical significance
(at 5% compared to 10%). Moreover, changes in the oil and gas markets in the USA Granger cause changes in
political risk in Germany (at the 5% level); Changes in political risk in the UK Granger cause changes in the EMU
oil and gas market (at the 10% level). Changes in political risk in the UK Granger cause changes in political risk in
France (at the 10% level).
There is dual Granger causality at the 10% level between changes in political risk in the UK and changes in
political risk in Germany (the relationship is stronger running from Germany to the UK). Changes in the world oil
and gas market Granger cause changes in political risk in Germany (at the 5% level). It is however significant at the
10% level that changes in the EMU oil and gas market Granger cause changes in the UK oil and gas market. Overall
the results again demonstrate that the relationships between country oil and gas market changes and political risk
ratings changes are stronger between the world oil and gas markets and the USA and UK markets and USA and UK
political risk ratings and oil and gas markets.
260

5. Conclusion
It is concluded from this short analysis that, in unlagged data, the oil markets of the US and the UK, in that order,
have a stronger relationship with each other and with the global oil and gas market than with the EMU. To a lesser
extent political risk changes in the UK have a small, but significant relationship with the global oil and gas market.
The model that includes these variables has strong explanatory power. The political influence of the major EMU
countries (That is, France, Germany and Italy) on the global oil and gas market is not statistically significant.
When a VAR is introduced with optimally lagged data the strength of the explanatory power of the model is
confirmed. When both pair-wise Granger causality and VAR based causality tests are applied it is confirmed that the
stronger relationships are between the world oil and gas market and those in the USA and the UK. The causality
tests also indicate stronger causal relationships between the USA and the UK in both oil and gas markets and in
political risk ratings. The major relationships in political risk in unlagged data, when political risk ratings only are
considered, are between the US and the UK.
All of this provides evidence that during a period of volatility over the past 7 years in both oil and gas markets
and in political environment, the USA and the UK oil and gas markets are closer to the world market than the EMU.
In addition the UK has been closer to the USA than it has been to the major EMU economies, and might indicate
that, from the view point of energy stock markets sectors, that the UK is understandably not yet fully committed
either politically or economically to take the next step in European integration by formally joining the EMU.
6. References
Asche, F., Osmundsen, P., and Tveteras, R., (2000) “European Market Integration for Gas? Volume Flexibility and
Political Risk”, CESifo Working Paper, Number 358, November.
Asche, F., (2006), “The UK Market for Natural Gas, Oil and Electricity: Are the Prices Decoupled?” The Energy
Journal, Volume 27, Issue 2, 27-40.
Bachmeier, L J., (2006), “Testing for Market Integration Crude Oil, Coal and Natural Gas”, The Energy Journal,
Volume 27, Issue 2, 55-72.
Bourke, P., and Shanmugam, B., 1990, “Risks in International Lending”, in Bourke (Ed), An Introduction to Bank
Lending, Sydney: Addison-Wesley Publishing.
Cantor, R., and Packer, F., 1996, “Determinants and Impact of Sovereign Credit Ratings”, FRBNY Economic Policy
Review, October 37-54.
Eng, G., (2006) “A Formula for LNG Pricing”, Ministry for Economic Development New Zealand,
http://www.med.govt.nz/Templates/multipageDocumentTOC___23939.aspx
ICRG. (2000-2009), International Country Risk Guide, The PRS Group Inc.
Mazighi, Ahmed El Hashemite, (2005), “Henry Hub and National Balancing Point Prices: What Will be the
International Gas Price Reference”, Organisation of Petroleum Exporting Countries, Research Paper,
September.
Okugu, B E., (2002), “Issues in Global Natural Gas: A Primer and Analysis”, IMF Working Paper, Number 40,
February.
Pindyck, R. S., (1999) “The Long-Run Evolution of Energy Prices”, The Energy Journal, Volume 20, Number 2,
pp1-27.
261

Appendix 1
840
800
760
720
OGW
680
01 02 03 04 05 06 07 08
5,000
4,750
4,500
4,250
4,000
3,750
OGUK
3,500
01 02 03 04 05 06 07 08
800
760
720
680
OGUS
640
01 02 03 04 05 06 07 08
5,000
4,000
3,000
2,000
OIGEMU
1,000
01 02 03 04 05 06 07 08
Where OGW, OGUS, OGUK, OIGEMU are the price indices for the oil and gas sector of the stock markets for the
world, the USA, the UK and the EMU.
262

Appendix 2
Note: PRUK, PRUSA, PRFRAN, PRGERM, PRITAL are the political risk ratings for the UK, USA, France,
Germany and Italy. The variable was not a significant explanatory variable of UK political risk and is omitted from
Equation 2).
24
20
16
12
PRUK
8
01 02 03 04 05 06 07 08
26
24
22
20
18
16
01 02 03 04 05 06 07 08
36
32
28
24
20
PRFRAN
PRITAL
16
01 02 03 04 05 06 07 08
25.0
22.5
20.0
17.5
15.0
12.5
PRUSA
10.0
01 02 03 04 05 06 07 08
20
18
16
14
12
PRGERM
10
01 02 03 04 05 06 07 08
263

THE PROGRESS OF NATURAL GAS MARKET LIBERALISATION IN THE UK AND USA: DE-
COUPLING OF OIL AND GAS PRICES
John Simpson, School of Economics and Finance, Curtin University, Perth, Western Australia.
Email: simpsonj@cbs.curtin.edu.au
www.cbs.curtin.edu.au
Abstract. According to the theory of market liberalisation, oil and gas prices should de-couple as de-regulation of natural gas markets
is progressed. This paper expands previous de-coupling studies (primarily, the UK study by Panagiotidis and Rutledge, 2007) and
compares the degree of de-coupling in the USA and the UK in differing periods of oil price volatility. Regression, cointegration and
causality methodology reveals that progress has been made in both the US and the UK in de-regulation with perhaps stronger evidence
of de-coupling of oil and gas markets in the UK, but it cannot be said that the nexus between oil and gas prices in either country has
been severed when long-term relationships between the variables are considered. As a final comment there is evidence in the shortterm
(particularly in the US) of the importance of gas futures markets as drivers of spot gas prices thus re-emphasising the importance
of price expectations that embody country specific economic forecasts, and seasonality and storage factors.
Keywords: Oil prices, gas prices, cointegration, vector autoregressive models, de-coupling, de-regulation.
JEL classification codes: C22; C52; O13; Q43.
1. Introduction
The connection between natural gas and oil markets, the degree of integration and the corresponding volatility and
similarity of volatility of these markets has been extensively studied. For example, Krichene (2002), in a supply and
demand model examined world markets for crude oil and natural gas and finds that both markets became highly
volatile following the oil shock of 1973. The elasticity estimates assisted in the explanation of the market power of
oil producers. Price volatility responding to shocks supported demand price and income elasticities found in energy
studies. Ewing et al (2002) looked at time varying volatility in oil and gas markets across markets and find that
common patterns of volatility emerge that might be of interest to financial market participants.
Adeleman and Watkins (2005) find a degree of stochastic similarity of movement in oil and gas reserve prices
for the period 1982 to 2003 in the USA using market transaction data. They find that both oil and gas current values
rose after 2000 with oil rising sharper than gas in 2003. A study by Regnier (2007) of crude oil, petroleum and gas
prices over a period from 1945 to 2005 finds that these prices are more volatile than prices for 95% of products sold
by domestic producers with oil prices showing greater volatility since the 1973 oil crisis.
In relation to the direct connection between oil and gas prices, a party to party gas price bargaining model was
expounded by Okogu (2002). The model posits that one of the principles of gas pricing is to relate the price of gas to
its value in the market for oil as the major competing fuel, thus implying a strong interrelationship between oil and
gas prices. The implication of such a model is that market power by State or privately owned monopolies can extract
rent from consumers of gas when oil is the only other source of energy. Eng (2006), in debating the price that New
Zealand should pay for its natural gas imports, alluded to the differences in the Japanese and Chinese pricing
models, both of which show the accepted relationship between oil and gas prices, based on data from such sources
as the Asia-Pacific Energy Research Centre and the Institute of Energy Economics of Japan. Again these export
pricing models imply a strong connection between oil and gas prices.
However, Serlitis and Rangel-Ruiz (2004) explored common features in North American energy markets in
shared trends and cycles between oil and gas markets. The study examined Henry Hub Gas prices and WTI crude oil
prices and finds de-coupling of oil and gas prices as a result of de-regulation in the USA.
De-regulation and liberalisation of gas markets implies the removal of the nexus between oil and gas prices.
Reforms in the institutional environment and market place encourage “gas to gas” competition. Gas and oil prices
are said to de-couple and welfare benefits to consumers emerge with a fall in the price of gas. In deregulated markets
the law of One Price should apply. As revisited by Asche (2000), in an integrated market, prices of homogenous
264

goods from different producers and suppliers should move together. Price differentials should only indicate
differences in transportation costs and quality in that homogeneous product. This implies that in a deregulated
natural gas market there should only be “gas on gas” competition and there should be no connection between oil and
gas prices.
The USA and the UK are two important Western economies that have undertaken positive steps to deregulate
their gas markets. The USA gas market is around 6 times larger than the UK market in terms of volumes consumed.
De-regulation of gas markets in the USA began in 1984 with a separation of natural gas supply from interstate
pipeline transportation, deregulated natural gas production and the wholesale market, and competition was
introduced in interstate pipeline transportation. In the UK in 1986, the British government privatised British Gas and
further reforms required the unbundling of supply and transportation and the releasing of some gas supplies to
competitors.
The evidence on market de-coupling is mixed. Silverstovs et el (2005) investigated the degree of integration of
natural gas markets in Europe, North America and Japan in the period early 1990s to 2004 using a principal
components and a cointegration approach where oil and gas markets interacted. They find high levels of gas market
integration within Europe, between Europe and Japan as well as within the North American market.
Mazighi (2005) noted that the UK’s National Balancing Point (NBP) gas price was significantly related to oil
prices. There is also evidence of a statistically significant relationship between oil and gas prices and industrial stock
prices. In testing the long-term behaviour of the UK National Balancing Point (NBP) gas prices he also finds a
relationship between the changes in the volume of manufactured production using regression analysis. As oil is used
as a source of industrial power it follows that there is a relationship between industrial stock prices as well as
alternative energy prices. Mazighi (2005) finds that more than 80% of gas price changes in the US market were not
driven by their fundamental values. In other words other factors such as, oil price changes need to be considered to
account for gas price changes. However, Mazighi (2005) suggests that, in the long-term and in accordance with
economic theory, the evolution of prices of natural gas and any other homogenous commodity is guided by supply
and demand.
Asche (2006) also examined whether or not de-coupling of natural gas prices from prices of other energy
commodities (such as oil and electricity) had taken place in the liberalised UK and in the regulated continental gas
markets after the Interconnector had integrated these markets after 1998. Asche finds that monthly price data from
1995 to 1998 indicated a highly integrated market where wholesale demand appeared to be for energy generally,
rather than specifically for oil or gas.
By 2003 the UK gas market was highly liberalised, according to Panagiotidis and Rutledge (2007), who
investigated the relationship between UK wholesale gas prices and Brent oil prices over the period 1996 to 2003 to
test whether or not orthodox liberalisation theory applied and whether or not oil and gas prices had de-coupled.
Using cointegration techniques and tests of exogeneity of oil prices through impulse response functions, their
findings generally did not support the assumption of de-coupling of prices in the relatively highly liberalised UK
market. The results may at least have indicated that progress in de-regulation had been made.
Studies of the connection between oil and gas markets have suffered because of the use of spot prices only when
a growing body of evidence impresses the need to take into account gas price expectations embodied in futures
prices and thus prices that embody forecasts of macro-economic data as well as seasonality and storage factors. For
example, Modjtahedi and Movassagh (2005) studied natural gas futures prices and find spot and futures prices are
non-stationary with trends due to positive drifts in the random walk components of the prices. They find that market
forecast errors are stationary and that futures prices are less than expected future spot prices (implying futures prices
are backward dated). They also find that the bias in futures prices is time varying and that futures have statistically
significant market timing ability.
Mu (2007) examined how weather shocks impact asset price dynamics in the US natural gas futures market
revealing a significant weather effect on the conditional mean and volatility of gas futures returns. Marzo and
Zagaglia (2007) modelled the joint movements of daily returns on one month futures for crude oil and natural gas
265

using a multivariate GARCH 1 with dynamic conditional correlations and elliptical distributions. They find that the
conditional correlation between the futures prices of natural gas and crude oil had risen over the receding 5 years,
but the correlation was low on average over most of the sample suggesting that futures markets did not have an
established history of pricing natural gas as a function of oil market developments. Geman and Ohana (2009)
remind their readers that it is central in the theory of storage, that there is a role for inventory in explaining the shape
of the forward curve and spot price volatility in commodity markets. They find that the negative relationship
between price volatility and inventory is globally significant for crude oil and the negative correlation applies only
during periods of scarcity and increases during winter months for natural gas.
Again, the evidence of relationships between oil and gas spot markets, as they are also influenced by futures
markets, is mixed. Overall, the forgoing studies are useful in providing information and empirical evidence relating
to the similarity of oil and gas spot and futures price volatility, the connection between oil and gas spot and futures
prices and the integration of gas markets over a period of de-regulation of natural gas markets particularly in the
USA and the UK. However, this paper does not delve into detailed analysis of de-regulation policies in each
country.
Initially assuming the existence of a relationships between gas and oil markets, the study considers a full period
(from the beginning of January 2001 to the end of May 2010) and interim periods of stable oil prices (from early
January 2001 to early June 2003); a period of rapidly rising oil prices (from early June 2003 to early June 2007) and
a period of downward volatility of oil prices (from early June 2007 until the end of May 2010. This latter period
coincides with the period of the global financial crisis). Perusal of the graphs of oil prices from 2001 to 2010 in
Figure 1 indicates the above described broad periods of price movement, sourced from the data in this study. It is
deemed appropriate to divide the study into these periods to partially control for time varying relationships and to
attempt to reduce the impact of apparent structural breaks in the data.
160
140
120
100
80
60
40
20
0
OILOPEC
01 02 03 04 05 06 07 08 09
Figure 1: Oil Price Volatility
The study compares price relationships in the US and the UK over the different periods of oil price volatility. The
issues covered in this paper are as follows:
1. Are unlagged spot oil price changes, the gas futures price changes and gas spot price change relationships
statistically significant?
2. In optimally lagged data, are the level series price relationships statistically significant and have these price
relationships changed?
3. In lagged data are there long-term cointegrating relationships in spot oil prices and spot and gas futures
prices?
4. In short-term dynamics, which markets are the major exogenous forces in the lagged models?
1 Generalised Autoregressive Conditional Heteroskedasticity model.
266

In other words, the central issue remains as to whether or not natural gas market liberalisation policies and deregulation
legislation in the USA and the UK are working. In examining these issues it is expected that further
information will be provided in relation to each of the above studies, but specifically this study is expected to
expand that of Panagiotidis and Rutledge (2007). That is, the study expands to provides a comparison of the
progress of oil and gas price de-coupling and therefore gas market liberalisation in the UK as well as the USA
markets; the study updates the period of analysis of previous studies by capturing the period of the recent global
financial crisis; it allows for structural breaks in data and time varying relationships by investigating sub-periods of
oil price stability and upward and downward oil price volatility; and it takes into account gas futures prices.
2. The Model and Method
In preliminary analysis, an unlagged model of first differences of the price index series is examined over a full
period from 01/01/2001 to 31/05/2010 and over the sub-periods described above for both US and UK markets. The
relative connectivity between oil, spot gas and gas futures price changes is examined. The preliminary analysis
provides information on the strength of relationships in each spot natural gas market and how these relationships
change in the various sub-periods.
Based on evidence of the studies of the nexus between oil and gas prices (For example, Krichene 2002; Ewing et al.,
2002; Okogu, 2002; Serlitis & Rangel-Ruiz, 2004; Adeleman & Watkins, 2005; Mazhigi, 2005; Asche, 2006; Eng,
2006; Regnier, 2007; Panagiotidis & Rutledge, 2007) and taking into account gas futures price interaction (For
example, Modjtahedi and Movassagh, 2005; Mu, 2007; Marzo & Zagaglia, 2007 and Geman & Ohana, 2009), the
following unlagged and lagged models are proposed for testing, with the various series logarithmically transformed.
The unlagged models are as follows:
� P � � � � �P
) � � ( �P
) � e
1)
gst
t
1 ( t ost
2t
gft
t
Where;
� Pgs , �P
t gft
and os P � are the changes in the spot gas price, the gas futures price and the spot oil price
respectively in each of the country markets.
In the main analysis, optimally lagged level series data are initially examined in a vector autoregressive model
(VAR) for each of USA and UK markets. If the series examined are cointegrated a vector error correction model
(VECM) is specified over the various periods of the study in order to confirm long-term equilibrium relationships
and to test for short-run dynamics and exogeneity.
Based on Equation 1 specified in level series, the following model in its functional form is specified. Note all
variables on the right hand side of the equation are specified in both an unlagged and a lagged form from t �1
to
t � n with n being the optimal lagged deduced from lag exclusion tests and lag order information criteria. The
endogenous variable is also lagged on the right hand side of the equation.
P � f P , P )
2)
t
gs
( os gf
3. The Data and Methodology
There is a two stage methodology in this paper. In preliminary analysis is found in unit root tests that level series
prices and errors of those relationships are non-stationary. First differences of those prices and the errors of the first
differenced relationships are stationary. The processes are therefore integrated and non-stationary. However, the
study first examines unlagged relationships in a regression model. This is specified in price changes to remove the
problem of serial correlation in the errors that is found if the model was specified in level series. The next stage of
the method is to examine a lagged bivariate model specified in level series prices. The optimal lags are decided with
lag length criteria and lag order tests. If cointegration is discovered a VECM is specified to confirm cointegration
and test for exogeneity.
267

Daily price data for the full period of the study are obtained from the DataStream database for all of the
variables in the models. The spot gas price in the USA is the Henry Hub (HH) gas price. The HH is an index in
dollars per million cubic metres of British Thermal Units. The delivery point is a pipeline interchange near Erath,
Louisiana, where a number of interstate and intrastate pipelines interconnect through a header system operated by
the Sabine Pipe Line. It is also the standard delivery point for the NYMEX natural gas futures contract in the US. It
is considered a representative indicator of US gas prices.
The spot gas price in the UK is the National Balancing Point (NBP) gas price for UK or London. The NBP is a
virtual trading location for the sale and purchase and exchange of UK natural gas. It is the pricing and delivery point
for the Intercontinental Exchange (ICE) natural gas futures contract. It is the most liquid gas trading point in Europe
and is a major influence on the price that domestic consumers pay for their gas at home. Gas at the NBP trades in
pence per therm. It is similar in concept to the Henry Hub in the United States, but differs in that it is not an actual
physical location 2 .
The spot oil price is the Organisation of the Petroleum Exporting Countries (OPEC) oil price. The OPEC cartel
(that is, there is a formal/explicit agreement among competing firms) consists of net oil-exporting countries
primarily made up of Algeria, Angola, Ecuador, Iran, Iraq, Kuwait, Libya, Nigeria, Qatar, Saudi Arabia, the United
Arab Emirates, and Venezuela. The OPEC prices reflect the market power of the group of net oil exporting countries
that control the exports collectively of around 40% of the world’s oil requirements. These oil prices, though formed
through the market power induced by cartel behaviour, are assumed the major driver of the global crude oil price.
The UK gas futures price is the ICE London or UK natural gas futures price for 6 months. It is a Reuter’s
continuation series, which gives the data for 6 months forward. The USA gas futures price is the NYMEX natural
gas futures prices for 6 months, which is 6 month forward rate. It starts at the 6th nearest contract month, which
forms the first values for the continuous series until the first business day of the nearest contract month when, at this
point, the next contract month is taken. These gas futures prices are considered representative of UK and US gas
futures markets respectively. The gas futures prices selected represent gas market price expectations as the data
embody various global and country economic forecast data and information that impact price expectations such as,
exchange rate, inflation, interest rates, growth rates and also storage factors and seasonal effects on gas demand in
the USA and in Europe and the UK. Due to unavailability of data for the NBP, the UK study cannot examine the
first sub-period of relative oil price stability. It examines for the UK the second and third periods individually and
the full period for the UK thus covers both sub-periods of upward and downward oil price volatility.
4. Preliminary Findings
The findings of the analysis are reported as preliminary analysis, preliminary findings and main findings as follows:
Preliminary analysis of all unlagged level series for both the USA and the UK and in all periods of the study
indicates skewness and kurtosis problems, which indicate lack of normality and uniformity (tests from Jarque &
Bera, 1987). This is a violation of the assumptions of ordinary least squares (OLS) regressions and indicates, in turn,
problems in serial correlation of the errors for each of the regression relationships. First differencing removes the
problems of serial correlation in the errors of the first difference relationships (according to Durbin and Watson
(DW) tests (Durbin and Watson, 1971), but White tests indicate that heteroskedasticity problems remain in the
errors thus indicating model misspecification. An autoregressive conditional heteroskedasticity (ARCH) model is
deemed more suitable for analysis. The results of this analysis are reported in Table 1.
268

Regression Model
(Equation 1)
Adjusted R Square z Statistics:
Spot Oil /Gas
Futures in price
changes
Standard Error of
Regression
Durbin-
Watson
Statistic
Variance Equation
Coefficient
ARCH/GARCH
US Spot Gas Price
Changes: Period 1
0.0011 1.4534/0.7073 0.5386 2.0750 2.1763/0.3418
Period 2 0.0735* 6.3355*/10.1860* 0.3060* 1.9919* 0.1349*/0.8705*
Period 3 0.0990* 7.9450*/5.6669* 0.2106* 2.1873* 0.1669*/0.7524*
Full Period 0.0366* 10.4862*/10.8608* 0.3572* 2.0778* 0.2061*/0.8609*
UK Spot Gas: Price
Changes
Period 1
0.0002 0.2275/0.3931 6.3800 2.2445 0.2276*/0.8805*
Period 2 -0.0006 0.6469/-0.7428 3.0404 2.2748 0.1979*/0.7781*
Full Period 0.0039* 1.1086/4.9908* 5.2515 2.2574* 0.1866*/0.8782*
Note: Significance levels for spot oil z statistics and ARCH/GARCH terms are, no asterisk means no significance; * denotes
significant at 1% level and ** denotes significant at the 5% level and *** denotes significant at the 10% level.
Table 1: Results of Spot Gas First Differenced Models
Whilst the explanatory power of the models for USA spot gas prices changes is low over all periods of the
study, it has improved over the periods 2 and 3. The results are significant at the 1% level and serial correlation does
not detract from the reliability of the model’s parameters as indicated by DW statistics. The z statistics for each of
spot oil and gas futures are statistically significant over the full period of the study and for sub-periods 2 and 3. The
magnitude of the z statistic for spot oil has increased through the second sub-period (upward volatility of oil prices)
and the third period (downward volatility of oil prices) and is greatest in magnitude over the full period.
In unlagged data, it is evident that the USA gas market has not de-coupled with the oil market; the relationship
between gas and oil markets having increased over the full period studied. However, the z statistics for US gas
futures are also significant at the 1% level over the second, third and full periods. The magnitude of the z statistic for
US gas futures is greatest over the full period of the study (and greater than that statistic for spot oil over the full
period) and is stronger than that for spot oil in the second sub-period of upward oil price volatility than in the first
and third sub-periods (that is, in periods of oil price stability and downward oil price volatility respectively). This
indicates that economic forecasts, seasonality and storage factors embodied in US gas futures price changes are
important in US gas markets over periods of upward price volatility, downward volatility and over the full period.
This evidence indicates that, though the nexus between gas and oil prices has not been broken, sound progress has
been made in deregulation of US gas markets as seen in the connection of gas futures price changes to spot gas price
changes.
Evidence of de-coupling in the UK gas market is provided in unlagged data with the lack of significance of the
spot oil price change variable in any period of the study. The explanatory power in the models for all periods is
lower for the UK spot gas market. None of the z statistics for the independent variables (that is, the spot oil and UK
gas futures variables) is statistically significant in any period except in the full period of study where the significant
variable is gas futures price changes. However, it appears that economic expectations, seasonality and storage issues
as reflected in gas futures price changes are not as important in UK gas markets. In the UK the nexus between oil
and gas appears to have been broken and gas price expectations embodied in gas futures price changes are
significant over the full period of the study. In the UK it appears that deregulation policies have acted to liberalise
the spot gas market over the full period of the study. In the next stage of the preliminary analysis the level series, the
first differenced series and the respective errors of these relationships for each country and each period under study
are tested for unit roots using Augmented Dickey and Fuller (ADF) tests (Dickey & Fuller, 1981). The results of
these tests are shown in Table 2.
269

Variable t statistic (level series prices) t statistic (first differences)
EQUATION ONE VARIABLES:
US spot gas:
Period 1 -3.6194* -16.0758*
Period 2 -2.4680 -30.3681*
Period 3 -1.1756 -28.2260*
Full Period -3.2471** -30.8007*
Spot Oil:
Period 1 -1.9851 -20.5592*
Period 2 -1.2215 -25.4412*
Period 3 -1.2526 -20.5072*
Full period -1.3347 -25.2201*
US gas futures:
Period 1 -0.2257 -27.6259*
Period 2 -1.6665 -32.1152*
Period 3 -1.1074 -29.1183*
Full Period -1.8914 -50.5273*
US spot gas errors:
Period 1 -5.5641* -15.0709*
Period 2 -3.9899* -32.2051*
Period 3 -2.0991 -27.2291*
Full Period -2.9969** -30.5471*
EQUATION ONE VARIABLES:
UK spot gas:
Period 1 -4.1836* -18.6215*
Period 2 -2.6305*** -23.1402*
Full Period -4.6458* -23.7836*
Spot Oil:
Period 1 -0.9171 -26.5189*
Period 2 -1.2494 -20.4901*
Full Period -1.4947 -22.1242*
UK gas futures:
Period 1 -1.6575 -31.4909*
Period 2 -1.0020 -25.9100*
Full Period -2.0604 -40.8704*
UK spot gas errors:
Period 1 -4.4892* -18.3592*
Period 2 -4.3887* -23.1257*
Full Period -5.0251* -23.6352*
Note: No asterisk means non-significance. ADF test results are shown. PP unit root tests are not reported, but confirm the results of the ADF
tests. *Significance levels are at 1%, ** at 5% and *** 10%. The full period for the USA is from 1/1/2001 to 31/5/2010. The first period for the
USA is from 1/1/2001 to 2/6/2003 (oil price stability), the second period for the USA is from 2/6/2003 to 1/6/2007 (upward oil price volatility)
and the third period USA is from 1/6/2007 to 31/5/2010 (downward oil price volatility). The first period in the UK is the period of upward
volatility from 2/6/2003 to 1/6/2007. This is due to limitations in NBP spot gas price data, which commence from 2/6/2003. The second period
for the UK is from 1/6/2007 to 31/5/2010 (period of oil price downward volatility). The full period for the UK is thus from 2/6/2003 to
31/5/2010.
Table 2: Unit Root Tests
Table 2 results indicate overall, that the level series and errors of the level series relationships are non-stationary and
the first differenced series and the errors of the first differenced relationships are stationary. This enables a
conclusion that, in each case for each country and for each period under study, the processes are integrated and nonstationary
and this in turn enables a move to the main analysis. That is, to apply all level series to a VAR based
Johansen (Johansen, 1988) cointegration test. If cointegration is present, a VECM 3 is specified to confirm long-term
relationships and to test short-term dynamics of those relationships in Granger Causality tests (Granger, 1988).
3 Once it is ascertained that the variables are I (1) and optimally lagged, a vector error correction model (VECM) is used in order to confirm
cointegration and test causality. The VECM is a re-parameterised version of the unrestricted VAR and is appropriate when our variables are I (1)
and cointegrated. In the presence of I (1) variables but no cointegration, causality would be studied from the VAR model specified in first
differences.
270

5. Main Findings
Lag order and lag exclusion tests are conducted to ascribe an optimal lag for cointegration and causality testing. The
models were initially tested for stability over all periods of the study for both US and UK markets using stability
condition tests. The findings are that the models are stable, with no root lying outside the relative unit root circle.
The results for the lag order and then cointegration and causality tests are shown in Table 3.
Model Number of Cointegrating
Relationships According to
Trace and Eigen-value tests
EQUATION TWO:
US Spot Gas:
Period 1
Period 2
Period 3
Full Period
EQUATION TWO:
UK Spot Gas:
Period 1
Period 2
Full Period
Optimal Lag
Order in
Days
2** 3*
1** (According to Trace test
only)
4*
None 3*
2** 4*
2** 3*
None 2*
2** 3*
Granger Causality (Chi Square Statistic)
US Gas Futures Granger causes US Spot Gas (135.4754*). Spot
gas causes gas futures (6.2797***). Note stronger Granger
causality of gas futures to spot gas.
No significant Granger causality between Spot Oil and Spot
US Gas.
US Gas Futures Granger causes US Spot Gas (120.7309*).
No significant Granger causality between Spot Oil and US
Spot Gas.
Within the model for this sub-period US Gas Futures significantly
Granger causes Spot Oil (64.1979*)
US Gas Futures Granger causes US Spot Gas (249.1941*).
No significant Granger causality between Spot Oil and US
Spot Gas.
Within the model for this sub-period US Gas Futures significantly
Granger causes Spot Oil (73.4092*) as does US Spot Gas
(9.4335**)
US Gas Futures Granger causes US Spot Gas (346.2631*).
No significant Granger causality between Spot Oil and US
Spot Gas.
Within the model and over the full period, US Gas Futures
significantly Granger cause Spot Oil (127.8207*)
No significant Granger causality between Spot Oil and UK
Spot Gas.
No significant Granger causality from UK Gas Futures to UK Spot
Gas.
Within the model for this period, UK Gas Futures significantly
Granger causes Spot Oil (15.2450*).
No significant Granger causality between Spot Oil and UK
Spot Gas.
Gas futures Granger cause spot gas (33.1187*).
Granger causality is significant from Spot Oil to UK Gas Futures
(11.6501*) and from UK Gas Futures to Spot Oil (86.7800*) with
the latter the stronger causal relationship.
No significant Granger causality from UK Gas Futures to UK Spot
Gas.
No significant Granger causality between Spot Oil and UK
Spot Gas.
Within this model, Spot Oil Granger causes UK Gas Futures
(8.5459***) and UK Gas Futures Granger causes Spot Oil
(83.7156*). The latter driver relationship is stronger according to
the magnitude of the Chi Square statistic.
Note: The Johansen cointegration tests take the assumption that there is a linear deterministic trend in the data. Optimal lags are decided based on
the majority significance of the Likelihood Ratio, the Final Prediction Error, the Akaike, Schwarz and Hannan-Quinn information criteria. The
number of cointegrating equations is based on both maximum eigenvalues and trace statistics. In the number of cointegrating relationships, no
271

asterisk means no significance. * denotes significance at the 1% level, ** at the 5% level and *** at the 10% level. The causality tests show
similar levels of Chi Squared statistical significance.
Table 3: Optimal Lags and Cointegration and Causality Test Results
In both markets over the long-term there is evidence of cointegration in the first sub-period for the US (the
period of oil price stability), the first sub-period for the UK (the period of upward price volatility) and in each
market over the full periods of the studies. This represents evidence that whilst short-term causal relationships may
show de-coupling evidence, there remains a long-term relationship between oil and gas prices and gas futures prices
whereby these variables move in a similar way and come together to stability. This represents evidence that in the
long-term the markets have not fully de-coupled and that in both markets deregulation policies still have some
distance to go in achieving full gas market liberalisation.
The most important finding, for both the US and the UK markets over the sub-periods of the study and over the
full period, is that there is no significant short-term causal relationship between spot oil and spot gas. In addition, it
is evident that in the US, gas futures Granger cause gas spot prices over the various sub-periods and over the full
period of the study. In the UK there is significant Granger causality running from gas futures to spot gas over the
full period of the study. There is however evidence within the model, of Granger causality between gas futures and
spot oil for both markets. Overall, in the short-term there is evidence of de-coupling in both markets, but the nexus
between oil and gas futures prices is not fully broken.
6. Conclusion
The theoretical base for this paper lies in the re-visitation of market liberalisation theory by Asche (2000), but
focuses on updating and expanding a UK cointegration and causality study by Panagiotidis and Rutledge (2007).
Just as mixed evidence is produced by the authors reviewed, mixed evidence is provided in this paper. The paper
commences with the assumption that there exists a relationship between oil and gas prices as implied, for example,
by Okogu (2002) and Eng (2006). Mazhigi (2005) finds the UK gas price was significantly related to oil prices.
Researchers such as, Krichene (2002), Ewing et al., (2002), Adeleman and Watkins (2005) and Regnier (2007) find
that oil and gas markets possess similar trends in stochastic volatility.The importance of gas futures in their
embodiment of price expectations based on economic, seasonal and storage information is put forward by
researchers such as Modjtahedi and Movassagh (2005), Mu (2007), Marzo and Zagaglia (2007) and Geman and
Ohana (2009). Serlitis and Rangel-Ruiz (2004), find de-coupling of oil and gas prices as a result of de-regulation in
the USA. Siliverstovs et el (2005), find high levels of gas market integration within Europe and North America.
Asche (2006) finds that monthly price data from 1995 to 1998 in the UK indicated a highly integrated gas market.
Panagiotidis and Rutledge (2007) findings generally did not support the assumption of de-coupling of prices in the
relatively highly liberalised UK market, but imply that progress was being made in deregulation policies.
The results of the study in this paper are summarised as follows: In unlagged data and regression analysis it is
evident that overall the UK has achieved progress in de-coupling of oil and gas markets. The nexus between spot oil
and spot gas markets in the US is stronger, but so is the relationship between US gas futures and US spot gas. When
data are lagged and considered in short-term causal relationships there is strong evidence of de-coupling of spot oil
and spot gas prices in both markets over all sub-periods and full periods of the study. This de-coupling effect in the
short term is supported by evidence of significant causality running from gas futures to spot gas, though within the
multivariate model, there is evidence a causal relationship between gas futures and spot oil prices for both countries.
However, when long-term relationships are considered in lagged data there is evidence of cointegration in
multivariate models where spot gas, gas futures and oil prices interact and this means that the nexus between oil and
gas prices has not been severed in either market. For the US this nexus between spot oil, gas futures and spot gas
markets appears stronger in the first period of oil price stability from 1/1/2001 to the 2/6/2003, but also over the full
period studied from 1/1/2001 to 31/5/2010. There is no discernable connection between these markets during the
period global financial crisis and partial recovery, which included rapidly falling oil prices. In the UK, the nexus
between spot oil, gas futures and spot gas appears strongest in the period of upward oil price volatility from
2/6/2003 to 1/6/2007, and over the full period of that study from 2/6/2003 to 31/5/2010. Again there is no
discernable connection between these markets in the UK over the period of the global financial crisis and partial
recovery.
272

Progress has been made in both the US and the UK in de-regulation with perhaps stronger evidence of decoupling
of oil and gas markets in the UK, but it cannot be said that the nexus between oil and gas prices in either
country has been severed when long-term relationships between the variables are considered. There is no
discernable evidence to indicate that any relationships identified in this study vary significantly from one sub-period
to another. As a final comment there is evidence in the short-term (particularly in the US) of the importance of gas
futures markets as drivers of spot gas prices thus re-emphasising the importance of price expectations that embody
economic forecasts, and seasonality and storage factors.
7. References
Adeleman, M A., and Watkins, G C., (2005), “US Oil and Natural Gas Reserve Prices, 1982-2003”, Energy
Economics, Volume 27, Issue 4, July, Pages 553-571.
Asche, F., (2000), “European Market Integration for Gas? Volume Flexibility and Political Risk”, CESIFO Working
Paper, Number 358, November.
Asche, F., (2006), “The UK Market for Natural Gas, Oil and Electricity: Are the Prices Decoupled?” The Energy
Journal, Volume 27, Issue 2, 27-40.
Dickey, D. A., and Fuller, W. A., (1981), “Likelihood Ratio Statistics for Autoregressive Time Series within a Unit
Root”, Econometrica, Volume 49, 1022-1057.
Durbin, J., and Watson, G. S., (1971), “Testing for Serial Correlation in Least Squares Regression-111”, Biometrika,
Volume 58, 1-42.
Eng, G., (2006) “A Formula for LNG Pricing”, Ministry for Economic Development New Zealand,
http://www.med.govt.nz/Templates/multipageDocumentTOC___23939.aspx.
Ewing, B T., Malik, F., and Ozfidan, O., (2002), Volatility Transmission in the Oil and Natural Gas Markets”,
Energy Economics, Volume 24, Issue 6, November, Pages 525-538.
Geman, H and Ohana, S., (2009), “Forward Curves, Scarcity and Price Volatility in Oil and Natural Gas Markets”,
Energy Economics, Volume 31, Issue 4, July, Pages 576-585.
Granger, C. W. J., (1988), “Some Recent Developments in a Concept of Causality”, Journal of Econometrics,
Volume 39, 199-211.
Jarque, C. M., and Bera, A. K., 1987, “A Test for Normality of Observations and Regression Residuals”
International Statistical Review, Volume 55, pp. 163-172.
Johansen, S, (1988), “Statistical Analysis of Cointegration Vectors”, Journal of Economic Dynamics and Control,
Volume 12, Pages 231-254.
Krichene, N., (2002), “World Crude Oil and Natural Gas: A Demand and Supply Model”, Energy Economics,
Volume 24, Issue 6, November, Pages 557-576.
Marzo, M., and Zagaglia, P., (2007), “A Note on the Conditional Correlation between Energy Prices: Evidence from
Future Markets”, Energy Economics, Volume 30, Issue 5, September, Pages 2454-2458.
Mazighi, Ahmed El Hashemite, (2005), “Henry Hub and National Balancing Point Prices: What Will be the
International Gas Price Reference”, Organisation of Petroleum Exporting Countries, Research Paper,
September.
Modjtahedi, B., and Movassagh, N., (2005), “Natural Gas Futures: Bias, Predictive Performance, and the Theory of
Storage”, Energy Economics, Volume 27, Issue 4, July, Pages 617-637.
Mu, X., (2007) “Weather, Storage and Natural Gas Price Dynamics: Fundamentals and Volatility”, Energy
Economics, Volume 29, Issue 1, January, Pages 46-63.
Okogu, B E., (2002), “Issues in Global Natural Gas: A Primer and Analysis”, IMF Working Paper, Number 40,
February.
Panagiotidis, T., and Rutledge, E., (2007), “Oil and Gas Markets in the UK: Evidence from a Cointegration
Approach”, Energy Economics, Volume 29, Issue 2, March, Pages 329-347.
Regnier, E., (2007), “Oil and Energy Price Volatility”, Energy Economics, Volume 29, Issue 3, May, Pages 405-
427.
Serlitis, A., and Rangel-Ruiz, R., (2004), “Testing for Common Features in North American Energy Markets”,
Energy Economics, Volume 26, Issue 3, May, Pages 401-414.
Silverstovs, B., L’Hegaret, G., Neumann, A., and von Hirschhausen, C., (2005), “International Market Integration
for Natural Gas? A Cointegration Analysis of Prices in Europe, North America and Japan”, Energy Economics,
Volume 27, Issue 4, July, Pages 603-615.
273

QUANTITATIVE EASING ENGINEERED BY THE FED, AND PRICES OF INTERNATIONALLY
TRADED AND DOLLAR DENOMINATED COMMODITIES AND PRECIOUS METALS
Gueorgui I. Kolev, EDHEC Business School, France
Email: gueorgui.kolev@edhec.edu
Abstract. I study the impact of quantitative easing by the FED on the prices of internationally traded and dollar denominated
commodities (oil) and precious metals (gold, silver, platinum and palladium). Finite distributed lag models suggest that the long run
multipliers in regressions of the log of precious metals or commodities prices on the log of the US monetary base is about one, i.e., a
permanent one percent increase in the US monetary base results in one percent increase in the prices of these commodities and
precious metals. In other words, the quantitative easing actions by the FED are purely inflationary as predicted by classical economic
theory. I also present an event study of the quantitative easing announcement effects on prices, and these are small for precious metals
and negative for oil. Overall this study suggests certain dissonance between the price behavior on one hand in the bond market, and
on the other in the commodities and precious metals markets. It also suggests that taking the US CPI reported by the Bureau of Labor
Statistics to be identical with the concept of inflation can be quite misleading. Further research is needed to determine why the implied
inflation in the US CPI and the US government debt market is so different from the implied inflation in the US dollar denominated
and internationally traded commodities and precious metals.
Keywords: quantitative easing, monetary policy, precious metals’ prices, oil prices
JEL classification: E5, G1
1 Introduction
Towards the end of September 2008, the American FED started unprecedented monetary expansion doubling the US
monetary base. This has been done through what Taylor (2009) calls mondustrial policy – financing certain sectors
and firms by printing money.
Cochrane (2011) worries that the US might enter deflation. As the nominal interest rates are close to 0 and so
the US treasury bonds become effectively money, he argues, the FED has not enough influence over the events –
whether the FED prints money or not does not matter for the US economy. He bases his observations on the fact that
the US consumer price index has been flat since September 2008, and the spreads between US nominal treasury
bonds and US real bonds (Treasury Inflation Protected Securities) has been negligible.
I study the prices of commodities and precious metals that are internationally traded and denominated in US
dollars, mostly after September 2008, when the quantitative easing has been started by the FED. Looking at the
growth in prices of these internationally traded goods since September 2008, one sees that the US is already in a
state of very high inflation, one might even say hyperinflation. I suggest that the mondustrial policy carried out by
the FED might have not been so innocuous, and might threaten the solvency of the US government and their ability
to borrow money in the future.
The main point of this paper is to show that looking at the Consumer Price Index (CPI) and trying to decipher
inflation is not a good idea. The CPI is an imaginary number that bureaucrats come up with, most probably trying to
pursue political agenda. Even if the US Bureau of Labor Statistics are politically impartial, and are trying to do their
job the best they can, still the CPI that they produce is not subjected to market discipline. If they get it wrong, by say
reporting too high or too low CPI, no arbitrageur can enter and correct the “mispricing”, the CPI is not traded.
On the other hand, looking at commodities and precious metals, it is very easy to see whether the US dollar is
inflating, or not. Commodities and precious metals are commonly quoted in US dollars. The exchanges where they
are traded have no incentive to misrepresent their value in dollars. If their value in dollars does not reflect their fair
value, any arbitrageur is free to enter and by his trades and attempts to make money out of the mispricing, correct
the mispricing.
274

2 Preliminary Graphical Analysis
Figure 1 below depicts the US monetary base measured in dollars, the spot price of oil (Crude Oil, Light-Sweet,
Cushing, Oklahoma) and the price of gold (London, afternoon fixing). All the series are normalized by subtracting
the mean of the series and dividing by the standard deviation. We can see from this picture the unprecedented
growth in the US monetary base that started in September 2008. Before that, including the period before January
2005 that is not in the graph, the US monetary base was growing at a steady, negligible rate. Then sometime in
September 2008, the FED in a desperate move, trying to save the sinking US economy, doubled the monetary base.
The first reaction to this is that the US government is monetizing its debt.
Looking at the core US CPI, nothing happens in September 2008 (Cochrane, 2011, figure 6). If anything, the
rate of inflation measured by the core CPI decreases between July 2008 and January 2009. This is very puzzling
from the standard classical economic theory point of view. If we consider the identity MV = PY (Mankiw, 2010,
ch.4), where M is the supply of money, V is the velocity of money, P is the price level, and Y is the real output of
the US economy, we see that M has doubled, Y has not changed, so for P to stay fixed, the velocity of money V
must have halved, the latter being extremely implausible. The other alternative is that the CPI is a very misleading
measure of the price level P.
Looking at the dollar denominated prices of oil and gold in Figure 1 below, we see a radically different story.
The moment when the FED started printing out money, the prices of oil and gold started growing accordingly. This
is exactly what one would expect from a classical economic theory point of view.
-2 0 2 4
01jan2005 01jul2006 01jan2008 01jul2009 01jan2011
date
monetary_base oil_price
gold_price
Figure 1. Time series evolution of the US monetary base, the price of gold and the price of oil. All variables are standardized by subtracting the
mean and dividing by the standard deviation. The heavy vertical line corresponds to 24 September 2008 when the unprecedented growth in the
US money supply started, the dashed and dotted vertical lines correspond to dates of FOMC quantitative easing announcements, 25 November
2008, 1 December 2008, 16 December 2008, 28 January 2009, 18 March 2009, 12 August 2009, 23 September 2009, 4 November 2009, 10
August 2010, and 21 September 2010.
Focusing on the period where there was substantial variation in the US monetary base, that is after and
including September 2008, Figure 2 below displays scatter plots of standardized oil and gold prices versus the
standardized US monetary base. The values are contemporaneous, so these pictures do not capture any dynamics.
The static relation between the gold price and the monetary base is almost linear. The thick line is the 45 degree line,
and the dotted line is the best linear fit between the price of gold and the contemporaneous monetary base. They
almost overlap, so for one standard deviation of increase in the monetary base after and including September 2008,
the price of gold increases by one standard deviation too.
Figure 2 below shows that the relation between the price of oil and the US monetary base is not that simple.
This is to be expected as oil is also the single most important input to production, whereas gold is mostly used as an
investment asset. Therefore the price of oil is expected to depend heavily on the state of the world economy, and the
US economy plays a major role in the world industrial output. Figure 2 reveals a V shaped relation between the
normalized price of oil and the normalized US monetary base. The right hand of the V cloud of points is more dense
275

and, although the linear relation between the two variables over the whole sample is almost flat, the relation in the
right hand side of the V is close to a standard deviation increase in monetary base resulting in a standard deviation
increase in the standardized price of oil, i.e., close to the 45 degree line.
-2 -1 0 1 2 3
-.5 0 .5 1 1.5 2
monetary_base
gold_price monetary_base
Fitted values
-2 -1 0 1 2 3
-.5 0 .5 1 1.5 2
monetary_base
oil_price monetary_base
Fitted values
Figure 2. Scatter plots of gold and oil prices versus the US monetary base. Data points are at weekly frequency (measured on Wednesdays) and
cover subsample over which the US monetary base exhibits variation, from 1 September 2008 to 20 October 2010. Variables are standardized as
in Figure 1 by subtracting the mean and dividing by the standard deviation, computed over the sample in Figure 1 (from 1 January 2005 to 20
October 2010). The heavy sloped line is the 45 degree line, and the dashed sloped line is the best linear fit to the scatter.
3 Dynamic Regression Analysis
To further understand how the US monetary base affects the prices of oil, gold and other precious metals, I fit finite
distributed lag models in Table 1 below. It is believed that the FED does not take into account prices of oil, gold,
other precious metals, stock prices, housing prices etc., in setting their monetary policy. What the FED targets is the
federal funds rate, keeping an eye on inflation and the output gap. Empirical research by monetary economist have
shown that what is known as Taylor rule (Orphanides, 2008) best describes the behavior of the FED, at least on
average and most of the time. Therefore it seems plausible to assume that, for example, in a regression of natural
logarithm of the price of gold on the contemporaneous and lagged values of the natural logarithm of the US
monetary base, that the contemporaneous and lagged values of the monetary base are strictly exogenous. Hence I
maintain the strict exogeneity assumption in the regressions in Table 1 below:
ln(Pt) = α + ∑j=0,1,..,n βj ln(Mt-j) + εt, E(εtMs) = 0, for each t and s,
where ln(P) is in the various regressions the natural logarithm of the price of oil, gold, silver, platinum and
palladium, ln(M) is the natural logarithm of the US monetary base.
In the present context the most interesting estimate coming from the finite distributed lag models below is the
Cumulative Dynamic Multiplier (also known as the long run multiplier), that shows the effect of a permanent unit
increase in the regressor on the dependent variable in question. The long run multiplier is given by the sum of the
contemporaneous effect and the effects of the lags.
CDM = ∑j=0,1,..,n βj.
Note that classical monetary theory has a prediction for the value of the long run multiplier – one percent permanent
increase in the supply of money should result in exactly one percent increase in the price level, so CDM=1,
according to classical economic theory. (As Milton Friedman famously said, “Inflation is always and everywhere a
monetary phenomenon”, see Friedman, 1970.) That is, of course, if we believe that the prices of oil, gold and other
precious metals are representative of the theoretical concept of inflation.
276

In the bottom panel of Table 1 we see that when fitting the finite distributed lag models by Ordinary Least
Squares (OLS), there are signs of severe positive autocorrelation (the autoregressive parameter r of the residual is
always above 0.7, and most of the time above 0.8, and the Durbin-Watson statistic DW0 is very small, always below
0.5), see Panel C, Table 1. From this, and also from theoretical considerations we might conclude that there is a unit
root in the oil, precious metals and monetary base time series. Therefore in fitting the model by OLS we would be
running spurious regression. To take care of this problem I follow McCallum (2010) and Kolev (2011) and estimate
the finite distributed lag models by feasible Generalized Least Squares (GLS). McCallum (2010) and Kolev (2011)
show that when feasible GLS is used in the presence of heavy autocorrelation, one obtains efficient estimates and
test statistics having correct size, even if one or both of the regressor and regressand contain a unit root. By using
feasible GLS one in effect lets the data determine what is the order of (quasi)differencing necessary to render the
time series complying with the classical regression requirements. Hence I estimate the finite distributed lag models
in Table 1 by the Cochrane & Orcutt (1949) feasible GLS procedure.
I selected the number of lags to be included in the models by testing down for each variable, starting from 14
lags. It turned out that the regression that required the largest number of lags is where the dependent variable is log
oil price, 10 lags. Although the precious metals required less lags, I then applied the 10 lags structure across all of
the dependent variables, with an eye on ease of comparison of models. Standard regression theory suggests that
including lags that are potentially zero in the population is not too onerous, it just results in loss of some efficiency.
Looking across Table 1, Panel A, we see that coefficients on the contemporaneous and lagged values of
monetary base are mostly positive for the precious metals (but there are some which are negative here too), and
somewhat mixed for oil. Therefore we focus on the value of the Cumulative Dynamic Multipliers (CDM), which are
displayed in Panel B of Table 1. The first two rows of Panel B display the estimates of the CDM and the p-value
associate with the null hypothesis that the CDM is equal to 0. We see that the CDM range between .77 (for oil) and
1.56 (for palladium) and the null hypothesis that the CDMs are 0 can be rejected at any significance level. What we
read from this information is that, for example, a one percent permanent increase in the US monetary base leads to a
.77% increase in the price of oil quoted in USD. Similarly one percent increase in the US monetary base leads to a
.77% increase in the price of gold quoted in USD.
The third and the fourth rows of Panel B display the estimate of the quantity CDM - 1, and the associated test of
the null hypothesis that the CDM=1. As we see the null hypothesis that CDM=1 cannot be rejected at standard
significance levels for any of the time series, and the closest it comes to rejection is for gold, where the p-value of
the test is .11. What we gather from Panel B of Table 1 is that the long term effect, i.e., the effect of one percent
permanent increase in the US monetary base on the price of oil and precious metals in the long run, is positive, and
most probably equal to 1, as predicted by classical monetary theory.
(1) (2) (3) (4) (5)
lnoil lngold lnsilver lnplatinum lnpalladium
Panel A
lnmonetary_base -0.0309 0.113 0.00467 0.355
(0.275) (0.113) (0.203)
**
0.571
(0.171)
*
(0.290)
L.lnmonetary_base 0.697 ***
0.170
(0.252)
*
0.0886 0.163 0.109
(0.101) (0.183) (0.161) (0.262)
L2.lnmonetary_base 0.301 0.0259 0.0228 0.226 0.0751
(0.252) (0.102) (0.183) (0.163) (0.264)
L3.lnmonetary_base -0.211 -0.247
(0.252)
**
-0.192 0.0808 0.136
(0.101) (0.183) (0.163) (0.263)
L4.lnmonetary_base 0.444 0.214
(0.273)
*
0.451
(0.112)
**
0.129 0.0661
(0.198) (0.184) (0.295)
L5.lnmonetary_base -0.664 **
0.0460 -0.0896 -0.240 -0.216
(0.269) (0.110) (0.195) (0.179) (0.288)
L6.lnmonetary_base -0.728 ***
0.157 0.0299 -0.146 -0.0820
(0.272) (0.116) (0.197) (0.190) (0.305)
L7.lnmonetary_base 0.650 ***
0.222
(0.232)
**
0.408
(0.0971)
**
0.219 0.340
(0.168) (0.159) (0.254)
L8.lnmonetary_base -0.363 -0.0774 -0.0171 0.0563 0.0800
277

(0.231) (0.0970) (0.167) (0.157) (0.252)
L9.lnmonetary_base 0.0517 0.0744 0.0808 0.152 0.134
(0.220) (0.0887) (0.160) (0.143) (0.230)
L10.lnmonetary_base 0.622 ***
0.0759 0.116 0.300
(0.219) (0.0882) (0.160)
**
0.342
(0.137) (0.227)
Constant -6.874 *
-4.174
(3.817)
**
-10.23
(2.021)
***
-11.49
(3.214)
***
-16.60
(1.825)
***
Panel B
(4.234)
Cumulative Dynamic Multiplier (CDM) .769 .773 .903 1.296 1.556
Ho: CDM=0, p-value 0.005 0.000 0.000 0.000 0.000
CDM - 1 -.230 -.226 -.096 .296 .556
Ho: CDM = 1, p-value
Panel C
0.385 0.110 0.665 0.021 0.062
Observations 106 100 102 100 100
R 2
0.857 0.992 0.730 0.990 0.947
F 51.17 1057.3 22.16 804.7 142.3
r 0.841 0.922 0.887 0.733 0.862
DW0 0.271 0.198 0.280 0.497 0.223
DW 1.674 1.935 1.858 1.535 1.236
Table 1: Standard errors in parentheses (* p < 0.10, ** p < 0.05, *** p < 0.01). The data frequency is weekly, measurements on variables are
taken on Wednesdays, and the sample runs from 1 September 2008 to 20 October 2010. The regressions are estimated by the Cochrane & Orcutt
(1949) feasible GLS procedure.
4 An Event Study of FOMC Announcements
In this section I present an event study of the short term effects of the FOMC (Federal Open Market Committee)
announcements of quantitative easing. Table 2 below displays the average price change from the closing price on the
day before the announcement, to the closing price on one day after the announcement of quantitative easing by the
FOMC. The event dates are given in the first column, for more details on these quantitative easing public
announcements by the FOMC see Gagnon, Raskin, Remache & Sack (2010), Krishnamurthy & Vissing-Jorgensen
(2011) and Wright (2011). For example, the first row of Panel A of Table 2 tells us that on the 25 November 2008
the FOMC announced they will undertake quantitative easing. The average percentage change, measured in per day
terms, between the closing price of oil on the 24 November 2008 and the closing price of oil on the 26 November
2008, was an increase of 0.53%. The last row in Panel A of Table 2 displays the average price change on non-event
dates.
It should be pointed out that the FOMC has not been very forthcoming about its quantitative easing activities.
By looking at the data we see that when the FOMC announces that they have decided to carry out quantitative
easing, the action has already taken place. In other words the FOMC announcements seem to endorse retrospectively
what the trading desks of the FED have already done through open market operations, a point made by Taylor
(2009) as well. It is not clear who decides on these matters, two possibilities being that the decision is taken by the
FOMC, but they hide it from the public for a while, the other being that anonymous people in the FED system take
the decisions, and the FOMC by its announcements fulfills formal requirements of due process after the fact. This
can be clearly seen by looking at Figure 1. The monetary base in the US grows from 915 030 mln USD on 24
September 2008 to 1 483 267 mln USD on 19 November, amounting to 62% growth. Nevertheless, the first
announcement of the FOMC mentioning quantitative easing takes place on the 25 November 2008, long after the
fact. Therefore the results of the event study should be read cautiously.
FOMC quantitative easing (1) (2) (3) (4) (5)
announcement date
Panel A
D.lnoil D.lngold D.lnsilver D.lnplatinum D.lnpalladium
25nov2008 0.00529 -0.00612 0.0128 0.0140 0.00256
(0.0233)
1dec2008 -0.0475
(0.0330)
(0.00975)
0.00257
(0.0138)
278
(0.0168)
-0.0518 **
(0.0238)
(0.0132)
0
(0.0187)
(0.0193)
-0.0116
(0.0273)

16dec2008 -0.0524 **
0.0259
(0.0233)
***
0.0305
(0.00975)
*
0.0124 0.00850
(0.0168) (0.0132) (0.0193)
28jan2009 -0.00108 -0.00293 -0.00719 0.00263 0.00526
(0.0233) (0.00975) (0.0168) (0.0132) (0.0193)
18mar2009 0.0248 0.0219
(0.0233)
**
0.0104 0.0187 0.0201
(0.00975) (0.0168) (0.0132) (0.0193)
12aug2009 0.00793 0.00567 0.0243 0.00677 0.00182
(0.0233) (0.00975) (0.0168) (0.0132) (0.0193)
23sept2009 -0.0420 *
-0.00210 -0.0141 -0.00188 -0.00667
(0.0233) (0.00975) (0.0168) (0.0132) (0.0193)
4nov2009 0.000377 0.0130 0.0305
(0.0233) (0.00975)
*
0.0127 0.0185
(0.0168) (0.0132) (0.0193)
10aug2010 -0.0211 0.00104 -0.00984 -0.00581 -0.0104
(0.0233) (0.00975) (0.0168) (0.0132) (0.0193)
21sept2010 -0.0124 0.00554 0.00405 0.00493 0.00649
(0.0233) (0.00975) (0.0168) (0.0132) (0.0193)
non-event -0.000795 -0.0000741 -0.000276 -0.000361 0.000780
Panel B
(0.00169) (0.000709) (0.00122) (0.000959) (0.00140)
Cumulative Event Effect (CEE) -.137 *
.064 **
.029 .064 .033
Ho: (CEE – non-event) = 0, p-value
Panel C
0.077 0.047 0.592 0.139 0.597
Observations 398 398 398 398 398
R 2
0.0325 0.0377 0.0390 0.0147 0.00851
Table 2: Standard errors in parentheses (* p < 0.10, ** p < 0.05, *** p < 0.01). The data frequency is daily, and the sample runs from 17
September 2008 to 20 October 2010. Announcement effects, standard errors and test statistics are computed by running a regression of the daily
change (between closing price today and closing price yesterday) in the log of the price in question on a mutually exclusive and exhaustive set of
indicator variables (without a constant). For example 25nov2008 is an indicator variable that takes the value of 1 if the date is the FOMC
announcement date 25 November 2008 or the next day 26 November 2008, and 0 otherwise. The estimates on the event indicators give the
average daily price change in percentage points associated with the given event date, assuming that the announcement effect took place only on
the day of the announcement and on the next day, i.e., two day event window. The indicator non-event takes the value of 1 if all the event
indicators are 0, and gives the average daily price change in percentage points associated with dates on which no announcements occurred. The
Cumulative Event Effect (CEE) is the sum of the estimated effects on all the event dates.
Looking at Panel B of Table 2 we see that the cumulative effect of all the quantitative easing announcements by
the FOMC is positive for precious metals, but very small in magnitude, and statistically significant only for gold.
(Significance is measured by testing the null hypothesis that the cumulative announcement effect price change is the
same as the average price change on non-announcement dates.) All the announcements by the FOMC resulted in
only 6.4% higher price level as measured by the increase in the price of gold, which is a negligible short term effect.
The FOMC announcements of quantitative easing resulted in lower price of oil in the short run. The cumulative
announcement effect for the price of oil is -13.7% and it is significant at the 10% significance level.
The negative effect for oil can probably be rationalized by recognizing that oil is a commodity heavily used in
production and the US is a major world consumer of oil. When the FOMC announces quantitative easing, the
markets realize that there will be more money chasing given amount of output (inflationary effect suggesting a
positive cumulative impact on the price of oil), but also that the state of the US economy is bad, and therefore there
will be less production and less demand for oil (deflationary effect suggesting a negative cumulative impact on the
price of oil). Empirically it turns out that the second effect dominates.
5 Summary
I study the effects of quantitative easing engineered by the FED on the prices of internationally traded and dollar
denominated commodities (oil) and precious metals (gold, silver, platinum and palladium). Sometime in September
2008 the FED started historically unprecedented expansion of the US monetary base. Curiously this expansion has
not resulted in any inflation so far, as measured by the US CPI index. Similarly the bond market for US government
279

debt offers nominal returns close to 0, which suggests not only that the real interest rate is close to 0 but also that the
bond holders expect that there will be no inflation in the years to come.
I present estimates from finite distributed lag models which suggest that one percent increase in the US
monetary base results in one percent increase in the prices of internationally traded and dollar denominated
commodities and precious metals, as predicted by classical economic theory. The long run multipliers for oil, gold,
silver, platinum and palladium are all about 1, estimated from feasible GLS regressions of the log of the price on the
log of the monetary base. The long run multipliers are all significantly different from 0, and the null hypothesis that
they are equal to 1 cannot be rejected at the 10% significance level. If we take the prices of oil and precious metals
as legitimate measures of inflation, we see that the US is already in a state of very high inflation. Every percent
increase in the monetary base has resulted in a percent increase in the prices of commodities and precious metals. In
view of my findings, the behavior of the US debt bond market, and the US CPI index is puzzling.
To complement the above evidence of long run price effects of quantitative easing by the FED, I present an
event study of the price effects of quantitative easing announcements by the FOMC. I take the event window to be
the day of the quantitative easing announcement and the next day. Here the price increases are small for precious
metals ranging between cumulative price increase resulting from all the FOMC announcements of 2.9% for silver
and 6.4% price increase for gold. The cumulative effect for oil is even negative 13.7%, i.e., the price of oil actually
decreases on the quantitative easing announcement dates. These results probably indicate that the news that the
markets read from the FOMC announcements of quantitative easing is that the FED believes that the US economy is
in a very bad shape. Hence in terms of price effects of these announcements, the bad economic conditions outweigh
the monetary expansion content. There is also clear evidence that the FOMC makes quantitative easing
announcements long after monetary expansion has already taken place, thereby retroactively admitting what they
have been doing, instead of disclosing their intentions for the future.
Overall, the results here are consistent with the long/short term dichotomy in economics. I do not mean to imply
that the FED should target commodities and precious metal prices when setting monetary policy (Mankiw & Reis,
2003). The point here is that looking at commodities and precious metal prices one can see inflation that is otherwise
invisible only upon looking at the US CPI and the bond market. The present study also suggests certain dissonance
between the markets for US government bonds, and the markets for commodities and precious metals. The implied
inflation expectations are vastly different. This raises the possibility that the money printing activities of the FED
caught investors in US government debt asleep at the wheel.
6 References
Cochrane, John (2011). Understanding policy in the great recession: Some unpleasant fiscal arithmetic. European
Economic Review, 55, 2–30.
Cochrane, D., & G. H. Orcutt (1949). Application of least squares regression to relationships containing
autocorrelated error terms. Journal of the American Statistical Association, 44(245), 32–61.
Friedman, Milton (1970). Wincott Memorial Lecture, London, September 16, 1970.
Gagnon, Joseph, Matthew Raskin, Julie Remache, & Brian Sack (2010). Large-Scale Asset Purchases by the Federal
Reserve: Did They Work? Federal Reserve Bank of New York Staff Report no. 441, March.
Kolev, Gueorgui I. (2011). The "spurious regression problem" in the classical regression model framework.
Economics Bulletin, 31(1), 925-937.
Krishnamurthy, Arvind, & Annette Vissing-Jorgensen (2011). The Effects of Quantitative Easing on Interest Rates.
Brookings Papers on Economic Activity, forthcoming, Fall 2011.
Mankiw, N. Gregory (2010). Macroeconomics, 7 th edition. Worth Publishers.
Mankiw, N. Gregory, & Ricardo Reis (2003). What Measure of Inflation Should a Central Bank Target? Journal of
the European Economic Association, 1(5), 1058–1086.
McCallum, Bennett T. (2010). Is the spurious regression problem spurious? Economics Letters, 107(3), 321-323.
Orphanides, Athanasios (2008). Taylor rules. The New Palgrave Dictionary of Economics, 2nd Edition. v. 8,. 200–
04.
Taylor, John (2009). The Need to return to a monetary framework. Business Economics, 44, 63 – 72.
Wright, Jonathan H. (2011). What does Monetary Policy do to Long-Term Interest Rates at the Zero Lower Bound?
Working paper, Department of Economics, Johns Hopkins University.
280

AGRI-BUBBLES ABSORB CHEAP MONEY: THEORIES AND PRACTICE IN CONTEMPORARY
WORLD
Evdokimov Alexandre Ivanovich & Soboleva Olga Valerjevna,
Saint-Petersburg State University of Economy and Finance, Russia
Email: ier2002@mail.ru, prof-evdokimov-alex@yandex.ru, www.finec.ru
Abstract. Agflation, biflation, stagflation, cycle development and world food crisis have become key words for economic discussions.
Since 2002 stock markets booming accompanied by crude oil and following wheat and other grains prices’ rally compelled market
analysts to reconsider core factors of upward trend and the role of speculators. In 2008 Global financial crisis was expected to correct
speculative trend and burst market bubbles however, it did not happen. High volatility of world prices has been remained as yet both
in capital and commodity markets. In summer 2010 world food prices began to grow rapidly and this trend continued in the beginning
2011. That makes political efforts of overcoming the Global financial crisis barren. We examined stock and commodity markets’
behavior and the fundamental reasons of upward trend. We have come to accept that current active growth of stock and commodity
markets are not supported by real business activity but by Central Banks which follow ‘cheap money’ policy.
Keywords: biflation, biofuels, bubble, ‘cheap money’ policy, global financial crisis, income-absorption theory, incomes parity, market
equilibrium, the Relative Strength Index (RSI), speculative activity
JEL classification: E2, E3, E5, F4, G01, G1, H5, I3.
1 Introduction
Since 2002 basic commodity prices grew actively with a slight correction for crude oil in 2003 and 2006, for wheat
and corn in 2004. General trend for world commodity prices remained ascending till the beginning of the Global
financial crisis in 2008. The main stock markets had four years of fluctuating but enthusiastic growth till the mid of
2007. Referring to the Financial Times published data the Global financial crisis dropped stock market indexes to
28% - 70% of their level recorded in 2007. World prices for commodities fell by 39% for wheat, 25% for corn, 54%
for crude oil regards 2007. Many analysts predicted slow and uneven recovery of the world economy.
Several reasons of the crisis are on agenda - financial globalization, convergence of banking and security
exchange markets, increasing speculative activity, innovations in the market of derivatives such as structured
investment vehicles (SIVs), or collateralized debt obligations (CDOs) that led to mortgage crisis in the USA (OECD
Report, 2008). And finally more analysts agree that the Global financial crisis coincided with the end of
Kondratiev’s business cycle. As Kondratiev’s cycles usually last 35-60 years, it is considered that the current
recession will last for several years.
However, in 2009 stock markets opposed pessimistic forecasts and showed 25-30% growth regards 2008,
particularly, Russia’s stock market grew by 175% regards 2008. Oil prices recovered by 74% regards 2008, grains
joined price rally in the mid 2010 - wheat month future price grew by 78%, corn – by 85%, oil – by 27% regards the
end of 2009. Encountering with such extreme prices jumps the questions therefore naturally arise: whether the
current soaring food commodity prices are the result of a speculative bubble or any confident fundamental reasons
support ascending prices.
For this paper we aim to provide our account for an up-to-date role of speculative activity in the stock and
commodity markets, to summarize fundamental factors contributing to upward trend, and to propose the theoretical
basis for food prices’ growth in the period of business cycle recession.
2 Critical Analysis of Contemporary Food Commodity Prices Growth and it’s Theoretical Basis
There are different approaches to the critical analysis of the current upward trend of the commodity prices but we
deliberately simplify data analysis by calculating the Relative Strength Index (RSI), a technical indicator/oscillator
which shows the overbought or the oversold moments for a certain commodity/stock item in exchange trade.
Observing the behavior of price movements and RSI movements during 2001-2010 we tried to form an overview for
281

the role of speculators in present days. The most promising approach for the research is to combine technical and
fundamental analyses of price movements being outlined. For this purpose we used the report of the United States
Department of Agriculture (USDA) which was published in 2008 (Ronald Trostle, 2008). In this report supply and
demand factors which contributed to the recent increase in food commodity prices were investigated. However, we
believe that in the circumstances of the Global financial crisis it is insufficient to use fundamental factors which are
originated from global supply and demand. Conditions of financial markets and monetary systems of different
countries are to be considered as important factors affecting commodity prices growth. So we expanded fundamental
analysis by two theoretical approaches: aspiration of achieving incomes parity, and absorption of money surplus due
to prices growth.
2.1 Balancing Role of Market Speculators is Doubtful
The principle assumption of balancing role of market speculators was accepted in the end of 1970s (Friedman M,
1968), however in 1990s it was demolished by numerous analysts who investigated the behavior of market
speculators in the low effective markets of developing and emerging countries. High effective markets are
characterized as having high business transparency and information penetration. Earlier it was considered that the
markets of developed countries had high effectiveness. Market speculators in the high effective markets act
beforehand. Due to this all market operators could know about market signals and could undertake precautionary
actions to reduce risks. For example, when the market price exceeds the level of market equilibrium, speculators
begin to sell overbought item and their actions pull down the price to equilibrium level.
For the last twenty years the balancing role of speculators has been abated both in the effective and ineffective
markets. It was mentioned that ineffective markets suffer from speculative activity because speculators here do not
adjust to market equilibrium but they form market trends themselves. They continue to buy overbought item with
the hope that the market price will increase further. Their confidence in ascending prices is partially supported by
generally high anticipations of inflation and by foreign investments inflows which may also contribute to inflation
growth in developing countries. Since the end of 1980s as market agents began to develop in the result of
consolidation in banking sector and other financial intermediaries financial globalization formed the worsen
competitive environment in developed, developing and emerging markets because one or few big investors, for
example, hedge funds, could affect market prices. Hence at present time speculative activity is not the problem for
developing countries only. That is illustrated by RSI and price movements.
RSI was offered in 1978 to indicate overbought and oversold moments. RSI oscillator is usually used in the
complex with other technical indicators, volume data, support and resistance lines, directional movements, etc. The
index normally varies in the range of 30-70. If RSI exceeds 30 or 70 level that is the technical signal of anticipated
change of current trend – downturn in the case of exceeding 70 point, and vice-versa in the case of exceeding 30
point. Market bubble is the term to characterize the overbought moment when numerous agents continue to buy
stock item/commodity being encouraged by own positive anticipations and there is no resistance to the ascending
trend. It is considered that bubbles burst (i.e. change towards rapid decrease) just after the moment when RSI
exceeds 70. Though, it does not always happen.
We analyzed RSI for wheat, crude oil and some stock markets (RTS Russia, FTSE 100, SP 500 indexes) and we
found out that during 2001-2010 there were moments when markets did not switch towards downturn trend when
RSI indicated overbought state. That happened with SP 500 in 2003, FTSE 100 in 2005, RTS in 2005-2006, wheat
in 2006, and in July and December in 2010, with crude oil in 2009. The illustration for wheat price’s behavior in
February 2010 - March 2011 is provided in Figure 1. In the moments when RSI exceeds 70 point market cool down
but the trend does not change in direction like market agents pause waiting to additional trigger to support prices
rally. Oppositely, when RSI reaches 30 point the market agents always switch to upward trend. This confirms the
market readiness to accept high prices.
We exercised linear regression for months’ indexes and month future prices published by Financial Times for
the period of March 2001 – December 2010 and we outlined (Table 1): strong ties between FTSE 100 and SP 500
indexes; strong ties between RTS and FTSE 100/SP 500 indexes; strong ties between oil future prices and wheat
future prices; strong ties between RTS and oil /wheat prices; weak ties between wheat future prices and FTSE 100
282

SP 500; middle ties between oil future prices and FTSE 100 /SP 500. It proves that mature financial markets such as
London or New-York are less responsive for commodity price movements compared to young emerged markets
such as Russia. That may be interpreted by the dominant role of speculative capital in ineffective markets and
therefore irrationally formed market trends. Durbin-Watson statistics confirms the existence of a positive
autocorrelation in the data raw that normally illustrates the current future prices’ dependence on previous market
expectancies.
During the sharp phase of the Global financial crisis market speculators were accused of their main role in
forming high volatility of prices, bubbles and financial crises. Nevertheless, it could be partially accepted because
we know the examples of stock market bubbles, such as IT/telecome papers crisis in 2002 in the USA, when there
was no affection of bank sectors and general economic slowdown.
close price, $ cents per
bushel
volume deals
1000
800
600
400
200
0
1500000
1000000
500000
Cutler's RSI
0
100
80
60
40
20
0
R (correlationcoefficient)
Wheat Front Month Futures WC1:CBT
Figure 1 Wheat price movements and month’s RSI
R2 (determination
coefficient)
F- statistics
empirical
F- statistics table, P-
0,05
Feb 2010
Mar 2010
Apr 2010
May 2010
Jun 2010
Jul 2010
Aug 2010
Sep 2010
Oct 2010
Nov 2010
Dec 2010
Jan 2011
Feb 2011
Mar 2011
F-statistics results
Durbin-Watson
statistics
t-statistics for
constant
t--statistics for
independent
variable
t-statistics table, P-
0,05
FTSE 100 (y) - SP 500 (X) 0,924 0,853 679,3 19,49 significant 0,144 2,80 26,06 1,66 strong tie
RTS (y) - SP 500 (X) 0,712 0,506 120,1 19,49 significant 0,075 -6,95 10,96 1,66 strong tie
RTS (y) - FTSE 100 (X) 0,745 0,555 146 19,49 significant 0,081 -7,96 12,08 1,66 strong tie
SP 500 (y) - Oil price (x) 0,519 0,27 43,21 19,49 significant 0,098 27,79 6,57 1,66 middle tie
FTSE 100 (y) - Oil price (x) 0,569 0,323 55,94 19,49 significant 0,103 29,13 7,48 1,66 middle tie
RTS (y) - Oil price (x) 0,891 0,795 453,1 19,49 significant 0,142 -3,12 21,29 1,66 strong tie
SP 500 (y) - Wheat price (x) 0,39 0,152 20,94 19,49 significant 0,092 24,09 4,58 1,66 weak tie
FTSE 100 (y) - wheat price (x) 0,431 0,186 26,76 19,49 significant 0,098 24,83 5,17 1,66 weak tie
283
Linear model results

R (correlationcoefficient)
R2 (determination
coefficient)
F- statistics
empirical
F- statistics table, P-
0,05
F-statistics results
Durbin-Watson
statistics
t-statistics for
constant
t--statistics for
independent
variable
t-statistics table, P-
0,05
RTS (y) - wheat price (x) 0,78 0,608 181,5 19,49 significant 0,164 -1,86 13,47 1,66 strong tie
Wheat price (y) - SP 500 (x) 0,39 0,152 20,94 19,49 significant 0,101 -0,07 4,58 1,66 weak tie
Wheat price (y) - FTSE 100 (x) 0,431 0,186 26,76 19,49 significant 0,11 -0,67 5,17 1,66 weak tie
Wheat price (y) - RTS (x) 0,78 0,608 181,5 19,49 significant 0,217 10,71 13,47 1,66 strong tie
Wheat (y) - oil (x) 0,799 0,639 206,7 19,49 significant 0,269 5,36 14,38 1,66 strong tie
Table 1: Results of linear single factor regression
Summing-up we believe that the balancing role of market speculators have been changed since 1970s.
Speculative activities do contribute to higher volatility of market prices and speculators are powerful to form upward
trends and bubbles. However, the reason for prices rally is not only originated from trader’s psychology to pursuit
short-term profitability. There must be some fundamental factors that support increasing willingness of market
speculators to take additional risks.
2.2 Fundamental Factors Contributing to Ascending Food Commodity Prices
In 2008 Ronald Trostle and USDA’s team published a research paper of supply and demand factors that contributed
to the increase in food commodity prices since 2006 till the financial crisis in 2008. The analysts divided factors into
two groups. Demand factors are: increasing population, rapid economic growth and therefore rising purchasing
power of developing countries, rising per capita meat consumption, declining demand for stocks of food
commodities, rapid expansion of biofuels production, dollar devaluation, large foreign exchange reserves,
aggressive purchases by importers. Supply factors are: slowing growth in agricultural production, escalating crude
oil price, rising farm production costs, adverse weather. We tested topicality of those factors in the light of the
consequences of the Global financial crisis.
Increasing population is admitted a stable factor that will form the general higher demand on food in future
hence, it will push food prices up. We do not doubt that adverse weather may cause disruption of food supply in any
times of any governments. Slowing growth in agricultural production seems reasonable because most of developed
and some emerging countries have already reached the maximum level of agricultural productivity while principally
new advanced technologies are not available. Victor M. Polterovich (Polterovich V.M, 2009) asserts that
biotechnologies and genetic engineering have not become an impulse for significant increase of agricultural
productivity and profitability of farmers. The global strategy of development of agriculture capacity in middle and
less developed countries stumbles on political, social instabilities in these countries that raise investment risks.
Rapid economic growth and rising purchasing power of developing countries have not definitely formed yet.
Most of economists forecast sooner recovery of developing countries after the Global financial crisis compared to
developed countries. We accept the idea, however, we wonder whether the economic growth induces the same
active growth of real incomes of the most of population in developing countries. Practically we found out that real
incomes of population in developing countries increase less actively compared to national product growth.
Economies of developing countries are currently dependent on their exports and purchasing capacity of developed
countries, so if purchasing power of developed countries decreases that will affect incomes of developing countries
as well. At the same time the living standards in developing countries, particularly agricultural exporters, remain
extremely low and it will definitely take several decades to eliminate this divergence between developing and
developed countries. Moreover, analysts must remember that when real incomes grow consumers tend to increase
non-food purchases keeping conservative dietary pattern.
284
Linear model results

Escalating crude oil price is a strong factor for recent increasing popularity of biotechnologies and some
changes of production policy of agricultural exporters. This factor has recently induced a rapid expansion of biofuels
production in some countries, and therefore a shrink of food grains’ supply. Currently the level of areas for biofuel
grains presents negligible share of the world’s areas so this factor is not fundamentally strong yet though it is
alarming. Switching to biofuel grains production is a young trend, but market speculators triggered price rally
beforehand. In addition the increase in energy prices affects farm production costs, particularly in developed
countries which use energy consuming technologies. Energy prices growth will remain a strong factor contributing
to ascending food commodity prices.
Declining demand for stocks of food commodities is not considered as a factor of demand of the end consumers
but is recognized as a result of changes of the governments’ policy aiming to reduce budget expenses for
maintaining big stocks. In the recent report of European Commission Directorate-General for Agriculture and Rural
Development (The Common Agricultural Policy Explained, 2008) in the chapter ‘A record to be proud of’ the
reporters boasts of reducing level of public storage of cereals, beef and butter compared to 1985-1995 level.
Aggressive purchases by importers, a factor mentioned by USDA analysts, are the result of shortage of food
commodity stocks. Decline of stocks is a strong basis for anxious speculative activity and therefore increasing food
prices volatility.
Financial markets turmoil is not accented by USDA analysts in 2008. Though in their report the dollar
devaluation and large foreign exchange reserves of developing countries are mentioned among demand factors that
push food prices up. We consider that contemporary monetary policies of developed and developing countries lead
to the large US dollar reserves nevertheless this is not the direct cause of the food prices growth. The only logical
explanation for large US dollar reserves’ influence on the rapid price growth appears when we estimate the
speculators’ confidence in future upward trend. These are trade speculators who include global money liquidity into
expectations of high level of future prices. And finally, most of economists anticipate US dollar devaluation or even
dollar crush, meantime since 2008 US dollar has depreciated gradually, not impressively. We conclude that the
dollar’s moderate devaluation could not be a strong factor for rapid growth of food commodity prices in 2010.
2.3 Some Theoretical Approaches for Long-Term Upward Trend of Food Commodity Prices
When we collected analytical discussions of the reasons for rapid growth of food prices we elicited scarce
theoretical basis for development of certain assumptions and forecasts. For example, fundamental factors of supply
and demand which have been discussed above do not apply to economic theories except for market equilibrium.
Market equilibrium can be difficult to attain in the case when exporting and importing countries form unpredictable
food commodity stocks for future consumption as well as in the case of inflation spiral. So, the market price factors
have to be investigated carefully with attention for market specifics and deviations.
The Global financial crisis has already brought its fruitful results for the science as more economists examine
the role of monetary policy for the economic development, and the ability of monetary authorities to decide
economic cycle’s problems. The most advances have already been developing the idea of global equilibrium both in
financial and business sectors. We agree that the processes of globalization contribute to the extent and the depth of
the Global financial crisis, but we have not discovered any confident theory for global equilibrium. For our paper
research we reconsidered two theoretical approaches pertaining to financial study that may explain the reasons of
long-term upward trend of food commodity prices.
2.3.1 Food Commodity Prices Growth Means to Correct Incomes Disparity
The problem of incomes parity has been discussed for years since the Great Depression in 1930s (Stine O.C, 1937).
In general view all groups of society involved in different economic activities in the country should be guaranteed of
comparable satisfactory standards of living. Otherwise employees leave less profitable activity in order to get higher
incomes in other industries (sectors of economy). Adam Smith’s laissez-faire principal confronts incomes parity as
the market community itself can define the most profitable activities and other less profitable ones naturally degrade.
However, in the XX century social principals of economics came on agenda of all economic formations, partly due
to the achievements of the former USSR (Union of Soviet Socialist Republics). Incomes parity is the social
285

theoretical approach of economics study. After the Second World War more countries admitted the role of
government in distribution of incomes from the most lucrative to less succeeded business activities through the
systems of support and social guarantees.
In the post crisis period of 1930s the USA became a pioneer of intervention into food commodity markets and
budgetary support of incomes in agriculture. Though the American system of incomes’ support was not ideal, the
USA demonstrated the most effective system of inventing and reforming the instruments of support. In 1980s the
USA actively lobbied the General Agreement on Tariffs and Trade, the GATT negotiations (currently World Trade
Organization, WTO), towards reducing the level of agriculture support, appealing to reconsider the ability of
farming to compete fairly. However, we have to emphasize that before liberal reforms American farmers had long
history of fruitful years of high profits and strong government support. This has formed US farmers as confident
competitors at present. Currently the incomes of US farmers are diversified significantly, only 12% of total incomes
of an average farm family in the USA were attained due to agricultural activity in 2003-2008 (The 2008/2009 World
Economic Crisis…., 2009).
In the end of 1980s there was a period of the highest level of agriculture support in the developed countries.
However, GATT/WTO agricultural agreements obliged member states to reduce support and to reform support
instruments. Countries are gradually decreasing direct payments which are subjects to production volumes, areas,
livestock quantity. Many developed countries currently pay fixed amounts to support farmer’s incomes, and
calculations for such payments are usually based on historical level of money transfers. Such payments are not
corrected according to the changes of average incomes of employees in other sectors of economy.
Discussions about the methods of calculation of incomes parity are always on the agenda. At present there is no
comparable statistics data on incomes per employee in agriculture across numerous countries. So, we decided to
compare the constant cumulative growth of gross value added (GVA) in agriculture and other sectors (Figure 2).
Data for graphs is available on http://data.un.org/Explorer.aspx. It is obvious that agriculture activity is less
profitable/productive compared to other industries and sectors of economy. The same trends were observed for
South Africa, Jordan, Algeria, Mexico, Poland, Canada, New Zeeland, Spain, Norway, France – not presenting in
Figure 2. The disparity of GVA growth rates is more visible in the period since 1990s when IT, telecommunications,
financial and other services became a core driver of economic growth.
An emerging concept of ‘biflation’ introduced by F. Osborne Brown in 2003 could be accepted as a mean for
achieving incomes parity. Biflation describes the situation when commodity-based assets’ prices increase while
debt-based assets’ prices decrease. Debt-based goods are fancy houses, automobiles, luxury goods, etc. In the post
crisis period all sectors of economy that produce debt-based goods will face shrinking demand. These sectors will
record numerous bankruptcies and layoffs. On other hand consumers will continue to buy staples at any prices
soaring, although it may induce social unrests all over the world, particularly in developing countries. The Global
financial crisis exposed disbalance of overestimated services, debt-based goods, and underestimated value of food
commodities and other basic industries of economy. So we suppose that the current rapid growth of food commodity
prices reflects the implicit intention of economic agents to develop incomes of agricultural activity to the level of
industrial and services sectors.
2.3.2 Food Commodity Prices Growth Absorbs Cheap Money
Income-absorption theory was introduced by Alexander S. in 1952 (Alexander S, 1953) to explain the effect of
elimination of incomes surplus in the period of national currency devaluation and exports volumes growth. When
the country pursues to develop export volumes the monetary authorities apply ‘cheap money’ policy. Interest rates,
municipal bond’s rate and official banks’ reserve rates are reduced. In the result of such actions exchange rate of
national currency decreases while foreign demand grow being attracted by cheaper exports from the country with
devaluated currency. However, at the same time import prices increase and if the country has a significant portion of
imported goods in the consumers’ basket, then both imported and domestic goods’ prices will grow. The growth of
domestic prices absorbs the surplus of national incomes earned from exports development.
286

cumulative growth index,
const. prices
cumulative growth index,
const. prices
3,5
3
2,5
2
1,5
1
0,5
0
1972
1975
8000
7000
6000
5000
4000
3000
2000
1000
0
United States of America
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
GVA in other sectors excluding agriculture
GVA in agriculture, hunting, forestry, fishing
Russia
2008
1992
1994
1996
1998
2000
2002
2004
2006
2008
GVA in other sectors excluding agriculture
GVA in agriculture, hunting, forestry, fishing
Figure 2 Dynamics of Constant Growth of Gross Value Added
Greece
Income absorption may not happen in the case of a strict monetary policy when money liquidity is limited and
controlled. However, strict monetary policy is not the case of contemporary world. Since 1990s many developed
countries decreased the cost of money. Figure 3 illustrates the dynamics of money value (one year deposit rates) and
wheat price constant growth (cumulative index). Data for calculation is available on http://data.un.org/Explorer.aspx
and http://faostat.fao.org/site/570/DesktopDefault.aspx?PageID=570#ancor. The same trends were observed for
South Africa, Jordan, Japan, Brazil, Argentina, Canada, Spain, France. There is no doubt that this is a global trend
for money becoming cheaper. Partially it is supported by foreign investors who have access to cheap money in one
country and can invest at higher income rates in other countries. Due to income absorption effect ‘Cheap money’
policy become ineffective in stimulating business activity. In the period of economy recession when there is no
principally new technology/product that could restart business cycle ‘cheap money’ policy produces inflation that
absorb money surplus.
We consider that present-day eagerness to accept higher prices have to be supported by some stronger
macroeconomic factor than the psychology of trade speculators and consumers. Most of consumers, especially in
developing countries, accept ascending prices. Such irrational behavior is described as a case of ‘ratchet effect’ –
when prices reach their top level once consumers will anticipate further prices growth and these anticipations will
bring on higher inflation. The desire to improve living standards and therefore to increase consumption wins over
rational behavior and preference to savings. Here we emphasize that such irrational consumer’s behavior is actively
supported by easy access to credits. For the last 15 years banks have switched to serve consumers as they admit to
having higher margin from retail banking compared to corporate clients. The Global financial crisis initially shrank
banks’ liquidity, but Central Banks began to support commercial banks injecting cheap money into heated markets.
287
cumulative growth index,
const prices
cumulative growth index,
const prices
4
3,5
3
2,5
2
1,5
1
0,5
0
3
2,5
2
1,5
1
0,5
0
1971
1975
1972
1975
1978
1981
Australia
1984
1987
1979
1983
1987
1991
1995
1999
1990
1993
1996
1999
2002
2003
2007
GVA in other sectors excluding agriculture
GVA in agriculture, hunting, forestry, fishing
GVA in other sectors excluding agriculture
GVA in agriculture, hunting, forestry, fishing
2005
2008

deposit rate, %
deposit rate,%
18
16
14
12
10
8
6
4
2
0
60,0
50,0
40,0
30,0
20,0
10,0
3 Summary
1970
1974
1978
United States of America
1982
1986
1990
1994
1998
2002
2006
deposit interest rate, annual
wheat price index
Russia
0,0
1996 1999 2002 2005 2008
deposit interest rate, annual
wheat price index
6,0
5,0
4,0
3,0
2,0
1,0
0,0
16,0
14,0
12,0
10,0
8,0
6,0
4,0
2,0
0,0
wheat price cumulative
index
wheat price
cumulative index
Figure 3. Dynamics of Wheat Price Growth and Costs of Money
Australia
Political obligation of any government is to guarantee satisfactory standards of living. It is problematically to
achieve in the environment of extreme world food prices growth. According to numerous analytical reviews the
recent unrests in Northern Africa and Arabic countries were happened partly due to the food commodity prices rally
commenced in summer 2010. Economists have to understand the current market environment and find the effective
instruments to prevent or overcome negative market consequences.
Our brief investigation has brought us to the general conclusion that the current economic and financial
environment form favorable conditions for food prices continuing growth in future. Though theories and empirical
observations verify the upward prices trend, it is the governments’ responsibility to influence on the character of
such growth. Incomes disparity and cheap money absorption call for a new approach to the governmental
regulations in XXI century. Wise governments should combine social aims together with the exigent requirements
for financial markets behavior. Appetite of portfolio investors for short-term high revenues should be restrained and
the governments should develop regulative means in the framework for long-term strategic development.
288
deposit rate,%
deposit rate, %
20,0
15,0
10,0
5,0
0,0
25,0
20,0
15,0
10,0
5,0
0,0
1970
1974
1978
1982
1970
1974
1978
1982
1986
1990
1994
1998
2002
2006
deposit interest rate, annual
wheat price index
1986
Greece
1990
1994
1998
2002
2006
deposit interest rate, annual
wheat price index
10,0
8,0
wheat price
cumulative index
6,0
4,0
2,0
0,0
35,0
30,0
25,0
20,0
15,0
10,0
5,0
0,0
wheat price
cumulative index

4 References
Alexander S. Effects of a Devaluation on a Trade Balance IMF Staff Papers. 1952. Vol. 2 Apr. (pp. 263-278)
Financial Market Highlights – May 2008: The Recent Financial Market Turmoil, Contagion Risks and Policy
Responses. Financial Market Trends, No. 94, Volume 2008/1 – June 2008, (pp. 20). //Url:
http://www.oecd.org/dataoecd/55/51/40850026.pdf
Friedman M. The Case for Flexible Exchange Rates. Readings in International Economics /Ed. R.E. Caves, H.G.
Johnson, Homewood (III.), 1968. (pp. 426)
Polterovich Vitor M. Mechanism of Global Economic Crisis and Problems of Technologies Modernization. The
Journal of New Economic Association, March 2009. //Url: http://www.econorus.org/sub.phtml?id=21
Ronald Trostle Global Agricultural Supply and Demand: Factors Contributing to the Recent Increase in Food
Commodity Prices. A Report from the Economic Research Service USDA, May 2008, (pp. 30). //Url:
www.ers.usda.gov
Stine O.C. Income Parity for Agriculture. The Conference on Research in Income and Wealth. Studies in Income
and Wealth, Volume 1, Part 8. (pp. 317-345) //Url: http://www.nber.org/chapters/c8143
The Common Agricultural Policy Explained. European Commission Directorate-General for Agriculture and Rural
Development, 2008, (pp. 20) //Url: http://ec.europa.eu/agriculture/
The 2008/2009 World Economic Crisis What It Means for U.S. Agriculture, A Report from the Economic Research
Service USDA, March, 2009, (pp. 30) //Url: http://www.ers.usda.gov/Publications/WRS0902/
289

THE INFORMATION CONTENT OF IMPLIED VOLATILITY IN THE CRUDE OIL FUTURES MARKET
Asyl Bakanova
University of Lugano and Swiss Finance Institute
E-mail: asyl.bakanova@usi.ch
Abstract. In this paper, we evaluate the information content of an option-implied volatility of the light, sweet crude oil futures traded at
New York Mercantile Exchange (NYMEX). This measure of volatility is calculated using model-free methodology that is independent
from any option pricing model. We do find that the option prices contain important information for predicting future realized volatility.
We also find that implied volatility outperforms historical volatility as a predictor of future realized volatility and subsumes all information
contained in historical data.
Keywords: Volatility forecasts, implied volatility, extreme value, crude oil.
1. Introduction
Financial market volatility plays a very important role in the theory and practice of asset pricing, risk management,
portfolio selection and hedging. Because of its importance, both market participants and financial academics have long
been interested in estimating and predicting future volatility.
Volatility models that fall into one of two categories, the ARCH family and the stochastic volatility family, have
been commonly used in modeling volatility for estimation and forecasting. These models are based on historical data.
Recently there has been a growing interest in extracting volatility from prices of options. This is because if markets are
efficient and the option pricing model is correct, then the implied volatility calculated from option prices should be an
unbiased and efficient estimator of future realized volatility, that is, it should correctly subsume information contained
in all other variables including the asset's price history.
The hypothesis that implied volatility (IV) is a rational forecast of subsequently realized volatility (RV) has been
frequently tested in the literature 1 . Empirical research across countries and markets so far has failed to provide a
definitive answer. Early research on the predictive content of IV found that IV explains variation in future volatilities
better than historical volatility (HV) (see, for example, Latane and Rendleman (1976), Chiras and Manaster (1978),
Schmalensee and Trippi (1978) and Beckers (1981)).
In subsequent research, Kumar and Shastri (1990), Randolph et al. (1991), Day and Lewis (1992), Lamoureuax and
Lastrapes (1993), and Canina and Figlewski (1993) found that IV is a poor forecast of the subsequently RV over the
remaining life of the option. Specifically, Day and Lewis (1992) and Lamoureux and Lastrapes (1993) find that IV has
some predictive power, but that GARCH and/or HV improve this predictive power and Canina and Figlewski (1993)
show the absence of correlation between IV and future RV over the remaining life of the option.
But the findings in the papers above are subject to a few problems in their research designs, such as maturity
mismatch and/or overlapping samples, among others. Overcoming these problems, more recent papers (e.g., Jorion
(1995), Fleming (1998), Moraux et al. (1999), Bates (2000), Blair et al. (2001), Simon (2003), Corrado and Miller
(2005)) confirm that IV still outperforms other volatility measures in forecasting future volatility, although there is
some evidence that it is a biased forecast. Christensen and Prabhala (1998), using monthly non-overlapping data, find
that IV in at-the-money one-month OEX call options is an unbiased and efficient forecast of ex-post RV after the 1987
stock market crash.
Szakmary et al. (2003) find that for a large majority of the 35 futures options markets IV, though not a completely
unbiased predictor of future volatility, outperforms the HV as a predictor of future volatility, and that HV is subsumed
by IV for most of the 35 markets examined. IV from options written on crude oil futures was examined by Day and
Lewis (1993). They compare it to a simple HV and out-of-sample GARCH volatility forecasts and find extremely good
forecasting performance for IV in this market. Although the results are somewhat mixed, the overall opinion seems to
be that IV has predictive power and therefore is a useful measure of expected future volatility. However, given the
equivocal results and conclusions across different options markets, it is clear that further research on the predictive
power of IV is needed.
Most of these studies have focused on individual stocks and stock indices, bonds, and currencies. In this paper we
analyze the implied volatility in the light, sweet crude oil market, which deserves an attention for a number of reasons.
1 Poon and Granger (2003) provide an extensive review of the literature on volatility forecasting.
290

Crude oil is the biggest and most widely traded commodity market in the world and the light, sweet crude oil futures
contract is the world's most liquid and largest-volume futures contract trading on a physical commodity. Since one of
the characteristics of prices in the oil markets is volatility, this market is a very promising area for testing volatility
models.
In particular, we test whether the IV is a better predictor of future RV and whether it reveals incremental
information beyond that contained in historical returns. We calculate IV of light, sweet crude oil futures from options
prices based on the concept of the fair value of future variance that appeared first in Dupire (1994) and Neuberger
(1994) and was improved further by various researchers (e.g., Carr and Madan (1998), Demeterfi et al. (1999), Britten-
Jones and Neuberger (2000), Carr and Wu (2006) and Jiang and Tian (2005)). This measure is calculated directly from
market observables, such as the market prices of options and interest rates, independent of any pricing model. Thus the
measurement error resulting from model misspecification is reduced.
As volatility proxy, we use the range-based -- or extreme value -- estimators proposed separately by Garman and
Klass (1980), Parkinson (1980), Rogers and Satchell (1991), and Yang and Zhang (2000). According to Alizadeh,
Brandt and Diebold (2002), the log range is nearly Gaussian, more robust to microstructure noise and much less noisy
than alternative volatility measures such as log absolute or squared returns. In addition, these estimators require the
daily open, close, high and low price data that are readily available for most financial markets. To our knowledge, this
is the first study to apply model-free methodology for implied volatility and the range-based estimators in the light,
sweet crude oil futures market.
Our findings can be summarized as follows. We find strong indications that the implied volatility obtained from
option prices, though slightly biased, indeed contains important information for predicting realized volatility at a
monthly frequency. It is also significant in the multiple regression where historical volatility is included, which means
that implied volatility subsumes the information content of historical volatility. The performance of option price based
predictions of future volatility is substantially improved by applying the instrumental variable approach to correct for
error in the predicted volatility variable. Finally, we find that implied volatility has better predictive power during more
volatile subperiod following September 11 terrorist attacks.
The paper proceeds as follows. Section 2 describes the data. Section 3 presents the methodology used for
construction of implied volatility, extreme value estimators and analysis of the information content of implied volatility.
Section 4 describes statistical properties of implied and realized volatilities the results for the forecasting performance
of the volatility index in terms of future realized volatility. In Section 5 we conduct robustness analysis. The
conclusions drawn from this study are presented in Section 6.
2. Data description
The dataset for this study contains daily time series of light, sweet crude oil futures and American-style options written
on these futures which are both traded on the New York Mercantile Exchange (NYMEX) for the period from January
02, 1996 through December 14, 2006. 2 On the NYMEX, futures and futures options are traded on the same floor, which
facilitates hedging, arbitrage, and speculation. Moreover, both markets close at the same time and their prices are
observed simultaneously which reduces the non-synchronicity biases and other measurement errors.
We consider only options at the two nearest maturities. When the time to the nearest maturity is less than seven
calendar days, the next two nearest maturities are used. We match all puts and calls by trading date, maturity, and strike.
For each pair, we drop strikes for which put/call price is less than $0.01. 3 Generally, a large number of options meet
these selection criteria. Since the options we use are American type, their prices could be slightly higher than prices of
the corresponding European options.
The futures high, low, open and closing prices are taken from the corresponding nearest futures contracts. We use
only the futures contracts with the same contract months as options to ensure the best match between the implied
volatility and the realized volatility calculated from subsequent futures prices. To eliminate any effect at the time of
rollover, we compute daily futures returns using only price data from the identical contract. Therefore, on the day of
rollover, we gather futures prices for both the nearby and first-deferred contracts, so that the daily return on the day
after rollover is measured with the same contract month. For the proxy of the risk-free interest rate, we use the rates of
the Treasury bill that expires closest to the option expiration date.
2
NYMEX does list European-style options. However, the trading history is much shorter and liquidity is much lower than for the American-style
options.
3
The reason for requiring option prices to exceed the given thresholds is that crude oil options are quoted with a precision of 0.01 USD.
291

Figure 1 plots the closing prices of the futures data. Figure 2 plots the returns series actually used in this study, which is
of the futures data.
Table 1 reports summary statistics for daily returns. Crude oil futures returns conform to several stylized facts
which have been extensively documented, such as fat tails and excess kurtosis.
3. Methodology
3.1. Range-based volatility estimators
Table 1. Descriptive statistics
Statistic Returns
Mean
Std. Dev.
Skewness
Kurtosis
Jarque-Bera
P-value
0.043
2.216
-0.296
5.250
612.04
0.000
Figure 1. Closing prices of light, sweet crude oil prices
Figure 2. Returns of light, sweet crude oil futures
The idea of using information on daily high and low prices, as well as the opening and closing prices, goes back to
Parkinson (1980) and Garman and Klass (1980) with further contributions by Beckers (1983), Ball and Torous (1984),
Wiggins (1991), Rogers and Satchell (1991), Kunitomo (1992), Yang and Zhang (2000).
Define the following variables: the opening price of the trading day t, the closing price of the trading
day t, the highest price of the trading day t, the lowest price of the trading day t. Traditionally, the
unconditional realized volatility of asset returns has been estimated using the series of closing prices as the daily
squared return:
292

(1)
However, using closing prices alone to estimate volatility ignores information contained in intraday high and low
prices. Assuming an underlying geometric Brownian motion with no drift, Parkinson (1980) developed an estimator
which uses the daily high and low prices of the asset for estimating its volatility:
(2)
Since the price path is not observable when the market is closed, Garman and Klass (1980) suggest a method to
mitigate the effect of discontinuous observations by including opening and closing prices along with the highest and
lowest prices as follows:
(3)
Rogers and Satchell (1991) relaxed the assumption of no drift and proposed an estimator which is given by:
(4)
Finally, Yang and Zhang (2000) proposed new improvements by presenting an estimator that is independent of any
drift and consistent in the presence of opening price jumps. This estimator can be interpreted as a weighted average of
the Rogers and Satchell (1991) estimator, the close-open volatility and the open-close volatility with the weights chosen
to minimize the variance of estimator:
(5)
with:
where , , and is the Rogers-Satchell (1991) estimator.
3.2. Model-free implied volatility
Britten-Jones and Neuberger (2000) proposed an implied volatility measure derived entirely from no-arbitrage condition
rather than from any specific model and calculated directly from market observables, independent of any pricing model.
In addition, it incorporates information from the volatility skew by using a wider range of strike prices rather just in-themoney
options.
This methodology, based on the theory for variance swap contracts and with pricing obtained from the full crosssection
of options prices, has been adopted by the Chicago Board of Exchange (CBOE) in constructing the monthly
implied volatility index, known as VIX, from on S&P500 index option prices. Following the VIX methodology with
some modifications, we calculate implied volatility using the following equation:
(6)
where: is the difference between strike prices defined as ; Ki is the strike price of the i-th
out-of-the-money option (a call if Ki > F and a put otherwise); Q(Ki,T) is the settlement price of the option with strike
price Ki; F is a forward index level derived from the nearest to the money option prices by using put-call parity, such
that ; K0 is the first strike below the forward index level F; T is expiration date
for all the options involved in this calculation; r is the risk free rate to expiration.
is calculated at two of the nearest maturities of the available options, T1 and T2. Then, we interpolate between
and to obtain an estimate at 30-day maturity:
(7)
where and denote the number of actual days to expiration for the two maturities.
3.3. The information content of implied volatility
IV has been regarded as an unbiased expectation of the RV under the assumption that the market is informationally
efficient and the option pricing model is specified correctly. Consistent with the previous literature, to test whether the
293

implied volatility index has a significant amount of information over the historical volatility, we examine the following
three hypotheses:
H1. Implied volatility is an unbiased estimator of the future realized volatility.
H2. Implied volatility has more explanatory power than the historical volatility in forecasting realized volatility.
H3. Implied volatility efficiently incorporates all information regarding future volatility; historical volatility contains no
information beyond what is already included in implied volatility.
To test the above hypotheses, we use the following regression models commonly used in the literature:
(8)
(9)
(10)
If, as our hypothesis H1 earlier stated, IV is an unbiased predictor of the RV, we should expect and
in regression (8). Moreover, if implied volatility is efficient, the residuals from regression (8) should be
white noise and uncorrelated with any variable in the market's information set. If, in accordance with hypothesis H2, IV
includes more information (i.e., current market information) than HV, then IV should have greater explanatory power
than HV, and we would expect a higher R 2 from regression (8) than regression (9). Finally, if hypothesis H3 is correct,
then when IV and HV appear in the same regression, as in (10), we would expect since HV should have no
explanatory power beyond that already contained in IV.
For the analysis on the information content of implied volatility, we use non-overlapping observations by
computing realized volatility separately for each calendar month, following the example of Christensen and Prabhala
(1998), since non-overlapping data results in more robust econometric findings.
4. Empirical results
Figure 3 shows the daily level of the implied volatility for the entire sample. In Figure 4 we present the various
estimates of daily realized volatility. The peaks of these estimates are approximately synchronous, but the general
behavior of the series differs, both in the range of variances and persistence phenomenon. Estimators using range data
are less volatile than the classical estimator. The Augmented Dickey-Fuller test strongly rejects the presence of a unit
root in all the series.
Descriptive statistics for the levels and logarithms of both realized and implied volatilities are provided in Table 2
for the entire sample in Panel A, for the first subperiod January 1996 through September 2001 in Panel B, and for the
second subperiod October 2001 through December 2006 - in Panel C. Both average implied volatility and average log
implied volatility exceed the means of the corresponding realized volatility.
Figure 3. Model-free implied volatility at daily frequency
294

RV C
RV P
Panel A: Full period - 01/1996 to 12/1996
Mean
St. Dev.
Kurtosis
Skewness
0.335
0.090
6.040
1.489
0.315
0.068
5.988
1.335
Panel B: Subperiod 01/1996 to 09/2001
Mean
St. Dev.
Kurtosis
Skewness
0.344
0.092
5.509
1.252
0.318
0.066
4.165
0.852
Panel C: Subperiod 10/2001 to 12/2006
Mean
St. Dev.
Kurtosis
Skewness
0.326
0.088
7.463
1.849
logRV C
0.312
0.071
7.906
1.802
logRV P
Panel A: Full period - 01/1996 to 12/1996
Mean
St. Dev.
Kurtosis
Skewness
-0.488
0.106
3.673
0.627
-0.511
0.088
3.779
0.536
Panel B: Subperiod 01/1996 to 09/2001
Mean
St. Dev.
Kurtosis
Skewness
-0.477
0.109
3.466
0.368
-0.507
0.087
3.298
0.218
Panel C: Subperiod 10/2001 to 12/2006
Mean
St. Dev.
Kurtosis
Skewness
-0.500
0.103
4.414
0.963
-0.515
0.089
4.682
0.888
Table 2. Descriptive statistics
RV GK
0.328
0.074
7.066
1.575
0.328
0.071
4.385
0.954
0.326
0.077
9.739
2.168
logRV GK
-0.494
0.089
4.170
0.683
-0.493
0.091
3.240
0.291
-0.496
0.089
5.568
1.162
RV RS
0.330
0.078
8.149
1.766
0.331
0.076
5.065
1.071
0.329
0.081
11.306
2.440
logRV RS
-0.492
0.093
4.492
0.746
-0.491
0.095
3.383
0.305
-0.493
0.091
6.259
1.328
Figure 4. Daily level of realized volatilities
RV YZ
0.307
0.077
5.574
1.898
0.309
0.075
5.374
1.221
0.306
0.079
9.207
2.604
logRV YZ
-0.524
0.097
4.666
0.838
-0.522
0.100
3.513
0.408
-0.526
0.094
6.645
1.424
IV
0.372
0.072
4.367
0.983
0.354
0.059
3.599
0.589
0.392
0.079
3.807
0.951
logIV
-0.437
0.080
3.126
0.412
-0.457
0.072
2.851
0.115
-0.415
0.084
2.819
0.477
Table 3 reports the ordinary least-square estimates for regressions (8) - (10). From the first regression it is seen that
implied volatility does contain information about realized volatility. However, we cannot conclude that the logarithm of
implied volatility is an unbiased estimator of realized volatility. The coefficient is 0.6722 and is significantly different
from zero, but also significantly less than unity although the intercept is statistically not different from zero at 5%
significance level. An F-test rejects the joint hypothesis and at 1% significance level.
This conclusion is found to be robust across a variety of asset markets (see Neeley (2004)) and has thus provided
the motivation for several attempted explanations of this common finding. As Christensen and Prabhala (1998)
suggests, the results may be affected by errors in variables (EIV) which induces a bias in both slope coefficients.
295

Consistent estimation in presence of the possible errors in variables problem may be achieved using an instrumental
variable method that we present in the next section.
Despite its biasedness, implied volatility remains a better predictor than past realized volatility. Indeed, taken alone,
historical volatility is statistically significant, 0.5001, but its predictive power is quite inferior to the implied volatility
predictive power. If we put in the same regression implied volatility and past volatility, we obtain interesting results.
We see that the slope coefficient for implied volatility remains statistically significant in the multiple regression. The
coefficient on historical volatility decreases strongly and is insignificant, which indicates that it does not contain
information beyond that in implied volatility. It is sufficiently precise that it subsumes the information content of
historical volatility.
Table 3. Information content of implied volatility: OLS estimates
Intercept IV HV Adj. R 2
0.1151
(0.044)
0.2076
(0.0352)
0.1132
(0.0397)
5. Robustness analysis
0.6722
(0.1145)
-
0.6547
(0.2167)
-
0.5001
(0.0945)
0.0266
(0.1981)
0.2561
0.1246
0.2504
Wald test Durbin
Watson
–
0.005 2.0385
We perform several exercises to verify the robustness of our results. We evaluate the impacts of error-in-variable
problems on our regressions where the implied volatility is used as a regressor and we analyze whether the information
efficiency of implied volatility varies significantly over different subsample periods.
5.1. Instrumental variable
The fact that the slope estimate is significantly below the null hypothesis of one could be either due to implied volatility
being a biased forecast or due to the bias induced by the error-in-variable problem. Christensen and Prabhala (1998)
assume that the error-in-variables (EIV) problem causes implied volatility to appear both biased and inefficient 4 and
propose to use instrumental variable framework as a way of correcting EIV problems in OEX implied volatility.
Within this framework, to correct for EIV, the following equations are used:
(11)
(12)
Under this procedure, implied volatility is first regressed on an instrument , which is correlated with true
implied volatility at time t but is not correlated with the measurement error associated with implied volatility sampled
one month later. With as the instrument, we estimate regression (11) using OLS. Then we reestimate
specifications (8) - (10) by replacing implied volatility, , with fitted values from the regression (11). We use the
same procedure for specification (10). First, we regress on both and and use the fitted values of
from this regression for specification (10).
Table 4 reports estimates based on this IV Procedure. Panel A reports estimates of the first-step regressions (11) and
(12), while Panel B reports the estimates of (8) and (10). The estimates in Panel B provide evidence that implied
volatility is much less biased and efficient. The point estimates of β1 in both specifications (8) and (10) are 0.892 and
0.824, respectively. Also the IV estimate of β2 is not significantly different from zero, indicating that implied volatility
is efficient.
4 The EIV problem has two effects. It generates a downward bias for the slope coefficient of implied volatility and the an upward bias for the slope
coefficient of past volatility, which explains the underestimation of implied volatility and the over estimation of past volatility. As a result, the usual
OLS will lead to false conclusions concerning implied volatility predictive power.
296
0.000
0.004
2.147
2.068

Panel A: first stage regressions estimates
Dependent variable:
Table 4. Information content of implied volatility: Instrumental variables estimates
Intercept IV HV Adj. R 2
0.128
(0.000)
0.1333
(0.000)
0.656
(0.000)
0.747
(0.000)
Panel B: second stage IV estimates
Dependent variable:
-
-0.127
(0.123)
0.428
0.434
Intercept IV HV Adj. R 2
0.02
(0.666)
-0.02
(0.683)
5.2. Subperiod analysis
0.88
(0.000)
0.83
(0.000)
-
0.06
(0.481)
0.29
0.29
Durbin-Watson
2.04
2.17
Durbin-Watson
2.00
To control for the time period effect, we reestimate specifications (8) and (10) separately for two separate subperiods:
pre-September 11 subperiod (January 1996 to August 2001) and a post-September 11 subperiod (October 2001 to
December 2006). Panel A of Table 5 reports estimates for the first subperiod and Panel B for the second subperiod.
Dependent variable:
Panel A: 01/1996 – 08/2001
Table 5: Information content of implied volatility: subperiod analysis
Intercept IV HV Adj. R 2
0.117
(0.046)
0.244
(0.000)
0.108
(0.072)
0.701
(0.000)
-
-
0.22
0.333
(0.008)
0.09
0.631
0.092
0.21
(0.002)
Panel B: 10/2001-12/2006
(0.500)
Intercept IV HV Adj. R 2
0.037
0.822
-
0.48
(0.395)
(0.000)
0.217
-
0.390
0.18
(0.000)
(0.000)
0.038
0.907
-0.092
0.48
(0.389)
(0.000)
(0.429)
2.14
Durbin-Watson
2.04
2.15
2.17
Durbin-Watson
2.04
The estimate of the slope coefficient for implied volatility in the first subperiod is smaller than corresponding estimates
in the second subperiod. These results suggest that there was a regime shift with implied volatility becoming less biased
after the attacks.
6. Conclusion
In this paper, we construct a model-free implied volatility from the options on light, sweet crude oil futures and
different measures for realized volatility for these futures. The main question we address is whether volatility implied
by the option prices predicts future realized volatility. We find that implied volatility does predict future realized
volatility alone as well as with the past volatility. We also find that historical volatility does not add any information
beyond that in implied volatility. Hence, we cannot reject the hypothesis that the volatility implied by option prices is an
efficient, although it is slightly biased estimator of realized volatility.
The implied volatility appears to be less biased and more efficient once we account for error-in-variables and apply
instrumental variable approach. This result provides support for the use of option pricing theory even for light, sweet
crude oil options. We also find that in the light, sweet crude oil market, the implied volatility performs better during the
more volatile period following September 11 terrorist attacks.
297
2.15
2.17

The main issue that deserves attention in the future research is to understand whether in the crude oil market the
bias is caused by OLS estimation method giving biased parameter estimates or by inefficiency of the options market,
i.e. implied volatility being an inefficient forecast of future volatility.
7. References
Alizadeh, S., M.W. Brandt and F.X. Diebold (2002). Range-based estimation of stochastic volatility models, Journal of
Finance, 57(3), 1047-1092.
Ball, C.A., and W.N. Torous (1984). The maximum likelihood estimation of security price volatility: Theory, evidence,
and application to option pricing, Journal of Business, 57, 97–112.
Bates, D.S. (2000). Post-87 crash fears in S&P 500 futures options, Journal of Econometrics, 94, 181-238.
Beckers, S. (1981). Standard Deviations Implied in Option Prices as Predictors of Future Stock Price Volatility, Journal
of Banking and Finance, 5, 363-381.
Black, F., and M. Scholes (1973). The Pricing of Options and Corporate Liabilities, Journal of Political Economy, 81,
637-659.
Blair, B., S.-H. Poon and S.J. Taylor (2001). Forecasting S&P 100 volatility: The incremental information content of
implied volatilities and high frequency index returns, Journal of Econometrics, 105, 5-26.
Britten-Jones, M., and A. Neuberger (2000). Option prices, implied price processes, and stochastic volatility, Journal of
Finance, 55, 839–866.
Canina, L., and S. Figlewski (1993). The informational content of implied volatility, Review of Financial Studies, 6, 3,
659-681.
Carr, P., and D. Madan (1998). Towards a theory of volatility trading. In Jarrow, Robert A. ed.: Volatility: New
estimation techniques for pricing derivatives, Risk Publications, London.
Carr, P., and Wu, L. (2006). A tale of two indices, Journal of Derivatives, 13, 13–29.
Chiras, D., and S. Manaster (1978). The information content of option prices and a test of market efficiency, Journal of
Financial Economics, 6, 213-234.
Christensen, B.J., and N.R. Prabhala (1998). The relation between implied and realized volatility, Journal of Financial
Economics, 50, 2, 125-150.
Corrado, C.J., Miller, T.W. (2005). The Forecast Quality of CBOE Implied Volatility Indexes, Journal of Futures
Markets, 25, 339-373.
Day, T.E., and C.M. Lewis (1993). Forecasting futures market volatility, Journal of Derivatives, 1, 33-50.
Demeterfi, K., Derman, E., Kamal, M., Zou, J. (1999). More than you ever wanted to know about volatility swaps,
Journal of Derivatives, 6, 9–32.
Dupire, B. (1994). Pricing with a smile, Risk 7(1), 18-20.
Fleming, J. (1998). The quality of market volatility forecasts implied by S&P 100 index option prices, Journal of
Empirical Finance, 5, 317-345.
Garman, M.B., and M.J. Klass (1980). On the estimation of security price volatilities from historical data, Journal of
Business, 53, 1, 67-78.
Jiang, G. J., and Y. S. Tian (2005). Model-free implied volatility and its information content, Review of Financial
Studies, 18(4), 1305-1342.
Jorion, P. (1995). Predicting volatility in the foreign exchange market, Journal of Finance, 50, 2, 507-528.
Kumar, R., and Shastri, K. (1990). The Predictive Ability of Stock Prices Implied in Option Premia, in Advances in
Futures and Options Research, Greenwich, CT: JAI Press, Vol. 4.
Kunitomo, N, (1992). Improving the parkinson method of estimating security price volatilities, Journal of Business, 65,
295–302.
Lamoureux, C., and W. Lastrapes (1993). Forecasting stock-return variance: toward an understanding of stochastic
implied volatilities, Review of Financial Studies, 6, 2, 293-326.
Latane, H., and R.J. Rendleman (1976). Standard deviations of stock price ratios implied in option prices, Journal of
Finance, 31, 2, 369-381.
Moraux, F., Navatte, P., and Villa, C. (1999). The Predictive Power of the French Market Volatility Index: A Multi
Horizons Study, European Finance Review, 2, 303-320.
Neuberger, A. (1994). The log contract: a new instrument to hedge volatility, Journal of Portfolio Management, Winter,
74-80.
298

Parkinson, M. (1980). The extreme value method for estimating the variance of the rate of return, Journal of Business,
53, 61-65.
Poon, S.-H., and C.W.J. Granger (2003). Forecasting financial market volatility: a review, Journal of Economic
Literature, 41, 2, 478-539.
Randolph, W.L., and M. Najand (1991). A test of two models in forecasting stock index futures price volatility, Journal
of Futures Markets, 11, 2, 179-190.
Rogers, L.C.G., and S.E. Satchell (1991). Estimating variance from high, low, and closing prices, Annals of Applied
Probability, 1, 504-512.
Schmalensee, R., and R.R. Trippi (1978). Common stock volatility expectations implied by option premia, Journal of
Finance, 33, 1, 129-147.
Simon, D.P. (2003). The Nasdaq Volatility Index During and After the Bubble, Journal of Derivatives, 11, 2, 9-22.
Szakmary, A., E. Ors, J.K. Kim and W.D. Davidson III (2003). The predictive power of implied volatility: Evidence
from 35 futures markets, Journal of Banking and Finance, 27, 2151-2175.
Wiggins, J. B. (1992). Estimating the volatility of S&P 500 futures prices using the extreme value method, Journal of
Futures Markets, 12, 265–273.
Yang D. and Q. Zhang, (2000). Drift-independent Volatility Estimation based on high, low, open, and close prices,
Journal of Business, 73(3), 477-491.
299

LITHIUM POWER: IS THE FUTURE GREEN?
Satyarth Pandey & Veena Choudhary
PGDM-DCP(2010-12)
Institute of Management Technology, Ghzaziabad, India
Email: Satyarthpandey@hotmail.com, Veena207@gmail.com, www.imt.edu
Abstract. The world has been suffering under the pressure of oil for the past so many decades. Sometimes it’s become the cause for
war among some nations, allegedly between USA and Iraq, and sometimes results in disharmony among some, Russia and Ukraine
2005-2006 gas dispute.
This helplessness has forced the research community to search for some alternative sources of fuel to drive our day-to day activities
with as minimum dependency and cost as possible. Some promising alternatives have taken the centre stage “hydrogen fuel cells”
being one of them ,but finding ways to store volatile hydrogen safely and bring down the costs of fuel-cell ingredients, which
currently include the fantastically expensive element platinum, has proved difficult.
Hence the proven alternative that has seen its rise as an economical and efficient fuel material is Lithium – the force behind the
electrical revolution in all sectors including electronics and automobile. The metal is "energy dense" yet “light” enough to be used in
cell phones, laptops or electric cars.
This paper analyses the rise and the future prospects of Lithium based products and whether it will be able to become an alternative
that can make the “black-gold” redundant and this long going search of humanity for something more viable and less scarce in nature.
This paper further presents an analysis of the nations that can gain a pivotal role if Lithium is to rise as the fuel source of the future.
Keywords: Lithium, Li, Fuel, Energy source, Lithium as a fuel, Lithium and oil
JEL classification: Q42
1 Introduction
The rising demand and high cost of oil has been an issue for the brain masters to consider for more than a decade .So
helpless is the scenario of world’s oil dependency that the world and people living in it have started clashing to get
their hands on to as much supply as possible to gain upper hand in world’s economic environment. This helplessness
and the degree of dependence on oil together have led people on search for different alternative of fuel than oil.
The two alternatives that have seen their rise in the recent times are hydrogen fuel cells and Lithium based
lithium-Ion (Li-Ion) batteries. Hydrogen fuel cells through their potent power generation have been a lucrative
prospect for the researchers. But, this option also suffers from its intrinsic pitfalls. Researchers have been fighting on
two of the major issues associated with Hydrogen cells namely reducing cost and improving durability.
Lithium on the other hand has risen like a superstar in the recent times. The advantage with Lithium based Liion
batteries is that they weigh less and have longer storage capacity. This is what makes it the preferred choice of
the current ever growing industry.
The purpose of this paper is to analyze the prospects associated with Lithium such as its demand, availability
and economical impact on the supplying nations.
2 Research & Methodology
The primary purpose of this paper is to highlight and analyze the possibility of using Lithium as a fuel. Answers to
questions such as “What would be the economic impact worldwide if Lithium replaces oil as a fuel?” sought after.
As the knowledge and information regarding the study is limited and more rigorous study has to be done to
collect all the required information the research can be categorized as an Exploratory Research. As the data was
collected and analyzed over a period of time, it was a longitudinal study and was qualitative in nature with all
information being theoretical.
Collation of secondary data and obtaining relevant information from it formed the major research methodology.
300

3 Lithium – The Future Fuel
3.1 What is Lithium?
The Lithium is considered to be the lightest solid element on earth. It is highly reactive metal and quickly tarnishes
in air after just a few minutes. Due to its high reactivity, it only appears naturally in the form of compounds.
According to the Handbook of Lithium and Natural Calcium, "Lithium is a comparatively rare element,
although it is found in many rocks and some brines, but always in very low concentrations. There are a fairly large
number of both lithium mineral and brine deposits but only comparatively a few of them are of actual or potential
commercial value. Many are very small; others are too low in grade."
For a long period of time, lithium was not considered much to have any practical applications in the real world.
But, over the years, lithium’s commercial applications have expanded tremendously. First, the pharmaceuticals
industry discovered that lithium had properties that affected brain chemistry (i.e. mood stabilizers used to treat bipolar
disorder). And, later, lithium was discovered to have ideal qualities for laptop, camera, and mobile phone
batteries.
Although new technologies are—by definition—subject to change, lithium is currently the best raw material for
making rechargeable batteries. Today, lithium-Ion (Li-Ion) batteries power most of the world’s laptops, mobile
phones and cameras. But lithium enthusiasts are also pinning their hopes on the electric car.
3.2 The Process and Players
The Lithium occurs majorly in two types of deposits: (a) Spodumene - a hard silicate mineral (i.e. glass), and (2)
Brine Salt Lake Deposits – dry salt lakes containing lithium chloride (in South America these are called “salares”).
Today, most of the world’s lithium comes from dry salt lakes because these deposits are more economically
viable for making Li-Ion batteries. These lakes result when pools of salt water containing lithium chloride (LiCl)
accumulate in places lacking drainage. Over the centuries the water evaporates leaving a dense layer of salt behind.
Underneath the salt crust is a layer of brine — salty groundwater with a high concentration of lithium chloride. It is
this brine that is pumped out and converted to lithium.
At present, the greatest part of the world’s accessible lithium reserves (over 80%) is in the so- called “Lithium
Triangle”, where the borders of Argentina, Bolivia, and Chile meet. (Evans, 2008; MIR 2008).
According to the United States Geological Survey (USGS), Chile provides 61% of lithium exports to the U.S.
and Argentina is the source of 36%. Chile has a reserve of an estimated 3 million tons and Argentina weighs in with
400,000 tons, while Bolivia’s reserve is calculated as being about 5.4 million tons.
Chile is the world’s largest producer — not only because Chile already has highly developed mining, transport
and processing infrastructure, but also because its climate and geography is favorable for the optimal solar
evaporation that is central to producing lithium. Neighboring Bolivia purportedly has the largest known reserves but
it does not currently produce any lithium. The cost of extracting lithium from Bolivia’s Salar de Uyuni is unknown.
Its ratio of magnesium to lithium – 30:1 units against 6.5:1 in Chile’s Atacama Desert – could put extraction costs at
about USD $5,000 a ton, compared with USD $1,500 in Chile. Currently, lithium is not traded as a commodity so
investors are looking into companies which sell the metal compounds.
301

Figure 1: World Based Lithium Reserves
Figure 2: Lithium Reserves in South American Countries
From figure 1 and figure 2 it is pretty much clear that South America holds the key to Lithium revolution and
the major player in this energy politics seems to be Bolivia who is yet to open its hands to the world.
It is clear that Bolivia is going to be a major player in this economic race and it has already started moving
keeping in mind its future perspective.
Bolivia’s president Evo Morales has a strict agenda of natural resource nationalization. This political agenda
aims to end of the decades of plunder, epitomized by the faded city of Potosi, where hundreds of thousands died in
the silver mines that funded Spain’s Armada. To this effect, Bolivia has ruled out private investment in producing
lithium – however it has left the door open to alliances for value-added products in which the government retains a
60% stake. Any partner would be required to build a plant to make lithium-battery powered cars.
The stakes for Bolivia are considerable. By taking advantage of these substantial natural resources, he can
realistically lift his country out of poverty and increase its role as a significant player on the global market. Even
302

prior to increased demand for lithium, the Bolivia’s mining industry was producing significant gains for the country,
leading to a 9.4% increase in its GDP.
3.3 Rise of Lithium
3.3.1 Industrial Demand and Application of Lithium
The wide variety of Lithium applications can be made from figure 3. Currently, Lithium is covering a huge sector of
applications and making its major impact in the world of batteries.
Figure 3: Industrial Application Of Lithium
Over the past years Lithium-Ion batteries have become the talk of the town. The commercial viability of these
batteries ranges from mobile to car powering batteries.
This has given the hope that Lithium power batteries could actually replace petroleum products in the
automotive industries.
The figure 4 below provides us a clear insight into the Lithium consumption by the end user in various
application areas and provides an estimate of what will be the consumption scenario in the year 2020.
303

Figure 4: Usage of Lithium in Different Industry Segments
The figure 3 & figure 4 were just to create the premise for strong demand of Lithium in various sectors. Due to
growth in each sector especially in the battery segment the demand for Lithium is expected to rise meteorically.
Figure 5: Trend Analysis of Demand of Lithium (2002-2020)
Figure 5 clearly depicts the rise in the demand over the years with an expected rise of up to 360% in the year
2020 from its demand in the year 2002.
3.3.2 Lithium – The Battery Segment
The lithium based battery segment may be termed as the critical area of Lithium usage. This segment is on the watch
list of many people owing to the fact that the Li-ion batteries are the most sought after for use in electric vehicles.
304

According to lithium producer Chemetall, there could be 6 million lithium powered vehicles by 2018. Each of
these vehicles would likely have a 10kWh (kilowatt hour) Li-Ion battery. And, each of these batteries requires
roughly 0.3kg of lithium metal equivalent per kWh of capacity. Therefore, this amounts to a total of 18,000 tons
(6,000,000 vehicles x 0.3kg x 10) of lithium (metal) or 84,000 tons of (Li2CO3). This number equates to almost the
entire 2008 world production of lithium and assumes that all of the lithium would be used to make electric vehicles
(i.e. none in laptops or cell phones). The idea of millions of cars powered by electric batteries is the fuel for the
current enthusiasm over lithium. According to the TRU Group, which is one of the most important consultants in the
lithium market, by 2020 38% of the lithium batteries used will be found in vehicles.
According to Chemetall analysis the following figure depicts the expected rise in the demand of electrical
vehicle.
Figure 6: Expected Rise in Demand of Electrical Vehicle
Though other battery options are available in the market such as Nickel Metal Hydride(Ni-MH), Nickel
Cadmium (Ni-Cd) but Lithium ions batteries tip the others in the competition because of the facts that it has higher
charge density, smaller size, lighter weight and a low self-discharge rate of approximately 5% per month, compared
with over 30% per month in common Ni-MH batteries 10% per month in nickel cadmium batteries. In addition to
the above points Lithium-Ion Batteries deliver performance that helps to protect the environment with features such
as improved charge efficiency without memory effect.
This is one area where Lithium is expected to resolve the global oil crises by lowering the demand for
petroleum for automotive purposes.
3.3.3 The Green Effect
Oil, currently, is also face of the heat on the ecological front. Oil, starting from its process of extraction to its
application is considered harmful for the environment because of large volume of “carbon footprints”.
In contrast to Oil Lithium is extremely environmental friendly option. The extraction process of Lithium is an
eco-friendly one with minimum energy consumption where in a small amount is consumed by pumping the brine
from the ground, while the energy used for the evaporation is solar. Also the desirable salts of lithium or potassium
305

chloride and the by-products such as magnesium or sodium chloride are substances that are already present in the
soil and not carcinogenic or dangerous to the environment.
The obvious question that arises is “what about after the battery discharges?” The answer to above question is
plain and simple. The Lithium ion battery is non-biodegradable, but the substances released are not harmful to the
environment.
3.3.4 Issues and Challenges
According to a research of William Tahil, Research Director-Meridian International Research, Lithium ion
batteries may be sustainable for portable electronics goods, it is not sustainable for EV applications. A balanced
scientific and economic analysis concerning the sustainability of Li-Ion technology for EV applications has not been
performed.
It, further, states that Lithium is not a traded metal but raw Lithium Carbonate was until recently valued at about
$1/kg. During 2005 and 2006 this rose to over $5/kg and apparently some Japanese Li-Ion battery manufacturers are
now offering $10/kg or $10,000 per ton, a tenfold increase in 2 years. This will only continue to rise as supply is
limited to the few brine and salt deposits.
The projected costs for Li-Ion batteries are still in the order of $300 - $450 per kWh even in high production
volume. A 30kWh Li-Ion battery would therefore cost at least $9,000: prohibitive for the mass market. Therefore, he
recommends that the factors – Performance, Safety, Cost, Simplicity, Industrial Availability as well as the very
significant Geostrategic and Environmental Protection implications of dependence on Lithium - should make the
ZnAir and NaNiFeCl batteries the prime choice for meeting the urgent need to reduce the consumption of oil
immediately at all costs or face the consequences of a meltdown in civilization.
4 Summary
To summarize the paper we say that over the past few years we have seen great rise in demand of Lithium in all
segments.
The segment which is has the biggest to gain from Lithium is the battery segment because of rise in demand in
the automobile industry.Associated with this rise are some apprehensions and doubts whether the world actually
holds the amount of Lithium capable to sustain the growing auto demand but we can only know this when that time
actually comes till then it is sure that Lithium will keep on making its gorund in this sector.
Linked with the rise of Lithium lies the fortune of South America especially Bolivia.The country is already
realizing the prospects it holds and is taking steps politically to maintain its edge.
5 References
Global Lithium Availability:A constraint for Electric Vehicles By Paul Gruber & Pablo Medina
The Trouble with Lithium : Implications of Future PHEV Production for Lithium Demand By William Tahil
Trends of R&D on Materials for High-power and Large-capacity Lithium-ion Batteries for Vehicles Applications
By Hiroshi Kawamoto\
http://www.lithiumalliance.org
http://www.darkmatterpolitics.typepad.com
http://www.fmclithium.com/
306

307

CORPORATE FINANCE
308

309

PREDICTING BUSINESS FAILURE USING DATA-MINING METHODS
Sami BEN JABEUR, Youssef FAHMI
Université de Bretagne-Sud (IREA)
benjabeursami@yahoo.fr, fyoussef@hotmail.com
Abstract: The aim of this paper to compare between two statistical methods in predicting corporate financial distress. We will use the
PLS (Partial Least-Squares) discriminant analysis and support vector machine (SVM). The PLS discriminant analysis (PLS-DA)
regression is a method connecting a qualitative variable dependent to a unit on quantitative or qualitative explanatory variables. The
SVM may be viewed as non-parametric techniques. It is based on the use of so-called kernel function which allows optimal separation
of data. In this work we propose to use a French firm for which some financial ratios are calculated.
Keywords: financial distress prediction, PLS discriminant analysis, Support Vector Machine
1. Introduction
This paper joins within the framework of the research works on the models of forecast of bankruptcies, which could
be used to detect the money troubles of the SME (small and medium-sized enterprise).
The research works on the forecast of the financial difficulties of companies are of a big importance for all the
partners of a company. From a manager point of view, have tools of forecast of the financial failures allows to take
in time strategic measures and management corrective appropriate to prevent these failures (Hooted and al 2007).
For other partners, such tools contribute to reduce the informative asymmetry with the company, to detect quickly
the vulnerable companies and to optimize the allocation of their capital (financier, human being, social). The
forecast of the financial difficulties so constitutes a means of diagnosis of the performance of companies and ends in
a classification of these last ones among the failing or not failing companies. The forecast of Failing companies can
help in particular to prevent the difficulties before they are translated by a financial crisis and to set up the necessary
measures (assistants, restructuring) before it is too late.
The use of the models of forecast of the bankruptcy of companies can vary according to tools and
methodologies used, the reserved sample, the explanatory variables chosen as well as the method of validation of the
results.
Before presenting the main methods of forecast used in this research work, it is useful to call back(to remind)
quickly the definitions of the financial distress.
2. Definition of the financial distress
During these last years, the annual flow of failures of companies did not stop growing and this tendency becomes
more marked during the periods of crisis. According to the INSEE( NATIONAL INSTITUTE FOR STATISTICS
AND ECONOMIC STUDIES), the number of failures of companies affected 52103 in 2010.
Table 1: Failures of French companies (published in the BODACC) between 1993 and 2010
Year 1993 1994 1995 1996 1997 1998 1999 2000 2001
Number of
failing
companies
55174 57494 52640 56858 53261 47525 42132 38346 36941
Year 2002 2003 2004 2005 2006 2007 2008 2009 2010
Number of
failing
companies
38202 43068 42368 43138 40407 42840 51254 53739 52103
310

The movement of the failures concerns in particular the companies from 10 to 50 employees and those whose
turnover is included between 500.000 euro and 2 million euro. In 2009 and in 2010, the fiscal measures of reflation
saved between 15.000 and 30.000 companies. SME (SMALL AND MEDIUM-SIZED ENTERPRISE), defined as
the companies of less than 250 employees and less than 50 million euro of turnover, benefited from 40 % of these
measures, that is 6,4 billion euro.
Even if the failure can have multiple reasons (fall in demand and difficulties of access to outlets, problem of
management, bad strategic choices and errors of management, competence and formation of the team, the
obsolescence of the production tool, the undercapitalization, the frauds), it is generally translated by financial
distress and a deterioration of the financial situation of the company (reduction of the activity, the decrease of the
profitability, the cash flow problems, the financial imbalance).
Wruck (1990) defines the financial distress as being a situation where the cash flows are insufficient to cover
the current bonds. These obligations can include the debts suppliers, the expenses. According to Baldwin and Scott (
1983 ), when the situation of a company degrades in the point where she cannot face her financial constraints, the
firm enters a state of financial distress. These same authors assure that this situation is the result of a bad economic
situation, a decline of their performances and a low quality of their management. From their part, Ooghe and Van
Wymeersch ( 1996 ) explain the financial distress in particular by the problems of liquidity (because of an
insufficient profitability and of a lack of resources) and of a high level of debts which pulls problems of solvency of
the company.
Several surrounding areas of forecast of the financial distress of companies were proposed since the end of the
sixties, using financial ratios and accounting data. The purpose is to differentiate the failing companies and the not
failing companies and to build an explanatory model of the failure of companies. The most used techniques are the
univarite approach, the multivariate approach, the discriminating analysis multiple shelf space, the multiple
regression, the logit regression, the models based on the artificial neuronal approach. These methods of forecast
present a joint representation concerning the evaluation of the risk of defect of companies, in the sense that they base
themselves on the financial analysis and the exploitation of the knowledge ex comment of the future of companies
(Redone 2004).
In this work, we are going to apply the discriminating approach and the SVM approach in the detection of the
failing companies. Both approaches will lean on a sample of 800 companies during period 2006 in 2008, as well as
on an recourse to 33 financial ratios.
3. Analysis method
3.1. The discriminating analysis PLS
There are several versions of the algorithm of univariate regression PLS1. They different at the level of the
normalizations (standardizations) and the intermediate calculations, but they give quite the same regression.
According to Bastien, Esposito and Tenenhaus ( 2005 ), the algorithm of the discriminating regression PLS ( PLS-
DA) can decompose as follows: we can repeat this process by using in the same way residues Y2, X21, X2k of the
regressions of Y, X1, Xk on t1, t2.
The algorithm PLS1 spells then:
Stage 1: X0=X; Y0=Y (initialization)
Stage 2: for h=1,…, r ;
X�h
�1y
h �1
Stage 2.1: w h �
y�h
�1yh
�1
Stage 2.2: Standards w in 1 h
X h �1w
h
Stage 2.3: t h �
w�
w
h
h
311

Stage 2.4: p
h
X�h
�1t
�
t�
t
Stage 2.5: X h � X h�1
� t hp'
h
y�h
�1t
h
Stage 2.6: ch
�
t�
t
Stage2.7: u
�
h�1
h
ch
Stage2.8: yh � yh�1
� cht
h
y
h
h
h
h
h
Once we retained the number of the constituents, we make a discriminating analysis on these constituents and not on
the variables of origin.
3.2. The principles of the method SVM
Without going into the technical details, we expose very briefly the principles of the method "Support Vector
Machines".
Support Vector Machines (SVM) are techniques of classification based on the statistical theory of the learning
(Vapnik on 1995, 1998). The SVM can be used to resolve problems of regression, that is predict the numerical value
of a variable, or a discrimination, that is decide to which class belongs a sample.
As method of classification, the approach SVM represents a kind of discriminating analysis generalized, made in a
space of rather big dimension so that exists a linear separation. She takes place in two stages. In the first one, a not
linear transformation makes pass of the space of origin in a space of bigger dimension but endowed with a scalar
product. Assist stage: in we look for a linear separator:
f ( x)
� a � x � b , which is a hyperplane to separate the whole learning points so that all points of the
same class are on the same side of the hyperplane. The latter must simultaneously fulfill two conditions:
- it separates the groups (accuracy of the model), in the sense f (x)
>0 => class A and f (x)
� 0 => class B;
- it is the farthest from all observations (model robustness), knowing that the distance of an observation x to the
hyperplane is a � x � b / a . the margin being
2 / a
2
Given points ( i, i ) y x with i y =1 if i x is in A and i y =-1 if i x is in B, find the linear separator f ( x)
� a � x � b
equivalent to finding a pair (a, b) simultaneously satisfies two conditions:
- For all i, ( a � x � b)
�1
(good separation).
-
yi i
2
a is minimum (maximum margin).
SVMs seek, among all possible hyperplanes, the one that maximizes the distance between the hyperplane decision
and the nearest points of each class.
When the two populations are not perfectly discriminated or separated but overlapping, a terme must be added to to
each of the two previous expressions.
The solution f (x)
expressed as a function of inner products x � x�
After transformation � , it is expressed in
terms of inner products �(
x) ��(
x�)
. The amount ) ( ) ( ) , ( x x x x k � � � � � � is called the nucleus. In the
algorithm, the kernel k and not � chosen and we can calculate k( x,
x�)
without showing � . The calculations are
then made in the original space and become simpler and faster. This is why it is called core or kernel machine.
Examples of cores include:
- Linear k( x,
x�)
� x � x�;
- polynomial
x�
) � ( x � x
d
; if d = 2, ( 1, 2)
x x
2
2
2x�2
) � ( xx
) ;
k ( x,
�)
�(
x) ��(
x�)
� ( x x�
� x �
1
1
312
x � and ) , 2 , ( ) (
2
2
� x � x x x x then
1
1
2
2

2
x�
x�
-
2
2�
Gaussian (RBF) k(
x,
x�
) � e ; one of the most commonly used
- Sigmoid k( x,
x�
) � tanh��
( x � x�)
��
� where � is the gain and � the threshold.
4. Methodological choice
4.1. The construction of the sample
Our approach to data collection involves several steps: the choice of the database, the selection of companies and the
indicators of financial failure.
4.1.1. Presentation of data:
We use the database DIANE (instant access to data from French companies for economic analysis) for our sample.
This database provides access to a fund composed of more than one million businesses, as these are the companies
that publish their annual accounts within the registries of commercial courts. Collected by Coface Services,
information on company accounts and is enriched with lots of related information. With ten years of historical
accounts, DIANE is the basis of financial information and general on French enterprises the most comprehensive.
4.1.2. The survey sample:
A company was considered deficient if it has been a first event declaration to the judicial tribunal of commerce
during 2009. The data studied is organized so that the accounting year is available for 2008, 2007 and 2006. The
final sample obtained upon completion of this rigorous selection process consists of 800 companies split into two
sub-samples: 400 healthy firms and 400 failing firms. This choice was dictated by itself, because of constraints
arising from the availability of identifying information of companies.
4.2. The selection of indicators and financial ratios
Selecting financial ratios has been a more logical and methodological choice in order to provide a relevant and
credible battery capable of meeting the objectives and expectations of external reviews. Financial ratios were
selected according:
- their recurrence in French literature (Bardos, 1995, Bank of France’s work) and international literature (Altman,
1968, 1984, Conan and Holder 1979; Rose and Giroux,1984, Remade, 2004).
- their relevance to financial analysis, incorporating the basic ratios in most existing models of detecting
bankruptcy: liquidity ratios, profitability management, productivity and financial structure. Thus, a series of 33
ratios (R01 to R33) was retained among those commonly used in literature and which have a significant
informational content in the analysis of the financial situation of the company.
5. Results
5.1. The classification of entreprises by the method of PLS discriminate analysis
PLS-DA analysis allowed retaining six components, the classification results are presented in the following table:
313

1, 2 or 3 year before failure Nature of the firm
Group
Trust
real
Number
%
healthy
H: healthy / D: defaulting
Table 2:Model validation (PLS-DA)
T-1: According to this table, we can say that the results obtained using the PLS-DA method are attractive
compared to discriminate analysis (DA1). Indeed, the rate of correct classification of companies is about 96.50%, an
error of type 1 (percentage of failing firms considered healthy) of 5.75% and a type 2 error (percentage of nonfailing
companies considered risky) of 1.25%. According to this analysis PLS-DA was able to identify 377 healthy
firms, a rate of 94.25%, with only five failing firms are classified as healthy, a rate of 98.75%.
T-2: In the group of healthy companies, the model holds 126 companies as being dysfunctional; thus they are
really not. Also in the group of failing firms, 137 of them are considered healthy, which is not the case.
The study of the “two years before failure” model, reveals a decrease in classification rate of healthy firms from
94.25% to 68.5%; and 98.75% of companies falling to 65.75%. So the more the forecast is, the less the accuracy of
the model is.
T-3: Looking at this table, we can say that the results obtained using the PLS method are attractive compared to
discriminate analysis. Indeed, the rate of correct classification of companies is about 60.5%, a second type error of
38.75% and a first type error is about 40.25%.
Regarding the predictive power of the model in the span of time from two to three years before the failure, we
find a drop in clearance rates in healthy firms from 68.5% to 61.25% and 65.75% of failing companies to 59.75 %.
Thus, the greater the forecast horizon is far, the more the ability of the model is reduced. In conclusion, we detect a
superiority of PLS regression in one, two and three years before failure compared to traditional discriminate
analysis. Indeed, one year before failure, the rate of correct classification of firms is 95.9% for AD1 and 96.5% for
the analysis PLS-DA1. Two years before the failure, the rate of correct classification is 64.8% to 67,125% for AD2
and PLS-DA 2. Regarding the third year, the rate of correct classification is 59.25% for AD 3 and 61.25% for PLS-
DA3. These results were confirmed by previous work applied to scientific fields other than the prediction of failure
by Nguyen and Roke (2004), Bastien et al. (2005), who believe that the role of restrictive assumptions of
discriminate analysis from the PLS regression can be successfully applied to financial data.
5.2. The classification of enterprises by the SVM method
Allocation group under the model
T-1 T-2 T-3
S D Total H D Total H D Total
377 23 400 274 126 400 245 155 400
Defaulting 5 395 400 137 263 400 161 239 400
healthy
94,25 5,75 100 68,5 31,5 100 61,25 38,75 100
Defaulting 1,25 98,75 100 34,25 65,75 100 40,25 59,75 100
The results of the classification of enterprises by the SVM method are presented in the following table:
314

1, 2 or 3 year Before the
failure
Group Trust
real
number
%
Nature of the firm
Table 2: Validation of SVM model
T-1: one year before failure, the rate of correct classification of firms is around94.875, a type 2 error of 7.5%
and a type 1 error of 2.75%. The SVM method was able to identify 370 of healthy companies, a rate of 92,5% while
30 failed businesses are classified as healthy firms. Similarly, it was able to identify 389 firms failed, a rate of
97.25%. Healthy companies classified as failing firms are 11 in number.
T-2: Two years before the failure, the SVM approach gave a rate of correct classification of the order of
65.625% with a type 2 error of 38% and a type 1 error of 30.75%. This method was able to identify 248 healthy 248
healthy firms, a rate of 62% while 152 companies are ranked among failed business. In addition, it was able to
identify 277 failed companies, a rate of 69.25%. Healthy companies classified as failing firms are 123.
T-3: Three years before the failure, the SVM approach gave a rate of correct classification of approximately
59,25% with a type 2 error of 29% and a type 1 error of 52.5 %. This method was able to identify 284 healthy
companies, a rate of 71% while 116 companies are ranked among failedbusinesses. In addition, it was able to
identify 190 failed companies, a rate of 47.5%. Healthy companies classified as failing firms are 210.
5.3. Comparison of two methods of classification
The following table compares the performance in anticipation of the PLS-DA and SVM approaches.
Healthy firms Failing firms Good ranking
PLS-DA SVM PLS-DA SVM PLS-DA SVM
T-1 94,25 % 92,5 % 98,75 % 97,25 % 96,5 % 94,875 %
T-2 68,5 % 62 % 65,5 % 69,25 % 67 % 65,625 %
T-3 61,25 % 71 % 59,75 % 47,5 % 60,5 % 59,25 %
Table 3: PLS-DA and SVM Performance approaches
Allocation group under the model
T-1 T-2 T-3
S D Total H D Total S D Total
healthy 370 30 400 248 152 400 284 116 400
Defaulting 11 389 400 123 277 400 210 190 400
healthy 92,5 7,5 100 62 38 100 71 29 100
Defaulting 2,75 97,25 100 30,75 69,25 100 52,5 47,5 100
The rates of correct classification provided by the two methods are very close. The rate of correct classification
of non-failing firms is the share of non-failing firms correctly classified all non-failing firms. Similarly, the rate of
correct classification is failing firms from failing firms correctly classified in all the failing companies.
The comparison of the overall performance of the two methods shows a clear superiority to the PLS-DA. A year
before the failure, the PLS-DA is more effective than SVM approach since it leads to a correct classification rate of
96.5%, against 94.875 for the SVM method. Two years before the failure rates were 67% for the PLS-DA and
65.625% for the SVM method. Finally, three years before failure, the rates were 60.5% for the PLS-DA and 59, 25%
for the SVM method.
315

The SVM method is superior to PLS-DA method only in terms of the classification of two failed businesses before
failure (69.25% against 65.5% for the PLS-DA), and for classification non-failing firms three years before failure
(71% against 61.25% for the PLS-DA).
6. Conclusion
As has been demonstrated by the previous studies and recent research in this field, we find that the predictive power
is weakened for models that use less information, or to recent models that are intended to be used to predict business
failure over a more distant period of time. In conclusion, we detected a clear strength in the PLS-DA method in
terms of one and three years before failure-analysis compared to the SVM method. We have shown that the
principles of PLS regression could be without difficulty discriminate analysis, PLS regression has an advantage in
terms of classification, taking into account important variables in the analysis, it can also solve the problem of
correlation between different variables while the majority of conventional methods grounded before such a problem.
7. Bibliography
Altman E.I. (1968), Financial ratios, discriminant analysis and the prediction of corporate bankruptcy, Journal of
Finance, 23(4), September, 589-609.
Albert A., Anderson J. (1984), On the Existence of Maximum Likelihood Estimates in Logistic Regression Models,
Biometrika, 71(1), 1-10.
Amemiya T. (1981), Qualitative Response Models: A Survey, Journal of Economic Literature, 19(4), 481-536.
Banque de france, Gourieroux C., Foulcher S., Tiomo A. (2003), La structure par termes des taux de défauts et
ratings, Cahiers Etudes et recherchesdel'Observatoiredesentreprises, Direction desEntreprises, Banque de France,
1.34.
Banque de France, Stili D. (2002), Détection précoce du risque de défaillance dans le secteur de la construction.
Cahiers Etudes et recherches de l’observatoire des entreprises, Direction des entreprises, Banque de France, 1.70.
Banque de France, Planes B. (2000), Détection précoce du risque de défaillance dans le secteur hôtels/restaurants:
SCORE BDFHR, Cahiers Etudes et recherchesdel'Observatoiredesentreprises, Direction des Entreprises, Banque de
France, 1.55.
Bardos M. (2008), Scoring sur données d’entreprises : instrument de diagnostic individuel et outil d’analyse de
portefeuille d’une clientèle, Revue MODULA, 38, 159 – 177.
Bardos M. (2001) : Analyse discriminante: application au risque et scoring financier, Dunod.
Bardos M. (1998), Detecting the risk of company failure at the Banque de France, Journal of Banking and Finance,
22, 1405-1419.
Bardos M., Zhu W.H. (1997), Comparaison de l'analyse discriminante linéaire et des réseaux neuronaux :
application à la détection de défaillance d'entreprises, Revue Statistique Appliquée, XLV (4), 65-92.
Bastien P, Vinzi V. E, Tenenhaus M. (2005), PLS generalized linear regression, Computational Statistics and Data
Analysis, 48-1.17-46.
Ben Jabeur.S.(2009), Prévision de la détresse financière des entreprises : approche par l’analyse discriminante
PLS, AFFI 2009 colloque international de l’association française de la finance, 13-14-15 Mai, Brest.
Chen W-S., Yin-Kuan Du Y-K.(2009), Using neural networks and data mining techniques for the financial distress
prediction model, Expert Systems with Applications 36, 4075-4086.
Chou H-I., Li H., Yin X. (2010), The effects of financial distress and capital structure on the work effort of outside
directors, Journal of Empirical Finance 17, 300–312.
Ding Y., Song X., Zen Y. (2008), Forecasting financial condition of Chinese listed companies based on support
vector machine, Expert Systems with Applications 34, 3081–3089.
Esposito Vinzi V, Trinchera L, Squillacciotti S, Tenenhaus M. (2008), REBUS-PLS: A response-based procedure
for detecting unit segments in PLS path modeling, Applied Stochastic Models in Business and Industry, 24-5, 439-
458.
Fisher N. (2004), Fusion statistique de fichiers de données, Thèse de doctorat, CNAM, Paris.
Fort G., Lambert-Lacroix S. (2005), Classification using Partial Least Squares with Penalized Logistic Regression,
Bioinformatics, 21(7), 1104-1111.
Franks J. R., Sussman O. (2005), Financial distress and bank restructuring of small to medium size U.K. companies,
The Review of Finance 9, 65–96.
316

Guilhot B. (2000), Défaillances d'entreprise : soixante-dix ans d'analyse théoriques et empiriques, Revue Française
de Gestion, 130, 52-67.
Hamza T., Bagdadi K. (2008), Profil et déterminants financiers de la défaillance des PME tunisiennes (1993-2003),
Banque & Marchés, 93, 45-62.
Hui L., Jie S., Jian W. (2010), Predicting business failure using classification and regression tree: An empirical
comparison with popular classical statistical methods and top classification mining methods, Expert Systems with
Applications 37, 5895–5904.
Jones S., Hensher D.A. (2004), Predicting firm financial distress: a mixed Logit model, The Accounting Review,
79(4), 1011-1038.
Koenig G. (1985), Entreprises en difficulté: des symptômes aux remèdes, Revue Française de Gestion,
janvier/février, 84-92.
Koenig G. (1991), Difficultés d’entreprise et inertie active », Direction et Gestion, 126-127.
Lin T. H. (2009), A cross model study of corporate financial distress prediction in Taiwan: Multiple discriminant
analysis, logit, probit and neural networks models, Neurocomputing 72, 3507-3516.
Li H., Jie Sun J. (2009), Predicting business failure using multiple case-based reasoning combined with support
vector machine, Expert Systems with Applications 36, 10085–10096.
Li H, Sun J, Wu J. (2010), Predicting business failure using classification and regression tree: An empirical
comparison with popular classical statistical methods and top classification mining methods, Expert Systems with
Applications 37, 5895–5904
Nguyen D., Rocke D. (2004), On partial least squares dimension reduction for micro-array-based classification: a
simulation study, Comput. Stat. Data Anal, 46, 407-425.
Ohlson J. A. (1980), Financial Ratios and the Probabilistic Prediction of Bankruptcy, Journal of Accounting
Research, 18-1,109-131.
Ohlson J.A. (1980), Financial ratios and the probabilistic prediction of bankruptcy, Journal of Accounting
Research, spring, 109-131.
Pages J., Tenenhaus M. (2001), Multiple factor analysis combined with PLS path modeling, Revue de Statistique
Appliquée, XLIV (2), 35-60.
Pindado J., Rodrigues L., Torre C. (2008), Estimating financial distress likelihood, Journal of Business Research 61,
995–1003.
Platt H. D., Platt M. B. (1990), Development of a Class of Stable Predictive Variables: The Case of Bankruptcy
Prediction, Journal of Business Finance and Accounting, 17, 1, pp. 31-51.
Platt H. D., Platt M. B. (2002), Predicting Corporate Financial Distress: Reflections on Choice-Based Sample Bias,
Journal of Economics and Finance, vol. 26, 2, 184-199.
Refait C. (2004), La prévision de la faillite fondée sur l’analyse financière de l’entreprise: un état des lieux,
Économie et Prévision, 162, 129-147.
Ross S.A., Westerfield R.W., Jaffe J.F. (2005), Finanzas corporativas, Ed. McGraw Hill (7ª edición).
Saporta G., Preda C. (2002), Régression PLS sur un Processus stochastique, Revue de statistique Appliquée, 50 (2),
27-45.
Stéphane T (2007), Data Mining et statistique décisionnelle, Éditions Technip, 3 éme édition revue et enrichie, juin
2007.
Sun J., Li H. (2010), Financial distress early warning based on group decision making, Computers & Operations
Research 36, 885 – 906.
Tenenhaus M. (1998), La régression PLS : théorie et pratique, Technip, Paris.
Tenenhaus M. (1995), Nouvelles méthodes de régression PLS, Les cahiers de recherche, CR 540.
Tenenhaus M. (2000), La Régression Logistique PLS, Journées d’Etudes en Statistique, Modèles Statistiques pour
données Qualitatives, 261-273.
Tenenhaus M., Gauchi J.P., Menardo C. (1995), Régression PLS et Applications, Revue de Statistiques Appliquée,
43(1), 7-63.
Wei-Sen C., Yin-Kuan D. (2009), Using neural networks and data mining techniques for the financial distress
prediction model, Expert Systems with Applications 36, 4075–4086.
Wruck K.H. (1990), Financial distress, reorganization, and organizational efficiency, Journal of Financial
Economies, 27, 419-444.
Zopounidis C. (1995), Evaluation du risque de défaillance de l'entreprise : Méthodes et cas d'application,
Economica, série Techniques de Gestion, Paris.
Zhongsheng H., Yu W., Xiaoyan Xu, Zhang B., Liang L. (2007), Predicting corporate financial distress based on
integration of support vector machine and logistic regression, Expert Systems with Applications 33, 434–440.
317

SIMULTANEOUS DETERMINATION OF CORPORATE DECISIONS: AN EMPIRICAL
INVESTIGATION USING UK PANEL DATA
Qingwei Meng, Birmingham Business School, University of Birmingham
Email: qxm645@bham.ac.uk
Abstract: We empirically investigate the joint determination of corporate investment, financing, and payout decisions under financial
constraints and uncertainty, using a panel of UK-listed firms. We model these corporate decisions within a simultaneous equations
system, where we treat each decision as endogenous and allow for their contemporaneous interdependences, as implied by the flow-offunds
framework. We find that capital investment and dividend payout as competing uses of limited funds are negatively interrelated,
while both of them are positively interrelated to net amounts of new debt issued during the corresponding period, suggesting the
existence of a joint determination of corporate decisions under financial constraints. We also offer the first attempt to examine the
effects of uncertainty on the set of corporate decisions within the simultaneous equations system. We find that the effect of uncertainty
on corporate investment is significant and positive, while that of uncertainty on dividend payouts is significant and negative. We
further observe that the simultaneity among the corporate decisions is more pronounced for firms which are financially moreconstrained,
while the effect of uncertainty is more significant for firms which are financially less-constrained. Accordingly, our results
suggest that financial constraints intensify the simultaneity among corporate decisions, and reduce managerial flexibility in response to
uncertainty. Therefore, this paper reveals new insight into the complex interdependence of corporate behaviour by UK-listed firms,
under financial constraints and uncertainty.
Keywords: Corporate investment; Debt financing; Dividend payout; Uncertainty; UK-listed firms
JEL classification: G31; G32; G35
References:
Abdul, H., & Wang, S. (2008). Uncertainty and investment evidence from a panel of Chinese firms. Structural
Change and Economic Dynamics, 19(3), 237-248.
Abel, A. B. (1983). Optimal investment under uncertainty. American Economic Review, 73(1), 228-233.
Almeida, H., & Campello, M. (2007). Financial constraints, asset tangibility, and corporate investment. Review of
Financial Studies, 20(5), 1429-1460.
Baker, H. K., Powell, G. E., & Veit, E. T. (2002). Revisiting managerial perspectives on dividend policy. Journal of
Economics and Finance, 26(3), 267-283.
Baum, C. F., Stephan, A., & Talavera, O. (2009). The effects of uncertainty on the leverage of nonfinancial firms.
Economic Inquiry, 47(2), 216-225.
Baum, C. F., Caglayan, M., & Talhadura, O. (2008). Uncertainty determinants of firm investment. Economics
Letters, 98(3), 282-287.
Bharath, S. T., Pasquariello, P., & Wu, G. (2009). Does asymmetric information drive capital structure decisions?
Review of Financial Studies, 22(8), 3211-3243.
Bloom, N., Bond, S., & Van Reenen, J. (2007). Uncertainty and Investment Dynamics. Review of Economic Studies,
74(2), 391-415.
Bond, S. R., & Cummins, J. G. (2004). Uncertainty and investment: An empirical investigation using data on
analysts’ profits forecasts. Finance and Economics Discussion Series Working Paper, No. 2004(20).
Bond, S., & Meghir, C. (1994). Dynamic investment models and the firm’s financial policy. Review of Economic
Studies, 61(2), 197-222.
Bond, S., Elston, J. A., Mairesse, J., & Mulkay, B. (2003). Financial factors and investment in Belgium, France,
Germany, and the United Kingdom: A comparison using company panel data. Review of Economics and
Statistics, 85(1),153-165.
Brailsford, T. J., Oliver, B. R., & Pua, S. L. H. (2002). On the relation between ownership structure and capital
structure. Accounting and Finance, 42(1), 1-26.
318

Brav, A., Graham, J. R., & Harvey, C. R., & Michaely, R. (2005). Payout policy in the 21st century. Journal of
Financial Economics, 77(3), 483-527.
Bulan, L. T. (2005). Real options, irreversible investment and firm uncertainty: New evidence from U.S. firms.
Review of Financial Economics, 14(3-4), 255-279.
Carruth, A., Dickerson, A., & Henley, A. (2000). What do we know about investment under uncertainty? Journal of
Economic Surveys, 14(2), 119-153.
Chay, J. B., & Suh, J. (2009). Payout policy and cash-flow uncertainty. Journal of Financial Economics, 93(1), 88-
107.
Cleary, S., Povel, P., & Raith, M. (2007). The U-shaped investment curve: Theory and evidence. Journal of
Financial and Quantitative Analysis, 42(1), 1-39.
Denis, D. J., & Osobov, I. (2008). Why do firms pay dividends? International evidence on the determinants of
dividend policy. Journal of Financial Economics, 89(1), 62-82.
Dhrymes, P. J., & Kurz, M. (1967). Investment, dividend, and external finance behaviour of firms. In R. Ferber
(Ed.), Determinants of Investment Behaviour (pp. 427-467). New York: Columbia University Press.
Ding, X., & Murinde, V. (2010). Simultaneous financial decision-making: Evidence from UK Firms. Strategic
Change, 19(1-2), 45-56.
Dixit, A. K., & Pindyck, R. S. (1994). Investment under Uncertainty. Princeton, N.J.: Princeton University Press.
Erickson, T., & Whited, T. M. (2000). Measurement error and the relationship between investment and Q. Journal
of Political Economy, 108(5), 1027-1057.
Fama, E. F., & French, K. R. (2001). Disappearing dividends: Changing firm characteristics or lower propensity to
pay? Journal of Financial Economics, 60(1), 3-43.
Fazzari, S. M., Hubbard, R. G., & Peterson, B. C. (1988). Financing constraints and corporate investment. Brookings
Papers on Economic Activity, 1988(1), 141-206.
Ferris, S. P., Jayaraman, N., & Sabherwal, S. (2009). Catering effects in corporate dividend policy: The international
evidence. Journal of Banking and Finance, 33(9), 1730-1738.
Flannery, M. J., & Rangan, K. P. (2006). Partial adjustment toward target capital structures. Journal of Financial
Economics, 79(3), 469-506.
Frank, M. Z., & Goyal, V. K. (2003). Testing the pecking order theory of capital structure. Journal of Financial
Economics, 67(2), 217-248.
Frank, M. Z., & Goyal, V. K. (2009). Capital structure decisions: Which factors are reliably important? Financial
Management, 38(1), 1-37.
Ghosal, V., & Loungani, P. (2000). The differential impact of uncertainty on investment in small and large
businesses. Review of Economic and Statistics, 82(2), 338-349.
Guariglia, A. (2008). Internal financial constraints, external financial constraints, and investment choices: Evidence
from a panel of UK firms. Journal of Banking and Finance, 32(9), 1795-1809.
Gugler, K. (2003). Corporate governance, dividend payout policy, and the interrelation between dividends, R&D,
and capital investment. Journal of Banking and Finance, 27(7), 1297-1321.
Hartman, R. (1972). The effects of price and cost uncertainty on investment. Journal of Economic Theory, 5(2), 258-
266.
Hennessy, C. A., Levy, A., & Whited, T. M. (2007). Testing Q theory with financing frictions. Journal of Financial
Economics, 83(3), 691-717.
Khan, T. (2006). Company dividends and ownership structure: Evidence from UK panel data. Economic Journal,
116(510), 172-189.
Lee, Y. T., Liu, Y. J., Roll, R., & Subrahmanyam, A. (2006). Taxes and dividend clientele: Evidence from trading
and ownership structure. Journal of Banking and Finance, 30(1), 229-246.
Lensink, R., & Murinde, V. (2006). The inverted-U hypothesis for the effect of uncertainty on investment: Evidence
from UK firms. European Journal of Finance, 12(2), 95-105.
319

Lensink, R., & Sterken, E. (2000). Capital market imperfections, uncertainty and corporate investment in the Czech
Republic. Economic of Planning, 33(1-2), 63-70.
Mahajan, A., & Tartaroglu, S. (2008). Equity market timing and capital structure: International evidence. Journal of
Banking and Finance, 32(5), 754-766.
Marsh, P. (1982). The choice between debt and equity: an empirical study. Journal of Finance, 37(1), 121-144.
Mason, R., & Weeds, H. (2010). Investment, uncertainty and pre-emption. International Journal of Industrial
Organization, 28(3), 278-287.
McCabe, G. M. (1979). The empirical relationship between investment and financing: A new look. Journal of
Financial and Quantitative Analysis, 14(1), 119-135.
McDonald, J. G., Jacquillat, B., & Nussenbaum, M. (1975). Dividend, investment and financing decisions:
Empirical evidence on French firms. Journal of Financial and Quantitative Analysis, 10(5), 741-755.
Miller, H. M., & Rock, K. (1985). Dividend policy under asymmetric information. Journal of Finance, 40(4), 1031-
1051.
Mougoué, M., & Mukherjee, T. K. (1994). An investigation into the causality among firms’ dividend, investment,
and financing decisions. Journal of Financial Research, 17(4), 517-530.
Mueller, D. C. (1967). The firm decision process: An econometric investigation. Quarterly Journal of Economics,
81(1), 58-87.
Myers, S. C., & Majluf, N. S. (1984). Corporate financing and investment decisions when firms have information
that investors do not have. Journal of Financial Economics, 13(2), 187-221.
Noronha, G. M., Shome, D. K., & Morgan, G. E. (1996). The monitoring rationale for dividends and the interaction
of capital structure and dividend decisions. Journal of Banking and Finance, 20(3), 439-454.
Peterson, P. P., & Benesh, G. A. (1983). A reexamination of the empirical relationship between investment and
financing decisions. Journal of Financial and Quantitative Analysis, 18(4), 439-453.
Ravid, S. A. (1988). On interactions of production and financial decisions. Financial Management, 17(3), 87-99.
Sant, R., & Cowan, A. R. (1994). Do dividends signal earnings? The case of omitted dividends, Journal of Banking
and Finance, 18(6), 1113-1133.
Shyam-Sunder, L., & Myers, S. C. (1999). Testing static tradeoff against pecking order models of capital structure.
Journal of Financial Economics, 51(2), 219-244.
Skinner, D. J., & Soltes, E. (2011). What do dividends tell us about earnings quality? Review of Accounting Studies,
16(1), 1-18.
Wang, D. H. M. (2010). Corporate investment, financing, and dividend policies in the high-tech industry. Journal of
Business Research, 63(5), 486-489.
320

LEVERAGE ADJUSTMENT AND COST OF CAPITAL
Sven Husmann, Michael Soucek & Antonina Waszczuk, European University Viadrina, Frankfurt (Oder), Germany
Email: euv36052@europa-uni.de, www.europa-uni.de
Abstract The classical approach to obtain unlevered cost of capital using CAPM is to adjust beta coefficients with one of the formulas
proposed by e.g. Modigliani & Miller (1959, 1963) or Miles & Ezzell (1985) to remove the financial effects of the leverage. As a
result we obtain the implicit unlevered beta. The static character of such an adjustment demands an assumption about the time point
related leverage ratio which is not in line with its time varying character. This study investigates the link between the capital structure
theory and the asset pricing models for the American stock market. The unlevered beta is a measure of the systematic risk of a purely
self-financed company. To capture the leverage dynamics it is reasonable to estimate the risk parameter from the in advance unlevered
returns. However, in a majority of cases the implicit unlevered risk parameters differ from the estimates based on adjusted returns.
This difference is statistically significant in 70% of cases for CAPM and in almost all cases for Fama & French (1993) risk loadings.
The error is smaller for Miles & Ezzell adjustment formula. We proved that the traditional approach to unlever risk parameters does
not provide correct results and the unlevered parameters have to be estimated from in advance adjusted returns. The market value of
debt has been estimated using Merton model (1974).
Keywords: CAPM, Fama-French Model, Modigliani-Miller Adjustment, Miles-Ezzell Adjustment, Unlevering, Asset Beta
JEL classification: G11, G12, G32
1 Introduction
The relevance of capital structure and the closely related estimation of firm-specific cost of capital is one of the most
important questions in modern finance. Based on the results of Modigliani & Miller (1959, 1963, further MM) the
relation between the values of levered and fictive unlevered enterprises was constructed which considers the
financing policy and tax shields. The well known studies of Hamada (1969, 1972), Bowman (1980) and Conine
(1980) create a link between the MM proposition III and Capital Asset Pricing Model of Lintner and Sharpe (1964)
or Mossin (1966) and derive the adjustment formulas for firm-individual beta risk factors. The resulting asset beta is
interpreted as the measure of systematic risk of a fictive unlevered firm which captures exclusively its business risk.
Miles & Ezzell (1980, 1985, further ME) extend the concept of MM and show an analogous relation for the cost of
capital under the assumption of value-oriented financing policy, i.e. under the assumption of deterministic debt-toequity
ratio. Both adjustment formulas are classical text book methods to separate financial and business risks.
Numerous authors investigate the empirical validity of MM theory. At the same time the question of an actual
consistency of adjustment formulas with classical asset pricing models is left unexplored. This aspect is discussed to
some extent only in relation to CAPM (Hamada (1972), Marston & Perry (1996)), while the context of multifactor
pricing models has not been investigated so far. A short theoretical introduction to Fama & French (further FF)
Three Factor Model was delivered by Lally (2004) and Dempsey (2009) but to our knowledge there has been no
empirical investigation dealing with this question.
The estimated parameters of commonly used asset pricing models reflect the information about a dynamic change of
leverage. However the use of classical adjustment formulas leave the non-static character of capital structure
neglected, since the leverage mimicking variable has to be approximated at a given point of time. The main aim of
our study is to quantify the error which emerges when the adjustment formula is applied directly on the risk
parameter. We show that the resulting loss of information is significant. If both analyzed asset pricing models are
valid then the commonly used method does not deliver a reliable reflection of the operative risk of the examined
firm. We argue that the direct estimation of unlevered parameters is a much sounder method of unlevering.
In comparison to earlier studies we extend our analysis to both most prominent capital structure models of MM and
ME. Additionally we check the relevance of the choice of possibly realistic leverage variable. The use of book
values for estimation of leverage ratio delivers little information concerning the dynamics of capital structure and
seems to be the least appropriate approximation for our study. Furthermore the estimation of leverage ratio in yearly
intervals causes abrupt changes of capital structure which may be incompatible with theory. For that reason the main
part of our paper uses the leverage in market values where the value of debt is estimated using Merton model (1974)
instead of the commonly used approximation with its book value.
321

The remainder of the paper is structured as follows: second and third section describe dataset and methodology. The
fourth section discusses the results. The fifth section concludes the paper.
2 Data
Our sample consists of stocks listed on NYSE and NASDAQ between 1989 and 2008. We limit our analysis to nonfinancial
1 stocks for which price, market capitalization (given as price times shares outstanding), pretax income,
income tax, interest expense on debt and total liabilities are available in Thomson DataStream Database for the
whole study period. To control for extreme observations we exclude stocks with leverage ratio higher than 10. Such
companies are either threatened with insolvency or even already insolvent. The final dataset consists of 418 stocks 2 .
Our study regards estimation of the following further variables: tax rate is proxy with quotient of income tax and
pretax income at time t 3 . The firm-specific cost of debt is calculated as the ratio of interest expenses and total
liabilities (book value) or market value of debt from Merton model (market value). In contrary to common
procedure of calculating costs of debt from Credit Default Spreads and Ratings we reach out to firms’ accounting
data. Because of the close relation between adjustment formulas and tax-shield we are interested in the exact actual
interest and tax expenses. The leverage ratio in book (market) values is given as the quotient of total liabilities and
book (market) value of equity.
2.1 Merton Model
The most studies on this topic approximate the leverage ratio using market value of equity and book value of certain
balance sheet debt accounts (Sweeney, Warga and Winters 1997). Nevertheless, according to the theory of capital
structure, the leverage in market values should be considered. In our paper, we estimate the market value of debt
using Merton model (Merton 1974). This model assumes the equity being an European option on the company’s
assets. It follows:
where
d
1
E
0
�rT
= V0
N ( d1
) � De N ( d 2 )
(1)
2
ln(
V0/
D)
� ( r ��
V /2) T)
=
and d 2 d1
��
V T
� T
=
V
� N(
d ) � V .
(2)
E E0 = 1 V 0
E0 and 0 V are the market value of equity and total value of a firm respectively, and e � und � V stand for the
volatilities of the market values, r is the risk free rate, T is the maturity of the hypothetical option and D is the
book value of debt. The parameters V0 and � V are unknown and need to be estimated. We solve the system of
equations iteratively using Newton-Raphson approach. Based on Sreedhar & Shumway (2008) we use the sum of
equity market value and total liabilities as a starting value for V 0 . The volatility � e is estimated using last 12
monthly returns and the start value for � V is ) ( =
E
� V � e . The market value of debt is the difference between
E � D
V0 and market value of equity.
1
Exclusion of financial stocks is a common procedure due to the different accounting standards relevant for such firms and characteristically high
leverage of such firms.
2
The survivorship bias problem is not relevant for our study.
3
If the calculated tax rate is higher than 39% it is replaced with this rate which is the highest tax rate in the U.S. since 1988 for incomes between
$100 000 and $335 000.
322

2.2 Methodology
We evaluate two asset pricing models, the Capital Asset Pricing Model of Sharpe, Linter and Mossin and the Fama
& French (1993) three factor model. To estimate risk parameters we use simple OLS regressions:
L
it
f
L
it
f
i
L
it
r � r = � � � ( E[
r ] � r ) � �
(3)
i
L
it
m
f
m
L
it
f
r � r = � � b ( E[
r ] � r ) � s SMB � h HML � �
(4)
The levered parameters for each stock i in our sample are estimated with rolling regression using five-years of past
monthly levered returns L
r it . 4
t
rm stands for monthly market return approximated with value-weighted average of all
stocks used in our study and f r for monthly risk-free return5 . Proxy portfolios SMBt and HMLt are constructed by
replications of Fama & French (1993) methodology. Risk parameters of fictive unlevered firms are calculated with
the help of adjustment formulas in spirit of MM and ME framework. In the case of non-zero covariance between
cost of debt and proxy portfolios the formula of Conine (1980) should be applied in relation to MM propositions
L
it
U
impl
it
U
impl
it
F
it
it
L
it
� = � � ( � � � )(1�
s ) L
(5)
F
By the adjustment in spirit of Hamada (1972) and Bowman (1979, 1980), � it is set to 0. To unlever according to
ME the following equation is to be applied
F
it
F
it
it
L U
impl
U
impl
F 1�
r (1�
sit
)
� it = � it � ( � it � � it )
Lit
(6)
1�
r
Here stands for tax rate, r for cost of debt and Lit for leverage ratio of an individual stock at time t. Lally
(2004) shows that this adjustment is analogically applicable for risk coefficients in the FF three factor model. We
use the simple mean of firm-specific leverage ratio and tax rate over the estimation period in the formulas above. As
U
it
impl
U
it
impl
U
it
impl
U
it
impl
a result we obtain � or b , s and h . We call this estimates, the implicit risk parametrs.
In most of studies which work with the assumption of possible covariance between risk parameters and the costs of
debt the latter ones are obtained with the help of credit default spreads. In contrary, we rely on relevant balance
F
it
sheet data. To obtain our cost of debt – risk loadings � we regress the cost of debt on given proxy portfolios.
Alternatively the risk parameters of an unlevered firm can be obtained directly from its unlevered returns. For MM
adjustment with firm-specific market or book value of debt, tax rate and cost of debt, the following holds:
while accordingly for Miles & Ezzell we have
U
it
r
L
F
rit
� rit
(1�
sit
) Lit
=
(1�
(1�
s ) L )
4 Because of our restriction each rolling regression uses exactly 60 observations when estimating risk parameters.
5 Risk-free rate of return is taken from Ken French online data library.
323
it
it
F
it
it
t
it
(7)

U
it
r
=
L
it
r
� r
F
it
1�
r
F
F
it
(1�
s ) L
1�
r
1�
rit
(1�
sit
) Lit
(1�
)
F
1�
r
The obtained unlevered returns are then regressed (in analogy to its levered counterparts) on proxy-portfolios of
CAPM and three factor model. The resulting regression coefficients are the directly estimated risk parameters of a
fictive unlevered enterprise
Uˆ
�it or U it b ˆ , U it
s ˆ and
it
it
F
it
it
U
hit ˆ . We can show that CAPM and three factor model have very
similar explanatory power for both levered and unlevered returns on both sector and industries level. The tables are
omitted in order to save space. As long as the applied models explain both fictive unlevered returns as well as the
returns observable on the market with similar power, the obtained unlevered risk parameters can be interpreted as
the true values of firms’ operative risk. If the adjustment formulas (5) and (6) used in the determination of the risk
parameters are correct, both parameters should have similar values.
3 Results
The main result of our study is obtained by comparison of the implicit and directly estimated risk parameters. We
analyze both prominent unlevering approaches based on Modigliani & Miller (1963) and Miles & Ezzell (1985)
using various leverage estimators. Since we divided the data into industry sector portfolios, we assume that the
operative risk within an industry sector or industry class is similar. Is this is the case then the standard deviation of
the unlevered parameters must be significantly lower than the one obtained from levered parameters (Hamada
(1972)). This fact could be verified for almost 100% of cases using the F-Test. The table has been omitted in order
to save space. As described above, because the leverage has a time–varying character, considering time point
estimates of the leverage ratio causes information loss when estimating operational (business) risk. The aim of our
study is to verify if this error is significant. The practitioners and most researchers use the arithmetic mean to
capture the leverage over the estimation period. To obtain implicit unlevered risk loadings we also use this
approximation in our study. The results do not change if we consider the values from the actual period of unlevering
which is also a common approach. Tables 1 and 3 summarize the results of the comparison of the implicit and direct
estimated unlevered parameters for CAPM and Fama & French (1993) models. In following, we assume the direct
parameters estimated from the unlevered returns are a valid measure of business risk. We observe that under the
assumption that the asset pricing models hold for unlevered returns the adjustment of the levered risk parameters
does not provide correct results. The influence of the variation in the leverage ratio and the tax shield cannot be
captured with implicit unlevered risk parameters. For the CAPM, debt-equity ratio measured in market values and
MM adjustment, we find only in 25% (8 of 32) of the cases the implicit and direct estimated unlevered parameters
being not significantly different on the industry sector level. This is also the case for only two out of nine industry
classes which is less than 25%. In around 80% of cases the difference is not equal to zero. Using the adjustment by
Conine (1980) and estimating the cost of debt with Merton model the number of insignificant differences slightly
increases, from 15,63% to 25%. Compared to MM, the ME adjustment is very sensitive to consideration of the cost
of debt beta. Omitting the
F
� increases the number of “correct” implicit estimates of unlevered risk loadings. The
reason is the important role the cost of debt plays in the ME adjustment. Even if the covariance between the cost of
debt and the market risk premium is generally insignificant, it still has an impact on the accuracy of the estimates.
324
(8)

Modigliani/Miller Miles/Ezzell
Leverage
Book
Corr.
F
K Sec. T

approximation of the market value of debt with the book value of total liabilities yields similar results as when using
Merton model. Nevertheless, the advantage of using Merton model is the possibility to calculate monthly varying
cost of debt.
Leverage
Book
Corr.
F
K
Modigliani/Miller
Sec. (3) (2) Ind. (3) (2)
Yes 0 (0%) 2 (6,25%) 0 (0%) 1 (11,1%)
No 0 (0%) 2 (6,25%) 0 (0%) 1 (11,1%)
Market
HML 0 (0%) 2 (6,25%) 0 (0%) 1 (0%)
Yes 0 (0%) 3 (9,38%) 0 (0%) 0 (0%)
No 0 (0%) 2 (6,25%) 0 (0%) 1 (11,1%)
TL to MV
HML 0 (0%) 3 (9,38%) 0 (0%) 0 (0%)
Yes 1 (3,13%) 2 (6,25%) 0 (0%) 0 (0%)
No 1 (3,13%) 2 (6,25%) 0 (0%) 0 (0%)
Book
HML 1 (3,13%)
Miles/Ezzell
2 (6,25%) 0 (0%) 0 (0%)
Yes 0 (0%) 1 (3,13%) 0 (0%) 1 (11,1%)
No 0 (0%) 2 (6,25%) 0 (0%) 0 (0%)
Market
HML 0 (0%) 1 (3,13%) 0 (0%) 0 (0%)
Yes 0 (0%) 2 (6,25%) 0 (0%) 1 (11,1%)
No 0 (0%) 3 (9,38%) 0 (0%) 0 (0%)
TL/MV
HML 0 (0%) 2 (6,25%) 0 (0%) 0 (0%)
Yes 1 (3,13%) 3 (9,38%) 0 (0%) 1 (11,0%)
No 0 (0%) 3 (9,38%) 0 (0%) 0 (0%)
HML 1 (3,13%) 2 (6,25%) 0 (0%) 0 (0%)
Table 3: Fama/French Three Factor Model Results
Table shows that for the all (or at least 2) directly estimated and implicit risk loadings the amount and share of industry sectors and classes are not
significantly different. We consider 32 industry sectors from 9 industry classes. The column Corr.
F
F
K stands for consideration of the � while
calculating the implicit unlevered risk loadings. In the HML row only the covariance between cost of debt and HML proxy portfolio is
considered.
4 Conclusions
To separate the financial and operational risk of a firm, the researchers and the practitioners use the adjustment
formulas. In connection with the common asset pricing models one may use this formulas to obtain risk parameters
of a hypothetically unlevered firm. The two most important adjustment formulas are based on the work of
Modigliani & Miller (1958, 1962) and Miles & Ezzell (1985). In the first step, the market parameters and risk
loadings for a levered firm are estimated. Next, these levered risk loadings are unlevered using adjustment formulas
mentioned above. Nevertheless, the static assumption about the firm’s leverage causes an information loss. Since the
company’s leverage has a time varying character, the approximation of the leverage aith a constat value causes
wrong results. We conclude that the direct unlevering of the stock returns and the direct estimation of unlevered
parameters out of the unlevered returns is the correct approach to obtain the operational risk. We also advise to
model monthly variation in the market value of debt, since the estimation of leverage ratio in yearly intervals, e.g.
using book values, causes abrupt changes of capital structure which may be incompatible with theory. Even if both
adjustment formulas contain the error described above, we show that it is smaller when using Miles & Ezzell
adjustement. Additionally, we that the results react only slightly on the consideration of the covariance between the
cost of debt and the market portfolios.
326

Leverage
Book
Corr.
F
K
Modigliani/Miller
MRP SMB HML
Yes 0.0442 0.0495 0.0737
Market
No 0.0442 0.0495 0.0739
Yes 0.0548 0.0575 0.0839
TL to MV
No 0.0539 0.0566 0.0825
Yes 0.0540 0.0570 0.0827
Book
No 0.0540
Miles/Ezzell
0.0571 0.0828
Yes 0.0003 0.0003 0.0005
Market
No 0.0015 0.0022 0.0033
Yes 0.0156 0.0221 0.0340
TL/MV
No 0.0016 0.0022 0.0035
Yes 0.0004 0.0003 0.0005
No 0.0015 0.0022 0.4869
Table 4: Fama/French Three Factor Model Average Root Mean Squared Error
The table shows the mean squared error between the direct estimated and implicit risk loadings for the FF three factor model. We calculate the
error for all stock between January 1995 and December 2008. The risk parametrs are estimated using five years rolling regression. The column
Leverage shows the way of approximation of the company’s leverage.
5 References
Bowman R. G. (1979). The Theoretical Relationship Between Systematic Risk and Financial (Accounting)
Variables. The Journal of Finance, 34(3), 617-630.
Bowman R. G. (1980). The Importance of a Market Value Measurement of Debt in Assessing Leverage. Journal of
Accounting Research, 18, 242-254.
Bharath S., & Shumway T. (2008). Forecasting Default with the Merton Distance To Default Model, Review of
Financial Studies, 21(3), 1339-1369.
Conine, T. (1980). Corporate Debt and Corporate Taxes: An Extension. The Journal of Finance, 35(4), 1033-1037.
Dempsey, M. (2009). The Fama-French three factor model and leverage: compatibility with the Modigliani and
Miller propositions. Investment Management and Financial Innovations, 6(1), 48-53.
Fama, E. F., & French, K. R. (1992). The Cross-Section of Expected Stock Returns. The Journal of Finance, 47(2),
427 465.
Fama, E. F., & French, K. R. (1993). Common Risk Factors in the returns on stocks and bonds. The Journal of
Financial Economics, 33(1), 3-56.
Fama, E. F., & French, K. R. (1997). Industry costs of equity. The Journal of Financial Economics, 43(1), 153-193.
Hamada, R. (1969). Portfolio Analysis, Market Equilibrium and Corporation Finance. The Journal of Finance,
24(1), 13-31.
Hamada, R. (1972). The Effect of the Firm's Capital Structure on the Systematic Risk of Common Stock. The
Journal of Finance, 27(2), 435-452.
Lally, M. (2004). The Fama-French Model, Leverage, and the Modigliani-Miller Propositions. The Journal of
Financial Research, 27(3), 341-349.
327

Lintner, J. (1965). The Valuation of Risk Assets and the Selection of Risky Investments in Stocks Portfolios and
Capital Budgets. Review of Economics and Statistics, 47(1), 13-37.
Marston, F., & Perry, S. (1996). Implied Penalties for Financial Leverage: Theory Versus Empirical Evidence.
Quarterly Journal of Business and Economics, 35, 77-97.
Merton, R.C. (1974). On the Pricing of Corporate Debt: the Risk Structure of Interest Rates. The Journal of Finance,
29(2), 449-470.
Miles, J.A., & Ezzell, J. R. (1980). The Weighted Average Cost of Capital, Perfect Capital Markets, and Project
Life: A Clarification. The Journal of Financial and Quantitative Analysis, 15(3), 719-730.
Miles, J.A., & Ezzell, J. R. (1985). Reformulating Tax Shield Valuation: A Note. The Journal of Finance, 40(5),
1485-1492.
Modigliani, F., & Miller, M. (1958). The Cost of Capital, Corporation Finance, and the Theory of Investment.
American Economic Review, 48(3), 261-297.
Modigliani, F., & Miller, M. (1963). Corporate Income Taxes and the Cost of Capital: A Correction. American
Economic Review, 53(3), 333-391.
Mossin, J. (1966). Equilibrium in a Capital Asset Market. Econometrica, 34(4), 768-783.
Schaefer, S. M., & Strebulayev, I. A. (2008). Structural models of credit risk are useful: Evidence from hedge ratios
on corporate bonds. The Journal of Financial Economics, 90(1), 1-19.
Sharpe, W. F. (1964). Capital asset prices: a theory of market equilibrium under conditions of risk. The Journal of
Finance, 19(3), 425-442.
Sweeney, R. J., Warga, A. D., & Winters, D. (1997). The Market Value of Debt, Market Versus Book Value of
Debt, and Returns to Assets. Financial Management, 26(1), 5-21.
328

DOES LEVERAGE AFFECT LABOUR PRODUCTIVITY? A COMPARATIVE STUDY OF LOCAL AND
MULTINATIONAL COMPANIES OF THE BALTIC COUNTRIES
Mari Avarmaa 1
Tallinn University of Technology
Akadeemia tee 3, Tallinn 12618
Phone: +372 6204 057
E-mail: mari_avarmaa@hotmail.com
Aaro Hazak
Tallinn University of Technology
Akadeemia tee 3, Tallinn 12618
Phone: +372 6204 057
E-mail: aaro.hazak@tseba.ttu.ee
Kadri Männasoo
Tallinn University of Technology
Akadeemia tee 3, Tallinn 12618
Phone: +372 6204 057
E-mail: kadri.mannasoo@tseba.ttu.ee
Abstract. This paper investigates the impact of leverage on labour productivity of companies operating in the Baltic countries, with a
focus on differences between local and multinational companies. We employ a fixed effects regression model on company level data,
covering the period from 2001 to 2008. Our results demonstrate that the impact of leverage on labour productivity is non-linear and it
differs dramatically between local and multinational companies. In the case of local companies, at low levels of leverage, an increase
in external financing tends to bring along an improvement in labour productivity, while at higher levels of leverage, an increase in
debt financing appears to result in a loss of labour productivity. For multinational companies, leverage does not seem to have a
significant impact on labour productivity. Although debt overhang is believed to be an issue in Baltic countries in general, local
companies with low leverage might be able to increase labour productivity by additional lending.
Keywords: labour productivity, leverage, multinational companies, local companies, Baltic countries
JEL classification: G32, D24
1 Introduction
The Baltic countries have been characterised by fast economic growth after regaining independence, demonstrating
GDP growth rates of average 5% during 1996-20092. The increase in private sector credit has been even more
striking. Between 2000 and 2007, the share of private sector credit in GDP doubled in Lithuania and tripled in
Estonia and Latvia. 3 After the period of rapid growth followed by the vast economic downturn of 2008 and 2009, the
question faced by the countries is, how to sustain development.
The neoclassical growth theory drawn on the seminal work of Solow (1956) demonstrates that productivity
growth is one of the main drivers for long-term per capita GDP growth. This relationship has found strong empirical
support (e.g. Hall and Jones 1999, OECD 2003) including in the context of Baltic countries (Schadler et al. 2006;
Arratibel et al. 2007). Understanding the determinants of productivity on a micro level, as well as the related
challenges and opportunities in a broader context, is therefore one of the key elements for exploring the paths for
economic growth.
While access to credit has been largely seen as a prerequisite for economic success (King and Levine 1993), the
recent lending booms have rather demonstrated the risks to company viability resulting from excessive debt
financing, highlighted by the global crisis of 2008/09. The impact of leverage on productivity and long-term growth
hence deserves closer scrutiny.
In the decades to come, the Baltic economies will be increasingly under pressure as a result of the ageing
population. As a consequence, output growth needs to be achieved with limited increase of labour force, and
improvements in labour productivity are essential for sustaining growth.
The Baltic countries have been successful in attracting foreign investments. Empirical evidence shows that
foreign direct investments play an important role in the labour productivity growth in the region (Bijsterbosch and
Kolasa 2009). It would therefore be interesting to understand the drivers of labour productivity of multinational
companies operating in the Baltic countries, and to identify whether these are significantly different from the
determinants of labour productivity of local companies.
1 Corresponding author
2 Eurostat
3 European Commission 2010
329

This paper focuses on company financing and ownership as determinants of labour productivity. Our aim is to
study the relationship between leverage and labour productivity comparing the MNCs and local companies in the
Baltic countries. Although the areas of capital structure and productivity have both been widely researched, the
linkage between company financing, ownership structure and labour productivity has received limited attention in
previous literature. Empirical evidence indicates that MNCs in the Baltic countries have more flexibility in their
financing decisions compared to local firms as the negative impact of credit constraints on leverage is much stronger
in the case of local companies (Avarmaa et al. forthcoming). We seek to investigate whether such flexibility leads to
any advantages for MNCs in achieving higher labour productivity. We perform a panel data regression analysis on a
sample of 3,676 Baltic companies covering the period of 2001 to 2008. According to our knowledge, this is the first
empirical research on the relationships between leverage and productivity covering the three Baltic countries. We
contribute to the literature by investigating whether the impact of leverage on labour productivity is different for
local companies and MNCs.
The article is set up as follows. The next section provides overview of the literature on the relationships
between leverage and productivity, as well as on the productivity differences between foreign and local firms,
Section 3 presents the regression model and data, Section 4 explains our results, and the last section concludes the
paper.
2 Literature overview
The classics of corporate finance theories offer some predictions on the influence of leverage on productivity. The
agency theory of capital structure explains that debt functions as a monitoring device over managers (Jensen and
Meckling, 1976), meaning that higher debt levels might thus result in higher efficiency and productivity. The
signalling theory of capital structure suggests that since better performing companies use the issuance of debt as a
signal about their quality (Ross, 1977), higher debt might be associated with higher productivity. On the other hand,
the debt overhang concept by Myers (1977) explicates that high leverage can cause firms to underinvest, since the
benefits of new capital investments accrue largely to debt holders instead of equity holders. This leads ultimately to
weaker firm performance. The law of diminishing returns shows that every additional unit of (labour) input results
in a diminishing increase in output. Coricelli et al. (2010) offers a similar explanation, pointing out that excessive
leverage could lead to overcapacity and therefore result in lower productivity.
As regards empirical research, several works show that leverage has a negative impact on productivity. Nucci et
al. (2005) find a negative relationship between leverage and productivity in a sample of Italian companies. They
show that there is a negative causal relationship from firm’s leverage to its propensity to innovate, and that
innovativeness leads to higher productivity. Ghosh (2009) makes a similar conclusion on a sample of Indian hightech
firms. Based on their quantile regression analysis of a sample of Portuguese companies, Nunes et al. (2007) also
show that the relationship between leverage and labour productivity is negative, except for the most productive firms
in the case of which higher leverage tends to increase productivity. In contrast to the above papers, Kale et al. (2007)
find on a sample of US companies a positive concave relationship between leverage and labour productivity, in line
with the agency theory. Hossain et al. (2005) analyse the components of productivity growth in US food
manufacturing industry and find that increases in dividends contribute to the productivity growth, in line with the
signalling theory.
Out of the limited research on the relationships between leverage and productivity as well as company
ownership in Central and Eastern Europe (CEE), Coricelli et al. (2010) have focussed on the impact of leverage on
total factor productivity (TFP) growth in twelve CEE countries (including Latvia) and found the relationship to be
non-linear. The impact of foreign ownership on TFP growth appeared to be insignificant, except for the subsample
with non-zero debt where a positive effect was found. Gatti and Love (2006) find on a Bulgarian sample that access
to credit is positively associated with productivity. Moreno Badia and Slootmaekers (2008) have investigated the
relationship between productivity and financial constraints in Estonia. They conclude that financial constraints do
not have an impact on productivity in most sectors, with the exception of R&D, where financial constraints have a
large negative impact on productivity. They find that companies with majority foreign ownership are more
productive.
330

Within the broad area of productivity related research, productivity differences between foreign and domestic
companies have received increasing attention during the last two decades. Pfaffermayr and Bellak (2000) have
summarised the main reasons for performance differences between foreign and domestic firms that have emerged
from existing research. They have pointed out the following factors: the firm-specific assets (such as production
process, reputation or brand) of multinational companies transferred from and to affiliates; the more narrow
specialisation of foreign-owned firms due to belonging to a larger group; the access of foreign-owned companies to
new technologies and opportunities for learning; different accounting practices, and different corporate governance
structures. The review paper by Bellak (2004) provides a detailed discussion on the sources of productivity gaps
between foreign-owned and domestic companies.
Empirical evidence on the productivity gap between foreign and domestic corporations is mixed, while the
existence of such a gap tends to be supported. Girma et al. (1999) find that there is a productivity and wage gap
between foreign and domestic firms in the manufacturing sector of the UK. Oulton (1998a) finds labour productivity
of foreign manufacturing plants to be higher compared to the UK-owned plants as well as labour productivity of
foreign companies to be better in the non-manufacturing sector in the UK (1998b). Greenaway et al. (2009) show
that there is a U-shape relationship between foreign ownership and productivity in China, suggesting that foreign
ownership is associated with improved performance only as long as it is accompanied by some degree of local
participation. In their quantile regression analysis of foreign-owned and domestic corporations in Greece, Dimelis
and Louri (2002) found that in the middle-productivity range, foreign firms exhibit higher efficiency while foreign
ownership does not matter among the very productive and least productive firms. Nunes et al. (2007) show that
foreign ownership increases labour productivity for all but the least productive firms in Portugal. In their plant-level
comparative analysis of labour productivity in foreign and domestic establishments in Canada, Globerman et al.
(1994), however, found no significant differences of productivity between these two groups after controlling for
factors such as size and capital intensity.
An area of research directly related to productivity differences is the study of productivity spillovers where the
focus is on the indirect benefits of FDI to productivity in the host country. Generally, productivity spillovers are said
to take place when the entry or presence of MNCs lead to productivity or efficiency benefits in the host country’s
local firms, and the MNCs are not able to internalize the full value of these benefits. (Blomström and Kokko 1998).
In this area, several studies have been performed based on data from CEE countries. Vahter (2004) has studied the
productivity spillovers in the manufacturing sector of Slovenia and Estonia and found that in both countries foreign
firms exhibit higher labour productivity compared to domestic companies. Positive spillover effects were found only
in Slovenia. Vahter and Masso (2006) have studied spillover effects in Estonia for 1995-2002 and found that
foreign firms demonstrate higher total factor productivity than domestic firms, the results regarding the existence of
spillover effects were mixed. Gersl et al. (2007) have investigated productivity spillovers in CEE countries and
found that the effects differ across countries and depend on various firm-, industry- and country-specific
characteristics.
Some authors have considered the impact of financing when analysing productivity gaps. Explaining the higher
productivity of foreign firms operating in UK compared to the local companies, Oulton (1998b) has pointed out that
local companies might face higher cost of capital than foreign-owned companies while foreign companies are likely
to be less constrained by the UK financial markets. Analyzing productivity gaps in Greece, Dimelis and Louri
(2002) have found a positive and significant effect of leverage as one of the control variables on labour productivity.
Greenaway et al. (2009), on the other hand, have found no significant relationship between leverage and labour
productivity on a Chinese sample of foreign and local companies.
In our study, we seek to link the relationship between financing and productivity with productivity gaps
between foreign (multinational) and local companies. Our main focus is on the impact of leverage on productivity of
MNCs and local companies in the Baltic countries.
331

3 The data and model
3.1 The model
We use panel data regression analysis to study the determinants of labour productivity. Drawing on the work of
Dimelis and Louri (2002), we use an augmented version of Cobb-Douglas production function for our empirical
model. Like these authors, we have included leverage as one of the independent variables. In order to allow for the
differences in the impact of leverage on productivity between multinational and local companies, we have included
and interaction term between leverage and dummy variable for the multinational companies. The model is
complemented with additional control variables derived from the findings of previous research. We model labour
productivity of an i-th company at time t as follows:
Log(Y/Lit) = β1LEVit +β2LEV 2 it + β3LEV×MNCit +β4LEV 2 ×MNCit +β5AGEit + β6AGE 2 it+ +β7SIZEit + β8TANGit +
β9HHIit + β10LEV×SKILLit+αi + uit
where α denotes firm-level fixed effects. The variables are explained in Table 1 below.
Variable Abbreviatio
n
Measurement Expected sign
Labour Productivity ln(Y/L) Ln (Sales/number of employees) dependent variable
Adjusted Leverage LEV (Short-term debt+Long-term liabilities)/(Total assets-
Current liabilities+Short-term debt)
Long-term Leverage LEV Long-term debt/ (Total assets-Current liabilities+Shortterm
debt)
non-linear
non-linear
Age AGE Number of years from incorporation non-linear
Size SIZE Ln of total assets +
Tangibility TANG Fixed assets/Total Assets -
Herfindahl index HHI Squared sum of market shares in all firms in the industry
based on 2-digit US SIC codes
Skill-intensive
industry
SKILL 1 if belonging to skill-intensive industry, otherwise 0 +
Multinationality MNC 1 if more than 50% owned by a foreign company,
otherwise 0
+
Table 1. Variables used in the regression model
We employ a fixed effects model since it helps to control for unobserved heterogeneity between the firms that is
constant over time and correlated with independent variables. The Hausman test showed that a fixed effects model
was to be preferred to a random effects model. Robust standard errors have been employed, which control for the
bias in the presence of heteroskedasticity and for the within-cluster serial correlation.
There are various ways for measuring productivity. Syverson (2010) has brought out issues related to the
measurement choice, concluding that the results of previous productivity research are generally not sensitive to the
method of measuring productivity. The most common measure of productivity in company-level research appears to
be TFP (e.g. Nucci et al. 2005, Ghosh 2009, Chen 2010, Coricelli et al. 2010). We however concentrate on studying
labour productivity as one of the key factors for economic growth under the aging population. Several previous
works on productivity have used value added per employee for measuring labour productivity (Globerman et al.
1999, Oulton 1998a, 1998b, Doms and Jensen 1998, Girma et al. 1999). Due to data limitations, we have not been
able to calculate value added for our data set. Therefore, logarithm of sales per employee was used as a measure of
productivity (Y/L), similarly to Dimelis and Louri (2002) and Pfaffermayr and Bellak (2000). Since sales are
332
+

influenced by inflation, real sales figures have been used. In order to arrive at real sales, industry-level price-index
deflators obtained from Eurostat have been used.
The main independent variable of interest is leverage (LEV). We have used two alternative measures for
leverage in our regression. First, we have included an adjusted measure of leverage, calculated similarly to several
studies on capital structure (Rajan and Zingales 1995, Jog and Tang 2001, Huizinga et al. 2008). This measure takes
into account that some assets on balance sheet are offset by specific non-debt liabilities. Previous studies on
productivity have used either the ratio of short and long term debt to net worth (Dimelis and Louri 2002) or the ratio
of total liabilities to total assets (Greenaway et al. 2009, Weill 2008) to calculate leverage. We believe that our
approach represents a more appropriate measurement of leverage. To make a consideration of the specifics of longterm
financing compared to short term financing, we have employed long-term leverage as the alternative to
adjusted leverage. While long-term investments should generally be financed from long-term financial resources,
long-term debt could be more difficult to obtain compared to short-term debt. We have used the same denominator
for the long-term leverage as for the adjusted leverage due to the above mentioned advantages of such measurement.
As some of the previous works have identified a non-linear relationship between leverage and labour productivity
(see Appendix 1), we have included both leverage (LEV), and squared leverage (LEV2), in our regression model.
Possible endogeneity of leverage was tested with Davidson-MacKinnon test and the exogeneity of leverage was
supported.
Previous literature has brought out that larger firms tend to benefit from economies of scale. A comprehensive
discussion of the reasons for the positive impact of company size on productivity is offered by Leung et al. (2008).
Empirical evidence confirms this positive relationship (Dimelis and Louri 2002, Greenaway et al. 2009, Moreno
Badia and Slootmaekers 2008). Company size has been measured in previous research by the logarithm of total
assets (Dimelis and Louri 2002, Greenaway et al. 2009) or by the number of employees (Kale et al. 2007, Hazak and
Männasoo 2010). We prefer the logarithm of total assets as in our case labour productivity is calculated based on the
number of employees. In order to eliminate the impact of inflation, real values of assets have been used.
In order to control for impact of the capital factor, we have included tangibility in the regression. The results of
previous research are inconclusive regarding the relationship between tangibility and productivity. Weill (2008) has
found a negative relationship between tangibility and cost efficiency in all the seven European countries included in
his sample. In addition to industry effects, he explains the relationship with the fact that a higher tangibility level
means lower working capital and therefore lower managerial performance. Greenaway et al. (2009) have found a
negative relationship between TFP and tangibility in China, while the influence of tangibility on labour productivity
remained insignificant. In their quantile regression analysis on a sample of Portuguese firms, Nunes et al. (2007)
found a negative relationship between tangibility and labour productivity in most cases, except for the firms with
very high productivity. They explain the outcome with the tendency that firms with high R&D investments tend to
have less fixed assets. Chen (2010), on the other hand, has found a positive relationship between collateral
(measured by tangible fixed assets by total assets) and TFP in China but the magnitude of the impact was small. She
concludes that firms’ ability to collateralise external borrowing can improve their productivity.
As productivity is considered to vary by the overall level of innovativeness in the industry, the impact of
leverage is observed separately for skill-intensive and non-skill intensive industries. We constructed a dummy
variable for skill-intensive industries (SKILL) and interacted this with the leverage variable (LEV×SKILL). The
classification of industries is based on the Pavitt taxonomy (Pavitt 1984) whereby industries are divided to four
classes – scale intensive, specialised suppliers, science based, and suppliers dominated. We consider the first three
classes as skill-intensive. The concordance between the two-digit US SIC codes and Pavitt’s categories is made
based on Greenhalgh and Rogers (2004). For the industries missing from the latter paper, we have used the
classification according to NACE codes from Pianta and Bogliacino (2008). Since level of labour productivity tends
to be industry-specific, we also control for the impact by including interaction terms between year and sector
dummies. For that purpose, industries are divided to four sectors (manufacturing, trade, construction, and service).
Productivity is considered to be influenced by product market competition. A comprehensive discussion on the
impact of competition on productivity is provided by Vahter (2006). He shows that in the empirical literature, a
positive relationship between competition and productivity is generally found. To control for the intensity of product
market competition, we have included the Herfindahl index (HHI) as an independent variable, similarly to Kale et
al. (2007). The index is calculated as a squared sum of market shares of all firms in the industry based on the 2-digit
333

US SIC-codes. However, as Vahter (2006) has pointed out, the Herfindahl index is based on a certain classification
of industries and thus could be misleading. Since there is no other appropriate proxy for competition available we
have used HHI despite the mentioned shortcoming.
We have divided the sample into two subsets – multinational and non-multinational companies. If more than
50% of a company is directly owned by a foreign company, it is classified as a multinational company (MNC) and
otherwise as a non-multinational (i.e. local) company. The terms “local company” and “non-multinational company”
are used interchangeably in this paper.
As the main focus of our article is the impact of leverage on productivity in the comparative perspective of
multinational and local firms, and considering that multinationality does not vary much over time, we have
interacted the MNC dummy with leverage (LEV×MNC) and the squared term of leverage (LEV2×MNC). In order
to test whether the coefficients for leverage and leverage squared are significantly different for MNCs and local
companies, the Chow test was performed. The independent variables were interacted with the MNC dummy and the
interaction terms were included in the regression. The null hypothesis that the coefficients are equal was rejected
with 5% significance.
3.2 Data
We have extracted data on companies operating in Estonia, Latvia and Lithuania from the Amadeus database
compiled by Bureau van Dijk. The database provides financial statements and information regarding the ownership
structure of private and publicly owned European companies. Our sample covers the period from 2001 to 2008.
Companies in the public utilities and financial sector (US SIC codes 4000-4999 and 6000-6999) are excluded from
the analysis due to their fundamentally different financial structure. Branches of foreign companies, cooperative
companies and partnerships are also excluded from the sample since their legal form makes financial decisions
different from regular limited liability companies. Similarly to Weill (2008), unconsolidated data are used. For every
company, data are included in the sample for those years for which financial information was available in sufficient
level of detail and all components of assets and liabilities were non-negative. In order to avoid the unjustified
influence of outliers to the regression results, the upper 2% of observations of labour productivity was eliminated.
For the same reason, for companies established before 1991, we have counted their age starting from year 1991
when the Baltic countries regained their independence and the regulatory frameworks for operating a company were
fundamentally changed. In case ownership data were missing for a certain year, the latest available information on
ownership was used. The companies for which no data on the number of employees were available were dropped
from the sample.
As our focus is on the analysis of the labour productivity of multinational companies compared to nonmultinationals,
we aimed to have an equal number of multinational and non-multinational companies in the sample.
We therefore included all multinational companies that met our criteria and randomly selected the same number of
local companies from each of the three countries. As a result, our sample consists of 18,401 company-year
observations whereof 50% belong to multinational companies. 50% of observations are from Estonia, 26% from
Latvia, and 24% from Lithuania. The total number of companies included in the sample is 3,676.
Appendix 2 provides descriptive statistics regarding the two subsamples. On average, MNCs appear to be twice
as productive as the non-multinational companies operating in the Baltic countries – the mean value of real labour
productivity of non-multinationals is 83 thousand euros per employee compared to 152 thousand euros in
multinationals. It becomes evident that MNCs are overall considerably bigger than local companies in terms of
sales, assets and headcount but are relatively less leveraged and carry relatively less tangible assets. As discussed in
Section 4, the different size and productivity levels for MNCs compared to local companies tend to have impact on
the relationship between leverage and labour productivity.
Average labour productivity by company age is presented in Figure 1. The figure reveals that for both local
companies and MNCs labour productivity increases rapidly after the start-up phase and starts decreasing gradually
thereafter.
334

Figure 1. Labour productivity (in thousands of euros) by company age
Average labour productivity has been constantly growing throughout the nine years in the Baltic countries
(Figure 2), especially for local companies. At the same time, the level of leverage has not increased considerably.
There is a slight upward trend for both adjusted leverage and long-term leverage of local companies for the boom
years of 2005 to 2007, and a drop in 2008, in relation with the financial and economic crisis. Trends of leverage of
MNCs, on the other hand, are relatively stable throughout the years under review.
Figure 2. Average labour productivity (in thousands of euros) and leverage of multinational and local companies by years.
Average labour productivity calculated for ten leverage brackets with a step of 10% (Figure 3) indicates that the
relationship between leverage and labour productivity tends to be non-linear and the nature of this relationship
seems to differ for local and multinational companies. The nature of this relationship is to be studied in a more
sophisticated regression analysis, presented in the next section.
335

Labour productivity
4 Results
180
160
140
120
100
80
60
40
20
0
Labour productivity by levels of
adjusted leverage
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
No of local No of MNC
Productivity of local Productivity of MNC
2,000
1,500
1,000
500
0
Number of observations
L a bo ur pr o duc tiv ity
180
160
140
120
100
80
60
40
20
0
Labour productivity by levels of longterm
leverage
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
No of local No of MNC
Productivity of local Productivity of MNC
Figure 3. Average labour productivity (in thousands of euros) by levels of adjusted and long-term leverage
In our panel regression analysis, we find support for the prediction that the relationship between leverage and labour
productivity in the Baltic countries is non-linear (See table 2 below). Namely, results for Model 1 show that at low
levels of adjusted leverage, increase in debt tends to bring along an increase in labour productivity, while in highly
leveraged companies an increase in debt financing appears to lead to a decrease in labour productivity. This outcome
is similar to Kale et al. (2007) who find a non-linear relationship between leverage and labour productivity on a
sample of US companies. Kale et al. (2007) argue that debt functions as a disciplining mechanism up to a certain
breakpoint starting from where the threat of financial distress or underinvestment due to debt overhang problem
begins to outweigh the incentives from the bonding mechanism. We believe that the positive coefficient of leverage
might also show that the lack of debt financing sets limits to companies’ productivity. In case of long-term leverage
(Model 2) the relationship between leverage and labour productivity is also non-linear. The squared term of leverage
is negative and significant while leverage remains insignificant, indicating that long-term leverage tends to have a
negative impact on leverage.
Labour Productivity No interaction terms With interaction terms Subsample with LEV>0
Adjusted
leverage (1)
Long-term
leverage (2)
Adjusted
leverage (3)
Long-term
leverage (4)
Adjusted
leverage (5)
2,000
1,500
1,000
500
0
N um ber o f o bse r v a tio ns
Long-term
leverage (6)
Leverage 0.15** 0.08 0.35*** 0.24** 0.27** 0.22*
Leverage 2
(0.08) (0.08) (0.10) (0.10) (0.11) (0.12)
-0.38*** -0.33*** -0.54*** -0.50*** -0.43*** -0.40**
(0.08) (0.11) (0.11) (0.15) (0.12) (0.16)
Leverage×MNC -0.41*** -0.35** -0.30** -0.12
(0.14) (0.16) (0.15) (0.18)
Leverage 2 ×MNC 0.33** 0.37* 0.22 0.08
(0.17) (0.32) (0.17) (0.24)
Age -0.01 -0.01 -0.01 -0.01 -0.02 -0.01
Age 2
(0.03) (0.03) (0.03) (0.03) (0.03) (0.02)
0.00 0.00 0.00 0.00 0.00 0.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)
Tangibility -0.56*** -0.56*** -0.57*** -0.57*** -0.61*** -0.59***
(0.05) (0.05) (0.05) (0.05) (0.06) (0.07)
336

Size 0.34*** 0.34*** 0.34*** 0.33*** 0.28*** 0.27***
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
HHI -0.07 -0.06 -0.06 -0.06 0.06 0.02
(0.15) (0.15) (0.15) (0.15) (0.17) (0.24)
Leverage×skill 0.10* 0.05 0.11* 0.07* 0.06 -0.07
(0.06) (0.07) (0.06) (0.07) (0.06) (0.08)
Constant 1.93*** 1.96*** 1.96*** 1.97*** 2.34*** 2.44***
(0.25) (0.25) (0.25) (0.24) (0.27) (0.27)
No of obs 18 401 18 401 18 401 18 401 14 458 10 371
R 2
0,92 0,92 0,92 0,92 0,93 0,93
Company fixed effects yes yes yes yes yes yes
Sector-year interactions yes yes yes yes yes yes
Year dummies yes yes yes yes yes yes
Table 2. Regression Results.4
Our results indicate that the relationship between financing and labour productivity is considerably different for
MNCs compared to local companies. For adjusted leverage as well as long-term leverage (Models 3 and 4
respectively), interaction term for leverage of MNCs is negative and significant, while the coefficient for interaction
term between squared leverage and MNC is positive and significant. This implies that labour productivity of MNCs,
in contrast to local companies, appears to decrease slightly as a reaction to increased leverage. The relationship is
illustrated in Figure 4. At high levels of leverage the negative impact of leverage on labour productivity is less
severe in case on multinational companies compared to the local companies.
The leverage breakpoint starting from where the impact of adjusted leverage for local companies becomes
negative is 42%, while the average level of adjusted leverage for local companies is 33% and the median value 27%.
Thus, for more than half of the observations, additional leverage might bring along improvements of labour
productivity. On the other hand, for MNCs additional leverage does not seem to have any positive impact on labour
productivity.
4 *, ** and *** indicate significance at 10%, 5%, and 1% level respectively. Robust standard errors in parentheses.
337

N u m b e r o f o b s e r v a tio n s ( c o lu m n s )
2,000
1,800
1,600
1,400
1,200
1,000
800
600
400
200
0
2,000
1,800
1,600
1,400
1,200
1,000
800
600
400
200
0
MNCs
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Leverage
Local companies
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
180
170
160
150
140
130
100
90
80
70
60
50
L a b o u r p r o d u c t iv ity ( lin e ) , E U R in th o u s a n d s
N u m b e r o f o b s e r v a t io n s ( c o lu m n s )
1,800
1,600
1,400
1,200
1,000
800
600
400
200
0
1,600
1,400
1,200
1,000
800
600
400
200
MNCs
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
0
Local companies
0% 10% 20% 30% 40% 50% 60% 70%
Leverage
80% 90% 100%
Figure 4. Impact of long-term leverage (left chart) and adjusted leverage (right chart) on labour productivity of MNCs and local companies.
The above relationship indicates that the availability of debt financing does not considerably limit the
productivity of MNCs operating in the Baltics unlike local companies. On the other hand, excess leverage does not
appear to jeopardize the productivity of MNCs as severely as local companies. A possible explanation for the
different impact of leverage on the labour productivity of MNCs might be that in their case the disciplining role of
debt is weaker compared to local companies. Belonging to a corporate group, MNCs are able to utilise intra-group
financial resources being therefore less dependent from external debt providers. In addition, the part of financing
that comes in the form of intra-group lending might not function as a monitoring device. Also, the threat of financial
distress is in general lower for MNCs due to potential support from the corporate group. As the size of the
operations of the subsidiaries of multinational groups’ in the Baltic countries tends to be relatively small compared
to the size of the entire group, providing financing for such operations is not likely to be significantly constrained. In
some cases, maintaining presence in the Baltic market might be of higher priority for corporate groups than
improving short-run results.
Another factor that is likely to cause differences in the impact of leverage on labour productivity of MNCs
compared to local companies is the relatively larger size of MNCs. According to the classical theory of diminishing
marginal return, an additional unit of labour input tends to result in a decreasing growth of output. Thus, there is less
potential for productivity growth, and probably less opportunities to impact this by additional leverage.
As mentioned in Section 3, compared to local companies, MNCs operate at much higher productivity levels
starting from the establishment (Figure 1). The higher productivity may be related to the advantages of MNCs
described in Section 2. It is therefore more difficult for MNCs to achieve additional improvements in productivity
by financing additional investments.
As regards the control variables, the relationship between labour productivity and company size is positive, as
expected, reflecting the existence of economies of scale in terms of labour productivity. A 1% change in assets
appears to result in a 0.34% change in productivity.We find the relationship between tangibility and labour
productivity to be negative. In our sample, a 1% reduction in tangibility results in an increase of labour productivity
by 0.57%. This might be explained by trade-off theory of capital structure (Kraus and Litzenberger 1973) whereby
firms rich in intangible assets have less collateral, higher bankruptcy risk and are thus less leveraged. At the same
time, innovative firms are proved to be highly productive (Egger and Keuschnigg 2010). The interaction term
between leverage and high-skilled industries is positive, supporting the argument that skill-intensive sectors seem to
benefit more from higher leverage than others. This could be related to the fact that their tendency to innovate
338
180
170
160
150
140
130
100
90
80
70
60
50
L a b o u r p r o d u c ti v ity ( l in e ) , E U R in th o u s a n d s

creates a need for higher financing while the innovative activities might not be transparent for outside agents, and
innovative firms are therefore credit rationed (Egger and Keuschnigg 2010). Our outcome is similar to the findings
of Moreno Badia and Slootmakers (2008) who found that credit constraints limit productivity in the R&D industry
in Estonia. However, in our case, skill-intensive industries include more industries than R&D. Company age and the
Herfindahl index remained insignificant in explaining labour productivity. In case of the Herfindahl index, this
might be related to the shortcomings of proxying as described in Section 3.
We have tested robustness of the results by running the regression on a limited subsample where the leverage of
observations is above zero. In case of adjusted leverage (Model 5), the outcome remained the same with the
exception that the interaction term between MNC and squared leverage is insignificant in explaining labour
productivity for the subsample, potentially explained by the small size of the subsample. For long-term leverage
(Model 6) the impact of leverage also stayed the same while the interaction terms remained insignificant since for
this regression the sample size was almost two times smaller compared to the main sample.
5 Summary
We found the impact of leverage on labour productivity to be considerably different for MNCs and local companies
operating in the Baltic countries. While there appears to be a positive concave relationship between leverage and
labour productivity for local companies, the impact is slightly negative in case of MNCs. We show that at moderate
levels of leverage (up to an adjusted leverage of 42%), lending tends to have a positive impact on labour
productivity of local companies in the Baltic countries. On the other hand, at high levels of leverage, there appears
to be a severe negative impact of leverage on labour productivity.
For MNCs additional leverage does not tend to bring along any improvements in labour productivity, while at
high levels of leverage, the negative impact of leverage on labour productivity is not as harmful as in case of local
companies. The different impact of leverage on MNCs can be explained by the weaker role of debt as a monitoring
device, lower bankruptcy risk, and the lower marginal productivity related to the higher productivity levels and
relatively bigger size of the multinational companies.
Although debt overhang problem is considered to be threatening the Baltic economies (Hertzberg 2010), for
many local companies, additional leverage might bring along improvements of labour productivity. This finding
may be of interest to companies, financing institutions and policy makers.
6 Acknowledgements
We are grateful to Professor Karsten Staehr, Dr Juan Carlos Cuestas, and the participants of the TTÜ doctoral
seminar for valuable comments and to the Estonian Science Foundation (grant no ETF8796) for financial support.
7 References
Arratibel, O., Heinz, F. Martin, R., Przybyla, M., Rawdanowicz, L., Serafini, R., Zumer, T. (2007). Determinants of
growth in the central and eastern European EU member states - a production function approach. European
Central Bank. Occasional Paper Series. No. 61.
Avarmaa, M., Hazak, A., Männasoo, K. (2011). Formation of the capital structures of multinational and local
companies in the Baltic countries. Baltic Journal of Economics. Forthcoming.
Bellak, C. (2004). How Domestic and Foreign Firms Differ and Why Does it Matter? Journal of Economic Surveys,
18 (4), 483-514.
Bijsterbosch, M., Kolasa, M. (2009). FDI and productivity convergence in central and eastern Europe an industrylevel
investigation. ECB, Working Paper Series. No 992.
Blomström, M. Kokko, A. (1998). Multinational corporations and spillovers. Journal of Economic Surveys. 12 (2),
247-278.
Castellani, D., Zanfei, A. (2004). Multinationals, Innovation and Productivity. Evidence from Italian Manufacturing
Firms. University of Urbino. http://www.fscpo.unict.it/pdf/Castellani%20(paper).pdf (06.12.2010)
339

Chen, M. (2010). Financial Effects and Firm Productivity: Evidence from Chinese Manufacturing Data. University of
Nottingham. 9 th Annual Postgraduate Conference, www.nottingham.ac.uk/GEP.
Coricelli, F., Driffeld, N., Pal, S., Roland, I. (2010) Leverage and Productivity Growth in Emerging Economies: Is
There A Threshold Effect? Brunel University Economics and Finance Working Paper Series, Working
Paper No. 10-21.
Dimelis, S., Louri, H. (2002). Foreign Ownership and Production Efficiency: a Quantile Regression Analysis. Oxford
Economic Papers, 54, 449-469.
Doms, M., and Jensen, B. (1998). Comparing Wages, Skills, and Productivity between Domestically and Foreign
Owned Manufacturing Establishments in the United States. In Geography and Ownership as Bases for
Economic Accounting, 235-258.
Egger, P. Keuschnigg, C. (2010). Innovation, Trade, and Finance. University of St. Gallen. Working Paper Series.
Working paper 2010-08.
Eurostat. http://epp.eurostat.ec.europa.eu/portal/page/portal/eurostat/home/ (12.02.11)
European Commision. (2010). Economic Policy Challenges in the Baltics. Cross-Country Study. Occasional Papers,
58
Gatti, R., Love, I. (2006). Does Access to Credit Improve Productivity? Evidence from Bulgarian Firms. World
Bank, Policy Research Paper No. 3921.
Gersl, A., Rubene, E., Zumer, T. (2007). Foreign direct investment and productivity spillovers: Updated evidence
from the Central and Eastern European countries. Czech National Bank, Working Paper Series, 8/2007.
Girma, S., Greenaway, D., Wakelin, K. (1999). Wages, Productivity and Foreign Ownership in UK Manufacturing.
Centre for Research on Globalization and Labour Markets, Research Paper No. 99/14.
Ghosh, S. (2009) Productivity and Financial Structure: Evidence from Indian High-Tech Firms. MPRA Paper, No.
19467.
Globerman, S., Ries, J., Vertinsky, I. (1994). The Economic Performance of Foreign Affiliates in Canada. The
Canadian Journal of Economics, 27 (1), 143-156.
Greenaway, D., Guariglia, A., Yu, Z. (2009). The More the Better? Foreign Ownership and Corporate Performance
in China. University of Nottingham, Research Paper No. 2009/05.
Greenhalgh, C., Rogers, M. (2004). The Value of Innovation: The Interaction of Competition, R&D and IP.
University of Oxford, Department of Economics. Economics Series Working Papers, 192.
Hall, R., Jones, C. (1999). Why Do Some Countries Produce So Much More Output per Worker than Others? The
Quarterly Journal of Economics, 114 (1), 83-116.
Hazak, A., Männasoo, K. (2010). Indicators of Corporate Default – An EU Based Empirical Study. Transformations
in Business & Economics, 9 (1), 62–76.
Herzberg, V. (2010). Assessing the Risk of Private Sector Debt Overhang in the Baltic Countries. IMF Working
Paper. WP/10/250.
Hossain, F., Jain, R., Govindasamy, R. 2005. Financial Structure, Production and Productivity: Evidence from the
U.S. Food Manufacturing Industry. Agricultural Economics, Vol. 33, No.S3, 399-410
Huizinga, H., Laeven, L., Nicodame, G. (2008). Capital Structure and International Debt Shifting. Journal of
Financial Economics, 88 (1), 80-118.
Jensen, M., Meckling, W. (1976). Theory of the Firm: Managerial Behavior, Agency Costs, and Ownership
Structure. Journal of Financial Economics, 3 (4), 305-360.
Jog, V., Tang, J. (2001). Tax Reforms, Debt Shifting and Tax Revenues: Multinational Corporations in Canada.
International Tax and Public Finance, 8 (1), 5-25.
Kale, J., Ryan, H. Wang, L. (2007). Debt as a Bonding Mechanism: Evidence from the Relations between Employee
Productivity, Capital Structure, and Outside Employment Opportunities. 18th Annual Conference on
Financial Economics and Accounting, NYU, http://w4.stern.nyu.edu/salomon/docs/conferences/Kale-Ryan-
Wang-.pdf. (06.12.2010)
King, R., Levine, R. (1993). Finance and growth: Schumpeter might be right. Quarterly Journal of Economics, 108
(3), 717-737
Kraus, A. Litzenberger, R. (1973). A State Preference Model of Optimal Financial Leverage. Journal of Finance, 28,
911-922.
Moreno Badia, M., Slootmaekers, V. (2008). The Missing Link Between Financial Constraints and Productivity.
LICOS Discussion Paper Series, Discussion Paper 208/2008.
Myers, S. (1977). Determinants of Corporate Borrowing, Journal of Financial Economics, 5, 147-75
Myers, S. (1984). Capital Structure Puzzle. NBER Working Paper Series, No. 1393.
340

Nucci, F., Pozzolo, A., Schivardi, F. (2005). Is firm’s productivity related to its financial structure? Evidence from
microeconomic data. Rivista di Politica Economica, SIPI Spa, 95(1), 269-290.
Nunes, P., Sequera, T., Serrasqueiro, Z. (2007). Firms’ leverage and labour productivity: a quantile approach in
Portugese firms. Applied Economics, 39 (14), 1783-1788.
OECD. The Sources of Economic Growth in OECD Countries.
http://www.oecd.org/dac/ictcd/docs/otherdocs/OtherOECD_eco_growth.pdf
Oulton, N. (1998a). Investment, Capital and Foreign Ownership in UK Manufacturing. NIESR Discussion Paper. No
141.
Oulton, N. (1998b). Labour Productivity and Foreign Ownership in the UK. NIESR Discussion Paper. No 143.
Pavitt, K. (1984). Sectoral Patterns of Technical Change: Towards a Taxonomy and Theory. Research Policy. 13,
343–373
Pfaffermayr, M., Bellak, C. (2000). Why Foreign-Owned Firms Are Different. A Conceptual Framework and
Empirical Evidence for Austria. HWWA Discussion Paper Series.
Pianta, M., Bogliacino, F. (2008). The Impact of R&D and Innovation on Economic Performance and Employment:
A Quantitative Analysis Based on Innovation Survey Data, University of Urbino, Faculty of Economics.
Rajan, R., Zingales, L. (1995). What Do We Know about Capital Structure? Some Evidence from International Data.
Journal of Finance, 50 (5), 1421-1460.
Ross, S. (1977). The Determination of Financial Structure: The Incentive Signaling Approach. Bell Journal of
Economics, 8, 23-40
Schadler, S. Mody, A., Abiad, A., Leigh, D. (2006). Growth in the Central and Eastern European countries of the
European Union. IMF. Occasional Paper. No 252.
Solow, R. M. (1956). A Contribution to the Theory of Economic Growth. Quarterly Journal of Economics, 70, 65-94
Syverson, C. (2010). What Determines Productivity. NBER Working Paper Series, No 15712.
Vahter, P. (2004). The Effect of Foreign Direct Investment on Labour Productivity: Evidence From Estonia and
Slovenia. Tartu University Press.
Vahter, P. (2006). Productivity in Estonian Enterprises: The Role of Innovation and Competition. Bank of Estonia
Working Paper Series, No. 7.
Vahter, P., Masso, J. (2006). Home versus Host Country Effects of FDI: Searching for New Evidence of Productivity
Spillovers. William Davidson Institute. Working Paper No. 820.
Weill, L. (2008). Leverage and Corporate Performance: Does Institutional Environment Matter? Small Business
Economics, 30 (3), 251-265
341

Appendix 1. Summary of previous studies on the impact of leverage on productivity.
Authors Producitivity measure Formula for leverage Sign for leverage
Dimelis and Louri 2002 Labour productivity (Short-term debt+long-term debt)/total assets +
Nucci et al. 2005 TFP Debt/total assets -
Kale et al. 2007 Labour productivity (Book value of long-term debt+short-term debt) /
(book value of debt+market value of equity)
Nunes et al. 2007 Labour productivity Total liabilities/total assets -
non-linear
Weill 2008 Cost efficiency Total liabilities/total assets varies by country
Ghosh 2009 TFP Total debt/total assets -
Greenaway et al. 2009 TFP Total liabilities/total assets -
Coricelli et al. 2010 TFP growth Total debt/total assets non-linear
Appendix 2. Descriptive statistics for multinational and local companies. (Monetary values in thousands of
euros)
Wilcoxon
rank-sum
Mean Median Sd Min Max No of obs test (z)
Total Assets
Equity
Long-term Debt
Short-term
Debt
Sales
Net Profit
Adjusted
Leverage
Long-term
Leverage
No of
employees
Labour
Productivity
Age
Tangibility
Local 1,959 749 5,473 0 174,424 9,282 -45.8***
MNC 5,673 1,933 12,658 1 272,140 9,119
Local 872 267 3,400 0 156,138 9,282 -35.2***
MNC 2,571 633 7,073 0 132,309 9,119
Local 294 17 1,620 0 64,669 9,282 8.6***
MNC 744 0 4,133 0 104,508 9,119
Local 213 20 936 0 22,489 9,282 3.6***
MNC 510 7 1,996 0 79,583 9,119
Local 3,361 1,392 7,354 0 174,582 9,282 -46.7***
MNC 9,248 3,569 17,825 1 401,879 9,119
Local 192 55 797 -7,538 39,981 9,282 -19.8***
MNC 471 123 1,816 -15,657 49,357 9,119
Local 0.33 0.27 0.29 0.00 1.00 9,282 10.8***
MNC 0.30 0.18 0.31 0.00 1.00 9,119
Local 0.11 0.03 0.16 0.00 0.99 9,282 18.0***
MNC 0.08 0.00 0.16 0.00 0.99 9,119
Local 50 27 77 1 1,557 9,282 -13.3***
MNC 86 33 157 1 4,985 9,119
Local 83 39 126 0 1,024 9,282 -39.2***
MNC 152 86 176 0 1,030 9,119
Local 9.0 9.3 4.2 0.1 17.0 9,282 5.7***
MNC 8.6 8.8 4.1 0.1 17.0 9,119
Local 0.33 0.30 0.24 0.00 1.00 9,282 22.4***
MNC 0.26 0.17 0.25 0.00 1.00 9,119
Appendix 3. Number of observations per industry.
342

Agriculture,
forestry,
and fishing
Mining Construction Manufacturing
Wholesale
trade
Retail trade Services Total
Local 29 43 49 48 177 57 48 83
418 29 1,657 2,520 2,450 1,333 875 9,282
5% 0% 18% 27% 26% 14% 9% 100%
MNC 99 41 115 78 258 87 89 152
95 110 429 2,965 3,589 612 1,319 9,119
1% 1% 5% 33% 39% 7% 14% 100%
Total 42 41 63 64 225 67 73 117
513 139 2,086 5,485 6,039 1,945 2,194 18,401
3% 1% 11% 30% 33% 11% 12% 100%
Appendix 4. Pairwise correlations between variables.
Labour
Productivity
Labour Productivty 1.00
Adjusted
Leverage
Adjusted Leverage -0.03*** 1.00
Long-term Leverage -0.09*** 0.68*** 1.00
Tangibility -0.32*** 0.31*** 0.36*** 1.00
Long-term
Leverage Tangibility Size Age HHI
Size 0.15*** 0.06*** 0.06*** 0.10*** 1.00
Age -0.06*** -0.10*** -0.05*** 0.09*** 0.09*** 1.00
HHI -0.13*** 0.03*** 0.04*** 0.16*** 0.15*** 0.03*** 1.00
*** denotes significance at 1% level.
343

MULTISTAGE INVESTMENT OPTIONS, TIME-TO-BUILD AND FINANCING CONSTRAINTS
Elettra Agliardi, University of Bologna and Rimini Centre for Economic Analysis, e-mail: elettra.agliardi@unibo.it
Nicos Koussis, Frederick University Cyprus and University of Bologna, e-mail: bus.kn@fit.ac.cy
Abstract. A dynamic investment options model with “time-to-build”, debt and equity constraints is studied to evaluate the effects on
firm value and leverage choices. It is shown that a firm is more likely to face financing constraints with short term debt. With “timeto-build”
and a tax scheme with full loss offset provision the firm increases initial leverage in order to reduce the impact of delayed
cash flow receipts resulting from “time-to-build”. Under no deductibility for losses, the firm would reduce initial debt significantly.
The joint impact of “time-to-build” and financing constraints causes a significant decrease in firm values.
Keywords: investment options, optimal capital structure; time-to-build; financing constraints; binomial lattice models; real options.
JEL classification: G3;G32;G33; G1
1 Introduction
Recent theoretical developments in corporate finance building on Leland (1994) have provided a unified framework
for the analysis of investment and financing decisions of the firm (see, e.g. Mauer and Sarkar, 2005; Sundaresan and
Wang, 2007; Hackbarth and Mauer, 2010). These papers do not deal with the effects of equity and debt financing
constraints, which are considered in empirical literature extensively (see, f.e., , Rauh, 2006, Hubbard et al., 1995,
Whited and Wu 2006, Titman et al. 2004). In this paper we develop a comprehensive real option model using a
numerical binomial tree approach that extends Broadie and Kaya (2007), allowing for optimal capital structure,
multiple debt issues with different priority rules, multiple investment option stages, different tax schemes and equity
and debt financing constraints. Moreover, we analyze how time-to-build affects leverage choices and how the joint
presence of time-to-build and financing constraints would affect firm values and leverage choices over time. To our
knowledge, this problem and this framework have not been tackled within the literature so far.
There are very few theoretical studies considering the impact of time-to-build on the valuation of an investment
project. Theoretical work using a real option approach with time-to-build has focused on the case without optimal
capital structure (see Majd and Pindyck, 1987, Bar-Ilan and Strange, 1996 and 1998). In Majd and Pindyck (1987)
there is a maximum rate at which construction proceeds, so that it takes time before the project is completed and
begins to generate revenue. Investment proceeds continuously until the project is completed, although construction
can be stopped and later restarted without a cost. In contrast, in our paper investment decisions are made discretely,
rather than continuously, the investment option comes to the end of its useful life, instead of being infinitely lived,
and optimal capital structure and financial constraints are introduced. Koussis et al. (2007) analyze a similar case
called “time-to-learn” where the firms learn new information about the project with a time lag. An interesting result
which emerges from the above mentioned papers is that the usual relationship between volatility and the opportunity
cost and the timing of investment may be reversed in the presence of time-to-build, i.e., an increase in volatility or a
decrease in the opportunity cost may accelerate investment. This result holds because an increase in time-to-build
causes a reduction in the value (moneyness) of the option, so that an increase in volatility or a decrease in the
opportunity cost may sufficiently increase the value of the project triggering earlier investment. Given the
prevalence of both time-to-build and financing constraints in different industries, we also investigate their joint
impact on firm values.
Our results show that with short term debt the firm optimally sets coupon levels to be high, often exceeding
revenue levels at the time of debt issue. In the presence of non-negative net worth (equity) constraints an increase in
volatility hurts firm value by decreasing both the value of unlevered assets and the tax benefits of debt. This
decrease in the tax benefits of debt is caused by the decrease in the use of debt in early stages, although subsequent
leverage may increase. With debt financing constraints higher volatility increases firm value for out-of-the money
options by increasing the option value component of firm value (since the firm cannot exploit high tax benefits of
debt). For in-the-money options an increase in volatility has little impact on firm values. Our analysis further
explores how different tax regimes may have an impact on firm values and leverage choice. Most trade-off models
of the capital structure assume a full loss offset taxation scheme. Alternative tax schemes include the asymmetric
344

tax-scheme, where the tax benefits of debt are lower when the firm incurs losses, or a no loss offset scheme (e.g.,
Tserlukevich, 2008, Agliardi and Agliardi, 2009). In our paper we compare the effects of a full loss offset tax regime
with the no loss offset case. We show that in the latter firms behave considerably more conservatively in the use of
debt for short term maturity, with a substantial firm value reduction compared to the full offset scheme. Sensitivity
analysis results show in general similar directional effects between the two tax schemes (both in the presence and
absence of financing constraints). The assumptions concerning the tax regime becomes particularly important in the
case of time-to-build. If full loss offset provisions are assumed, we find that firms will issue debt prior to
completion of the project to benefit from the resulting tax benefits and mitigate the time-to-build constraints. With
no loss offset the use of initial debt is substantially reduced while subsequent debt value is higher, at least for short
term time-to-build constraints. Our results with time-to-build and full loss offset reveal that debt levels increase to
alleviate the impact of delayed cash flow receipt due to time-to-build. With low volatility the impact of time-to-build
on firm value is reduced since the firm can borrow more heavily in order to alleviate the impact of time-to-build. A
similar effect holds for low opportunity cost-competitive erosion. Equity financing constraints have an important
effect on firm values in the presence of time-to-build (decreasing it by 31% for 5 year time-to-build horizons and by
43% to 10 year time-to-build horizon). For low volatility and high opportunity cost the impact of equity financing
constraints with time-to-build is more significant. Debt financing constraints impact on firm values is highly
significant reaching 42% for plausible parameter values. At low opportunity cost debt financing constraints
significantly reduce bankruptcy risk (because of the high debt capacity in the unconstrained case) and the firm
balances the leverage levels between the initial and the subsequent debt issue. With a no loss offset tax scheme
initial debt levels are reduced significantly for short to medium term horizons with initial leverage increasing
slightly only for long time-to-build horizons. Under this tax scheme we similarly observe that at lower volatility or
lower opportunity cost firm value is increased for all time-to-build horizons. The percentage decrease in firm value
due to time-to-build remains less severe with lower volatility or lower opportunity cost like in the full offset case.
However, due to the lower tax benefits under the no loss offset scheme the differences between high and low
volatility and high and low opportunity cost are lower under all time-to-build horizons.
Overall, we demonstrate the flexibility of the lattice method to incorporate several features that are embedded in
many investment decisions characterized by lengthy construction periods and financing constraints.
2 The Model
Consider a firm whose revenues follow a geometric Brownian motion of the form dP � P(
adt � �dZ)
where α ,
σ > 0 are constant parameters, r is the risk-free interest rate and dZ is the increment of a standard Wiener process.
The firm pays an operational cost C so that total earnings before interest and taxes is P-C. It can decide whether to
invest at time 1 T by paying an irreversible fixed cost I1 and choosing a mix of debt 1( ) P D and equity ) ( 1 P E =
I1 � D1(
P)
to finance the investment cost. After the first investment stage, subsequent investment stages may
follow with maturities T i , i ,... S 3 , 2 � relative to the prior stage (so that actual time for option i is the
accumulated time i T T T 1 � 2 � ... ). At each investment stage the firm may decide to issue new debt and rebalance
its capital structure. Tax-deductible coupon payments Ri are due at each period i and the principal debt (face
value) Fi is paid at maturity. Debt maturity for each debt issue is specified by TDi withTD1 � TF
,
TD2 � TF
� ( T1
� T2
) , TD3 � TF
� ( T1
� T2
� T3
) etc. where TF denotes the firm life. In order to
accommodate the choice of different coupon levels at each investment stage we employ forward-backward
algorithm. The algorithm first creates the pre-investment stage tree with N1 steps. At each revenue level at the end
nodes of the first investment stage, several lattices are created that capture the next operational phase and default
decisions conditional on the choice of the coupon level. Coupon levels depend on revenues P at each state which is
discretized through the choice of C n points and the use of a maximum of cmax points. This implies a coupon grid
{ 0,
1 2 cmax
� P,
� P,....
P}
. Investment stages are approximated by lattices with sizes that are defined relative
n n
nc c
c
345

to the tree used for the pre-investment stage. Denoting the pre-investment stage with N1 the size of the i
subsequent interval ( i � 2,
3,...
) becomes
� Ti
�
N i �
�
� � N1
T �
� . The last period (after
� 1 �
T S ) is approximated by
T ( T T ... T ) T
F � �
� � � � � � �
�
� �
N � �
�
F 1 2
T1
S �1
S
� � N1
� � N1
.
� � T1
�
A standard formulation of the lattice parameters for the up and down jumps and the up and down probabilities (see
Cox, Ross and Rubinstein, 1979) requires that
�dt
u � e ,
dt 1
d � e �
��
, pu
( r�
� ) dt
e � d
�
, pd � 1 � pu
,
where
u u � d
TF
dt � and δ is an opportunity cost parameter. We keep track of the following information at each node
N
F
of the binomial tree: unlevered assets ( U
V ), tax benefits of debt (TB), bankruptcy costs ( BC ), equity (E), debt
issues ( 1, 2 ,... D
L
D ) and levered firm value ( V ). We denote by b the proportion of the value of the firm that is
lost in case of bankruptcy and assume that debt holders will be reimbursed following an absolute priority rule. It
implies that the debt holders’ claim in case of default is:
u ~
D1t � min�( 1�
b)
Vt
, R1
�t
� D1t
�
(1)
j 1
u
~
D jt � min[( 1�
b)
Vt
� � Dit
, R j�t
� D jt ]
i 1
�
, j = 2,….S
�
where it D~ denotes the expected continuation value for debt issue i in case the firm does not default at t , that is,
~
�rdt
Dit
� ( pu
Dt�dt
, h � ( 1�
pu
) Dt�dt
, l ) e . Cash inflows (revenues) and outflows (costs and interest payments)
as well as decisions occur every time step � t . �t can be controlled by a variable N dec that specifies the number
of decision points within each unit period 1 . At the end of the operational phase
are calculated as follows:
TF equity and the other variables
�
S
S
debt
debt �
ET � max�(
P � C � � Ri
I i )( 1�
� ) �t
� � Fi
I i , 0�
(2)
F
�
i�1
i�1
�
where � >0 is the corporate tax and
debt
Ii is an indicator that takes the value of 1 if debt issue i has not expired
and zero otherwise. If E TF
� 0 , then V P C t
u
TF � ( � )( 1�
� ) � , TBTF
� S
debt �
� � ��
Ri
Ii
��t
� i�1
�
, BC TF
� 0 ,
DiTF � Ri
�t
� Fi
,
L
VTF
� ETF
S
debt
� � Di
Ii
,
i�1
otherwise if ETF
2 � 0 (i.e., bankruptcy occurs), then
V P C t
u
TF � ( � )( 1�
� ) � , TB TF
� 0 , BCTF u L
� bVT
, V
F TF
� ETF
� DT
. Debt values at maturity in the event
F
of default depend on the priority rule. Under the absolute priority rule we get:
D
U �( 1�
b)
V , R �t
� �
D1
TF � min TF
1 F1
(3)
jTF
j 1
u
� min[( 1�
b)
V � � DiT
, R j�t
� Fj
]
F
i 1
�
, j = 2,….S
�
1 Thus, Δt = 1/Ndec. Each Δt interval is approximated by a sub-tree NΔt. To maintain accuracy discounting occurs for the interval dt = Ti/Ni. In
principle, the decisions can be made as dense as possible approximating the continuous decision limit when Ndec tends to infinity.
2 If the value of unleveled assets turns negative then the value of all variables are set to zero.
346

Prior to the maturity of the operational phase (and after all investments have taken place) the values of each of these
variables are calculated as follows:
�
S
debt ~ �
Et � max�(
P � C � � Ri
I i )( 1�
� ) �t
� Et
, 0�
�
i�1
�
u
~
~
S
u
� debt �
If E t � 0 , then Vt
� ( P � C)(
1�
� ) �t
�Vt
, BCt
� 0 � BCt
, TBt
� � ��
Ri
I i ��t
+ TB t
� i�1
�
~
,
~
S
L
debt
u
~ u
D it � Ri�t
� Di,
t , Vt
� Et
� � Dit
Ii
, whereas, if E t � 0 , then Vt
� ( P � C)(
1�
� ) �t
� Vt
,
i�1
S
u
L
debt
BCt � bVt
and TB t � 0 , Vt
� Et
� � Dit
Ii
, where ~
xt
denotes the expected discounted value of
i�1
rdt
variable x and equals
~
�
xt
� ( pu
xt
�dt
, h � ( 1�
pu
) xt
�dt
, l ) e .
For points within the lattice not involving a decision to default or not, which are used for increased accuracy, the
values of each variable are the discounted expected values of the variables of the following period. At the maturity
of each investment option stage i occurring at time t, where t takes values according to the specified investment
maturities, the levered firm value includes the equity value plus the amount of debt received at i plus the expected
values of debt raised in the future minus the total cost which includes the investment paid at i stage and the
expected cost to be paid in the future:
S ~
S ~
S
L
u
~
V t � max[ Et
� Dit
� � Dkt
� ( Ii
� � I k ), 0]
� max[ Vt
� TBt
� BCt
� ( Ii
� � I k ), 0]
(4)
k �i�1
k�i
�1
k �i�1
Bankruptcy in periods between investment stages (and prior to the final investment) is triggered when the earnings
net of cost and coupon payments plus the expected levered firm value (which includes expected equity value,
expected cash received by debt issues and expected costs to be paid) are negative. Thus, the bankruptcy condition
for any time t prior to the last investment stage is:
�
I
L
debt ~ L �
Vt � max�(
P � C � � Ri
Ii
)( 1�
� ) �t
� Vt
, 0�
(5)
�
i�1
�
and the values at that stage are calculated as above, depending on whether there is default or not. The values
obtained at the first investment stage are discounted at t = 0. The value of the firm at time zero involves the sum of
the present value of equity, all expected debt issues minus the expected present value of the investment costs. This is
equivalent to the expected present value of the unlevered assets plus the expected present value of the tax benefits
minus the expected present value of bankruptcy and the investment costs.
In what follows we introduce the possibility that a firm may face equity or debt financing constraints or both. In
order to incorporate equity financing constraints we investigate the case where equity holders face constraints
where net worth remains positive at all times, i.e., E t � 0 , that is, we do not allow for negative equity values
between stages, implying that no equity infusion of cash is allowed. Alternative constraints where partial infusion of
cash prior to investment equity could be easily analyzed in this framework. Observe that this condition does not
imply that there is no equity financing since equity holders can finance part or all of the investment cost at the time
investments take place. However, if at any point in time after investment equity value drops to zero default is
triggered without considering additional infusion of cash. This type of constraints may reflect difficulties in raising
new external finance in the case where the firm is not performing well or the inability of current equity holders to
infuse new cash because of personal financing constraints. Debt constraints can be easily incorporated in this
framework, for example, by allowing coupon rates to be only a fraction of the revenue level at which the leverage
decision is made. In the following numerical simulations both equity and debt financing constraints will be
examined in a two stage model.
3 Simulation results: Unconstrained and Constrained Debt and Equity
Table 1 presents sensitivity results with respect to volatility for unconstrained debt where both debt maturity issues are
equal (TD1 = TD2 = 5). The firm optimally selects high coupon levels which exceed current revenue levels. As we shall
347

subsequently see, it implies that debt financing constraints become binding. Observe that equity value (net worth) is
negative in this case. This is because of the high coupon and debt levels used. One can also find that firm values are
higher than what can be obtained in a longer debt maturity case, for all levels of volatility and for different levels of
moneyness (see also Agliardi and Koussis, 2010, for a single debt issue). Equity values are increasing in volatility for
both low and high revenue level. Debt values are strictly decreasing and thus total leverage (LevT) is decreasing in
volatility. With short term debt equity holders want to accelerate the receipt of tax benefits of debt. Since the initial debt
expires before the exercise of the second stage investment there are no remaining net benefits, the firm is able to borrow
more heavily initially and thus we register an option effect related to unlevered assets.
Table 2 shows sensitivity results with respect to volatility when the firm faces equity financing constraints. Firm
values decrease as a function of volatility because the debt levels are more significantly decreased than the equity
increase. Observe that for both the out-of-the-money and the in-the-money case an increase in volatility reduces firm
values by reducing the value of unlevered assets and the tax benefits of debt. Under positive net worth constraints firms
issue less debt initially in order to avoid distress and behave conservatively. This is consistent with empirical studies
(Graham, 2000 and DeAngelo et al., 2010) providing a possible explanation of growth firms borrowing conservatively.
In Table 3 the results under debt financing constraints are examined. Under our parameters, debt constraints for both
the out-of-the-money and the in-the-money case are binding, resulting in the maximum allowable coupon being used. In
the out-of-the-money case an increase in volatility causes an increase in firm value which is caused by an increase in the
option value of unlevered assets (value of unlevered assets net of expected costs). In this case a higher volatility helps
alleviate the impact of financing constraints which reduce the tax benefits of debt. In order to evaluate the amount of
leverage we calculate three measures. All measures use the time 0 values of equity and debt (despite the fact that some
debt may be issued at a future date). The first measure (Lev1) includes only the first debt issue over the total value of
equity plus all debt. The second measure (Lev2) includes the proportion of the second debt issue over the value of equity
plus all debt issues. The third measure is total leverage over the total value of equity plus debt (LevT). We observe that
both the initial and subsequent debt are decreased but lev1 and lev2 remain rather flat with lev1 being higher than lev2
(like in the unconstrained case). In the in-the-money case an increase in volatility has little impact on firm value when the
firm faces debt financing constraints. This occurs because both the value of unlevered assets and the net benefits of debt
remain flat. In comparison with the unconstrained case the percentage drop in value is more substantial for the out-ofthe-money
case. For in-the-money the absolute value drop is significant while the percentage drop is less important than
out-of-the-money case. In the unconstrained case we observe that lev1 is much higher than lev2. While this remains in the
constrained debt case this gap is now reduced.
Finally, we can compare the impact of two alternative tax regimes on firm value and leverage. From the comparison
between a full loss offset and a no loss offset scheme both in the out-of-the-money case and in-the-money case, we can
see that the difference in firm value under the two regimes is substantial, with firm values even halved in some
occasions. Coupon levels under the full loss offset case are always higher than the corresponding cases of no loss
offset. This is shown also in the large tax benefit differences in the two schemes with the full loss offset exhibiting
much higher values (while bankruptcy costs are of the same order of magnitude). This difference exists despite the
large advantage obtained under the no loss offset case, because of no taxes paid in the loss region.
4 Time-to-build
Let us now investigate the impact of time-to-build on firm values in the presence and in the absence of
financing constraints. First, we consider without financing constraints the financing choices of the firm facing timeto-build
restrictions. It can be shown that the results critically depend on the tax scheme. Our second goal is to
investigate the importance of financing constraints when the firm faces time-to-build restrictions and analyze the
adjustments in the financing policy.
Table 4 analyzes the case without financing constraints with sensitivity with respect to volatility and the
opportunity cost-competitive erosion. With time-to-build the firm receives no cash flows until the project is
completed, i.e., cash flows initiate after the second investment is implemented, and the useful life of the project is
reduced by the time-to-build horizon 3 . In the base case, we observe that firm value, the value of unlevered assets and
3 Had we assumed the operational period remains fixed we will simply capturing an option to delay the exercise of the investment option.
348

the net benefits of debt are reduced with time-to-build. Interestingly, debt 1 and debt 2 increase with time-to-build
and are only decreased in very long time-to-build horizons. Lev1, Lev2 and LevT reflect this pattern of debt values
since they are increasing as time-to-build increases and then remain flat for long time-to-build horizons. Equity
values are decreasing with time-to-build reflecting the increase in the debt due (except for very long time-to-build
horizons where debt due is decreased).
Investigating the case of low volatility reveals some interesting insights. Since we are using short-term debt an
increase in volatility reduces firm value. Based on standard real options model (e.g., Majd and Pindyck, 1987 and
Bar-Ilan and Strange, 1996 and 1998) with time-to-build one could expect that higher volatility may have been
beneficial in order to increase the option value of completing the project successfully. The results, however, show
that with low volatility the firm borrows more heavily to alleviate the impact of time-to-build. This is reflected in the
large differentials between the low and high volatility case with respect to the tax benefits of debt. The ability of the
firm to raise more debt under low volatility reduces the percentage decrease in firm values at higher time-to-build
horizons (compared to the high volatility case). Thus in the case of time-to-build with optimal capital structure a low
volatility may be preferred in contrast to the traditional result without debt financing where a high volatility would
be preferable. A similar effect exists for the case of low δ. A low δ reduces the impact of time-to-build on firm
values since the debt capacity is higher and the increase in the tax benefits is more significant.
Table 5 investigates the impact of positive net worth equity constraints in the presence of time-to-build. The
results show that positive net worth equity constraints can cause severe reductions in firm values in the presence of
time-to-build. For the base case the decrease in firm value reaches 31% for 5 year time-to-build horizon and 43% for
10 years of time-to-build (when comparing the corresponding cases of no constraints and equity constraints). For
lower volatility the decrease in values due to positive net worth equity constraints is more severe. The reason is that
at low σ the unconstrained firm would optimally choose to borrow more at t = 0 but that would increase the
probability of negative equity worth (in those states that revenues would not suffice to cover coupon payments).
Therefore, positive net worth equity constraints indirectly impose constraints on the optimal amount of leverage
causing severe reductions in firm values. Interestingly, the firm in this case will optimally shifts emphasis on
obtaining higher leverage in subsequent debt issues. For lower opportunity cost δ the percentage reduction in firm
values due to positive net worth equity constraints are less important (but still significant). Indeed, with lower δ the
firm can retain positive equity values at the unconstrained levels more easily (despite borrowing more heavily). With
lower δ and for short time-to-build the firm can retain positive equity values even using high initial debt levels.
However, as the time-to-build becomes longer the subsequent leverage level becomes more important.
Table 6 analyzes the case with debt financing constraints. Debt financing constraints cause a decrease in value
of about 15-16% even for the case without time-to-build. Time-to-build makes the impact of debt financing
constraints more significant reaching firm reductions of 33% for time-to-build of 5 years for the base case, 42%
reduction for the low volatility case and 36% for the low opportunity cost case. The results also show that the initial
leverage appears significant than the subsequent leverage level. With debt constraints we observe that the firm value
differences between low and high volatility are not significant. Low coupon levels due to the constraints reduce the
risk of bankruptcy (as indicated by the low bankruptcy costs) even for the high volatility (base case) thus values of
unlevered assets, and tax benefits do not change with different volatility levels. Leverage choices remain very
similar to the higher volatility case with lev1 being more important than lev2. When δ is small bankruptcy is
practically non-existent because of the constrained low levels of coupons used and the high debt capacity if the firm
was not constrained. Debt appears to be equally balanced between the first and the second debt issue.
With no loss offset the results are materially different regarding the use of initial leverage. One can find (these
results are not shown for brevity) that the initial debt levels are drastically reduced and are close to zero for short to
medium term horizons. Initial leverage increases slightly only for long time-to-build horizons. Some similarities
with the full loss offset however still remain. With lower volatility or lower opportunity cost firm values are higher
than the base case for all time-to-build horizons. Furthermore, the percentage decrease in firm value due to time-tobuild
is less severe with lower volatility or lower opportunity cost. However, the reduced impact of time-to-build
under low volatility or lower opportunity cost is diminished because now the tax benefits of low volatility or low
opportunity cost are less important.
349

5 References
Agliardi, Elettra and Rossella Agliardi, 2009, Progressive taxation and corporate liquidation policy: analysis and
policy implications, Journal of Policy Modeling 31, 144-154
Agliardi, Elettra and Nicos Koussis, 2011, Optimal capital structure and investment options in finite horizon.
Finance Research Letters, 8, 28-36
Bar-Ilan, Avner, and William C. Strange. 1996, Investment Lags, American Economic Review 86, 610-622.
Bar-Ilan, Avner, and William C. Strange. 1998, A Model of Sequential Investment, Journal of Economic Dynamics
and Control 22, 437-463.
Broadie, M and Kaya, O. 2007, A Binomial Lattice Method for Pricing Corporate Debt and Modeling Chapter 11
Proceedings, Journal of Financial and Quantitative Analysis, 42, 279-312.
DeAngelo, Harry, Linda DeAngelo, Toni M. Whited. 2010, Capital Structure Dynamics and Transitory Debt,
Journal of Financial Economics, doi: 10.1016/j.jfineco.2010.09.005
Cox, J.C, Ross S.A. and Rubinstein, M. 1979, Option Pricing: a Simplified Approach, Journal of Financial
Economics, 7, 229-263.
Graham, J. R, 2000, How big are the tax benefits of debt? Journal of Finance, 55, 1901-1941
Hackbarth, D and D.C: Mauer, 2010, Optimal priority structure, capital structure and investment, mimeo
Hubbard, Glenn R., Anil K. Kashyap, and Toni M. Whited. 1995, Internal Finance and Firm Value, Journal of
Money, Credit, and Banking 27, 683-701.
Koussis, N., Martzoukos S. and Trigeorgis, L. 2007, Real R&D Options With Time-to-Learn and Learning by
Doing, Annals of Operations Research 151, 23-59.
Leland, H. 1994, Corporate Debt Value, Bond Covenants, and Optimal Capital Structure, Journal of Finance 49,
1213-1252.
Majd, S., and R. Pindyck. 1987, Time to Build, Option Value, and Investment Decisions.” Journal of Financial
Economics 18, 7-27.
Mauer, D.C., and Sarkar S. 2005, Real Options, Agency Conflicts and Optimal Capital Structure, Journal of
Banking and Finance 29, 1405-1428.
Rauh, Joshua D. 2006, Investment and Financing Constraints: Evidence from the Funding of Corporate Pension
Plans, Journal of Finance LXI, 33-71.
Sundaresan, S. and Wang, N. 2007, Dynamic Investment, Capital Structure, and Debt Overhang, Working Paper,
Columbia University.
Titman, Sheridan, Stathis Tompaidis, and Sergey Tsypliakov. 2004, Market Imperfections, Investment Flexibility,
and Default Spreads, Journal of Finance LIX, 165-205.
Tserlukevich, Y, 2008, Can real options explain financing behaviour? Journal of Financial Economics,89, 232-252
Uhrig-Homburg, Marliese. 2005, Cash-Flow Shortage as an Endogenous Bankruptcy Reason, Journal of Banking and
Finance 29, 1509-1534.
Whited, Toni M., and Guojun Wu. 2006, Financial Constraints Risk, Review of Financial Studies 19, 531-559.
Table 1. Short term debt with no constraints: Sensitivity to volatility
Panel A: Full tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 42.776 79.275 48.971 2.497 -16.666 82.289 60.126 50 32.973 16.5 0.65 0.48 1.13
σ = 0.20 34.600 74.358 39.691 5.624 -10.602 71.392 47.635 50 23.826 16 0.66 0.44 1.10
σ = 0.30 31.324 71.545 34.110 7.928 -7.660 69.885 35.501 50 16.402 18 0.72 0.36 1.08
σ = 0.40 31.170 70.621 31.914 7.883 -4.414 66.008 33.057 50 13.481 18 0.70 0.35 1.05
In-the-money case (P=30)
350

Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 327.563 239.867 174.825 0.119 -85.046 311.372 188.246 50 37.009 60 0.75 0.45 1.21
σ = 0.20 305.948 239.331 158.112 8.114 -70.534 290.180 169.683 50 33.382 58.5 0.75 0.44 1.18
σ = 0.30 281.229 235.359 137.561 15.097 -50.305 270.946 137.182 50 26.594 60 0.76 0.38 1.14
σ = 0.40 261.456 232.054 120.325 18.968 -29.342 239.926 122.827 50 21.954 57 0.72 0.37 1.09
Panel B: No loss tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 20.880 90.616 9.121 1.234 6.441 51.235 46.893 50 33.690 10 0.49 0.45 0.94
σ = 0.20 18.648 80.113 7.484 3.662 10.589 49.026 32.577 50 23.544 10.5 0.53 0.35 0.89
σ = 0.30 18.081 74.959 11.589 3.266 15.942 41.811 28.664 50 18.337 9.5 0.48 0.33 0.82
σ = 0.40 18.789 72.575 7.024 4.377 16.944 43.935 21.503 50 13.594 11 0.53 0.26 0.79
In-the-money case (P=30)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 239.558 239.867 3.207 0.068 -33.153 210.205 149.519 50 37.013 40.5 0.64 0.46 1.10
σ = 0.20 232.003 239.331 18.424 2.673 10.398 183.872 122.748 50 35.015 36 0.58 0.39 0.97
σ = 0.30 220.276 236.941 24.103 6.657 44.030 147.851 109.133 50 30.737 30 0.49 0.36 0.85
σ = 0.40 210.889 232.054 37.569 6.652 73.111 133.044 80.370 50 25.637 28.5 0.46 0.28 0.74
Notes: The model with no equity or debt constraints is used. Base case parameters are: P =10 (out-of-money) or P = 30 (in-the-money), C = 0,
risk-free rate r = 0.06, volatility σ = 0.2, competitive erosion δ = 0.06, investment cost I1 = 50, I2 = 50, b = 0.5, tax rate τ = 0.35 and T1 = 0, T2 = 5
(time of second option relative to the first), TF = 20 and debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal coupon is chosen
based on nc =20 discretization points for each price level with maximum coupon level points being equal to the price levels (cmax = 40). In all
tables Ndec = 1 (yearly decisions) with NΔt = 24 steps per year.
Table 2. Short term debt with equity financing constraints: Sensitivity to volatility
Panel A: Full tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 31.037 79.813 37.319 1.766 6.973 41.641 66.751 50 34.328 8 0.36 0.58 0.94
σ = 0.20 26.521 75.918 31.386 4.739 8.151 43.397 51.017 50 26.044 9 0.42 0.50 0.92
σ = 0.30 25.687 73.651 28.337 3.745 13.534 41.104 43.604 50 22.556 9 0.42 0.44 0.86
σ = 0.40 24.783 70.621 25.231 5.394 12.975 41.002 36.481 50 15.675 10 0.45 0.40 0.86
In-the-money case (P=30)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 262.944 239.867 111.901 3.165 25.721 117.089 205.792 50 35.659 22.5 0.34 0.59 0.93
σ = 0.20 246.437 239.845 97.930 7.525 42.926 108.555 178.769 50 33.813 21 0.33 0.54 0.87
σ = 0.30 238.636 238.742 90.632 10.115 50.195 119.431 149.634 50 30.624 24 0.37 0.47 0.84
σ = 0.40 227.516 236.392 81.142 12.759 60.181 107.942 136.652 50 27.259 22.5 0.35 0.45 0.80
Panel B: No loss tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 20.344 90.616 5.580 3.031 6.533 50.890 44.813 50 31.893 10 0.50 0.44 0.94
σ = 0.20 18.418 80.113 7.254 3.662 12.527 46.858 32.577 50 23.544 10 0.51 0.35 0.86
σ = 0.30 17.694 74.959 11.173 2.916 17.587 39.806 29.040 50 18.740 9 0.46 0.34 0.80
σ = 0.40 18.594 72.537 6.868 4.377 18.579 42.104 21.505 50 13.594 10.5 0.51 0.26 0.77
In-the-money case (P=30)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 232.521 278.015 35.441 2.410 31.301 140.250 146.878 50 35.909 27 0.44 0.46 0.90
σ = 0.20 225.487 266.640 39.048 5.579 50.751 138.169 120.298 50 33.732 27 0.45 0.39 0.84
σ = 0.30 214.369 260.385 37.767 7.416 72.379 113.045 110.525 50 31.581 22.5 0.38 0.37 0.76
σ = 0.40 208.050 251.882 34.250 12.701 75.449 125.344 81.928 50 24.670 27 0.44 0.29 0.73
Notes: The model with equity constraints (debt unconstrained) is used. Base case parameters are: P =10 (out-of-money) or P = 30 (in-the-money),
C = 0, risk-free rate r = 0.06, volatility σ = 0.2, competitive erosion δ = 0.06, investment cost I1 = 50, I2 = 50, b = 0.5, tax rate τ = 0.35 and T1 = 0,
T2 = 5 (time of second option relative to the first), TF = 20 and debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal coupon is
chosen based on nc =20 discretization points for each price level with maximum coupon level points being equal to the price levels (cmax = 40). In
all tables Ndec = 1 (yearly decisions) with NΔt = 24 steps per year.
Table 3. Short term debt with debt financing constraints: Sensitivity to volatility
Panel A: Full tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 16.445 78.310 23.245 0.055 35.030 38.588 27.882 50 35.055 7.5 0.38 0.27 0.65
σ = 0.20 16.565 72.407 21.097 0.540 32.149 36.517 24.299 50 26.400 7.5 0.39 0.26 0.65
σ = 0.30 18.571 71.545 20.081 1.009 32.234 34.908 23.476 50 22.046 7.5 0.39 0.26 0.64
σ = 0.40 20.681 70.621 19.077 1.247 32.700 33.175 22.577 50 17.770 7.5 0.38 0.26 0.63
In-the-money case (P=30)
351

Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 223.994 239.867 71.169 0.000 107.696 116.807 86.532 50 37.041 22.5 0.38 0.28 0.65
σ = 0.20 223.946 239.779 71.110 0.024 107.669 116.736 86.461 50 36.919 22.5 0.38 0.28 0.65
σ = 0.30 222.648 238.024 69.682 0.628 107.359 114.875 84.844 50 34.430 22.5 0.37 0.28 0.65
σ = 0.40 220.579 236.392 66.866 2.025 108.161 111.162 81.910 50 30.654 22.5 0.37 0.27 0.64
Panel B: No loss tax offset
Out-of-the-money case (P = 10)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 15.154 78.310 18.882 0.139 33.715 38.563 27.816 50 34.940 7.5 0.39 0.28 0.66
σ = 0.20 15.077 72.407 15.082 0.829 30.786 36.301 24.107 50 26.117 7.5 0.40 0.26 0.66
σ = 0.30 15.966 69.030 13.169 1.381 29.578 33.984 21.801 50 19.397 7.5 0.40 0.26 0.65
σ = 0.40 17.634 70.621 11.414 2.648 30.723 32.392 21.267 50 16.748 7.5 0.38 0.25 0.64
In-the-money case (P=30)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
σ = 0.10 219.912 239.867 57.481 0.000 103.614 116.807 86.532 50 37.041 22.5 0.38 0.28 0.66
σ = 0.20 218.014 239.701 48.337 0.146 101.824 116.625 86.297 50 36.733 22.5 0.38 0.28 0.67
σ = 0.30 213.319 236.941 42.946 1.555 99.504 113.854 83.168 50 33.208 22.5 0.38 0.28 0.66
σ = 0.40 207.898 234.578 36.599 4.438 99.439 108.880 77.966 50 28.386 22.5 0.38 0.27 0.65
Notes: The model with debt constraints (without equity constraints) is used. Base case parameters are: P =10 (out-of-money) or P = 30 (in-themoney)
, C = 0, risk-free rate r = 0.06, volatility σ = 0.2, competitive erosion δ = 0.06, investment cost I1 = 50, I2 = 50, b = 0.5, tax rate τ = 0.35
and T1 = 0, T2 = 5 (time of second option relative to the first), TF = 20 and debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal
coupon is chosen based on nc =20 discretization points for each price level with maximum coupon level points being equal to the price levels
(cmax = 15) implying that coupons cannot exceed 75% of revenue (P) level at the time of the financing decision. In all tables Ndec = 1 (yearly
decisions) with NΔt = 24 steps per year.
Table 4. Time-to-build without financing constraints: Sensitivity to model parameters
Panel A: Base case
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 72.624 79.956 33.902 1.234 14.527 49.665 48.432 20 20.000 9.75 0.44 0.43 0.87
T 2=1 68.256 73.282 36.361 2.632 0.491 57.813 48.708 20 18.756 11.5 0.54 0.46 1.00
T 2=2 66.139 65.781 39.224 2.129 -11.322 69.032 45.166 20 16.737 14 0.67 0.44 1.11
T 2=3 63.855 60.223 40.815 1.798 -19.171 69.411 49.001 20 15.385 14 0.70 0.49 1.19
T 2=4 61.404 55.652 41.946 1.781 -25.811 67.493 54.136 20 14.413 13.5 0.70 0.56 1.27
T 2=5 59.215 50.846 43.985 3.122 -37.085 73.999 54.794 20 12.494 15.5 0.81 0.60 1.40
T 2=6 51.063 46.199 39.547 2.917 -33.080 66.920 48.990 20 11.766 14 0.81 0.59 1.40
T 2=7 43.361 41.622 35.513 2.706 -29.743 59.753 44.418 20 11.067 12.5 0.80 0.60 1.40
T 2=8 34.935 37.323 30.421 2.413 -24.000 50.301 39.030 20 10.397 10.5 0.77 0.60 1.37
T 2=9 28.838 33.188 27.453 2.176 -22.149 45.403 35.211 20 9.627 9.5 0.78 0.60 1.38
T 2=10 21.921 29.367 23.467 1.911 -18.037 38.298 30.661 20 9.001 8 0.75 0.60 1.35
Panel B: Lower volatility (σ = 0.1)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 79.406 79.956 40.118 0.668 4.115 57.980 57.312 20 20.000 11.25 0.49 0.48 0.97
T 2=1 77.070 73.440 43.070 0.611 -7.770 79.899 43.770 20 18.830 15.5 0.69 0.38 1.07
T 2=2 75.262 67.154 46.325 0.602 -20.083 84.985 47.976 20 17.616 16.5 0.75 0.43 1.18
T 2=3 72.927 61.477 48.520 0.876 -30.383 89.384 50.119 20 16.194 17.5 0.82 0.46 1.28
T 2=4 70.811 56.136 50.614 0.580 -39.022 87.111 58.080 20 15.358 17 0.82 0.55 1.37
T 2=5 68.645 51.027 52.587 0.504 -47.642 87.066 63.686 20 14.464 17 0.84 0.62 1.46
T 2=6 61.236 46.211 49.528 0.878 -47.526 79.386 63.002 20 13.625 15.5 0.84 0.66 1.50
T 2=7 53.048 41.677 44.997 0.796 -43.482 71.698 57.660 20 12.829 14 0.83 0.67 1.51
T 2=8 44.689 37.406 40.077 0.712 -38.448 64.018 51.200 20 12.082 12.5 0.83 0.67 1.50
T 2=9 35.833 33.384 34.460 0.633 -31.879 53.793 45.296 20 11.378 10.5 0.80 0.67 1.47
T 2=10 28.235 29.596 29.913 0.558 -27.072 46.118 39.905 20 10.715 9 0.78 0.68 1.46
Panel C: Lower opportunity cost (δ = 0.02)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 120.073 112.578 50.614 3.120 12.340 75.426 72.306 20 20.000 15 0.47 0.45 0.92
T 2=1 118.984 105.944 54.906 3.075 -2.173 93.384 66.564 20 18.791 18.5 0.59 0.42 1.01
T 2=2 118.172 98.573 59.218 2.453 -16.310 100.722 70.926 20 17.165 20 0.65 0.46 1.10
T 2=3 118.326 93.142 63.495 1.856 -28.489 100.222 83.049 20 16.455 19.5 0.65 0.54 1.18
T 2=4 116.906 87.102 66.529 1.291 -39.035 103.007 88.368 20 15.434 20 0.68 0.58 1.26
T 2=5 113.603 81.313 67.802 0.939 -46.483 103.140 91.518 20 14.573 20 0.70 0.62 1.31
T 2=6 106.761 75.458 66.179 1.429 -50.303 101.990 88.521 20 13.447 20 0.73 0.63 1.36
T 2=7 98.139 69.691 62.422 1.314 -48.864 99.401 80.262 20 12.661 19.5 0.76 0.61 1.37
T 2=8 89.146 64.039 59.925 3.666 -54.582 98.335 76.544 20 11.151 20 0.82 0.64 1.45
T 2=9 78.908 58.491 54.242 3.328 -48.901 90.935 67.369 20 10.496 18.5 0.83 0.62 1.45
T 2=10 68.655 53.058 48.475 2.999 -42.966 81.130 60.370 20 9.879 16.5 0.82 0.61 1.44
352

Notes: The model with no equity or debt constraints is used. Base case parameters are: P =10 , C = 0, risk-free rate r = 0.06, volatility σ = 0.2,
competitive erosion δ = 0.06, investment cost I1 = 20, I2 = 20, b = 0.5, tax rate τ = 0.35 and T1 = 0,. T2 ranges from 0 (no-time-to-build) to timeto-build
of 10 years. When time-to-build exists the firm foregoes all cash flows until full completion of the project. Useful life after the
investment completion is constant at TF = 20. Debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal coupon is chosen based on
nc =20 discretization points for each price level with maximum coupon level points being equal to the price levels (cmax = 40). In all tables Ndec =
1 (yearly decisions) with NΔt = 12 steps per year.
Table 5. Time-to-build with equity financing constraints: Sensitivity to model parameters
Panel A: Base case
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 72.624 79.956 33.902 1.234 14.527 49.665 48.432 20 20.000 9.75 0.44 0.43 0.87
T 2=1 68.256 73.282 36.361 2.632 0.491 57.813 48.708 20 18.756 11.5 0.54 0.46 1.00
T 2=2 62.727 67.189 34.316 1.512 0.434 43.526 56.034 20 17.266 8.5 0.44 0.56 1.00
T 2=3 55.376 61.549 31.359 1.276 0.758 28.433 62.441 20 16.256 5.5 0.31 0.68 0.99
T 2=4 47.418 56.130 28.085 1.618 0.736 18.254 63.607 20 15.179 3.5 0.22 0.77 0.99
T 2=5 40.682 51.015 25.232 1.274 1.608 13.147 60.217 20 14.290 2.5 0.18 0.80 0.98
T 2=6 34.428 46.199 22.831 1.149 1.499 13.100 53.281 20 13.453 2.5 0.19 0.78 0.98
T 2=7 28.116 41.622 20.259 1.063 1.872 10.496 48.450 20 12.701 2 0.17 0.80 0.97
T 2=8 22.641 37.323 18.157 1.007 1.588 10.478 42.408 20 11.832 2 0.19 0.78 0.97
T 2=9 17.416 33.188 16.126 0.656 1.928 7.806 38.923 20 11.241 1.5 0.16 0.80 0.96
T 2=10 12.490 29.367 14.145 0.794 1.508 7.868 33.342 20 10.227 1.5 0.18 0.78 0.96
Panel B: Lower volatility (σ = 0.1)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 79.406 79.956 40.118 0.668 4.115 57.980 57.312 20 20.000 11.25 0.49 0.48 0.97
T 2=1 72.593 73.456 38.586 0.612 0.572 25.957 84.900 20 18.835 5 0.23 0.76 0.99
T 2=2 64.549 67.334 35.513 0.559 0.263 12.979 89.047 20 17.738 2.5 0.13 0.87 1.00
T 2=3 56.690 61.569 32.348 0.522 0.451 7.787 85.157 20 16.705 1.5 0.08 0.91 1.00
T 2=4 49.101 56.140 29.159 0.466 1.055 5.191 78.587 20 15.733 1 0.06 0.93 0.99
T 2=5 42.417 51.027 26.736 0.584 0.206 5.246 71.727 20 14.762 1 0.07 0.93 1.00
T 2=6 35.909 46.211 24.116 0.464 0.496 5.191 64.175 20 13.953 1 0.07 0.92 0.99
T 2=7 29.435 41.677 21.320 0.421 1.240 2.596 58.740 20 13.141 0.5 0.04 0.94 0.98
T 2=8 23.688 37.406 19.033 0.375 1.308 2.596 52.160 20 12.376 0.5 0.05 0.93 0.98
T 2=9 18.340 33.384 16.943 0.332 1.254 2.596 46.145 20 11.655 0.5 0.05 0.92 0.97
T 2=10 13.364 29.596 15.035 0.291 1.091 2.596 40.653 20 10.976 0.5 0.06 0.92 0.98
Panel C: Lower opportunity cost (δ = 0.02)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 120.073 112.578 50.614 3.120 12.340 75.426 72.306 20 20.000 15 0.47 0.45 0.92
T 2=1 118.132 106.069 53.831 2.936 0.226 78.982 77.756 20 18.832 15.5 0.50 0.50 1.00
T 2=2 110.592 99.704 50.270 1.911 2.522 54.229 91.312 20 17.471 10.5 0.37 0.62 0.98
T 2=3 102.546 93.460 46.774 1.313 3.967 41.419 93.533 20 16.374 8 0.30 0.67 0.97
T 2=4 94.176 87.339 43.251 0.998 5.020 31.161 93.411 20 15.416 6 0.24 0.72 0.96
T 2=5 86.334 81.340 40.642 1.147 3.568 26.018 91.248 20 14.501 5 0.22 0.76 0.97
T 2=6 78.257 75.458 37.508 0.890 4.020 18.222 89.835 20 13.819 3.5 0.16 0.80 0.96
T 2=7 70.461 69.691 34.597 0.815 3.809 18.209 81.456 20 13.013 3.5 0.18 0.79 0.96
T 2=8 60.845 64.039 31.113 2.199 1.859 10.589 80.505 20 12.108 2 0.11 0.87 0.98
T 2=9 53.254 58.491 28.073 1.729 2.897 10.411 71.527 20 11.581 2 0.12 0.84 0.97
T 2=10 46.145 53.058 25.533 1.548 2.544 10.405 64.094 20 10.898 2 0.14 0.83 0.97
Notes: The model with equity constraints and no debt constraints is used. Base case parameters are: P =10 , C = 0, risk-free rate r = 0.06,
volatility σ = 0.2, competitive erosion δ = 0.06, investment cost I1 = 20, I2 = 20, b = 0.5, tax rate τ = 0.35 and T1 = 0,. T2 ranges from 0 (no-timeto-build)
to time-to-build of 10 years. When time-to-build exists the firm foregoes all cash flows until full completion of the project. Useful life
after the investment completion is constant at TF = 20. Debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal coupon is chosen
based on nc =20 discretization points for each price level with maximum coupon level points being equal to the price levels (cmax = 40). In all
tables Ndec = 1 (yearly decisions) with NΔt = 12 steps per year.
Table 6. Time-to-build with debt financing constraints: Sensitivity to model parameters
Panel A: Base case
353

Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 66.997 79.956 27.140 0.098 29.357 38.869 38.771 20 20.000 7.5 0.36 0.36 0.73
T 2=1 60.943 73.456 26.387 0.065 24.322 38.895 36.561 20 18.835 7.5 0.39 0.37 0.76
T 2=2 55.246 67.302 25.677 0.017 19.583 38.896 34.483 20 17.717 7.5 0.42 0.37 0.79
T 2=3 49.812 61.500 24.963 0.012 15.116 38.864 32.470 20 16.639 7.5 0.45 0.38 0.83
T 2=4 44.654 56.057 24.238 0.065 10.913 38.805 30.511 20 15.575 7.5 0.48 0.38 0.86
T 2=5 39.809 50.948 23.535 0.148 6.945 38.757 28.633 20 14.526 7.5 0.52 0.39 0.91
T 2=6 35.137 46.144 22.788 0.278 3.266 38.590 26.798 20 13.517 7.5 0.56 0.39 0.95
T 2=7 30.505 41.566 21.837 0.579 -0.148 38.168 24.803 20 12.318 7.5 0.61 0.39 1.00
T 2=8 26.180 37.250 20.955 0.808 -3.283 37.730 22.951 20 11.218 7.5 0.66 0.40 1.06
T 2=9 21.076 33.049 19.182 1.617 -5.808 36.158 20.264 20 9.538 7.5 0.71 0.40 1.11
T 2=10 16.693 29.221 17.767 1.568 -6.911 33.501 18.830 20 8.727 7 0.74 0.41 1.15
Panel B: Lower volatility (σ = 0.1)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 67.210 79.956 27.255 0.000 29.339 38.936 38.936 20 20.000 7.5 0.36 0.36 0.73
T 2=1 61.082 73.456 26.461 0.000 24.313 38.936 36.668 20 18.835 7.5 0.39 0.37 0.76
T 2=2 55.310 67.334 25.714 0.000 19.580 38.936 34.533 20 17.738 7.5 0.42 0.37 0.79
T 2=3 49.874 61.569 25.010 0.000 15.122 38.936 32.522 20 16.705 7.5 0.45 0.38 0.83
T 2=4 44.754 56.140 24.347 0.000 10.924 38.936 30.628 20 15.733 7.5 0.48 0.38 0.86
T 2=5 39.933 51.027 23.723 0.000 6.970 38.936 28.844 20 14.816 7.5 0.52 0.39 0.91
T 2=6 35.393 46.211 23.135 0.000 3.247 38.935 27.164 20 13.953 7.5 0.56 0.39 0.95
T 2=7 31.117 41.677 22.581 0.000 -0.260 38.935 25.582 20 13.141 7.5 0.61 0.40 1.00
T 2=8 27.083 37.406 22.052 0.006 -3.561 38.928 24.085 20 12.370 7.5 0.65 0.41 1.06
T 2=9 23.224 33.383 21.497 0.053 -6.646 38.856 22.617 20 11.603 7.5 0.71 0.41 1.12
T 2=10 18.429 29.595 19.821 0.272 -7.760 35.929 20.975 20 10.715 7 0.73 0.43 1.16
Panel C: Lower opportunity cost (δ = 0.02)
Firm Unlevered TB BC Equity Debt 1 Debt 2 Inv1 Inv2 Coupon 1 Lev 1 Lev 2 Lev T
T 2=0 99,832 112,578 27,254 0,000 61,962 38,935 38,935 20 20,000 7,5 0,28 0,28 0,56
T 2=1 94,227 106,078 26,985 0,000 55,963 38,935 38,164 20 18,835 7,5 0,29 0,29 0,58
T 2=2 88,689 99,707 26,720 0,000 50,083 38,935 37,409 20 17,738 7,5 0,31 0,30 0,60
T 2=3 83,216 93,460 26,460 0,000 44,319 38,934 36,667 20 16,704 7,5 0,32 0,31 0,63
T 2=4 77,813 87,337 26,205 0,001 38,669 38,933 35,940 20 15,729 7,5 0,34 0,32 0,66
T 2=5 72,480 81,337 25,956 0,002 33,130 38,933 35,228 20 14,811 7,5 0,36 0,33 0,69
T 2=6 67,215 75,456 25,707 0,007 27,700 38,931 34,525 20 13,940 7,5 0,38 0,34 0,73
T 2=7 62,024 69,688 25,467 0,007 22,378 38,930 33,840 20 13,124 7,5 0,41 0,36 0,76
T 2=8 56,905 64,022 25,230 0,006 17,155 38,929 33,162 20 12,341 7,5 0,44 0,37 0,81
T 2=9 51,851 58,477 24,988 0,014 12,042 38,918 32,491 20 11,600 7,5 0,47 0,39 0,86
T 2=10 46,876 53,015 24,750 0,018 7,014 38,913 31,820 20 10,871 7,5 0,50 0,41 0,91
Notes: The model with debt constraints (without equity constraints) is used. Base case parameters are: P =10 , C = 0, risk-free rate r = 0.06,
volatility σ = 0.2, competitive erosion δ = 0.06, investment cost I1 = 20, I2 = 20, b = 0.5, tax rate τ = 0.35 and T1 = 0,. T2 ranges from 0 (no-timeto-build)
to time-to-build of 10 years. When time-to-build exists the firm foregoes all cash flows until full completion of the project Useful life
after the investment completion is constant at TF = 20. Debt maturity TD1 = 5 and TD2 = 5 assuming zero principal. An optimal coupon is chosen
based on nc =20 discretization points for each price level with maximum coupon level points being equal to 75% of the price levels (cmax = 15) .
In all tables Ndec = 1 (yearly decisions) with NΔt = 12 steps per year.
354

RE-EXAMINING CAPITAL STRUCTURE TESTS: AN EMPIRICAL ANALYSIS IN THE AIRLINE
INDUSTRY
Kruse Sebastian, Kalogeras Nikos, Semeijn Janjaap
School of Business & Economics, Maastricht University, the Netherlands
Email: n.kalogeras@maastrichtuniversity.nl
Abstract. Research has not reached a consensus as to whether Trade-off or Pecking Order theories best explain actual firms’ financing
behavior. Recently, different authors have examined the impact of competitive strategy and ownership structure on capital structure
decisions, providing empirical support for important interactions. Yet, this type of research has not been included in mainstream
finance theories of capital structure. Our central hypothesis is that existing empirical capital structure tests may be biased since they do
not account for the attributes of competitive strategy and ownership type. We recognize that these two attributes should be included in
the existing empirical tests, and hence, we may improve their explanatory power as well as shed more light on the debate among
Trade-off or Pecking Order theories. Specifically, we modify three existing empirical models to include competitive strategy and
ownership structure and test our central hypothesis in the airline industry. Our empirical results show strong support for the impact of
competitive strategy and ownership structure on capital structure choices. Moreoever, our results demonstrate that existing empirical
tests for Trade-off and Pecking Order theory may indeed be biased. Financial managers and policy makers may gain some useful
insights from these results. First, our results indicate that managers, CEOs, and CFOs should account for the strategic focus of firms
over time when making capital structure decisions. Second, managers should weigh carefully the benefits of lower interest rates
against the reduced ability to issue seasoned equity when thinking about state investors. Finally, our results highlight that managers
should consider the high costs of information asymmetries emphasized by Pecking Order theory, as well as the tax benefit versus
bankruptcy cost argument provided by Trade-off theory
Keywords: capital structure choice, competitive strategy, ownership type, airline industry.
JEL classification: G30; G32.
1 Introduction
Fifty-two years after the seminal work of Modigliani and Miller (1958), there is no consensus among financial
scholars concerning how firms determine their capital structure. Two main theories have emerged: Trade-off theory
(Modigliani & Miller, 1958, 1963) and Pecking Order theory (Myers, 1984; Myers & Majluf, 1984). On the one
hand, Trade-off theorists argue that firms weigh costs and benefits of debt and equity and make their leverage
decision accordingly. Consequently, an optimal leverage ratio exists for each firm. On the other hand, Pecking Order
proponents assert that high costs of information asymmetries drive firms’ preferences for internal to external
finance. As a result, firms only use external finance when all internal means are exhausted. Furthermore, firms
always prefer debt to equity, when debt is available. Thus, Pecking Order theory posits that no optimal leverage ratio
exists and firm leverage is the outcome of past financing needs.
Empirical finance literature does not provide conclusive references favoring either Trade-off or Pecking Order
theories in order to determine and predict capital structure choices (Frank & Goyal, 2008). In fact, both theories
have received some empirical support. In empirical analysis, most of the variables were found to be crosssectionally
correlated to a firm’s leverage. This result has implications consistent with Trade-off theories (Rajan &
Zingales, 1995). However, other studies indicated that profitability is negatively correlated to leverage, as Pecking
Order theory predicts (e.g., Frank & Goyal, 2009). Moreover, some authors strongly support that the target
adjustment processes are consistent with Trade-off theories (e.g., Flannery & Rangan, 2006), while others find the
target-adjustment process to be too slow to matter at all (e.g., Fama & French, 2002). Overall, evidence so far is
mixed and often similar tests have yielded different results depending on the sample employed.
Yet, strategy scholars have begun recently to examine the interaction of strategic variables with capital
structure. Specifically, theorists in strategy suggest that competitive strategy and ownership structure are important
determinants of capital structure (e.g., Williamson, 1988; Barton & Gordon, 1988). First, firms that follow a
differentiation strategy usually have lower degrees of collateralizable assets, and in turn, reduce their leverage
capacity. Moreover, a firm’s differentiation strategy may lead to an excess borrowing capacity for strategic
acquisitions or investments. Empirical evidence provides some support that differentiation strategies are indeed
associated with lower levels of leverage (e.g., Balakrishnan & Fox, 1993). Second, state-owned firms usually enjoy
implicit guarantees on their credit, which decreases their cost of debt. That is, state-owned firms tend to rely more
on debt financing compared to firms financed entirely by private investors (Dewenter & Malatesta, 2001). Hence,
355

theory and evidence from strategy scholars favor the idea that competitive strategy and ownership structure have an
important effect on the level of a firm’s leverage.
Building on strategic management advances (e.g., Williamson, 1988; Balakrishnan, & Fox, 1993; O’Brien,
2003), we hypothesize that there is an important interaction between competitive strategy, ownership structure and
the empirical debate contrasting Trade-off and Pecking Order theory. Specifically, we aim to answer the following
research question: Does the omission of competitive strategy and ownership structure bias the results of existing
empirical finance tests designed to distinguish between Trade-off and Pecking Order theories? To the best of our
knowledge, this is the first empirical study that attempts to examine the relationship between competitive strategy,
ownership structure and a potential bias in existing and commonly used capital structure tests.
The intuition behind this rational is two-fold. First, firms that follow a differentiation strategy may have lower
levels of leverage than low-cost firms. Existing tests, such as the financing deficit test (e.g., Shyam-Sunder &
Myers, 1999), do not account for the fact that there may be systematic differences in the level of leverage according
to different firm strategies. However, these systematic differences in leverage may have a large impact on the test
results and may lead to false conclusions. Second, state-owned firms’ behavior may be consistent with Pecking
Order implications since their capital structure formation relies heavily on debt and they are reluctant to issue equity.
However, the behavior of these firms is not caused by information asymmetries, but it results from the ownership
structure of these firms. State-owned firms are able to obtain debt cheaper and may be restricted from issuing equity.
That is, empirical Pecking Order tests might yield wrong conclusions when a selected sample consists of stateowned
firms. Furthermore, empirical tests designed to detect mean reversion behavior regarding leverage levels may
also yield biased results because state-owned firms cannot decrease their leverage by issuing equity easily
(Dewenter & Malatesta, 2001).
We perform three empirical tests that are designed to detect Pecking Order and Trade-off behavior. Specifically,
we build on existing literature and then extend each test to accommodate our new control variables. First, we test
three cross-sectional variables in order to examine the impact on leverage level of both competitive strategy and
ownership structure. Second, Trade-off theory predicts that firms try to achieve a target level of leverage, but firms
may move away from this target over time. Consequently, we measure the speed of the adjustment process to a
leverage target (Shyam-Sunder & Myers, 1999). Third, Pecking Order theory predicts that financing deficits will be
funded with additional debt issues, which is tested using a financing deficit test (Shyam-Sunder & Myers, 1999). In
order to examine the extent to which airlines may follow Pecking order theory, we modify the extended financing
deficit test (Frank & Goyal, 2009) by including our control variables (i.e. competitive strategy and ownership
structure). Our central proposition is that competitive strategy and ownership structure may bias the results of
existing empirical capital structure tests. Our findings indicate that partial state-ownership is associated with more
Trade-off and less Pecking Order behavior, while the relevant theory suggests opposite behavior. Moreover, a lowcost
strategy is associated with less Pecking Order behavior. In the next sections, we discuss in more detail the
formation of our hypothesis and the results of our empirical study.
2 Hypothesis formation
To be comprehensive in our empirical analysis, we use three types of empirical tests. First, we examine the crosssectional
correlations of factors often indicated as being correlated to firms’ leverage (e.g. Rajan & Zingales (1996)).
Second, we focus on the mean reverting behavior of firms, which is a major characteristic of a firm’s Trade-off
behavior (Shyam-Sunder & Myers, 1999). Third, we use the financing deficit methodology (Frank & Goyal, 2003)
to examine the extent to which firms follow Pecking Order behavior. We structure our hypothesis generation along
these three empirical tests.
2.1 Cross-sectional correlation
To examine the cross-sectional correlations, we use the three variables most consistently associated with leverage in
cross-sectional analyses: profitability, size and M/B. Using this cross-sectional analytical framework, we provide a
detailed picture of the financing behavior of the firms in our sample.
It has been theorized (e.g., Titman, 1984; Balakrishnan & Fox, 1993) that firms following a differentiation
strategy will most likely have lower levels of leverage compared to firms following a low-cost strategy. Usually, a
differentiation strategy entails a high level of firm-specific investments, which cannot easily be collateralized.
Furthermore, differentiated firms often maintain excess borrowing capacity for pursuing strategic acquisitions. Lowcost
firms often have a lower degree of firm-specific investment and often to not pursue any strategic acquisitions
(O’Brien, 2003). Hence, differentiated firms have an incentive to keep their leverage lower than low-cost firms. This
356

elationship has been confirmed almost unanimously in the empirical literature (e.g., Titman, 1984; O’Brien, 2003).
We also expect to find a differentiation strategy to be associated with lower leverage compared to a low-cost
strategy in our sample, hence:
H1: Firms that follow a differentiation strategy are less leveraged than firms that follow a low-cost strategy.
Dewenter and Malatesta (2001) argue that state-owned firms have a lower leverage level than privately-owned
firms for at least two reasons. First, state-owned firms are often prohibited from issuing equity, which may dilute the
ownership of the state. Second, state-ownership serves as an implicit state guarantee for the survival of the firm,
inducing lenders to offer debt to the firm at lower rates compared to privately-owned firms. Several researchers
(e.g., Titman and Wessels, 1988) provide empirical support that links state ownership to a higher degree of leverage.
Following Dewenter & Malatesta (2001) closely, we also expect state ownership to be related to a higher degree of
leverage compared to private ownership.
H2: State-owned firms are more leveraged than privately-owned firms.
Trade-off theorists argue that leverage should increase as profitability increases because a firm’s tax shield also
increases in value when profits are higher. However, Pecking Order theory posits that firms with a high profitability
will have lower leverage as they rely on their internal funds for financing. Empirical evidence strongly favors the
Pecking Order proposition (Rajan & Zingales, 1995). Since there is no evidence supporting a positive relationship
between profitability and leverage, we also expect profitability to be negatively related to leverage.
H3: A firm’s profitability is negatively related to its leverage.
Trade-off theories support a positive relationship between firm size and leverage. That is, large-sized firms have
a lower risk of bankruptcy; thereby they may reduce financial distress costs (e.g., Ang, et al. 1982; Warner, 1977).
Pecking Order theory does not provide a clear interpretation on the role of size in its relation to leverage. Moreover,
most empirical studies rather support the Trade-off theory point of view and, therefore, we also expect a positive
relation between leverage and firm size, as follows:.
H4: A firm’s size is positively related to its leverage.
In accordance with Trade-off theory, the market-to-book ratio is usually associated with high growth
opportunities, which increase the costs of financial distress (Kraus & Litzenberger, 1973). Growth opportunities also
increase potential agency conflicts. This happens because the investment decisions of firms are less transparent to
outsiders (Jensen & Meckling, 1976). Finally, growth options are difficult to collateralize, reducing the maximum
debt capacity of firms (e.g., DeAngelo & Masulis, 1980). Hence, the market-to-book ratio as a measure of growth
options is theorized to be negatively related to leverage in Trade-off theories (Frank & Goyal, 2008). In contrast,
Pecking Order theories claim that, under ceteris paribus conditions, higher growth options lead to more financing
needs and therefore to larger debt accumulation over time (Myers, 1984). Empirical evidence, such as Barclay et al.
(2006), usually finds a negative relation of the market-to-book ratio to leverage. Consistent with this empirical
evidence, we also expect:
H5: A firm’s M/B is negatively related to its leverage.
2.2 Target adjustment
All Trade-off theoretical streams, whether they are based on theoretical premises regarding tax benefits vs.
bankruptcy costs (Kraus & Litzenberger, 1973), agency costs (Jensen & Meckling, 1976) or transaction costs
(Fisher, et al. 1989), support that firms should exhibit some sort of target adjustment behavior in their capital
structure decisions. In other words, if a test can prove that firms follow a target adjustment procedure in their capital
structure decisions, then this is usually taken as evidence in favor of Trade-off theories (Frank & Goyal, 2003).
Empirical studies often use moderate target adjustment procedures in diverse samples (e.g., Shyam-Sunder &
Myers, 2001; Jalilvand & Harris, 1984; Hovakimian, et al. 2001). Although we focus on only one industry, we also
expect to see moderate target adjustment behavior:
357

H6: Firms in our sample exhibit moderate target adjustment behavior.
Competitive strategy has been linked theoretically to leverage levels within firms (e.g., Titman, 1984;
Balakrishnan & Fox, 1993). Empirical studies have confirmed these theoretical linkages (e.g., Balakrishnan & Fox,
1993; Jordan, et al. 1998; O’Brien, 2003). To the best of our knowledge, there are no theoretical predictions
regarding the impact of competitive strategy on the adjustment behavior of firms. Thus, we expect to find no
significant differences in target adjustment behavior in our sample according to competitive strategy.
H7: There are no significant differences in target adjustment behavior between low-cost and differentiated firms.
Dewenter and Malatesta (2001) have discussed how a firm’s state ownership is linked to a higher degree of
leverage. One of the reasons for claiming that state-ownership leads to higher leverage may be the reduced ability of
state-owned firms to issue equity in the secondary markets. That is, often state-owned firms are not allowed to issue
additional equity because these equity issuances would dilute the state ownership, which is against the interests of
the firm’s owner (the state). Here, we adopt the premises made by Dewenter & Malatesta (2001) to our decision
context. If state ownership leads to reduced ability to issue equity, then surely state-owned firms cannot revert back
to pre-defined leverage targets. This premise may be illustrated by using the following example: Suppose that a
shock to the equity market decreases the overall valuations of all firms in an economy (e.g., a financial crisis
occurs). In this situation, firms are moved away from their leverage targets because their equity values decreased.
Hence, firms may have an incentive to issue equity to move back to their leverage targets. Private firms in this
scenario, which have no general restrictions on their equity issues simply issue seasoned equity to restore their target
capital structure. State-owned firms, on the other hand, may find themselves in the situation that they would like to
issue equity, but they are restricted from doing so due to the fact that their state owner is not in favor of diluting its
ownership. The state-owned firms in the economy may then find it hard to revert back to their leverage targets. 1 In
an empirical sample including a considerable number of state-owned firms, slow target adjustment behavior would
be observed and the researcher may conclude that Trade-off theories do not explain this finding. Yet, the slow target
adjustment may simply be attributed to the fact that state-owned companies are incapable to revert back to their
targets. In line with this reasoning, we predict that state-owned firms revert back to their targets more slowly than
privately-owned firms. Therefore, we hypothesize that:
H8: State-owned firms revert back to their leverage target more slowly than privately-owned firms.
2.3 Financing deficit
The Pecking Order theory predicts that firms follow a clear hierarchy in terms of financing (Myers, 1984). First,
firms use internal funds. Second, firms rely on financing instruments with a low degree of adverse selection costs,
e.g., debt. Third, firms only use instruments with high adverse selection costs, such as equity, as a last resort when
their full debt capacity is exhausted (Myers & Majluf, 1984). The financing deficit methodology allows the testing
for one of the central propositions of Pecking Order theory, namely the strong reliance on debt as the primary
financing tool to close financing gaps (Shyam-Sunder & Myers, 1999). In essence, this methodology answers a
simple question: Does a firm use debt as its primary tool for financing, given that its internal funds are exhausted?
Frank and Goyal (2003) use a large sample of firms and show that firms finance their deficits only to a very low
extent with debt, contradicting Pecking Order theory. The authors find a Pecking Order coefficient of roughly 0.28.
Such a coefficient is very small in magnitude and would contradict the predictions of Pecking Order theory.
H9: We expect that the Pecking Order coefficient will be 0.28 over the full sample.
Suppose that two firms exist: A and B. Firm A follows a differentiation strategy and firm B follows a low-cost
strategy. Firm A has a high degree of investments in firm-specific assets, which are not easily collateralized (e.g.,
Titman, 1984; Balakrishnan & Fox, 1993). Moreover, firm A maintains an excess borrowing capacity for future
acquisitions (O’Brien, 2003). Firm B has a much lower degree of firm-specific assets and does not value future
acquisitions that much. Hence, firm A will have a leverage level much lower than firm B. This may happen because
firm A follows a differentiation strategy. Let’s suppose that firm A has a leverage level, measured as debt-over-total-
1 State-owned firms also have the option to issue equity to their state, so that state ownership is actually increased. However, this is not always an
option, as overall shocks to the economy, e.g. financial crises, may also deter states from investing in companies.
358

assets of 40% and firm B has a leverage level, measured exactly in the same way, of 60%. Both firms have a
financing deficit of $1 million in year t. To keep their leverage targets at this point in time, firm A issues $400,000
in debt and $600,000 in common stock. Firm B also intends to keep its leverage level constant and issues $600,000
in debt and $400,000 in common stock. In year t+1, both firms maintain the same leverage levels as in period t (e.g.,
40% and 60% respectively). If the financing deficit methodology will be applied to a sample including firms only
similar to firm A, one may arrive at the conclusion that those two firms finance themselves mostly with equity,
hence the Pecking Order predictions will be rejected. Yet, if the sample includes firms similar to firm B, the findings
may show that the B-like firms mostly use debt to finance their deficits, hence, the Pecking Order theory will be
confirmed. That is, firms A and B simply have different target levels of leverage and finance their deficits
accordingly. Here, we claim that the financing deficit test does not account for the fact that different types of firms
have different levels of leverage. Instead, this test is based on the assumption that all firms have the same target
level. Therefore, we argue that the existing financing deficit test results may be biased because competitive strategy
is not included as a control variable. Following this reasoning, we expect that low-cost firms exhibit more Pecking
Order behavior than firms using a differentiation strategy, hence:
H10: Firms that follow a low-cost strategy exhibit more Pecking Order behavior compared to firms that follow a
differentiation strategy.
Analytical works (e.g., Dewenter & Malatesta, 2001) as well as empirical evidence (e.g., Titman & Wessels,
1988) support that state-owned firms have leverage levels higher than privately-owned firms. Yet, following our
discussion in the previous paragraph (H10), the financing deficit test does not account for these systematic
differences in leverage. Thus, in a sample with a high proportion of state-owned firms, researchers may infer that
firms follow the Pecking Order (high reliance on debt), whereas researchers may come to the completely opposite
conclusion when analyzing a sample including low proportion of state-owned firms. Hence, one may support that
empirical tests may be biased without including state-ownership as a control variable. In line with this reasoning, we
expect that state-owned firms exhibit more Pecking Order behavior than entirely privately-owned firms, hence:
H11: (Partially) state-owned airlines follow mostly a Pecking Order behavior.
3. Sample
To address our research objectives, we use the decision context of the airline industry. We can easily identify
competitive strategy by splitting airlines into two types. First, some airlines offer a full network across the globe
with a high level of service. These airlines follow a differentiation strategy. Second, many airlines offer point-topoint
service at very low costs and relatively low levels of service. These airlines follow a low-cost strategy. 2
Moreover, many firms in the airline industry have a history of state ownership, while most of these firms have
recently been (partially) privatized. Since many governments still hold large stakes in their national airlines, the
airline industry seems to be a prominent decision context for testing the impact of state ownership on capital
structure decisions.
3.1 Data
We used the database “Company.Info” that provides financial information and industry-specific overviews.
Specifically, we selected all firms classified under Transport – Air Transport – Airlines (63 firms). These 63 firms
represent all air transportation firms. We excluded all firms involved only in helicopter or cargo transportation.
Furthermore, we excluded those firms that are not listed in Compustat, assuming that these firms do not have any
financial data publicly available. Our final sample includes 44 airline firms. We obtained quarterly and yearly
accounting data from Compustat. Because data for non-American firms has been available only since 1998, we
restricted our data collection from 1998 to 2009. To complete missing data we also used DataStream and Thomson
One Banker. In the case of missing data records, we reported the data as missing or, if only one observation is
missing, we interpolated between the previous and following observations.
To control for competitive strategy, we followed Kalogeras et al. (2009)’s classification scheme closely. We
assigned a dummy variable to each airline indicating whether it follows a differentiation strategy (full-service
2 Strictly speaking, there are regional airlines, which may follow a different strategy than differentiation or low-cost. However, these airlines are
often relatively small-sized in terms of revenue and market capitalization. Thus, we exclude these companies from our analysis.
359

airlines offering a wide network and high customer service) or low-cost strategy (low-cost airlines offering point-topoint
service at low customer services with a focus on offering low prices). With respect to ownership type, we
followed the classification of Jenkinson (1998): we define ownership categories asymmetrically to account for the
fact that a relatively small governmental stake in an otherwise privately-owned airline may carry significant veto
power, while the same is not necessarily true for a state-owned airline in which a private investor holds an
equivalent stake. Thus, we consider airlines to be privately-owned only in cases where private parties hold 99% or
more of the airline’s outstanding voting stock. In contrast, we categorize an airline as a purely public sector-owned
airline if a state, or several states, own 95% or more of its equity. All other airlines are categorized as “mixed
ownership” (Backx, et al. 2002). To obtain our data using this classification procedure, we used annual reports of
the selected airlines. Apparently, our sample only contains airlines with private and mixed ownership (see Table 2).
This finding may highlight the fact that there are no airlines with pure state ownership anymore due to the
continuous liberalization of the airline markets during the last 30 years.
The airlines in our sample have a mean ratio of total liabilities to assets of 70.8%, which is four percentage
points higher than the mean of this ratio over all industries in the US, but one percentage point lower than the mean
of this ratio in Germany (Rajan & Zingales, 1995). When looking at long-term debt as a measure of leverage, we
recognize that airlines tend to be highly leveraged: 29.9% long-term debt relative to assets is higher than the mean of
this ratio in the US and much higher than the mean of this ratio in Germany. Thus, the selected airlines seem to be
relatively high-leveraged. The selected airlines tend to have very low market valuations. This is most probably due
to low growth opportunities. Airlines in our sample have a very low (2.70) market-to-book ratio (M/B). The
standard deviation for M/B is relatively high, indicating large dispersion in valuations. Furthermore, we observe an
average profitability of only 2.1%, which may explain the low market valuations. Finally, the descriptive
information presented in Table 3 indicates that the average sales of an airline are roughly $4 billion, and its median
sales are around $1 billion.
4. Empirical Modeling Approach
4.1 Cross-sectional correlation
For the cross-sectional analysis we defined leverage as long-term debt divided by total assets. We expected longterm
debt to be a more precise measure of corporate leverage than, say, total liabilities, since total liabilities include
many items, such as accounts payable, which are used for transaction purposes and do not necessarily reflect
corporate leverage. We used the book value of long-term debt as market values are not available for all firms (Frank
& Goyal, 2008). Bowman (1980) finds that no large differences exist between using market and book values, since
correlation between the two measures is very high. Following Rajan and Zingales (1995) closely, we specified our
empirical model for testing the cross-sectional correlations as follows:
M
Leverage i = α + β1
* + β2
* Size + β3*
Pr ofitability
+ β4
* Stategy + β5
* Ownership + εi
B
The market-to-book ratio is defined as market value of equity over book value of equity, each measured at the end
of every quarter. We used profitability in our cross-sectional model to examine airlines’ ability to serve interest
payments and hence to capture the financial-related risks. Therefore, we used quarterly operating cash flows
normalized by the book value of assets as a proxy. Although this measures cash flows rather than profits, we
followed the rationale of Rajan and Zingales (1995) and considered that it is a better measure to capture debt-serving
ability of the firms in our sample. We also used the natural logarithm of quarterly sales in US dollars for capturing
the effect of Size. Next, we expanded the specification of Rajan and Zingales’s (1995) modeling framework by
accounting for the influence of competitive strategy and ownership structure on airlines’ leverage.
4.2 Trade-off theory: Target Adjustment
We applied the target adjustment model introduced by Shyam-Sunder and Myers (1999). In this model, we checked
whether the net changes in debt exhibit a pattern that brings firms closer to a predefined leverage target. In formulae:
*
ΔD = α+ β (D - D ) + ε
it
ta
where: ΔDit is the change in debt from period t-1 to period t for firm i, D * it is the target debt level for firm i and Dit-1
is the leverage in period t-1 for firm i. In other words, we are testing whether new debt issues are used to move the
leverage level back to a predefined target. If target adjustment takes place, βta should be larger than 0, but smaller
360
it
it-1
it

than 1, since we expect significant adjustment costs that prevent perfect adjustment. Consistent with prior research
findings (e.g., Jalilvand & Harris, 1984), we selected the average leverage of firm i over our entire sample period as
the leverage target for firm i. Usage of other targets (e.g., rolling averages) provided similar results. We also chose
the proportion of long-term debt to book value of assets as our proxy for corporate leverage (see: Shyam-Sunder &
Myers, 1999). We treated all observations as equally important (Frank & Goyal, 2003).
4.3 Pecking Order: Financing Deficit
Building on Shyam-Sunder and Myers (1999) and Frank and Goyal (2003), we specified a model to examine
whether the financing needs of a company are met by new debt issues, as predicted by the Pecking Order. We
specified the financing needs, or the financing deficit, of an airline firm as follows:
D EFt
= DIVt
+ CapExt
+ ΔWorkingCapital
t - OperatingCashFlows
t
In line with this model’s specification, the financing deficit of an airline grows larger when more dividends are paid,
more cash flows are used for capital expenditures, and working capital is increased. The financing deficit decreases
when operating cash flows are higher. Our dataset contained the items’ dividends, capital expenditures and operating
cash flows. To estimate the change in working capital we used the common definition of working capital as current
assets minus current liabilities. 3 We examined whether this financing deficit is filled by new debt issues by applying
the model of Frank and Goyal (2003). Formally:
Δ D = a + β * DEF + εi
it
po
where: ΔDit represents the change in debt from one period to the next. The ΔDit should be the amount of new debt
issued when DEF is negative and the amount of old debt retired when DEFit is positive (financing surplus). If βPO
equals exactly one, then it implies that the financing deficit always corresponds to entirely new debt issues or
retirements. That is, when there is a financing gap of $100 million in year t, then a company would issue exactly
$100 million to finance that gap. Naturally, it must be smaller than one because firms also use equity to close
financing gaps. However, if βPO is reasonably close to one, then this would favor the Pecking Order explanation of
capital structure behavior. Since new debt issues are unavailable for non-US firms, we estimate the amount of debt
issued or retired in any year as the difference between the level of long-term debt between one year and the next for
the airlines included in our sample. To be consistent with the target adjustment model, we also followed Frank and
Goyal (2003) and used simple regressions. Finally, we used a truncation procedure at the 4.5% level for the
exclusion of outliers.
5. Results
5.1 Cross-sectional analysis
First, we checked whether differentiated airlines are less leveraged than low-cost airlines and whether state-owned
airlines are more leveraged than private-owned airlines. The results in Table 1 reveal that differentiated airlines are
13 percentage points more leveraged than low-cost airlines in terms of total liabilities over assets. The findings with
respect to the long-term debt over assets, however, show that differentiated airlines are roughly four percentage
points less leveraged than low-cost airlines. This finding confirms our H1. We rather consider that long-term debt
over assets is the more useful measure, since it excludes items used for ongoing business such as supplier credit and
it is focused on the pure debt portion in balance sheets. However, it is not quite clear what causes the ratio of total
liabilities over assets to be substantially higher among differentiated airlines compared to low-cost airlines. That is,
we partially confirm H1: differentiated airlines are less leveraged than low-cost airlines in terms of long-term debt
over assets.
3 For non-US companies this is the most precise measure available to us. To be consistent we also apply this measure to US firms.
361
it
t

Shown values are averages for the individual groups of the average ratios for each airline over the entire sample
Competitive Strategy Ownership
Differentiation Low-cost Private Mixed
Total liabilities / Assets 73.55% 60.80% 72.84% 69.11%
Long-term debt / Assets 29.64% 33.31% 32.24% 27.46%
Table 1– Descriptive Statistics of Airline’s Competitive Strategy and Ownership Structure
The results regarding ownership structure do not confirm that state-owned firms should be more leveraged than
privately-owned airlines (H2). Privately-owned airlines are more leveraged than airlines with mixed ownership
structure, regardless of which measure is used. This is a striking result since state-owned airlines can obtain debt
cheaper. We may only assume that there is a correlation between state-owned airlines and airlines that follow a
differentiation strategy. This correlation effect might have caused this pattern in long-term debt. That is, we examine
empirically our intuition; indeed, ownership and competitive strategy are correlated significantly. 4
Next, we performed cross-sectional regressions for determining the leverage of airlines with the M/B,
profitability and size of the selected airlines. The results of these regressions are presented in Table 2.
Variable Meaning Significance Coefficient Stand. Error t-statistic p-value
α Constant *** 0.50 0.04 14.30 0.00
β1 M/B ** 0.01 0.00 2.20 0.03
β2 Profitability *** -1.57 0.27 -5.91 0.00
β3 Size *** -0.03 0.01 -5.58 0.00
Legend: * = 10% significance, ** = 5% significance, *** = 1% significance
Leveragei = α + β1 * M/B + β2 * Profitability + β3 * Size + εi
Dependent variable: Long-term debt / Assets
R 2 = 0.122 ; Adjusted R 2 = 0.118
Table 2 – Results of Cross-sectional Analysis
In contrast to other studies (e.g. Rajan & Zingales, 1996; Barclay, Morellec & Smith, 2006), our results suggest that
the market-to-book ratio is positively related to leverage. This would imply that firms with more growth
opportunities have higher leverage than firms with fewer growth opportunities. This explanation is consistent with
Pecking Order Theory: higher growth opportunities increase the need for external finance. Although the coefficient
is significant, the impact on leverage is very small in magnitude: an increase in the market-to-book ratio from 1 to 2
would only lead to a 1-percentage point increase in leverage. Based on this finding, we argue instead that the
economic impact of market-to-book on leverage of airlines is virtually zero.
The direction of profitability is consistent with our hypothesis, meaning that airlines with higher profitability
tend to have lower leverage. This result clearly favors Pecking Order descriptions of capital structure (H3):
profitability is negatively related to leverage. It should also be mentioned that the magnitude of the profitability
coefficient is by far the highest compared to the other coefficients. A 1% increase in profitability lowers leverage by
1.57 percentage points. Hence, the leverage of airlines is related significantly and very strongly to their profitability.
Finally, we find a negative sign for the size coefficient. Since size is measured as a logarithmic variable, a 1%
increase in the sales of an airline results in a lower leverage level 0.0003 percentage points. An increase in sales of
50% would lead to a decrease in leverage of 0.0015 percentage points. This impact on leverage seems rather low to
infer any economically meaningful impact on an airline’s leverage in reality. Hence, it seems that the leverage of
airlines is fairly independent of size.
4 coefficient = -0.342, p-value < 0.00
362

Finally, we re-performed our regression by also accounting for the two control variables (i.e., strategy and
ownership) in order to determine an airline’s leverage.
Leveragei = α + β1 * M/B + β2 * Profitability + β3 * Size + β4 * Strategy + β5 * Ownership + εi
Base level Strategy = Differentiation
Base level Ownership = Private
Variable Meaning Significance Coefficient Stand. Error t-statistic p-value
α Constant *** 0.51 0.04 14.70 0.00
β1 M/B 0.00 0.00 0.91 0.36
β2 Profitability *** -2.07 0.26 -8.06 0.00
β3 Size *** -0.02 0.01 -4.94 0.00
β4 Strategy 0.02 0.01 1.39 0.17
β5 Ownership *** -0.09 0.01 -8.18 0.00
Legend: * = 10% significance, ** = 5% significance, *** = 1% significance
Dependent variable: Long-term debt / Assets
R 2 = 0.231 ; Adjusted R 2 = 0.225
Table 3 – Results of Cross-Sectional Analysis including Airlines’ Strategy & Ownership Type
The R² almost doubles from 12.2 (see Table 2) to 23.1%. This increase in explanatory power provides some support
for our intuition that our two control variables are important in explaining capital structure choices. Yet, we find no
significant results for the impact of the market-to-book ratio when we control for airlines’ strategy and ownership
type. This finding confirms our intuition that the market-to-book ratio is not important in explaining capital structure
in our sample. Furthermore, the impact of size decreases in magnitude, but it remains significant. Finally, the
coefficient for profitability is now even more pronounced, indicating a leverage decrease of 2.07 percentage points
for every 1 percentage point increase in profits. In contrast to the results based on descriptive statistics (see Table 4),
we observe no difference in the level of leverage according to competitive strategy. This is another striking result,
since we used long-term debt over assets as our dependent variable. The descriptive statistics also indicated that
differentiated airlines are less leveraged than low-cost ones. We hypothesize that the competitive strategy dummy is
not significant because the ownership dummy is included at the same time and these two variables are significantly
correlated. Indeed, the ownership dummy is highly significant and suggests that airlines with mixed ownership have
a leverage level 9 percentage points lower than private airlines. In the descriptive statistics, we were only able to
detect a difference of roughly 5 percentage points. 5 This finding is not consistent with prior research which argued
that state ownership of firms should lead to higher leverage (e.g., Titman & Wessels, 1988; Dewenter & Malatesta,
2001).
5.2 Target Adjustment
We applied Shyam-Sunder and Myers (1999)’s method/test to examine target adjustment without accounting for
airlines’ strategy and control type. A target adjustment coefficient of 1 would indicate perfect adjustment, so that
each firm is at its target every single year. The results demonstrate stronger target adjustment processes (target
adjustment coefficient = 0.41) than those indicated by Shyam-Sunder and Myers (1999) (target adjustment
coefficient = 0.28). This finding is in favor of Trade-off explanations for capital structure since Pecking Order
theory does not have an explanation even for these moderate adjustment processes. Next, we included interaction
terms between the two control variables and the target adjustment coefficient in our empirical model. This enabled
us to examine whether differences exist regarding the adjustment speed between differentiated and low-cost airlines
and between privately and partially state-owned airlines. First, we observe that the explanatory power of our model
increases by almost ten percentage points. Again, we may interpret this as providing some evidence that our two
new variables contribute to a more comprehensive model. Second, the target-adjustment coefficient has a value of
around 0.28, which indicates moderate target adjustment processes and it is roughly twice as large as the one found
5 This may depend on the measurements used in our descriptive statistics: We used an average of the average leverage of each airline in the
descriptive statistics, whereas in the regression we treat each quarterly data separately.
363

y Shyam-Sunder and Myers (1999). Third, the findings show that neither the strategy nor the ownership dummy is
significant. This means that no significant differences exist due to competitive strategy, followed by the airlines and
their ownership type regarding the level of new debt issued or retired every year. Fourth, the interaction term of
competitive strategy and the adjustment coefficient is insignificant. As hypothesized, there appears to be no
difference in target adjustment behavior between airlines with a differentiation and low-cost strategy.
ΔDit = α + bta (D*it-Dit-1) + β1 * Strategy + β2 * Ownership + β3 * Strategy * (D*it-Dit-1) + β4 * Ownership * (D*it-Dit-1) + eit
Base level Strategy = Differentiation
Base level Ownership = Private
Variable Meaning Significance Coefficient Stand. Error t-statistic p-value
α Constant ** 85.13 40.47 2.10 0.04
bta Target adjustment coefficient *** 0.28 0.04 7.71 0.00
β1 Strategy 5.28 80.19 0.07 0.95
β2 Ownership -95.42 61.83 -1.54 0.12
β3 Interaction with Strategy -0.08 0.24 -0.32 0.75
β4 Interaction with Ownership *** 0.44 0.07 6.74 0.00
Legend: * = 10% significance, ** = 5% significance, *** = 1% significance
R 2 = 0.430 ; Adjusted R 2 = 0.421
Table 4 – Target Adjustment Test including Control Variables and Interaction Terms
Finally, the interaction term of ownership structure and target adjustment is significant and very large. For
interpretation, the interaction term and the target adjustment coefficient need to be added. Hence, airlines with
partial state ownership revert back to their leverage target much more quickly than privately-owned airlines
(adjustment coefficient = 0.72). 6 To the best of the authors’ knowledge, there is no underlying theoretical
explanation regarding the impact of airlines’ state-ownership status which should lead to stronger target adjustment.
It could be that when companies move away from their leverage target, they often issue more debt. Consequently,
there is a need to rebalance capital structure by issuing equity. This new equity issue is not in the interest of a state
that already holds shares in an airline. It may be that either new investors come in, and hence the state ownership
will be diluted or the state has to buy the issued equity. That is, the state will have to pay for the re-adjustment to the
target level either by diluted ownership or by buying up the issued shares.
5.3 Financing Deficit
We performed the original financing deficit test without additional control variables to check if we observe the same
behavior as Shyam-Sunder and Myers (1999) and Frank and Goyal (2003) observed in their respective studies.
Based on the results fo this test our Pecking Order coefficient is 0.53 with an R² of 38.3 percent. The estimated level
of this coefficient implies that only 53% of the financing deficit of airlines may be “filled in” by new debt issues.
Shyam Sunder and Myers (1999) find a coefficient of 0.85 (R² = 86%), but restrict their sample to firms for which
continuous data is available. Frank and Goyal (2003) find a coefficient of 0.28 (R² = 27%) by expanding the sample
of Shyam-Sunder and Myers (1999) to include firms with reporting gaps. Thereby, our coefficient’s level stands in
the middle of those two prior estimations, both in terms of magnitude and explanatory power. Our findings do not
reconcile this finding with the Pecking Order behaviour, which predicts that the financing deficit will always be
filled with new debt issues, except when debt capacity is exhausted.
6 This result is computed as follows: the sum of the general target adjustment coefficient (0.28) + the interaction term (0.44) = 0.72.
364

ΔDit = α + bpo (DEFit) + β1 * Strategy + β2 * Ownership + β3 * Strategy * (DEFit) + β4 * Ownership * (DEFit) + εit
Variable Meaning Significance Coefficient Stand. Error t-statistic p-value
α Constant *** 152.75 42.07 3.63 0.00
bpo Pecking Order Coefficient *** 0.65 0.05 13.95 0.00
β1 Strategy -7.88 81.85 -0.10 0.92
β2 Ownership ** -71.34 62.08 -1.15 0.25
β3 Interaction with Strategy ** -0.27 0.13 -2.02 0.05
β4 Interaction with Ownership *** -0.45 0.10 -4.72 0.00
Legend: * = 10% significance, ** = 5% significance, *** = 1% significance
Base level Strategy = Differentiation
Base level Ownership = Private
R 2 = 0.439 ; Adjusted R 2 = 0.429
Table 5 – Frank and Goyal (2003)’s Test including Control Variables & Interaction Terms
The results of Table 5 provide challenging insights into financing deficit levels when we account for strategy
and ownership type of the airlines, and their interaction terms are also included. Similarly to our previous test
results, the explanatory power rises by 10 percentage points when we include the control variables. The Pecking
Order coefficient increases to 0.65. Again, the level of this coefficient ranges between the one indicated by Shyam-
Sunder and Myers (1999) and the one by Frank and Goyal (2003). That is, we rather support once more that a
coefficient of 0.65 is hard to reconcile with Pecking Order behavior. Furthermore, the insignificant result for the
Strategy dummy may imply that there is no difference in the level of new debt issues between low-cost airlines and
those that follow a differentiation strategy. Yet, the Ownership dummy equals -71.34 and is significant. This implies
that airlines with state ownership issue roughly $71 million less in new debt than privately-owned airlines in each
year. The state-owned airlines tend to be less leveraged than private ones. Hence, they will issue less new debt each
year to replace maturing portions of debt. We need to be careful regarding interpretation though, because the
financing deficit test does not control for firm size. So, if private airlines are larger in size than state-owned airlines,
it may imply that state-owned airlines issue less debt due to their smaller size.
6. Conclusion
On balance, our results support the intuition that a firm’s competitive strategy and ownership structure have an
influence on capital structure and may bias the results of the existing empirical tests. We used the decision context
of the airline industry. First, we find differences in leverage according to strategy and ownership structure in
descriptive statistics and cross-sectional regressions. Second, we find that partially state-owned airlines exhibit much
stronger target adjustment processes than privately-owned airlines. Third, partially state-owned airlines show less
Pecking Order behavior than differentiated airlines. This result corroborates the intuition that state ownership may
be associated with Trade-off behavior. Fourth, we find that low-cost airlines exhibit less Pecking Order behavior
than differentiated airlines, which suggests a significant role for competitive strategy in capital structure tests. These
findings expand prior research findings on the impact of competitive strategy (Barton & Gordon, 1988;
Balakrishnan & Fox, 1993) and ownership structure (Titman & Wessels, 1988; Dewenter & Malatesta, 2001) on a
firm’s corporate behavior. At a practical level, financing managers and practitioners may recognize the role of
strategy in capital structure decision-making. That is, managers and policy makers in an industry (e.g., airline)
should carefully weigh the benefits of lower interest rates against the reduced ability to issue seasoned equity when
considering state investing behavior in a specific industry. For instance, industry managers may consider the high
costs of information asymmetries emphasized by Pecking Order theory as well as the tax benefit versus bankruptcy
cost argument provided by Trade-off theory.
Several caveats and challenges should be mentioned. First, we used a rather limited sample and focused on only
one industry. Future research may examine the potential biases of existing tests for multiple industries and larger
samples. Second, we used an international sample of airlines that necessitates currency conversions, which
otherwise may have an influence on our results. Finally, the decision context of the airline industry may not be
365

comparable to many other industries because of its high concentration and competitiveness. These unique
characteristics of the airline industry may have been mirrored in the financial data of airlines, hence making it
difficult to generalize our results.
7. References
Ang, J., Chua, J., & McConnell, J. (1982). The Administrative Costs of Corporate Bankruptcy: A Note. Journal of Finance, 37(March), 219-226.
Balakrishnan, S., & Fox, I. (1993). Asset Specificity, Firm Heterogeneity and Capital Structure. Strategic Management Journal, 14(1), 3-16.
Barclay, M., Morellec, E., & Smith, C. (2006). On the Debt Capacity of Growth Options. Journal of Business, 79(1), 37-59.
Barton, S. L., & Gordon, P. J. (1988). Corporate Strategy and Capital Structure. Strategic Management Journal, 9(6), 623-632.
Bowman, J. (1980). The Importance of a Market Value Measurement of Debt in Assessing Leverage. Journal of Accounting Research, 18(1),
242-254.
DeAngelo, H., & Masulis, R. W. (1980). Optimal Capital Structure under Corporate and Personal Taxation. Journal of Financial Economics,
8(1), 3 - 29.
Dewenter, K. L., & Malatesta, P. H. (2001). State-Owned and Privately Owned Firms: An Empirical Analysis of Profitability, Leverage, and
Labor Intensity. American Economic Review, 91 (1), 321-334.
Fama, E., & French, K.R. (2002). Financing Decisions: Who Issues Stock. Journal of Financial Economics, 76 (June 2005), 549-582.
Flannery, M. J., & Rangan, K. P. (2006). Partial Adjustment toward Target Capital Structure. Journal of Financial Economics , 79(3), 469-506.
Frank, M. Z., & Goyal, V. K. (2008). Trade-off and Pecking Order Theories of Debt. In B. E. Eckbo (Ed) Handbook of Corporate Finance:
Empirical Corporate Finance (Vol. 2, Chapter 12), Nortrh Holland-Amsterdam: Elsevier Handbooks in Empirical Finance.
Frank, M. Z., & Goyal, V. K. (2009). Capital Structure Decisions: Which Factors are Reliably Important? Financial Management, 38(1), 1-37.
Frank, M., & Goyal, V. (2003). Testing the Pecking Order Theory of Capital Structure. Journal of Financial Economics, 67(2), 217-248.
Hovakimian, A., Opler, T., & Titman, S. (2001). The Debt-equity Choice. Journal of Financial and Quantitative Analysis, 36(1), 1-24.
Jalilvand, A., & Harris, R. S. (1984). Corporate Behavior in Adjusting to Capital Structure and Dividend Targets: An Econometric Study. Journal
of Finance, 39(1), 127-145.
Jalilvand, A., & Harris, R. S. (1984). Corporate Behavior in Adjusting to Capital Structure and Dividend Targets: An Econometric Study. Journal
of Finance, 39, 127-145.
Jenkinson, T. (1998). Corporate Governance and Privatization via Initial Public Offering. Corporate Governance, State-owned Enterprises and
Privatizations - OECD Proceedings (pp. 87-118), Paris, France: OECD Publishing.
Jensen, M. C., & Meckling, W. H. (1976). Agency Costs and the Theory of the Firm. Journal of Financial Economics , 3, 305-360.
Jordan, J., Lowe, J., & Taylor, P. (1998). Strategy and Financial Policy in UK Small Firms. Journal of Business Finance and Accounting, 25, 1-
27.
Kalogeras, N., Labryga, T., Kruse, S., and J. Semeijn. (2010). Determining Capital Structure Choices in Transportation Industry: A Comparison
Among the Main Theoretical Angles. International Conference on Global Trends in the Efficiency & Risk Management of Financial
Services, EURO-Working Group on Efficiency and Productivity Analysis (EWG-EPA), Chania, Crete, July, 2-4, 2010, Greece, EU, pp.
52.
Kraus, A., & Litzenberger, R. H. (1973). A State-preference Model of Optimal Financial Leverage. Journal of Finance, 28, 911-922.
Modigliani, F., & Miller, M. H. (1958). The Cost of Capital, Corporation Finance and the Theory of Investment. The American Economic
Review, 48, 261-297.
Modigliani, F., & Miller, M. H. (1963). Corporate Income Taxes and the Cost of Capital: A Correction. The American Economic Review, 53, 433
- 443.
Myers, S. C. (1984). The Capital Structure Puzzle. The Journal of Finance, 39, 575 - 592.
Myers, S. C., & Majluf, N. S. (1984). Corporate Financing and Investment Decisions when Firms have Information that Investors do not have.
Journal of Financial Economics, 13, 187 - 221.
O'Brien, J. P. (2003). The Capital Structure Implications of Pursuing a Strategy of Innovation. Strategic Management Journal, 24, 415-431.
Rajan, R. G., & Zingales, L. (1995). What Do We Know About Capital Structure? Some Evidence From International Data. Journal of Finance,
50, 1421-1460.
Shyam-Sunder, L., & Myers, S. C. (1999). Testing Static Trade-off Against Pecking Order Models of Capital Structure. Journal of Financial
Economics, 51, 129 - 244.
Titman, S. (1984). The Effect of Capital Structure on a Firm's Liquidation Decision. Journal of Financial Economics, 13(March), 137-151.
Titman, S., & Wessels, R. (1988). The Determinants of Capital Structure Choice. The Journal of Finance, 43(1), 1 - 19.
Warner, J. (1977). Bankruptcy Costs: Some Evidence. Journal of Finance, 32(2), 337-347.
Williamson, O. (1988). Corporate Finance and Corporate Governance. Journal of Finance, 43(3), 567-591.
366

DEUNDERSTANDING THE DIFFERENCE IN PREMIUM PAID IN ACQUISITION OF PUBLIC
COMPANIES
Nahum Biger 1 and Eli Ziskind
Abstract. The variance in acquisition premium of different companies is driven by the difference in the set of factors mentioned
above, between different acquisition targets.Understanding the drivers of acquisition premium variance is an important subject in
investment theory as it allows investors seeking to invest in publicly listed companies, undergoing or likely to undergo merger or
acquisition processes, to better analyze the effect of such processes on a target's share price (due to potential premium to be
introduced).This research is aimed at explaining the variance in premium paid in acquisition transactions. It breaks down the
transaction data set into geographic regions, as well as by identifying specific variables that have a significant correlation with
premium variance.This research suggests a different approach in analyzing the variance of premium paid in acquisition transactions by
suggesting a multi-variable regression analysis which divides the data set into geographic regions. A sample of 607 mergers
completed between December 1995 and September 2009 was studied.
1. Introduction
Since the early 1980s the number of merger and acquisition transactions has gradually increased. Companies have
been acquired by private individuals, funds, competitors etc.
An acquisition price is typically driven by multiple factors, comprising among other: projected and historic
financial performance, anticipated value of synergies between the acquired company (“target”) and the acquirer, the
stake that is being acquired (namely, whether it is a control stake or not), market share and technological advantage
of the target, etc.
Acquisition premium is normally referred to as the difference between the market value of a target’s share
capital and the price paid for the share capital by an acquirer of the shares.
The purpose of the research is to try and explain part of the variance in acquisition premium paid in different
acquisition transactions using a multi-variable regression analysis and to opine on whether the variance could be
better explained while breaking the dataset into transactions origin in different regions.
Acquisition premium is a well debated subject in the economic literature. Many researches have elaborated on
their view of how to measure an acquisition premium in the past. Leonce, Bargeron, Frederik, Schlingemann, Stulz
and Zutter (2008) measure the premium by the difference between the price paid for the share capital of a target and
the market value of a target’s share capital as of 3 days, or 42 days prior to the acquisition announcement;
Gaspara,Massab and Matosb (2005) choose the days to be 63 and 123 days prior to the acquisition announcement;
Lammanen (2007) uses 1 day, 1 week and 1 month prior to the announcement date; Porrini (2006) refers to 4 weeks
prior to the acquisition announcement, and Song and Ralph (2000) defined acquisition premium as the difference
between the highest price paid per share and the target share price four weeks prior to the announcement date
While different researches have used different definitions for acquisition premium, the majority seem to be
using a premium calculated as the difference between the share price paid by the acquirer for the target’s shares and
the average market price of the target share measured by a time range of between 4 and 8 weeks prior the
announcement date.
Thus, researchers disagree about the time frame on which a premium is measured. While normally a researcher
would tend to compare the acquisition price to the latest market price available, prior to the transaction, doing so
could bias the results and provide an indication of a lower premium.
1 Carmel Academic Center, Haifa, Israel. The paper is based on Mr. Ziskind’s thesis at the University of Haifa, Haifa, Israel
367

The reason for that could be the effect information leakages may have on the share price of a target. As a
transaction is getting close towards an announcement, individuals comprising, among others, employees of the target
and acquirer, consultants, professional service providers etc. are becoming aware of the anticipated announcement
and could leak information to the market. Hence, the longer time frame (prior to the announcement) a premium
would be applied on, the lower the effect of information leakages will be. However, taking a too long time frame for
premium calculation would be inaccurate as well due to the outdated market price (which will not reflect the effect
of recent corporate announcement by the target to the market).
In order to minimize the likelihood of using a biased acquisition premium and due to the limited available
information, we have calculated acquisition premium based on the difference between the highest price paid per
share within an acquisition of a target and the target share price four weeks, one week and one day prior to the
announcement date.
Prior research into the premium variance include Bargeron, Schlingemann, Stulz and Zutter (2008) who found
that premium paid in acquisitions were 35% higher when the acquirer was a public, rather than private, company.
The authors have not explained this finding. They also found that premium paid by strategic acquirers (corporate
companies or “corporates”) were 63% higher than those paid by financial acquirers (private equity firms\funds).
They suggested that corporates pay more for acquisitions as they expect to benefit from synergies with their existing
activities rather than financial acquirers which usually have no synergies with their existing activities.
Moeller, Schlingmann, and Stulz (2004) report that large firms tend to pay higher premium (in percentage of
acquisition price) relative to smaller size firms. They conjecture that larger firms could realize more synergies from
acquisitions due to the usage of larger integration teams or having more experience in synergies realization.
Hanouna, Sarin, and Shapiro (2001) found that higher percentage premium was paid in acquisition of control
transactions. They explain that an acquirer is probably willing to pay a higher price when acquiring control as it
provides the acquirer with better access to the target as well as ability to better utilize the target’s resources and
realize synergies. They also found that 30% higher premium was paid in US transactions (i.e. both the target and the
acquirer are US based) relative to cross-border transactions.
Former researchers have found that the size premium paid in acquisitions is driven by the type of acquirer,
general economic conditions (e.g. under active merger & acquisition- “M&A”) and market conditions. Premium
tend to be higher due to the general positive prospect of financial results (of the targets) and the competitive bid
atmosphere, size (measured by market capitalization or revenues for example) of the acquirer and the percentage
acquired in the target (in particular, whether control has been acquired).
It I clear that the variance in acquisition premium percentage is driven by multiple factors. In addition,
transactions happening in different global regions might be completed with different premia as regions tend to differ
from one another in terms of the level of development, competition, opportunities, projected growth etc.
For example a retailer operating in a stable western market such as France, would provide its potential acquirer
with a relatively clear growth prospect. Revenues of such a company would normally tend to be stable at the long
run as they are closely correlated with the steady long term growth of the population and GDP per capita. A retailer
based in India however, could provide its potential acquirer with a more rapid growth profile due to the fast growing
population rate and constantly increasing GDP per capita. It would also tend to be associated with a higher risk level
due to the unstable market environment and lack of market “education” or “tradition”.
The pitfalls of measuring the variance in acquisition premium are that having too many variables could lead to
biased as well as non-significant results. In addition there might be some autocorrelation that might obscure the
results.
Another problem is the lack of variables that represent insider information about the target, held by the acquirer,
ahead of the contemplated transaction. Such information could comprise, among others, contracts with major clients,
intellectual property rights and any otherinformation collected by the potential acquirer during the due diligence
process or along the transaction which will enable the acquirer to evaluate the size of synergies, as well as added
value contributed by other factors. Such a variable does not exist and could not be represented by a proxy.
368

The pitfalls mentioned below have been avoided by:
� Using the “step wise routine” to eliminate variables that do not provide further explanation to the regression’s
results and hence “maximizing’ the R-squared of the regression
� The inability to find a variable that provides details on the “non-publicly available” information could be offset
by the fact that a final price, achieved in an acquisition transaction, comprises multiple factors and is being
finally determined by negotiations which usually tend to be bounded by historic premium paid in similar
transactions or in transactions completed in specific regions throughout a given time frame, which lowers the
magnitude of missing a variable from the regression equation.
The study provides a unique insight to the existing literature on global merger & acquisition transactions as it
will use the “step wise routine” method in order to try and explain the variance in premium paid in M&A
transactions by using multiple variables as well as try opine on whether the variance could be better explained by
breaking the dataset into transactions completed in different geographic regions.
2. Research Hypotheses
1. Variance in acquisition premium could be better explained in specific geographies, relative to global data set.
2. Variance in acquisition premium is positively correlated with: change of control, revenue growth, EBITDA
growth, use of advisors and deal attitude.
3. Variance in acquisition premium is negatively correlated with: target LTM return, implied equity to book
value, the ratio between enterprise value and the last twelve months (“LTM”) revenues, the ratio between
enterprise value to LTM EBITDA and the ratio between market capitalization and net income.
3. Data description
Data base: transaction database with the following characteristics:
Transactions sourced out of M&A Monitor service, CapitalIQ website, SDC data-source and Reuters, including data
on 607 merger and acquisition transactions that have been announced and completed (i.e. not “pending”) between
December 1995 and September 2009; Target companies that have been listed for trade on public stock exchanges.
Transactions in which a significant minority stake or control have been transferred to the acquirer;
Transactions that have been completed in the following regions: the Americas (North and South), Europe, Asia
Middle East and Africa (“MENA”); Transactions in which premium could be identified for 1-day, 1-week and 1month
prior to the transaction announcement; Transactions comprising both strategic and financial types of
acquirers;Targets with projected revenues, EBITDA and net income for 1 and 2 years after the date of the
acquisition
Dependent variables: percentage premium paid, measures trice: One month prior, one week prior and one day prior
to the announcement of the transaction.
Explanatory (independent) variables:
� Transaction value – represents the value of the target in millions of US$ as of the transaction date. Transaction
value should theoretically be positively correlated with the absolute value (in millions of US$) of premium paid
in a transaction, as premium is paid in percentage of a company’s value. However, as premium is measured in
% rather than in absolute value, transaction value should not be correlated to premium.
� Percentage sought – represents the % of shares (out of the total share capital of the target) that have been
acquired by the acquirer. This variable should be positively correlated with premium size as theoretically, the
more percentages an acquirer acquires in a target the higher ability the acquirer will have to implement
synergies and cost cutting processes and increase the return on its investment. Hence percentage variable
represented in percentages should be positively correlated with the premium size.
� Target Market Capitalization transaction date -1 day – represents the market value of the target in millions of
US$ one day prior to the acquisition announcement. This variable should theoretically be positively correlated
with the absolute value (in millions of US$) of premium paid in a transaction, as premium is paid in percentage
369

of a company’s value. However, as premium is measured in % rather than in absolute value, the variable should
not be correlated to premium.
� Target total revenue LTM – represents the revenues of the target in millions of US$ over the last twelve months
prior to the date of the acquisition. This variable should theoretically be positively correlated with the absolute
value (in millions of US$) of premium paid in a transaction, as premium is paid in percentage of a company’s
value and the bigger revenues are, the larger would the company’s value be. However, as premium is measured
in % rather than in absolute value, the variable should not be correlated to premium.
� Target total EBITDA LTM – represents the EBITDA of the target in millions of US$ over the last twelve
months prior to the date of the acquisition. This variable should theoretically be positively correlated with the
absolute value (in millions of US$) of premium paid in a transaction, as premium is paid in percentage of a
company’s value and the bigger EBITDA is, the larger would the company’s value be. However, as premium is
measured in % rather than in absolute value, the variable should not be correlated to premium.
� Target LTM return – represents the change (in percentage) between the target’s share value as of 1 year prior to
the date of the acquisition and the value as of the last day prior to the acquisition date. This variable should
theoretically be negatively correlated with the premium paid in a transaction, as the higher the return is, the
larger has the share price of the target increased of the last twelve months. Hence if a target’s share price
increased significantly, the target would be perceived as less attractive in terms of value and the acquirer would
logically be willing to pay a lower premium for it.
� Revenue growth – represents the change (in percentage) between the target’s revenue over the last twelve
months. This variable should theoretically be positively correlated with the premium paid in a transaction, as the
higher the revenue growth is, the more attractive the target is as it embeds attractive growth opportunity and
value enhancement. Hence, the acquirer will be willing to pay more for a company with a better growth
potential and revenue growth could serve as a proxy for growth prospects.
� EBITDA growth – represents the change (in percentage) between the target’s EBITDA over the last twelve
months. This variable should theoretically be positively correlated with the premium paid in a transaction, as the
higher the EBITDA growth is, the more attractive the target is as it embeds attractive growth opportunity and
value enhancement. Hence, the acquirer will be willing to pay more for a company with a better growth
potential and EBITDA growth could serve as a proxy for growth prospects.
� Last twelve months EBITDA margin– represents a target’s EBITDA margin of revenues (in percentage) over
the last twelve months. This variable should theoretically not be correlated with the premium paid in a
transaction as different businesses have different margin levels and premium paid in their acquisitions should
not be subject to margin.
� Implied equity value to book value – represents a target’s equity value (market value of share capital), divided
by the target’s book value (accounting value of the assets less accounting value of the liabilities) in percentages.
This variable should theoretically be negatively correlated with the premium paid in a transaction as the higher
the ratio is, the higher the target is valued (as the market estimates a premium exists over the book value of the
shares). And should a target be highly valued, normally an acquirer would tend to pay a lower premium on top
of the value.
� Enterprise value to last twelve months revenues – represents a target’s enterprise value (market value of share
capital plus the net debt), divided by the target’s revenues over the last twelve months prior to the acquisition in
percentages. This variable should theoretically be negatively correlated with the premium paid in a transaction
as the higher the multiple is, the higher the target is valued (as the market compensate the target with a higher
valuation relative to its actual financial performances). And should a target be highly valued, normally an
acquirer would tend to pay a lower premium on top of the value.
� Enterprise value to last twelve months EBITDA – represents a target’s enterprise value (market value of share
capital plus the net debt), divided by the target’s EBITDA over the last twelve months prior to the acquisition in
percentages. This variable should theoretically be negatively correlated with the premium paid in a transaction
as the higher the multiple is, the higher the target is valued (as the market compensate the target with a higher
valuation relative to its actual financial performances). And should a target be highly valued, normally an
acquirer would tend to pay a lower premium on top of the value.
� Market capitalization to last twelve months net income – represents a target’s market value of its share capital,
divided by the target’s net income over the last twelve months prior to the acquisition in percentages. This
variable should theoretically be negatively correlated with the premium paid in a transaction as the higher the
multiple is, the higher the target is valued (as the market compensate the target with a higher valuation relative
370

to its actual financial performances). And should a target be highly valued, normally an acquirer would tend to
pay a lower premium on top of the value.
� Use of advisors – represents a dummy variable which is valued 0 when no advisors have been used in a
transaction and 1 when advisors have been used. This variable should theoretically be positively correlated with
the premium paid in a transaction as advisors are being hired by targets in order to assist the target to be sold at
its full potential value (i.e. including maximum premium). Advisors are running auction processes and their
added value is the additional premium paid.
� Majority / minority – represents a dummy variable which is valued 1 when the transaction represents a majority
(over 50% of the shares) transaction and 0 when the transaction represents a minority transaction. This variable
should theoretically be positively correlated with the premium paid in a transaction as majority acquisition
allows the acquirer to consolidate the target’s financial statements into its own and also to more efficiently
execute synergies measures.
� Change of control – represents a dummy variable which is valued 1 when the transaction represents a control
(legal or effective) transfer and 0 when the transaction does not represent a control transfer. It should be noted
that control could be legal (above 50% of the shares in most of the markets or above 30% in the UK) or
effective could be lower than 50% but with rights to appoint board members). This variable should theoretically
be positively correlated with the premium paid in a transaction as majority acquisition should allow the acquirer
to control the target, navigate its activities and implement any measures.
� Deal attitude – represents a dummy variable which is valued 1 when the transaction has been executed at
friendly terms (the board of the target welcomed the acquirer and cooperated during the transaction) and 0 when
the transaction has been executed at a hostile manner (the board of the target did not welcome the acquisition
and the acquirer had to submit a hostile tender offer to the target’s shareholders). This variable should
theoretically be negatively correlated with the premium paid in a transaction as hostile bids are not
recommended by the board of directors of the target, hence the shareholders of the target should be convinced
to actually accept the offer (which often requires the acquirer to pay a significant premium).
Descriptive statistics for the explanatory variables
Geography Global The Americas
Premium 1-month 1-week 1-day 1-month 1-week 1-day
Average 34% 29% 24% 36.8% 31.1% 26.8%
Median 31% 26% 22% 33.1% 28.8% 24.6%
Max 99% 93% 69% 98.8% 92.9% 68.7%
Min 0% 1% (4%) 0.3% 1.1% (1.6%)
STD 19% 17% 16% 19.7% 17.2% 15.8%
Mode 33% 33% 7% 33.3% 33.3% 33.3%
Skewness 0.7 0.6 0.7 0.6 0.6 0.6
Variance 4% 3% 2% 3.9% 3.0% 2.5%
Geography Europe Asia and MENA
Premium 1-month 1-week 1-day 1-month 1-week 1-day
Average 28% 22% 19% 33.0% 27.2% 22.1%
Median 24% 18% 14% 31.7% 26.7% 21.4%
Max 84% 66% 67% 70.9% 72.0% 64.8%
Min 0% 4% 0% 3.3% 5.0% (4.4%)
STD 17% 14% 13% 17.0% 14.8% 14.9%
Mode 24% 17% 7% 33.3% 32.7% 14.4%
Skewness 1.0 0.9 1.1 0.4 0.5 0.8
Variance 3% 2% 2% 2.9% 2.2% 2.2%
4. Overview of the research method
The analysis used OLS regressions under different scenarios and limitations as follows:
� Stage 1 – regressions implemented with and without dummy variables, for each of the three dependent
variables: premium paid measured by 1- month, 1-week and 1-day prior to transaction announcement;
371

� Stage 2 –separate regression broken into 4 geographic regions: global (all regions grouped together), Europe,
North America and ASIA & Middle East & Africa (“MENA”).
� Stage 3 – Step-wise regressions, with and without dummy variables, for the same dependent variables.
� Stage 4 – Step wise regressions, with and without dummy variables, broken into 4 geographic regions.
The table below provides summary results of the regressions.
Region Global Americas Europe Asia/MENA
General data
1-month premia
Average premium 34.4% 36.8% 27.6% 33.0%
Median premium 31.3% 33.1% 23.7% 31.7%
Standard deviation 19.2% 19.7% 17.2% 17.0%
1-week premia
Average premium 28.8% 31.1% 22.1% 27.1%
Median premium 26.4% 28.8% 18.4% 26.7%
Standard deviation 16.6% 17.2% 14.0% 14.6%
1-day premia
Average premium 24.5% 26.8% 18.8% 22.0%
Median premium 22.0% 24.6% 14.5% 20.9%
Standard deviation 15.6% 15.8% 13.3% 14.7%
Regressions statistic information
standard
regression
step-wise
routine
standard
regression
step-wise
routine
standard
regression
step-wise
routine
standard
regression
step-wise
routine
1-month premia
Includding dummy variables
R-squared 7.8% 7.5% 7.1% 5.8% 27.0% 20.4% 33.1% 22.2%
% of significant variables 16.7% 100.0% 11.1% 75.0% – 100.0% 5.6% 66.7%
Excludding dummy variables
R-squared 5.8% 6.1% 6.4% 5.8% 21.8% 15.7% 27.8% 19.0%
% of significant variables 14.3% 83.1% 14.3% 75.0% – 75.0% 14.3% 100.0%
1-week premia
Includding dummy variables
R-squared 8.6% 7.8% 8.5% 6.9% 22.9% 10.1% 32.5% 28.6%
% of significant variables 22.2% 100.0% 22.2% 50.0% – 100.0% 5.6% 83.3%
Excludding dummy variables
R-squared 6.0% 4.7% 6.3% 5.0% 19.3% 10.1% 24.4% 16.7%
% of significant variables 14.3% 60.0% 21.4% 100.0% – 100.0% 14.3% 100.0%
1-day premia
Includding dummy variables
R-squared 11.4% 10.6% 13.9% 11.8% 25.1% 17.4% 33.1% 26.0%
% of significant variables 16.7% 71.4% 27.8% 83.3% 5.6% 100.0% 5.6% 80.0%
Excludding dummy variables
R-squared 9.8% 1.3% 10.6% 8.9% 23.0% 19.1% 32.5% 23.3%
% of significant variables 28.6% – 28.6% 100.0% 14.3% 80.0% 7.1% 80.0%
A few clear insights could be driven from the table above:
1. Generally, standard deviation in premium paid has decreased in the different regions (with the exception of
the Americas) relative to the global results.
2. Regressions run on the global market data set, without breakdown per market, provided less significant
results (lower R-squared) and had a lower number of significant variables with the exception of the
Americas
3. Amongst the regions, regressions implemented on Europe and ASIA & MENA have yielded more
significant results
4. Across the regressions run, those that combined dummy variables, generally had higher R-squared, yet
lower number of significant variables
5. Where regressions seemed more effective (Europe and Asia & MENA) regressions implemented with 1
month premium as the explained variable have provided better results in most cases (higher R-squared)
and had more significant variables.
372

5. Result
Variable Coefficient
Number of times appear in
regressions (out of 48)
% of coefficient regressions
usage of variables
Change of Control (yes=1, no=0) Positive 31 71.0%
Deal Attitude (friednly=1, hostile=0) Positive 26 53.8%
EBITDA growth LTM Negative 34 61.8%
Enterprise value / EBITDA LTM Negative 15 73.3%
Enterprise value / revenue LTM Positive 19 89.5%
Implied Equity Value/Book Value Positive 27 29.6%
LTM EBITDA margin% Positive 30 50.0%
Revenue growth LTM Positive 30 80.0%
Target Enterprise value transaction date Negative 37 83.8%
Target LTM Return % (%) Negative 43 81.4%
Target Market Capitalizatiat transaction date -1 day Negative 28 60.7%
The insights from the table above could be summarized as follows:
1. Change of control has a positive correlation with premium paid in acquisitions. It showed consistent
positive correlation results and mostly the variable was significant
2. Deal attitude had a positive correlation with premium paid (i.e. a friendly deal attitude provided a higher
premium usually).
3. EBITDA growth LTM had a negative correlation with premium paid. This implies that companies with a
higher growth rates received lower premium upon acquisition.
4. Enterprise value / revenue LTA and Enterprise value / EBITDA LTM had opposite correlation results. Both
of these two variables measure the correlation between premium and market valuation. However,
Enterprise value / revenue LTM showed more significant results and was found positive.
5. Implied equity value / book value had positive correlation to premium, effectively implying that companies
with higher market valuations were acquired at higher premium. However, results were not mostly
significant.
6. LTM EBITDA margin had positive correlation to premium paid. Thus, companies with higher margins
receive higher premium upon acquisition.
7. Revenue growth LTM had positive correlation to premium paid. Hence, companies with higher growth
rates receive higher premium upon acquisition.
8. Target enterprise value at transaction date had negative correlation to premium paid, implying that
companies with relative high enterprise value received lower premium. The results appeared in most of
the regressions and had significant values throughout most of the cases.
9. Target LTM return had negative correlation to premium paid, implying that companies whose share price
had increased significantly over the last year received lower premium. The results appeared in most of
the regressions and had significant values throughout most of the cases.
10. Target market capitalization at transaction date – 1day had negative correlation to premium paid, implying
that companies with relative high market value received lower premium. The results did not appear in
all of the regressions and had low significant values throughout most the cases.
6. Conclusions
Hypotheses validation
1. Variance in acquisition premium could be better explained in specific geographies, relative to global data set.
The hypothesis was partially validated by the regression results as the regressions were found more significant and
with more significant variables for Europe and Asia & MENA than for the global data set.
373

The exceptions are transactions in America which came out with poor regression results relative to the global data
set. It may be useful for future researches to focus on the Americas markets and try to further break the data set into
industry segments in order to isolate groups of transactions and reach better results.
The clear conclusion is that while measuring the variance in premium paid in merger and acquisition transactions, it
is important to separate the dataset into geographic regions, rather than to take a global date set. This could be
explained by the different conditions in different geographies that should be taken into account while analyzing the
variance in premium paid for acquisitions such as: political risk, growth rates, risk premium and tax regime (e.g.
more favorable in emerging markets rather than in western economies). In addition there are differences in:
� Profits sharing and ability to take cash out of the country (e.g. corporate companies which acquire Indian firms
often tend to have difficulties to take out dividends outside of India)
� General acquisition preferences for a region (e.g. Asia is considered to be a selected acquisition target for
financial services and insurance companies as the market is considered to be untapped and has a significant
growth opportunities and are hence willing to pay high premium in order to penetrate this market)
Hence, the hypothesis seems to be valid and apply for Europe and Asia & MENA.
It should be noted that the common practice for valuation exercises in professional firms is to compare acquisition
premium on a sector and geographic basis, typically for the abovementioned arguments.
2. Variance in acquisition premium is positively correlated with the following variables: change of control,
revenue growth, EBITDA growth, use of advisors, minority / majority, change of control and deal attitude
� Change of control – we found a positive correlation between change of control and premium paid, confirming
the hypothesis. Additionally, no correlation was found between the percentage sought and premium paid,
indicating that acquirers were willing to pay premium in order to secure control over a target but not to acquirer
more shares (as more shares in a target would not necessarily secure control over a target). Acquirers would not
typically pay a higher premium for more shares at the target. They would only do so if the shares acquired
provide control over the target. The above finding confirms the control premium hypothesis (i.e. premium in
control acquisition transactions should be larger than premium paid in transactions in which control was not
secured).
� Revenue growth – a positive and significant correlation was found between revenue growth and premium paid.
This implies two conclusions: 1. acquirers view revenue growth as a relevant indication for company growth
and value creation (rather than net income, EBITDA or cash flow growth); 2. Acquirers are willing to pay a
premium for a company which showed historic growth (revenue growth was taken as the LTM growth). Hence
the hypothesis was confirmed.
� EBITDA growth – as opposed to the hypothesis above, the hypothesis in this case was denied as a negative
correlation was found between EBITDA growth and premium paid (however, only in 60% of the cases the
correlation was significant). The above could imply that acquirers do not necessarily see EBITDA growth as a
value accretive exercise. On the contrary, acquirers could view companies that focus on EBITDA growth as
more focused on margin improvement rather than securing control over a market (as implied by revenue
growth). The conclusion is that acquirers seems willing to compensate companies for revenue growth and see
such a growth as a value adding phenomenon, yet do not see much of a value added by EBITDA growth.
� Use of advisors – former researches (e.g. Porrini, "Are investment bankers good for acquisition premiums”)
implied that the use of advisors is positively correlated with acquisition premium. We didn’t find a significant
positive correlation between the two variables. The explanation for the difference in results could be driven by
asymmetric information theory: over the last 10 years, with the wider penetration of internet, companies have
much more information on merger & acquisition transactions at their disposal and hence do not need advisors to
better understand acquisition premium paid in similar transaction. In the past, such information was exclusively
at the disposal of advisors and hence a company that entered acquisition negotiations without using advisors did
not have access to such information. As our research is based on more recent transaction database, it is likely
that the access to information factor is already embedded in the premium taken into consideration in the sample
used.
� Deal attitude – in practical discussions, it is often argued that a positive transaction attitude should provide a
higher premium in an acquisition. The logical explanation to the above is that a cooperative management and
374

oard, which are supportive of the transaction, would provide the acquirer better access to information would
work with the acquirer in cooperation in order to achieve the acquisition goals and would hence allow the
acquirer to realize higher synergies from the acquisition. This enables the acquirer to offer a higher premium on
the target. The research confirms the above hypothesis, and the positive correlation assumption.
3. Variance in acquisition premium is negatively correlated with the following variables: target LTM return,
implied equity value to book value, enterprise value to last twelve months (“LTM”) revenues, enterprise value
to LTM EBITDA and market capitalization to net income. One would expect to find a negative correlation of
each of the abovementioned variables with the acquisition premium in a significant amount of regressions run.
� Target LTM return – the hypothesis of negative correlation between target LTM return and premium paid is
confirmed by the regression. The above implies that acquirers are not willing to pay a significant premium for
highly valuated companies. Highly valuated companies are those whose share price has significantly increased
over the last 12 months. The negative correlation was found in most of the regressions (43 / 48) and at a
significance level of 90%. Hence, the conclusion is that acquirers tend to pay a lower premium for companies
whose share price increased over the last 12 months
� Implied equity value to book value – even though the hypothesis above, of a negative correlation between
implied equity value to book value and premium, is confirmed by the regression, the correlation appeared in
less than 40% of the cases and at low significance levels, implying that there is no strong correlation. The above
indicates that acquirers do not view book value to equity value as a proxy for valuation of a target and hence do
not decide on premium size based on that variable
� Enterprise value / revenue – the hypothesis of a negative correlation between enterprise value / revenue and
premium paid is denied for this case. The conclusion of the above is that acquirers mostly do not value targets
based on enterprise value / LTM revenue, otherwise, I would have found a negative correlation which would
imply the low tendency of acquirers to pay a high premium over the shares of a highly valued target
� Enterprise value / LTM EBITDA – the hypothesis of a negative correlation between enterprise value / LTM
EBITDA and premium paid is confirmed for this case. The implied conclusion is that acquirers value targets
based on enterprise value / LTM EBITDA and would not pay a high premium over a target which is highly
valued (i.e. has a high enterprise value / EBITDA ratio). The above conclusion is enhanced by the fact that a
negative correlation between Enterprise value / LTM revenue and premium paid was not found, implying that
enterprise value / LTM EBITDA is a proxy acquirer’s use for valuation purposes
� Market capitalization to net income – the regressions run did not show any correlation between market
capitalization to net income and premium paid. Hence the hypothesis is denied. The conclusion driven is that
acquirers do not view the mentioned variable as a proxy for valuation (hence net income is not seen as a reliable
representative of a company’s business outcome)
4. Variance in acquisition premium is better explained when using 1-month premium, relative to 1-week / 1-day.
The hypothesis above could be verified by consistently finding better regression results for 1-month premium,
relative to 1-week and 1-day
Even though there was no absolute indication of best correlation results for regressions run with 1-day, 1-week or 1month,
in most of the cases, results run on 1-month regressions were more accurate and significant, implying that
the hypothesis was correct.
The abovementioned result is a key issue in merger & acquisition transactions. While a company contemplates an
acquisition of a public company, it usually refers its acquisition proposal to a certain share price of the target.
However, during the negotiations process, leakages of information may occur and would typically tend to push the
share price of the target up, as a potential transaction would normally be executed at a premium.
Hence, a premium is always referred to a share price which is unaffected by any leakages of information and the
longer the time between the share price reference point and the date of the acquisition offer announcement, the less
affected the share price will be.
375

However, taking a share price which is too far away from the announcement would be inaccurate as the share price
will not reflect other corporate activities that happened after the date taken into consideration and could affect the
share price also.
In practice, premium does not tend to be measured at more than 1 month prior to the acquisition date and the typical
timing of measurements is: 1-month, 1-week and 1-day prior to the transaction announcement date. Interestingly in
all regressions results using 1-month premium were more significant and typically had a larger number of significant
variables relative to 1-week and 1-day premium.
5. Other significant findings
� LTM EBITDA margin% – positive correlation of at 5% significance level between LTM EBITDA margin%
and premium paid. The might imply that acquirers view LTM EBITDA margin as a significant indication of a
target’s value. The larger the EBITDA margin is, the higher the premium an acquirer is willing to pay
� Enterprise value and market capitalization: a significant negative correlation was found between the absolute
value of both and premium paid. Hence the larger a size of a company (as reflected in its enterprise value and
market capitalization) the lower % premium an acquirer is willing to pay. The above could be explained by the
fact that the larger a company, the more resources an acquirer will have to raise. Raising significant resources
could negatively affect the acquirer’s ability to offer a premium.
7. References
Bargeron Leonce L., Frederik P. Schlingemann, Rene M. Stulz, Chad J. Zutter, (2008), Why do private acquirers
pay so little compared to public acquirers?, Journal of Financial Economics, vol. 89-2008, pp. 375–390
Jose-Miguel Gaspara, Massimo Massab, Pedro Matosb, (2005), Shareholder investment horizons and the market for
corporate control, Journal of Financial Economics, vol. 76 (2005) pp. 135–165
Moeller, B. Sara, Schlingmann, P. Frederik, Stulz, M. Rene, (2004), Firm size and the Gains from acquisitions,
Journal of Financial Economics, vol. 73 (2004), pp. 201-228.
Moon H. Song and Ralph A., (2000), Abnormal returns to rivals of acquisition targets: A test of the acquisition
probability hypothesis, Journal of Financial Economics, vol. 55, issue 2, pp. 143-171
Porrini Patrizia, (2006), Are investment bankers good for acquisition premiums, Journal of Business Research, vol.
59 (2006), issue 1, pp. 90-99
376

THE SPECIALITIES OF THE SMALL- AND MEDIUM-SIZE ENTERPRISES’ FINANCING AND THE
DETERMINANTS OF THEIR GROWTH IN HUNGARY
Zsuzsanna Széles, PhD,
Institute of Finance and Accounting, Szent István University, Hungary
Email: Szeles.Zsuzsanna@gtk.szie.hu
Zoltán Szabó, PhD
Institute of Marketing, Szent István University, Hungary
Email: Szabó.Zoltán@gtk.szie.hu
Abstract. The most task of the financing politics is the own and foreign resource rate determine. The equity capital is a part of the
company’s capital, which is not to be repaid. The bigger the rate of the company’s equity capital is, the bigger financial independence
it has. A significant rate of the equity capital makes the company more resistant to a crisis. As it is the private property which is
reduced in case of a loss, the company can more easily cope with temporary difficulties if the rate of the equity capital is high. The
primary private resource is the called-up share, which is increased in every year by the retained profit. Self-financing means that the
company’s developments are financed from the company’s own resources.
The SMEs must be facing during their operations the following problems and challenges: functions of the financial financing, low
capitalization, low risk-bearing capacity, chronic underfinancing, liquidity problems, economies of scale and transaction costs, lack of
transparency (lack of clarity), insufficient funds.
Our paper is a brief summary of the theoretical and empirical considerations related to the possible determinants of firm growth with a
special focus on SME firms. After examining the key factors influencing the decision on a company’s capital structure, we present
some facts on the situation of SMEs in Europe and especially Hungary, in order to see what type of governmental programs seem to
be reasonable on economic grounds in fostering the growth of small firms.
Keywords: financing, SME, growth
JEL classification: G2, G3
1. Introduction
The role and importance of the enterprises has increased in the course of the past decades. Before the change of
regime we could witness the expansion of large companies, but after the political turnover in 1989-90 a remarkable
realignment occurred in the size structure of the enterprises. The „reversed pyramid” turned back, and nowadays the
hour-glass effect” is typical. It means that the entrepreneurial sector stands on the wide base of small enterprises,
above this there is a narrower stratum of large companies, while the number of the medium sized enterprises is very
low. In Hungary and in the neighbouring countries, one out of three employees works in a small or medium-sized
enterprise. Approximately 20 million SMEs operate in the EU which contribute to the employment, as well as the
income generation and the export of their country. Far more small enterprises exist after the 2004 enlargement of the
European Union than before. These enterprises play an important role in the preservation of the adaptability and
dynamism of the economy and the intensity of market competition. (Szirmai, 2003)
In our days, large companies cannot be considered as the engine of the economy. Therefore the literature takes
increasing notice of the sector of small and medium-sized enterprises. The importance of this sector lies in the fact
that small enterprises, by virtue of their size, are more flexible, hereby they are able to adapt more quickly to the
changes occurring in the environment, react to the arising threats, and exploit the opportunities residing in them. The
feature which can become an advantage, can be a disadvantage at the same time: in a strong market competition it is
hard to keep up. A remarkable part of the newly established enterprises is not able to strengthen and subsist in the
market in the long run. A lot of enterprises close down in the first few years, and there is a considerable number of
companies that stagnate on a certain level, and are not able to develop. This phenomenon means that the sector of
the medium-sized enterprises is too narrow. There are not enough micro and small enterprises, which are able to
grow in a higher degree and join the medium size category. According to various authors’ opinion, the financial
difficulties are the main barriers to the subsistence, undisturbed operation and growth of SMEs. And the reason
behind this problem is not the lack of capital-owners and creditors.
377

In Hungary in 2008 the number of registered enterprises has exceeded 1.2 million. In previous years, the
business sector structure has changed: in the services sectors we observe growing number of businesses, in
industrial, agricultural and commercial sectors the numbers were decreased. The spatial distribution of firms is the
fact that Budapest and Pest County, has most businesses. The number of firms related to the population is
outstanding in Budapest. In addition, a higher than average proportion of firms is to observe in the Gyor-Moson-
Sopron, Pest counties. In other counties, especially in Borsod-Abaúj-Zemplén, Nógrád, Jasz-Nagykun-Szolnok,
Békés and Szabolcs-Szatmár-Bereg counties, it is lower than the average rate of business. Number of businesses
show that the vast majority, 96% are micro enterprises, with their proportion of the small ones has more than 99%.
The micro and small businesses employ half of the employees, medium firms employ 20 percents of employees.
The micro and small enterprises' share of employment increased slightly, the large companies reduced. In 2007, a
large 37 percent of the total net sales was generated, in the medium enterprises 20, the micro and small enterprises,
43 percent realized. Then the micro and small enterprises in total exports had 22 and 15. The small and mediumsized
structural features of recent years is not, or hardly changed. Their management is characterised with high
labour and low capital-intensity. They have greater share of employment, such as turnover or income production.
2. The specialities of the small- and medium-size enterprises financing
The financing means to get the necessary money for the enterprise operation independent from what the aim of the
capital getting was.
The most task of the financing politics is the own and foreign resource rate determination. It means decision
about resource structure. The equity capital is a part of the company’s capital, which is not to be repaid. It can come
both from external and internal resources: from the registered capital, from the retained profit, from state support,
and it can be increased by issuing new shares, increasing the registered capital, fusing with another company or by
an acquisition. The bigger the rate of the company’s equity capital is, the bigger financial independence it has. A
significant rate of the equity capital makes the company more resistant to a crisis. As it is the private property which
is reduced in case of a loss, the company can more easily cope with temporary difficulties if the rate of the equity
capital is high. The primary private resource is the called-up share (that can be increased with additional external
capital or with other financial contribution), which is increased in every year by the retained profit. Self-financing
means that the company’s developments are financed from the company’s own (private) resources.
The external resource parts are state support (the investment support is part of the own resource – capital reserves),
issue of the new stocks or union with other enterprise.
The financing resources of the SMEs:
1. Internal resource
2. External resource
a/ External debt, non-institutionalized forms of financing
The external debt, financing is a non-institutional involvement with an additional source of funds, which
specializes in financial institutions and non-financial companies we use, but also from people who have
given their principal activity and are not engaged in granting loans.
Groups:
� Ownership or shareholder's loan,
� Family loan and loan form friends,
� Suppliers credit,
� Customer deposit,
� Public warehousing, warehouse receipts.
b / External, capital-related financing options:
� Informal investors,
� Business Angels,
� Risk capital.
c/ External debt, the institutional financing forms:
378

� Bank credit,
� Leasing,
� Factoring.
d/ State involvement in the financing of SMEs
Reasons for government involvement:
� Creation of employment (flexible),
� to stimulate competition,
� keep the economy on the move,
� Moving the production and distribution chains,
� diversification of products and services market (product variety, flexibility in meeting the unique
needs),
� The important tool to achieve social objectives (eg, regions lagging behind).
The direct role of the state assets:
� Tenders announced form Domestic sources,
� Co-financed by the European Union announced proposals,
� Access to credit assistance programs,
� Capital programmes,
� Guarantees, Collateral security.
The SMEs must be facing during their operations the following problems and challenges (Béza et al, 2007):
� functions of the financial financing,
� low capitalization,
� low risk-bearing capacity,
� chronic underfinancing, liquidity problems,
� economies of scale and transaction costs,
� lack of transparency (lack of clarity),
� insufficient funds.
Credits
Micro Small Medium Total
2009 2010 2009 2010 2009 2010 2009 2010
Credits over the year 336 338 288 308 266 749 257 968 201 685 217 696 804 772 763 972
HUF Loans 189 294 196 060 132 328 138 276 103 288 115 138 424 910 449 474
FX Loans 147 044 92 248 134 421 119 692 98 397 102 558 379 862 314 498
Credits within the year 747 888 785 371 841 678 1 064 941 971 522 1 407 979 2 561 089 3 258 290
HUF Loans 579 571 644 864 635 679 832 928 690 941 929 456 1 906 191 2 407 248
FX Loans 168 317 140 507 205 999 232 013 280 581 478 523 654 897 851 042
Total cerdit at the end
of the year
Loan classificaton
1 496 817 1 407 035 1 286 129 1 349 614 1 237 046 1 286 007 4 019 992 4 042 655
Problem-free 1 018 389 859 003 883 789 841 986 1 059 280 957 333 2 961 458 2 658 322
Special Monitoring 286 905 314 065 201 189 263 036 76 777 190 256 564 871 767 357
Problematic
transaction
191 524 233 966 201 150 244 592 100 989 138 418 493 663 616 977
Below average 54 654 67 173 90 109 64 891 22 499 32 699 167 263 164 763
Doubtful 60 055 86 225 43 468 98 876 43 444 42 739 146 968 227 841
Bad 76 815 80 568 67 573 80 825 35 046 62 980 179 433 224 373
Table 1: Credits for SMEs provided by Loan Banks 2010 (Million HUF)
Source: Hungarian Financial Supervisory Authority
379

The credit of the SMEs has been increased by 0.6% over the previous year in 2010. The credit of the micro
enterprises has been decreased by 6 %, the credit of the small enterprises has been increased by 4.9%. Problematic
tranactions are increased by 25%. Table 1 show the details between 2009 and 2010 by classification.
Lending to small and medium-sized enterprises may be supported by government backed lending schemes. As
of the third quarter of 2010 two new elements were added to the state-backed lending scheme: the existing overdraft
programme (Széchenyi card) was supplemented by a working capital loan programme and an investment loan
programme, both with subsidised interest rates and state guarantees (with the participation of Garantiqa Zrt.). While
the subsidised interest rates encourage borrowing on the demand side, the guarantee provided by the state improves
banks‟ willingness to lend. The efficiency of the programme can be measured by the fact that currently more than 20
per cent of the SME loan portfolio is covered by the guarantee of Garantiqa Zrt. and respectively the counterguarantee
of the state (Figure 1). This implies that the decline in lending would be more pronounced without the
guarantee schemes. Although guarantee programmes need considerable public sources, reallocation to such targets
need to be taken into account, because substantial stimulation of economic growth can be achieved through such
programmes.
Figure 1: Loans outstanding to the SME sector backed the guarantee of Garantiqa Zrt.
Source Hungarian National Bank
The factoring takes 10% of the short term loans gained from banks of small and medium scale companies,
showing that factoring is not a marginal-importance financial service, but a meaningful element of the short term
financing. The Garantiqa Hitelgarancia Zrt. (Garantiqa Credit Warranty Inc.) was founded in 1992, incorporating the
Hungarian state, most important domestic commercial banks, different saving banks and some trade unions of
enterprises. Their aim is to support the operation of the domestic small and medium scale enterprises, by providing
them guaranty that means cash bail and lending such organizations that try to be engaged in the employee
shareholder program. In 2007 the Pénzügyi Szervezetek Állami Felügyelete (PSZÁF, Hungarian Financial
Supervisory Authority) authorised for Garantiqa Credit Warranty Inc. the operation which is equal in terms to the
ones regarding the credit institutions.
The Garantiqa Credit Warranty Inc. fulfils all the requirements that are regarded to the credit institutions,
namely the applied routines of management, control, risk management, needed capital and calculations and these
were appreciated by the Hungarian Financial Supervisory Authority.
With this action the circle of activities of Garantiqa Credit Warranty Inc. became broadened to support the
development of the small and medium scale enterprises. It can provide bail to all kinds of source involvement: bank
warranty, leasing and factoring transactions, risk capital involvement, domestic and EU tenders, applications.
380

Above these they support the borrowing and bond issuing activities of local governments and their enterprises
by providing vouching for them.
With guaranteeing it obligates itself to pay instead of the debtors to the financial institution in case of situations
when the debtor does not fulfil the financial duties. The application of the enterprise willing to request the source,
will be sent by the financial institutions (commercial banks, saving banks, leasing and factoring partners) in a
standardised way to Garantiqa (there is no direct connection between the Credit Warranty Inc. and the clients).
Further possibility for is that the source-requiring company can directly turn to Grantiqa in case of need for tender
warranty related to the enterprise. This activity makes it possible for such local governments and enterprises that
have a realistic business plan to reach sources that they would not be able to or they would not be able to get in any
other ways due to their too risky classification.
Basic aim is to develop the domestic small and medium scale enterprises by spreading and making the factoring
financing available in a much broader circle of clients. For the factoring contract there may be a need for assurance.
Those small and medium scale enterprises that would most need this kind of service are the ones that are lack of the
required assurance. Garantiqa can help them by providing the assurance towards the factoring company.
The delayed turnaround in corporate lending can be mainly attributed to credit supply constraints. In our
forecast in November 2010 we anticipated a turnaround in corporate lending in the first quarter of 2011, with
significant downside risks. According to our updated lending forecast, an upturn (i.e. a net increase in the
outstanding amount) in corporate lending may be expected only around the third to fourth quarter of 2011. Given
that the capacity utilisation of firms may reach its historic average in the first quarter of 2011, a further expansion on
firms‟ export markets or a possible upswing in internal demand may lead to an increase in the demand for
investment loans. The demand of the SME sector for working capital loans may also increase. There is a risk that the
aforementioned duality in the corporate segment will remain persistently. Owing to tight credit supply constraints,
only a part of credit demand is expected to be met, which may cause costs to the real economy.
Figure 2: The turnover of the factoring market in Hungary
About all the four years were analysed we can state that the market shrinking was the same in the bank factoring
as well in the non-bank background factoring companies. The turnover of the non-bank factoring between 2005 and
2009 remained below 230 billion HUF, meanwhile the bank factoring turnover in the analysed period has almost
reached 600 billion HUF. Form 2008 to 2009 the turnover of bank factoring companies decreased from 592 billion
HUF to 536.1 Billion HUF, and in case of the non-bank factoring it dropped from 226.3 billion HUF to 202.7 billion
HUF. The data of the analysed companies in the previous years show an increasing tend at the bank and non-bank
factoring companies as well until 2008. The bank factoring companies realised a higher rate of growth than the nonbank
ones.
Nowadays, the economic and financial crisis is the biggest, most complex and most difficult economic
challenge. In the current economic challenges in their business will stay alive, only companies that are capable and
381

innovative ideas. The approach to the economic challenges can be very diverse, these are rising unemployment,
credit constrained firms / households etc.
The majority of challenges for domestic medium-sized companies are in a changing regulatory environment, as
well as problems that they have a minimum of internal resources on how to make operations more efficient.
(González-Páramo, 2006)
The management structure is determined by the history of companies. Many of the „emerging” companies are
coming from smaller companies, family businesses managed segment. They had good ideas and marketable
products. However, in parallel, not every company has developed the share of responsibilities and tasks, the modern
management approach. It is possible that the leader of a medium-sized companies do not have serious economic and
legal knowledge, and no separate department in the organization, which can continues the important legislative
changes and new business opportunities for example in the taxation.
The Hungarian ambitious medium-sized enterprises often would like to save the tax and do not care about risk
reduction options. Many companies put to focus on the direct benefit instead of the uncertainties inherent in their
operations. Many times, bank financing cause problems. Basically, the medium-sized enterprises are risky for banks.
Of course, the enterprise has to strive to the utmost to fulfil the conditions imposed by banks, in terms of its risks,
however no matter how this will make it. For example, the bank has outstanding for a total restructuring of the
structure group wants to carry less risk to the banking point of view. It could mean that the bank would like to be
preferably in the competence of the group's most profitable company, even if originally this company did not have
belonged to the bank guarantee. (Világgazdaság, 2010)
3. Company liquidations and winding up
It is still a difficult situation for the enterprises, but with a slower increase in liquidations. Liquidations in the first
quarter of 2010 slowed the growth rate over the same period the previous year, but the national average was to lie
behind the sharp regional disparities.
The national figures mask considerable regional disparities are: examining the 2010 first quarter in Western
Hungary it has been decreasing for months, the Central region and in Eastern Hungary is still a growing number of
companies went into liquidation. The evolution of liquidations is robustly supported by the split in the economic
sphere: in the western regions of the company it fell considerably less, while the eastern regions - Central Hungary,
along with - plenty of double-digit increase in the extent of creditors by the launch of the new number of
liquidations. Peak increases are to be seen in Northern Ireland, where 43.1 percent higher the number of insolvent
companies was in a year.
The final settlement terms have not seen such a sharp deviation in most regions, fewer firms were closed in the
previous year. The largest decrease in the Northern Plains region showed 32 per cent. Growth only in Northern
Hungary and the dominant weight were recorded in the region of Central India. The Central region has increased the
fastest in the number of closed firms with 26.7 percent in the previous year.
However, in the country none of the regions increased the number of new companies, except in Southern Great
Plain anywhere in the country we observe two digits (20 percent) as a decline in the January-March period, a year in
retrospect. In the Southern Great Plains 9.8 percent fewer new company is incorporated as a year earlier.
The majority of small and medium-sized enterprises are exposed to exchange rate depreciation, but that
exchange rate risk management techniques are almost unknown to them. The main reason for ignoring FX risks is
that FX risk management tools are thought to be expensive, complicated or ineffective. The majority of enterprises
think there are no suitable tools to manage FX risks or they expect external solutions, such as the introduction of the
euro to decrease their risks. (Bodnár, 2009)
382

4. Material and method
The research has been conducted on the basis of the secondary data. The data was available from the Hungarian
National Bank and Hungarian Financial Supervisory Authority databases.
The analytical trend calculation is the most often used way of the trend calculation. The permanent tendency of
the time series can be expressed by certain well-fitting function. In the course of fitting the function, similarly to the
regression calculation, using the least square method we search for the trend best fitting to the values of the time
series. So the analytical trend is the specific function, where the differences of the square amounts between the
values of the same dates in the time series and the function’s own values is the least. (Szűcs, 2004)
�
( y � yˆ )
2
i i
where yi : the i th power of the time series
ŷi : the value of the trend with the i th (i=1,…n)
date
� min.
During the trend calculation I used the following types of function:
� linear trend,
� exponential trend
The trend calculation is very similar to a regression model with two variables, in which the result variable is the
value of the time series, the explanatory variable is the trend variable representing the progress of time. Now I will
mention two of the differences between the regression analysis with two variables and the analytical trend
calculation (Rappai, 2001):
- while in the regression model theoretically the order of observations can be changed at will, the order in the
trend calculation is defined (determined by the time);
- theoretically the value of the explanatory variable of the regression model is free, the value of the trend variable
in the trend function is usually interpreted on the set of the whole numbers, so the difference between the values
following each other is usually 1.
When choosing a type of trend they were important points of view if I am searching for the reason on the rate of
the pace of increase, and that how long is the time of series I have.
4.1. Linear trend
The basic tendency can be expressed by a linear function if the development of the time series is steady and the
rate of the time change is permanent.
The general form of the linear function:
where ŷ : the value of the trend
x : the values of the time changes equidistant from each other
a and b: the unknown parameters of the function
The aim is to estimate the parameters that can be determined with the standard equations. The standard
equations are the equations where the ( y
2
� yˆ
) � min.
function’s primary partial derivates are equalised
with 0. (Szűcs, 2004)
4.2. Exponential trend
� i i
We use it I the relative change of the examined time series and the pace of the change is about permanent. We often
use it in social-economic time series. A row of the economic growth’s index-numbers, the tendency of the
population or in time of inflation almost all current price index-numbers – in a limited interval – show an
exponential increase (or decrease), that is why they can be described with the exponential trend. (Hunyadi & Vita,
2002)
The equation of the exponential function:
383

x
yˆ � ab
Where a : the trend value belonging to the x=0 period of time,
b : the average pace of time change
x=n+1,n+2,…n+k : n indicates the number of examined periods
5. Results
Figure 3 shows total credits of SME sector within the year between 2001 and 2010. We cannot put a trend line to the
credits over the year, the difference was so high between the examined years in this sector.
Figure 3: Total credits within the year of SME sector from financial institutions (Million HUF)
Source Own writing based on the data from Hungarian Financial Supervisory Authority
In case of the HUF and FX loans I fit an exponential trend on the volume of the credits of the SME sector. The
average yearly trend increase is 14.7% at the total loan.
The exponential trend equation of the total credits within the year is:
y = 708.7e 0.147x
R 2 = 0.888
y = 708.8*1.147 x
The average yearly trend between 2001 and 2010 increase is 14.7% at the HUF loans. The trend value is 708.8
if x = 0. (2001)
The exponential trend equation of the HUF loans is:
y = 578.8e 0.129x
R 2 = 0.847
y = 578.8*1.129 x
The average yearly trend between 2001 and 2010 increase is 12.9% at the FX loans. The trend value is 578.8 if
x = 0. (2001).
The striped column means FX (foreign) loans and its linear trend equations:
y = 79.75x + 34,36
R 2 = 0.918
The fittings of the trends are close (R 2 is 0.918). The average yearly FX increase is 79.75 billion HUF.
384

5 000,000
4 500,000
4 000,000
3 500,000
3 000,000
2 500,000
2 000,000
1 500,000
1 000,000
500,000
0,000
y = 394,1x+ 600,52
R 2 = 0,9285
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Total credit Total credit (real data) Lineáris (Total credit)
Figure 4: Total credits of SME sector from financial institutions (Million HUF)
Source Own writing based on the data from Hungarian Financial Supervisory Authority
Figure 4 shows total credits of SME sector between 2001 and 2010. The linear trend is a solid line between
2001 and 2008, the forecast (in a square with broken line) is from 2009 to 2010. The fittings of the trends are close.
The average yearly credit increase is 394.1 Million HUF. We can see big differences between before and after crisis
total credits have been 2900 Million HUF in 2009 and the forecast is 4200 Million HUF. It is a huge go-down.
6. Summary
The micro and small businesses employ half of the employees, medium firms employ 20 percents of employees. The
micro and small enterprises' share of employment increased slightly, the large companies reduced. In 2007, a large
37 percent of the total net sales was generated, in the medium enterprises 20, the micro and small enterprises, 43
percent realized. Then the micro and small enterprises in total exports had 22 and 15. Lending to small and mediumsized
enterprises may be supported by government backed lending schemes. As of the third quarter of 2010 two new
elements were added to the state-backed lending scheme: the existing overdraft programme (Széchenyi card) was
supplemented by a working capital loan programme and an investment loan programme, both with subsidised
interest rates and state guarantees. The demand of the SME sector for working capital loans may also increase.
There is a risk that the aforementioned duality in the corporate segment will remain persistently. Owing to tight
credit supply constraints, only a part of credit demand is expected to be met, which may cause costs to the real
economy.
The average yearly total credit of SME sector increase is 394.1 Million HUF between 2001 and 2008. We can
see big differences between before and after crisis total credits have been 2900 Million HUF in 2009 and the
forecast is 4200 Million HUF. It is huge go-down.
7. References
Szirmai P. (2003): Szemelvenygyőjtemeny a kis- es kozepes vallalkozasok a magyar es nemzetkozi gazdasagban
cimő targyhoz, Kisvallalkozasfejlesztesi Kozpont, Budapest,
Apatini K. (1999): A kis- es kozepvallalkozasok finanszirozasa, KJK Kerszov, Budapest, 1999
Szerb L. (2005): Global Entrepreneurship Monitor 2005 Magyarorszag. A vallalkozoi aktivitast es a vallalkozast
befolyasolo tenyezık alakulasa Magyarorszagon az Europai Unios csatlakozas utan, Pecsi Tudomanyegyetem
Kozgazdasagtudomanyi Kar
Kallay L., Imreh Sz. (2004): A kis- es kozepvallalkozas-fejlesztes gazdasagtana, Aula Kiado, Budapest, 2004
Kallay L. (2002): Paradigmavaltas a kisvallalkozas-fejlesztesben, Kozgazdasagi Szemle, XLIX. evf., 2002. juliusaugusztus,
pp. 557-573.
385

Béza D., Csapó K., Farkas Sz., Filep J., Szerb L. (2007): Kisvállalkozások finanszírozása. Perfekt Kiadó, Budapest.
pp. 27-37.
J. M. González-Páramo (2006): Corporate finance and Monetary policy: the role of small and medium-sized
enterprises. ECB Conference on Corporate Finance and Monetary Policy. 6 p.
Világgazdaság online (2010): KOMOLY KIHÍVÁSOK ÉS MEGOLDÁSOK
http://www.vg.hu/gazdasag/adozas/komoly-kihivasok-es-megoldasok-310044 April 2010
Kohegyi K. (2001): A vallalkozasok finanszirozasa, Cegvezetes, 2001. oktober, pp. 138-149.
Arvai Zs. (2002): A vallalatfinanszirozas uj fejlodesi iranyai, MNB Muhelytanulmanyok (26)
MNB - Hungarian National Bank (2011): Report of financial stability (April 2011). MNB, Budapest. 79. p.
Bodnár K. (2009): Exchange rate ecpossure of Hungarian enterprises. MNB, 2009. Occasional 80. 4 p.
PSZÁF - Hungarian Financial Supervisory Authority (2010): Aggregated data of financial banksector in 2010.
386

387

CORPORATE GOVERNANCE
388

389

CORPORATE GOVERNANCE PRACTICES AND THEIR IMPACT ON FIRM’S CAPITAL
STRUCTURE AND PERFORMANCE; CASE OF PAKISTANI TEXTILE SECTOR
Prof. Dr. Hayat M. Awan. Khuram Shahzad Bukahri & Rameez Mahmood Ansari,
Bahauddin Zakariya University Multan, Pakistan.
Abstract
Manuscript Type: Empirical.
Research Question/Issue: This paper investigates the corporate governance practices currently practiced in Pakistani firms
and the relationship between internal corporate governance structures and capital structure and internal corporate
governance structures and firm performance in Pakistani Textile sector.
Research Findings/Insights: The study used a sample of 100 manufacturing companies listed at Karachi Stock Exchange
(KSE) and Lahore Stock Exchange (LSE) from Textile sector. Regression Analysis and Structural Equation Modeling are
used to determine the relationship between internal corporate governance structures and capital structure and internal
corporate governance structures and firm performance. In finding the relationship of internal corporate governance
structures with capital structures, significant relationship was found with No of Board of Directors, Board Composition,
CEO Duality, Board Skill, Audit Committee Composition, Chairman Independent or Non-executive, CFO attend All Board
Meetings and Separate Management Consultants. In finding the relationship of internal corporate governance structures
with firm performance, significant relationship was found with CEO Duality, Audit Committee Composition, CFO attend
All Board Meetings and Company Chairman in Audit Committee. Findings also suggested that all internal corporate
governance structures taken together impact capital structure negatively and firm performance positively.
Theoretical/Academic Implications: The findings support the theories of corporate governance. Our findings are consistent
with the theoretical models that good corporate governance practices lead to decrease in capital structure and improved
performance. Future research should include both internal and external corporate governance structures and find their
relationship with capital structure and firm performance or find relationship with different proxies of capital structure or
firm performance.
Practitioner/Policy Implication: Our results stress on the importance of corporate governance practices for firms to reduce
their debt structure and improve their performance. And if firms want to get more capital and improve investor’s
confidence, they would have to improve their corporate governance practices.
Keywords: Corporate Governance, Capital Structure, Firm Performance
1. Introduction
Mechanisms that protect the interests of the shareholders are known as Corporate Governance mechanisms.
Good corporate governance helps in economic development. Last two decades have seen an increasing
intensity of research on the subject of corporate governance. Firms having weaker governance structures
face more agency problems and managers of those firm’s get more private benefits, due to weak
governance structures (Core et al., 1999). According to Chuanrommanee and Swierczek (2007), corporate
governance practices in financial corporations of the ASEAN countries are consistent with the international
practices. Corporate Governance has become one the important research area in Pakistan after publication
of SECP Corporate Governance Code 2002, for publicly listed companies. The code was met with a lot of
criticism in the start and there was lot of difficulties in implementing and enforcing it. However, despite
these criticisms, the Code has been the reason for the start of a new era of corporate governance in
Pakistan. Rais and Saeed (2005) argued that the acceptance of Corporate Governance Code has improved
overall structure of the corporation and environment of the businesses by ensuring transparency and
accountability in reporting framework.
Corporate failures such as Enron, WorldCom, One-Tel, Ansett, Parmalat etc have awaken the need to
strengthen corporate governance practices not only in the developed world but also in the developing
world. Most of the corporate governance research has been done on the developed economies (Rajagopalan
and Zhang, 2008). And limited researches are there to address the issues related to corporate governance
that can be applicable to the developing economies. Pakistani corporations are family-controlled from the
past. Ownership is very concentrated. The families control the firms through pyramidal and tunneling
ownership structures. Ghani and Ashraf (2005) argued that families manage business groups in Pakistan,
which contains combinations of different business entities. They also suggested that firms having external
shareholders affiliated with these business groups have weaker corporate governance mechanisms and are
390

less transparent. And due to this, market value of those firms is discounted without even the concern that
they are producing greater profits. Institutional framework has to be strengthened in order to improve the
corporate governance mechanisms of a country.
The main focus of this paper is to find the corporate governance practices currently practiced in
Pakistani firms. And to find the relationship between corporate governance and capital structure and
corporate governance and firm performance in Pakistani textile sector by using KSE and LSE listed firms.
Several studies have tested the hypothesis of finding relationship between characteristics of corporate
governance and capital structure and between characteristics of corporate governance and performance.
However, very few studies have in conducted in context of Pakistan or Pakistani Listed Firms and are
limited in finding the relationship with few characteristics and structures of Corporate Governance.
This paper is organized as follows: the next section provides literature review and development of
hypothesis. The fourth section describes the methodology used. The penultimate section discusses the
results. Finally, the last section concludes the results and concludes the discussion.
2. Literature Review and Hypothesis Development
2.1. Corporate Governance and Capital Structure
According to the modern theories, agency cost is one of the main components of capital structure and
corporate governance mechanisms reduce agency problems, therefore both are linked. Claessens et al.
(2002) argues that good corporate governance mechanisms help firms through better access to financing
and lower cost of capital.
The boards of directors are responsible for managing the overall firm and firm operations. They play a
vital role in deciding about the financial mix. A significant relationship is found between capital structure
and board size by Pfeffer and Salancick (1978). Berger et al. (1997) found that large board of directors
have low leverage levels and larger boards also exert pressure to enhance firm performance. On the other
side, Jensen (1986) states that high leverage levels have large boards. Wen et al. (2002) found positive
relationship between capital structure and board size. And large board size is associated with higher debt
levels. Abor (2007) found a positive correlation between board size and capital structure.
Non executive directors are essential part of modern corporate governance mechanisms. Pfeffer and
Salancick (1978) states that presence of non executive directors reduces uncertainties about the company
and help in raising capital. And presence of higher numbers of non executive directors lead to higher
leverage levels. Jensen (1986) and Berger et al. (1997) also found the same, those companies having high
leverage levels have relatively more external directors. Abor (2007) concluded that there is a positive
correlation between board composition and capital structure. On the other side, Wen et al. (2002)
concluded that a significant negative relationship between leverage levels and number of external directors.
And that the presence of external directors leads to low leverage levels.
CEO/Chair duality is also one of the important feature of corporate governance. This can directly effect
the capital structure decisions of the company. Fama and Jensen (1983) argue that role of CEO and
chairman should be separate, as CEO is the chief decision management authority and chairman is the chief
decision control authority. Fosberg (2004) found that separate CEO and chairman have higher leverage
levels and results in optimal amount of debt. Abor (2007) concluded that there is a positive correlation
between CEO duality and capital structure.
2.2. Corporate Governance and Firm Performance
Black et al. (2006) concludes that firm’s having higher governance scores have a high market value. Chen
(2008) suggested that establishing effective governance mechanisms leads to improvement in firm’s value.
Harford et al. (2008) concluded that poorly governed firm’s destroy firm’s value.
391

Mintzberg (1983) and Kosnik (1990) argue that large board size negatively influences performance of
firm’s. Lipton and Lorsch (1992) and Jensen (1993) argue that larger board size turns less effective because
of poorer communication and decision making process. Van den Berghe and Levrau (2004) argue that
increasing the number of directors increase expertise, knowledge and skills than smaller boards. A high
negative relationship between performance of firm and board size was found by De Andres, Azofra and
Lopez (2005). Analysis by Dalton and Dalton (2005) found that superior performance resulted from larger
boards while Bhagat and Black (1999) and Hermalin and Wiesbach (2003) proposed an opposite view.
Brown and Caylor (2006) showed firms with board size in between 6 and 15 yield higher returns on equity.
Jackling and Johl (2009) found that large board size impacts performance positively.
Baysinger and Butler (1985) found that board with more outside directors performed better than other
firms. Fosberg (1989) found no relationship between proportion of outside directors and firm’s
performance. Rosenstein and Wyatt (1990) found that there is slight increase in stock prices when more
outside directors are appointed by firms. Hermalin and Weisbach (1991) found no relationship between
board composition and firm value. Yermack (1996) showed that proportion of non-executive directors does
not affect performance of firm significantly. Agrawal and Knoeber (1996) while conducting a study on US
firms found negative relationship between proportion of outside directors and performance of firms.
Shrader et al. (1997) found negative relationship between proportion of women on board and firm’s value.
Bhagat and Black (1999) found no significant between board independence and firm’s performance in a
long run in case of US firms. Roberts et al. (2005) suggests that the active participation of an independent
director brings to team, independent ideas. And can be helpful in better functioning the organization and
board. In support of the above assumption, Brown and Caylor (2006) found that firms with more
independent directors performed well than others with higher ROE, greater profits, more dividends and
higher repurchase of stock. And the most important factor in the above study affecting firm’s performance
was independence of directors. Chan and Li (2008) found that firm value in enhanced by presence of
expert-independent directors in board. Jackling and Johl (2009) found that large number of independent or
outside directors on board impacts performance positively.
Rechner and Dalton (1991) found that firms with CEO duality performed better. Daily and Dalton
(1992), while conducting a study on entrepreneurial firms, found no relationship between CEO duality and
performance of firm’s. Peel and O’Donnell (1995) showed that splitting both roles lead to improved
performance. Yermack (1996) argued that those firms are more valuable which have separate CEO and
chairman position. Brickley et al. (1997) concluded that CEO duality doesn’t leads to inferior performance.
Sanda et al. (2003) also found a positive relationship between separate CEO and chairman positions and
firm’s performance. Brown and Caylor (2006) also concluded the same that when CEO and chairman
positions are separate, firms are more valuable. Kang and Zardkoohi (2005) concluded regarding studies of
the relationship between CEO/Chair duality and other measures that the results are complex. They
proposed that if such duality exists as a reward, it might result in positive performance. But if the reason is
to increase the CEO’s power than it may have a negative affect on the performance of the firm. Elsayed
(2007) concluded that CEO duality has no effect on performance. But it varies from industry to industry.
Uzun et al. (2004) found that higher audit committee independence resulted in a lower chance of fraud.
So, audit committee reduces agency cost and improves over all performance. Brown and Caylor (2006)
found positive relationship between dividend yield and independent audit committees. But found no
relationship between independent audit committee and performance. Chan and Li (2008) found that firm
value in enhanced by presence of expert-independent directors in audit committee.
Lybaert (1998) stated that better performance of firms is due to higher and better level of education
among entrepreneurs. On the other side, Powel (1991) stated that occupational and professional affiliations
of qualified managers with firms may have a negative effect. Lawrie (1998) stated that the gaps in
management expertise are considered less as an barrier in development of SME.
What happens at meetings of board and how many directors attend those meetings tells shareholders,
how seriously the governance responsibilities are met. According to Lipton and Lorsch (1992), more
meetings of board results in improved performance. A positive association exists between frequency of
392

meetings and performance of firm and also in between the attendance of directors and performance of firm
(Brown and Caylor, 2006).
3. Methodology
The study is a quantitative approach, basically an exploratory type of study. Textile sector is used as the
main population of study. KSE and LSE listed firms is used in the sample, selected on the basis of response
of firms on Corporate Governance structures and availability of secondary data on the other required
variables. Non-probability sampling is used to select firms. A sample of 150 firms was taken in the start,
but final sample include 100 firms, which were having complete data of all internal corporate governance
structures and financial data of period from 2005 to 2009. Data on required variables is collected through
primary and secondary sources. Data on internal Corporate Governance structures are collected through
self administrated survey, mail survey, phone survey, interviews and annual reports. Data related to
financial part of the study is collected from Annual Reports. Regression analysis and structural equation
modeling is used.
3.1. Capital Structure Measure
Debt ratio is used as the proxy of capital structure. It is measured by dividing total debt by total assets.
Ratio of debt to assets was used in the past by Short and Keasey (1999) and Holderness et al. (1999) in
their studies.
3.2. Firm Performance Measure
Return on Assets is used as proxy for performance. It is measured by dividing net earnings by total assets.
Muth and Donaldson (1998) and Erhardt et al. (2003) have used ROA in measuring firm performance in
their studies.
3.3. Corporate Governance Measures
No. Of Board Meetings, Board Composition, Board Size, CFO Attend All Board Meetings, Directors In
Audit Committee, CEO Duality, Board Skill, Chairman Of Audit Committee is Non-executive or not, No
Of Audit Committee Meetings In A Year, Is Company Chairman In Audit Committee, No Of Audit
Committee Members, Is Chairman Of Company Independent or Non-executive, Audit Committee
Composition and Are Separate Management Consultants hired or not are used as the measures of Corporate
Governance.
Table 1 – Measures of Corporate Governance
1. Number of Board Meetings is measured as total number of board meetings.
2. Board Composition is measured by dividing number of independent or non-executive directors by total number of board of
directors.
3. Board Size is measured as total number of board of directors.
4. CFO attend All Board Meetings is 1 if CFO attends all board meetings and 0 if otherwise.
5. Directors in Audit Committee are measured by dividing number of directors in audit committee by total number of audit
committee members.
6. CEO Duality is 1 if CEO is chairman as well and 0 if otherwise.
7. Board Skill is measured by dividing number of directors having professional degree or qualification by total number of board of
directors.
8. Chairman of Audit Committee is Non-executive is 1 if chairman of audit committee is non-executive and 0 if otherwise.
9. No of Audit Committee Meetings in a Year is measured as total number of audit committee meetings in a year.
10. Company Chairman in Audit Committee is 1 if company chairman is in audit committee and 0 if otherwise.
11. No of Audit Committee Members is measured as total number of audit committee members.
12. Chairman of Company Independent or Non-executive is 1 if chairman of company is independent or non-executive and 0 if
otherwise.
13. Audit Committee Composition is measured by dividing number of independent or non-executive members by total number of
audit committee members.
14. Separate Management Consultants is 1 if separate management consultants are hired by company and 0 if otherwise.
393

3.4. Control Variables
Age of firm and size of firm is used as control variables. Age of Firm is measured as logarithm of number
of years between the observation year and the year of incorporation. Age of firm has been used as a control
variable in various studies like Li and Simerly (1998), Dimelis and Louri (2002) and Gedajlovic et al.
(2005). To measure age of firm, number of years from the time of listing was used by Drobetz et al. (2004).
Size of firm is measured as logarithm of book value of assets. Logarithm of book value of assets is populist
in most of the researches, like in the studies of Renneboog (2000), Bauer et al. (2004) and Drobetz et al.
(2004).
4. Results
4.1. Corporate Governance Practices
81% of organizations were having 7 directors serving on their board, 97% of organizations were having
independent or non-executive directors, 25% of organizations were having 4 independent or non-executive
directors, 93% of organizations were not having representatives of minority shareholders, 5% of
organizations were having 1 representative of minority shareholders, 70% of organizations were not having
independent or non-executive chairman, 55% of organizations were not having the same individual,
holding the posts of Chairman and Chief Executive. In all organizations functions of Chairman were clearly
defined by Board of Directors, in all organizations functions of CEO were clearly defined by Board of
Directors, all organizations were having Statement on Ethics and Ethical Business Practices, all
organizations were having terms of appointment and remuneration package of CEO and the Executive-
Directors approved by the BOD. 46% of organizations were having 4 meetings of the BOD in the past 12
months, 38% of organizations were having 5 members of board holding professional degree’s, 50% of
organizations CFO did not attend the board meetings, all of them were having degrees in accounting,
economics and business administration. But most of them were CA, ACA and FCA. Company Secretary of
all organizations attend all board meetings, all were having degrees in accounting, economics and business
administration. But most of them were Masters, MBA and CA. In all organizations, external auditors or
his/her spouse or children were not holding shares in company. All organizations were having an audit
committee, 91% of organizations were having 3 members in their audit committee, 95% of organizations
were having 3 directors in their audit committee, 73% of organizations were not having company chairman
as members in their audit committee, 57% of organizations were having 2 non-executive members in audit
committee, 61% of organizations were having the non-executives chairman of their audit committee, 74%
of organizations were having 4 meetings of Audit Committee in past 12 months, all organizations CFO
attend meetings of the Audit Committee. All organizations were having an internal auditing department,
head of internal auditing department of all organizations were having direct access to Chairman Audit
Committee, all organizations were having external auditors, appointed as per prescribed guidelines, 57% of
organizations were having separate Management Consultants, hired to advice.
4.2. Corporate Governance and Capital Structure
Model
Table 3 – Corporate Governance impacting Debt Ratio
Unstandardized Coefficients
B Std. Error Beta
Standardized
Coefficients t Sig.
(Constant) .641 .320 2.001 .046
Size Of Firm -.073 .018 -.205 -4.179 .000
Board Size .070 .015 .226 4.615 .000
Board Composition .341 .059 .359 5.752 .000
CEO Duality .047 .024 .120 1.977 .049
Board Skill .182 .067 .144 2.705 .007
394

Audit Committee Composition -.230 .064 -.284 -3.597 .000
Chairman Independent or Non-executive -.059 .029 -.139 -2.027 .043
CFO Attend All Board Meetings -.032 .019 -.083 -1.672 .095
Separate Management Consultants -.056 .020 -.144 -2.853 .005
Dependent Variable: Debt Ratio
R Square= 0.243 S.E= 17.2 F=7.486 Sig.= 0.000
In finding the relationship of Corporate Governance structures with Debt ratio, which is the dependent
variable and used as the proxy of Capital structure, while Size of firm and Age of firm are used as control
variables, the value of R square indicate that 24.3 percent variation is explained by the independent
variables and the remaining 75.7 percent is due to some other factors. ANOVA table indicate that overall
regression is significant. Results show that Board Size, Board Composition, CEO Duality, Board Skill,
Directors In Audit Committee, Company Chairman In Audit Committee, Chairman Of Audit Committee
Non-executive and No Of Audit Committee Meetings In A Year are positively related to capital structure.
And Audit Committee Composition, Chairman Independent or Non-executive, No Of Board Meetings,
CFO Attend All Board Meetings, No Of Audit Committee Members and Separate Management
Consultants are negatively related to capital structure. But significant relationship was found with Board
Size, Board Composition, CEO Duality, Board Skill, Audit Committee Composition, Chairman
Independent or Non-executive, CFO Attend All Board Meetings and Separate Management Consultants.
And non-significant relationship was found with No of Board Meetings, No of Audit Committee Members,
Directors in Audit Committee, Company Chairman in Audit Committee, Chairman of Audit Committee
Non-executive and No of Audit Committee Meetings in a Year.
4.3. Corporate Governance and Return on Assets
Model
Unstandardized Coefficients
B Std. Error Beta
Standardized
Coefficients t Sig.
(Constant) .041 .141 .293 .770
Debt Ratio -.181 .023 -.409 -7.975 .000
CEO Duality -.019 .010 -.111 -1.828 .068
Audit Committee Composition -.051 .029 -.141 -1.777 .076
CFO Attend All Board Meetings -.041 .008 -.237 -4.779 .000
Company Chairman In Audit Committee .026 .010 .134 2.625 .009
Dependent Variable: Return On Assets
R Square= 0.259 S.E= 0.07543 F= 7.638 Sig.= 0.000
Table 4 – Corporate Governance impacting Return on Assets in presence of Debt Ratio
In finding the relationship of Corporate Governance structures with Return on Assets, which is the
dependent variable and used as the proxy of Firm Performance, while Debt ratio, Size of firm and Age of
firm used as control variables, the value of R square indicate that 25.9 percent variation in is explained by
the independent variables and the remaining 74.1 percent is due to some other factors. ANOVA table
indicate that overall regression is significant. Results show that Board Size, Board Composition, Board
Skill, Company Chairman In Audit Committee and Separate Management Consultants are positively
related to capital structure. And CEO Duality, Audit Committee Composition, Chairman Independent or
Non-executive, CFO Attend All Board Meetings, No Of Audit Committee Members, Directors In Audit
Committee and Chairman Of Audit Committee Non-executive are negatively related to capital structure.
No relationship was found with No Of Board Meetings and No Of Audit Committee Meetings In A Year.
Significant relationship was found with Debt ratio, CEO Duality, Audit Committee Composition, CFO
Attend All Board Meetings and Company Chairman in Audit Committee. Non-significant relationship was
found with Size of Firm, Age of Firm, Board Size, Board Composition, Board Skill, Chairman Independent
or Non-executive, No of Board Meetings, No of Audit Committee Members, Directors in Audit
395

Committee, Chairman of Audit Committee Non-executive, No of Audit Committee Meetings in a Year and
Separate Management Consultants.
4.4. Corporate Governance Impacting Debt Ratio and ROE
Figure 1 - Corporate Governance Impacting Debt Ratio and ROA
Chi-Square = 741.47 df=101 p-value = 0.00000 RMSEA = 0.13
The data fits the model well and model is significant. Corporate Governance structures taken together
significantly impact Capital Structure negatively. Corporate Governance structures taken together impact
Return on Assets positively and Debt ratio impacts the Return on Assets negatively. But impact of Debt
ratio is greater in comparison to Corporate Governance measures. And impact of Corporate Governance
becomes slightly greater in one-to-one relationship, as compared to the results in presence of Debt ratio.
5. Conclusion and Discussion
Corporate Governance has become one the important research area in Pakistan after publication of SECP
Corporate Governance Code 2002, for publicly listed companies. Corporate failures such as Enron,
WorldCom, One-Tel, Ansett, Parmalat etc have awaken the need to strengthen corporate governance
practices not only in the developed world but also in the developing world. Pakistani corporations are
family-controlled from the past. Ownership is very concentrated. The families control the firms through
pyramidal and tunneling ownership structures. The relationship of variables of corporate governance has
been widely researched but very little variables have been taken. In this research a large number of
corporate governance Structures are taken and their relationship is found with capital structure and firm
performance.
In finding the relationship of Corporate Governance structures with Debt ratio, while Size of Firm and
Age of Firm used as control variables, Negative relationship was present in Size of Firm and Capital
Structure, the reason could be, as the size of the firm increase, the size of the assets increase and the
creditworthiness of the organization increase and cost of raising debt decrease and overall cost of debt also
decreases and it is particularly true about the textile sector because of nature of ownership, which is
predominantly closely held with the family or kinship within the family. Positive relationship was present
in No of Board of Directors and Capital Structure as found by Wen et al. (2002) and Abor (2007), the
reason could be, as the no of board members increase, the capability to raise debt increase due to joint
investment portfolio of the directors and greater number of board of directors put pressure on managers,
through strict monitoring to use more debt in order to increase the firm’s value. Positive relationship was
present in Board Composition and Capital Structure as found by Jensen (1986), Berger et al. (1997) and
Abor (2007), the reason could be, as the number of independent director’s increase, the creditworthiness of
the organization increase and decision making of management becomes better and more board members
could have those useful links on financing facilities, so more debt is used in comparison to equity. Positive
relationship was present in CEO Duality and Capital Structure as found by Abor (2007), the reason could
396

e as the control increase, there would be active involvement and no conflict between the CEO and
chairman and better allocation of resources and effective strategy about use of debt. Positive relationship
was present in Board Skill and Capital Structure, the reason could be, as there would more educated
directors, the board will be more effective and decision making of management becomes better and they
can raise more at low cost debt as compared to equity by having knowledgeable and useful links. Negative
relationship was present in Audit Committee Composition and Capital Structure, Chairman Independent or
Non-executive and Capital Structure, CFO attend All Board Meetings and Capital Structure and Separate
Management Consultants being hired and Capital Structure, the reason could be, as there will be more
independent or non-executives, CFO attending all board meetings and separate management consultants
being hired, there would be strict monitoring and regulatory mechanisms due to negative orientation
towards more debt in comparison to assets and positive orientation towards stable cash flows and they
would influence the board to increase the assets as well in comparison to debt.
In finding the relationship of Corporate Governance structures with Return on assets, while Debt ratio,
Size of Firm and Age of Firm used as control variables, Negative relationship was present in Debt ratio and
Return on Assets , the result is consistent with many studies done in the developing countries and the
reason could be that costs like interest payments, bankruptcy costs and agency costs are greater than tax
benefits of using more debt and can also be attributed increase in rate of interest in last few years. Negative
relationship was present in CEO Duality and Return on Assets as found by as found by Sanda et al. (2003),
the reason could be as CEO is chairman as well, he/she is not able manage both duties actively at the same
time and due to burden of duties, performance is affected badly. Negative relationship was present in Audit
Committee Composition and Return on Assets and CFO attend All Board Meetings and Return on Assets,
the reason could be, as there would be more independent directors or CFO attending all meetings, they will
influence the board to invest less in the risky projects and would prefer more stable projects and would
result in lower returns. Positive relationship was present in Company Chairman in Audit Committee and
Return on Assets, the reason could be, as the company chairman will be there in the audit committee, he
will serve as monitoring mechanism to the decisions of the board and he will keep close eye on the
financials of the company and would deal with discrepancies in time.
Corporate Governance structures taken together significantly impact Capital structure negatively. If
Corporate Governance becomes strong, Debt ratio will decrease, which is true as there will be more
Corporate Governance structures, there would be strict monitoring and regulatory mechanisms and to
protect the interests of the shareholders, managers will pursue the policy of lower debt levels, in order to
mitigate extra risks related to with higher debt levels. Corporate Governance measures taken together
impacts Return on Assets positively and Debt ratio impacts the Return on Assets negatively. If Corporate
Governance becomes strong, Return on Assets will increase, which is true as there will strong corporate
governance structures, there will be strict monitoring and regulatory mechanisms, effective accounting
standards and better control systems, that will result in better utilization of firm’s resources and improved
Return on Assets.
Overall, there is relationship between Corporate Governance structures and Capital Structure and
Corporate Governance structures and Firm Performance. In some cases, it is positive and in other cases it is
negative. In some cases, it is highly significant and in other cases it is slightly significant. If all Corporate
Governance structures are taken together, it impacts Capital Structure negatively and Return on Assets
positively.
The above study has outlined the internal corporate governance structures and their relationship with
capital structure and firm performance. It adds to the literature and opens new avenues for future studies, in
which both internal as well as external corporate governance structures can be taken and than their
relationship can be found with different proxies of capital structure and firm performance. And due to lack
of time and large number of firms, only one sector was taken, future studies can incorporate different
sectors and can also find differences in those sectors related to corporate governance structures and their
relationship with capital structure and firm performance.
397

6. References:
Abor, J. 2007. Corporate governance and financing decisions of Ghanaian listed firms. Corporate
Governance, 7: 83-92.
Agrawal, A. & Knoeber, C. R. 1996. Firm performance and mechanisms to control agency problems
between managers and shareholders. Journal of Financial and Quantitative Analysis, 31:377-397.
Bauer, R., Gunster, N. & Otten, R. 2004. Empirical evidence on corporate governance in Europe: The
effect on stock returns, firm value and performance. The Journal of Asset Management, 5:91-104.
Baysinger, B. D. & Butler, H. N. 1985. Corporate governance and board of directors performance effects of
changes in board composition. Journal of Law Economics and Organization, 1:101-124.
Berger, P. G., Ofek, E. & Yermack, D. L. 1997. Managerial Entrenchment and Capital Structure Decisions.
Journal of Finance, 52:1411-1438.
Bhagat, S. & Black, B. 1999. The uncertain relationship between board composition and firm performance.
The Business Lawyer, 3:921-963.
Black, B. S., Hasung, J. & Woochan, K. 2006. Does Corporate Governance Predict Firms’ Market Values?
Evidence from Korea. Journal of Law, Economics and Organization, 22:366-413.
Brickley, J. A., Coles, J. L. & Jarrell, G. 1997. Leadership Structure: Separating the CEO and Chairman of
the Board. Journal of Corporate Finance, 3:189-220.
Brown, L. D. & Caylor, M. L. 2006. Corporate Governance and Firm Valuation. Journal of Accounting and
Public Policy, 25:409-434.
Chan, K. C. & Li, J. 2008. Audit Committee and Firm Value: Evidence on Outside Top Executives as
Expert-Independent Directors. Corporate Governance, 16:16-31.
Cheema, A., Bari, F. & Saddique, O. 2003. Corporate Governance in Pakistan:Ownership, Control and the
Law. Lahore University of Management Sciences, Lahore.
Chen, Y. 2008. Corporate governance and cash holdings: listed new economy versus old economy firms.
Corporate Governance, 16:430-442.
Chuanrommanee, W. & Swierczek, F. W. 2007. Corporate Governance in ASEAN Financial Corporations:
reality or illusion? Corporate Governance, 15:272-283.
Core, J. E., Holthausen, R. W. & Larcker, D. F. 1999. Corporate governance, chief executive officer
compensation, and firm performance. Journal of Financial Economics, 51:371-406.
Claessens, S. & Fan, J. P. H. 2002. Corporate Governance in Asia: A Survey. International Review of
Finance, 3:71-103.
Daily, C. M. & Dalton, D. R. 1992. The Relationship Between Governance Structure and Corporate
Performance in Entrepreneurial Firms. Journal of Business Venturing, 7:375-386.
Dalton, C. M. & Dalton, D. R. 2005. Boards of directors: Utilizing empirical evidence in developing
practical prescriptions. British Journal of Management, 16:91-97.
De Andres, P., Azofra, V. & Lopez, F. 2005. Corporate boards in OECD countries: Size, composition,
functioning and effectiveness. Corporate Governance, 13:197-210.
Dimelis, S. & Louri, H. 2002. Foreign Ownership and Production Efficiency: a Quantile Regression
Analysis. Oxford Economic Papers, 54:449-469.
Drobetz, W., Schillhofer, A. & Zimmermann, H. 2004. Corporate governance and expected stock returns:
evidence from Germany. European Financial Management, 10:267-293.
Elsayed, K. 2007. Does CEO duality really affect corporate performance? Corporate Governance, 15:1203-
1214.
Erhardt, N. L., Werbel, J. & Sharder, C. B. 2003. Board of director diversity and firm financial
performance. Corporate Governance, 11: 102-111.
Fama, E. F. & Jensen, M. 1983. Separation of Ownership and Control. Journal of Law and Economics,
26:301-325.
398

Fosberg, R. 1989. Outside directors and managerial monitoring. Akron Business and Economic Review,
20:24-32.
Fosberg, R. H. 2004. Agency problems and debt financing: leadership structure effects. Corporate
Governance: International Journal of Business in Society, 4:31-38.
Gedajlovic, E., Yoshikawa, T. & Hashimoto, M. 2005. Ownership structure, investment behavior and firm
performance in Japanese manufacturing industries. Organization Studies, 26:7-35.
Ghani, W. I. & Ashraf, J. 2005. Corporate Governance, Business Group Affiliation and Firm Performance:
Descriptive Evidence from Pakistan. CMER Working Paper No. 05-35.
Hasan, A. & Butt, S. A. 2009. Impact of Ownership Structure and Corporate Governance on Capital
Structure of Pakistani Listed Companies. International Journal of Business and Management, 4:50-57.
Harford, J., Mansi, S. A. & Maxwell, W. F. 2008. Corporate governance and cash holdings. Journal of
Financial Economics, 87:535-555.
Hermalin, B. E. & Weisbach, M. S. 1991. The effects of board composition and direct incentives on firm
performance. Financial Management, 20:101-112.
Hermalin, B. E. & Weisbach, M. S. 2003. Boards of directors as an endogenously determined institution: A
survey of the economic literature. Economic Policy Review-Federal Reserve Bank of New York, 9:7-
26.
Holderness, C., Kroszner, R. & Sheehan, D. 1999. Were the Good Old Days That Good? The Evolution of
Managerial Stock Ownership Since the Great Depression. Journal of Finance, 54:435-469.
ICAP MIES-5. Available at www.icap.org.pk/mies/mies5.pdf (Accessed on 10/08/2010)
Jackling, B. & Johl, S. 2009. Board Structure and Firm Performance: Evidence from India’s Top
Companies. Corporate Governance: An International Review, 17:492-509.
Javid, A. Y. & Iqbal, R. 2010. Corporate Governance in Pakistan: Corporate Valuation, Ownership and
Financing. PIDE Working Papers 2010:57.
Jensen, M. C. 1986. Agency costs of free cash flow, corporate finance and takeovers. American Economic
Review, 76:323-329.
Jensen, M. C. 1993. The Modern Industrial Revolution, Exit, and the Failure of Internal Control Systems.
Journal of Finance:831-880.
Kang, E. & Zardkoohi, A. 2005. Board leadership structure and firm performance. Corporate Governance,
13:785-799.
Kosnik, R. D. 1990. Effects of board demography and directors’ incentives on corporate greenmail
decisions. Academy of Management Journal, 33:129-150.
Lawrie, A. 1998. Small Firms Survey: Skills. British Chambers of Commerce, London.
Li, M. & Simerly, R. Y. 1998. The moderating effect of environmental dynamism on the ownership and
performance relationship. Strategic Management Journal, 19:169-179.
Lipton, M. & Lorsch, J. 1992. A Modest Proposal for Improved Corporate Governance. The Business
Lawyer, 48:59-77.
Lybaert, N. 1998 The information use in a SME: its importance and some elements of influence. Small
Business Economics, 10:171-191.
Mintzberg, H. 1983. Power in and around organizations. Englewood Cliffs: Prentice Hall.
Mir, S. & Nishat, M. 2004. Corporate Governance Structure and Firm Performance in Pakistan: An
Empirical Study. Paper presented at Second Annual Conference in Corporate Governance. Lahore
University of Management Sciences, Lahore.
Muth, M. M. & Donaldson, L. 1998. Stewardship theory and board structure: A contingency approach.
Corporate Governance, 6:5-27.
Peel, M. & O’Donnell, E. 1995. Board structure, corporate performance, and auditor independence.
Corporate Governance, 3:207-217.
399

Pfeffer, J. & Salancik, G. R. 1978. The external control of organizations:a resource dependence
perspective. New York: Harper and Row.
Powell, W. W. 1991. Expanding the scope of institutional analysis. The new institutionalism in
organizational analysis. Chicago: University of Chicago Press.
Rais, R. B. & Saeed, A. 2005. Regulatory Impact Assessment of SECP’s Corporate Governance Code in
Pakistan. Lahore University of Management Sciences, Lahore, CMER Working Paper 06-39.
Rajagopalan, N. & Zhang, Y. 2008. Corporate governance reforms in China and India: Challenges and
opportunities. Business Horizons, 51:55-64.
Rechner, P. L. & Dalton, D. R. 1991. CEO duality and organizational performance: a longitudinal analysis.
Strategic Management Journal, 12:155-160.
Renneboog, L. 2000. Ownership, managerial control and the corporate governance of companies listed on
the Brussels Stock Exchange, Journal of Banking and Finance, 24:1959-1995.
Roberts, J., McNulty, T. & Stiles, P. 2005. Beyond agency conceptions of the work of the nonexecutive
director: Creating accountability in the boardroom. British Journal of Management, 16:5-26.
Rosenstein, S. & Wyatt, J. 1990. Outside directors, board effectiveness and shareholders’ wealth. Journal of
Financial Economics, 26:175-191.
Sanda, A. U., Mukaila, A. S. & Garba, T. 2003. Corporate Governance Mechanisms and Firm Financial
Performance in Nigeria. Final Report Presented to the Biannual Research Workshop of the AERC,
Nairobi, Kenya, 24-29.
Shaheen, R. & Nishat, M. 2005. Corporate Governance and Firm Performance: An Exploratory Analysis.
Paper presented in the Conference of Lahore School of Management Sciences, Lahore.
Short, H. & Keasey, K. 1999. Managerial ownership and the performance of firms: evidence from the UK.
Journal of Corporate Finance, 5:79-101.
Shrader, B., Blackburn, V. B. & Iles, P. 1997. Women in management and firm financial value: an
exploratory study. Journal of Managerial Issues, 9:355-372.
Uzun, H., Szewczyk, S. H. & Varma, R. 2004. Board composition and corporate fraud. Financial Analysts
Journal, 60:33-43.
Van den Berghe, L. A. A. & Levrau, A. 2004. Evaluating boards of directors: What constitutes a good
corporate board? Corporate Governance, 12:461-478.
Wen, Y., Rwegasira, K. & Bilderbeek, J. 2002. Corporate Governance and Capital Structure Decisions of
Chinese Listed Firms. Corporate Governance, 10:75-83.
Yermack, D. 1996. Higher market valuation of companies with a small board of directors. Journal of
Financial Economics, 40:185-221.
400

CORPORATE GOVERNANCE AND COMPLIANCE WITH IFRSs - MENA EVIDENCE
Marwa Hassaan & Omneya Abdelsalam
Aston Business School, Birmingham, UK
Email: hassanm3@aston.ac.uk
Abstract This paper aims to examine the influence of corporate governance as a newly introduced concept in the MENA region on
the levels of compliance with IFRSs disclosure requirements by companies listed on two leading stock exchanges in the region; CASE
and ASE. This study employs a cross-sectional analysis of all non-financial companies listed on both stock exchanges for the fiscal
year ending December, 2007. Using a disclosure index derived from mandatory IFRSs disclosure requirements for the fiscal year
beginning January, 2007; this study measures the levels of compliance by companies listed on focus stock exchanges. An innovative
theoretical foundation is deployed, in which compliance is interpretable through three lenses; the institutional isomorphism theory,
secrecy versus transparency as one of Gray's accounting sub-cultural values and economic-based theories. This study extends the
financial reporting literature, cross-national comparative financial disclosure literature and the emerging markets disclosure literature
by carrying out one of the first comprehensive comparative studies in the MENA region. Results provide evidence of de jure but not
de facto compliance with IFRSs disclosure requirements by scrutinised MENA countries. In broad terms the influence of corporate
governance best practices on the levels of compliance with IFRSs is limited as it is not yet part of the cultural values in the MENA
society. The results of the multivariate analysis shows that levels of compliance with the IFRSs disclosure requirements can best be
explained by ownership structure. Companies with a higher proportion of public ownership, comply less with IFRSs disclosure
requirements. These findings are consistent with the notions of the proposed theoretical foundation.
Keywords IFRSs, MENA region, Egypt, Jordan, CASE, ASE, Role Duality, Board Composition, Ownership Structure, Isomorphism
and Gray’s Accounting Sub-culture Model
1 Introduction
In a global economy, the financial reporting practices by companies around the world are a key issue. Globalization
of the capital markets has increased the need for high-quality, comparable financial information consequently,
pressure has been increasing for adoption of a single set of accounting standards worldwide (Joshi et al., 2008).
The use of the IFRSs is expected to improve the comparability of financial statements, enhances corporate
transparency, and hence increases the quality of financial reporting worldwide (Daske et al., 2008). Furthermore, for
the Middle East and North Africa (MENA) capital markets as emerging economies, compliance with the IFRSs may be
essential in order to foster their economic transition (CIPE, 2003) as it would attract more direct foreign investments.
This study is motivated by a belief that achieving de facto compliance with the IFRSs by the MENA region listed
companies is not an easy task. It is an on-going process which requires a strong support from researchers, capital
market authorities, accounting regulators, business firms, accounting practitioners and other stakeholders.
The MENA region capital markets that are under scrutiny in this study (Egypt and Jordan) mandated the adoption of
the IFRSs by companies listed on their stock exchanges in 1997 and 1998 respectively. Consequently, they can be
considered as early adopters of the IFRSs compared to the European Union countries that only required companies
listed on their stock exchanges to prepare their financial statements in accordance with the IFRSs in 2005 and
mandated their adoption in 2007. This fact raises the need to investigate the obstacles to full compliance with the
IFRSs especially after the introduction of corporate governance concept in the MENA region which is supposed to
enhance the levels of disclosure and transparency by publicly listed companies.
The review of prior compliance literature reveals that there is a shortage in the number of financial disclosure studies
that investigated emerging capital markets in general and the MENA capital markets in particular. Consequently, this
study is expected to contribute to filling this gap. On the other hand, as financial disclosure lies at the core of all
corporate governance statutes and codes, investigating the association between corporate governance as a newly
introduced concept in the region and the levels of compliance with the IFRSs disclosure requirements is expected to
enrich financial disclosure as well as corporate governance literature. Moreover, this study to the best of the
researchers' knowledge is the first to use the institutional isomorphism theory in providing a theoretical foundation for
the impact of corporate governance structures on the levels of compliance with the IFRSs in the MENA region.
This paper addresses five research questions:
1. What is the extent of compliance with the IFRSs disclosure requirements by companies listed on the two
selected MENA region stock exchanges?
2. How could differences in levels of compliance with the IFRSs be explained by board of directors (BOD)
independence?
401

3. How could differences in levels of compliance with the IFRSs be explained by role duality?
4. How could differences in levels of compliance with the IFRSs be explained by ownership structure?
5. To what extent do institutional isomorphism theory, cultural theories and economic-based theories help to
explain the levels of compliance with the IFRSs disclosure requirements within the MENA context?
In order to answer the research questions the remaining part of the paper is organized as follows: A literature review is
provided in Section 2. Section 3 develops and formulates research hypotheses. Section 4 describes sample selection,
data collection, and research methods. Results and analysis are presented in Section 5. Finally, Section 6 concludes.
2 Literature Review and Formulation of Research Hypothesis
Financial disclosure is a rich field of empirical enquiry (Healy & Palepu, 2001). More recently, researchers became
more concerned with investigating the issue of adopting a unified set of accounting standards worldwide. This line
of research is concerned with investigating the applicability of full compliance with the IFRSs and the association
between levels of compliance with the IFRSs and disclosure environment attributes. The importance of evaluating
the levels of compliance with mandatory disclosure requirements, proved by the findings of this line of research
which reports the existence of low level of compliance with mandatory disclosure requirements particularly in
developing countries (e.g., Abdelsalam & Weetman, 2003; Glaum & Street, 2003; Owusu-Ansah & Yeoh, 2005;
Samaha, 2006; Dahawy & Conover, 2007; Al-Shammari et al., 2008).
For emerging capital markets good corporate governance practices may be essential for a sound success of their reform
programmes and for reserving a healthy investment environment. Following a series of events that took place over the
last two decades addressed by the Asian financial crisis and the wide spread of high-profile corporate scandals such as
WorldCom and Enron, empirical research into accounting disclosure practices began to consider the impact of
corporate governance structures on disclosure practices. The development of corporate governance is a global
occurance thus is influenced by legal, cultural, ownership and other structural differences (Mallin, 2009: 13). To date
corporate governance does not have a widely accepted paradigm or theoretical foundation (Tricker, 2009: 233).
Transparency, fairness and accountability are the core values of corporate governance. Stemming from the desire to
enhance access to more capital that is necessary to achieve economic development and globalise their economies,
corporate governance practices have been brought in the spotlight in developing countries. In this regard many
researchers highlight the influence of corporate board composition and ownership structure (e.g., Eng & Mak, 2003,
Ghazali & Weetman, 2006; Ezat & El-Masry, 2008) on disclosure practices of companies listed on emerging stock
exchanges. However, we suggest that as corporate governance culture was initiated in developed countries and as it is
newly introduced in developing countries, their contribution to enhancing capital markets' performance is subject to the
extent to which the requirements for good corporate governance practices are consistent with the existing values, past
experiences and the needs of all parties involved in the financial reporting process. Otherwise, it is expected to take
some time until the impact of corporate governance culture can be measured. This is because it needs developing an
understanding, forming a favourable attitude and belief and developing the skills required to apply corporate
governance best practices.
3 Development of Hypotheses
We have chosen variables to represent country (stock exchange in which the company is listed), particular aspects of ownership
and board composition (role duality and board composition). In each case we state an expectation based on prior literature.
3.1 Country
although Egypt and Jordan are not homogeneous in terms of their capital markets capacities to practice and enforce
compliance with the IFRSs and corporate governance principles (CIPE, 2003) mainly due to the shortage in
qualified accountants in the Jordanian context, the similarities in their legal, economic and cultural contexts are
expected to reduce the differences in the levels of compliance with the IFRSs between both countries. In addition,
cultural similarities and the novelty of corporate governance concept in both jurisdictions are expected to reduce
differences in the dominant corporate governance structures between Egypt and Jordan. Accordingly, the first
research hypothesis can be stated as follows:
H1: There is no significant statistical differences between the Egyptian and the Jordanian contexts.
402

This hypothesis can be further divided into the following two hypotheses.
H1a: There is no significant statistical differences between Egypt and Jordan in the levels of compliance with the
IFRSs disclosure requirements.
H1b: There is no significant statistical differences between Egypt and Jordan in the dominant corporate governance structures.
3.2 Role Duality
Role dualityis a governance issue that concerns with whether the chief executive officer (CEO) is also thechair of the board of directors.
Separating the two positions has the potential to improve the monitoring function of the board and to reduce the
advantages gained by withholding information, hence to improve the quality of reporting (Arcay & Vazquez, 2005).
Combining both roles reduces the availability of independent evaluation of the CEO's performance as the CEOs
themselves will select which information to be provided to other directors (Jensen, 1993). In addition, dual role
creates a strong individual power base that could impair board independence thus the effectiveness of its governing
function may be compromised (Abdelsalam & Elmasry, 2008).
On the other side, some researchers argue that the separation of the two positions is not essential for better
performance (Dahya et al., 1996; Gul & Leung, 2004). Furthermore, Rechner & Dalton (1991) suggest that role
duality leads to a clear unfettered leadership of boards and companies.
With respect to the MENA region listed companies including Egypt and Jordan, role duality is common as the majority
of companies are family-owned which makes it difficult from their perspectives to induce an owner of a company who
invested money to step aside and allow others to manage his money (CIPE, 2003; IFC & Hawkamah, 2008).
The results of prior research investigated the association between role duality and levels of financial disclosure are
mixed. Some studies show that role duality is significantly associated with a lower level of financial disclosure (e.g.,
Haniffa & Cooke, 2002; Gul & Leung, 2004; Arcay & Vazquez, 2005; Abdelsalam & Elamasry, 2008) while Ghazali
& Weetman (2006) report an insignificant relationship. On the other hand, some empirical research show that there is
no association between role duality and financial disclosure or reporting quality (Cheng & Courtenay, 2006; Ghazali &
Weetman, 2006) and one study (Felo, 2009) reports a positive relationship between role duality and financial disclosure
practices. The contradictory nature of these results raises the need to re-examine this relationship and make it difficult
to predict the type of the relationship between role duality and levels of compliance with IFRSs in the scrutinised
MENA capital markets. Accordingly, the second research hypothesis can be stated as follows:
H2: There are no statistically significant differences in the levels of compliance with the IFRSs disclosure
requirements between companies that separate the positions of the CEO and the Chair and those that do not.
3.3 Board independence
Board independence is an issue that concerns with the number of independent non-executive directors (outsiders)
compared to the number of executive directors (insiders) on firm boards.
Typically a board with more independent directors is expected to be more effective in monitoring management and
hence leads to improved financial disclosure (Haniffa & Cooke, 2002; Dey, 2008). It is expected that insiders cannot
effectively monitor themselves on behalf of shareholders (Muslu, 2005).
Dominance of non-executive directors in the board will maximize their ability to enforce management to meet the
disclosure requirements of different stakeholder groups (Haniffa & Cooke, 2002; Ghazali & Weetman, 2006).
Non-executive directors further to their prestige, expertise and contacts are respected due to their wisdom and
independence (Haniffa & Cooke, 2002).
Opponents argue that increasing the number of non-executive directors on the board result in excessive monitoring
(Baysinger & Butler, 1985), or lack of genuine independence (Demb & Neubauer, 1992). Insiders also can provide
operational efficiency (Muslu, 2005).
Concerning the attribute toward board independence in the MENA region, the existence of a number of non-executive
directors on the board is recognized in companies listed on scrutinised MENA stock exchanges. However, the issue is
that in most cases non-executive board members lack independence or they may lack experience (CIPE, 2003: 37).
Findings of prior research that examined the association between board independence and financial disclosure are mixed
which makes it difficult to predict the relationship between board independence and levels of compliance with the IFRSs
in the scrutinised MENA capital markets thus raises the need to revisit this issue. Some researchers report a positive
relationship (e.g., Arcay & Vazquez, 2005; Cheng & Courtenay, 2006; Abdelsalam & Street, 2007; Abdelsalam &
Elmasry, 2008; Ezat & Elmasry, 2008; Felo, 2009). In contrast, some researchers report a negative relationship (Eng &
Mak, 2003; Gul & Leung, 2004; Muslu, 2005) while other researchers did not find any relationship (Haniffa & Cooke,
2002; Ghazali & Weetman, 2006). Accordingly, the third research hypothesis can be stated as follows:
403

H3: There is no relationship between BOD independence and the extent of compliance with the IFRSs disclosure requirements.
3.4 Ownership Structure
Ownership structure is defined by Denis & McConnell (2003: 3) as ‘The identities of a firm's equity holders and the
sizes of their positions’.
The ownership structure of a firm may be a possible determinant of firm disclosure practices (Eng & Mak, 2003; Arcay &
Vazquez, 2005). High levels of concentration of capital may be accompanied by the owner’s considerable involvement in
the firm’s management, which in turn may lead to unrestricted access to information thus may limit the demand hence the
supply for company information and vice versa (Haniffa & Cooke, 2002; Arcay & Vazquez, 2005; Ezat & El-Masry, 2008).
On the other hand, when share ownership is widely held, the potential for conflicts of interests between the principal
and the agent is greater than in closely held companies. As a result, disclosure is likely to be greater in widely held
companies to enable the principal to effectively monitor whether his/her economic interests are optimized and whether
the agent acts in the best interests of the principal as an owner of the firm (Fama & Jensen, 1983; Chen & Gray, 2002).
The review of the dominant ownership structures in the focus MENA stock exchanges reveals that shares of most
companies are family owned or government owned (CIPE, 2003; Al-Htaybat, 2005; Naser et al., 2006; Tricker,
2009). The dominance of family ownership is expected to have a negative impact on the level of financial disclosure
for two reasons; firstly, they have direct access to company information (Naser et al., 2006). Secondly, the secretive
nature of these societies makes family shareholders encourage management to keep disclosure at minimum levels as
long as costs of compliance exceeds costs of non- compliance regardless of the impact on the interests of minority
shareholders. With respect to the impact of dominant government ownership there are two distinct points of view.
The first argues that in the case of dominant government ownership, disclosure levels will be low as government can
directly request any information they need from company management (Naser et al., 2006). The other point of view
sees dominant governmental ownership advantageous as based on agency theory it will improve disclosure practices
as the government will encourage management to use competent disclosure policies (Suwaidan, 1997) thus to reduce
monitoring costs management will choose to comply with the IFRSs.
Based on the review of the patterns of ownership structure in the listed MENA companies under scrutiny in this study
and the availability of ownership structure related data for these companies, this study examines the influence of
ownership structure on the levels of compliance with the IFRSs in the scrutinised MENA capital markets using four
distinct measures; government ownership (defined as the percentage of company shares owned by the government),
management ownership (defined as the percentage of company shares owned by company management and other
board members), private ownership (defined as the percentage of company shares owned by private shareholders) and
public ownership (defined as the percentage of company shares owned by the free float that is less than 5%).
3.4.1 Government ownership
The results of prior studies that examined the association between government ownership and levels of financial
disclosure are mixed. For instance, Eng & Mak, (2003) report a positive relationship while Naser et al. (2006) report
a negative relationship and Ghazali & Weetman (2006) find a negative but insignificant relationship.
Accordingly, built on the above discussion research hypothesis (4a) can be stated as follows:
H4a: There is no relationship between government ownership ratio and the extent of compliance with the IFRSs
disclosure requirements.
3.4.2 Management Ownership
The results of the majority of prior studies that examined the association between management ownership and levels
of financial disclosure show a negative association (e.g., Eng & Mak, 2003; Arcay & Vazquez, 2005; Ghazali &
Weetman, 2006; Abdelsalam & El-Masry, 2008). Thus, the effect of managerial ownership on levels of compliance
with the IFRSs is expected to be substitutive. Accordingly, research hypothesis (4b) can be stated as follows:
H4b: There is a negative relationship between management ownership ratio and the levels of compliance with the IFRSs.
3.4.3 Private Ownership
Similar to government ownership, there is no consensus among prior reseachers regarding the influence of private
ownership on the levels of compliance with financial disclosure requirements. Some researchers report that private
ownership may be a complementary (Diamond & Verrenchia, 1991 as cited in Haniffa & Cooke, 2002). On the
contrary, some researchers report that concentrated ownership and financial disclosure are substitutes (dominance of
404

private shareholders reduces levels of financial disclosure) such as Naser et al. (2006) while others report no
association between the dominance of private ownership and levels of financial disclosure (Suwaidan, 1997; Depoers,
2000; Omar, 2007). Accordingly, research hypothesis (4c) can be stated as follows:
H4c: There is no relationship between private ownership ratio and the extent of compliance with IFRSs disclosure requirements.
3.4.4 Public Ownership
The results of most prior studies show a positive association between public ownership and levels of financial
disclosure (e.g., Haniffa & Cooke, 2002; Al-Htaybat, 2005; Arcay & Vazquez, 2005). Accordingly, research
hypothesis (4d) can be stated as follows:
H4d: There is a positive relationship between public ownership ratio and the extent of compliance with the IFRSs
disclosure requirements.
3.5 Control Variables
The review of prior financial disclosure studies led to the decision to incorporate BOD size and three firm-specific
characteristics in the multivariate analysis as control variables; company size (total assets), type of business activity
(non-manufacturing versus manufacturing) and type of audit firm (Big 4 versus non Big 4).
Prior research results are mixed regarding the direction of the association between BOD size and disclosure
practices. Hence, we include BOD size as a control variable but do not predict the direction of the association.
Based on the findings of prior researchers (Eng & Mak, 2003; Akhtaruddin, 2005; Aksu & Kosdag, 2006 and Naser
et al., 2006) we include firm size as a control variable and we anticipate a positive relationship between firm size
and levels of compliance with the IFRSs disclosure requirements.
Based on the evidence provided by prior research that the type of business activity influences its disclosure practices (e.g,
Cooke, 1992; Haniffa & Cooke, 2002), type of business activity will be employed in this study as a control variable.
Based on the evidence provided by prior research that there is a relationship between the type of auditor and levels
of company disclosure (e.g., Patton & Zelenka, 1997; Glaum & Street 2003), type of auditor will be employed in
this study as a control variable and it is expected that auditing by a Big 4 audit firm has a positive impact on the
levels of compliance with the IFRSs.
4 Methodology
4.1 Sample Selection
This study applies to the annual reports for the fiscal year ending December 2007 for the entire population of nonfinancial
companies listed on scrutinised MENA region stock exchanges with total 311 companies (145 companies
from Egypt and 166 from Jordan). After excluding companies for which no corporate governance information for
2007 was available or those for which the 2007 complete annual report was not available in either Zawya database
or EGID (Egypt for Information Dissemination Company), the final sample contained 102 companies (75 from
Egypt and 27 from Jordan).
4.2 Disclosure Checklist
To meet the purpose of this study the researchers use a self-constructed disclosure checklist based on the IFRSs
required to be followed by the IASB in preparing the financial statements for the fiscal year beginning January
2007.The disclosure index employed in this study includes 275 IFRSs based items 1 . Thus it can be considered
amongst the most comprehensive mandatory disclosure indices applied in the MENA region. The disclosure index
for each company was calculated as the ratio of the total actual score awarded to the maximum possible score of
relevant items applicable for that company.
The calculation of the disclosure index (dependent variable) for each company under this approach is as follows:
DI = ADS/ MD
Where:
DI refers to the disclosure index (0≤ DI≤1)
1 The disclosure index employed in this study is not presented here to save space but available from authors upon request
405

ADS refers to the aggregate disclosure score for a particular company
MD refers to the maximum score possible for that company (≤ 275 items)
4.3 Data Collection and Regression Model
The data on the chosen independent variables (see below) were obtained from Zawya database and from EGID. The
following multiple regression model is proposed:
Yj= β0+ β1 countryj + β2 role dualityj + β3 board independencej + β4 government ownership ratioj + β5
management ownership ratioj + β6 private ownership ratioj + β7 public ownership ratioj + β8 board sizej + β9 total
assetsj + β10 type of business activityj + β11 type of audit firmj + Ɛj
Where:
Yj= Disclosure index for companies (j=1,…, 102) which denotes the dependent variable;
β0= The intercept;
Ɛj= Error term
The regression model incorporates two categorical test variables, country (1 if Jordan, 0 if Egypt) and role duality (1 if
the chairman is not the CEO, 0 if the chairman is the CEO). It also incorporates two categorical control variables, type
of business activity (1 if non- manufacturing; 0 if manufacturing) and type of audit firm (1 if a big 4, 0 if not a big 4).
5 Results and Analysis
5.1 Descriptive Statistics
The results of the analysis of the 102 annual reports of companies listed on the two scrutinised stock exchanges,
CASE (75 companies) and ASE (27 companies) as demonstrated in table 1 show that the average level of
compliance with the IFRSs disclosure requirements is 80% in Egypt and 78% in Jordan. This implies that there is a
relative similarity between the levels of compliance with the IFRSs disclosure requirements between the two stock
exchanges. However, it is also recognised that the minimum level of compliance with the IFRSs disclosure
requirements by companies listed on CASE is higher than that of companies listed on ASE (68% and 59%
respectively). This difference may be attributed to the shortage in qualified accountants in Jordan.
Country
Egypt
Jordan
Minimum
%
68
59
Maximum
%
91
87
Table 1: Comparison between Egypt and Jordan in terms of the levels of compliance
with the IFRSs disclosure requirements (total score)
Mean
%
80
78
Applicability
75
27
Number of companies
got above 50%
75
27
5.2 Testing Differences between the Egyptian and the Jordanian Contexts (H1)
Percentage of companies
got above 50%
100%
100%
In order to investigate whether there are significant statistical differences between the Egyptian and the Jordanian
contexts (the first research hypothesis), the researcher employed the Mann-Whitney U test 2 .
Mann-Whitney U test has been used to examine whether there are significant statistical differences among companies'
disclosure indices (dependent variables), and among independent variables (board independence, role duality, ownership
structure, BOD size, firm size, business activity and audit firm) between companies listed on CASE and those listed on
ASE. Accordingly the first research hypothesis has been further divided into the following two sub-hypotheses.
5.2.1 Significant Differences between Egypt and Jordan in terms of the Levels of Compliance with the IFRSs
Disclosure requirements
The results demonstrated in table 2 show that there are no statistically significant differences between the two groups of
companies: those listed on CASE and those listed on ASE with respect to total disclosure score (the aggregate level of
compliance with the IFRSs disclosure requirements) which is the main dependent variable in this study (P>.05).
2 Normality tests were performed using Shapiro-Wilk and K-S tests (not presented here to save space but available from authors upon request)
showed that data distribution is not normal thus following prior research (e.g., Ghazali & Weetman, 2006) the researcher decided to use nonparametric
tests and to use normal scores in multivariate analysis.
406
Rank
NA
NA

Table 2: Mann-Whitney U Test Results for the Dependent Variable
Disclosure Index (N=102) Mann-Whitney U Wilcoxon W Z Asymp. Sig. (2-tailed)
Total Score 895.500 .013 -.887 .375
5.2.2 Significant Differences between Egypt and Jordan in terms of the Independent Variables
The results demonstrated in table 3 show that there are no statistically significant differences (P>.05) with respect to
all independent variables except two of the test variables: role duality and board independence (P

Constant .027 .094 .287 .775
Public ownership ratio -.281 .115 -.238 -2.436 .017 1.000 1.000
As seen in table 4 the adjusted R 2 = .047. This implies that approximately 5% of the variation in the aggregate
disclosure index is explained by variation in public ownership ratio. In addition, the model reaches statistical
significance whereas F= 5.932 and the significance =.017 (

esults in the restriction of information to preserve power inequalities. On the other side, this lends support to agency
theory (as government, private and management stockholders are in a position to get access to all company
information they need and as public investors do not demand more compliance with IFRSs disclosure requirements
and do not put any pressures on BOD or management to improve compliance, monitoring costs will be low so will
reduce management incentives to comply with the IFRSs disclosure requirements. In addition, this result is
consistent with the notions of cost-benefit analysis as the weak enforcement of the IFRSs disclosure requirements
and weak sanctions if any causes non-compliance costs to be less than compliance costs for companies listed on
scrutinised MENA stock exchanges, thus direct management incentives toward non-compliance. All of this
contributes to the problem of decoupling as companies listed on MENA stock exchanges state that they prepare their
financial statements in accordance with the IFRSs, while none of them fully comply with these requirements.
On the other side, low explanatory power of the regression model calls for carrying out a number of interviews with
different parties involved directly or indirectly with financial reporting process in the MENA emerging markets to
explore the other factors that best explain levels of compliance with the IFRSs disclosure requirements in the region
that are difficult to quantify and analyse using quantitative research methods.
References
Abd-Elsalam, O. H. & Weetman, P. (2003). `Introducing International Accounting Standards to an Emerging Capital
Market: Relative Familiarity & Language Effect in Egypt'. Journal of International Accounting. 12, pp. 63-84.
Abdelsalam, O. & El-Masry, A. (2008). ‘The Impact of Board Independence and Ownership Structure on the Timeliness
of Corporate Internet Reporting of Irish-listed Companies’. Managerial Finance. Vol. 34 (12). pp. 907-918.
Akhtaruddin, M. (2005), `Corporate Mandatory Disclosure Practices in Bangladesh'. The International Journal
ofAccounting. 40, pp. 399-422.
Aksu, M. & Kosedag, A. (2006). `Transparency and Disclosure Scores and their Determinants in Istanbul Stock
Exchange'. Corporate Governance. Vol. 14 (4). pp. 277- 296.
Al-Htaybat, K. (2005). ‘Financial Disclosure Practices: Theoretical Foundation and an Empirical Investigation on
Jordanian Printed and Internet Formats’. PhD Thesis. University of Southampton. UK.
Al-Shammari, B., Brown, P. & Tarca, A. (2008). ‘An Investigation of Compliance with International Accounting Standards by Listed
Companies in the Gulf Co-operation Council Member States’. The International Journal of Accounting. Vol. 43. pp. 425-447.
Arcay, M.R.B. & Vazquez, M.F.M. (2005). ‘Corporate Characteristics, Governance Rules and the Extent of
Voluntary Disclosure in Spain’. Advances in Accounting. Vol. 21. pp. 299-331.
Al-Jifri, K. (2008). ‘Annual Report Disclosure in a Developing Country: The Case of the UAE’. Advances in
Accounting Incorporating Advances in International Accounting. Vol. 24. pp. 93-100.
Baysinger, B., & Butler, H. (1985). ‘Corporate Governance and Board of Directors: Performance Effects of
Changes in Board Composition’. Journal of Law. Economics and Organization 1: 101–24.
CIPE (2003). ‘Corporate Governance in Morocco, Egypt, Lebanon and Jordan- Countries of the MENA Region’.
Middle East and North Africa Corporate Governance Workshop. The Center for International Private Enterprise.
Chau, G. K., & Gray, S. J. (2002). ‘Ownership Structure and Corporate Voluntary Disclosure in Hong Kong and
Singapore’. The International Journal of Accounting. Vol. 37. pp. 247–265.
Cheng, E.C.M. & Courtenay, S.M. (2006). Board Composition, Regulatory Regime, and Voluntary Disclosure. The
International Journal of Accounting. Vol. 41. pp. 262-289.
Dahawy, K. & Conover, T. (2007). ‘Accounting Disclosure in Companies Listed on the Egyptian Stock Exchange’.
Middle Eastern Finance and Economics. Issue (1). pp. 5-20.
Dahya, J., Lonie, A.A. & Power, D.M. (1996). ‘The case for separating the roles of Chairman and CEO: an analysis
of stock market and accounting data’, Corporate Governance – An International Review. Vol. 4 (1). pp. 71-77.
Daske, H. (2006). `Economic Benefits of Adopting IFRS or US-GAAP- Have the Expected Cost of Equity Capital
Really Decreased?'. Journal of Business Finance and Accounting Research. Vol. 33 (3). pp. 329-373.
Dey, A. (2008). ‘Corporate Governance and Agency Conflicts’. Journal of Accounting Research. Vol. 46 (5). pp.1143-1181.
Depoers, F. (2000), `A Cost-Benefit Study of Voluntary Disclosure: Some Empirical Evidence from French Listed
Companies', The European Accounting Review, 9, (2), pp.245-263.
409

Diamond, D. W. & Verrecchia, R. E. (1991), `Disclosure, Liquidity and the Cost of Capital'. The Journal of Finance.
Vol. 46 (4). September. pp. 1325-1359.
Eng, L.L. & Mak, Y.T. (2003). ‘Corporate Governance and Voluntary Disclosure’. Journal of Accounting and
Public Policy. Vol. 22 (4). pp. 325-345.
Ezat, A., El-Masry, A. (2008). ‘The Impact of Corporate Governance on the Timeliness of Corporate Internet
Reporting by Egyptian Listed Companies’. Managerial Finance. Vol. 34 (12). pp. 848-867.
Fama, E. F. & Jensen, M.C. (1983). `Separation of Ownership and Control', Journal of Law and Economics. Vol. 88. pp. 301-325.
Felo, A.J. (2009). ‘Voluntary Disclosure Transparency, Board Independence and Expertise, and CEO Duality’.
Working Paper. http:// ssrn.com/abstract=1373942.
Forker, J. (1992). ‘Corporate Governance and Disclosure Quality’. Accounting and Business Research. Vol. 22 (86). pp. 111-124.
Field, A. (2005). Discovering Statistics Using SPSS. (2 nd Edition). London: Sage Publications Ltd
Ghazali, N. A. M. & Weetman, P. (2006), `Perpetuating Traditional Influences: Voluntary Disclosure in Malaysia
Following the Economic Crisis', Journal of International Accounting, Auditing and Taxation. 15, pp. 226-248.
Glaum, M. & Street, D. L. (2003). `Compliance with the Disclosure Requirements of Germany's New Market: IAS
versus US GAAP'. Journal of International Financial Management and Accounting. Vol. 14 (1). pp. 64-100.
Gul, F.A. & Leung, S. (2004), ‘Board Leadership, Outside Directors Expertise and Voluntary Corporate
Disclosures’. Journal of Accounting and Public Policy. Vol. 23, pp. 351-79.
Gray, S. J. (1988). ‘Towards a theory of cultural influence on the development of accounting systems
Internationally’. Abacus. Vol. 24 (1). March. pp. 1–15.
Haniffa, R. M. & Cooke, T. E. (2002). `Culture, Corporate Governance and Disclosure in Malaysian Corporations'.
Abacus. Vol. 38 (3). pp. 317-349.
Healy, P. M. & Palepu, G. (2001). `Information Asymmetry Corporate Disclosure, and the Capital Markets: A
Review of the Empirical Disclosure Literature'. Journal of Accounting and Economics. Vol.31. pp. 405-440.
Jensen, M.C. (1993). ‘The Modern Industrial Revolution, Exit, and the Failure of Internal Control Systems’. The
Journal of Finance. Vol. 48. pp. 831-880.
Joshi, P.L., Bresmer, W.G. & Al-Ajmi, J. (2008). ‘Perceptions of Accounting Professionals in the Adoption and
Implementation of a Single Set of Global Accounting Standards: Evidence from Bahrain’. Advances in
Accounting Incorporating Advances in International Accounting. Vol. 24. pp. 41-48.
Mallin, C.A. (2009). Corporate Governance. (3 rd Edition). New York: Oxford University Press.
Muslu,V. (2005). ‘Effect of Board Independence on Incentive Compensation and Compensation Disclosure:
Evidence from Europe’.Working Paper. available: http://www.ssrn.com. Accessed: 23-9-2009.
Nasr, K., Al-Hussaini, A., Al-Kwari, D. & Nuseibeh, R. (2006). ‘Determinants of Extent of Corporate Social Disclosure in
Developing Countries: The Case of Qatar’. Advances in International Accounting. Vol. 19. pp. 1-23.
Omar, B.F.A (2007). ‘Exploring the Aggregate, Mandatory and Voluntary Financial Disclosure Behaviour Under a
New Regulatory Environment: The Case of Jordan’. PhD Thesis. The University of Hull. UK.
Owusu-Ansah,S. & Yeoh, J.(2005). `The Effect of Legislation on Corporate Disclosure Practices'. Abacus. Vol.41, No. 1,pp.92-109.
Patton, J. & Zelenka, I. (1997). ‘An empirical analysis of the determinants of the extent of disclosure in annual reports of
joint stock companies in the Czech Republic’. The European Accounting Review. Vol.6 (4). pp 605-626.
Samaha, K. (2006). 'Compliance with International Accounting Standards: Some Empirical Evidence from the Cairo &
Alexandria Sock Exchange'. Accounting, Management & Insurance Review. Issue (7). Cairo University Press.
Suwaidan,M.S.(1997).`VoluntaryDisclosureof AccountingInformation:thecaseofJordan'.PhDThesis.UniversityofAberdeen.UK.
Tricker, Bob (2009). Corporate Governance: Principles, Policies and Practices. New York: Oxford University Press.
IFC & Hawkamah (2008). 'A Corporate Governance Survey of Listed Companies and Banks Across the Middle East
and North Africa'. Available: http://www.ifc.org/ifcext/mena.nsf. Accessed 25/2/2010.
410

BUSINESS PERFORMANCE EVALUATION MODELS AND DECISION SUPPORT SYSTEM FOR THE
ELECTRONIC INDUSTRY
Wu Wen, Department of Information Management, Lunghwa University of Science and Technology, Taiwan, R.O.C.
Email:wenwu@mail.lhu.edu.tw
Abstract. In the paper, a decision support system is built for evaluating business performance for manufacturing companies of the
electronic industry. To select better performance indices in the decision support system, we adopt literature review, expert’s
questionnaire, and factor analysis-principal component analysis. Through two-phase expert meeting and questionnaires, 16 out of 28
indices are selected. Next, based on the result of factor analysis, there are four factors (i.e., categories) named as (1) Profitability
Ability-return on assets, return on stockholders’ equity, return on investment, and net profit margin, (2) Efficiency Ability- average
collection period, accounts receivable turnover, inventory turnover, working capital, and (3) Liquidity- current ratio, quick ratio, and
cash ratio. Additionally, the non-financial 5 indices are included as the fourth factor. Besides, an artificial neural network model is
created for forecasting net profit of a company. Finally, analysis hierarchy process (AHP) is adopted to evaluate business performance
and a management decision support system is built for providing some managerial suggestions to middle or top managers for
conducting business.
Keywords: Factor analysis; Principal component analysis; Analytic hierarchy process; Business Performance financial index
1 Introduction
Accurate business performance evaluation is a key to success for enterprises. Particularly, a company in the
competitive environment of the 21st century requires substantial financial and non-financial analysis, rapid response,
efficient management, and high quality of products and services for maintaining superiority. In the past, manual selfannual
reports on financial statements such as income statement and balance sheet would be done and used to
examine a company’s performance. Meanwhile, without considering other competitors in the same industry, a
company will lead to self-satisfaction. Furthermore, in the modern era, it is unwise to merely calculate models by
hand. Conversely, we should build a business performance evaluation decision support system for providing optimal
suggestions to the managers for conducting their company for facing the rapid changing global environment.
Therefore, this paper employs basic statistics, principal component analysis, and analytic hierarchy process method
to build a business performance evaluation model and decision support system for middle or top managers to
conduct their company.
As we know that business performance can be measured by using financial and non-financial factors. Based on
literature review, we have studied many financial and non-financial indices related to business performance.
Grafton, Lillis, and Widener (Grafton et al., 2010) used a structure equation model to analyze the degree of
commonality between measures identified as decision-facilitating and decision-influencing, which is significantly
associated with the use of decision-facilitating measures for both feedback and feed-forward control. They also
suggested managers to use the multiple financial and non-financial performance indicators increasingly incorporated
in contemporary performance measurement systems. Saranga and Moser (Saranga & Moser, 2010) believed that
owing to the potential to strategically influence both operational performance as well as financial performance
outcomes, Purchasing and Supply Management (PSM) today is increasingly playing an important role to senior
managers. They developed a comprehensive performance measurement framework using the classical and two-stage
Value Chain Data Envelopment Analysis models, which make use of multiple PSM measures at various stages. In
their paper, a single efficiency measure that estimates the all-round performance of a PSM function was provided.
Eddy, Paula, and Van (Eddy et al., 2010) surveyed that an organization and the presentation of performance
measures are affected by how evaluators weight financial and non-financial measures. Through two experiments,
they found that when the performance differences belong to the financial category, evaluators use a BSC-format
place more weight on financial category measures than evaluators using an unformatted scorecard. The second
finding is that when performance markers are added to the scorecards, evaluators use a BSC-format weight measures
in any category containing a performance difference more heavily than evaluators using an unformatted scorecard.
By studying financial performance indices, Sohn et al. (Sohn et al., 2007) proposed a structural equation model
(SEM) to examine the relationship between technology evaluation factors and the financial performance. It can be
411

used for not only for the effective management of the technology credit funds for small and medium enterprises
(SME) but also for evaluating financial performance of SMEs based on the technology evaluation of companies.
Their results showed that the operation ability of manager has the highest direct effect on the finance performance
index (FPI) and the level of technology has the highest indirect effect on the FPI. Knowledge and experience of
manager as well as marketing of technology have positive effect on the FPI. Ocal et al. (Ocal et al., 2007) used
factor analysis to select the financial indicators for evaluating financial trend of Turkish construction industry. They
collected 5 years of data starting from 1997 to 2001 for 28 Istanbul Stock Exchange traded construction companies.
In the factor analysis, there were 25 ratios adopted. According to the values of the correlation matrix, 9 ratios had a
weak correction with the others and could be removed. The results of factor analysis showed that 5 factors would be
extracted (i.e., eigenvalues are larger than 1). They are named as liquidity factor, capital structure and profitability
factor, activity efficiency factor, profit margin and growth factor, and assets structure factor. Wen et al. (Wen et al.,
2008) presented a knowledge-based decision support system for evaluating enterprise performance. The KDSS
system provides not only company’s various financial data query, but also enterprise performance based on
knowledge reasoning. The system integrates a database, a knowledge base, an inference engine, and a model base.
The model in the model base adopts 12 key financial indicates, debt to total assets ratio, permanent capital to fixed
assets ratio, current ratio, quick ratio, accounts receivable turnover, average accounts receivable collection,
inventory turnover ratio, fixed asset turnover ratio, total asset turnover ratio, return on assets, return on equity, and
net profit ratio to assess business performance. Through a rating mechanism, a company can know what position it
locates in its industry. Tseng et al. (Tseng et al., 2009) developed a performance evaluation model for assessing
high-tech manufacturing companies based on a new set of financial and non-financial performance indicators. A
data envelopment analysis (DEA), an analytic hierarchy process (AHP), and a fuzzy multi-criteria decision-making
approach are employed in the model. The data were collected from 5 target companies that produce large-size TFT-
LCD panels. Their results showed that the companies focused on competition performance and financial
performance. The paper also strengths to increase market share, sales growth rate, maintaining steady, and sufficient
upstream materials and supplies, and enhance the ability to obtain critical technology and patents. Lin et al. (Lin et
al.,2005) applied a structural equation model to supply chain quality management and organization performance.
The questionnaire data from both Taiwan and Hong Kong’s supply chain firms were collected. Both data show that
there have direct effects on the relationship between: QM practices and supplier participation, supplier participation
and organization performance, QM practices and supplier selection, supplier participation and supplier selection.
Moreover, the relationship between supplier selection and organizational performance has indirect influences. QM
practice and organizational performance have also indirect effects. Hoque (Hoque, 2004) surveyed and discussed the
impact on performance of two factors, strategy and environmental uncertainty, from 52 samples of manufacturing.
His result shows that management’s strategy choice is positively related to performance remarkably. But there is no
evidence to prove the relationship between environmental uncertainty and performance. For financial analysis, a
better financial company should have 4 abilities: liquidity/debt paying ability, financial structure (stability),
activity/efficiency ability, and profitability (Gibson, 2009; McGuigan et al., 2009; Qcal et al., 2007; Shon et al.,
2007; Tseng et al., 2009; Wen et al., 2008). Other many performance related papers have been published (Wwyer et
al., 2003; Lam, 2004; Lin et al., 2005).
2 Methods
According to literature review, we collected and filtered 28 indices in common use for evaluating business
performance. Among the 28 indices, 18 indices are financial indices and 10 indices are non-financial indices. The
main participants, who filled in the questionnaire, are 34 experts from electronic companies, from academics, and
from accounting department. Through two-phase expert meeting and questionnaires, we collect, discuss and
evaluate all the 28 indictors. The survey adopts a 5-point Likert scale (1=Not important at all, 5=very important).
In the first phase, in order to choose critical indicators, we utilize 3 approaches to reduce the indices from 28 to
18. First, based on total number of “Important” and “Very Important,” shown in questionnaires for financial indices,
the index of which total number is less than or equal to 27 is deleted and for non-financial indices, the index of
which total number is less than or equal to 28 is deleted (see Table 1). After counting the total number of
“Important” and “Very Important,” 10 out of 28 indices are removed. Therefore, times interest earned, debt-toequity,
total asset turnover, fixed asset turnover, gross profit margin, productivity, number of patents, upstream
materials and supplies, downstream tactical alliances are erased.
412

NO
Indictor
Important Very Important Total
1 current ratio 24 6 30
2 quick ratio 21 9 30
3 cash ratio 20 13 33
4 working capital 18 13 31
5 Permanent capital to fixed assets 22 6 28
6 Debt ratio 16 16 32
7 Times interest earned 9 10 19
8 Debt-to-equity 19 8 27
9 Inventory turnover 14 14 28
10 Total asset turnover 15 5 20
11 Accounts receivable turnover 14 16 30
12 Fixed asset turnover 11 5 16
13 Average collection period 16 13 29
14 Return on assets 19 10 29
15 Return on stockholders’ equity 14 16 30
16 Return on investment 20 10 30
17 Net profit margin 16 15 31
18 Gross profit margin 13 13 26
19 Product competitiveness 13 20 33
20 Market share 19 11 30
21 Productivity 20 8 28
22 Product quality level 13 18 31
23 Number of patents 14 25 25
24 R&D expenditure ratio 18 26 26
25 Ability to obtain critical technology 11 33 33
26 Capability to improve manufacturing
processes
19 29 29
27 Upstream materials and supplies 19 27 27
28 Downstream tactical alliances 17 27 27
Table 1 Financial and non-financial indices-total number of important and very important
Second, the study also examines communalities of the 18 indices as criteria to select index. We divided the 18
indices into two groups-financial indices and non-financial indices. The financial group has 13 indices and the nonfinancial
group has 5 indices. For the financial group, in line with the principal component -the values of
communalities, we deleted the indicators, permanent capital to fixed assets and debt ratio of which extraction values
of communalities are lower than 0.2 (see Table 2). Thus, 11 out of 13 indices are chosen as (1) Profitability Abilityreturn
on assets, return on stockholders’ equity, return on investment, and net profit margin, (2) Efficiency Abilityaverage
collection period, accounts receivable turnover, inventory turnover, working capital, and (3) Liquiditycurrent
ratio, quick ratio, and cash ratio. For the non-financial group, none of index’s communalities is lower than
0.2. Thus, all 5 non-financial indices are product competitiveness, market share, product quality level, ability to
improve manufacturing processes.
3 System Implementation
In the second phase, when the 16 critical indices are chosen, we have invited 15 experts from the electronic industry,
academic, and accounting to participate a survey via an AHP electronic questionnaire. Basically, there are 5 pairwise
comparison matrices. The first pair-wise comparison matrix is for business performance, which shows various
abilities including debt paying-efficiency ability, efficiency-profitability ability, debt paying ability, and nonfinancial
ability. The second pair-wise comparison matrix is for debt-efficiency ability consisted of cash ratio,
working capital, inventory turnover, and accounts receivable turnover. The third pair-wise comparison matrix is for
efficiency-profitability ability composed of average collection period, return on assets, return on stockholders’
equity, and return on investment. The fourth pair-wise comparison matrix is debt paying ability including current
413

atio and quick ratio. The fifth pair-wise comparison matrix is for non-financial ability which represents product
competitiveness, market share, product quality level, ability to obtain critical technology, capability to improve
manufacturing processes. Each ability is compared to itself and the rest of others in row on top of the square matrix.
The values (i.e., judgment) of a matrix mean the importance of an item when each item in left column is compared
to the items in the row on top. Therefore, the values for the diagonal of the matrix are 1 because the ability is
compared to itself. For example, number 5 in the (1, 2) or first-row second-column position shows that debt payingefficiency
ability is moderately more important than efficiency-profitability ability. However, the value of 1/2 in the
(2, 3) shows that debt paying ability is minor more than efficiency-profitability ability (see Figure 1 and Figure 2).
Figure 1. AHP electric questionnaire.
414

Figure 2. AHP electric questionnaire (cont.).
The system will check the consistence for every matrix. Hence, once all the five matrices pass the test of
consistence, all average weights are able to be found as shown in Figure 3. In Figure 3, we also list the deciles of
each financial index based on the real value for a company’s financial index which is compared with the average
value of the financial index in the electronic industry. Consequently, the total scores of a company for its business
performance can be calculated.
Figure 3. The total scores of business performance for a company.
415

Figure 4. The broken line graph for business performance for various companies during the recent 2003-2008 years.
Figure 4 presents the broken line graph for business performance for various companies during the recent 2003-
2008 years. Through the diagram, we are easily able to understand whether the business performance of a company
make an improvement or a degeneracy. Furthermore, an artificial neural network model is created for forecasting net
profit of a company. Figure 5 represents an artificial neural network model’s training and test processes. X1, X2, and
X3 present the three input variables in the input layer. Y represents the output variable in the out layer. In other
words, we use the net profit of the first quarter, the net profit of the second quarter, and the net profit of the third
quarter to predict the net profit of the fourth quarter. We set up mean square of error (MSE) is less than 0.001 to end
the training process. Thus, Figure 5 also presents the net profit for the second quarter of fiscal 2010 which is
predicted by using the net profit of the third quarter of fiscal 2009, the net profit of the fourth quarter of fiscal 2009,
and the first quarter of fiscal 2010.
Figure 5. An artificial neural network model for predicting net profit of a manufacturing company of the electronic industry.
4 Conclusions and future work
This paper constructs a business performance evaluation model by using financial and non-financial indices and
constructs a decision support system for providing vital suggestions for top managers for conducting their company.
416

To create the model, we examined more than 40 indices for evaluating business performance in light of literature
review. 28 out of 40 indices are chosen as critical indices. Next, a two-phase expert meeting and questionnaire have
been done to count the total number of “Important” and “Very Important” for each financial and non-financial
index. After that, we also used the principal component -the values of communalities, to select the indictors.
Consequently, 16 out of 28 indices are remained for evaluating business performance. Finally, the principal
component is adopted for running factor analysis. Based on the result of factor analysis, there are four factors named
as (1) Profitability Ability-return on assets, return on stockholders’ equity, return on investment, and net profit
margin, (2) Efficiency Ability- average collection period, accounts receivable turnover, inventory turnover, working
capital, and (3) Liquidity- current ratio, quick ratio, and cash ratio. Additionally, the non-financial 5 indices are
included as the fourth factor. Consequently, there are four factors with 16 indices to process the AHP method to
determine the weight of each index.
Next, a management decision support system is built for providing some managerial suggestions to middle or
top managers for conducting business. In the decision support system, some important information such as financial
information can be collected via online access and an artificial neural network model is created for forecasting net
profit of a company. Additionally, analysis hierarchy process (AHP) is adopted to evaluate business performance.
The system is also able to provide critical knowledge to top managers for knowing what business performance it is
and what position in the electronic industry the company is. Finally the DSS can automatically provide vital private
and internal suggestions to top-level managers for running the company.
5 Acknowledgements
This Research Project was sponsored by NSC:98-2410-H-262-005-MY2, Republic of China.
6 References
S. Dwyer, O. C. Richard & K. Chadwick, “Genderdiversity in management and firm performance: the influence of
growth orientation and organizational culture,” Journal of Business Research, Vol. 56: pp. 1009-1019, 2003.
C. Eddy, M.G. Paula & V.D. Van, “Financial versus non-financial information: The impact of information
organization and presentation in a Balanced Scorecard,” Accounting, Organizations and Society, vol. 35, pp.
565–578, 2010.
C.H. Gibson, Financial reporting and analysis, South-western CENGAGE Learning, 2009.
J. Grafton, A.M. Lillis, & S.K Widener, “The role of performance measurement and evaluation in building
organizational capabilities and performance,” Accounting, Organizations and Society, vol. 35, pp. 689–706,
2010.
Z. Hoque, “A contingency model of the association between strategy, environmental uncertainty and performance
measurement: impact on organizational performance,” International Business Review, vol. 13: pp. 485-502,
2004.
M. Lam, “Neural network techniques for financial performance prediction:integrating fundamental and technical
analysis,” Decision Support systems, vol. 37: pp. 567-581, 2004.
M. J. Lebas, “Performance measurement and performance management,” Internatinal Journal of Production
Economics, 41: pp. 23-35, 1995.
C. Lin, W. S. Chow, C. N. Madu, C. H. Kuei, & P. P. Yu, “A structural equation model of supply chain quality
management and organizational performance,” Int. J. Production Economics, 96, pp. 355-365, 2005.
J.R. McGuigan, W.J. Kretlow, & R.C. Moyer, Contemporary corporate finance. South-western CENGAGE
Learning, 2009.
M. E. Qcal, E. L. Oral, E. Erdis, & G. Vural, “Industry financial ratios-application of factor analysis in Turkish
construction industry,” Building and Environment, vol. 42, pp. 385-392, 2007.
H. Saranga, & R. Moser, “Performance evaluation of purchasing and supply management using value chain DEA
approach,” European Journal of Operational Research, vol. 207, pp. 197–205, 2010.
S. Y. Shon, H. S. Kim, & T. H. Moon, “Predicting the financial performance index of technology fund for SEM
using structural equation model,” Expert Systems with Applications, vol. 32, pp. 890-898, 2007.
F. M. Tseng, Y. J. Chiu, & J. S. Chen, “Measuring business performance in the high-tech manufacturing industry: A
case study of Taiwan’s large-sized TFT-LCD panel companies,” Omega, vol. 37, pp. 686-697, 2009.
W. Wen, Y. H. Chen, & I. C. Chen “A knowledge-based decision support system for measuring enterprise
performance,” Knowledge-based system, vol. 21, pp. 148-163, 2008.
417

APPLYING THE CONCEPT OF FAIR VALUE AT BALANCE SHEET ITEMS. THE CASE OF
ROMANIA
Marinela – Daniela Manea, Valahia University of Targovişte, Romania
Post-doctorate researcher at Alexandru Ioan Cuza” University of Iaşi
Email: marinelamanea7@yahoo.com
Abstract: In the last period, in what assets are concerned, it has been agreed to set certain payments (particularly those regarding
immobilizations) at the estimated value of future flows of incomes provided to entities – recoverable or covering value – rather than at
past expenses expressed at their historical cost, a fact which is synonymous to a considerable radical change in accounting. According
to such a perspective, the objective of accounting is that of going deeper into and analyzing future, precisely for transposing it into
balance sheet items.
Under the current circumstances, which involve the scientific research within the filed, even if the Romanian accounting practice has
registered a significant normative progress, by taking over, sometimes up to the point of identifying itself with them, the concepts and
definitions existing within financial report standards, Romanian norms do no not offer too many alternatives, by preserving their
character of regulations strictly explained, a fact which does not provide the possibility of choosing an accounting remedy or applying
a policy established through a professional judgement correlated with the normative requisites.
The current research focuses on the way that just value concept is applied, by acknowledging that there has not been yet identified a
communication alternative between accounting norms and policies, between the freedom to choose accounting procedures and the
duty to provide relevant credible information to users.
Keywords: fair value, annual financial statement, tangible fixed assets
JEL classification: M 41
1. Introduction:
Internationally, two sets of accounting rules have the supremacy nowadays: the international ones, developed by
ISAB, and the American ones, developed by FASB. IASB has accepted, in its own way, FASB’s rules’ influence by
participating at joint projects with FASB, whose purpose is to limit the differences between IFRS and US GAAP. It
must be noted that fair value’s measurement is one of the convergent issues between the two sets of rules, IASB
having already published international rules of financial reporting, IFRS 1 to IFRS 8. From these, the American
rules’ influence is obvious within IFRS 5 “Nonconcurring assets held for sale and discontinuous operations”,
respectively IFRS 3 “Entities combinations”. ISAB has not yet issued a rule exclusively dedicated to fair value, but
only limited to bringing a few changes to IAS 16 “Tangible properties” and developing IFRS-es requiring the use
of fair value, without detailing how to obtain them. If initially IASB was equidistant from the prescribed evaluation
bases, in recent years, it was requested a stronger reliance on fair value and, therefore, its extensive employment.
International accounting rules are becoming more and more future-oriented, seeking to favor information users’
predictions. Thus IASB once indicated that the fair presentation of the results and of an entity's financial position is
more important than applying a set of rules.
2. Applications of fair value in the discovery of financial statements elements
Representations of the fair value for the recognition of patrimonial items, respectively for later notice can be at least
the following [Deaconu A.] applications:
� coverage of intangible fixed assets with a potential buyer, starting with the second year of holding in entity, at
fair value;
� recognition of tangible/intangible imobilisation depreciated either at the utilisation value or at the net selling
price;
� for the tangible/intangible immobilization made as a contribution to the social capital, reflection in accounting
is realized at their market value;
� immobilized tangible/intangible assets from exchange transactions are evaluated at the market value,
respectively based on the best estimates, where market information is insufficient;
� immobilized tangible/intangible assets from entity combinations are recognized in the patrimony at market
value;
418

� measurement and reflection in accounting of the residual value of tangible/intangible immobilization is realized
at market value;
� tangible assets’ recognition, such as special equipment, rarely traded on active markets at replacement cost –
representation of fair value in accounting;
� both initial and subsequent recognition of tangible immobilization held for sale at the lowest value from book
value and the fair value diminished with the sales’ cost;
� subsequent recognition of property investments, in agreement with the IAS 40 conditions “Property
investments” at market value;
� initial registration of the goods acquired through leasing at the lowest value from the good’s market value and
the updated minimal leasing value;
� initial recognition of long-term debts at their updated, market value;
� the subsequent recognition (at the end of the financial exercise) of the goods stock, products, merchandise etc. is
performed at the net realizable value (market value), precisely for identifying any value losses/reappreciation;
� at the initial recognition (at the harvest time) of the agricultural production stocks, related to the biological
assets, their market value it is used;
� initial recognition of commercial debts related to the current activity’s income, in accordance with the IAS 18
conditions “Income” is obtained at the fair value – updated of the sums which are to be received;
� at the financial exercise’s end, subsequent recognition of debts and commercial obligations denominated in
foreign currencies is achieved at their updated value;
� the debts related to the employee’s benefits are initially recognized at the corresponding value of the provided
service, respectively at an updated value if the contributions are payable within one year after the accounting
period when they have been rendered;
� at the initial recognition, long term liabilities are evaluated at their settlement value (reimbursement value) or at
the updated value;
� subsequent recognition, at the financial year-end of cash denominated in foreign currencies is achieved at the
value obtained from the sums in foreign currency.
3. The application of fair value in Romania
For Romania the fair value is still new. It is difficult for the professional accountants to clarify it on a conceptual
level – as it will be seen from the research performed in this material – but even more problematical is its practical
application. Regarding this matter, there can be noticed an attitude of reticence justified in the academic and
professional environment in connection with the possibility of introducing an accounting system including the fair
value, or, why not, based on the fair value. What could be the excuse for this behavior?
Mainly, the system focused on the fair value as a management representative based on value, is directed towards
maximizing the shareholders and creditors’ fortune. Under the circumstances of the Romanian instable economical
environment, where the bankruptcy phenomenon is manifesting strongly for too many entities, does the concern of
protecting these shareholders and creditors even exist? Since he brought funds as well, will the minority shareholder
be granted protection and will his economical interests be pursued? Is it possible, in the Romanian economy, to
obtain an accounting support guaranteeing to the minority shareholder the profitability corresponding to the invested
capital and the risks he has taken? Here are so many questions that must be answered by the Romanian economical
environment.
Carrying on with the idea of the minority shareholder’s protection, the Romanian accounting should find the
best solution for evaluating the patrimonial items mirrored in the financial situations at the lowest cost from the
historical cost and the fair value. At the time being, in Romania, at least on a conceptual level, a combination of the
two systems has been adopted – historical cost and fair value, with an important significance for the former. Beyond
the conditions of the accounting regulations, putting into practice the system based upon the fair value remains a
necessity rather than an actual fact.
How did this situation come to be, actually? One possible explanation would be that, for the moment, too many
hindrances are restricting the applicability of the concept, such as:
� a profound study and theoretical reflection rather poor for the alternative methods of evaluation in accounting;
419

� the insufficient theoretical development of the fair value concept by not knowing in practice its succesive
modelling stages;
� the scarcity of information on the market, as a result of the imperfect economical conditions;
� the opacity and lack of vision of the professional accountants who are not willing to accept the change of a
familiar, easy to work with evaluation system, with another, more complex one that requires alterations and
estimations difficult to achieve;
� the Romanian present accounting system’s reduced ability to apply evaluation at a fair value, which requires
specialised professionals – usually, the fair value is the evaluators’ attribute.
�
Concerning the last point, only a few entities can afford hiring a professional evaluator, respectively creating a
specialized department because this implies a sustained financial effort. Using outer consultants is not an option for
Romanian companies either, especially during this crisis period. This is why, in our opinion, if professional
accountants would have the necessary expertise, they could successfully shape the fair value, at a minimum cost for
the entity for which they are providing accounting services. Additionally, ensuring the estimation’s objectivity can
be guaranteed through the ethical standards provided by an accredited organism to which that specific professional
accountant has adhered.
3.1 Applications of fair value for stocks
In the case of free of charge stocks – either by finding something extra on the inventory or in the case of receiving
it as a donation – evaluation presupposes the employment of fair value at the entry moment, materialized in the
market price. Although not specifically mentioned Romanian accounting regulations have used, throughout the time,
several names to designate the fair value, like: utility value (Accounting Law, nr.82/1991) and market price or other
value (OMFP nr. 1752/2005). For the assets received through donation, both for the Romanian regulations and for
the international ones, there is a consensus regarding the use of any form of fair value.
For the stocks acquired at a global cost, specific for coupled elements, it is necessary that every good is
individually attributed the aquisition/ production cost, proportional with the market/net realization value, at the tiem
of the evaluation.
To obtain the net realization value of inventories at the end of the financial year, we will take into consideration
the purpose for which they are held, as follows:
� for raw materials – meant for consumption – it is recommended to establish the cost of replacement;
� for finished products, semi-finished goods and merchandise – meant for sale – it is right to determine the market
value and the deduction of the items related to the amount of net realization such as: commercial discounts,
administration and selling expenses to be made, production costs to be made for products in progress.
3.2 Application of fair value for intangible assets
In case of intangible assets, the fair value is shaped in the following situations: in case of a value depreciation, in
case of review, when determining their residual value, at the time of the exchange of assets, in business
combinations.
Impairment of intangible assets is currently inlcuded in the general rules applicable to all assets, relinquishing
the completion required by IAS 38 „Intangible assets”, which stated that the entity should estimate the recoverable
amount at least once a year, even when there were no indications of impairment for assets not available for use,
respectively for those amortized for a period longer than 20 years. In the curent version, for each intangible asset the
entity will compare its recoverable amount with the book value at the end of each financial year, and whenever there
is an indication of a likely loss of value. Also, for assets with indefinite life period IAS 36 the ”Impairment” requires
annual testing, regardless of wether there is any indication of impairment.
Revaluation of intangible assets, as an alternative treatment for subsequent evaluation, goes through the same
steps applicable to tangible assets. Things to be considered: estimating the fair value of the asset, the correction of
accumulated depreciation and accounting of the surplus or decrease value as a result of the re-evaluation. Problems
arise in identifying an active market for such assets beacuse they are either unique elements – such as trademarks,
420

patents, film publishing rights, or although they can be traded, such operations are rare and negotiated irectly
between buyer and seller, and prices are not generally available.
The most appropiate method for revaluation of the intangible fixed assests is the one of the comparison market,
especially for those assets for which there is a potential buyer or a specific market – certain software products,
concessions, certain types of fishing licenses, taxis, production quotas.
For intangible assets estimating the residual value is rarely possible and due to their specific nature – being
dedicated just to the current user, respectively a very long amortization period (20 years and even more) in which it
is used economic. It is usually considered by international normalizers, for intangible assets, a zero residual value.
But there are intangible assets such as software products, with a shorter life period and who can resist morally, in
such way as to provide utility at the end of their useful life. In this cases it is estimated that the residual value is a
fair value determined through the market comparison approach. The starting point in estimating the residual value is
the selling price of similar assets (worked in similar conditions and had a similar life period with the evaluated asset)
minus the estimated sale costs. The value judgement is currently made in relation with the expected age and
condition for the asset at the end of its useful life. In other words, not using changes of price or of other variables
that would develop during the useful life time, but actual values. It should be noted that the estimations are not made
for the present time, but for a future time, which is why the residual value will periodically reshape, depending on
the evolution of market conditions speific to the asset.
For the purchase of intangible assets in return for the surrender of non-monetary assets, combined or not with
monetary asset, is accepted by the international accounting standards fair value measurment. Exceptions to this rule
are the following cases:
� the exchange transaction has no commercial character, wich means that future cash flows generated from the
activity does not change as a result of the exchange transaction;
� the given asset is, in turn, valued at the fair value;
� can not reliably measure the fair value, neither for the received asset, nor for the given asset.
The most appropiate method of measuring the fair value appears to be the market comparisons method [Ristea M.,
Dumitru C., Irimescu A.], but where information about transactions with similar goods are not enough valuation
models are applyed. Also, if you can reliably measure the fair value of both assets subjected to the exchange, it is
advisable o use the fair value of the given asset. Unless it is clear that the fair value of the received asset is more
well-argued, prefer the latter.
The current Romanian norms account for the exchange of assets as a simultaneous sale-purchase operation,
although the final outcome is zero.
For entries in the entity´s assets of intangibles from business combinations, their fair value recognition from the
entry date is applyed. One can estimate the fair value of intangible assets resulting from a business combination, in
most cases, except those assets that are not separable but whose fair value can not be reliably estimated, given the
existence of several variables that can not be evaluated. As the evaluation methodology , is preferable the market
comparison method – when there is an active market and the actual prices are known, or indirect estimation method
based on income. When none of the above-mentioned evaluation techniques do not provide a credible assessment of
intangible costs resulting from a combination of business the asset will not be recognized, but included in the
goodwill.
3.3 Applications of the fair value for tangible assets
In the case of tangible assets the fair value is shaped in the following situations: when noticing value depreciation,
at revaluation, when determining their residual value, when exchanging assets, within leasing contracts.
IFRS treats the the tangible assets’ depreciation through the IAS 36 standard „The assets’ depreciation”,
specifying the necessity of going through the depreciation tests’ steps depending on the internal and external
existing clues available for the entity. When measuring the recoverable value, the entity calculates two values: the
net fair value and the utility value, and then proceeds to the comparison of the two aggregates, choosing, in the end,
the biggest of the two.
421

The net fair value or the net selling price is, in fact, the market value reduced with the cost of sale. For those
productive tangible assets such as machineries and equipment, more difficult to trade in active markets, there must
be carried analogies with recent transactions of the entity for similar assets, compliance with reasons that will take
into account variables such as production capacity, the operating state, the age etc.
If the asset can be traded on an active market, the fair value is determined by reference to the market value of
the tangible asset, the best clue being considered to be the price of a strong sale commitment registered after a
market transaction. The disposal costs are deducted from the market price. As disposal costs there can be mentioned
the legal costs, the notary fees related to the sale, stamp taxes and similar charges, postal charges, costs meant to
ensure the asset’s good condition for sale, travel costs etc.
For the assets traded on the market for which there is no strong sales commitment. The starting point for the fair
value’s measurement is the asset’s market value, by analyzing the recent selling prices (or the sale offers) of fixed
assets identical or similar to the analyzed one, in order to reach an indication about its value. The method is known
in evaluation under the name of sales comparison approach. Because it is usually difficult to find comparable assets
identical to the one involved, there must be applied corrections to the prices of the fixed assets similarly sold to
ensure their comparability on the account of the differences between their essential characteristics called elements of
comparison.
The elements on which the comparisons and adjustments are made are taken from the market and they mirror
what the buyers believe to be the causes for the price differences they are willing to pay.
If comparable assets are superior to the one analyzed in a certain characteristic of theirs, then their price is
corrected downwards. Vice versa, if the prices of these assets are inferior to the analyzed one, then a positive
correction will be made. Ideally, when such configurations and evaluations are made, the conclusions ought to be
based on sales of similar or identical assets (usually similar) which have been traded on the market.
Unfortunately, sales of assets that are identical to the analyzed one are very rare. Practically, the market analysis
will reveal sales of similar and not identical assets [Manea M.], and this represents the similarity analysis on which
the authorized specialist bases his opinion upon value. As comparison elements, we can mention:
� the origin and the actual age (the shaping of the actual age of the comparable will be attempted) ;
� the state (condition): It is known that the differences from the asset’s state (condition) affects the selling price of
the similar assets;
� the capacity: Ideally, the comparable assey should have the same production capacity (or a similar one) with
that of the analyzed asset. On a contrary case, there might be imposed the correction of the comparable asset’s
selling price to reflect the capacity differences.
� the characteristics (accessories): The evaluator specialist should compare the analyzed asset with assets that
show the same characteristics and accessories;
� the location: the geographical location of a comparable asset’s sale can affect the selling price. Additionally,
even an asset’s physical location, within an installation can influence the selling price;
� the producer: Where it is possible, the assesor specialist should perform the comparison of the targetted asset to
the sales of similar assets realised by the same producer;
� the parties’ motivation: this represents an important comparison problem, especially for the big machineries.
The specialist assesors should analyze and understand tboth the buyer’s and the seller’s motivation and in what
way the analyzed asset’s value influences this motivation;
� the price: In many cases, especially for big properties, the transaction’s price should be investigated and
expressed in monetary terms (cash);
� the quality: the comparable asset’s quality should be equivalent to that of the analyzed immobilization.
Otherwise, the specialist should either give up on the comparable asset or perform an adequate correction;
� the quantity: The unit prices can vary considerably depending on the sold quantity, which will also be correlated
with the market conditions: a buyer’s market suggest some important quantity’s availability, while a seller’s
market suggest a limited quantity;
� the selling date: The specialist assessor should obtain pieces of information about the sales recorded within a
reasonable period of time from the analysis’ date. This is important especially in the cases of unstable markets.
Theoretically, comparable sales should be close to the analysis’ date, but they are not always easy to obtain.
When this kind of sales is recorded off the „reasonable” period if time, but they must be taken into account, the
422

specialist should justify this and make the adequate corrections because the pieces of information have a low
degree of interest;
� the sale type: the sale’s type and the terms indicate different price levels belonging to various ways of
commercializing (therefore, the value premises).
Generally, the sales comparison method can be used for any of the value’s premises, for any commercializing
form. If the prices from before the beginning of the transaction would have been announced, the result could be the
market value. If auctions would be considered a basis for comparison, the result could be forced liquidation value.
Sales comparison approach or the modeling technique is not applicable when the asset involved is unique. Even
in the situation in which the asset in not unique, the approach is not applicable unless there is an active market for
that element. An inactive market, or one where there is a limited number of sales comparable with the asset
involved, often show a lack of demand as the existence of economical depreciation: where there is an inactive
market, the analyzed element could more adequately be evaluated through the analogy method in accordance with
the paragraph’s 27 regulations of IAS 36 norm: “The assets’ depreciation”. Consequently, it is possible for the fair
value to be determined even if the asset is not traded on an active market. In these cases, the entity will analyze the
available pieces of information regarding the possible past transactions, with similar assets, for which the market
selling prices are known; likewise, if there are offers for similar assets and the prices are situated at approximately
equal values, there can be made an estimation for the fair value without the selling cost. This way, especially for
those productive tangible assets as machineries and equipment that are more difficult to trade on active markets, it is
necessary to carry out analogies with recent trades of the enterprise regarding similar assets, compliance with
reasoning that will take into account variables as production capacity, operating state, age etc.
Returning to the sales comparison approach, pertinent questions pop up: how is market information obtained and
where is it obtained from?
How market information is obtained: one method of getting market information is contacting the used
equipment’s dealers who are familiar with the type of equipment that is analyzed, in order to find out recent selling
prices or current solicitated prices. It is preferable to make the relation with more than one dealer, in order to assure
a certain information coherence.
Where can one get market information:
� used equipment dealers;
� new equipment dealers;
� sales brochures;
� newspapers commercials;
� private sales;
� auctions.
New and used equipment dealers are good information sources. If a customer bought a mechanical device from a
used equipment dealer, this is a good piece of information. Even more, that dealer can be very cooperative about
providing information. The specialist assessor should ask the customer’s permission to contact the dealer who sold
him the equipment, because the customer could be against communicating the piece of information about the sale,
especially if the dealer has connections with his competitors. The pieces of information obtained from the market
can be uses by the specialist in shaping the analyzed asset’s value; in its simplest form, sales comparison method
representation presupposes the relation:
The immobilized comparable asset’s price +/- corrections = The analyzed fixed asset’s price
For determining the price more techniques and methods can be employed:
� the identification method;
� the assimilation method;
� the cost percentage method.
The identification method. This technique establishes an asset’s value through comparison to an identical
replacement that has a known selling price. A possible example can be the cost for a fork manufacturing lift, with
known model, age, capacity and condition. In this case, specialized guides are usually used.
423

The assimilation method. This technique establishes the value relying on the analysis of some assets that have
key parameters close in size, but not identical (similar assets), using a utility measure (size, capacity) as a
comparison basis.
The cost percentage method. This technique only implies establishing the ratio between the selling price and the
current gross cost of an asset on sale date. With sufficient information, a specialist can make statistical analyses and
he can establish relations that have appeared on the market, between age, selling price (or the demanded one) and a
new one’s price.
Tangible fixed assets revaluation process translates into determining the fair value by the assessment experts.
Romanian rule does not elaborate on what are the methods for obtaining fair value. It just points out that if fair value
can not be determined because of lack of active market, that asset should be recognized at its cost or, where
appropriate, the last reviewed value (less the cumulative value adjustments). As a value type to be expected when
reviewing tangible assets, we believe that the use value should be determined (income approach) or net replacement
cost (cost approach). Only in the event that management intends and argues the sale of property, we can get market
value (market comparison approach).
The residual value as a possible sale price of the asset at the end of his utility period in the entity is thus a
market value. Assessor specialists will quantify the size of the allegedly selling costs operation will be deducted
from the market value. As we can not currently know from the market the price of similar future transactions, in
shaping the market value we will not use the approach method based on market comparisons, but we will use either
the income approach or the cost-based approach.
If the income-based approach can be applied in a credible manner – but customizing the value to the current
operating conditions of the asset and to its current user, but not taking into account the possible acquisition by a
future user and an alternative use, cost-based approach requires modeling of variables difficult to quantify. It is
about the reconsideration of direct and indirect cost components of the fixed asset at the time of its sale – modeling
requires information difficult to quantify and which lead us to the idea that no approach is perfect.
As in the case of intangible fixed assets, the exchange of tangible assets is treated in Romanian accounting as a
sale operation, disregarding the requirement of applying the fair value if the transaction has a commercial nature,
with the modeling option for the fair value of the given asset rather than of the received asset.
In terms of leasing contracts, between the two types of contracts – financial and operational, only the first one
involves news about the applications of fair value.
If operational leasing generates costs to the tenant, respectively incomes to the tenant – both are recognized in
the profit and loss account, in addition the leased asset remains in the lessor´s stock and will be applyed to it all its
assessments tangible fixed assets, financial leasing in return involves several types of estimations included in the
scope of fair value as follows:
� fair value of the asset that is subject to the leasing contract, requiring the estimation of a market value in our
opinion because the transfer of property is only partially achieved;
� modeling a residual value of the asset for the end of the lease, both tenant and the landlord, which is also a
market value for that date adjusted with the accumulated depreciation;
� leasing gross investment consisting of both the tenant and the landlord in the estimation of the minimum lease
payments to receive;
� present value of minimum lease payments obtained by discounting the debt totaling periodic lease lease
payments and the residual value at the end of the lease. To obtain the present value of minimum lease payments
we use the implicit interest rate specific to the contract.
3.4 Applications of fair value for assets held for sale
Until the date of classification as held for sale fixed assets are valued in accordance with their specific standards
applicable at the date of entry into property or during use. From the time of their classification as held for sale, the
424

evaluative criteria is covered by IFRS 5 „Assets held for sale and discontinued operations” given by the minimum
between the net book value and the fair value[IASB].
If since their entry into heritage assets are classified as held for sale the evaluation involves comparison between
the accounting value that it would have received if it had not been classified as such (ie acquisition cost) and the fair
value minus the selling costs. For assets or group of assets acquired during a business fusion, assessment involves
modeling only the net fair value because the book value (its costs) is not relevant at the acquisition date.
In determining the fair net value between the sales costs we identify: transportation expenses, handling,
advertising, sales commissions given to the inermediaries etc. If for the materialization of the sale the one-year
period is exceeded, the costs of sale should be updated. Also, after updating the sales costs we may add other items
of expenditure that could increase that value. Increasing the size of the sales costs will not be recorded as a reduction
in asset value due to operating expenses, but as a financial expense.
3.5 Applications of the fair value for income
For income, fair value correspond to either:
� the fair value of assets/services covered by the transaction, when the entry of the counterparty transaction is not
delayed;
� the fair amount of money – cash or cash equivalent – to be received, for the case the entry of the counterpart
transaction is delayed.
For the case when the counterpart transaction entry is not delayed, the fair value is determined by agreement
between seller and buyer; for the first one, the evaluation basis that determined the sales prices is the net realizable
value, for the last one, the transaction evaluation basis is the historical cost. The two bases correspond to the
situation in wich the asset is sold.
If the counterpart transaction entry is delayed, the fair value is established – the updated value of the amounts to
be received. The present value of revenue is obtained by applying to the initial value a discount rate. This will
penalize the baseline considering the delay of its receipt. For the discount rate the international normes mention a
neutral rate – interest rate, because delaying of the commercial transaction counterparty collection is considered as a
financial transaction.
4. Testing the fair value concept application in Romanian accounting practice
To test how the fair value concept is implemented into Romanian accounting practice we started from the
consideration that shaping it has a great effect on the assets held by the entity and thus our research methodology has
focused on applying the concept in this direction. We proceeded to the achievement of applied research through a
survey of companies in the South-Muntenia region, research materialized in shaping and interpreting the results of a
questionnaire distributed to managers, and professional accountants. It sought to identify those entitiesthat use in
accounting practice the concept of fair value, taking into account the applicable law, the tax limitation that do not
motivate the professional accountants to model two sets of financial statements – one according to the tax provisions
and another taking into account the economic reality and which would result in a true accounting image.
The questionnaire distributed compiles a set of 16 questions and it is listed below:
QUESTIONNAIRE
This questionnaire is aimed at highlighting the practical aspects of applying the concept of fair value to the
recognition of those assets and preparing financial statements. The data you provide is strictly confidential and used
solely for scientific research.
I. Informations about the company
Company name:
Legal form of company: SA, Ltd., SNC, SCS, SCA
Company field of activity:
425

Type of company: JV, SME, MI
II. Financial information
Turnover: > 100.000 EUR
101.000 EUR – 500.000 EUR
500.001 EUR – 3.000.000 EUR
3.000.001 EUR – 5.000.000 EUR
5.000.000 EUR – 7.300.000 EUR
< 7.300.000 EUR
Number of employees: 1 – 9 employees
10 – 30 employees
30 – 50 employees
Over 50 employees
Total assets:
Net tangible assets:
III. Information on applying the concept of fair value in Romanian accounting practice
1. In your company is the fair value concept known by the accounting professionals?
YES
NO
2. Does your company use the contribution of professional accountants in shaping the fair value (or is there a
specialized department for evaluation?)
YES
NO
If the answer is YES, go to question number 3!
3. Who is the decision maker in your company regarding the application of the fair value aggregate in the
evaluation?
4. What goals do you have in mind when determining the fair value of various assets held by the company?
5. Do the accounting professionals from your company have enough working tools in modeling the fair value?
YES
NO
6. Do the accounting professionals in your company know the three modeling approaches of fair value: market
value, utility value, replacement value?
YES
NO
7. In determining fair value for assets does the market information has priority?
YES
NO
8. When you can not find market prices for identical assets do you collect market prices for similar assets, adjusting
them to the existing differences?
YES
NO
9. In the absence of any market information do your professional accountants use, in measuring the fair value,
assessment techniques based on results and costs?
YES
NO
10. In the last three years, was there any fair value modeling for assets hold by your company?
YES
NO
11. Please indicate the fair value quantification method by checking the appropriate boxes:
Market value
Income capitalization method
Discounted cash flow method
426

Depreciated replacement cost method
Other, ie _________________________________________
12. In determining the market value through the modeling technique, the comparable asset price was obtained by
using one of the following methods:
Identification method
Assimilation method
Percentage of cost method
Other, namely _________________________________________
13. The estate assets have been revalued in the last 3 years?
YES
NO
14. Please specify, by checking the appropriate boxes, at whose request was the revaluation of the tangible fixed
assets made?
Bank
Shareholders
State
Others, namely _________________________________________
15. Where the reassessment results recorded in the accounts?
YES
NO
16. Specify what is the reassessment process occurrence on the company´s financial statements?
5. The results’ interpretation
The following conclusions were drawn:
� The Romanian entities, through the practitioner professional accountants are familiar with the international
standards’ regulations regarding the fair value;
� From all the entities that responded to our questionnaire, 90% have applied the fair value’s concept in the
evaluation of various types of assets
� Although the methods of modeling the fair value are not concretely detailed in the Romanian accounting
standards, the Romanian entities can turn to the specialized assessors’ help, who have their own work tools for
its quantification
� In the action taken, approximately 85% of the questioned entities used external assessors in the fair value’s
evaluation process
� Considering the high costs for the sustenance of a specialized evaluation department, the Romanian entities do
not use this solution in over 80% of the cases, preferring the external assessors’ help
� Although the fair value’s modeling techniques are well known, at least theoretically, even by the professional
accountants, there must exist a period of adjustment for the decision factors in the Romanian entities in order to
realize the importance and the necessity of the specialized accountants in modeling this aggregate;
� Although there have been established fair values for the owned assets, these haven’t all been used in the
financial reports, as a consequence of the restrictive fiscal practices that limit the concept’s applicability and
recognition.
6. Instead of conclusions…
Lately, the fair value has become an important concept, defining the current accounting rules, conferring new
qualities of reflecting the economical reality regarding the entity and its assets, also contributing to the increase of
the professional accountant’s role who becomes a veritable consultant for the entity. The Romanian accounting
regulations have recorded a significant progress by taking over, sometimes up to identification, the definitions and
conditions from the international accounting standards, however without realizing conciliation between accounting
rules and accounting policies, between the freedom of choosing accounting procedures and the obligation of
providing the users relevant and reasonable information. Although the application of the fair value’s concept in the
427

Romanian accounting practice represents a novelty, because this is still the expert assessors’ attribute, who have
sufficient tools and modeling techniques, we believe that in the near future the fair value’s measurement will be
attributed to the professional accountant, much better grounded in the entity’s decision process.
7. Acknowledgement
This work was cofinanced from the European Social Fund through Sectorial Operational Programme Human
Resources Development 2007-2013, project number POSDRU/1.5/S/59184“Performance and excellence in
postdoctoral research in Romanian economics science domain.”
8. References:
1. IASB, International Financial Reporting Standards (IFRS), Official rules issued on January 1 st 2009, translation,
CECCAR Publishing, Bucharest, 2009;
2. Deaconu A., The fair value. Accounting concept, Economical Publishing, Bucharest, 2009, chapter 5, pp. 200-
330;
3. Feleagă N., Feleagă L., Evaluation models and regulations in international accounting, Journal of Theoretical and
Applied Economics;
4. Manea M., Measurement and evaluation of depreciation to the assets, Academic Book Publishing, Bucharest,
2007, pp.30-34;
5. Ristea M., Dumitru C., Irimescu A., Manea L., Nichita M., Sahlian D., Policies and accounting treatments of
assets, Economic Tribune Publishing, Bucharest 2007, pp. 142-144;
6. Minister Of Public Finance’s order nr 3.055/2009 for the approval of accouting regulations in conformity with
European Directives, published in the Official Gazette nr 766 bis/10.11.2009
428

THE IMPACT OF THE INTELLECTUAL CAPITAL ON THE COST OF CAPITAL: BRICS 1 CASE 2 .
Elvina R. Bayburina & Alexandra Brainis, National Research University – Higher School of Economics, Russia
Email: elvina.bayburina@gmail.com, abrainis@yahoo.co.uk, http://en.cfcenter.ru/
Abstract. Intellectual capital is an important value driver to form the sustainable corporate performance and reduce stakeholders’
risks. Both companies and capital markets realize that there is a wide range of decisions to be made to improve the control and
reporting systems currently being used. Weighted average cost of capital also plays a great role to access and to regulate the corporate
risk. Although the impact of the Intellectual capital (IC) on the cost of capital is significant and this fact is investigated and accepted,
however there are few researches and investigations on the subject. The research conducted in this paper is quite unique and very
essential, as special evidence to the internal managers and owners of the companies why the growth of IC expenditures can reduce the
cost of capital of the company. The considered time period is 5 years and the considered sample includes companies from BRICS
(Brazil, Russia, India, China, South Africa). It was statistically proved that some of the components and subcomponents of the
Intellectual capital have a significant influence on the corporate cost of capital, as well as the special conditions of the crisis period
and special features of some industries. These findings should definitely help the “insiders” and “outsiders” of the company to create
decision-making schemes with less risky and random strategic solutions.
Key Words: Intellectual capital, financial architecture, innovations, sustainable competitive advantage, emerging markets, BRICS
JEL: G32, G34, G35, L21, L26, M14, M51, M52, O31, O32, O34
1. Introduction
The modern economy has been utterly transformed in recent years. The production notions and means have had to
be totally revised. To survive and to succeed each company now possesses Intellectual capital which must be well
managed and exploited in order to succeed (Stewart, 1997). Intellectual capital is an important value driver in
today’s companies. Traditional financial statements do not provide the relevant information for managers or
investors to understand how their resources – many of which are intangible – create value in the future. Intellectual
capital statements are designed to bridge this gap by providing information about how intellectual resources create
future value. Intellectual statements can be used as tools to communicate over the knowledge-based strategy
externally and it is obvious that it can be significantly helpful in creating internal management decisions. Both firms
and capital markets realize that something has to be done to improve the control and reporting systems currently
being used. The vast majority of the scholars investigating the Intellectual capital argue that the financial reporting
system is incapable of explaining “new” resources such as relationship, internally generated assets, and knowledge.
Disclosing information on such factors is likely to lower the cost of capital because it decreases the uncertainty
about future prospects of a company and provides a more precise valuation of the company (Botosan, 1997). Below
it can be found several reasons why the disclosure on Intellectual capital information is greatly important for the
capital market situation:
1) decreased asymmetry of information, the problem of exploiting the insider information available only for the
internal managers whereas the outstanding shareholders have no idea of its existence. (Lev, 2001);
2) decreased cost of capital (Lev, 2001);
3) increased engagement; the minor shareholders may be encouraged because they get the access to the
information on intangibles which was presented only in the meetings with larger investors (Holland, 2001);
4) decreased volatility and the danger of incorrect valuations of firms, which leads to investors and banks placing
a higher risk level to organizations (Mouritsen et al., 2004).
As it was mentioned above Intellectual capital(IC) leads to decrease the corporate cost of capital. It can be
explained by the positive effect of an open strategy on the better assessment of future wealth creation capabilities
(Burgman & Roos, 2007). Furthermore, “new sources of wealth, not readily identified in the value chain” such as
1 Brazil, Russia, India, China, South Africa. On 24th of December 2010 South Africa received an official invitation to enter BRIC. The entrance
has previously been approved by members of BRIC, however, according to many experts, South Africa is not on the level of other BRICcountries
by some economic measures.
2 The results of the project “Researches of corporate financial decisions of the companies of Russia and other countries with emerging capital
markets under conditions of global transformation of the capital markets and formation of economy of innovative type”, carried out within the
framework of the Programme of Fundamental Studies of the Higher School of Economics in 2011, are presented in this paper.
429

value networks, value shops, and the matching business risks force companies to find alternative ways to disclose
this information in a credible way (Burgman & Roos, 2007). Taking into account the above mentioned it is obvious
that the firms spending more resources on IC enjoy a higher market capitalization and lower effective cost of capital.
But one question still remains unanswered: how to define the Intellectual capital in the financial statements and
public information about the firm? Therefore, it is essential to make a brief overview of the articles dedicated to the
Intellectual capital investigation.
2. Intellectual capital overview
2.1 Background approach
It is difficult to provide precise definitions for intangible assets and IC. Thus, the definitions found in the literature
are decidedly broad. According to Stewart IC can be defined through its three main components: human, structural,
and customer capital (Stewart, 1997). “Human capital is the capacity of individuals to provide solutions for their
customers. Structural capital transforms know-how into the group’s property. Customer capital allows relations with
customers to be perpetuated”. However, a single definition of IC adopted by one company may not generalize to
other companies because IC is closely tied to the industry and the specific company it serves. Other definitions in
the literature include Moore who defines IC as customer capital, innovation capital and organizational capital
(Moore, 1996). A more comprehensive and generic framework is presented by Brooking (Brooking, 1996). This
framework has the following categories of IC:
1) market assets (consisting of service or product brands, backlog, customer loyalty, etc.);
2) intellectual property assets (patents, know-how, trade secrets, etc.);
3) human-centered assets (education, work-related knowledge, vocational qualifications, etc.);
4) infrastructure assets (management philosophy, corporate culture, networking systems, etc.).
Brooking’s (1996) framework was modified by his followers (Guthrie et al., 2003). The updated version of IC
has three categories (components) and 18 subcomponents. The three IC categories are:
(1) internal capital with six subcomponents (e.g. intellectual property);
(2) external capital with seven subcomponents (e.g. brands); and
(3) human capital with five subcomponents (e.g. training).
In addition, it is highly crucial to pay attention to the definition of the Intellectual capital and its components
given by Mouritsen et al., (2004). It is said in his study that the intellectual resources comprise of the firm’s
knowledge. In a business context this knowledge is used to improve a firm’s innovation capability, processes and
performance. Below the authors define four types of knowledge resource, i.e. employees, customers, partners,
virtual infrastructure and technologies.
1) Employees and knowledge resources with inherent attributes such as skills and personal competences,
experience, educations, commitment, or willingness to adapt. Groups of employees produce beneficial emergent
qualities.
2) Knowledge resources based on customers and partners. Especially the relationship to customers, users, and other
partners such as suppliers, their satisfaction and loyalty, their referral to the company, insight into users’ and
customers’ needs and the degree of co-operation with customers and users in product and process development.
3) The virtual infrastructure can knowledge resource as it includes procedures and routines. These can be the
company’s innovation processes and quality procedures, management and control processes and mechanisms for
handling information.
4) Technologies are knowledge based assets as they refer to the technological support of the other three knowledge
resources. Focus is usually on the company’s IT systems (software and hardware) such as Internet, IT intensity, IT
competencies and IT usage.
After the detailed examination of the papers on IC defining it was decided to use the definition and the structure
of the IC components given by Bayburina and Golovko (Bayburina & Golovko, 2008). Intellectual capital is the
aggregation of the key qualitative, intangible by the meaning characteristics, which give an opportunity to have the
430

competitive advantage and provide a distinguishing valuable offer. This definition gives the broadest understanding
of IC because it includes the commonly used term of only intangible assets included in. Thus, Intellectual capital is
a complexity of knowledge, accrued experience of employees and intellectual property. It is important to mention
that not all objects of engineering design, marketing mix tools, customer relationship principles are the objects of the
intellectual property (Bayburina, 2007; Bayburina, Golovko, 2008). To define IC components it was assumed to use
the classification represented in the research papers by Ivashkovskaya and Bayburina, Bayburina, Bayburina and
Golovko (Ivashkovskaya and Bayburina ,2007; Bayburina, 2007; Bayburina & Golovko, 2008,2009,2010). These
components are as follows (the certain subcomponents are enclosed in brackets):
1) Human capital (personnel’s knowledge and skills, their key competences and loyalty);
2) Process capital (company’s infrastructure – IT systems, business processes, operational processes, the plan of
changing the organizational structure, launching and implementation of the basic and auxiliary business processes);
3) Client capital (whole set of the client relationship characterized mostly by the loyalty and also by the activities
directed to improve the relationship, e.g. performing the environmental protection, servicing and settlement of
disputes, personal data protection and confidentiality guarantee etc.);
4) Innovation capital (development and implementation of the new technologies, investments in the fixed assets
etc.);
5) Network capital (aggregation of the company relationships with the outstanding shareholders, such as banks,
investors, creditors, suppliers, government).
2.2. Measuring cost of capital
Weighted average cost of capital also plays a great role in the company’s life. Globally, there two ways of
implementation of the cost of capital theory: to attract more investors and to get as precise calculation of the existing
and future projects as possible. Although the impact of the cost of capital is significant and this fact is broadly
accepted, concerning Intellectual capital there are very few investigations on this theme. Weighted average cost of
capital is a calculation of a cost of capital of the company according to which each source of capital is
proportionately weighted. All capital categories (equity and debt), such as common and preferred stock, bonds and
different types of long-term debt are weighted according to their market value and calculated in WACC. Ceteris
paribus WACC (corporate cost of capital) is to some extent the measure for evaluation of corporate risk. The
WACC equation is the cost of each capital component multiplied by the market valuation of weights and calculated
as follows:
, where:
– Weighted average cost of capital,
– cost of equity,
– cost of long–term debt,
– market value of equity of a company,
–market value of long-term debt of a company,
– market value of the company. The meaning of this variable is calculated as follows: ,
– weight of market value of equity,
– weight of market value of long-term debt,
–corporate tax rate.
Most of the researches are dedicated exactly to the cost of equity. This fact can be explained by the practical
usage of the cost of equity in everyday calculations, whereas the cost of debt remains the same most of the time.
While observing the cost of equity the scholars have come to some interesting conclusions. Botosan (1997), for
example, reports that firms with a low analyst following benefit from lower costs of equity when their disclosure
levels are higher. It was extended and proved after by Botosan and Plumlee in 2002 (Botosan & Plumlee, 2002).
Botosan examines the association between disclosure level and the cost of equity capital by regressing firm-specific
estimates of cost of equity capital on market beta, firm size and a self-constructed measure of disclosure level.
Overall, empirical research generally documents a negative association between public disclosure and the cost of
equity (Frankel et al., 1995; Welker, 1995; Healy et al., 1999; Lang and Lundholm, 2000). There are also some
investigations related to the influence of the other factors, such as market timing (Song, 1993), information
environment (Willis, 1991), intangible investments (Shangguan, 2005), legal institutions and securities regulations
(Hail, Leuz, 2005), corporate social responsibility (El Ghoul, Guedhami et al. 2010) etc. But none of the scholars
431

provides the research on the influence of the IC on the cost of capital, while conducting the similar researches on IC
impacts on the Market Capitalization, Intellectual Enterprise Value (Bayburina & Golovko, 2008, 2009, 2010), etc.
Following Bayburina and Golovko, Intellectual Enterprise Value, IEV, is the difference between Market value of
company’s Equity and the Book value of its Equity (Bayburina, Golovko, 2008). The calculated delta reflects the
excess of the Market capitalization over the Book value, which on the panel data of over 5 years help to eliminate
short-term speculative reactions; this delta also reflects how “the market” (stakeholders and mainly investors)
measures the level of the company’s internal efficiency (Bayburina, Golovko, 2008). By this pattern the value is
generated by its intellectual part. So, Bayburina and Golovko conclude, that the Intellectual value of the company is
a part of the total value, created through the process of the intellectual components’ accumulation (Bayburina,
Golovko, 2008). And moreover this value can be traced by the external stakeholders of the company. The models
developed in this study showed the significant effect of the IC components on the estimating variables (Bayburina,
Golovko, 2009; Abdolmohammadi, 2005; O’Donnell, O’Regan, Coates, 2000; Brennan, Connell, 2000; Petty,
Guthrie, 2000, etc.). But taking into consideration IC impact on the cost of capital it is possible to assume with high
percentage of confidence, that the research conducted in this paper is quite essential, to make the evidence to
managers and owners of the companies why IC expenditures can decrease the cost of capital and the corporate risk.
3. Hypotheses
Authors of the research designed the following hypotheses.
Hypothesis 1. Negative influence of human capital components on the company’s cost of capital. Human capital
refers to the accumulated value of investments in employee training, competence, and future. The term focuses on
the value of what the individual can produce; human capital thus encompasses individual value in an economic
sense. Human capital can be further sub-classified as, the employees’ competence, relationship ability and values.
Employees of the company can be regarded as the most valuable asset, what was fully proved in the time of crisis,
because of the employees’ ability to generate the ideas, keep the knowledge and forecast the internal changes
because of the deepest understanding of the company’s internal environment. As the indicators for the human
capital it was decided to include the number of employees, the personnel expenditures and the usage of assets,
which was calculated as the book value of total assets divided by the number of employees. It was supposed that this
parameter characterizes the efficiency of exploring the assets by the employees of the company.
Hypothesis 2. Negative influence of the process capital components on the company’s cost of capital. As an
indicator for the process capital the operating expenses were taken, because the increase in process expenditures
means to some extent, that the company has a tendency to maintain and improve the operational process, which
directly reflects the risks of company’s bankruptcy and instability.
Hypothesis 3. Direct influence of the innovation capital components on the company’s cost of capital. It is
assumed by the authors that if the company has the opportunity and bears substantial capital expenditures, the
company may have more opportunities and can qualitatively improve its production, intensifies, renews and creates
assets. That’s why the capital expenditures were taken as one of the indicators of the innovation capital. It reflects
the possibility of the company to implement innovations, its direction towards the optimization of the current assets
structure, replacement of old equipment by new more productive one. It shouldn’t be forgotten about the dividend
policy. From the one hand, the increase of dividends usually leads to a decrease of reinvestment in business, from
the other hand dividends are considered to be a signal for the market, however such a signal may have either a
positive, an improvement of the financials, or a negative nature of the influence. Companies with the higher level of
Intellectual capital tend to pay fewer dividends as the signal that they intend to implement responsibilities towards
all stakeholders. In addition the smaller dividend payout is considered as a signal of intentions to invest in the
sustainable development of the company. As the last, but not the least indicator of the innovation capital the R&D
expenditures should be mentioned. This kind of expenditures should be considered not as the current issues, but the
investments in the future improvements, potential and growth.
Hypothesis 4. Significant influence of a particular time period on the company’s cost of capital. In the time of
crisis all the key financial indicators have utterly changed, that crucially affected the indicators of the company’s
sustainability and effectiveness. Considering the panel date from 2005-2009 the impact of the time period on the
network activity and collaboration can hardly be excluded.
432

Hypothesis 5. Significant influence of a particular industry on the company’s cost of capital. This hypothesis is
mostly addressed to the second part of the empirical study where the groups of countries are considered, companies
of BRICS were divided into 2 main groups. Thus, a set of dummy-variable for each industry was developed.
4. Research model
According to all limitations and principles of the analysis the basic research model can be presented as following,
where cost of capital, WACC is the main dependent variable:
(2) WACCit=α+(ρ 1 it,… ρ n it) X IC + (β 1 it ,.….β k it )X FV +εit, where:
WACCit – a vector of Weighted average cost of capital, calculated according to the formula (1),
IC – a vector of Intellectual capital subcomponents;
FV – a vector of fundamental variables;
ε – a vector of random errors (“white noise”);
I – BRICS company index;
T – a year index;
N – a particular Intellectual capital component index;
K – a particular fundamental variable index.
4.1 Research model: fundamentals
Fundamental variables will also be included into the model as control variables, the combination of such
variables in the model will be defined during the research process. The research model includes the following
fundamental factors:
� Total revenues of the goods sold less adjustment on returns, discounts, insurance payouts, tax on sales, value
added tax (Sales)
� Book value of total assets (BVA)
� Return on book value of total assets (ROA)
� Return on book value of equity (ROE)
� Net income (NI)
� Operating income(OI)
� Earnings before interest, taxation, depreciation and amortization (EBITDA). The meaning of this variable is
calculated as follows: EBITDA=Operating income + D&A, where:
Operating income – income from the company’s operating activity;
D&A – depreciation and amortization.
� Book value of equity (BVE)
� Market value of equity, market capitalization (MC)
� Book leverage (Lev). The meaning of this variable is calculated as follows: Lev = Book value of total debt
/BVE
� Intellectual value distress (IVD). The meaning of this variable is calculated as follows: IVD = BVE / MC
� Size (LnBVA). The meaning of this variable is calculated as follows: Natural logarithm of BVA (lnBVA)
4.2 Research model: IC components and subcomponents
� Human capital
o Personnel expenses (PE)
o Number of employees (NE)
o BVA /NE
� Process capital
o Operating expenses (OE)
� Innovation capital
o Capital expenditures (Capex)
o Dividends paid (DP)
433

o Research &Development expenses (lnR&D). The meaning of this variable is calculated as follows: Natural
logarithm of Research & Development expenses
4.3 Research model: IC components and subcomponents
The influence of a particular time period was included into the model. The main distinctive feature is that the
networks capital is accumulated mostly by the interactions of the company with the external business environment.
Taking into consideration the network capital it is worth to mention that the specific of the world economy stage
influence the degree of stakeholders’ activity. Specific and intensity of the business network collaboration, M&A
type and frequency depend on each particular investigated year. The influence of the economic upturn business
activity, 2005-2007, is connected close to the possibility of creating value networks among companies. The period
of downturn and compressing of the business activity in 2008-2009, features of each industry on the development of
the company, features of each particular country on the development of the company were included into the model
to investigate. A set of corresponding dummy variables, which are proxy variables for estimation of each year
influence, was included into the model.
� D 05 – equals “1”, if the year is 2005 & “0” otherwise;
� D 06 – equals “1”, if the year is 2006 & “0” otherwise;
� D 07 – equals “1”, if the year is 2007 & “0” otherwise;
� D 08 – equals “1”, if the year is 2008 & “0” otherwise;
� D 09 – equals “1”, if the year is 2009 & “0” otherwise;
A set of corresponding dummy variables, which reflect industry influence, was included into the model.
� DBM – equals “1”, if the industry is «Basic materials», and “0” otherwise;
� DTel – equals “1”, if the industry is «Telecommunications», and “0” otherwise;
� DConsG – equals “1”, if the industry is «Consumer goods», and “0”otherwise;
� DConsS – equals “1”, if the industry is «Consumer services», and “0” otherwise;
� DH&C – equals “1”, if the industry is «Health & Care», and “0” otherwise;
� DO&G – equals “1”, if the industry is «Oil & Gas», and “0” otherwise;
� DFin – equals “1”, if the industry is «Financial», and “0” otherwise;
� DInd – equals “1”, if the industry is «Industrial», and “0” otherwise;
� DTech – equals “1 if the industry is «Technology», and “0” otherwise;
� DUt – equals “1”, if the industry is «Utilities», and “0” otherwise;
A set of corresponding dummy variables, which reflect industry influence, was included into the model.
� D BZ – equals “1”, if the country is «Brazil», and “0” otherwise;
� D RUS – equals “1, if the country is «Russia», and “0” otherwise;
� D IN – equals “1”, if the country is «India», and “0”otherwise;
� D CH – equals “1”, if the country is «China», and “0” otherwise;
� D SAR – equals “1”, if the country is «South Africa», and “0” otherwise;
Table 1 provides hypothetical signs of connection between the dependent variable and independent variables.
Independent variables Hypothetical sign of connection with the dependent variable
Sales -
Book value of total assets -
Net assets -
Net income -
Return on book value of total assets -
Return on book value of equity -
Operating income -
Book value of equity -
Market value of equity -
Book leverage +
EBITDA -
Intellectual value distress -
Size -
Personnel expenses -
434

Number of employees -
Book value of total assets/Number of employees -
Operating expenses -
Capital expenditures +
Dividends paid +
Research & Development expenses -
Time influence Significant
Industry influence Significant
4.4 Sample and source data.
The research conducted in this study is dedicated only to the BRICS companies because it is a widespread opinion
that the BRICS countries are the most interesting in terms of sample of emerging markets to investigate. The term
“BRIC” was launched by the economist from Goldman Sachs Investment Bank Jim O’Neill in 2003. After 2011
year conference in Davos, it was proposed to include South Africa in BRIC emerging markets group, because it
reflects all the issues related to Brazil, Russia, India and China. The BRIC countries as Brazil, Russia, India and
China are frequently referred to have attracted media and academic attention in the recent years. These countries are
different from one another over their culture, background, language, and the structure of their economies. However,
they have a common denominator: economic growth development in the BRICS has greatly exceeded growth
compared to the world's leading industrialized nations. Even after the economic crisis that started in 2007, they
continued outperforming the rest of the world. While in 2009 large economies shrunk as much as 6%, (e.g., Japan
and Germany), Brazil stayed steady, India grew 5.9%, and China 8.1%; only Russia was the group's bad performer
shrinking 7%. One more common characteristic is that the financial assets of these 5 countries are underestimated
and have a great potential of growth. The forecast of their future performance will further increase the interest on
these countries and also gives the reason to pick exactly this set of countries to develop the research in this study.
While all of BRICS countries are associated with the most fast developing markets, the potential of each country is
estimated as being at the different levels. After the crisis the scholars have divided the scope of countries into 2 parts
– “the source economies” (Brazil, Russia and South Africa) and “the pushovers” (India, China). That’s why, in this
paper it was decided to develop three models: for the 5 countries and for each group separately. Taking into account
the necessity to assess companies from different countries the final research sample was formed along with the
following criteria:
a) The sample is formed from BRICS companies listed from the year 2005 till 2010.
b) The sample is formed from BRICS companies about which there is enough disclosed data to calculate the cost
of capital, WACC.
c) The selection of such time period is caused by the length of Intellectual capital accumulation and, therefore,
the necessity to analyze panel data for not less than 5 years (Bayburina, Golovko, 2009).
d) Such research is of special interest, owing to the fact, that, from one hand, BRICS countries are leaders of
emerging economies, though, at the same time, they differ dramatically from each other. It should be mentioned,
that the information used in the study should be available in the open sources (e.g. annual reports or other
information on websites), thus making it useful and acceptable to trace for a wide range of users. In order to ensure
comparability, companies should have financial reports along with the National GAAP accounting standards. As the
results of previous studies show (Bayburina, Golovko, 2009), the number of companies, especially from India and
Brazil, which have IFRS financial reports in the period earlier than 2005, is not quite high and is not sufficient
enough for the research purposes, thus it is necessary to impose restrictions on usage of such type of reports in the
research.
e) For the purpose of comparability, the limitations on the companies’ capitalization prescribe to choose
companies with market capitalization in 2010 not less than $200mln.
f) The sample is formed from BRICS companies about which there is enough disclosed data to calculate the cost
of capital, WACC, according to the formula (1).
Statistical information for the research was collected from the Bloomberg data source and corporate web-sites.
Along with all the criteria being met the final sample was formed and it includes 273 companies from BRICS
countries, representing companies from 10 industries. To reach the main goal of the research the special research
model has been introduced and a series of linear regression tests has been held. The empirical panel data research
was fulfilled with the use of regression analysis software tools (program packages Stata 11.0, Eviews 7.0, Microsoft
Excel 2010).
435

5. Results
5.1 Data analysis
Test for normality of variables distribution, to fulfill the task a corresponding test for normality was held (Skewness-
Kurtosis Test for Normality.Data verification was held by the means of correlation analysis, special tests for
multicollinearity. Various types of OLS regression models were tested and VIF-tests for two samples were held.
Distinct models were chosen with meanings of VIF-tests not greater than critical levels, 10 for individual VIF
meanings and 6 for average meanings for group of factors according to the Stata criteria and thereon the final model
of the research was chosen. The meaning of VIF contributed to 3,13 for the chosen set of variables in the sample for
four countries. All variables of subsample “Brazil-Russia-South Africa” are normally distributed. The meaning of
VIF contributed to 1,27 for the chosen set of variables in subsample. All variables of subsample “India-China” are
normally distributed. The meaning of VIF contributed to 1,39 for the chosen set of variables in subsample.
5.2 Evaluation
In order to evaluate the influence of each independent factor a series of linear regression tests has been held for each
determined sample: BRIC, Brazil-Russia-South Africa, India-China (3 parts of empirical investigation). As the final
model of the research the authors of this article have chosen the model in which all the factors are significant (at no
less than 5% level of significance).
I. For the chosen BRIC models specification tests were held, tests for model specification selection which reflects
temporal structure of the data available. The authors have carried out the Wald Test, Breusch-Pagan Test, Hausman
Test. Wald test showed that the Pooled-up model is rejected compared to the Fixed Effect model. Breusch-Pagan
test showed that Pooled-up model is rejected compared to the Random Effect Model. Hausman test showed that the
Random Effect model is rejected compared to the Fixed Effect Model.
According to the results of the tests the following model with the Fixed Effect was chosen for the BRIC sample.
According to the results the Fixed Effect model of the regression has been chosen. So the basic criterion was the
highest value R2within=0,5304. The final model with the Fixed Effects for the BRIC sample is presented below:
Test Statistics
Wald test F test that all u_i=0; F(263, 859)=3,64 Prob>F=0,0000
Breuch-Pagan test chi2=96,40 Prob>chi2=0,0000
Hausman test chi2=44,68 Prob>chi2=0,0000
Table 2: BRIC model specification
The subcomponents of human capital such as book value of total assets divided by the number of employees
and the subcomponents of innovation capital such as Capex and R&D are significant. Taking the absolute meaning
of Capex as indicator will lead us to the negative impact on the company’s cost of capital, what will confirm the
authors’ expectations. Year 2008 and 2009 can be considered as the years of the economic downturn especially
concerning the emerging markets. The influence of the 2008 and 2009 years is significant and positive in the model.
The book value of assets is the significant fundamental variable, but its “weight” is quite small. The same can be
said about the equity variable but its influence is positive. It can be explained by the thoughts that the owners of the
company prefer to raise the rate of the cost of equity while enlarging the equity capital, because of increasing risks
and the capital “erosion”. The long-term data analysis gives the opportunity to eliminate speculative value
fluctuations. Accordingly the accumulation of the Intellectual capital is the time-demanding process: the
performance should be evaluated over the long-run horizon. The meaning of the constant in final BRIC model is
positive, that means that in general WACC of the BRIC companies was increasing over the investigated period
2005-2009, which is probably due to the unstable economic situation those years.
II. For the chosen Brazil-Russia-South Africa models specification tests were held, tests for model specification
selection which reflects temporal structure of the data available. The authors have carried out the Wald Test,
Breusch-Pagan Test, Hausman Test. Wald test showed that the Pooled-up model is rejected compared to the Fixed
436

Effect model. Breusch-Pagan test showed that Pooled-up model is rejected compared to the Random Effect Model.
Hausman test showed that the Fixed Effect model is not rejected compared to the Random Effect Model.
The sample of companies in the research is closer to the general (universal) set than the fixed set of data, the
Random Effect model is more appropriate than the Fixed Effect model. According to the results of the tests the
following model with the Random Effect was chosen for the Brazil-Russia-South Africa sample.
Test Statistics
Wald test F test that all u_i=0; F(120, 300)=2,06 Prob>F=0,0000
Breuch-Pagan test chi2=16,16 Prob>chi2=0,0000
Hausman test chi2=8,79 Prob>chi2=0,2684
Table 3: Brazil-Russia-South Africa model specification
For the subsample Brazilian, Russian and South African companies the final model with Random effect was
chosen, Wald statistics is acceptable (502,1). The final model with the Random Effect for Brazilian, Russian and
South African sample is presented below:
In this sample the personnel expenses became significant; the industry of Oil & Gas has a large “weight”, what
is very logical, due to the specialty of the data. The role of the crisis period remains the same (positive influence),
while among the fundamentals there are some changes – the Intellectual value distress and the leverage become
significant with a comparable coefficient.
III. For the chosen India-China models specification tests were held, tests for model specification selection which
reflects temporal structure of the data available. The authors have carried out the Wald Test, Breusch-Pagan Test,
Hausman Test. Wald test showed that the Pooled-up model is rejected compared to the Fixed Effect model. Breusch-
Pagan test showed that Pooled-up model is rejected compared to the Random Effect Model. Hausman test showed
that the Fixed Effect model is rejected compared to the Random Effect Model.
Test Statistics
Wald test F test that all u_i=0; F(93, 247)=2,38 Prob>F=0,0000
Breuch-Pagan test chi2=32,53 Prob>chi2=0,0000
Hausman test chi2=7,98 Prob>chi2=0,2397
Table 4: India-China model specification
The final model with Random Effect was chosen for the second subsample (India-China). The final criteria is
Wald statistics (361,54). The final model with the Random Effect for Indian and Chinese sample is presented below:
6. Conclusion.
The main goal of this research is to evaluate by the means of the panel data analysis the influence of the particular
components of the Intellectual capital on the company’s cost of equity. Due to the fact that the Intellectual capital is
the process of accumulation through a long period of time it was decided to consider a 5-year time period and use
the panel data analysis. Conducting the research the authors resulted in the proof that the components of the
Intellectual capital are rather significant for the company’s sustainable existence. Human and innovation capital can
be considered as the key factors reflecting the company’s cost of capital both for the BRICS sample and for the
“group” samples. Talking about the BRICS countries it is crucial to emphasize that the influence of the human
capital (Personnel expenses per the number of employees) and some of the components of the innovation capital
(R&D expenses) is larger than the impact of such fundamentals as Total assets and Equity. There is no doubt that the
employees of the company and their knowledge and skills are the basis for the company’s effectiveness. It was
proved by the recent crisis: the benchmark companies in all industries made their best to support their staff, to create
the benefit programs and motivation schemes to keep employees loyal. Also during the crisis most of the key
financials lost their trust and value, which caused the abrupt search among the investors the other indicators to make
437

making decisions less risky and random. All of the above mentioned lead to the fact that the Intellectual capital has
becoming more and more popular among both the outsiders and the insiders of the company. The significant
influence of the innovation capital can be explained by the following conclusions. The influence of the capital
expenditures means that the company has the opportunity and bears substantial capital expenditures; the company
may have more opportunities and qualitatively improve its production. It is some kind of a signal that the company
has the possibility to implement the innovations and tends to optimize the current asset structure by replacing the old
equipment by the new one. Research and development expenses are the synonym for the long-term basis for the
company’s future development and therefore future decrease in the cost of capital.
There is no surprise that some of the years turned to have a significant influence on the company’s cost of
capital. This research was conducted using the data from 2005 to 2009 what includes both the period of stabilization
and the crisis. It was shown by the regression that the years 2008 and 2009 negatively impact on the cost of capital
and it is not surprising due to the fact that the economy of all the countries till 2007 was overheated and the crisis
destroyed all the development programs and forecasts. The found influence of the particular industries is rather
logical because of the specific division of the countries into two groups (and companies correspondingly) – the
source economies and the pushovers. The first group felt sensitive to the Oil & Gas industry and the second – the
industry of Basic Materials and Consumer goods. To sum up, from this research it became clear to the authors that
the Intellectual capital components play a great role in the company’s sustainable development and effective
existence. By increasing expenses on the human resources and innovation capital of the company probably enjoy
much higher financial indicators and will be able to focus on the long-term macroeconomic strategies.
7. References
Abdolmohammadi, M. (2005) Intellectual capital disclosure and market capitalization, Journal of Intellectual
capital, Vol. 6(3), 397-416.
Bayburina, E. (2007) Intellectual Capital Investigation Techniques as the Key Trigger of the Sustainable Long-term
Company Development, Journal Corporate Finance (e-journal), Vol. 3, 85-101 (In Russian).
Bayburina, E. & Ivashkovskaya, I. (2007) Role of the Intellectual Capital in the Value creation Process of the Large
Russian Companies, The bulletin of Finance Academy, Vol. 4(44), 53-62 (In Russian).
Bayburina, E. & Golovko T. (2008) The Empirical Research of the Intellectual Value of the Large Russian
Companies and the Factors of its Growth, Journal of Corporate Finance (e-journal), Vol. 2(6), 5-19 (In
Russian).
Bayburina, E. & Golovko, T. (2009) Design of Sustainable Development: Intellectual Value of Large BRIC
Companies and Factors of their Growth, Journal of Knowledge Management (e-journal),Vol.7 (5), 535-558
Bayburina, E & Golovko, T.V. (2010) Synergetic effects of Intellectual capital: “hidden” reserves for innovation
development of BRIC large companies, Proceedings of the 7th International Conference on Applied Financial
Economics, 169-178.
Botosan, C. (1997), Disclosure level and the cost of equity capital, Accounting Review, Vol. 72(3), 323-49.
Botosan, C. & Plumlee, M. (2002) A re-examination of disclosure level and the expected cost of equity capital,
Journal of Accounting Research, Vol. 40(1), 21-40.
Brennan, N. & Connell, B. (2000) Intellectual capital: current issues and policy implications, Journal of Intellectual
Capital, Vol. 1(3), 206-240.
Brooking, A. (1996) Intellectual Capital, International Thompson Business Press, Boston, MA.
Burgman, R. & Roos, G. (2007) The importance of intellectual capital reporting: evidence and implications, Journal
of Intellectual Capital, Vol. 8(1), 7-51.
El Ghoul, S., Guedhami, O., et al. (2010) Does corporate social responsibility affect the cost of capital, Journal of
Banking and Finance, 1-19.
Frankel, R., McNichols, M., Wilson, G. (1995) Discretionary disclosure and external financing, Accounting Review
(January), 135-150.
438

Guthrie, J., Petty, R., Yongvanich, K., Ricceri, F. (2003), Intellectual capital reporting: content approaches to data
collection, paper presented at Performance Measurement Association Intellectual Capital Symposium,
Cranfield, October 1-2.
Hail, L., Leuz, C. (2006) International differences in the cost of capital: Do legal institutions and securities
regulation matter? Journal of Accounting Research, Vol. 44(3), 485-531.
Healy, P., Palepu K. (1993) The effect of firms' financial disclosure strategies on stock prices. Accounting Horizons
(March), 1-11.
Lang, M., Lundholm, R. (1993) Cross-sectional determinants of analyst ratings of corporate disclosures, Journal of
Accounting Research 31, 246-271.
Lev, B. (2001), Intangibles, Management, Measurement and Reporting, Brookings Institution, Washington, DC.
Moore, N.G. (1996), Measuring corporate IQ, Chief Executive, November, 36-9.
O’Donnell, D., O’Regan, P., Coates, B. (2000) Intellectual capital: a Habermasion introduction, Journal of
Intellectual capital, Vol. 1(2), 187-200.
Petty, R., Guthrie, J. (2000) Intellectual capital literature review: measurement, reporting and management, Journal
of Intellectual Capital, Vol. 1(2), 155-176.
Shangguan, Z. (2005) Intangible investments and the cost of equity capital, Dissertation, University of Connecticut.
Song, K. (1993) Market timing and the cost of capital of the firm, Dissertation, Louisiana State University.
Stewart, T.A. (1997), Intellectual Capital: The New Wealth of Organizations, Currency-Doubleday, New York, NY.
Willis, V. (1991) Analyst following, cost of equity capital and the information environment: evidence from the
electric utility industry, Dissertation, University of Colorado at Boulder
Welker, M. (1995) Disclosure policy, information asymmetry and liquidity in equity markets, Contemporary
Accounting Research (Spring), 801-828.
439

BRAND VALUE DRIVERS OF THE LARGEST BRICS COMPANIES 1
Elvina R. Bayburina & Nataliya Chernova, National Research University – Higher School of Economics, Russia
Email: elvina.bayburina@gmail.com, nataliya.s.chernova@gmail.com, http://en.cfcenter.ru/
Abstract.
The world financial crisis emphasized the problem of the brand valuation, booming brand expectations were ruined by the world
economic transformations. The brand valuation critics predominantly from academics and business actors cover both classical
financial and modern approaches towards valuation. Financial crisis made the liquidity and stable cash flows as the main current goal
to achieve of major companies of developed and emerging markets. The obvious contradiction between current values and
fundamental drivers, which are unique in the process of value creation form the research ground. Thus, the main research discussion
covers the question whether it is necessary to consider the fundamental, brand value drivers, which are different from financials and to
use the complicated analytical tools, whether the value of the brand is the fundamental or temporal corporate value driver. The
framework of the empirical research is based on the stakeholders’ theory and the intellectual capital concept. The authors investigate
the influence of brand on the corporate value within the companies of the BRICS region. This research is extremely urgent and has a
great importance because of the lack of brand’s concept investigation, including corporate finance. There is a little attention paying to
this issue, especially in the case of developing capital market countries. Over the last years, more and more companies have realized
the significance and the need of a strategic management of their brand portfolio. This is valid for all economic sectors: for fast moving
consumer goods as well as for industrial goods. “In accordance with a survey conducted by Pricewaterhouse Coopers and Sattler
(2001) on German brand products across all categories, the average percentage amounts to 56 %, for manufacturers of fast moving
consumer goods the figure is even higher and comes up to 62 % of the total company value.” (Sattler, Hogl, Hupp, 2002). The
methods of brand evaluation have essentially increased over the global market, as a well-focused brand management demands an
outstanding brand controlling.
The data sample is formed from BRICS 2 companies. The first reason of such a choice is the increased importance of these countries on
the global market. As Goldman Sachs has argued, since the Brazil, Russia, India and China are developing rapidly; by 2050 their
combined economies could exceed the combined economies of developed countries. The second reason is related to their original
difference from the developed countries. It is commonly known that these four countries have specific economic, political, social and
legislative structures. Therefore some laws, existing in the developed countries have no implementation there. This research provides
the information, which can help to operate better the brand portfolio for top-managers in the range of industries, to take part in buyouts
and mergers, to reshape the investment activity in more optimal way that is of great importance especially for BRICS companies.
Keywords: emerging markets, economic crisis, value drivers, intangibles, valuation, brand, brand value, branding, marketing,
intellectual capital, client capital, BRICS
JEL: L14, L24, G34, D85
1. Introduction
The world financial crisis emphasized the problem of the brand valuation, booming brand expectations were ruined
by the world economic transformations. The brand valuation critics predominantly from academics and business
actors cover both classical financial and modern approaches towards valuation. Financial crisis made the liquidity
and stable cash flows as the main current goal to achieve of major companies of developed and emerging markets.
The obvious contradiction between current values and fundamental drivers, which are unique in the process of value
creation form the research ground. For example, the acquisition of brand “Skype” was so promising for its buyer
“Ebay”, however a lot was omitted due to the obvious limitations of the solely financial approach to the brand
valuation, later after the acquisition the overall management results were very far away from desired goals. It should
be noted that still, those business actors who tend to survive and stand the economic crisis continued to run the value
of their brands on the basis of its internal fundamental drivers such as marketing and client capital, organizational
capital (innovations and technology), qualified personnel and human capital (Bayburina, Levkin, 2010). Thus, the
main research discussion covers the question whether it is necessary to consider the fundamental value drivers,
which are different from financials and to use the complicated analytical tools, whether the value of the brand is the
fundamental or temporal corporate value driver. The framework of the empirical research is based on the
stakeholders’ theory and the intellectual capital concept. However, looking through the scientific literature dealing
1 The results of the project “Researches of corporate financial decisions of the companies of Russia and other countries with emerging capital
markets under conditions of global transformation of the capital markets and formation of economy of innovative type”, carried out within the
framework of the Programme of Fundamental Studies of the Higher School of Economics in 2011, are presented in this paper.
2 Brazil, Russia, India, China, South Africa. On 24th of December 2010 South Africa received an official invitation to enter BRIC. The entrance
has previously been approved by members of BRIC, however, according to many experts, South Africa is not on the level of other BRICcountries
by some economic measures.
440

with the brand evaluation subject, it turns out that there is not until today a valid and comprehensible model. So the
main methodological purpose of this paper is to build the model, which will be useful for estimating the real brand
value and for following empirical researching of brand value. Therefore, this research paper has several objectives.
The first objective is to consider the existent brand valuations, based on different perspectives of the brand equity.
The review of the literature allows extracting the necessary approach. The second objective is to formulate, to
modify, the methodology of proposed research. Afterwards, the paper presents the main results of empirical research
and presents the main findings.
2. Brand value perspective
Tangible assets have stopped to be the general driver for corporate value growth and competitive advantage because
of globalization, technological and production development, which have intensified and have given new
opportunities for companies from emerging markets to succeed on the field of intangible assets. Keller (2007), who
provides the summary of the literature, lists a lot of financial benefits to build a strong brand and reshape the
marketing strategy towards it, that includes for sure greater customer loyalty, larger profits, less vulnerability;
competitive marketing programs and plans help to dispatch the consequences of economic crises, to originate less
elastic consumer response to price increases and more elastic consumer responses to price decreases, greater trade
cooperation and support, increased marketing communication effectiveness, increased licensing opportunities, and a
lot of additional brand extension opportunities. As a well-focused brand management demands an outstanding
brand controlling, the brand evaluations methods are to be used along with the implementation of a strategic brand
management throughout within the company, has essentially increased in the global market. Many authors devote
their investigation and research efforts to the concept of brand value, trying to measure the brand value and to give
right recommendations in terms of creating and building a brand. Nevertheless, companies as well as consultants are
uncertain about how to conduct a feasible brand assessment exactly.
Nowadays there are two main perspectives that may be adopted. The first one considers brand equity in the
context of marketing analytical lens. The second one is financially based. Each perspective has its advantages and
disadvantages. To support the launch of new brand, it is better to appraise brand perspectives based on the
customers’ and clients’ perception (marketing perspective and marketing analytical lens). However, if brand exist
and the core lies more on the brand’s performance within a certain analytical lens, then the financial evaluation of
the brand tends to be more helpful and meaty in terms of strategic corporate finance research and value based
management. The brand financial evaluation also becomes increasingly vital as recently, more and more brands
have to be investigated for their value due to company buy outs, mergers or the transfer of divisions, alliances
activity and different kinds of network collaboration towards foreign markets or even franchise, licensing. The
review of the brand valuation models shows that there are many works devoted to the assessment of brand equity,
but in spite of this fact, there is not until today a valid and comprehensible model. Most of researches are based on
assumptions, which it is almost impossible to prove empirically.
2.1 Background approach
In brand valuation theory all models can be split into three groups:
� traditional economic brand valuation models;
� client and behaviorally-oriented brand valuation models;
� holistic economic and behaviorally-oriented brand valuation models.
Further the most significant studies presented the above models are described.
2.1.1 Traditional economic brand valuation models
Capital market-oriented brand valuation (Simon and Sullivan, 1993). Simon and Sullivan determined brand equity
as “the incremental cash flows which accrue to branded products over and above the cash flows which would result
from the sale of unbranded products.” (Simon, Sullivan, 1993) To fulfill this definition, it estimates the current
market value of the company. The market value of the equity gives an understanding and impartial market
estimation of the possibilities and future cash flows. Then the authors extract the brand value from the value of the
other assets of the company, which consists of tangibles and its remaining intangible assets: the value of other
specific assets is not associated with the brand equity and market-specific factors that lead ultimately to imperfect
441

competition and market falls. Thus, the resultant estimate of brand equity is based on the market valuation of the
future cash flows.
Market value-oriented brand valuation. In a market value-oriented approach, the value of a brand is established
by the fair value of comparable brands. However, this procedure is reliable, if the market position of the reference
brand and the brand to be evaluated are more or less identical, this research assumption is hardly to be realistic, but
with no doubt to be vital.
Cost-oriented brand valuation. By this model the pure value of assets equals total assets of the company less all
obligations. There are two alternatives to the net asset value approach: to assess the brand value by means of the
costs (historical or perceived) which have arisen while brand was launched in the past, and to assess expenses which
would arise today respectively.
Earning capacity-oriented brand valuation (Kern’s model). According to this approach, the brand value is
calculated by capitalizing the value of potential earnings (earnings to be discounted) received by the company,
which possesses the brand with the earnings capacity.
Customer-oriented brand valuation. Sometimes renewals are not subject to the decision making in marketing.
So, the earning-capacity indicator in these models is based not only on the net income but on the average customer
contribution to net income margin attained by a brand product and brand loyalty rate. As it could be noticed,
traditional economic brand valuation models are related to the financial perspective of brand equity. They measure
the financial part of brand value and do not take into account the particular customers’ and clients’ influence to
brand value.
2.1.2 Client and behaviorally-oriented brand valuation models
Aaker brand valuation model (Aaker, 1991). The model points out five determinants of brand value: brand loyalty,
brand awareness, perceived quality, brand associations and other brand assets. It regards brand equity from a
consumer’s perspective. However, this model does not include the quantity value.
Kapferer brand valuation model (Kapferer, 1992). This approach assumes that the brand value reduces buying
risk of customers and is determined as the differential effect of brand knowledge on customer response to the
marketing mix efforts to promote the brand.
Keller brand valuation model (Keller, 1993). This model refers to the assumption that the brand value is
corresponding with the knowledge of the brand. It is based on comparison with an unbranded product. Brand
knowledge consists of brand awareness and brand image.
McKinsey brand valuation model (McKinsey, 1994). McKinsey defines the three “P” - Performance, Personality
and Presence - of the brand, the main components of the strong brand. McKinsey estimates the brand strength as the
function of those three “P”.
To sum up, client and behaviorally-oriented brand valuation models focus on the consumer’s behavior and
attitudes.
2.1.3 Holistic economic and behaviorally-oriented brand valuation models
These models include both financial and client-oriented indicators.
Semion brand value approach (Clifton et el., 2009). The approach defines Brand value drivers, such as:
financials of the company, brand strength, brand protection and brand image. Firstly, the author ascertain the
factor’s value for each of those determinants on the basis of the determined criteria, then the indicator for each
factor is summed and aggregated to give a single value. Secondly, the values are summarized to produce the
weighted factor. Afterwards, average earnings of the past three years are multiplied by the obtained weighted factor.
Market-oriented brand valuation model (Bekmeier-Feuerhahn, 1998). This model says that the brand value is
related to brand strength and brand earnings estimated by the means of market prices.
442

Interbrand brand value approach. In this model there are two key elements of the valuation of a brand: the
earnings referred to the brand and the brand strength multipliers. The multiplier shows the strength of the brand
itself. “It takes into account seven factors: market leadership, brand stability, brand extension possibilities, current
market prospect, and internationalization potential, adaptation to time, brand support and legal
protection.”(Molameni, Shahrokhi, 1998). Then the brand strength is correlated to a multiple such as the price
earning (P/E) ratio, which gauges the certainty in the future. The brand multiple then is be applied to earnings of the
brand to extract the value of brand equity. Interbrand Group has made a graph known as the “S - curve” to relate the
brand strength and multiplier. This curve is based on Interbrand’s examination of multipliers implicated in many of
brand P/E ratios over the accepted period. Although, this approach seems to be comprehensive, its shortcomings
can be pointed out, the brand weights are based on historical data.
Intellectual value concept and stakeholders’ approach. As it is said, majority of above mentioned valuation
models estimates only financial part of brand, others uncover clients’ and customers’ attitudes and behavior in
brand valuation. To determine Brand value drivers holistic economical and behaviorally-oriented brand valuation
models are to be used to include both financial estimates and comprehensive research. In this paper the combined
approach is proposed, as it provides a more precise estimate of Brand value drivers to raise and to boost the
corporate value according to the Intellectual capital concept. From the point of view of (Bayburina & Golovko,
2008) the intellectual capital is a complexity of key qualitative characteristics of the company, which are not always
objects of intellectual property, such as, for instance, competencies. Thus, intellectual capital is a complexity of
knowledge, accrued experience of employees and intellectual property. It is important to mention that not all objects
of engineering design are the objects of the intellectual property; however its importance for the process of the value
creation is difficult to underestimate, for example, in the companies of the high tech industry. Simultaneously it is
complicated to provide the full, complete, exact definition of the notion “intellectual capital of the company”. For
this purpose it is reasonable to concentrate on its structure and the content. In less details it is possible to form a
structure of intellectual capital, which consists of the 5 following components or the elements of the first level of
intellectual capital; this approach has been used by the authors (Bayburina & Ivashkovskaya, 2007; Bayburina &
Golovko, 2008; Bayburina & Golovko, 2009). Briefly, the structure of intellectual capital could be described as
follows:
1. human capital (key knowledge and abilities of the personnel);
2. process capital (key characteristics of the business processes of the company);
3. client capital (key features of the company which are necessary to manage customer relationship and loyalty);
4. innovation capital (renovation techniques to maintain the future growth of the company);
5. network capital (synergy which occurs from the interactions of the company).
The complexity of such characteristics represents the competitive advantage originated inside and within the
company. According to (Bayburina & Golovko, 2008) from the one hand the volatility of the business environment
from the other hand the opposite stakeholders interests stipulate the fact that value becomes the result of constant
strategic changes. In this case the value becomes a complicated intellectual parameter, which is managed and
defined by the multi-level combination of interactions of different groups of stakeholders (e.g. clients). By this
pattern the value is generated by its intellectual part. So, (Bayburina & Golovko, 2008) conclude, that the
Intellectual value of the company is a part of the total value, created through the process of the intellectual
components’ accumulation. This value exactly as the brand value can be “traced” and should be traced by the
external stakeholders of the company.” (Lev & Sougiannis, 1999; Singh & Van der Zahn, 2009).
2.2 Hypotheses
The methodology will utilize the combined approach, as the explanation of the results, which consequentially and
indirectly include attitude of consumers and clients. After determining the brand value, Brand value drivers, the
influence of different factors will be estimated; among those factors are well-known fundamental financials usually
used to be investigated according to the Intellectual capital concept in terms of value following Bayburina and
Golovko (Bayburina and Golovko, 2009). According to above mentioned Intellectual capital concept in terms of
value the hypotheses of this study are:
443

Hypothesis 1.Positive influence of the company size on the brand value. This hypothesis is based on the fact,
that the large company has a range of advantages. It has more stable position on the market, the consumers and the
suppliers try to cooperate with such a company, as it is more observable and reliable. The large size allows the
company to reduce production costs, to generate innovative products and to conduct great marketing strategies.
Hypothesis 2. Positive influence of the company’s cumulative marketing (advertising and promotion) expenses
on the brand value until the moment, after which brand value returns diminish along with expenses increase.
Advertising is one of the main components of marketing communications. It can increase brand name recognition
and a value premium for reputation. Advertising creates special associations of the particular brand, which
differentiates the products of a company and defends them from imitating by competitors (Mizik and Jacobson,
2003). Also it can be noticed that advertising is a kind of the entry barrier for the other company, that is why it is
necessary to spend the high amount of money to overthrow the established brand loyalty (Ho et al., 2005). Another
marketing communication is trade promotion, which also can create a brand value (Keller, 2007). Promotion
enhances brand attributes and help to identify the particular brand. In fact, altogether expense on promotional
activities compound about 60–75% of promotional budgets as being spent on sales promotion (Belch and Belch,
2007).
Hypothesis 3. Positive influence of the company’s expenses on R&D, until the moment, after which brand value
returns diminish along with expenses increase. It is assumed by the authors that companies with higher expenses on
R&D have the stronger. This can be true, as new technologies and company’s know-how can also be viewed as an
entry barrier, as the potential competitors need the high amount of money to get these advantages. To reach the
settled goals the empirical research model will be tested and hypotheses will be verified.
2.3 Research model
(1) Brand valueit=α+(ρ 1 it,… ρ n it) X BVD + (β 1 it ,.….β k it )X FV +εit,
(2) Brand value Ratioit=α+(ρ 1 it,… ρ n it) X IC + (β 1 it ,.….β k it )X FV +εit, where:
Brand valueit – a vector of Brand value meanings, calculated according to the formula (3),
Brand value Ratioit – a vector of Brand value ratios, calculated according to the formula (4),
BVD – a vector of Brand value drivers;
FV – a vector of fundamental variables;
ε – a vector of random errors (“white noise”);
I – BRICS company index;
T – a year index;
N – a particular Brand value driver index;
K – a particular fundamental variable index.
2.3.1 Dependent variable
The dependent variable is the Brand value. Two ways of calculation are considered. Firstly Brand value equals the
Market capitalization divided by the Net Assets.
(3)Brand valueit = Market Capitalizationit / Net Assetit, where:
I – BRICS company index;
T – a year index;
� Market Capitalizationit – Market value of all shares (USD);
� Net Assetsit - Book value of net assets (USD).
The second proxy dependent variable Brand value ratio, called the Q-tobin ratio, is calculated as the sum of the
Market capitalization and the Total assets divided by the sum of the Total Assets and the Total liabilities.
(4) Brand value Ratioit = Q-Tobinit = (Market Capitalizationit + Total Liabilitiesit) / (Total Assetsit + Total
Liabilitiesit), where:
I – BRICS company index;
T – a year index;
� Market Capitalizationit – Market value of all shares (USD);
� Total Liabilitiesit - Book value of total liabilities (USD);
� Total Assetsit - Book value of total assets (USD).
444

2.3.2 Independent variables
1. Independent variables: fundamentals
� Market Capitalization divided by Total Assets,
� Return on assets (ROA),
� Return on equity (ROE),
� Capital Expenditure divided by Total assets,
� Ln (Revenue),
� Ln (Market Capitalization),
� Ln (delta Market Capitalization),
� Ln (Total Assets),
� Ln (Net Assets),
� Ln (Ebit),
� Ln (Equity),
� Ln (Total Liabilities),
� Ln (Capital Expenditure),
� Ln (delta Capital Expenditure).
Fundamental variables were included in the research model as “control” variables.
2. Independent variables: Brand value drivers
� Marketing Expenses in present time period,
� Marketing Expenses in previous time period,
� R&D Expenses in previous time period,
� Ln (Marketing Expenses in present time period),
� Ln (Marketing Expenses in previous time period),
� Ln (R&D Expenses in present time period),
� Ln (R&D Expenses in previous time period),
� Ln (Total Assets divided by Number of Employees),
� Ln (Number of Employees).
Below there is a table, which presents Hypothetical signs of the coefficients.
Independent variables
Hypothetical sign of the coefficients
(Brand value)
Hypothetical sign of the coefficients (Q-
Tobin)
Market Capitalization divided by Total
Assets
+ +
Return on assets - +
Return on equity + -
Capital Expenditure divided by Total assets + +
Ln (Revenue) + +
Ln (Market Capitalization) + +
Ln (delta Market Capitalization) + +
Ln (Total Assets) - -
Ln (Net Assets) - -
Ln (Ebit) + +
Ln (Equity) - -
Ln (Total Liabilities) + -
Ln (Capital Expenditure) + +
Ln (delta Capital Expenditure) + +
Ln (Capital Expenditure divided by Total
assets)
+ +
Marketing Expenses in previous time period + +
R&D Expenses in previous time period + +
Total Assets divided by Number of
Employees
+
+
445

Independent variables
Hypothetical sign of the coefficients
(Brand value)
Hypothetical sign of the coefficients (Q-
Tobin)
Number of Employees + +
Ln (Marketing Expenses in present time
period)
+
+
Ln (Marketing Expenses in previous time
period)
+
+
Ln (R&D Expenses in present time period) + +
Ln (R&D Expenses in previous time period) + +
Ln (Total Assets divided by Number of
Employees)
+
+
Ln (Number of Employees) + +
Time influence Significant Significant
Industry influence Significant Significant
2.4 Sample and sources of data
Table 1. Hypothetical signs of the coefficients.
The research conducted in this study is dedicated only to BRICS companies because it is a widespread opinion that
the BRICS countries are the most interesting in terms of sample of emerging markets to investigate. The term
“BRIC” was launched by the economist from Goldman Sachs Investment Bank Jim O’Neill in 2003. After 2011
year conference in Davos, it was proposed to include South Africa in BRIC emerging markets group, because it
reflects all the issues related to Brazil, Russia, India and China. The BRIC countries as Brazil, Russia, India and
China are frequently referred to have attracted media and academic attention in the recent years. These countries are
different from one another over their culture, background, language, and the structure of their economies. However,
they have a common denominator: economic growth development in the BRICS has greatly exceeded growth
compared to the world's leading industrialized nations. Even after the economic crisis that started in 2007, they
continued outperforming the rest of the world. While in 2009 large economies shrunk as much as 6%, (e.g., Japan
and Germany), Brazil stayed steady, India grew 5.9%, and China 8.1%; only Russia was the group's bad performer
shrinking 7%. One more common characteristic is that the financial assets of these 5 countries are underestimated
and have a great potential of growth. The forecast of their future performance will further increase the interest on
these countries and also gives the reason to pick exactly this set of countries to develop the research in this study.
While all of BRICS countries are associated with the most fast developing markets, the potential of each
country is estimated as being at the different levels. After the crisis the scholars have divided the scope of countries
into 2 parts – “the source economies” (Brazil, Russia and South Africa) and “the pushovers” (India, China). That’s
why, in this paper it was decided to develop two models for each group separately. Taking into account the necessity
to assess companies from different countries the final research sample was formed along with the following criteria:
� The sample is formed from BRICS companies listed from the year 2005 till 2010.
� Such research is of special interest, owing to the fact, that, from one hand, BRICS countries are leaders of
emerging economies, though, at the same time, they differ dramatically from each other. It should be
mentioned, that the information used in the study should be available in the open sources (e.g. annual reports or
other information on websites), thus making it useful and acceptable to trace for a wide range of users. In order
to ensure comparability, companies should have financial reports along with the National GAAP accounting
standards. As the results of previous studies show (Bayburina, Golovko, 2009), the number of companies,
especially from India and Brazil, which have IFRS financial reports in the period earlier than 2005, is not quite
high and is not sufficient enough for the research purposes, thus it is necessary to impose restrictions on usage
of such type of reports in the research.
� The sample is formed from BRICS companies about which there is enough disclosed Brand data. Observation
period: 2005-2009 years (5 full financial years). Number of observations: 306.
Number of Companies Country of residence
71 Brazil
58 Russia
61 India
57 China
59 South Africa
Table 2. BRICS sample.
446

As the sources of data the following sources were used:
1. Bloomberg data base (fundamental factors’ and Brand value drivers).
2. Corporate official web-sites (Brand value drivers).
3. Corporate accounting reports (fundamental factors’ and Brand value drivers).
To reach the main goal of the research the special research model has been introduced and a series of tests has
been held. The data research was fulfilled with the use of software tools (program packages Stata 11.0, Eviews 7.0,
Microsoft Excel 2007).
3. Results
The pooled sample of BRICS was split into 2 parts: “India, China”, “Brazil, Russia, South Africa”. In order to verify
each hypothesis a series of linear regression tests has been held. Final models of the research have significant factors
(at no less than 5% level of significance; and the constant of the model is significant at 6,5% level of significance)
and the meaning of coefficient of determination is the highest (R2 (“India, China” Model)= 0.8593 and R2(“Brazil,
Russia, South Africa” Model)= 0.2442 ).
Test for normality. Both models were tested for normality of variables distribution, Skewness-Kurtosis Test for
Normality was held.
Test for Multicollinearity. All models were tested for the presence of multicollinearity. In every final model all
VIFs are less than 4. The meaning of VIF contributed to 1.89 for the chosen set of variables in subsample “India,
China” and 2.07 for the chosen set of variables in subsample “Brazil, Russia, South Africa”.
Test for Heteroskedasticity. All models were tested for the presence of heteroskedasticity. All tests for the final
model confirm their homoskedasticity.
“India, China”model. Model has several variables which are significant: Market Capitalization divided by
Total Assets, Return on equity, Ln (Equity), Ln (Marketing Expenses in present time period), Ln (Total Assets
divided by Number of Employees). To choose the final model specification it was necessary to hold model
specification tests, such as: the Wald Test, Breusch-Pagan Test, Hausman Test:
1. Wald Test showed that the Fixed Effect Model provides a better explanation than the Pooled-up Model
2. Breusch-Pagan Test showed that the Pooled-up Model provides a better explanation than the Random Effect
Model.
3. Hausman Test showed that the Fixed Effect Model provides a better explanation than the Random Effect Model.
4. The Fixed Effect Model is final.
Test Statistics
Wald test F test that all u_i=0; F(104, 298) = 6.48
Prob > F = 0.0000
Breuch-Pagan test chi2=115.50 Prob>chi2=0,0000
Hausman test chi2=67.08 Prob>chi2=0,0000
The final model is presented below:
Table 3. Specification tests for “India, China”model
(3) Brand valueit = 20.425+ 1.856 � MC/TAit + 3.357 �ROEit - 3.054 � LnEit+ 1.308 �lnTA/NEit + 1.025 � LnMEit,
where:
� MC/TA – Market Capitalization divided by Total Assets;
� ROE – Return on equity;
� LnE– Natural logarithm of Equity;
� LnTA/NE– Natural logarithm of Total Assets divided by Number of Employees;
� LnME – Natural logarithm of Marketing Expenses in present time period;
447

� I – BRICS company index;
� T – a year index.
“Brazil, Russia, South Africa” model. Model has several variables which are significant: Return on assets, Capital
Expenditure divided by Total assets, Ln (Revenue), Ln (Market Capitalization), R&D Expenses in previous time
period, Total Assets divided by Number of Employees, Number of Employees, Ln (R&D Expenses in present time
period). To choose the final model specification it was necessary to hold model specification tests, such as: the Wald
Test, Breusch-Pagan Test, Hausman Test:
1. Wald Test showed that the Pooled-up Model provides a better explanation than the Fixed Effect Model.
2. Breusch-Pagan Test showed that the Pooled-up Model provides a better explanation than the Random Effect
Model.
3. Hausman Test showed that the Fixed Effect Model provides a better explanation than the Random Effect Model.
4. The Pooled-up Model is final.
Test Statistics
Wald test F test that all u_i=0; F(168, 570) = 0.93
Prob > F = 0.6985
Breuch-Pagan test chi2= 0.32 Prob>chi2= 0.5727
Hausman test chi2=17.43 Prob>chi2=0.0078
The final model is presented below:
Table 4. Specification tests for “Brazil, Russia, South Africa” model
(4) Brand value Ratioit = -881.381 + 0 .006 � NE it + 0 .002 � TA/NE it + 0 .696 � R&D 0 it + 0 .401 � ROA it +
51.042 � LnMC it + 158.758 � Capex/TA it + 61.764 � LnR&D it - 0.003 � LnR it, where:
� NE – Number of Employees;
� TA/NE – Total Assets divided by Number of Employees;
� R&D 0 - R&D Expenses in previous time period;
� ROA–Return on assets;
� LnMC – Ln (Market Capitalization);
� Capex/TA - Capital Expenditure divided by Total assets;
� LnR&D - Ln (R&D Expenses in present time period);
� LnR- Ln (Revenue);
� I – BRICS company index;
� T – a year index.
4. Conclusions
The world financial crisis emphasized the problem of the brand valuation, booming brand expectations were ruined
by the world economic transformations. The brand valuation critics predominantly from academics and business
actors cover both classical financial and modern approaches towards valuation. Financial crisis made the liquidity
and stable cash flows as the main current goal to achieve of major companies of developed and emerging markets.
The obvious contradiction between current values and fundamental drivers, which are unique in the process of value
creation form the research ground. For example, the acquisition of brand “Skype” was so promising for its buyer
“Ebay”, however a lot was omitted due to the obvious limitations of the solely financial approach to the brand
valuation, later after the acquisition the overall management results were very far away from desired goals. It should
be noted that still, those business actors who tend to survive and stand the economic crisis continued to run the value
of their brands on the basis of its internal fundamental drivers such as marketing and client capital, organizational
capital (innovations and technology), qualified personnel and human capital (Bayburina, Levkin, 2010). Thus, the
main research discussion covers the question whether it is necessary to consider the fundamental value drivers,
which are different from financials and to use the complicated analytical tools, whether the value of the brand is the
fundamental or temporal corporate value driver. The framework of the empirical research is based on the
stakeholders’ theory and the intellectual capital concept. In this investigation the authors set a goal to estimate Brand
values drivers by the means of the panel data. Therefore the paper provides the empirical research of Brand value
drivers and Brand value within the large BRICS companies over the 5-year period.
448

The research shows that the Brand value drivers, non-fundamental components of the Intellectual capital are
rather significant in terms of sustainability. The greatest effect of Brand value drivers, non-fundamental
determinants can be noticed in “India, China” sample, with Brand value as dependent variable, and “Brazil, Russia,
South Africa” sample with Brand value ratio (Q-Tobin) as dependent variable. The contributions of marketing and
R&D expenses to Brand value are obvious. All coefficients are positive. It verifies the second and third hypothesis
and means that most companies of BRICS spend not so much for these needs. (Chu, Keh, 2006). Otherwise it
developed that was impossible to test those parts of both hypotheses, which suppose that there is trigger point, after
which returns to brand value diminish and expenses increase, still by the largest companies of BRICS this trigger
point is not achieved yet. The authors can conclude that BRICS companies don’t invest enough money into brand
promotion as it is executed in large multinationals of developed countries. The practical application of this result is
obvious: spending its money on the marketing and R&D can give the company an opportunity to create the
competitive advantage. Otherwise the efficiency of such expenses depends on each particular emerging country
features. In “India, China” subsample marketing expenses are significant and in “Brazil, Russia, South Africa” -
R&D expenses. It can be explained by the cost of marketing and R&D in different countries. In subsample “Brazil,
Russia, South Africa” R&D costs are quite low versus “India, China” subsample. Marketing costs are lower in
“India, China” subsample, than in “Brazil, Russia, South Africa” sample. It is possible to assume that for Indian and
Chinese large public companies it is necessary to develop the brand promotion programs and spend much more on
it, this is the field of inevitable and obvious Brand value boosting. It is possible to conduct that for Brazilian,
Russian, South African large public companies it is necessary to invest more over Research and Development
strategic roadmaps, with no doubt that is the obvious Brand value driver. In all cases it can be noticed, that the
increase of number of employees has the positive impact on the brand value. In all public companies of emerging
capital markets the large size allows the company to reduce production costs, to generate innovative products and to
conduct great marketing promotional activities. Summing up, the research shows that the Brand value drivers are
the matter of concern. For sure they have an influence on corporate value and can have a great importance in the
company’s existence taking into consideration the global competition. BRICS large companies should try to increase
expenses on marketing and R&D, as they have the enormous potential for the growth. In the future research it is
worth to pay attention to other components of brand value and to explore the changes conducted by the type of
industry and each particular year (post-crisis and after-crisis features).
5. References
Aaker, D. (1991) Managing Brand Equity, San Francisco: Free Press.
Aaker, D. (1996) Measuring Brand Equity across Products and Markets, California Management Review, 102-120.
Bayburina, E. (2007) Intellectual Capital Investigation Techniques as the Key Trigger of the Sustainable Long-term
Company Development, Corporate Finance (e-journal), Vol. 3, 85-101 (In Russian).
Bayburina, E. & Golovko, T. (2008) Intellectual Value of the company and the factors of its growth: a panel study
of the large Russian companies, Corporate Finance (e-journal), Vol. 6, 5-19 (In Russian).
Bayburina, E. & Golovko, T. (2009) Design of Sustainable Development: Intellectual Value of Large BRIC
Companies and Factors of their Growth, Journal of Knowledge Management (e-journal),Vol.7 (5), 535-558
Bayburina E. & Levkin D. (2010) Brand stabilizer bottom-line: empirical research of large Russian companies,
Contemporary Views on Business - Partnering for the Future, Conference Proceedings, Helsinky, 312-313.
Bekmeier-Feuerhahn, S. (1998) Marktorientierte Markenbewertung. Eine konsumenten- und unternehmensbezogene
Betrachtung, Wiesbaden.
Belch, G.E., Belch, M.A. (2007) Advertising and promotion: an integrated marketing communications perspective,
7th(ed.), McGraw Hill.
Clifton et el. (2009) Brands and branding, NY: Bloomberg Press.
Chu, S., Keh, H. (2006) Brand Value Creation: Analysis of the Interbrand-Business Week Brand Value Rankings.
Market Letters, 323-331.
Ferjani, M., Jedidi, K., Jagpal, S. (2009) A Conjoint Approach for Consumer- and Firm-Level Brand Valuation.
Journal of Marketing Research, 846–862.
449

Ho, Y.K., Keh, H.T., Ong, J. (2005) The effects of R&D and advertising on firm value: An examination of
manufacturing and non-manufacturing firms, IEEE Transactions on Engineering Management, 52(1), 3–14.
Jucaityte, I., Virvilaite, R. (2007) Integrated Model of Brand Valuation, Economics and Management, 376-383.
Kallapur, S., Kwan, S. (2004) The Value Relevance and Reliability of Brand Assets Recognized by U.K. Firms,
Accounting Review, 151-172.
Kamakura, W., Russell, G. (1993) Measuring Brand Value with Scanner Data, Elsevier Science Publishers B.V, 9-
22.
Kapferer, J.N. (1992). Strategic brand management: new approaches to creating and evaluating brand equity,
Kogan Page, London.
Keller, K. (1993) Conceptualizing, Measuring and Managing Customer-Based Brand Equity, Journal of Marketing,
57(1), 1-22.
Keller, K. (2007) Strategic Brand Management Building, Measuring and Managing Brand Equity, 3rd(ed.), Upper
Saddle River, NJ: Prentice Hall.
Lev, B., Sougiannis, T. (1999) Penetrating the book-to-market black box: the R&D effect, Journal of Business
Finance and Accounting, Vol. 26 (3/4), 419-49.
McKinsey, F. (1994.) Winning the right to brand, Brand Strategy Newsletter, November.
Mizik, N., Robert, J. (2003) Trading Off Between Value Creation and Value Appropriation: The Financial
Implications of Shifts in Strategic Emphasis, Journal of Marketing, Vol. 67(1), 63-76.
Molameni, R., Shahrokhi, M. (1998) Brand Equity Valuation: a Global Perspective, Journal of Product and Brand
Management, 275-290.
Sattler, H., Hogl, S., Hupp, O. (2002) Evaluation of the Financial Value of Brand, Research Papers on Marketing
and Retailing, University of Hamburg.
Singh, I., and Van der Zahn, M.. (2009) Intellectual capital prospectus disclosure and post-issue stock performance,
Journal of Intellectual Capital, Vol. 10(3), 425 – 450.
Simon, C., Sullivan, M. (1993) The Measurement and Determinants of Brand Equity: a Financial Approach,
Marketing Science, 28-52.
Virvilaite, R., Jucaityte, I. (2008) Brand Valuation: Viewpoint of Customer and Company, Engineering Economics,
28-52.
450

GOVERNANCE, BOARD INDEPENDENCE, SUB COMMITEES AND FIRM PERFORMANCE:
EVIDENCE FROM AUSTRALIA
Wanachan Singh, Suan Dusit Rajabhat University, Bangkok, Thailand
Robert T Evans, Curtin University, Perth, Australia
John P Evans, Curtin University, Perth, Australia
Email:John.Evans@curtin.edu.au , www.curtin.edu.au
Abstract. This study investigates whether the monitoring of company management by an independent board of directors serves to
enhance firm performance in Australia. The paper is of interest in that it explores the impact of sub-committees such as audit,
remuneration, and nomination as well as the quality and size of the audit firms. From the perspective of a regulator our findings have
implications and suggest that the imposition of sub-committees and the respective composition does not in itself align the interest of
managers and shareholders.
Keywords: Governance, board of directors, agency theory, Australia
1 Introduction
Agency theory defines the agency relationship where the principal (or owner) delegates tasks to an agent (or
manager). The theory highlights costs associated with the principal-agent relationship which include the
opportunistic behaviour or self-interest of the agent taking priority over the principal’s interest. Mallin (2004)
highlighted a number of dimensions to this including the agent misusing their power for financial or other
advantage, and the agent not taking appropriate risks in pursuance of the principal’s interests – often because
managers are more risk-averse than the companies they lead. Another cost arises due to the principal and agent
having access to different levels of information; the agent (manager) usually being in control of superior and more
detailed information than that of the owner (information asymmetry). This requires the owner to institute expensive
monitoring of the managers actions to redress the knowledge imbalance.
Fama (1980) argued that the boards of directors provide the most critical internally-based method for
monitoring the performance of managers. They have the ability to directly oversee the performance of managers and
to offer incentives to managers which reward performance in line with owner expectations, or equally, discipline
managers when these are not met. Fama and Jensen (1983) note that effective monitoring requires the board of
directors to act as an independent arbiter between management and owners, as collusion could result in an overall
loss to the owners. To assess the degree to which a board may act independently to monitor the performance of
managers, a number of key measures of board independence have been studied. These include the size and structure
of the board, with the former referring to the number of directors and the latter generally being the proportion of
non-executives, the number of sub-committees, the proportion of non-executives on subcommittees and whether the
CEO and chairperson positions are combined. If monitoring is effective, a positive relationship should exist between
the level of board independence and firm performance (Mura 2007; Choi, Park and Yoo 2007; Schmid and
Zimmermann, 2008).
There has been little research done on board-monitoring subcommittees particularly the existence of them and
their independence (Gales and Kesner 1994; Dalton et al. 1998). This study extends the work of previous researchers
to include the number of board subcommittees established by the main board as well as their composition as part of
the internal corporate governance mechanisms that potentially affect firm performance.
2 Background
A number of empirical studies have been undertaken to assess the degree to which the independence of the Board
(as reflected in the above characteristics) impact on company performance. Rosenstein and Wyatt (1990) assessed
whether share prices respond positively when additional outside directors are appointed to companies. They found a
statistically significant, but economically-small effect on prices averaging around 0.2%. They found this effect to be
slightly stronger when the appointment is for a director representing a financial institution versus those in other
451

usiness. Their (1997) study focused on the price effect of the appointment of insiders (managers) to the board.
They reported associated stock price decreases, although these were only significant when an insider is added to the
board of a company where inside directors own a significant proportion (5–25%) of the shares.
Consistent with the above study, Klein (1998) found only modest association between firm performance and
overall board composition with directors divided into insiders, outsiders and affiliates. She did however isolate
significant positive relationships between the percentage of executive directors on finance and investment subcommittees
with both accounting and market performance measures. Additionally, weak evidence is provided of a
positive relation between firm performance and the presence of at least one outside director holding at least five
percent of the firm’s equity.
Fich (2005) analysed 1,493 first-time director appointments to Fortune 1000 boards, for the period 1997–99, to
assess whether certain outside directors produce more positive share price reactions. The study found that appointees
who are CEOs of other companies result in a more positive share price reaction than for other appointments. In
addition, the study revealed positive long-term performance benefits in firms that appoint outside CEOs as directors.
However, the study found that the CEOs experienced negative stock-price effects at their own firms when they
accepted outside director positions and this was particularly pronounced if the CEOs’ own firms were faced with
significant growth opportunities.
Mura (2007) also examined the relationship between firm performance and board composition. The results,
which included a test for endogeneity among variables, confirmed that the direction of the relationship moves from
board composition to performance. A significant positive coefficient on the variable of the proportion of outside
directors showed that the proportion of non-executives on the board has a positive impact on firm performance. The
study supported the idea that non-executive directors are effective monitors of the firms’ management for this U.K.based
sample.
The level of outsider representation on the sub-committees of the main board has also been associated with the
independence of the board and reduced agency costs. For example in Australia, the Bosch Committee (1995)
proposed that audit committees should comprise a majority of non-executive directors. Stapledon and Lawrence
(1997) concluded in a study of the top 100 Australian companies that independent directors are better-placed to
effectively monitor executive management. Vance (1983), drew a similar conclusion believing that their
independence ensures both objectivity and that there are sufficient ‘checks and balances’ on managerial behaviour.
Menon and Williams (1994, p.125) similarly stated that the presence of executive managers on an audit committee
precludes them from being an objective monitor of management. Finally, Cotter and Silvester (2003, p.214) reported
that “independent audit committees can reduce agency costs by minimising the opportunistic selection of financial
accounting policies, and by increasing the credibility and accuracy of financial reporting”.
The independence of remuneration committees which determine the reward packages for senior executives is
also considered essential (Bosch 1995). Kesner (1988) believed this determination to be central to the monitoring
role of the board as it evaluates the performance of the managers in line with company goals and sets appropriate
rewards for performance. Cotter and Silvester (2003) noted that an independent committee is far more likely to
determine a fair and equitable reward package, thereby reducing agency costs, than if executives are responsible for
setting their own pay.
Laing and Weir (1999) studied the relationship between the governance structures proposed by the Cadbury
Committee (1992) and corporate performance in the U.K.. The sample studied consisted of 115 U.K. non-financial
quoted companies for the years 1992 and 1995. The study found a positive impact of board subcommittees on
performance (measured as return on assets) in both years. Companies with remuneration and audit committees
outperformed others. Additional evidence found was a significant improvement in performance for those firms
which introduced audit and remuneration committees during the periods under examination, suggesting that the
establishment of board committees is an effective monitoring mechanism. A further study by Weir and Laing (2000)
found the presence of a remuneration committee has a positive effect on performance as measured by market returns
but not on the accounting performance. Similarly, the return on assets is lower if firms have more outside directors
on the board, but this is not reflected in the market returns.
452

The independence of the board is also considered in the context of the leadership structure in operation, i.e. are
the roles of CEO and chair combined or separated? The empirical evidence on this question is limited however, but
generally favours that a separation of roles leads to improved performance. An empirical study by Rechner and
Dalton (1991) found that companies with separated roles outperform firms with combined roles. They assessed that
their findings “may provide empirical support for some strongly-worded admonitions about a governance structure
that includes the same individual serving simultaneously as CEO and board chairperson” (p.59).
Pi and Timme (1993) produced similar results in a sample of banks over the 1987–90 period. Their study, which
included controls for firm size and other variables, determined that the accounting-based measure (return on assets)
was higher for firms with separated roles. Baliga et al. (1996) used a market-based measure (Market Value Added)
to similarly determine that a dual leadership structure produced superior returns in the long run.
Elsayed (2007) examined the extent to which board leadership structure (as proxied by CEO duality) impacts
corporate performance in Egyptian-listed companies. Two alternative measures of corporate performance were used:
return on assets and Tobin’s Q ratio. The sample was taken from Egyptian public limited firms over the time period
2000–04. The data on board leadership structure, corporate performance and other related variables were available
for 92 firms in 19 different industrial sectors. The findings initially indicated that board leadership structure has no
direct impact on corporate performance. However, additional analysis revealed that the impact of CEO duality on
corporate performance varied with financial performance of firms and across industry types. Consistent with the
finding of Finkelstein and D’Aveni (1994), CEO duality is preferable in low-performance firms. Overall, the
findings support the conclusion of the meta-analysis of Rhoades, Rechner, and Sundaramurthy (2001) and the
argument of Boyd (1995) and Brickley, Coles, and Jarrell (1997), in that the relationship between board leadership
structure and corporate performance may vary within the context of firms and industry and that CEO duality will
only be advantageous for some firms whilst not for others.
Board sub-committees were investigated by Weir, Laing, and McKnight (2002) using 311 quoted, non-financial
U.K. firms covering the period 1994–96. They found little evidence that board structure affects performance. Their
results also indicated that the structure and quality of board subcommittees have little impact on performance – a
finding consistent with Klein (1998), Vafeas and Theodorou (1998) and Dalton et al. (1998). However, the study
found that companies in the top performance deciles have a greater proportion of independent directors both on their
boards and on their audit committees.
The literature on governance characteristics and performance is still developing, particularly in its application to
countries outside the U.S. and U.K. Generally, public policy has preceded in advance of rigorous empirical research
findings, with many countries imposing regulation or codes of conduct upon companies. These have usually been
based on the assumption that an independent board is in the best interest of shareholders, although the literature
provides only modest support for this view.
3 Hypothesis Development
Agency theory suggests the use of effective corporate governance mechanisms to mitigate manager–shareholder
conflicts and to monitor the performance of managers. Hypotheses relating to effective corporate governance
mechanisms are developed under three main categories: the proportion of independent directors who form part of
the board; the leadership structure of the CEO and chairperson; the existence and independence of various board
sub-committees; and the role of external auditors.
3.1.1 Proportion of outside directors
National corporate governance codes of conduct generally focus on how boards of directors should be structured in
order to generate independent control of companies. Most often they prescribe a minimum representation of nonexecutive
directors as a means of achieving sufficient independence from management (Bosch 1995; NACD 1996;
Holmstrom and Kaplan, 2003). Non-executive directors are believed to play an important role in monitoring, and
perhaps challenging if needed, management. This is supported in agency theory which suggests that effective
monitoring leads to a reduction in agency costs since managers have fewer opportunities to build their personal
wealth at the expense of shareholders.
453

Despite the above, empirical evidence on the value of non-executive directors on boards is mixed. Rosenstein
and Wyatt (1990) showed that a positive stock price reaction follows the appointment of non-executive directors to
company boards. Weir, Laing, and McKnight (2002) found that the presence of non-executive directors on U.K.
boards positively influenced the Tobin’s Q ratio of companies. Using a large panel dataset of U.K. firms for the
period 1991-2001, Mura (2007) found significant positive association between the proportional representation of
non-executives on the board and firm performance. Similarly, Choi, Park, and Yoo (2007) reported that outside
directors in Korea have a significant and positive effect on firm performance.
In contrast, Hermalin and Weisbach (1991), Mehran (1995), Klein (1998), Dalton et al. (1998), Vafeas and
Theodorou (1998) and Laing and Weir (1999) all found no significant performance relationship. A possible
explanation being that boards dominated by outsiders may lack company-specific operational knowledge to
effectively guide the company (Klien, 1998).
On balance, it is expected that a majority outsider representation best enables efficient monitoring activities
leading to a fall in agency costs. To investigate this relationship further, this study examines the relationship
between board composition and firm performance as follows:
Hypothesis 1:
The proportion of non-executive directors serving on the board will be positively associated with improved
monitoring of management and consequently, higher firm performance.
CEO duality
A question that has received growing attention in corporate governance literature is whether there is a relationship
between board leadership structure and corporate performance. It is often speculated that the presence of a
combined CEO/chair compromises the independence of the board as the individual has sufficient power to
unreasonably influence company decision-making (Cadbury 1992; Jensen 1993) and thereby reduce the board’s
ability to both monitor and discipline the management team. When the CEO is also the board chairman, a single
person holding both roles is more likely to dominate the board, as it “signals the absence of separation of the
decision management and the decision controls” (Fama and Jensen 1983, p. 314). This could render the board
ineffective in discharging its leadership and control duties (Daily and Dalton 1993; Jensen 1993), with CEOs free
“to pursue their own interests rather than the interests of shareholders” (Weisbach 1988, p. 435). A structure of this
type enhances CEO power and authority and compromises board independence (Finkelstein and D'Aveni 1994;
Rhoades, Rechner, and Sundaramurthy 2001), and eventually leads to the incapability of protecting the interest of
shareholders by boards of directors. Therefore, CEO duality is expected to increase agency costs and affect firm
performance negatively.
There have been some dissenting views, with researchers finding that the combination of the positions (CEO
and chairperson ) enhanced company performance (Brickley, Coles, and Jarrell 1997; Coles, McWilliams, and Sen
2001; Ying-Fen 2005).
While the empirical consensus around the relationship between CEO duality and corporate performance is not
resolved, both theoretical arguments and regulatory frameworks provide strong support for the separation of these
roles. Hence:
Hypothesis 2:
Separation of the roles of the Chief Executive Officer and the Chair of the Board will be positively related
to firm performance.
The existence of subcommittees
Board subcommittees are formed and used as another agency conflict-controlling mechanism by firms to organise
their boards in such a way that they can make most effective use of their directors, with much of the key decisionmaking
and implementation occurring at committee level (Kesner 1988; Bilimoria and Piderit 1994; Daily 1994,
1996; Ellstrand et al. 1999). It has been widely promulgated that boards of public companies should have separate
454

monitoring committees for auditing the company financial statements, supervising the compensation of executive
directors and controlling the selection process of new directors (Lipman 2007; The Sarbanes-Oxley Act of 2002).
The audit committee is responsible for nominating the outside auditor, overseeing the preparation of the financial
statements and annual reports, ensuring the efficacy of internal controls, and investigating allegations of material,
financial, ethical and legal irregularities (Anderson and Anthony 1986, cited in Ellstrand et al. 1999). The
remuneration committee is responsible for establishing the level of compensation for senior corporate executives
and corporate officers. In addition, the remuneration committee is charged with recommending appropriate
compensation for corporate directors (Fisher 1986, cited in Ellstrand et al. 1999). Finally, the nomination committee
is charged with the identification, selection, and evaluation of qualified candidates to serve in key positions within
the corporation. More specifically, this committee is responsible for the selection of the CEO, directors and other top
corporate executives (Vance 1983, cited in Ellstrand et al. 1999).
The relationship between board monitoring subcommittees and company performance (corporate value) is a
research area of corporate governance that has not been extensively studied (Gales and Kesner 1994; Dalton et al.
1998). Some early supportive empirical evidence is provided by Wild (1996) and Laing and Weir (1999) who
reported that a positive effect on firm performance resulted after the establishment of audit committees. According
to Main and Johnston (1993), Laing and Weir (1999) and Weir and Laing (2000), the presence of remuneration
committees is similarly positively-associated with improved performance of companies. Finally, Klein (1998) found
a positive (though weak) relationship between the presence of a remuneration committee and company performance.
These committees perform key functions; their presence contributes to the board’s monitoring role and
enhances the confidence of investors, not only in the reliability and fairness of company financial statements, but
also in the effectiveness of the corporate reward system (executive remuneration packages) and the quality of
appointed directors.
As a result, companies with such monitoring subcommittees are expected to outperform those without them. Hence:
Hypothesis 3a:
The number of board subcommittees established by the main board will be positively related to firm
performance.
The independence of board monitoring committees
A number of authors have argued that non-executive or independent directors on board subcommittees are more able
to exercise independent judgment and will therefore be more effective in performing their monitoring role (Laing
and Weir 1999; Stapledon and Lawrence 1997). Being independent from company executive positions, outside
directors are free to assess and evaluate management actions and judgments objectively as well as to make crucial
business decisions based upon moral grounds (Vance 1983; Ellstrand et al. 1999).
In the case of audit committees, independence will ensure that the financial viability and integrity of the
company are maintained and the interests of shareholders are being properly safeguarded. Cotter and Silvester
(2003) argue that “independent audit committees, thereby, can reduce agency costs by minimizing the opportunistic
selection of financial accounting policies, and by increasing the credibility and accuracy of financial reporting” (p.
214). There is some empirical support for this position with Weir, Laing and McKnight (2002) reporting that U.K.
firms with superior performance have a higher percentage of non-executives on the audit committee. Erickson, Park,
Reising, and Shin (2005) found a reduction in the negative impact of ownership concentration on the value of
Canadian firms when the proportion of outside directors on the audit committee increases.
The call for independence may similarly be applied to remuneration committees. “It is clearly important that the
remuneration committee should be able to arrive at its decisions independently so that suitable reward packages are
drawn up which would motivate and retain executive directors while protecting shareholder interests” (Laing and
Weir 1999, p. 458). As stated by Weir and Laing (2001, p. 88), “given that the aim of the remuneration committee is
to supervise the performance of the executive directors and to devise suitable reward packages, its effectiveness is
likely to be related to its structure and membership. It would therefore be expected that the remuneration committee
would be made up entirely of non-executive directors”. Perhaps more bluntly put by Williamson (1985, p. 313), who
commented that “the absence of an independent remuneration committee is akin to an executive’s writing their
employment contract with one hand and then signing it with the other”.
455

Finally, the independent structure of the nomination committee is also considered to be crucial. Ellstrand et al.
(1999) asserted that “a nominating committee that is composed of independent directors may be more likely to
appoint other independent directors who will be vigilant in monitoring the CEO”. Support is provided by Shivdasani
and Yermack (1999) who found that there is significant negative market reaction to and significantly lower
cumulative abnormal stock returns when there is an involvement of the CEO in the selection of company directors.
In relation to monitoring committees (audit, remuneration and nomination), the independence of each
committee member is expected to improve the ability to both monitor and discipline company management, in turn
reducing agency costs and increasing company performance. Hence:
Hypothesis 3b:
The independence of the audit committee will be positively related to firm performance.
Hypothesis 3c:
The independence of the remuneration committee will be positively related to firm performance.
Hypothesis 3d:
The independence of the nomination committee will be positively related to firm performance.
Auditor quality
An additional governance instrument through which shareholders can monitor managerial behaviour and
effectiveness is the independently-audited annual report to shareholders, which includes an auditor’s review of the
reliability of the financial statements prepared by management (Watts and Zimmerman, 1983). Auditors not only
provide shareholders with independently-ratified financial statements, but can also discover issues through internal
control systems, including fraud detection. This improves the credibility of financial statements issued by the
audited companies leading to lower contracting costs with external suppliers and lenders and therefore, a lower cost
of capital (Jensen and Meckling, 1976).
The quality of audit services may be defined as ‘the market-assessed joint probability that a given auditor will
both (a) discover a breach in the client’s accounting system, and (b) report the breach’ (DeAngelo 1981, p. 186).
From a shareholders perspective, the quality of the audit is not observable and the shareholder is therefore able to
make judgments based only on reputation of the auditor and possibly, price (where the fee is published). Given there
is no direct measure of either audit quality or reputations, the majority of studies in this area have relied upon auditor
size as a proxy measure for quality.
Extant empirical research supports the positive relationship between auditor size and quality, indicating that the
largest international audit firms provide above average quality of audit services (e.g. Palmrose 1988; Siew Hong and
Wong 1993; Niemi 2004). Niemi (2004) in summarizing the literature in this area finds that large audit firms are
associated with more accurate reports, lower litigation activity and greater compliance with GAAP reporting
requirements. They are also seen to be able to better withstand pressure from clients, and due to their greater
collateral, are seen to have more to lose in the case of an audit failure. This latter point was similarly revealed by
DeAngelo (1981) who found that the probability of an auditor finding and reporting on a problem in the accounting
system increased with audit firm size. Accordingly, auditor size as proxied by the participation of the ‘Big 4’ audit
firms will be used as an indicator of higher-quality auditors, lowering agency and capital costs which will ultimately
lead to better firm performance. Hence:
Hypothesis 4:
Auditor size will be positively related to firm performance.
Control Variables
A number of additional variables are included to control for other potential influences on the performance of firms.
Firm size
456

The firm’s market capitalisation is included to control for the potential effects of firm size on corporate
performance. Short and Keasey (1999) proposed two major avenues through which this effect may occur. Firstly, a
financing effect, in which larger firms find it easier to generate funds internally and to access funds from external
sources, lowering the overall cost of capital. Secondly, large firms may create higher entry barriers, thereby reducing
competition and benefitting from above-normal profits.
Debt ratio
The debt ratio is defined as the book value of total debt divided by total assets – and this influences company
performance in two ways. Firstly, the presence of debt ensures that management decisions and the firm’s operation
are being externally monitored by debt holders. Stiglitz (1985) contends that lenders, particularly banks, effectively
perform a function of management supervision. Secondly, the use of financial leverage creates contractual
obligations for managers to meet fixed future debt repayments, thereby reducing the funds available to management
for discretionary consumption of perks; moreover, debt requires management to become more efficient to reduce
both the probability of bankruptcy and the potential loss of their own reputation (Jensen and Meckling 1976;
Grossman and Hart 1982; Jensen 1986).
Industry classification
Industry effects account for the nature of the competitive environment in which a firm operates, including, the
number and size-dispersion of industry rivals and the rate of growth of the industry in general. Since performance
may also depend on industry affiliations, a number of studies have included a dummy variable to capture these
industry effects (Vafeas and Theodorou 1998; Ellstrand et al. 1999).
Board size
The disadvantages associated with large boards have been addressed by many authors. “When boards get beyond
seven or eight people, they are less likely to function effectively and are easier for the CEO to control” (Jensen
1993, p. 865). A board with “eight or fewer members engenders greater focus, participation, and genuine interaction
and debate” (Firstenberg and Malkiel 1994, p. 34). According to Goodstein, Gautam, and Boeker (1994), strategic
actions and changes are less likely to be initiated when there are a large number of board members. Yermack (1996),
who first empirically documented a significant inverse relation between board size and firm performance,
concluded that the costs associated with large boards (e.g. coordination, communication and director free-riding
costs) are not sufficiently offset by its benefits alone. This study includes consideration of the number of directors on
the board both as a control variable and also as an attempt to expand the literature linking board size with corporate
performance.
4 Sample Selection and Research Design
To examine the relationship between governance and firm performance, 250 companies were randomly selected
from a population of all companies listed in the Australian Stock Exchange in the 2005 financial year. The year
2005 was chosen as it was a period of heightened interest in governance and governance structures. This followed
from the collapse of companies such as Enron and Worldcom. Indeed the two years prior (2003and 2004) were
characterized by the high level of activity in governance reviews and legislative changes. A study conducted in the
year 2005 provided the first opportunity to assess governance and performance, incorporating mandatory changes in
the areas of sub-committees, audit and composition. Finance-related companies including banking, insurance and
trust companies were excluded from the sample, as their accounting reporting requirements and capital structure
varies greatly from other companies and would distort the overall results. Missing data arises as a result of
inadequate disclosure resulting in an inability to distinguish the role of CEO and board chair (Table 1).
250 companies randomly selected from a population of all Australian companies listed on the ASX. Finance-related 250
companies including banking, insurance and trust companies were excluded from the sample
Identities of CEOs and/or Chair of Board not disclosed resulting in insufficient information to determine the company (11)
leadership structure (CEO duality)
Sample remaining after listwise deletion 239
Table 1: Sample Size for the Governance-Performance Model
457

In the corporate governance literature, the dependent variable firm performance has been measured as either:
market-based (de Miguel 2004; Mura 2007); accounting-based (Dhnadirek and Tang 2003; Ng 2005); or both (Short
and Keasey 1999; Bonn 2004; Guedri and Hollandts 2008). The majority of studies have followed the prescription
of the Demsetz and Lehn (1985) study by using Tobin’s Q as the measure of firm performance. This is seen to have
an advantage over accounting performance by incorporating a current perspective of the position of the firm (as
determined by market price), rather than an historical perspective based on accounting results as measured by
accounting conventions (Demsetz and Villalonga 2001). In accepting this approach, this study employs Tobin’s Q—
which measures the degree to which the market values the firm above (or below) the book value of its assets—and
provides an assessment of the efficiency with which management is utilising those assets. Table 2 shows the mean
Tobin’s Q for the 2005 financial year (2.008), consistent with an earlier Australian survey of 114 listed companies in
the financial year 1999-2000 which found a mean Tobin’s Q of 1.80 (Welch, 2003). Approximately 75% of
Australian firms, report their Tobin’s Q to be higher than 1.0 in 2005.
Table 2: Mean, Median and Quartile Range for Tobin’s Q for the Year 2005
Country Min. Max. Mean Median Percentiles
Australia 0.000 10.430 2.008 1.460 0.790 1.018 2.393 3.845
Tobin’s Q is regressed against seven variables of corporate governance mechanisms to measure their impact on firm performance. The study,
therefore, specifically tests the following model:
Tobin’s Q = a + β1ProNED + β2Duality + β3Subcommittees + β4AuditIndp + β5RemuIndp + β6NomiIndp + β7Auditor + γControl Variables
Where:
Tobin’s Q =The sum of market value of equity and preferred shares and book value of total liabilities, divided by the book value of
total assets.
ProNED = The proportion of non-executive directors on the board
Duality = A dummy variable is set equal to 1 if the positions of CEO and COB are either held by
the same person or two different persons with the same family name and zero otherwise
Subcommittees = An index scored from 0–3, with 1 point scored for the disclosure of each of three board
subcommittees: Audit, Remuneration and Nomination
AuditIndp = The proportion of non-executive directors in the Audit committee
RemuIndp = The proportion of non-executive directors in the Remuneration committee
NomiIndp = The proportion of non-executive directors in the Nomination committee
Auditor = A dummy variable is set equal to 1 if the firm’s audit company is one of the following
large audit firms: KPMG, E&Y, PWC or DTT, and zero otherwise
Control Variables:
Firm Size = The natural log of market capitalisation.
Debt Ratio = Total liabilities as a proportion of total assets
Industry = 1 for companies in mining and resource sectors; and 0 otherwise
Board Size = The number of directors in the boards
5 Results
Two models are presented to address multicollinearity detected among the subcommittee and subcommittee
independent variables.
When both were included in the regression model an unacceptable high tolerance statistic of 10.741 was detected for
the subcommittee variable (Tabachnik and Fidell, 1996).
Model 1 (Table 3) produces the coefficients on the variables ProNED and Duality with expected signs—positive and
negative, respectively— consistent with agency theory. Contrary to expectations, the number of subcommittees
(Subcommittees) and auditor size (Auditor) produce a negative relationship with firm performance. Results are
similar when the number of board monitoring committee’s variable (Subcommittees) is removed as shown in Model
2. Among the control variables, the positive coefficients for firm size and debt variables are consistent with the
arguments discussed earlier. A negative and statistically significant coefficient on board size suggests larger boards
may not enhance company performance.
Table 3: Regression analysis of Tobin’s Q on governance practices, firm size, debt ratio, industry and board size for listed Australian companies
in 2005 (p-values in parentheses below coefficients)
_______________________________________________________________________
Variable a
(1) (2)
________________________________________________________________________
458
10 th
25 th
75 th
90 th

Constant –1.110 –1.167
(0.157) (0.131)
ProNED 0.924* 0.987*
(0.097) (0.067)
Duality –0.449* –0.458*
(0.095) (0.088)
Subcommittees –0.135
(0.639)
_
AuditIndp –0.369 –0.505
(0.396) (0.118)
RemuIndp –0.176 –0.309
(0.667) (0.290)
NomiIndp –0.324 –0.476
(0.465) (0.113)
Auditor –0.101 –0.095
(0.646) (0.665)
Firm Size 0.231*** 0.229***
(0.000) (0.000)
Debt Ratio 0.300 0.286
(0.345) (0.365)
Industry –0.190 –0.183
(0.383) (0.399)
Board Size –0.165** –0.159**
(0.036) (0.040)
________________________________________________________________________
R 2
0.135 0.135
Adjusted R 2
0.094 0.097
F-Statistic 3.233*** 3.546***
Sample Size 239 239
Where: ***p < 0.01, **p < 0.05, and *p < 0.10
Tobin’s Q =The sum of market value of equity and preferred shares and book value of total liabilities, divided by the book value of
total assets.
ProNED = The proportion of non-executive directors on the board
Duality = A dummy variable is set equal to 1 if the positions of CEO and COB are either held by
the same person or two different persons with the same family name and zero otherwise
Subcommittees = An index scored from 0–3, with 1 point scored for the disclosure of each of three board
subcommittees: Audit, Remuneration and Nomination
AuditIndp = The proportion of non-executive directors in the Audit committee
RemuIndp = The proportion of non-executive directors in the Remuneration committee
NomiIndp = The proportion of non-executive directors in the Nomination committee
Auditor = A dummy variable is set equal to 1 if the firm’s audit company is one of the following
large audit firms: KPMG, E&Y, PWC or DTT, and zero otherwise
Firm Size = The natural log of market capitalization.
Debt Ratio = Total liabilities as a proportion of total assets
Industry = 1 for companies in mining and resource sectors; and 0 otherwise
Board Size = The number of directors in the boards
Overall, the findings suggest an increase in the proportion of non-executive directors on boards will result in an
increase in corporate value as a result of effective monitoring of duties performed by independent directors.
Additionally, firms that employ the unitary leadership structure are shown to underperform those firms that assign
two unrelated persons to the posts of CEO and chair. The roles of subcommittees, irrespective of their independence,
and auditor quality did not produce significant results for this data.
The results for the control variables of firm size, financial leverage and board size were also found to be consistent
in direction across all models in both countries. Firm size and debt ratio are positively associated with performance,
although the latter is not significant for this data, while board size produces a significant negative relationship.
6 Conclusion
The results of the influences of the board monitoring level on firm performance were inconclusive. The data provide
no consistent evidence that regulatory pronouncements or shareholder agitation regarding establishment and
independence of board subcommittees are effective in enhancing firm performance. Nevertheless, two exceptions
were noted with respect to the NED representation and dual leadership structure which appear to be influential in
improving firm performance.
459

The proportion of non-executive directors was found to positively affect firm performance (H1) in line with the
evidence on U.S. firms provided by Rosenstein and Wyatt (1990), on U.K. firms provided by Mura (2007), and on
large Australian firms provided by Bonn (2004) and Bonn, Yoshikawa, and Phan (2004). This supports the
contention that non-executives on the board enhance firm performance as they are able to independently scrutinise
management action.
The leadership structure was found to be in the predicted direction (H2), where firms with a non-executive chair
outperform firms that assign a CEO as board chairman. The results found in this study are in line with others (e.g.
Rechner and Dalton 1991; Pi and Timme 1993; Daily and Dalton 1994; Rhoades et al. 2001), who suggest that
dividing these two positions enhances corporate performance.
The remaining governance-related hypotheses were not supported in this study. The subcommittee variable
(H3a) was not significant and contrary to expectation, was negatively-correlated to performance. Board committee
composition (H3b, c, d ) was similarly not statistically significant and negative. However, this result mirrors the
insignificant relationships generally discovered in previous studies into the influence of subcommittee independence
on firm performance (Klein 1998; Vafeas and Theodorou 1998; Ellstrand et al. 1999; Cotter and Silvester 2003; Hsu
2008). Audit quality as measured by the auditor size was not found to have any influence on firm performance.
The control variables of firm size and debt ratio are positively associated with performance and board size
produces a significant negative relationship. The firm size finding is particularly important in confirming the results
of a number of international studies and lends support for the contention that large boards are inefficient.
Overall, the results presented suggest that regulators concentration on board subcommittees and their
composition would seem to be ineffectual in aligning the interests of managers and shareholders. No support could
be found for the contention that an increase in the number of subcommittees and/or NED representation on those
subcommittees would improve firm performance. This has been supported by a growing number of studies that
failed to find a systematic relationship between the structure and composition of board subcommittees and corporate
financial performance (Klein 1998; Vafeas and Theodorou 1998; Ellstrand et al. 1999; Cotter and Silvester 2003;
Hsu 2008). If regulators continue to pursue an agenda of compulsorily requiring the establishment of
subcommittees, or insisting on a higher proportion of non-executive directors on those subcommittees, it may be
counterproductive to business generally. Further research is required to identify the causes of this disparity between
theory and practical outcomes, before more costly regulation is enacted.
The study does find support for NED representation on the board of directors and a dual leadership structure in
improving firm performance. It is notable that 13% of the Australian companies did not achieve a majority nonexecutive
director representation and 19% of Australian companies did not utilise a dual leadership structure. The
study provides an argument that these companies should consider increasing the proportion of non-executive
directors on the board and assigning two unrelated persons to the posts of CEO and chair of the board.
This research could be extended upon in a number of ways. A replication of the study employing longitudinal
data would assist in determining whether the relationships found were robust under differing economic conditions.
Tobin’s Q was employed as the dependent variable in this analysis, to provide an assessment of the efficiency with
which management is utilising assets. Future research could assess the impact of the governance mechanisms on
other performance-related variables, including accounting-based measures (ROA, ROE and ROI) – which provide
an historical view of performance based on accounting conventions. The possibility of using more refined measures
to assess the characteristics of the board and its committee members including their experience and expertise—
would enhance our understanding of the role of the committees and their ultimate impact on performance. This may
include qualitative aspects which investigate the management process and directly interrogate the participants (for
example, how influential are NEDs in the decision making process), rather than relying on reported board structures.
7 References
Bilimoria, D., and S. K. Piderit. 1994. Board committee membership: Effects of sex-based bias. Academy of
Management Journal 37 (6): 1453–1477.
Bonn, I. 2004. Board structure and firm performance: Evidence from Australia. Journal
460

of the Australian and New Zealand Academy of Management 10 (1): 14–24.
Bosch, H. 1995. The Director at Risk: Accountability in the Boardroom. Melbourne: FT Pitman Publishing.
Boyd, B. K. 1995. CEO duality and firm performance: A contingency model. Strategic Management Journal 16 (4):
301–312.
Brickley, J. A., J. L. Coles, and G. Jarrell. 1997. Leadership structure: Separating the CEO and chairman of the
board. Journal of Corporate Finance 3 (3): 189–220.
Cadbury, A. 1992. Committee on the Financial Aspects of Corporate Governance. London: GEE.
Cadbury Committee Report 1992. The financial aspects of corporate governance.
Choi, J. J., S. W. Park, and S. S. Yoo. 2007. The value of outside directors: Evidence from corporate governance
reform in Korea. Journal of Financial & Quantitative Analysis 42 (4): 941–962.
Coles, J. W., V. B. McWilliams, and N. Sen. 2001. An examination of the relationship of governance mechanisms to
performance. Journal of Management 27 (1): 23.
Cotter, J., and M. Silvester. 2003. Board and monitoring committee independence. Abacus 39 (2): 211–232.
Daily, C. M. 1994. Bankruptcy in strategic studies: Past and promise. Journal of Management 20 (2): 263.
Daily, C. M. 1996. Governance patterns in bankruptcy reorganizations. Strategic Management Journal 17 (5): 355–
375.
Daily, C. M., and D. R. Dalton. 1993. Board of directors leadership and structure: Control and performance
implications. Entrepreneurship: Theory & Practice 17 (3): 65–81.
Dalton, D. R., C. M. Daily, A. E. Ellstrand, and J. L. Johnson. 1998. Meta-analytic reviews of board composition,
leadership structure, and financial performance. Strategic Management Journal 19 (3): 269.
de Miguel, Alberto. 2004. Ownership structure and firm value: New evidence from Spain. Strategic Management
Journal 25 (12): 1199–1207.
DeAngelo, L. E. 1981. Auditor size and audit quality. Journal of Accounting and Economics 3 (3): 183–199.
Demsetz, H., and K. Lehn. 1985. The structure of corporate ownership: Causes and consequences. Journal of
Political Economy 93 (6): 1155–1177.
Demsetz, H., and B. Villalonga. 2001. Ownership structure and corporate performance. Journal of Corporate
Finance 7 (3): 209–233.
Dhnadirek, R., and J. Tang. 2003. Corporate governance problems in Thailand: Is ownership concentration the
cause? Asia Pacific Business Review 10 (2): 121–138.
Ellstrand, A. E., C. M. Daily, J. L. Johnson, and D. R. Dalton. 1999. Governance by committee: The influence of
board of directors' committee composition on corporate performance. Journal of Business Strategies 16 (1):
67–88.
Elsayed, K. 2007. Does CEO duality really affect corporate performance? Corporate Governance: An International
Review 15 (6): 1203–1214.
Erickson, J., Y. W. Park, J. Reising, and H.-H. Shin. 2005. Board composition and firm value under concentrated
ownership: the Canadian evidence. Pacific-Basin Finance Journal 13 (4): 387–410.
Fama, E. F. 1980. Agency problems and the theory of the firm. Journal of Political Economy 88 (2): 288–307.
Fama, E. F., and M. C. Jensen. 1983. Separation of ownership and control. Journal of Law & Economics 26 (2):
301–326.
Fich, E. M. 2005. Are some outside directors better than others? Evidence from director appointments by Fortune
1000 firms. Journal of Business 78 (5): 1943–1971.
Finkelstein, S., and R. A. D'Aveni. 1994. CEO duality as a double-edged sword: How boards of directors balance
entrenchment avoidance and unity of command. Academy of Management Journal 37 (5): 1079–1108.
Firstenberg, P. B., and B. G. Malkiel. 1994. The Twenty-first Century boardroom: Who will be in charge? Sloan
Management Review 36 (1): 27–35.
Gales, L. M., and I. F. Kesner. 1994. An analysis of board of director size and composition in bankrupt
organizations. Journal of Business Research 30 (3): 271–282.
Goodstein, J., K. Gautam, and W. Boeker. 1994. The effects of board size and diversity on strategic change.
Strategic Management Journal 15 (3): 241–250.
Grossman, S., and O. Hart. 1982. Corporate financial structure and managerial incentives.
In The Economics of Information and Uncertainty, ed. J. McCall, 107–137. Chicago: University of Chicago Press.
Guedri, Z., and X. Hollandts. 2008. Beyond dichotomy: The curvilinear impact of employee ownership on firm
performance. Corporate Governance: An International Review 16 (5): 460–474.
Hermalin, B. E., and M. S. Weisbach. 1991. The effects of board composition and direct incentives on firm
performance. FM: The Journal of the Financial Management Association 20 (4): 101–112.
461

Holmstrom, B., and S. N. Kaplan. 2003. The state of U.S. corporate governance: What’s right and what’s wrong?
Journal of Applied Corporate Finance 15 (3): 8–20.
Jensen, M. C. 1993. The modern industrial revolution, exit, and the failure of internal control systems. Journal of
Finance 48 (3): 831–880.
Jensen, M. C., and W. H. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownership
structure. Journal of Financial Economics 3 (4): 305–360.
Kesner, I. F. 1988. Directors' characteristics and committee membership: An investigation of type, occupation,
tenure, and gender. Academy of Management Journal 31 (1): 66–84.
Klein, A. 1998. Firm performance and board committee structure. Journal of Law and Economics 41 (1): 275–303.
Laing, D., and C. M. Weir. 1999. Governance structures, size and corporate performance in U.K. firms.
Management Decision 37 (5/6): 457.
Lipman, F. D. 2007. Summary of major corporate governance principles and best practices. International Journal of
Disclosure and Governance 4 (4): 309–319.
Main, B. G. M., and J. Johnston. 1993. Remuneration committees and corporate governance. Accounting & Business
Research 23 (91A): 351–362.
Mallin, C. A. 2004. Corporate governance. Oxford: Oxford University Press, 2004.
Mehran, H. 1995. Executive compensation structure, ownership, and firm performance. Journal of Financial
Economics 38 (2): 163–184.
Menon, K., and J. D. Williams. 1994. The use of audit committees for monitoring. Journal of Accounting & Public
Policy 13 (2): 121–139.
Mura, R. 2007. Firm Performance: Do non-executive directors have minds of their own? Evidence from U.K. panel
data. Financial Management (Blackwell Publishing Limited) 36 (3): 81–112.
NACD. 1996. A practical guide for corporate directors. Washington: National Association of Corporate Directors.
Ng, C. Y. M. 2005. An empirical study on the relationship between ownership and performance in a family-based
corporate environment. Journal of Accounting, Auditing & Finance 20 (2): 121–146.
Niemi, L. 2004. Auditor size and audit pricing: Evidence from small audit firms. European Accounting Review 13
(3): 541–560.
Palmrose, Z.-V. 1988. An analysis of auditor litigation and audit service quality. Accounting Review 63 (1): 55.
Pi, L., and S. G. Timme. 1993. Corporate control and bank efficiency. Journal of Banking & Finance 17 (2-3): 515–
530.
Rechner, P. K., and D. R. Dalton. 1991. CEO duality and organizational performance: A longitudinal analysis.
Strategic Management Journal 12 (2): 155–160.
Rhoades, D. L., P. L. Rechner, and C. Sundaramurthy. 2001. A meta-analysis of board leadership structure and
financial performance: “Are two heads better than one”? Corporate Governance: An International Review 9
(4): 311–319.
Rosenstein, S., and J. G. Wyatt. 1990. Outside directors, board independence, and shareholder wealth. Journal of
Financial Economics 26 (2): 175–191.
Rosenstein, S., and J. G. Wyatt. 1997. Inside directors, board effectiveness, and shareholder wealth. Journal of
Financial Economics 44 (2): 229–250.
Sarbanes-Oxley Act 2002. The Public Company Accounting Reform and Investor
Protection Act (H.R. 3763). http://www.whitehouse.gov/infocus/corporateresponsibility.
Schmid, M. M., and H. Zimmermann. 2008. Should chairman and CEO be separated? Leadership structure and firm
performance in Switzerland. Schmalenbach Business Review : ZFBF 60: 182–204.
Shivdasani, A., and D. Yermack. 1999. CEO involvement in the selection of new board members: An empirical
analysis. Journal of Finance 54 (5): 1829–1853.
Short, H., and K. Keasey. 1999. Managerial ownership and the performance of firms: Evidence from the U.K..
Journal of Corporate Finance 5 (1): 79–101.
Siew Hong, T., and T. J. Wong. 1993. Perceived auditor quality and the Earnings Response Coefficient. Accounting
Review 68 (2): 346–366.
Stapledon, G. P., and J. Lawrence. 1997. Board composition, structure and independence in Australia's largest listed
companies. Melbourne University Law Review 21 (1).
Stiglitz, J. E. 1985. Credit markets and the control of capital. Journal of Money, Credit & Banking 17 (2): 133–152.
Tabachnik, B. G., and L. S. Fidell. 1996. Using multivariate statistics (third edition). New York: Harper Collins.
Vafeas, N., and E. Theodorou. 1998. The relationship between board structure and firm performance in the U.K..
The British Accounting Review 30 (4): 383–407.
Vance, S. C. 1983. Corporate leadership, boards, directors and strategy. New York: McGraw-Hill.
462

Watts, R. L., and J. L. Zimmerman. 1983. Agency problems, auditing, and the theory of the firm: Some evidence.
Journal of Law & Economics 26 (3): 613–633.
Weir, C., D. Laing, and P. J. McKnight. 2002. Internal and external governance mechanisms: Their impact on the
performance of large U.K. public companies. Journal of Business Finance & Accounting 29 (5/6): 579–611.
Weir, C. M., and D. Laing. 2000. The performance-governance relationship: The effects of Cadbury compliance on
U.K. quoted companies. Journal of Management and Governance 4 (4): 265–281.
Weir, C. M., and D. Laing. 2001. Governance structures, director independence and corporate performance in the
U.K.. European Business Review 13 (2): 86–93.
Weisbach, M. S. 1988. Outside directors and CEO turnover. Journal of Financial Economics 20: 431–460.
Welch, E. 2003. The relationship between ownership structure and performance in listed australian companies.
Australian Journal of Management 28 (3): 287–305.
Wild, J. J. 1996. The audit committee and earnings quality. Journal of Accounting, Auditing & Finance 11 (2): 247–
276.
Williamson, O. E. 1985. The economic institutions of capitalism: Firms, markets, relational contracting. New York:
Fress Press.
Yermack, D. 1996. Higher market valuation of companies with a small board of directors. Journal of Financial
Economics 40 (2): 185–211.
Ying-Fen, L. 2005. Corporate governance, leadership structure and CEO compensation: Evidence from Taiwan.
Corporate Governance: An International Review 13 (6): 824–835.
463

COST-BENEFIT ANALYSIS OF THE FINANCIAL STATEMENTS CONVERSION: A CASE STUDY
FROM THE CZECH REPUBLIC
David Procházka, University of Economics, Prague, Czech Republic
Email: prochazd@vse.cz, https://webhosting.vse.cz/prochazd/
Abstract. EU Regulation No. 1606/2002 obliges companies listed on the EU stock exchanges to prepare their consolidated financial
statements in compliance with IFRS. Many Czech companies are under control of foreign companies (issuers of securities listed on
EU capital markets). As the IFRS statements are not relevant for the tax purposes, Czech companies prepare financial statements
according to the Czech legislative for statutory purposes. However, for consolidation purposes, they have to provide their parent
companies with financial statements prepared in compliance with IFRS as adopted by the EU.
The paper deals with three basic solutions, which can be applied when an entity is engaged in the conversion of financial statements
from one set of accounting standards to another set of standards. The first method uses conversion on the financial statements level.
The second method applies the conversion on the trial balance level. Finally, some companies prefer to implement specialised
accounting software, which enables to record all accounting transactions twice – both according to CAS and IFRS. Each method is
shortly described and its main cost-benefits are analysed. The theoretical findings will be supported by empirical evidence based on
questionnaire scrutinising practical experience of selected Czech companies involved in the process of financial statements
conversion. Moreover, a short description of expected future development is outlined.
Keywords: Conversion of Financial Statements; IFRS; Czech Accounting Standards; Dual (Financial) Accounting System
JEL classification: M41
1 Introduction
The adoption of the International Financial Reporting Standards has caused a radical change in financial reporting,
esp. in countries with the code-law tradition of accounting regulation. Increased usefulness of financial statements
prepared in accordance with the IFRS worldwide is evidenced by recent research. However, the implementation of
the IFRS into national legislation elicits costs for many subjects involved in the process. E.g. many national
legislations decree entities to prepare the IFRS statements for European stock exchanges and simultaneously to
prepare the financial statements based on national accounting standards for statutory and/or tax purposes. As a
consequence, entities have to maintain two different sets of accounting data. The conversion of financial statements
from one set to another is a complex and costly process. The paper’s main aim is to evaluate recent development in
this field in the Czech Republic.
2 Background
2.1 Literature overview
The IFRS adoption has increased the quality of disclosed information comparing to national GAAP (Barth et al.,
2008). The improvement in quality is significant across Europe (Aubert and Grudnitski, 2009) and esp. in countries
with code law (Morais and Curto, 2007), where accounting is closely linked to taxation systems. The increase in
value relevance is demonstrated in countries with significant level of discretion in financial reporting, such as Italy
(Cordazzo, 2008) or Spain (Ferrer et al., 2009). The positive influence of the IFRS adoption is also evidenced in
transitional countries, e.g. in Romania (Mustata et al., 2009), Poland (Jaruga et al., 2007), Russia (Bagaeva, 2009).
The evidence confirming the value relevance of the IFRS is also available for countries, which traditionally focus on
supplying the high-quality information for external users, such as United Kingdom (Christensen et al., 2009).
As the previous summary of literature shows, the benefits of IFRS adoption are significant and can be traced in
many areas of economy. On the other hand, not negligible costs are induced by the harmonisation process. Entities
are obliged to prepare financial statements, which are usually more complex than under local GAAP; investors have
to recustomise models computing intrinsic values of corporate securities, creditors need to adjust their risk
assessment models, state authorities solve the problem how to ensure control of tax duty fulfilment under new
accounting system, etc. Not all costs of the IFRS implementation are borne by those who take advantage of more
464

useful and comparable financial statements. Therefore, a comprehensive cost-benefit analysis of the IFRS
implementation is difficult because it is unfeasible to compare benefits of one group involved with costs of another
group.
2.2 Aims of the paper
The paper focuses on firms preparing financial statements in compliance with the IFRS. More specifically,
advantages, disadvantages and costs of various methods used for the preparing of IFRS financial statements will be
scrutinised. The problem seems to be pretty trivial at the first sight. The method used and cost incurred should be
very similar as under any else reporting system. However, we should have on mind that financial reporting is a
subject of state regulation. Besides IFRS statements for stock exchanges, entities are also bound to prepare financial
statements according to national accounting standards for statutory and/or tax purposes in many countries (e.g. the
EU member states). As a consequence, entities face to a problem how to ensure the preparation of two sets of
financial statements, usually substantially different, in order to meet all legal requirements with minimal costs. This
issue will be referred as the conversion of financial statements further in the text.
Advantages and disadvantages of various approaches to the conversion of financial statements will be
evaluated. Further, a short description of regulatory framework for financial reporting within the European Union
together with the analysis of accounting regulation in the Czech Republic will be performed. Finally, empirical
evidence will be presented to support some theoretical findings.
The Czech Republic as the “case country” has been chosen for two reasons. The conversion of financial
statements is an accounting issue of great importance because about 40% of Czech companies prepare two sets of
financial statements. Secondly, accounting profession has been striving to persuade the regulator of accounting in
the Czech Republic (i.e. the Ministry of Finance) to undertaken certain measures and thus to improve rather
unsatisfactory situation. After many years, the Ministry of Finance has reflected the effort of accounting profession
and amended the Act on Accounting by enabling selected entities to apply the IFRS on a voluntary basis. The
development in the Czech Republic can serve as an inspiration for the regulators of financial reporting in other
countries, both in positive and negative sense. We can assume that countries, whose accounting regulation is based
on the code law approach and accounting is subordinated to tax system requirements, face a similar problem.
3 Conversion of financial statements
The conversion of financial statements can be defined as a process when an entity prepares two or more sets of
financial statements for external users, each in compliance with distinct financial reporting standards. Financial
statements different from statutory accounts for bank credit purposes; financial statements based on national
legislation for tax purposes by listed companies; statements according to foreign GAAP for the purpose of
consolidation by parent company domiciled abroad or financial statements in accordance with generally accepted
principles instead of local GAAP for stock exchanges are common examples of this conversion.
There are three dimension of financial statements conversion:
� preparation of the first individual financial statements (opening balance sheet respectively) prepared in
compliance with alternative set of financial standards;
� reporting of individual financial statements and other figures needed for consolidation or other purposes in
regular intervals;
� consolidation of individual financial statements.
The number of particular steps in each mentioned phase may differ depending on the purpose of conversion.
Whether a company makes conversion on its own or whether it uses a template (e.g. prepared by parent company) is
another factor influencing the process of conversion.
The choice of a proper method of financial statements conversion is relevant especially in case of frequent
regular (e.g. monthly) reporting. No general advice, which solution to adopt, exists. Each entity should take into
account its specific conditions and chose an approach mixing benefits and cost in the most favourable manner. The
choice of conversion method appropriate for an entity depends on many factors, at least labour and ICT costs,
number and type of differences, frequency of and deadlines for reporting should be taken into consideration. There
are three basic approaches how to transform financial statements from one set of accounting standards to another set
(Mejzlík, 2006):
465

� conversion on financial statements level;
� conversion on trial balance level;
� dual accounting system.
3.1 Conversion on financial statements level
This method uses only the reclassification of items presented in financial statements. The main advantages of the
method are:
� easy and quick to implement;
� no specialised ICT is needed;
� low cost and labour burden;
� easy to check the correctness of adjustments.
The disadvantages are:
� applicable only if the number of differences is low (no measurement, recognition, accounting policies
issues);
� workable only for those cases when classification is the only difference.
3.2 Conversion on trial balance level
In this case, the list of accounts (trial balance) based on CAS rules is exported from an accounting software and then
adjustments are made in spreadsheets (like Excel). The following pros can be identified:
� no specialised ICT is needed (data spreadsheets are sufficient),
� applicable even if the number of differences is higher.
The cons of this method are:
� applicable only for difference in recognition of items (provisions, IFRS 3 recognised assets); however not
workable for different measurement issues (work-in progress) and accounting policies (depreciation);
� additional accounting expert for the IFRS should be employed => higher salary expenses;
� conversion system is designed by the expert => his/her substitution in case of illness or termination of the
job is questionable and sometimes even excluded;
� testing of the correctness and conclusiveness of the „conversion bridge“ is complicated and causes
additional problems esp. for auditors;
� the consistency of adjustments within periods and data relationships (e.g. retained earnings) is hard to hold;
� archiving of underlying documents outside the accounting system is an open issue.
Moreover, this method is indecisive as far as meeting reporting deadlines concerns. There are no additional
transactions to be recorded into the accounting system; however, after all transactions are recorded, the whole
conversion process has to be carried out.
Timely and error-free data recording is a limiting factor of this method. Most delays are caused because
deadlines for recording of transactions are not held, e.g. missing transactions (due to delay of underlying
documentation) are accounted for additionally and the whole conversion must be run once again. Predefined tables,
macros and other automatic calculations may mitigate the negative consequences of those delays although not in all
cases.
3.3 Dual accounting system
The accounting system is set up so it enables entities to record all transactions twice regarding on the different
requirements of the both accounting system. There are following plusses of this method:
� all types of differences could be included,
� conclusiveness and objectivity is secured as all „adjustments“ are incurred directly in SW modules,
� customisation of accounting SW is possible and more outputs for management are available,
� proper if number of difference is very high,
� possible integration with systems for consolidation reporting and other ICT systems.
This method has following minuses:
� implementation of a new or upgrade of current SW is needed => additional costs and changes in processes,
� more transactions are recorded (more workload and additional employees => high labour costs),
� question is how to record the transactions (all transaction to record twice in each module or to make special
module for different transactions only),
466

� way of archiving of the documents is not clear-cut (shall be documents numbered, ordered and stored
according to local GAAP or IFRS or in a combined manner?).
This method of conversion of financial statements brings uncertain results regarding the reporting deadlines.
More transactions need to be recorded into accounting software. However, if all is recorded, additional adjustments
are not needed, as both sets of financial statements can be exported directly from accounting software. Moreover,
recording of additional transactions do not cause any serious delays, because updated IFRS statements can be
retrieved from accounting software immediately.
4 Implementation of the IFRS in the Czech Republic
4.1 Regulation of financial reporting in the Czech Republic
As a member state of the EU, the regulation of accounting in the Czech Republic shall conform to the EU
legislation. The chief source of the EU guidance comes from the following documents.
� Fourth Council Directive 78/660/EEC of 25 July 1978 based on Article 54 (3) (g) of the Treaty on annual
accounts of certain types of companies (incl. subsequent amendments);
� Seventh Council Directive 83/349/EEC of 13 June 1983 based on the Article 54 (3) (g) of the Treaty on
consolidated accounts;
� Regulation (EC) 1606/2002 of the European Parliament and of the Council of 19 July 2002 on the
application of international accounting standards.
The Directives were incorporated directly into the Act on accounting already in 1990s.
“For each financial year starting on or after 1 January 2005, publicly traded companies governed by the law of
a Member State shall prepare their consolidated accounts in conformity with the international accounting
standards” in accordance with the Article 4 of Regulation (EC) 1606/2002.
Regulation on International Accounting Standards ordains a duty to prepare consolidated financial statements
by publicly traded companies. Member states may broaden the scope of entities, which are obliged/allowed to apply
the IFRS (e.g. in individual financial statements of listed companies or in individual/consolidate financial statements
of non-listed companies).
The main means of accounting regulation is the code law. The Czech accounting and financial reporting is
regulated by:
� Act No. 513/1991 Coll., Commercial code;
� Act No. 563/1991 Coll., on accounting;
� Decree of Ministry of Finance No. 500/2002 Coll., amending some enactments of Act No. 563/1991 Coll.,
on accounting, in the financial statements of business enterprises;
� Czech Accounting Standards (further “CAS”) for business enterprises subject to Decree of Ministry of
finance No. 500/2002 Coll.
Act on accounting until 2010
Despite the fact that regulations are generally binding in its entirety and directly applicable in all member states
of the EU without any further implementation in the national legislation, the provisions of Regulation (EC)
1606/2002 were incorporated directly in the Act No. 563/1991 Coll., on accounting. The obligation to prepare
consolidated financial statements accoutring to the International Accounting Standards as adopted by the EU by
companies listed on the EU capital markets is included in §23a, article 1. However, the Czech regulator of
accounting went one step forward and set up a duty to prepare individual financial statements by affected
companies. According to § 19, article 9 “entities, which are business companies and which are the issuers of
securities publicly traded on a regulated market in the member states of the European Union, shall apply the
International Accounting Standards as adopted by the EU for keeping their accounts and for preparation of
financial statements”.
In addition, IFRS can be applied in consolidated financial statements of non-listed companies voluntarily (§23a,
article 2). No other entities were allowed to choose the IFRS on voluntary basis.
Act on accounting from 2011
After several years of effort by academics and practitioners, the Ministry of Finance proposed an amendment of
Act on Accounting, which was approved by the Parliament in 2010. Starting from 2011, companies specified by the
Act are allowed to select the IFRS as the basis for preparation of individual financial statements, which are accepted
for statutory purposes. According to new §19a, articles 7 and 8, following three groups of entities may opt to use the
IFRS in their individual financial statements:
467

� parent companies preparing consolidated financial statements in compliance with the IFRS voluntarily
pursuant to §23a, article 2
� entities under control belonging to a consolidation group for which the consolidating company prepares
IFRS consolidated statements
� joint ventures belonging to a consolidation group for which the consolidating company prepares IFRS
consolidated statements
The amendment of the Act has changed the features of companies covered by Category II (see description
further in the text). From 2011, the Category II entities can opt whether to prepare two sets of financial statements
(both CAS and IFRS) or whether to prepare only one set of financial statements (only IFRS).
4.2 Issues regarding the implementation of Regulation (EC) 1606/2002
As far as financial reporting concerns, three groups of Czech companies can be recognised. To summarise,
provisions of the Czech Act on Accounting distinguished following groups of companies until 2010:
� Category I (big Czech companies that are publicly traded on stock exchanges in the EU markets – IFRS
reporting only):
These entities have both to account for their transactions and to prepare their financial statements (both
individual and consolidated) using the IFRS. These companies are not required to prepare their financial statements
according to the Czech Accounting Standards (further “CAS”) as financial statements prepared in accordance with
the IFRS are also accepted for the statutory purposes.
� Category II (Small and medium-sized enterprises – both CAS and IFRS reporting):
This category covers a diverse group of companies. The common feature of Category II is that companies in
question are not a direct issuer of publicly traded securities. Nevertheless, their owners are such issuers. Therefore,
the companies belonging to this category shall prepare their individual financial statements in accordance with the
CAS for statutory purposes. In addition, they shall provide their parent companies with financial statements and
other information needed for consolidation in compliance with the IFRS. Act on accounting did not permit any
voluntary application of the IFRS instead of the CAS by this kind of entities. In this group companies preparing
consolidated financial statements pursuant to §23a, article 2 voluntarily may be also subsided.
� Category III (Small and medium-sized enterprises – only CAS reporting):
Category III covers family owned companies and other companies that are neither direct, nor indirect issuer of
publicly traded securities. They shall account for and report in accordance with the CAS (again without possibility
to apply the IFRS voluntary).
Provisions of Act on accounting required mandatory application of the IFRS not only in consolidated financial
statements of listed companies (pursuant to Regulation 1606/2002), but also mandatory application of the IFRS in
their individual statements. Individual financial statements are accepted for statutory purposes levied by the
Commercial Code and are submitted to Business Register. That means, that listed companies (Category I
companies) are not engaged in the process of financial statements conversion, as both individual and consolidated
financial statements are prepared according to the IFRS and no additional set of financial statements prepared in
accordance with the Czech Accounting Standards (CAS) is necessary.
Conversion of financial statements is an important issue for companies covered by Category II. The majority of
Czech companies are not directly listed on stock exchanges (there are only 60 issuers on Prague Stock Exchange).
According to Act on accounting, all non-listed companies had to keep their accounts and prepare their individual
financial statements in accordance with the CAS. However, about 40% of Czech companies are under control of
foreign owners. A lot of them are domiciled in Germany, Netherlands, Austria and other EU member states and they
are often listed on stock exchanges. For the consolidation purposes, Czech companies must provide their parent
companies with the IFRS financial statements.
As a voluntary application of the IFRS in individual financial statements had not been allowed till the end of
2010, affected companies faced a problem of financial statements conversion. Statutory accounts were held in
compliance with the CAS; and consequently statutory statements had to be converted into IFRS statements.
The conversion is not a trivial issue as a huge number of differences between CAS and IFRS exist. PwC (2009)
published a comprehensive analysis, which comprises differences between the IFRS and CAS on 80 pages.
468

Therefore, the decision, which method of conversion to use, needs a deeper analysis by an entity’s management. All
relevant advantages, disadvantages, possible benefits and cost restrains should be taken into account.
The method of conversion on financial statements level is not appropriate for the vast of Czech companies, as
differences between CAS and IFRS are not insignificant. Remaining two approaches are therefore favoured by
Czech companies. The second method of financial statements conversions (on trial balance level with usage of
spreadsheet applications like Excel) represents “golden middle way”, as benefits and costs are balanced for the
majority of Czech companies reporting both under CAS and IFRS. Low level of conclusiveness together with the
dependence usually on the only one expert responsible for the conversion is offset by significant ICT cost savings,
because no specialised software is necessary under this approach. The last method (dual accounting system) is
applied by those Czech companies belonging to big supranational consolidation groups which use the same
accounting and reporting system for all group companies. Higher ICT and labour costs are counter-balanced by two
dataset of information. Moreover, the conclusiveness and consistency of accounting records is a valuable asset of
this method.
The impossibility to apply the IFRS voluntary produces high social costs regardless, which method of
conversion is chosen by entities. Scarce economic resources have to be employed in non-productive use. Academics
and accounting profession tried therefore persistently to persuade the Ministry of Finance to amend the Act on
accounting by enabling voluntary application of the IFRS in individual financial statements by Category II
companies. The Ministry of Finance finally recognised this proposal to be justified. Starting from 2011, Czech
companies, which are consolidated entities in the context of Regulation 1606/2002, can chose to prepare their
individual financial statement in accordance with the IFRS. In case of optional application of the IFRS, financial
statements conversion is not an issue anymore. However, companies may decide to maintain current status quo and
to prepare their individual statements further under CAS principles.
New provisions of the Act were enacted in December 2010. How many entities will utilise amendments of the
Act is still uncertain. As the implementation of new accounting software is a quite complicated project, it is highly
improbable that any companies have switched to the IFRS already from January 2011. First empirical evidence will
not be available sooner than next year. Author of the paper carried out a quick empirical pre-research to evaluate the
readiness of entities, external accounting firms, firms developing accounting software and auditors for the IFRS
transition. A questionnaire containing two sets of questions was answered by ten accounting or auditing firms. The
first group of questions relates to the possible extent of differences between the IFRS and CAS among various types
of companies.
Table 1: Differences between IFRS and CAS financial statements
Insignificant Rather insignificant Rather significant Significant
Manufacturers 0% 0% 60% 40%
Merchants 80% 20% 0% 0%
Services 0% 50% 40% 10%
Source: Authorial survey
The second group of questions focuses on evaluation of readiness for the IFRS adoption by various subjects.
Table 2: Readiness for voluntary application of IFRS by non-listed companies
Certainly not Rather not Rather yes Certainly yes
Entities 0% 10% 40% 50%
Accounting firms 10% 30% 50% 10%
Auditors 0% 50% 50% 0%
Software firms 10% 80% 10% 0%
Source: Authorial survey
Despite the fact, that the research is not fully representational because of restricted size of the sample, certain
tendencies can be derived from the respondents’ answers. The reactions to the first set of questions affirm a general
conclusion about high number of differences between the IFRS and CAS. According to professional accountants and
auditors, this issue is relevant esp. for manufacturing companies (which create a significant part of Czech gross
domestic product). In addition, differences are an issue for entities providing services, which usually struggle with
469

evenue recognition as there is no guidance on this issue (neither general, nor for the construction contracts) in CAS
and revenue recognition is mainly influenced by legal and tax matters.
As far as readiness for voluntary application of the IFRS concerns, companies are on the top of the list.
Companies, currently preparing both sets of financial statements, should have relatively smaller difficulties when
shifting from the CAS statutory accounts to the IFRS. On the other side, it is believed that Czech software firms are
not ready for the transition, which could bring problems for companies considering the voluntary IFRS adoption.
ICT solutions for keeping accounts according to the IFRS are offered by foreign software developers (such SAP,
etc.). However, costs of this solution can be prohibitive for affected companies, esp. for medium sized enterprises.
The replacement of local accounting software by ERP systems results in increased annual IT expenses up to ten
times in Czech companies according to the author’s experience.
The possible advantages from a voluntary shift to the IFRS in statutory accounting may be evaluated with
reference to experience of the Czech listed companies, which have to apply the IFRS obligatory both in consolidated
and individual financial statements. There are about 60 issuers on Prague Stock Exchange, from which were
excluded some issuers such as public sector institutions, financial institutions and issuers with domicile located
abroad. Representatives of twenty three companies remaining in the sample were asked for evaluating benefits and
costs of IFRS implementation in their companies. The benefits mentioned by already-adopters may serve as a useful
source of reference for those companies, who are considering utilising a new provision of Act on accounting, which
allows voluntary IFRS adoption in individual financial statements by designated entities. The answers of
respondents on benefits are summarized in Table 3.
Table 3: Benefits from IFRS implementation on companies’ level
Certainly yes Rather yes Rather not Certainly not
Easier access to financing by share capital 73% 27% 0% 0%
Easier access to financing by bonds 9% 55% 27% 9%
Easier access to financing by bank credits 0% 45% 36% 18%
Easier reporting within consolidation group 64% 36% 0% 0%
Increased relevance of data for management 0% 27% 55% 18%
Increased credibility for our trade partners 0% 64% 27% 9%
Increased reputation for general public 0% 45% 55% 0%
Source: Authorial survey
Once again, presented table cannot be supposed to represent generally valid inferences due to a restricted
sample. However, it can be said that results are not surprising for the most of answers and they correspond to
general expectations. As the IFRS are compulsory for listed companies, the connection between the IFRS
implementation and the possibility to raise share capital financing is very close. A successful issuance of bonds is
also supported by the IFRS adoption, as capital markets are the only source of available debt funds even in such an
undeveloped capital market like in the Czech Republic. Banks possess more tools for evaluating financial health of
applicants for bank credits; therefore IFRS adopters do not perceive the shift to IFRS to be really relevant for this
purpose. Based on personal talks with some representatives, IFRS statements play no role in banks’ assessment
whether to grant a company with requested credit or not. Nevertheless, IFRS statements are said to reduce costs of
preparing applications forms and other documentation required by bank significantly.
The IFRS implementation is not supposed to enhance data relevance for management purposes. It is a quite
interesting outcome, when taking into account relatively low relevance and usefulness of accounting principles set
up by Czech national accounting legislation. There are two possible explanations to this phenomenon. Firstly,
managers are accustomed to the former CAS, which they used for years, and their do not understand and/or do not
believe in data based on the IFRS. Secondly, due to weakness of the CAS companies were forced to developed highquality
management accounting systems removing those weak points and supplying relevant data for decisionmaking.
Therefore, a shift to IFRS has not produced any significant increase in data relevance from managers’ point
of view. This finding needs further attention and scrutiny.
Representatives are indecisive regarding the influence of IFRS statements and annual reports on company
reputation in the eyes of general public. On the other side, IFRS statements are supposed to be welcomed by trade
partners. They are usually applied for credit scoring of customers and other risk management procedures. Finally,
470

the IFRS implementation brought substantial advantages in the process of preparing consolidated financial
statements. Representatives of all listed companies assert that the IFRS adoption has eased reporting within
consolidation group. This may be the crucial supporting element for those Czech non-listed companies when
evaluating whether to adopt the IFRS voluntarily in individual financial statements as it is allowed by Act on
accounting starting from 2011.
5 Conclusions
If the IFRS are not allowed to be applied voluntarily in individual financial statements of listed companies or in
individual statements of non-listed companies belonging to a group consolidated in compliance with the IFRS, the
necessity of financial statements conversion occurs. The second case is a common practice in the Czech Republic
because Act on accounting did not enable non-listed entities to use the IFRS on an optional basis until 2010. The
financial statements conversion is therefore an issue for almost 40% of Czech companies. To keep accounts
according to two sets of relatively difference accounting standards in single accounting software is very costly
matter. Therefore, most entities have chosen to make the conversion through Excel and similar data spreadsheets on
trial balance level.
Such an approach reduces costs significantly; however accuracy and transparency of the conversion process
heavily rest on the abilities of a single expert who are in charge of the conversion. The conclusiveness of such
conversions is not high in general. The author of paper has experience from a significant number of companies, of
which financial statements converted from CAS to IFRS using trial balance method of conversion are not in the
compliance with all provisions of IFRS. Spreadsheets are not constructed to cope with all nuances of double-entry
accounting. With rising number of differences between the CAS and IFRS, omissions and computing mistakes are
common feature of this method of financial statement conversion. Mistakes and omissions remain even after being
checked by auditors. As a consequence, the quality of consolidated financial statements presented by parent
companies may be severely impaired, because a lot of Czech companies create a significant part of consolidated
figures. In my opinion, this is an issue of a fundamental importance not only for the Czech Republic, but also
worldwide. Nevertheless, this issue is not really addressed by either current practice or research.
In the context of previous doubts, the amendment of Act on accounting by the end of 2010, which enables
selected companies to apply the IFRS voluntarily, shall be welcomed. By allowing companies to apply the IFRS in
their individual statements, the Czech Republic follows pattern of financial reporting used e.g. in the Netherlands,
Denmark, Ireland, etc. The amendment should lead to presenting accounting information, which is more useful for
public. The second favourable effect would be the reducing in costs connected with recording transactions and
preparing financial statements under two different set of accounting standards. Finally, the risk of errors contained in
consolidated financial statements would substantially decrease, because the voluntary IFRS adoption in individual
financial statement means that all records are kept within accounting software and not outside in spreadsheets.
Claims to prepare individual statements compulsory according to national accounting legislation levy high costs
on companies and have other negative economic consequences. National regulator of accounting should enable
affected companies to prepare their individual statements on alternative basis, mostly in compliance with the IFRS.
The development in the Czech Republic can serve as a source of inspiration for all countries solving the relationship
of financial reporting standards applied for consolidated and individual financial statements. However, more robust
empirical evidence will be available earliest in 2013, as first companies will probably switch to the IFRS from year
2012 (as the Act was amended in December 2010).
There are some restrictions impairing inferences of this study. Firstly, not all companies are allowed to apply
the IFRS voluntarily. Only entities, which are subject of a full consolidation under the IFRS, are allowed to take
advantage of the option offered by the amendment of Act on accounting. As a consequence, companies classified as
investments in associate and consolidated by equity method are excluded from this option.
The crucial problem is, although, that Czech accounting is closely linked with the taxation system. For the
computation of current income tax, only net income according to the CAS is relevant. Therefore, all companies
regardless whether they prepare individual financial statements according to the IFRS compulsory or voluntarily
have to keep evidence of taxable income based on the CAS. Without releasing financial accounting from the income
471

tax law, the conversion of financial statements will remain a common practice for Czech companies. Author of the
paper is a member of a researcher team that is currently working on a study evaluating various approaches how tax
system can be separated from the financial reporting. The future conclusions of this study can be valid for all
countries, in which state regulator carries out the regulation of financial reporting mainly for the tax purposes.
6 Summary
The paper discusses various approaches to the conversion of financial statements and their advantages and
disadvantages are evaluated. This issue is relevant especially for countries, which have adopted the IFRS and
concurrently companies are required to prepare complementary set of financial statements according to national
GAAP. The empirical evidence in case of Czech non-listed companies is provided in the paper.
7 Acknowledgements
The paper is processed as an output of a research project „ Adoption of the IFRS and Its Influence on the Mobility of
Labour and Capital" supported by the Internal Science Foundation of the University of Economics, Prague
(registration number F1/4/2011).
8 References
Aubert, F. – Grudnitski, G. (2009). The Importance and Impact of Mandatory Adoption of International Financial
Reporting Standards in Europe. Tampere, 32nd Annual Congress of the European Accounting Association, 12.
5. 2009 – 15. 5. 2009.
Bagaeva, A. (2009). The IFRS and Accounting Quality in the Transitional Economy: A Case of Russia. Tampere,
32nd Annual Congress of the European Accounting Association, 12. 5. 2009 – 15. 5. 2009.
Barth, M. E. – Landsman, W. R. – Lang, M. H. (2008). International Accounting Standards and Accounting Quality.
Journal of Accounting Research, 2008, vol. 46, is. 3, p. 467–498.
Christensen, H. B. – Lee, E. – Walker, M. (2009). Do IFRS Reconciliations Convey Information? The Effect of
Debt Contracting. Journal of Accounting Research, vol. 47, is. 5, p. 1167–1199.
Cordazzo, M. (2008). The Impact Of IAS/IFRS On Accounting Practices: Evidences From Italian Listed Companies.
Rotterdam, 31st Annual Congress of the European Accounting Association, 22. 4. 2009 – 25. 4. 2009.
Ferrer, C. – Callao, S. – Jarne, J. I. – Lainez, J. A. (2009). IFRS Adoption in Spain and the United Kingdom: Effects
on Accounting Numbers and Relevance. Tampere, 32nd Annual Congress of the European Accounting
Association, 12. 5. 2009 – 15. 5. 2009.
Jaruga, A. – Fijalkowska, J. – Frendzel, M. – Jaruga-Baranowska, M. (2007). The Impact of IAS/IFRS on the
Accounting Regulations and Practical Implementation in Poland. Lisbon, 30th Annual Congress of the
European Accounting Association, 24. 4. 2009 – 27. 4. 2009.
Mejzlík, L. (2006). Accounting Information Systems [Text in Czech: Účetní informační systémy]. Prague:
Oeconomica Publishing house, 2006.
Morais, A. – Curto, J. D. (2007). IASB Standards Adoption: Value Relevance and the Influence of Country-Specific
Factors. Lisbon, 30th Annual Congress of the European Accounting Association, 24. 4. 2009 – 27. 4. 2009.
Mustata, R. – Matis, D. – Dragos, C. (2009). The Challenges of Accounting Harmonisation: Empirical Evidence of
the Romanian Experience. Tampere, 32nd Annual Congress of the European Accounting Association, 12. 5.
2009 – 15. 5. 2009.
PricewaterhouseCoopers Audit (2009). IFRS and Czech Accounting Standards – Similarities and Differences [Text
in Czech: IFRS a české účetní předpisy – podrobnosti a rozdíly]. Prague: PricewaterhouseCoopers Audit, 2009.
472

NEGOTIATING SUCCESSION STRATEGY IN FAMILY RUN SME’S
Christopher Milner
Dept. Strategy & Business Systems, University of Portsmouth
Richmond Building, Portland Street, Portsmouth, PO1 3DE, UK
Abstract. The entrepreneurial founders of family run organisations often choose intergenerational succession as their preferred exit
strategy. Statistics suggest that only 30% manage the succession process successfully. This paper tracks and analyses the experience
of six organizations, highlighting those factors which appear to support a successful transition. The comparison between current
thinking and founder perception brings the clear identification of gaps that hold the potential to serve as reasoning for failure, this
paper focusing upon the suggested and perceived attributes and methods of preparation of the successor. Thorough analysis allows for
the building of a contextual model of Intergenerational succession and a set of recommendations to serve family organisations to bring
a successful and sustainable transfer of its leadership.
Many founders choose intergeneration succession as their exit strategy, wanting their children to take over at the
helm of the organisation. Literature surrounding the succession process in family run companies is thin and
fragmented, and the limited attention it has been given seems to have a negative slant, understandable considering
many with a history of success fail to survive past the tenure of the founding generation (Birley, 1986; Kets de
Vries, 2001; Morris 1997 et al). This paper focuses on the gap between what previous research suggests as best
practice against that of founder perception, with particular interest towards the preparation of the successor.
Stepping down can prove much harder for founding owners, than their professional orientated ‘employed’
counterparts, as they can perceive their businesses as an extension of themselves. The choice of exit strategies
available can depend on a number of variables; these including the entrepreneurs’ values and wants for the long term
future of the company, the operational structure and financial performance, while the state of the economy and
market the organisation operates in is central to the options available.
According to Sharma et al. (2001) the single most cited obstacle to effective succession is the incumbent’s
inability to let go. With careful implementation and management of the succession plan, handover brings in the new
chapter, and should involve a process of mutual role adjustment (Handler 1990, 1992), where the incumbent’s role
gradually diminishes whilst the successor’s gradually increases. The author suggests that the transition should be
looked upon from evolutionary perspective through a balance of internal and external method, however there are
cases where a revolutionary is necessary due to the lack willingness form the founder’s side to let go of the
organization they have themselves built. There are a number of variables that collide at this critical point in the
process and so the management and planning for such is vital. The founder’s willingness to let go and the
successor’s desire and ability to take over should have been increased over the succession process.
1. Literature Review
The attributes of the incumbent, the successor, and the relationship between them serve as the foundation and
backbone of the potential success of the process, Handler (1992) reports however that very little is known about the
role played by either. Do they have trust and mutual respect for an effective transfer of tacit knowledge? Handler
(1990) suggests that mutual respect and understanding between generations is crucial; do they have a realistic
appreciation for the skills and abilities of what one has done and of what the other can do? Do they share similar
views to the current and future needs and strategic positioning and direction of the company, and is it important to
do so? Is the incumbent willing to truly hand over the reins, and does the successor have the competence, ability,
education and knowledge that may be necessary, with the passion and motivation to take them?
Founders of organisations have an entrepreneurial spirit, a need to be one’s own boss and build success though
their own direction. Kirzner (1985) describes them as individuals who demonstrate the alertness, character and
temperament needed to exploit a discovered opportunity, characterised by innovation, risk-taking, creativity and
growth, while Berry (1998) defines a firm run by an entrepreneurial founder ‘an independent enterprise, managed in
a personalised way’.
473

The characteristics of entrepreneurs can be divided into the instinctive, intuitive and impulsive ‘charismatic’ and
the more conservative and realistic ‘pragmatic’, these distinguished through the variables brought together from
research done by McCarthy (2001). The author suggests that a charismatic character with an obsessive nature and
less business acumen, is more likely to view the intergenerational succession process in a simplified state, and may
be more likely to be unwilling to let go of the organisation during the final stages of any future succession. In fact
the incumbent’s motivation and willingness to step down can prove to be a major obstacle, Dyer (1986), Handler
(1990), Lansberg (1999) & McGiven (1978) all suggest the importance of the predecessors ability to overcome
anxiety about succession, move beyond the denial stage, and being willing to confront succession and let go is
crucial to the success of the process. They must face normal fears such as losing control, power, and even part of his
or her identity and stature in the community (Potts et al. 2001)
As the individual being groomed to take over the management and leadership of an organisation, the successor
could be described as the key to the long term operational and financial future of the company. They are a central
figure to discovering what is necessary in order to successfully negate through an intergenerational succession
process.
The depth of literature on intergeneration successors is relatively thin, but the ability, in terms of their potential
leadership and business management is often questioned. It could be stated that the company, depending on the
position within its lifecycle, would benefit running as a professionally structured organisation with the appointment
of a more highly skilled, educated and experienced manger, rather than a family orientated one, but research shows
that family businesses are idiosyncratic to a high degree (Nooteboom, 1993; Pollak 1985) and are more likely to
appoint their potentially less competent offspring. The author would though suggest that there is high value in
mentoring and on the job training, which could result in contextual competence of the business, so whilst potentially
not as formally educated as a manager sourced externally, may not be any less competent, but hold competence of a
different type. Although theorists (Dyer et al,. 1986) have stated that the main cause of intergeneration succession
failure is the founding incumbent not letting go of the organisation, the successor also holds a great responsibility;
do they hold the ability, suitability and motivation to run the business, have they been properly prepared, and does
the founder have realistic expectations of them? Lee, Lim & Lim (2003) have written that when personal and
emotional factors determine who the next leader will be, poorly qualified successors can prove detrimental to
success and short and long term stability of the organisation.
Morris et al. (1997) and Chrisman et al. (1998) note the importance of experience, suggesting a successor who
has worked within the family firm, as well as other organisations will be better prepared. Next generation family
members are often working full time in the business, building organisation knowledge, but according to
Longenecker & Schoen (1978) they tend to characterise their role as limited but functional.
Whilst the theoretical and founders perception of how vital it is to train, educate and build the tools necessary to
run an organisation may prove to be different, one potential central problem that has been researched by Howorth
and Ali (2001), is that a number of patriarchal family owners fail to prepare the successor to succeed, and so
possibly putting their children in a very difficult and stressful position, and putting the future stability of the
organisation and prospective success of the succession process at risk. Whether this failure is due to lack of
appreciation of the importance of a well-structured succession plan to bring the successor ability and confidence, or
the founder having an unrealistic view of their children’s attributes and abilities, it is understandable to imagine why
some successors may find it difficult running the company, or to gain the respect and credibility of the staff.
Founders can have a large bearing on the culture of an organisation; the closer the founder’s and successor’s
thoughts on the objectives, strategy, values and ethics, the smoother the transition could be. Miller, Steier and Le
Breton-Miller’s (2003) research suggests three types of succession, depending on the actions of the successor once
they take over the management and leadership of the organisation; conservative, wavering and rebellious.
While some literature views succession as a short term handover, others see it as a multi stage long term process
(Handler, 1990), but a process Lansberg (1999) argues few family firms are capable of successfully completing.
Dyck et al (2002) analogy of the 4x100 relay race brings together the elements of sequence , timing , baton passing
technique and communication , while Miller, le Breton-Miller & Steiner’s (2004) integrative model of succession
succeeds in bringing a fuller picture to light. This model suggesting that respect, understanding and corporation is
vital between the founder and successor, supporting the thinking that the founder must be willing to let go and the
474

successor, if they have the motivation and commitment, alongside the ability and competence the choice whether
they want to take over.
Alongside identification of the perceived necessary attributes, it is important to pay attention to the ways in
which these attributes are developed and facilitated. While theory suggests that education is positively correlates to
a smooth transition and post succession performance (Morris et al. 1997) and that sets a foundation for areas
including business strategy and marketing; and that internal training (Neubauer & Lank, 1998), knowledge transfer,
and a growing involvement within the organisation (Cabrera-Suarez et al. 2001) and outside work experience
(Barach & Gantisky, 1995) are all important.
2. Data and Methodology
The data collection process followed a unique but appropriate two-stage format. Due to existing business
relationships, it was possible to conduct initial, informal conversations with participants over a period of months.
These were followed by a final, formal semi-structured and open-ended interview. The interviews are used as the
main source of primary data collection; generating rich and detailed accounts of the individual’s experience,
perceptions, plans and thinking.
The interviews have been transcribed, case study summarised, the results analysed, a summary matrix
formulated, enabling a clear comparison between current theory and founder perceptions. Whilst these will are not
available within this paper, the results and gap analysis are quite enlightening. Due to the depth of the study, many
conclusions could be put forth, but for the purpose of this paper, the most relevant trends found are offered.
With a small sample of subjects more appropriate than a large, those included within the study are all founders
(owning over 100% of shares) of family led SME’s located in the southern counties of England and operating within
the retail environment of which intergenerational succession is a current or future possibility.
Yin & Heald (1975, p.372) suggests “the case study method is mainly concerned with the analysis of qualitative
evidence in a reliable manner” and it is this approach that this research is based, the six case participants listed
within Figure 1 below.
3. Empirical Findings and Discussion
Figure 1: Case Participants
Conclusions drawn from the case studies suggest that each of the founders choosing succession as their exit strategy
expects the transfer to be successful, and on a general basis, whilst perceiving long term semi-formal planning,
475

implementation and management to be necessary, they have some very interesting views on what the process should
consist of, especially relating to the successors training needs.
Out of the six founders involved in the study, five of them have chosen intergeneration succession as their likely
exit strategy; reasons ranging from that of the founder of Company A who states that it is not something he has
chosen, but “one of those things that happens”, to the thoughts of the founder of Company F who being able to retire
while still taking an income and enabling the children to have a career (Company F). The founder of company D
though has chosen to sell his organisation, who questionings “why anyone would want to bring their son or daughter
into their business”.
Whilst some founders may be very content and look forward to the comfort and warmth of their next chapter,
literature shows that some feel great anxiety in the face of retirement (Dyer, 1986; Handler, 1990; Lansberg 1999).
All but one of the founders in the study stated that they are looking forward to retirement. Other than the founder of
Company D who states that he will miss “nothing at all”, the rest expect to miss aspects of running the organisation;
the founders of company E and F state that the day to day contact with customers, suppliers and staff as some of the
main elements. This is an interesting link in variables considering company D is the only one stating that “the
business being sold” as the likely exit strategy; those choosing intergenerational succession all intend to take up
figure head roles and so expect to remain involved within their organisation.
Whilst the Founder of company A has the strongest desire to continue, stating “even if I take a semi-retirement
backseat role, I will remain involved” and in reply to being asked for what reason he will leave his post, his reply
was “I will die” and owner of company F commenting that “if there was a problem and a figure head was needed, I
could step in and sort it out”, this potential evidence of founders themselves being potential obstacles to a
successful succession, by not letting go of the organisations reins.
Motivation and enthusiasm top the founders list of necessary attributes, the author agrees that this is a crucial
factor, but with a balance of competence and business acumen. Leadership and credibility come in a close second
followed by organisational knowledge and management and strategic ability. Morris et al (1997) states that
education is linked to positive succession, the founders don’t portray it in high esteem, company F’s stating that
“common sense is required rather than education”.
It is the successors that the founders see as the ones who must hold such attributes in order to successfully run
the organisation, and while they state motivation as the key, a question hangs whether the successor’s business
acumen and management ability is looked at from a parental perspective, or truly measured though a professions
eye. While the majority seemed very confident in their siblings current and prospective future attribute set and
abilities, the founder of company A states that his successor, other than motivation “needs to work on all of them,
she basically does not have most of them”, of which she openly concurs saying “sometimes I doubt my own
abilities…..it will take a lot of work”. Due to the limited pool and lack of potential options it is difficult to quantify
to what extent founders initially question their successors ability or suitability, especially those of whom the current
literature states make the decision of intergenerational succession for emotional and personal reasons. How well the
successor is prepared and primed for such an important role in such circumstances is then down to the succession
plan itself, and the development that takes place during the process.
Cabrera-Suarez et al (2001) suggests that the closer the relationship between the founder and successor and their
thoughts on the organisations objectives, performance, values and ethics, the smoother the succession may be.
Drawing on the case studies the relationship between the founder and successor appears to be of particular
importance. The founder of company C sees “personality crashes” as a potential problem, the founder of company
A, sees his relationship with his successor as a strong father/daughter bond, but while the finance director and
daughter agree; she recognises the problem of not being able to separate personal and professional life stating that
they “cross over most of the time”.
The founder of company D states “if you let one of your children take over your business, you will never accept
that they can do it better than you, and you will never let them”, alongside the ability and attribute set the successor
has or has built over the succession process, they may also need a strong character in the face of too much founder
involvement.
476

It is natural that family organisations have a much smaller pool of talent of which to draw, especially when it
comes to intergenerational succession, and so the development of the successor is of great consequence during the
process. A potentially successful successor should be motivated, have leadership ability alongside credibility,
intrinsic organisational knowledge, good management ability, experience, and an education (Howorth and Ali, 2001;
Morris et al. 1997; Christman et al. 1988; Handler, 1990), and this is where the gaps in theory and founder
perception become more visible. As can be seen in figure 2; while motivation, leadership ability and credibility hold
top positioning within the founders attribute list with organisational knowledge and management ability trailing,
leaving education and real management experience behind, and so the risk of handing the business over to a
unprepared or under qualified successor is increased.
The author concurs with Morris et al. (1997) who suggest that education positively correlates to a smooth
transition and post succession performance and sets a foundation for areas including business strategy and
marketing. Alongside this Internal training (Neubauer & Lank, 1998), knowledge transfer, and a growing
involvement within the organisation (Cabrera-Suarez et al. 2001) alongside outside work experience (Barach &
Gantisky, 1995) are all important, the founders again, have a different perspective as can be seen in Figure 3.
The founder of Company D stand apart from the other founding members, by questioning the reasoning behind
intergenerational succession and states outside work experience is essential as a learning platform, suggesting that
“working for a few good people and a few bad people” brings with it a greater understanding and knowledge base
and that young successors “need to stand on their own two feet”. The others founders expect that an internal
training scheme and gaining knowledge from the founders themselves, of which the potential successors in all but
one of the organization have already begun, will enable the successor to gain everything they need. The author
suggests that this is likely to be a natural consequence of the founders having children, and the children’s lifelong
involvement within the company. The author also puts forward that a training program originating entirely from
within the organisation is not enough to prepare the successor, especially when they have so far taken a backseat
role in the organisation, with limited functionality, this supporting by the work of Longenecker & Schoen (1978).
Figure 2: Successor Attribute Gap
477

Figure 3: Training Method Gap
To gain the credibility and respect of the staff that are considered by the author to be factors that may be
necessary in order to lead effectively, a broader learning platform could further enable the successor’s development
and potentially build a greater and more effective skill set. None of the founders taking part in the research though
saw formal education as a valuable tool; it could be assumed that they would prefer their children to learn through
the school of hard knocks that they themselves may have worked through, or as Ibrahim et al (2004) suggests, view
education as more of a cost burden than an investment in future growth.
The case studies suggest that founders much prefer internal training mechanisms, whilst some research,
(Howorth and Ali, 2001; Morris et al. 1997; Christman et al. 1988; Handler, 1990) takes into account the value of
outside contributors including work experience and education.
With careful implementation and management of the succession plan, handover brings in the new chapter for
the organisation, whilst some literature suggests that it should involve a process of mutual role adjustment (Handler
1990, 1992), the incumbents role gradually diminishes whilst the successors gradually increases. The analysis of the
case studies shows that the founders perceive the management of the handover to be crucial, and a gradual
responsibility changeover to facilitate its completion. While the founder of Company A does not ever see himself
fully stepping down, the rest intend to hold figure head roles; ranging from the founder of Company B who states he
will “always be available” to the thoughts of the founder of Company F who states “I think I will remain with the
business, less on the responsibility and decision making role, and more as sticking my head in the door, making sure
the suppliers and customers know I’m around, but only on a superficial level”. This founder goes on to say that
“you almost have to let go through gritted teeth, but knowing its overall the best thing to do”; handover clearly
serves as potentially critical period, some potential scenarios of are explored later in the paper.
While most the founders seem comfortable with their successors bringing in new impetus and young blood into
the organisation and have intentions of being available, acting as figure heads to their offspring, there is a possibility
they will be surprised at the effect of any of the three previously discussed succession types; conservative, wavering
and radical (Miller, Steier and Le Breton-Miller 2003). If they see nothing happening, poor strategy or radical
change, the research results suggest that they may find themselves wanting to take back some control; such a
situation could become very complicated and frustrating. The likely successor in company A, states “I think initially
I will sit back and see how things go, and then if something comes to light that might need changing, I will
implement those changes”, but while the financial director questions her true motivation and ability to do so. The
variables involves in a succession process are vast; the founder, successor, organisation, process plan and handover;
it is becoming increasingly evident why it can be such a dangerous time for any family company.
Clear and concise communication between the founders themselves and their successors is viewed as crucial by
the literature and founders; a healthy respect enabling the transfer of tacit knowledge and platform to train.
478

Communication of the succession plan itself can also prove vital throughout the company board to ensure
implementation; interestingly the founders of Company’s B and F state that keeping the staff within the company in
the loop, communicating to them the succession plan so they are aware of the future company structure should help
to ensure their involvement and motivation to help things transfer smoothly.
The founders agree that the succession process needs to be formally planed and managed, this supported by
researchers including Handler, (1990). Like any business strategy, true implementation is key to ensure it is achieves
its objectives; it may need amending from time to time, changing when unforeseen circumstances raise their head,
but the management of such is essential, the Finance Director of company A states in answer to how important it
management of the process is; “Most definitely, especially in a family owned business, it’s even more important
than in large corporate organisations”.
The results have brought forward a clear message of what is perceived to be necessary to help bring a successful
transfer and of the needed attributes and preparation of the potential successor. A strong respectful relationship
between a founder that is willing to let go and successor with the motivation and ability to take over, a shared long
term, multi stage and well managed formal succession plan in which to prepare the organisation and develop the
successors abilities and attributes through the use of internal and external facilitators, with a handover through
mutual adjustment.
The founders proved to be very interesting individuals, passionate about their organisations and very confident
about the future, whose thoughts on the many elements of the succession process were aligned to current thinking.
The clear gaps though identified in Figures 4 and 5 signify how family organisation do prefer to do things from
within; potentially losing the vital value of outside work experience and education, two sources of development that
the literature suggest positively correlates to a smooth transition and post succession performance (Morris et al.
1997; & Barach & Gantisky, 1995).
The literature talks of how crucial the choice of successor is, of how family businesses are idiosyncratic to a
high degree (Nooteboom, 1993), how personal and emotional factors can determine who the next leader will be
(Lee, Lim & Lim, 2003) and are more likely to appoint their potentially less competent offspring (Pollak 1985). It
may be true that some founders make the decision of intergeneration succession due personal and emotional needs
instead of strategic goals for the future performance of the company, but some do it as they believe it is the right
choice, or because it may be the only one available; family organisations usually have a smaller pool of talent on
which to draw, (Lansberg, 1999; Miller, Steier, & Le Breton-Miller, 2003) and so successor options can prove to be
slim, potentially relying on a unsuitable candidate.
4. Conclusion
A contextual5 stage model has been built from the results of the analysis; for the use in family run SME’s who have
come to a stage where the founder is reviewing potential exit strategies, and intergenerational succession is chosen.
Stage 1: Exit Strategy Choice
Stage 2: Successor Choice – Review of their (a) motivation (b) ability & (c) attributes: Successor Level vs.
Required Level
Stage 3: Review of relationship between founder and successor – review of their perception on shared vision &
needs of the succession process
Stage 4: Successor Development – Design & implementation to fill ability and attribute gaps through Internal (early
exposure, knowledge transfer – founder led & internal training program) & External methods (external work
experience & formal education)
Stage 5: Role Adjustment & Final Handover – Scenario planning and acceptance
Firstly, the internally based mechanisms; early exposure to the business can help build an important foundation
of organisational knowledge and understanding (Cabrera-Suarez et al., 2001), in Company A the successor has had
such, but in a role has proved to be sporadic and limited and due to this does not hold much credibility or respect
from the staff. As the leader of the organisation, the knowledge transfer between incumbent and successor is
479

crucial; company strategy, marketing, financial results and expectation, the list is endless, but the potential to gain a
rich first hand grasp of the needs, mechanisms, culture and people within the organisation may prove to be
invaluable. This is a stage where the relationship between the two is vital and mutual adjustment process (Handler
1992) can begin; a point in time where the founder gives, and the successor should respectfully take. While this
communication stream may never truly diminish, it is as important that the founder from this point should be coming
to terms with the idea of letting go of the organisation, as well as the successor getting ready to take it; the more
effective this is, the more probable a smooth final handover period.
Each of the participating organisations saw the value in a well-constructed internal training program, this is
supported by the finding of Neubauer & Lank (1998); an opportunity to build an deep understanding of each aspect
and department, to bring clarity to how the business works and how each part is linked, to build relationships,
respect and potential credibility through training, working and performing alongside staff at each level of the
company. This is a stage where clear objectives and timelines need to be set, a time where communication to the
staff involved must be clear and given in plenty of time; depending on the size and complicity of the organisation
and market it operates in. For example the within Company A the successor would potentially need to spend a
minimum of six weeks visiting the branches, two weeks each in the Operations, H.R. and Accounts Departments,
with a shorter introduction to Marketing and I.T. in order to gage a rich understanding into the intricacies of each.
This would bring an opportunity to work and spend valuable time with the customer/public facing staff, but also the
Finance Director and Operations Manager who alongside bring a greater depth of knowledge. This would paint a
realistic picture of the company’s economic position.
Externally based successor development has proved to be the largest chasm in perception between literature and
founder perception in the investigated cases. Outside work experience which Barach & Ganstisy (1995) state can
bring with it some very valuable lessons and even more so formal education which Morris et al. (1997) suggests
positively correlated with a smooth transition and post succession performance, are the two aspects that founders do
not believe to be very necessary. Whether this is down to limited personal exposure or their high confidence in
internal based training is open to question. However the conclusions from the case studies suggest that each family
organisation should consider the benefits of each, to enable successor development in order to increase their
knowledge base and understanding of the economy, and application of business concept and techniques.
Ibrahaim (2004) states this may be due to the perception of the firms that external training is more of a cost
burden than an investment in their future growth; however the author suggests that the founders perceive on the job
training and personal knowledge transfer as a more valuable platform of development than external alternatives, and
so it is not so much of a cost burden than a less effective option.
Stage 5 of the overall framework related to handover; the ideal scenario (Figure 4: Diagram 1) portrayed by the
literature (Handler 1990, 1992) and primary research (Founders B, C, D, E, F) is for the succession process to be
completed as planned; for the responsibility of the incumbent and successor to mutually adjust over a period of time.
If the organisation in question has a board of directors, such as Company, they can help facilitate the final crossover;
supporting the successor in their new role, whilst bringing confidence to the incumbent that all will be ok and
resisting their potential want to continue, helping transfer into their planned a figure head role. The type of
succession will at this point start to develop, the successor within Company A may prove to be what Miller, Steier
and Le Breton-Miller’s (2003) summarise as a conservative, stating that she will “sit back and see how things go”.
480

Figure 4: Handover Scenarios
Poor choice of successor can clearly be an obstacle to success, as can lack of confidence or insufficient
development on the part of the successor (McGiven, 1978; Schein, 1985). However, the founder’s willingness to
step down is tested at this critical stage (Sharma et al. 2001). If any of the above come into reality (Figure 4:
Diagram 2), a period of frustration can be encountered; a time where the important relationship (Cabrera-Suarez et
al. 2001) between the two can falter, and a period of organisational uncertainty for many staff can follow. Any such
obstacle needs to be managed; an organisation such as Company A whose successor expects “a difficult transition”
and the incumbent does not see himself fully retiring, the board of directors will have to play a pivotal management
role.
Handler (1990) has said that the realisation of poor or failing health can be a key stimulus for the incumbent to
start letting go of the business; Company A’s founder has been confirmed to have an incurable illness, so the future
is not as clear as one would like, in the unfortunate occurrence of his sudden death, the company should have a
contingency to help ensure its long term future (Figure 4: Diagram 3). In such a case the board of directors can take
a central role, the successors continued training and development is crucial to their ability to lead and manage the
organisation in the future.
Successful intergenerational succession process within a family owned and run business needs to be carefully
planned for, implemented and managed. The relationship between founder and chosen successor, the development
of the successor’s abilities, their motivation, and attributes including the exploitation and analysis within the review
of current thinking concludes that a leadership, organisational knowledge, management ability and credibility within
the organisation are deemed to be important. Training mechanisms including internal training programs, formal
education and outside experience all being reported to positively correlate with a smooth transition and postsuccession
performance, and with smooth final handover viewed as vital, it seems that the process is one that needs
to be thorough and complete; one that if any parts are missing or facilitated to a low level could potentially bring
failure.
481

5. References:
Barach, J.A. and Gantisky, J.B. (1995). Successful succession in family business. Family Business Review, 8(2),
131-155.
Berry, M. (1998) Strategic planning in small, high-tech companies, Long Range Planning, Vol. 31 pp. 455-66.
Birley, S. (1986). Succession in the family firm: the inheritors view. Journal of Small Business Management, 24(3):
36-43.
Cabrera-Suarez, K., De Saa-Parez, P., and Garcia-Almeida, D. (2001). The succession process from resource and
knowledge-based view of the family firm. Family Business Review, 14(1), 37-47.
Christman , J.J., Chua, J.H., Sharma, P. (1998). Important attributes of successors in family businesses: an
exploration study. Family Business Review 11 (1), 19-34.
Dyck, B., Mauws, M., Starke, F.A., and Mischke, G.A. (2002). Passing the baton: The importance of sequence,
timing, technique and communication in executive succession. Journal of Business Venturing 17: 143-162.
Dyer, W.G. Jr. (1986). Cultural change in family firms: Anticipating and managing business s and family
transitions. San Fransisco: Jossey-Bass.
Handler, W.C. (1990). Succession in family firms, a mutual role adjustment between entrepreneur and nextgeneration
family members. Entrepreneurship Theory and Practice, 15(1), 37-51.
Handler, W.C. (1992). The succession experience of the next generation. Family Business Review, 5 (3), 282 – 372.
Howorth, C., Ali, Z.A. (2001). Family business succession in Portugal. Family Business Review. 14 (3), 231-244.
Ibrahim, A.B., Soufani, K., Poutziouris, P. & Lam, J. (2004). Qualities of an effective successor: the role of
education and training. Education & Training Vol. 26. pp.474-480.
Kets de Vries, M. (2001). The Leadership Mystique, Prentice-Hall, London.
Kirzner, I. (1985). Discovery and the Capitalist Process. Chicago, IL: University of Chicago Press.
Lansberg, |. (1999). Succeeding Generations: Realizing the Dream of Families. Harvard Business School Press,
Boston.
Lee, K.S., Lim, G.H. & Lim, W.S. (2003). Family Business Succession: Appropriation Risk and Choice of
Successor. Academy of Management Review. Vol. 28, No4, 657-666.
Longenecker, J.G., & Schoen, J.E. (1978). Management Succession in the family business. Journal of Small
Business Management. July, 1-5.
McCarthy, B. (2001). Strategy is personality-driven, strategy is crisis driven: insights from entrepreneurial firms.
Management Decision 41/4, pp. 327-339
McGiven, C. (1978). The dynamics of management succession. Management Decision, 16(1), 32.
Miller, D., Steier, L., & Le Breton–Miller, I. (2003). Lost in Time: Intergenerational succession, change and failure
in family business. Journal of Business Venturing, 18(4), 513-531.
Miller, D., Steier, L., & Le Breton –Miller, I. (2004). Towards an Integrative Model of Effective FOB Succession.
Entrepreneurship Theory and Practice.
Morris, M.H., Williams, R.O., Allen, J.A., and Avila, R.A. (1997). Correlates of success in family business
transitions. Journal of Business Venturing, 12(5), 385-401.
Neubauer, F., & Lank, A.G. (1998). The family business. London: Macmillan.
Nooteboom, B. (1993). An analysis of specifically in transaction cost. Organisational Studies, 14:443-451.
Pollak, R.A. (1985). A transaction cost approach to families and households. Journal of Economic Literature, 23:
581-608.
Potts, T.L., Schoen, J.E., Engel Loeb,, M., & Hulme, F.S. (2001). Effective retirement for family business ownermanagers:
Perspectives of financial planners-Part 2. Journal of Financial Planning, 14(7), 86-96.
Sharma, P., Christman, J.J., Pablo, A.L., & Chua, J.H. (2001). Determinants of initial satisfaction with the
succession process in family firms: A conceptual model. Entrepreneurship Theory & Practice, 25(3), 17-35.
Yin, R.K. & Heald, K.A. (1975). Using the Case Survey Method to Analyze Policy Studies (Electronic Version)
Administrative Science Quarterly, Sept 1975, Vol 20. pp 371-381.
482

CREDIT RATING MODEL FOR SMALL AND MEDIUM ENTERPRISES WITH ORDERED LOGIT
REGRESSION: A CASE OF TURKEY
Res. Asst. Özlen ERKAL
Istanbul University - Avcılar Campus
Industrial Engineering Department
34320 Avcılar-Istanbul / TURKEY
ozlenerkal@istanbul.edu.tr
Res. Asst. Tugcen HATİPOĞLU
Sakarya University - Esentepe Campus
Institute of Science and Technology
54187 Serdivan-Sakarya/TURKEY
thatipoglu@sakarya.edu.tr
Abstract. Under the pressure of competition, the market is shared by a great deal of companies which differ in terms of their
productivity levels, adaptation abilities, trends or new technologies, strength levels through financial crisis and roles in enhancing the
employment. With their products and services Small and Medium Enterprises (SMEs) can compete with large-scale companies. On
the other hand, they can also support and complete the large-scale companies by constituting subsidiary industries. SMEs have an
important role in developing national economies. However, financial crisis affects the SMEs financially as well as the other
companies. Especially in crisis period, enterprises dealing with financial problems are expected to demand credits; while the banks are
more attentive to provide this demand because of non-performing credits. In order to determine the enterprises to provide credits, the
banks need some accurate statistical analysis. In this study for the year 2010, which is the liquidity crisis period, the financial distress
possibilities of 145 SMEs out of 250 enterprises that do credit application to a bank are evaluated with the help of ordered logit
regression by using relevant financial ratios. The analysis of data have pointed out that 7 of 19 financial ratios are regarded to be more
important than the others as being the main indicators for the banks in the usage of credit approval or denial decisions. In the study,
risk levels are determined with Altman Revised Z Score model and an early warning system is proposed.
Key Words: Financial distress, Altman Z Score, Ordered Logit Regression.
1. Introduction
Gaining sustainable economic growth, enhancing the employment degree and raising the standard of living in
countries would definitely contribute to the development of world economy. However, a sharp increase in financial
competition is observed in the intensified structure of financial environment which is triggered by globalization. The
market is shared by various financial or productive units with their own characteristic properties such as their
productivity levels, adaptation abilities, trends or new technologies, strength levels through financial crisis and roles
in enhancing the employment. Small and Medium Enterprises (SMEs) have an important role in both local and
global market environments while competing with large-scale companies. They also should exist in even a simple
market environment in order to support the large-scale companies by constituting them subsidiary industries. In
most economies, relatively small scaled enterprises are greater in number and SMEs are generally responsible for
driving innovation and competition. In such a multi-unit, dynamic and competitive market structure, maintaining the
financial stability can not be generally achieved. The credit-related businesses become progressively more diverse
and complex and they also grow rapidly, straining the limits of traditional methods for controlling and managing
credit risk.(Treacy & Carey ,2000) Moreover, the financial crisis may pose some additional persistent risks about the
economic health and may effect the overall consistency of the financial environment for an uncertain period of time.
The financial crash started in housing market in 2007 caused financial instability and spread in waves in years.
2010 was the year that the liquidity crisis exists intensively. Not only the real estate and bubble credit crisis but also
the processes performed based on them have triggered the conversion of this instability into a serious crisis
environment. (Ozturk & Govdere, 2010) Therefore during 2010, it would not be hard to say that a great deal of
companies needed financial help. The strength levels of the enterprises through the financial crisis differ through
their properties and enterprises usually confined to demand credits from the banks to get out of any kind of financial
problems.
SMEs’ financial decisions have crucial effects especially in the crisis period. For the least damage, some
supportive factors must be provided by banks such as granting SMEs credits when they are in financial need. On the
other side, in order to provide a better insight for the banks, the necessity of an accurate evaluation system for credit
risk scoring is inevitable within financial risk researches. The credit process in banks is to make decisions about the
credit approval or denial as a result of various financial assesments. The process policies would differ depending on
the organizational and financial structures of the potential borrowers. Therefore, in order to determine which
483

enterprises to assign credits, the banks need some certain and accurate statistical methods to analyze relevant
financial ratios. This issue is crucial in financial evaluation. Whether the banks accept the credit application and the
debt haven’t been repaid after the credit risk evaluation process; it would cause a huge economic loss. In other case,
whether some SMEs that have capabilities of repaying the credit have been rejected, it leads to block the way of
recovering the economy.
2. CREDIT RATING SYSTEMS
Many banks have introduced more structured or formal systems for approving loans, portfolio monitoring and
management reporting, analysis of the adequacy of loan loss reserves and loan pricing analysis. (Treacy & Carey
,2000) Rating judgements are known as the financial evaluations of potential borrowers’ ability to repay the debt.
The evaluation is fundamentally based on the historical borrowing and repayment data of the borrower, as well as
the availability of its assets and the extent of liabilities. Credit ratings have also been used for adjusting insurance
premiums, determining employment eligibility and establishing the amount of a utility or leasing deposit.
(Falavigna, 2010)
2.1 Altman Z-Score Model
The most famous credit risk model is Altman's Z-score model which employs accounting-ratio-based information in
determining and quantifying how the probability of default and a set of financial ratios are related. The Altman's Zscore
model relies mostly on information obtained from companies' financial statements (Leon Li & Miu, 2010) The
approach is a classification method which projects high-dimensional data onto a line and performs classification in
this one-dimensional space. The projection maximizes the distance between the means of the two classes while
minimizing the variance within each class. (Thomas Ng& Wong& Zhang, 2011) The Z-Score model is a linear
analysis in that five measures are objectively weighted and summed up to arrive at an overall score that then
becomes the basis for classification of firms into one of the priori groupings (distressed and nondistressed). (Altman,
2000)
However; Altman’s Revised Z score equation is adopted in this study which would let 3 different zones to
analyze instead of 2. The revised Z score model is presented as :
Z' = 0.717T1 + 0.847T2 + 3.107T3 + 0.420T4 + 0.998T5
Where :
T1 = (Current Assets-Current Liabilities) / Total Assets,
T2 = Retained Earnings / Total Assets,
T3 = Earnings before Interest and Taxes / Total Assets,
T4 = Book Value of Equity / Total Liabilities,
T5 = Sales/ Total Assets
Zones of Discrimination:
� Z' > 2.9 -“Safe” Zone
� 1.23 < Z' < 2. 9 -“Grey” Zone
� Z' < 1.23 -“Distress” Zone
Presumably, Altman’s Z-score is negatively correlated with the firm’s current and prospective financial
situation, and the firms with low Z-scores have relatively high demand for loans because without additional
borrowing, these firms may have to go out of business. However, profit-maximizing banks in a well-functioning
market economy typically would not be willing to advance additional loans to firms that have low Z-scores and,
therefore, are relatively likely to default.(Alexeev& Kim, 2008)
3. ORDERED (ORDINAL) LOGIT (LOGISTIC) REGRESSION
Ordinal logit models are extensively used in the literature on bond ratings, organizational ranking, and occupational
attainment, where the dependent variable has a natural ranking. In the case of a binary response variable, the
assumptions of linear regression are not valid. Logistic (Logit) regression is a method for fitting a regression curve,
484

y = f(x), when y consists of proportions, probabilities or binary coded (0-1 or failure - success) data. It is a nonlinear
regression model that forces the output (the predicted values) to be either 0 or 1 and is used when the dependent
variable is categorical. However, as with the other techniques, independent variables may be either continuous or
categorical in logit regression.
Although the outcome is discrete, the multinomial logit or probit models would fail to account for the ordinal
nature of the dependent variable. If the situation being modeled is unordered, an ordered model can lead to serious
biases in the estimation of the probabilities. On the other hand, if the type of event under study is ordered, an
ordered model loses efficiency rather than consistency ( Nam , 1997)
While examining access to credit markets, the underlying latent variable is the ability to observe and/or
demonstrate creditworthiness. Thus, both the borrower’s signaling efforts and the lender’s screening efforts are
involved. The cutpoints in the ordered logit models separate the different categories according to their ability to
demonstrate creditworthiness and the lender’s ability to evaluate this creditworthiness.(Joshi, 2005)
When an ordinal regression is fit, it is assumed that the relationships between the independent variables and the
logits are the same for all the logits. It means that the results are a set of parallel lines or planes. This assumption can
be checked by allowing the coefficients to vary, estimate, and then test whether or not they are all equal. (ASPC
v.13, 2011)
In order to analyze, it is often of interest to classify the units into various segments. For the banks, classifying
potential borrowers to their risk profile are usually defined from the outset as discrete categories with a certain
ordering. They form a finite and ordered partition into non-overlapping subsets. (Fok&Franses, 2002)
4. Application
In this study, the financial data related to 145 SMEs is used as sample that is determined as being more accurate data
within 250 SMEs that have done credit application to a public bank in year 2010, in Turkey. In order to analyze the
financial status of these enterprises, 19 characteristic financial ratios are selected at the outset.
X1 Liquidity ratio
X2 Current ratio
X3 Cash ratio
X4 Equity Multiplier (EM)
X5 Short term liabilities/Total assets
X6 Short term liabilities/Total liabilities
X7 Long term liabilities/ Total assets
X8 Long term liabilities/ Total liabilities
X9 Fixed assets/Equity
X10 Fixed assets/Continuous Equity
X11 Net Working Capital Turnover
X12 Current Assets Turnover
X13 Assets turnover (annual)
X14 Net Profit Margin (annual)
X15 Operational Profit Margin (OLPM)
(annual)
X16 Profit Capital Ratio – (annual)
X17 Return on Assets - annual
X18 Ebitda/Net sales
X19 Earnings Per Share (EPS)
Table 1: Selected financial ratios
485

Stepwise ordered logit model is preferred to find solution for possible multicollinearity problems and to
eliminate some of the correlated ratios from the analysis. At the first step, Altman Revised Z-score is calculated and
the financial status of firms are categorized in 3 groups as Safe, Grey and Distress. Safe enterprises are coded by
“2”, Grey enterprises are coded by “1” and Distress enterprises are coded by “0” according to their safety levels in
credit approval. The data is evaluated via SPSS Programme - Version 17 th .
N
Marginal
Percentage
Y .00 86 59.3%
1.00 18 12.4%
2.00 41 28.3%
Valid 145 100.0%
Missing 0
Total 145
Table 2: Case Processing Summary (Zones and Percentages)
The main aim is to determine the correct financial ratios for the banks to be used for the credit approval or
denial decisions when the financial status of the enterprise is addressed as an ordered logit variable and to propose
these ratios as the main indicators for the prementioned financial decisions. Testing the parallel lines assumption of
the model is crucial for the accuracy of gathered information.
Model
-2 Log
Likelihood Chi-Square df Sig.
Null Hypothesis 84.094
General 40.671 43.422 7 .100
*Link Function : logit
Table 3: Test of Parallel Lines* results
In connection with the inequation P>0.05, H1 is accepted. That means the assumption is ensured and the accuracy of
the results to be gained from the model is statistically proved. Then the model’s goodness of fit is evaluated.
Model
-2 Log
Likelihood Chi-Square df Sig.
Intercept Only 268.539
Final 84.094 184.445 7 .000
Link function: Logit.
Table 4: Model Fitting Information
Chi-Square df Sig.
Pearson 94.620 281 1.000
Deviance 84.094 281 1.000
Link function: Logit.
Table 5: Goodness-of-Fit results
486

Cox and Snell .870
Nagelkerke .854
McFadden .887
Link function: Logit.
Table 6: Pseudo R-Square Results
The inequation p0.05 in Table 5
indicates the success of the link function logit to hold the goodness of fit.
Table 6 which indicates the ability of providing explanation of the independent variables shows high ability of
explanation via the percentages of Cox and Snell R2 = %87, Nagelkerke R2 =%85 and McFadden R2=%88. The
output of the model is shown at below.
Estimate Std. Error Wald df Sig.
95% Confidence
Interval
Lower Upper
Bound Bound
Threshold [Y = .00] 26.529 5.239 25.645 1 .000 16.262 36.797
[Y = 1.00] 29.142 5.512 27.949 1 .000 18.338 39.946
Location X1 -1.015 .196 26.722 1 .000 -.630 1.400
X5 10.808 3.329 10.542 1 .001 4.284 17.331
X7 72.291 13.237 29.827 1 .000 46.347 98.234
X11 -.096 .045 4.483 1 .034 -.184 -.007
X13 -15.000 2.748 29.805 1 .000 -9.615 20.385
X16 -.243 .059 16.770 1 .000 -.127 .359
X17 -.204 .082 6.183 1 .013 -.365 -.043
Link function: Logit.
Table 7: Parameter Estimates Results
Because of Stepwise model is used, the output model is gained from the variables that all show significance.
In the study, significant X1 variable refers to the liquidity ratio. This ratio indicates the degree of safety for the
current assets in terms of liquidity level of the enterprise. As being an indicator for the sustainability of the activities
in current financial status and unexpected market conditions; the liquidity ratio is especially important for the
lenders in the sense of interpreting the borrowers’ liquidity level and adequacy for paying the debts back related to
the maturing liabilities.
Variable X5 is the ratio of short term liabilities/total assets. This ratio can also be used as a risk indicator for the
enterprises that indicates how much of the total assets is get funded by short term liabilities and how much of short
term liabilities is used within total assets. The highness of this ratio can be regarded as an indicator of the increase in
risk level and it also provides the probability of the increase in dividends for per shareholder by the leverage effect.
The ratio of short term liabilities/total assets can be used with Total Assets/ Active Assets ratio. If the ratio related to
the short term liabilities is low when the Total Assets/ Active Assets ratio is high; then a relatively less
problematical financial status can be observed.
Necessity for a lower percentage value than 33% for Short Term Liabilities/Active Assets ratio is adopted as
being a common criteria by the western financial agencies. However, in the countries that have problems about
resource procurement or have inflationary tendency this ratio is expected to be approximately 50%. Enterprises
should set a linkage between the maturity and the time the assets are used for.
Gaining liabilities with a different cast of mind may leave the enterprises in a difficult situation. Thus, the
enterprises that invest to current assets with their short term liabilities are noteworthy in the financial evaluation.
487

X7 variable is a ratio of long term liabilities/ total assets. This ratio shows how much long term liabilities in
total assets is and it gives information about the company’s power of providing long term funds. When this ratio is
high, the company should have its assets with long term resources simply and if it becomes greater, it would be an
indicator of having difficulties to pay amortization of the debt in recession period. In general, companies which
started doing high investments also have this financial ratio high. It can be interpreted as when this ratio does not
decrease after the investments, that investment’s added value is not satisfactory. In sectoral analysis, this ratio
requires to be paid attention. Besides, for a better interpretation, analyzing the short term liabilities should have been
caused drawing more reliable decisions.
Net working capital turnover which is shown by X11 variable is a ratio measure being how successful at sales
volume with minimum working capital (net working capital) for the sustaining activities. When this ratio is high, it
indicates that utilizing from net working capital efficiency and suitable, inventory and accounts receivables turnover
are high or inventories and receivables are requires low working capital, Short Term Financial Expenses at
organization is high and current ratio is low. On the other hand, when this ratio is low that means the company has a
greater working capital, inventory and accounts receivables turnover are low and the company’s cash value more
than needed. With increasing sales, net working capital is prone to increase. The main reason of this case is the
raising trade receivables with increasing demands.
X13, Assets turnover, is a ratio to determine the success of the company's sales revenue to it’s total assets. The
ratio financially shows whether the assets of the company grow essentially and whether or not the company invests
to the assets greater than planned before. This ratio should be regarded as a measurement of using technology or
consuming the assets within the company. If the fixed assets in total assets become relatively high, assets turnover
falls off. This case has been experienced mostly at capital intensive companies. On the other hand, it is normative to
have this ratio high at the companies that have low fixed assets (for instance trading companies and financial
companies) . Asset turnover is a significant indicator as a measure of the company’s profitability. When the other
circumstances are fixed, having this ratio high in a firm means that it’s profitability ratio would also be high. The
rate of profit in the future at the organizations ,which has a high long term assets (Fixed Assets) in consequence of
low assets turnover (particularly organizations have new investments) , is indefinite because of depending on
company’s demand changes. For this reason, we should use assets turnover as an indicator of risk. When the
company’s assets turnover is prone to decrease, the idle capacity might be the main reason of the case.
X16 variable refers to profit capital ratio measures companies’ performance of equity. Particularly comparing
companies in common sector this ratio is useful. If the ratio is high, it is the indicator of the company’s right
investment choices and ability to control related expenses.
X17 variable refers to return on assets that have been used for measuring the efficiency provided by the size of
the company. Return on assets indicates how efficient a company is at using the total assets including financial
investments. Basically Return on Assets is a combination of net profit margin and asset turnover.
5. Conclusion
Market is shared by various enterprises that are characteristically different and SMEs are the noteworthy productive
units for gaining sustainable innovation and competition. Financial crisis may have some forcing effects on SMEs
and these effects may cause serious damaging effects in financial structure of the market. For the financial support,
banks are generally the preferential choice of SMEs for the credit application in financial need. However, from the
sight of the banks; a set of SMEs should be chosen to determine the enterprises to approve the application.
Financial data of 145 SMEs that have done credit application in a public bank in 2010 which is liquidity crisis
period is analyzed in the study. For interpreting the financial status of the enterprises, 19 characteristic financial
ratios are selected at the outset. With Altman Revised Z score model the enterprises are divided into 3 zones in order
to separate the safety degree of them in granting credits. 86 enterprises (59.3%) are in Zone 0 and they would not be
able to pay any probable credit debts back. 18 enterprises are in Zone 1 (12.4%) that can not been clearly sure about
the financial power for the payment. 41 enterprises (59.3%) are in Zone 2 that are in safety zone for the banks to
grant credits. In order to analyze 19 different financial ratios to make the credit approval/denial decisions; 7 of these
ratios are proposed as being more noteworthy for the SMEs to represent their financial ability of paying the debts
488

ack. These 7 financial ratios are Liquidity Ratio, Short Term Liabilities/Total Assets, Short Term Liabilities/Active
Assets, Long Term Liabilities/ Total Assets, Net Working Capital Turnover, Assets Turnover, Return On Assets.
These ratios are the main indicators for the banks to determine the enterprises to grant credits. It is recommended for
SMEs to check relevant ratios permanently to keep the competitive edge in financial markets and for the banks to
pay more attention in financial decisions.
6. References
Alexeev, M., & Kim, S., (2008),The Korean financial crisis and the soft budget constraint ,Journal of Economic
Behavior & Organization, 68(1),178-193
Fok, D., & Franses, P.H., (2002), Ordered logit analysis for selectively sampled data, Computational Statistics &
Data Analysis, 40, 477 – 497
Joshi, M.G., (2005), Access to credit by hawkers: What is missing? Theory and evidence from India, Ph.D. Thesis,
The Ohio State University
Leon Li, M.,& Miu P, (2010), A hybrid bankruptcy prediction model with dynamic loadings on accounting-ratiobased
and market-based information: A binary quantile regression approach, Journal of Empirical Finance ,
17,818–833
Nam, D., (1997), Econometric analysis of highway incident duration, public perceptions and information for
advanced traveler information systems, Ph.D. Thesis. University of Washington
Özturk, S. & Govdere, B., (2010), The effects of the global financial crisis and Turkish economy, SDU The Journal
of Faculty of Economics and Administrative Sciences, 15(1), .377-397.
Thomas Ng, S., & Wong, J.M.V, & Zhang, J,(2011), Applying Z-score model to distinguish insolvent construction
companies in China, Hong Kong Habitat International
Treacy, W.F., & Carey M., (2000) , Credit risk rating systems at large US banks, Journal of Banking & Finance,
24, 167-201
http://www.cersi.it/itais2010/pdf/084.pdf, accesed at 30.04.2011
http://pages.stern.nyu.edu/~ealtman/Zscores.pdf, accessed at 13.04.2011
http://ww2.coastal.edu/kingw/statistics/R-tutorials/logistic.html, accesed at 20.04.2011
http://www.norusis.com/pdf/ASPC_v13.pdf, accessed at 25.04.2011
489

PRODUCTS WITH LONG AGING PERIOD IN THE AGRO-FOOD SYSTEM: ANALYSIS OF MEAT
SECTOR
Mattia Iotti, Giuseppe Bonazzi, Vlassios Salatas, University of Parma, Italy
Email: mattia.iotti@unipr.it , www.unipr.it
Abstract. The study aims to analyze the difference between profit and cash flow generation in firms operating in the meat sector with
long aging period of productions; this type of analysis could be useful in an applied field considering the long fresh meat aging period
of some productions, as ham, even considering the level of capital requirement to finance investment in land, building and machinery
in the sector. In fact, in general, this type of firm are often able to generate profit to provide a return to equityholders, but are not
always able to generate sufficient cash flow to pay debts and distribute dividends. About this topic, the paper would examine a panel
of firm in order to quantify the difference in applying an economic approach and a financial approach to evaluate investment,
comparing the results deriving from ratio analysis and cash flow analysis. The expected result of the paper is an application of ratio
analysis and cash flows analysis to be considered in small agro-food firm operating with aging of production.
Keywords: Long aging period, Accrual and financial approach, Free cash flow, Cover ratio
JEL classification: Q13 - Agricultural Markets and Marketing; Cooperatives; Agribusiness; Q14 - Agricultural Finance
1 Introduction
In Italy the meat sector is characterized by the herd of swine heavy pig that is bred to be processed for typical Italian
cold cuts production, especially the typical ham (PDO ham); for these hams are used fresh legs of pigs born, raised
and slaughtered in a defined area, and the pigs must have characteristics of quality defined in specific production
rules. The firms operating in the agro-food system that produce long aging period food have to face problems related
to the high level of capital requirement; often the firms have difficulties relating to the duration of the financial
cycle, because the firms require large investments in start-up activity for the acquisition of industrial buildings,
plants and equipments (Bonazzi et al., 2007); moreover, the capital requirement is inherent with the typical
production aging period, that requires large volumes of capital, then expanding the capital requirement for
equipment. In fact, the cycle of aging of the fresh meat causes a further expansion of capital requirements in order to
sustain the cycle of working capital. Considering the channel of sales frequently used by firms in the sector, as to
say namely large-scale distribution (GDO), it is to note an increase in average day extension in receiving payment
from customer, and this aspect of financial dynamic improve capital requirement for processing firms. In order to
comprehend the characteristics of management of this type of firm, it could be useful to compare economic and
financial results, in order to consider the attitude of the firm to generate cash flow sufficient to sustain the business
cycle, pay interest charge and distribute dividends to equityholders. In fact, in this type of firm, it is possible to note
a dyscrasia between economic and financial result, even because of the long period of aging of production. About
this topic, the paper would examine a panel of firm in the agro food system, operating in the long aging period food
production, in order to quantify the difference in applying an economic approach and a financial approach to
analyze the firm’s data in the sector, even comparing the results deriving from ratio analysis and cash flow analysis.
2 Methodology
2.1 Income statement analysis
The annual account, based on the rules of the Fourth EU Directive, is a general base, required by law, to evaluate
economic, financial and patrimonial aspects of the firm management; the annual account uses an accrual methods in
order to quantify profit to equityholders, considering positive and negative voices of revenue on an accrual basis.
This method allows to have attention to the creation of value and do not directly consider the moment in which it is
possible to achieve the cash inflow or there is the cash outflow. The income statement, as a part of annual account, is
used to express the economical result of a firm in term of profit, that is expressed using accrual methods. The
income statement considers the moment in which the value is created; in this way, the analysis of profitability is
performed using the analysis of annual economic accounts (Ferrero, et al., 2005). The annual accounts analysis
considers the data presented in tables required by law (Andrei et al., 2006), as defined by the European Union and
490

the national civil law in Europe (in Italy the civil code of commerce); in this way it is possible to compare data from
different firm, even from different countries (Andrei et al., 2006). The firm income considers the accrual basis
(Andrei et al., 2006) and expresses the moment of creation of value so the income statement is not dependent by the
generation of cash flow from operations. The scheme required by law in Italy is:
(1) Tω � ΔIα � Rβ � Sγ � Lδ � (D � A)ε � I � V � E � TX � Π
In (1) T is turnover (sales), I is the stock (inventory), where � is annual variation in order to apply accrual basis
analysis, R is raw materials, S is services, L is labour, D + A is depreciation and amortization, V is revaluations and
devaluation, E is extraordinary income or expenses, TX is income tax, � is profit; � is a multiplicative parameter
applied to respective quantity, in order to calculate the total amount of turnover (T�); respectively �, �, �, � and �
are multiplicative parameter applied to respective quantity, in order to calculate �I� (non monetary cost), R� is raw
materials cost, S� is services cost, L� is labour cost, (D + A)� is depreciation and amortization cost (non monetary
cost). So it is (Lagerkvist et al., 1996):
(2)
Tω �
I
�
i�1
T ω
i
i
; Iα �
�
j�1
�
k�1
Where in (2) i, j, k, l, m and n are respectively the different items of product sold (i) and items of production factor
bought (j, k, l, m and n). In (1) and (2) costs are joined having attention to the different nature of cost. It is possible
moreover to express as follow (3) considering that R� and S� express the external cost (Ce) defined as cost to
remunerate external production factor:
(3) Ce
� Rβ � Sγ
It could be useful to note that the pre tax civil income (��) is different from the pre tax fiscal income (��) in Italy,
because many type of cost with civil relevance have not fiscal relevance, it is:
And then, considering that Cnfr is cost with not fiscal relevance, and considering a tax rate t, is:
In order to obtain some more information about the capacity the generate source of finance, it is useful to apply the
value added reclassification form of the income statement (Ceccacci et al., 2008), that is expressed as follows:
(6) Tω � ΔIα � Ce
� VA ; VA � Lδ � EBITDA ; EBITDA � (D � A)ε � EBIT ; EBIT � I � V � E � TX � Π
In (6) EBITDA is earnings before interest, taxes, amortization and depreciation, EBIT is earnings before interest and
taxes. EBITDA expresses the capacity to generate cash and sustain the financial cycle considering and intermediate
income margin that is assumed to assimilate the creation of cash by income cycle, not considering non financial
cost, as depreciation and amortization.
2.2 Cash flow statement and balance sheet analysis
J
I α
j
j
; Rβ �
(4) Tω � ΔIα � Ce
� Lδ � (D � A)ε � I � V � E �
(5) (Πθ
� Cnfr
) � Πζ ; t � Πζ � TX
K
R
k
β
Πθ
k
; Sγ �
Then it could be applied the cash flows statement (Brealey et al., 2003) in order to quantify directly generation of
cash, deriving this analysis from the annual account; cash flow statement is also useful to analyze different sources
of cash flows (Shireves et al., 2000). It is that:
In (7) CF is cash flow, OCF is operating cash flow, UFCF is unlevered free cash flow, FCFE is the cash flow
available for equityholders (free cash flow). CF is profit (�) increased with costs that do not cause an outflow of
money (D + A) and the impact of income taxes (TX); OCF quantifies the absorption of net working capital (NWC)
and has a particular importance in the analysis of firms with great absorption of working capital, as long aging
L
�
l�1
S γ
l
l
; Lδ �
M
�
m�1
L
m
δ
m
; (D � A)ε �
�
�
(7) � � ( D � A) � TX � CF ; CF � Δ NWC � OCF ; OCF � Δ FA � UFCF ; UFCF�
DS � FCFE
491
N
�
n�1
(D � A) ε
n
n

period firm; UFCF is the sum of OCF and the absorption of capital resulting from investments in fixed assets (FA);
UFCF is the cash flow available for debt service (DS, debt service, where DS = K + I, with K is the principal and I is
interest charge). FCFE is the cash flow available for equityholders. The reclassification of the balance sheet
(Ceccacci et al., 2008) is conducted according to liquidity level:
(8) t
c
i
ar
TA � WC A � FA � WC A � WC A � WC A � FA
In (8) TA is total asset, WCtA is total investment (asset) in working capital, WCiA is working capital in inventories,
WCCA is working capital in cash, WCarA is working capital in account receivables, FA is fixed assets.
Reclassification of balance sheet liabilities is conducted according to the origin of sources of capital:
(9) TS � E � D � E � WC B � DF � DF
In (9) TS total liabilities and equity, E is equity, D is total debt, WCt B is total working capital liabilities, DFs is
short-term financial debts (due within 12 months), DFl is a medium/long term financial debts (duration over 12
months). The difference between WCtA and WCtB is net working capital (NWC). Expressed in this way, the
formulas allow to have a metric useful to compare economic and financial result, even considering traditional ratio
analysis (Roe, Roa, EBITDA / I, EBIT / I).
3 The meat sector
t
s
l
In Italy, the production of meat is one of the most important industry in the agro-food system, especially related to
the production of pork and bovine meat. The Italian production of pigs in 2009 is of 12,922,000 animals; 8,707,362
pigs are to produce cold cuts with protected designation of origin (PDO), on IPQ-INEQ data. The consumption of
pork, in Italy, is 37.68 kg per capita in 2009 and the self-supply rate was 68.9%. Genetic selection operates with two
addresses: for the cold cuts production, it is bred Italian Large White (LWI), Italian Landrance (LI) and Italian
Duroc (DI) even for production of meat for butcher is used race Petrain (P). In Italy operate 121 slaughterhouses in
the agro-food system (year 2008) that have slaughtered more than 9 million pigs; the maximum concentration of
slaughterhouses is in Lombardia region (38), Emilia Romagna region (27) and Piemonte region (18). All
slaughterhouses are inspected at least once a year to control information about the activity of slaughtering, in order
to detect the incoming streams of live pigs and outflows of raw material. This procedure, associated to a selection,
allows the exclusion of the fresh pork legs not suitable for production reducing the final index of non-compliance.
The consumption of cured meats in Italy in 2009 was 1.1745 million tons, having that the ham is the first cold cuts
for consumption in Italy (280.6 thousand tons), followed by cooked ham (275.8 thousand tons), Mortadella, with
173.9 thousand tons, and Salame, with 110,4 thousand tonnes. In Italy production of cold cuts with PDO and PGI
marks are 33, especially concentrated in northern regions. The Region that has the largest number of protected
products is Emilia Romagna (11 protected products, 7 PDO and 4 PGI), Lombardia (3 PDO and 6 PGI) and Veneto
(2 PDO and 4 PGI). In 2009 there were 8.680 milions slaughtered pigs certificates with PDO, mark 17.361 milions
available thigh for PDO production, 14.550 milions thighs started to PDO production, of which 9,429,462 for the
PDO Prosciutto di Parma and 2,521,213 for the PDO Prosciutto San Daniele.
In 2008, the herds for PDO production were 4,819 in 11 regions of central and northern Italy. The highest
concentration of farms is 1,936 farms in Lombardia, then we have Piemonte (970) and Emilia Romagna (926), so
that 79.5% of herds is located in these three regions. For the distribution of pigs per genotype, on 2008 data, there is
the prevalence of pigs from hybrid verro, representing 67.2% of the total.
Even in the meat sector, the European Community has implemented a strategy of diversification of farm
production in order to achieve a better balance between supply and demand in the markets, considering that
production, processing and distribution of agricultural products and foodstuffs has an important role in the
Community. Reg. (EC) No 510/2006 of 20 March 2006 regulates the protection of geographical indications and
designations of origin for agricultural products and foodstuffs. The verification of compliance with specifications
(Article 11) shall be performed before selling the product by one or more competent agency and / or one or more
control agency within the meaning of art. 2 of Reg (EC) No 882/2004; the control agency operates as a product
certification institution. Istituto Parma Qualità (IPQ), linked with Istituto Nord Est Qualità (INEQ) has implemented
a system that provides control and compliance requirements for origin of raw materials and production process
492

upstream in the chain. In this system of rules and controls, the farms must put the firm code and the month of birth
code on both legs, in order to have the slaughter of animals with at least nine months of life; in this way it is possible
within thirty days after the birth of the pig to exclude animals born outside the territory of origin. The transfer of
animals between farms should be documented so it could be easier the control issued on that by IPQ and INEQ. The
slaughterhouse must fill out a document for each day of production with a list of all lots of animals received and the
number of pigs slaughtered, codes of origin and origin. Moreover, the slaughterhouse puts on each thigh a stamp of
approval attesting to the code compliance of origin, provenance and quality.
Raw material arrives at the cold cuts production firm with the mark of identification and self-certification of
slaughter, having a copy of the document issued by the slaughterhouse. For every delivery, the production firm
analyzes the conformity of raw material and mark in an official register in which is possible to find the description
and all the identification elements of the product. IPQ and INEQ include these data in their database and check all
the documents produced for each lot of production. It is the responsibility of IPQ and INEQ to check all the quality
standards of the cured product by testing and verifying the minimum maturing period, the absence of morphological,
technical and taste defects. After these procedures, PDO certification mark is applied to product and it could be
marked on the product label.
4 Impact on local socio-economic system
The activities related to agriculture, processing industry and related services have a central role in the socioeconomic
system in the province of Parma. The food industry is the first industry in the province, with a turnover
(2008) of € 7,500 million, 36.6% of total industrial sales in the province, deriving € 973 million from exports. The
turnover of the mechanical food plant industry, that is the third largest industrial sector of the province, has a level
of € 2,200 million that is as 10.7% of total industry turnover. The sector of the food and plant for food, taken
together, amounted to 47.3% of the industry turnover of the province. The food sector with the highest turnover is
the sector of pasta, bread, pastries, frozen foods and related products, even considering the presence of some large
companies; turnover of the sector is, in 2008, € 3,000 milion (40.0% of provincial revenues in the food industry and
to 14.6% of industry turnover in general). The meat preserve industry generates 900 million euro turnover in 2008
(25.3% of sales in the food industry and 9.3% of industry sales in general).
The province of Parma has an important meat processing activities so the local socio-economic system is
characterized by the presence of a large number of firms in processing pork meat, especially in the production of
cold cuts. In the meat processing industry, having the analysis based on data made available by the Registry of firms
at the Chamber of Commerce of Parma, operates 446 firms and 143 local units of firms operating in the meat
processing industry in the province of Parma (10.13 main activity code of ATECO 2007 classification), as for a total
of 589 units in the sector in the province of Parma. In the meat sector work 4,399 staff employees and 322
independent operators, for a total of 4,721 unit of labour. The municipalities in the province with greater presence of
meat industry are Langhirano (123 companies, 41 local units, 1,140 staff employees and 72 independent operators),
Lesignano de' Bagni (38 firms, 10 local units, 352 staff employees, 18 independent operators) Felino (35 firms, 17
local units, 634 staff employees, 44 independent operators), Sala Baganza (26 firms, 17 local units, 609 staff
employees, 19 independent operators), 38 firms, 10 local units, 366 staff employees and 21 independent operators
operate in the municipality of Parma.
Parma PDO Ham is the most important production of cold cuts industry in the province of Parma and is
produced observing production regulations issued by the Consorzio del Prosciutto di Parma ensuring the respect of
EEC Regulation 2081/92 (now Regulation EC 510/06); Parma PDO Ham derive by processing thighs of heavy pig
that must be older than 9 months of age, weighing over 150 kg. The pig must is bred in the territory of 10 regions of
northern and central Italy but the production process must be done in one part of the province of Parma between the
Via Emilia, at a distance of at least 5 km from north, by the river Enza, to the east, and by the river Stirone, from the
west. The south limit of production is an altitude above sea level exceeding 900 meters. In addition to the Consorzio
del Prosciutto di Parma, Istituto Parma Qualità (IPQ) conducts necessary quality checks on the ham as "third and
independent part ".
Analyzing the firm per rank size, basing the analysis on produced hams per year, there is a concentration of
production in a small number of firms; with a total production analyzed of 9,429,642 hams processed in 181 plants,
493

with an average production of 52,096 hams per plant, 58.85% of production is concentrated in 25.41%
manufacturing plants that are characterized for an annual production of more than 100,000 hams. Manufacturing
plants with less than 25,000 hams per year produced 5.62% of the hams, involving 34.25% of plants; moreover it is
to note that the plants up to 1,000 pieces per year of production are 8.84% of the total (0.08% of production), while
the plants up to 10,000 pieces per year of production are the 22.10% of the total, with only 1,14% of the number of
Parma Ham in 2009.
Production plant
per rank size
Ham
(n.)
Ham
(%) Production plant (n.) Production plant (%)
0 – 1.000 7,652 0.08% 16 8.84%
1.001 – 10.000 136,870 1.45% 24 13.26%
10.001 – 25.000 384,970 4.08% 22 12.15%
25.001 – 50.000 2,067,900 21.93% 53 29.28%
50.001 – 75.000 1,282,433 13.60% 20 11.05%
75.001 – 100.000 2,161,614 22.92% 25 13.81%
100.001 – 200.000 2,316,990 24.57% 17 9.39%
> 200.000 1,071,033 11.36% 4 2.21%
Total 9,429,462 100.00% 181 100.00%
Source: IPQ
Table 1. Production plant per rank size (2009)
The consumption of Parma PDO Ham is for 79% on domestic market and 21% in foreign markets; 2,046,495 hams
are exported in 2009 (12,662 tons) with an estimated turnover of 181 million euro. France and the United States are
the most important foreign markets, and the whole European market accounts for 75.05% of exports, while the
American continent is 21.30% of exports, of which 18.81% in the USA alone; exports to other states are modest,
except Japan, that accounts for 4.28% of exports, approximately 87 thousand hams in 2008. During the last decade
there was an increasing in consumption of Parma ham sliced and packaged in boxes for sale in the refrigerated
counter. During the period 2005/2009 the increase in the number of meats sliced was equal to 83.2%, from 627,344
to 1,149,574 and the relative packages production increased from 30.885 million of 2005 to 54.796 milion of 2009.
The slicing process performed in the production chain make easier the consumption process, particularly in
foreign markets where the process of slicing made at the store or directly from the consumer is not always carried
out with the necessary expertise, thus penalizing the sale to final consumer. With regard to market, Parma sliced
ham, with a total production of 6,010,930 kg of food (1,865,490 kg for domestic consumption, up to 31.03%, and
4,145,440 kg, 68.97%, for export) confirms the presence of foreign demand for a product with a high level of
service. Even with respect to exports of sliced Parma PDO Ham, there is a concentration of demand in some foreign
markets, so the top 5 export destination markets are Britain, France, Belgium, Germany and the USA; these markets
generate 79.41% exports (49.12% for the first two target markets). Even if the sliced product is concentrated in
European market, that consumes 86.55% of exports, is also of relevance the USA market for the sliced product
(8.83% of the export market for sliced).
5 Data analysis
In this sector, the paper analyze the annual accounts (year 2009) of a sample of 50 firms in the area of Parma Ham
analyzing the balance sheet and income statement, on the basis of data made available by the local Chamber of
Commerce.
The analysis of the balance sheet shows an average TA of € 15.212 million to € 1.047 minimum and € 67.968
maximum, average WCtA / TA is 63.91% and FA / TA è 36,09%; �WCiA / TA is 37.17%. The other components of
working capital (WCcA+ WCarA) / TA, is 21.25%. The analysis around the sources of capital in the sample shows,
on average, that leverage is 2.495 and DER is 1.495.
494

Index Mean Median St. Dev. CV Min Max
Inventories 5,655,549 3,877,189 6,047,882 0.940 5,002 27,966,350
Account receivable 3,543,580 2.19,.453 3,805,939 0.931 125,932 17,823,799
Cash 325,810 30,002 632,808 0.515 22 2,806,379
Others 198,049 8,426 548,379 0.361 - 3,336,096
Working capital (inv.) 9,722,988 8,055,316 9,215,128 1.055 371,600 41,655,870
Fixed asset 5,489,484 3,404,495 6,090,518 0.901 108,167 34,006,538
Total Asset 15,212,472 11,587,352 13,892,680 1.095 1,047,065 67,968,144
Equity 6,096,607 4,678,132 5,738,354 1.06 295,708 24,674,954
Financial Debt (s) 3,001,646 1,646,256 4,067,973 0.738 - 20,520,274
Financial Debt (l) 2,490,703 904,519 3,488,198 0.714 - 13,155,378
Working Capital (debt) 3,623,516 2,367,118 3,411,318 1.062 321,802 18,096,661
Total Source 15,212,472 11,587,352 13,892,680 1.095 1,047,065 67,968,144
Source: firm data and elaborations
Table 2 – Investment and source (Firm Sample 50 firms – annual account 2009)
Index Mean Median St. Dev. CV Min Max ≤0 >0
EBITDA 458,565 305,627 620,765 0.74 - 192,552 2,750,309 5 45
EBIT 384,754 196,573 570,816 0.67 - 646,613 2,107,160 9 41
NET PROFIT 92,015 31,140 323,492 0.28 - 725,250 1,184,406 15 35
CF 31,969 232,527 514,245 0.06 - 206,715 2,539,986 6 44
OCF - 476,764 86,711 1,013,058 - 0.47 -1,683,759 3,800,912 21 29
UFCF - 593,887 - 88,701 835,774 - 0.71 -1,785,798 3,113,437 30 20
FCFE - 611,777 - 191,278 763,872 - 0.80 -2,110,005 2,088,789 33 17
Source: firm data and elaborations
Table 3 – Economic and financial margins (Firm Sample 50 firms – annual account 2009)
is 0.459 million €, 5 cases of EBITDA negative; EBIT average is 0.384 million €, 9 cases of EBIT negative on 50
firms; the average net profit is 0.092 million €, with 15 cases of negative net profit on 50 firms, so income analysis
of 50 firms sample in the 2009 shows in the sample, 35 firms generate profits (� > 0) and 15 generate losses (� ≤
0). In order to quantify the sources of liquidity, analyzing the creation of cash flow, data show a situation where
average CF is 0.039 million euro, 6 cases of CF negative, OCF average is -0.477 million €, 21 cases of OCF
negative on 50 firms; the average UFCF is -0.594 million €, with 30 cases of negative UFCF on 50 firms; average
FCFE is -0.612 million €, with 33 cases of negative FCFE on 50 firms. The analysis shows a great absorption of
capital in the working capital cycle, moreover having difficulty in generating a positive cash flow to serve debt
(UFCF) and to distribute dividends to shareholders (FCFE). In particular, working capital has a substantial effect on
the absorption of liquidity (5 cases of negative CF and 21 cases of negative OCF). It could be interesting to note that
there is a quite significance difference between net profit and FCFE means applying a t test (two sample paired per
means) as shown in table 4 with a significance level that is 92,61% (1-0,0739) in a 2 tails type analysis with a
reliability 95. A certain level of difference in means appears applying a t test (two sample paired per means) as
shown in table 4 with a significance level that is 88,75% (1-0,1125) in a 2 tails type analysis. A more clear
difference appears analyzing equity ratios (Roe and FCFE / E); the level of difference in means appears applying a t
test (two sample paired per means) with a reliability 95% as shown in table 4 with a significance level that is
99,88% (1-0,0012) in a 2 tails type analysis.
495

Analysis 1
NET PROFIT
Analysis 1
FCFE
Analysis 2
ROE
Analysis 2
FCFE / E
Means 216,3066 121,1119 0.845% -7.240%
Variance 730,616 261,527 0.299% 3.433%
Data 50 50 50 50
Pearson 0.9795 0.4689
Degree of freedom 49 49
Stat t 1.8264 3.4276
P(T 1;
even more (UFCF-I)/DF is > 0 in 17 cases. It is to note that EBITDA / I has 45 case of value > 0 and 38 > 1, but
OCF / I has 29 case of value > 0 and 25 > 1; applying to this couple of ratio the t test (two sample paired per means)
in order to quantify the significant level of means difference, it is as shown in table 4 with a significance level that is
86,55% (1-0,1345) in a 2 tails type analysis.
Index Mean Median St. Dev. CV Min Max >0 >1
ROE 0.8453% 1.0862% 5.4112% 0.16 -13.8454% 13.7024% 35 NU
ROA 2.4759% 2.4617% 3.1800% 0.78 -4.2638% 11.9884% 41 NU
EBITDA / I (interest) 133.136 2.653 621.967 0.214 - 21.738 4.232.818 45 38
EBIT / I (interest) 38.502 1.919 156.143 0.247 - 40.264 893.108 41 38
CF / I (interest) 105.316 1.883 521.214 0.202 - 23.032 3.617.948 44 32
OCF / I (interest) 71.063 1.085 369.030 0.193 - 69.246 2.547.273 29 25
UFCF / I (interest) -442.758 - 0.824 3.184.425 0.139 - 22.723.013 688.078 20 17
(UFCF - I (interest)) / DF 7.780 - 0.048 49.726 0.156 - 1.163 326.150 17 2
UFCF / DF 8.138 - 0.016 51.598 0.158 - 1.147 338.471 20 NU
FCFE / PN - 0.072 - 0.055 0.183 - 0.395 - 0.561 0.452 17 NU
Source: firm data and elaborations ; NU not useful
Table 5 – Economic and financial ratios (Firm Sample 50 firms – annual account 2009)
496

Analysis 3
EBITDA
Analysis 3
OCF
Analysis 4
EBITDA / I
Analysis 4
OCF / I
Means 458,564.62 294,736.44 133.1356 71.0631
Variance 393,212,880,156 1,047,230,428,214 394,737 138,962
Data 50 50 50 50
Pearson 0.7221 0.9618
Df 49 49
Stat t 1.6163 1.5219
P(T

ASS.I.CA. (2010), Rapporto Annuale 2009. Milano.
Binsbergen J., Graham J., Yang J. (2008), The cost of debt. Duke University, Working paper.
Bonazzi G. (2005), Prosciutto di Parma DOP e sistema dei controlli. In: Annali Facoltà Medicina Veterinaria di
Parma, Anno XXIV, 2004, Università degli Studi di Parma, Parma.
Bonazzi G., Iotti M., Salatas V. (2010), Financial Analysis of the Parma PDO Ham Firms. In: Proceedings of the 7th
International Conference on Applied Financial Economics, INEAG, Samos Island, Greece.
Brealey R.A., Myers S.C., Sandri S. (2003), Principi di finanza aziendale. Milano, McGraw-Hill.
Ceccacci G., Camanzi P., Rigato C. (2008), Basilea 2 per piccole e microimprese, Milano, Edizioni FAG.
Cleary S. (1999), The Relationship between Firm Investment and Financial Status, Journal of Finance, 54, 673-692.
Damodaran A. (1994). Damodaran on Valuation. Security Analysis for Investment and Corporate Finance. New
York, John Wiley & Sons.
Ferrero F., Dezzani F., Pisoni P., Puddu L. (2005), Le analisi di bilancio. Indici e flussi. Milano, Giuffrè.
Henry D. (1996), Cash Flow and Performance Measurement: Managing for Value, Financial Executives Research
Foundation. New York, Morristown.
Iotti M. (2009), La valutazione degli investimenti industriali. Milano, Franco Angeli
IPQ-INEQ, (2002), Certificare l’origine della qualità, IPQ-INEQ, Parma.
Kurshev A., & Strebulaev I. A. (2006), Firm size and leverage. Stanford, Stanford University.
Lagerkvist C.J., Andersson H. (1996), Taxes, inflation and financing – the rate of return to capital for the
agricultural firm, European review of agricultural economics, 23, 437 – 454.
Lewellen J.W., (2004), Predicting returns with financial ratios, Journal of Financial Economics 74, 209-235.
Sciarelli S. (2004), Fondamenti di economia e gestione delle imprese. Padova: CEDAM.
Shireves R.E., Wachowicz J.M. (2000), Free Cash Flow (FCF), Economic Value Added (EVATM), and Net Present
Value (NPV): A Reconciliation of Variations of Discounted-Cash-Flow (DCF) Valuation. Knoxville:
Tennessee University Press.
498

EMPLOYEE STOCK OPTIONS INCENTIVE EFFECTS: A CPT-BASED MODEL
Hamza BAHAJI, DRM Finance,Université de Paris Dauphine, France
Email: hbahaji@yahoo.fr
Abstract. This paper examines the incentives from stock options for loss-averse employees subject to probability weighting. Employing the
certainty equivalence principle, I built on insights from Cumulative Prospect Theory (CPT) to derive a continuous time model to value
options from the perspective of a representative employee. Consistent with a growing body of empirical and experimental studies, the model
predicts that the employee may overestimate the value of his options in-excess of their risk-neutral value. This is nevertheless in stark
contrast with a common finding of standard models based on the Expected Utility Theory (EUT) framework that options value to a riskaverse
undiversified employee is strictly lower than the value to risk-neutral outside investors. In particular, I proved that loss aversion and
probability weighting have countervailing effects on the option subjective value. In addition, for typical setting of preferences parameters
around the experimental estimates, and assuming the company is allowed to adjust existing compensation when making new stock option
grants, the model predicts that incentives are maximized for strike prices set around the stock price at inception. This finding is consistent
with companies’ actual compensation practices that standard EUT-based models have difficulties accommodating their existence.
Keywords: Stock options, Cumulative Prospect Theory, Incentives, Subjective value.
JEL Classification: J33, J44, G13, G32, M12
1 Introduction
Instead of an increasing interest for restricted stock and performance unit plans, the 2006 Hewitt Associates Total
Compensation Measurement survey has revealed that stock options are still the most prevalent long term incentive
vehicle 1 . The stated argument for the large use of executive stock options is that they align the interests of executives
and shareholders since they provide incentives for the manager to act in order to increase the firm value. The use of
stock options has even overtaken the traditional arena of executive population. Actually, firms’ compensation
practices show that stock options are issued to reward non-executive employees as well. In order to figure out the
reason why stock options may be attractive to employees it is crucial to assess the utility they receive from them.
Moreover, understanding how employee evaluates its stock options (i.e. their subjective value) allows assessing their
incentive power and the implied employee behaviour in terms of risk taking.
Most of the theoretical literature on stock options relies on the Expected Utility theory (EUT henceforth)
framework to derive models of option value from the employee perspective (Lambert et al, 1991; Hall and Merphy,
2000, 2002; Henderson, 2005). These models predict that the nontransferability of the options and the hedging
restrictions faced by the employee make him value his options below their issuance cost born by the company (i.e.
their risk neutral value). Moreover, standard normative models fail to predict stock options as part of the
compensation contract. Several quantitative studies taking place in principal-agent framework showed that EUTbased
models predict optimal compensation contracts which do not contain convex instruments like stock options
(Holmstrom and Milgrom, 1987; Dittmann and Maug, 2007).
This paper analyzes the valuation of stock options and their incentives effect to an employee exhibiting
preferences as described by Cumulative Prospect Theory (Tversky and Kahneman, 1992). It aims to propose an
alternative theoretical framework for the analysis of pay-to-performance sensitivity of equity-based compensation
that takes into account a number of prominent patterns of employee behavior that standard EUT cannot explain. This
work is motivated by recent empirical and theoretical researches on employee compensation incorporating CPTbased
models (Dittmann et al., 2008; Spalt, 2008). These models have proved successful in explaining some
observed compensation practices, and specifically the almost universal presence of stock options in the executive
compensation contracts that the EUT models have difficulties accommodating their existence. Therefore, they have
advanced CPT framework as a promising candidate for the analysis of equity-based compensation contracts.
1 80% of the responding participant companies to the survey have reported that stock option grants represent in 2006, on average, about 54% of
their global long-term incentives.
499

I drew on this theoretical framework to derive a continuous time model of the stock option subjective value
using the certainty equivalence principle. I then performed sensitivity analyses with respect to preferences-related
parameters and found that loss aversion and probability weighting have countervailing effects. In particular, I proved
that the option subjective value is increasing in probability weighting degree and decreasing in loss aversion. My
analyses also show that, for a given level of option moneyness, the subjective value of the option may lies strictly
above the Black and Scholes value (BS henceforth) when the effect of probability weighting tends to dominate that
of loss-aversion. These results lead to the conclusion that the lottery-like nature of stock-options, combining large
gains with small probabilities, may make them attractive to employees subject to probability weighting which is
consistent with the proposition that employee option value estimate may exceed the BS value (Lambert and Larcker,
2001 ; Hodge et al., 2006 ; Sawers et al., 2006 ; Hallock and Olson, 2006 ; Devers et al., 2007).
Furthermore, this work elaborates on incentives from stock options and on some implications in terms of design
aspects. Following previous researches, I defined incentives as the first order derivative of the subjective value with
respect to the stock price. A numerical analysis of the incentive function shows that stock option incentive effects
are increasing in employee’s degree of probability weighting and may even lie above incentives for a risk-neutral
individual. Moreover, I considered the incentive effects of setting the strike price of the option above or below the
stock price at inception. In this analysis, I relied on Hall’s and Murphy’s (2002) methodology in solving for the
exercise price that maximizes incentives holding constant the company cost of granting the options. I used this
approach to explore the situation where the company is allowed to adjust existing compensation when making new
stock option grants. For typical setting of preferences parameters around the experimental estimates from CPT
(Tversky and Kahneman, 1992), the model predicts that incentives are maximized for strike prices set around the
stock price at inception, which is consistent with companies’ actual compensation practices. Additional analyses
suggest also that loss-averse employees who are not subject to probability weighting, or even with very week
degrees of probability weighting, receiving options at high exercise price would willingly accept a cut in
compensation to receive instead deep discount options or restricted shares for those of them displaying more lossaversion.
This result is broadly consistent with Hall and Murphy (2002) and Henderson (2005) findings for nondiversified
risk-averse employees.
This article proceeds as follows. The first section describes the features of stock option value from the
perspective of a representative employee with preferences as described by CPT. Throughout this paper, we will refer
to this employee as a “CPT employee”. This section provides also numerical analyses on the model sensitivity to
preferences-related parameters. The next section introduces incentive effects of stock options for a CPT-employee
and examines some design implications in terms of strike price setting. The risk taking incentives question is
explored in the third and last section. Appendices provide proofs of the propositions in the first section.
2 Stock option value from a CPT-employee perspective
In this section, I develop a base-case model for analyzing the value of the stock-option contract from the perspective
of a representative employee with CPT-based preferences (subjective value henceforth). Specifically, I assume that
the employee is granted a European call option on the company’s stock, denoted by S, with maturity date t=T and
strike price K. These are the traditional features of executive stock options as reported in Johnson and Tian (2000)
and used by prior studies focused on stock option incentives (Lambert et al., 1991; Hall and Murphy, 2002;
Henderson, 2005). Often in practice, stock options are Bermudan-style options. Thus, my model relies on a naïve
setting in that it ignores complications related to early exercise or forfeitures.
2.1 Theoretical framework
2.1.1 Stock-option contract
The stock option contract is issued in t=0. The contract payoff at expiry, t=T, is � �
h S K �
� � . I make the
T T
assumption that the employee is not allowed to short-sell the company stock and that he can earn the risk-free rate r
from investing in a riskless asset. Moreover, the price dynamic of the stock is given by a geometric Brownian
motion represented by the following SDE:
dS � r � q S dt � � S dZ
(1)
� �
t t t t
500

2
Z is a standard Brownian motion with respect to the probability measure IP. � and q are respectively the total
t
variance of the stock price returns and the dividend yield.
2.1.2 Risk preferences
Following Tversky and Kahneman (1992), I consider that, to each gamble with a continuous random outcome
y IR
f y , the employee assigns the value:
� which the probability density function is denoted by � �
E � y� v � y�d f � y�
�
� � � �
IR
� �
Note that the expectation E �. � � is a function of two distinct functions. The first function, �. �
function, is assumed of the form:
�
� � ;
� �
(2)
v �
, called the value
�
� y �� y � �
v � y � � � � , where 0 � � � 1and
� � 1 (3)
�
�� �� � � y ; y � �
This formulation has some important features that distinguish it from the standard utility specification. First, the
utility is defined over gains and losses assessed based on the reference point denoted by � . The second important
feature is the shape of the value function. While it is convex over losses, it is concave over gains, which represents
the observation from psychology that people are risk averse over gains and become risk-seeking over losses.
Moreover, the value function has a kink at the origin introduced by the parameter � � 1.
This feature, known as
loss-aversion, gives a higher sensitivity to losses compared to gains. Finally, outcomes are treated separately from
other components of wealth, which reflects the well-documented phenomenon of narrow framing (Thaler, 1999).
The second function, �. a, b �
by the cumulative probability function F �. � , in order to transform them into decision weights according to:
a
� ��1 � F � y��
�
1
�
a a a
�� �F � y� � (1 � F � y�)
�
� , �F a b � y��
� � b
� F � y�
�
1
b � b b
�F � y� � (1 � F � y�)
�
; y � �
; y � �
where 0.28 � a � 1 and 0.28 � b � 1 (4)
� , is called the weighting function. It applies to cumulative probabilities, represented
��
This function stands for another piece of CPT, which is the nonlinear transformation of probabilities.
Specifically, it captures experimental evidence on people overweighting small probabilities and being more sensitive
to probability spreads at higher probability levels. The degree of probability weighting is controlled separately over
gains and losses by the weighting parameters a and b respectively. The more these parameters approach the lower
boundary at 0.28 the more the tails of probability distribution are overweighted. For instance, when a=b=1,
probability weighting assumption is relaxed. For simplicity, these parameters are assumed to be equal (i.e. a=b) and
� . in the rest of this paper. Finally, note that the lower boundary at 0.28 is
the weighting function will be denoted � �
�p
a
a technical condition to insure that ��a
, b � p�
is positive over ]0,1[ as required by the following first order condition:
a
a
�a � p �a � p� p�� p�
�1 � 1� � 1� � 0 .
2.2 Stock option subjective value
2.2.1 The model
In order to estimate the subjective value of the stock option contract described above, I use the certainty equivalence
principle. In particular, this value is defined as the cash amount, C , that leaves the employee indifferent between
� ,a
501

this amount and the uncertain payoff of the contract, h , irrespective of the composition of the remainder of his
T
private wealth. Formally, C is the solution of the following equation:
� ,a
� , a
rT � � IP � �a
� T ��
= � v �hT � d�a �F �S �
T ��
v C e E v h
� � �
IR
The left-hand side of the equation above represents the benefit for the employee from receiving the cash amount
C instead of the stock option contract at the inception of the latter. This amount is assumed to be placed into the
�
,a
risk-free asset over the whole lifetime T of the stock option contract. The other side of the equation gives the
expected utility to the employee from receiving the risky payoff implied by the value function v �. � � . Here, the
expectation relies on the transformed probability measure IP� . Let E represent the expectation in the right-
� ,a
EIP� a
hand side of (5). It follows from (5) and (3) that:
� � 1
�� � � �rT
��� � E e , if E 0
� , a � � � , a
��
� �
C � 1
� , a �� �
1 �
�� � � �rT
� � � E �e
, otherwise
� , a
�� �
�
�
� � �
�� �
�
Where E should write:
� ,a
With:
1
��
x
� �
, a a
I � g x � K �� � � u du � x dx
� � �
l1
� �� �
E � I � I � I (6.1)
1 2 3
� , a � , a � , a � , a
1
�
�
2
� � � � � � � � � � � (6.2) ; I � �� �K �� � g , a � x�� �� � a �u� du � � x� dx
� � � � � (6.3)
l2
� �
3
�
� ��� � �
� , a a � � � �
�
l2
a
(5)
(6)
l x
2 � � �
�� r�q� ��
T ��
T x
� �
� �� �
I
�
� ��
x dx
�
�
� (6.4) ; g( x) � Se
2 � (6.5)
2
� S � � � �
ln � r q T
K �
� � � � �
2
�
� � �
l1
� �
� �
� T
2
� S � � � �
ln � r q T
K
� � � � �
2
�
� �
(6.6); l2
� �
� �
� T
(6.7)
Here� �. � is the Gaussian density function and � a � p�
2.2.2 Reference point set up
��a
� �
�p
� p�
is the first-order partial derivative of �. a �
� .
Although CPT specifies the shape of the value function around the reference point, it does not provide guidance on
how people set their reference points. Neither does most of the psychological literature relying on the assumption
according to which the reference point is the Status quo. Instead, this literature admits both the existence and the
importance of non-status quo reference points since “there are situations in which gains and losses are coded
relative to an expectation or aspiration level that differs from the status quo” (Kahneman and Tversky, 1979).
In principle, employee would update the reference point in a way that fits with his own expectations regarding
the underlying share price at expiry. Intuitively, the employee could estimate the intrinsic value of the option based
on his future share price forecasts as he can rely on the BS value of the option disclosed by the firm. Following Spalt
(2008), I consider that the reference point parameter in the model,� , is the BS value 2 . Beyond the argument of
empirical evidence on employee exercise behaviour depending on non-status quo reference points (Huddart and
Lang, 1996; Heath et al, 1999), this assumption is supported by firms common practices in terms of stock option
2 To be more precise, the value used here is the expectation of the option payoff at expiring yielded by the BS model (i.e. the nondiscounted BS
value). Consistent with this specification, the probability measure IP used to derive the subjective value in (6) is the risk neutral probability
measure. Moreover, ignoring the probability weighting feature (i.e. a=1), this setting allows the subjective value implied by the model to
converge towards the risk neutral value (i.e. BS value) when the preferences of the employee tend to risk neutrality (i.e. α=λ=1).
502

compensation. Most of stock-option designers use the BS model in order to estimate the value of stock options as
constituents of the total compensation package. This value is usually announced to the employee at the inception of
the options. Moreover, the BS model is recommended in the FASB and the IASC guidelines for determining the fair
value of stock options (i.e. the amount an outside investor, with no hedge restrictions, would pay for the option) that
needs to be disclosed in the financial statements. These statements, comprising the BS value of the stock options, are
provided to shareholders as well as stakeholders including employees.
2.2.3 The impacts of preferences-related parameters: a numerical analysis
To provide a concrete outline on the profile of the subjective value yielded by the model relative to the risk neutral
value profile, I performed a numerical analysis 3 of the value of a 4-year call option (T=4) with a strike price K=100.
For the remaining option-related parameters, the figures were computed assuming no dividend payments
(q=0%), σ=30% and r=3%. Moreover, I set the curvature parameter of the value function (α) and the loss aversion
coefficient (λ) to respectively 0.88 and 2.25 based on experimental estimates from CPT (Tversky and Kahneman,
1992). Furthermore, in order to calibrate the probability weighting function, I used three different values of the
parameter a within the range of values estimated in the experimental literature 4 .
Figure 1 depicts the option value as a function of the stock price. The three blue curves represent the value
profiles from the perspective of three CPT employees with the same value function and different degrees of
probability weighting. At first sight, depending on the degree of probability weighting and the option moneyness
(S/K) the subjective value could lie either under or below the BS value. In contrast, standard EUT-based models
predict that the option value from a risk-averse employee perspective is systematically lower than the risk-neutral
value (Hall and Murphy, 2002; Henderson, 2005). These preliminary results are consistent thought with some
empirical findings suggesting that frequently employees are inclined to overestimate the value of their stock-options
compared with the BS value (Lambert and Larcker, 2001; Hodge et al., 2006; Sawers et al., 2006; Hallock and
Olson, 2006; Devers et al., 2007). On the other hand, the results presented in figure 1 show that the option subjective
value is increasing in probability weighting degree (i.e. decreasing in parameter a). Actually, given the asymmetric
profile of the option payoff, the expectation E in the subjective value formula (6) is positively affected by the
� ,a
emphasis put on the tail of the payoff distribution which is governed by the parameter a : the lower a the more
overweighed will be small probabilities and the more underweighted will be medium to large probabilities. This
lottery-like nature of an option, combining large gains with small probabilities, may make it attractive to a CPT
employee subject to probability weighting. This preliminary outcome leads to proposition 1 which stats that:
Proposition 1: the value of the stock option contract to a CPT- employee is increasing with respect to his
degree of probability weighting (i.e. is decreasing with respect to a) 5 .
Moreover, the effect of probability weighting is expected to increase with the skewness of the distribution of the
underlying stock price, which is captured by the volatility parameter σ given the Log-normality assumed in the
model. In order to show that, I performed a numerical analysis of the sensitivity of the subjective value to the
probability weighting degree as a function of the volatility σ and the parameter a. This sensitivity is defined as the
partial derivative of subjective value with respect to a. The results are reported in figure 2 in the form of a graph. It
shows that the sensitivity to parameter a is negative and locally decreasing in volatility. That means that the more
the share price is volatile the more the option will be attractive for a CPT employee subject to probability weighting.
This supports Spalt’s (2008) findings that the effect of probability weighting provides an economic rationale to
riskier firms (i.e. more volatile firms) for granting more stock options to non-executive employees.
Furthermore, I investigate the effect of loss aversion on the subjective value. The variable of interest here is λ.
In an analogical sense with the EUT framework, the option value from the perspective of a loss-averse employee is
expected to decrease with his degree of loss-aversion. To verify this I computed numerically the first order
derivative with respect to λ crossed over various levels of λ and moneyness, ranging from 0.05 to 1 and from 5% to
3 The integrals in (6.2), (6.3) and (6.4) were computed numerically.
4 Tversky and Kahneman (1992) got 0.65 on average (a=0.61 for gains and b=0.69 for losses). These results are corroborated by Abdellaoui’s
(2000) findings (a=0.60 for gains and b=0.70 for losses, hence an average of 0.65). In addition, Camerer and Ho (1994) obtained a=0.56 for gains
whereas Gonzales and Wu (1996) and Bleichrodt and Pinto (2000) found a=0.71 and a=0.67 respectively.
5 Proofs are available in appendices A and B of Bahaji H.(2010), Incentives from stock option grants: a behavioural approach, Working Paper,
DRM Finance, Paris Dauphine University. Available at: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1734831.
503

200% respectively. The outcome is reported in figure 3. It shows that the sensitivity to loss aversion is negative and
locally decreasing in moneyness. That means that the more the employee is loss averse, the less would the option be
worth to him. This conclusion is taken up in proposition 2 hereafter:
Proposition 2: the value of the stock option contract to a CPT- employee is decreasing with respect to his
degree of loss-aversion (i.e. a decreasing function of λ). (For proofs see footnote 5)
I performed a similar analysis in order to get a view on the effect of the curvature parameter α. In the same way,
I assessed locally the first order derivative with respect to α within a range of values from 0.05 to 1 and using share
prices ranging from 10 to 200. While not formally reported in this paper, the results show that this derivative is
locally increasing with the option moneyness for values of α above say 0.75. It also shows broadly that the
subjective value is an increasing monotone function of α over a range of values around the experimental estimate of
0.88 (from 0.7 to 1) irrespective of the option moneyness.
Figure-1: Option value against the stock
price. This figure is a plot of the
subjective value computed under
different probability weighting
parameters (the blue curves). It illustrates
the profile of the subjective value
compared to that of the risk neutral value
(red curve). The Parameters used are:
T=4; K=100; σ=30%; r=3%; q=0%;
�
3 Incentives from stock-options
Figure-2: Sensitivity to probability
weighting. This figure is a plot of the
partial derivative of the subjective value
with respect to “a” against both “a” and
the stock price volatility “σ”. It exhibits
the local effect of probability weighting
given the payoff distribution skewness
captured by “σ”. The derivative was
computed numerically based on the
following parameters: T=4; K=S=100;
r=3%; q=0%; �=2.25; α=0.88.
Figure-3: Sensitivity to loss aversion.
This figure is a plot of the partial
derivative of the subjective value with
respect to “λ” against both “λ” and the
stock price “S”. It exhibits the local
effect of loss aversion given the option
moneyness. The derivative was
computed numerically based on the
following parameters: T=4; σ=30%;
K=100; r=3%; q=0%; a=0.65; α=0.88.
Stock-options are incentive tools used within a principal-agent relationship to align the interests of the agent
(employee) on those of the principal (shareholders). The shareholders grant stock options in order to provide the
employee with incentives to make efforts that enhance the value of the firm, and thus their own wealth. Indeed,
assuming that employees are aware of how their actions affect the share price, option holdings will prompt them to
make efforts that increase share price. Therefore, the incentive from a single option grant will depend on the degree
of the sensitivity of the subjective value to the stock price.
3.1 The incentive measure
Following Jensen and Murphy (1990), Hall and Murphy (2000, 2002) and others, I defined the incentive effect as
the first order derivative of the subjective value with respect to share price which defines how the value from the
employee perspective changes with an incremental change in the stock price. A preliminary numerical analysis
relying on the setting reported in §2.2.3 shows that incentives are greatest for in-the-money 6 options and increasing
6 The terminology “at-the-money” is referring to stock-options with an exercise price equal to the stock price at inception. The expressions “Outof-
the-money” and “In-the-money” are also used throughout the paper to refer to options with strike price respectively below and above the grant
date stock price.
504

with the degree of probability weighting. Another result from the analysis is that, for sufficiently high level of
probability weighting, the option can give much more incentive to increase stock price than is reflected by the BS
delta. This is consistent with my previous finding that the subjective value can overstate the BS value. In addition,
consistently with the EUT-based models, the analysis results suggest that when the employee process probabilities
in a linear way (a=1) - which means that the probability weighting assumption is relaxed and only loss-aversion
matters - the incentives lie strictly under the BS delta whatever the level of the option moneyness. That is to say that
the options are less attractive for both risk-averse employee and loss-averse employee who are not subject to
probability weighting. Furthermore, with a hold constant at 0.65, I find that incentives are decreasing in lossaversion.
Conversely, with loss-aversion parameter λ set to 1, which means that the loss-aversion effect is
neutralized, incentives for a representative loss-neutral employee, with a degree of probability weighting equal to
experimental estimate of 0.65, overstate the BS delta.
3.2 Implications for stock option design: optimal strike price
Setting the strike price of standard stock options boils down to defining the threshold against which the performance
is assessed and, consequently, to determining the likelihood of a final payout. As stated in the previous section,
incentives increases in the option moneyness and, equivalently, decreases with strike price. In parallel, from the
shareholders perspective, granting in-the-money options is much more costly than granting out-of-the money or atthe-money
options (recall figure 1). This leads the firm to a trade-off to make when setting the exercise price of the
options in the sense that, holding his cost unchanged, she could either grant fewer options at a low strike price or
increase the grant size at higher exercise price.
I relied on the methodology from Hall and Murphy (2002) in figuring out the optimal exercise price satisfying
the double purpose of maximizing incentives and holding constant the firm’s cost of granting options. I considered
the situation where the employee and the firm are allowed to bargain efficiently over the terms of the compensation.
Thus, the firm is assumed to fund additional options by an adjustment to other compensation components that leaves
the employee indifferent between his initial package and the new package including the additional grant.
Let’s considerer then that the company is allowed to make an efficient adjustment to existing compensation
components (cash for example) to grant additional options to the employee. The impact of this adjustment should be
neutral with regard to the total compensation cost for the company. Moreover, assuming this adjustment involves
cash compensation, it must be attractive to the employee so that he’d be willing to give up some cash compensation
against extra options grant. Therefore, it must leave the employee at his initial total subjective value of the
compensation package. It follows that the strike price that maximizes total incentives for a given company cost is the
solution of this following optimization problem:
�nC
, a �K � �
max subject to n� �K � � nC ,a �K � � c and �
n � 0 (7)
K �S
Where
� ,a
� �
�nC
K
denotes the incentives from receiving n options with a strike price K , � �K � is the per-unit
�S
cost or the BS value of one option and c is a fixed constant. The constraint in (7) is the aggregation of the company
cost constraint and the employee value constraint used in Henderson (2005). This optimization problem was solved
numerically by varying the parameter K. First, the BS and the subjective values are computed for a given K which
enables to determine the grant size n in accordance with the constraint in (7) for K≠S. Then, the cost function is
assessed based on n and K. This procedure is reiterated recursively until the optimal value of K is found. In this
analysis, c was chosen such that for retained parameters, the number of granted at-the-money options n is around
1000, hence a total cost of €28 333.
The left-hand side sub-figure in figure-4 exhibits total incentives for different levels of K and probability
weighting parameter a, with risk aversion held constant at λ=2.25. For each combination of K and a the constraint in
(7) is solved for n, which allows to determine total incentives. The plots indicate that when the employee is deeply
subject to probability weighting (a≤ 0.475), total incentives are strictly decreasing throughout the depicted range of
strike prices (see the curves in blue). In this case, similar to EUT-based studies’ findings (Hall and Murphy, 2002;
Henderson, 2005), incentives are maximized through restricted stocks grant rather than stock options. The intuitive
reason behind that is that, given that the employee systematically values the options in excess of their BS value,
505

efficient trade-off over compensation allocation is made via the grant of equity-based instruments that employee
values at their actual cost, restricted stocks for instance. However, for lower degrees of probability weighting
(typically 0.475

*
the effect of loss-aversion on K by holding a constant at 0.65 and varying λ. It mainly shows that loss-aversion has
an effect opposite to that of probability weighting: when probability weighting effect is dominant 8 (i.e. the employee
may potentially put overstated value in the option, specifically for high strikes) optimal strike increases and total
incentives decreased with loss-aversion. Conversely, when loss-aversion effect is dominant, as stated before, the
model yields predictions comparable to that of the EUT-based models.
4 Conclusion
This paper proposes an alternative theoretical model of stock option subjective value to analyze their incentive
effects for employees. The model predictions ascertain the ability of CPT to explain some prominent incentive
patterns that EUT models have difficulties to capture. It mainly provides arguments on the well documented
tendency of employees to frequently value - under some circumstances - their options in excess of their cost to the
company. Specifically, these results highlight the economic rational for firms, in particular those with higher risk, to
widely use stock options in non-executive employee compensation (Spalt, 2008).
Loss aversion and probability weighting are the key features driving the subjective value in the CPT model.
These parameters have countervailing effects on the modeled subjective value. Depending on which of them is
dominant in the preferences calibration, the model yields different predictions regarding incentives. Specifically,
consistent with some behavioral patterns observed in many surveys and experimental studies on equity-based
compensation, the model predicts that, when the probability weighting feature prevails, the subjective value may
overstate the risk-neutral value of the option. In this case, assuming the company and the employee bargain
efficiently over the compensation components, incentives are maximized for strike prices set around the stock price
at inception for a representative employee with preferences calibration meeting the experimental estimates from
CPT (Tversky and Kahneman, 1992). Obviously, this finding is consistent with companies’ actual compensation
practices. Moreover, executives with such preferences profile may be prompted to act in order to increase share
price volatility. However, when the emphasis is put on the loss-aversion feature, by relaxing the probability
weighting assumption, the model yields results comparable to those of EUT-based models. In particular, the model
predicts that loss-averse employees who are not subject to probability weighting, or even with very low degrees of
probability weighting, receiving options at high exercise price would willingly accept a cut in compensation to
receive instead deep discount options or restricted shares for those of them displaying more loss-aversion.
Despite their practical interest, the conclusions from the results of this research should not be drawn without
underlining some of its limitations. The first one concerns the specification of the CPT model. Actually, although
the reference point specification in the model is consistent with both empirical evidence on people setting reference
points in a dynamic fashion and firms’ widespread use of the BS value as a standard for Financial and Human
Resources disclosers, empirical and experimental literature is still silent on how people set reference points when
assessing complex gambles like stock option payoffs. In addition, to keep the model tractable, only European-style
options were studied in this paper. The model is still however easily extendable to Bermudian-style options using
numerical schemes such like lattice approaches. The other limitation of this study is related to the large
heterogeneity in probability weighting that may exist across individuals (Wu and Gonzalez, 1996). The Tversky’s
and Kahneman’s (1992) weighting function used in the model provides only a fit to the median profile.
Finally, this work highlights - as did some previous eminent researches in this field (Dittmann et al., 2008;
Splat, 2008) - a number of future promising research directions in equity-based compensation incorporating CPT
framework. For instance, exploring the ability of CTP to explain the growing use of performance shares plans
instead of stock options in employee compensation would be of great interest for future research. Furthermore, given
that several empirical studies has documented that employee stock option exercise behaviour is also driven by
behavioural factors, a promising research direction incorporating CPT is studying its ability to predict exercise
patterns.
8 See the limit case of λ=1 where risk-aversion is ignored (the blue dashed curve).
507

5 References
Abdellaoui, M. (2000), “Parameter-Free Elicitation of Utility and Probability Weighting Functions”, Management
Science, Vol. 46 No. 11, pp. 1497-1512.
Bleichrodt, H. and Pinto, J.L. (2000), “A Parameter-Free Elicitation of the Probability Weighting Function in
Medical Decision Analysis”, Management Science, Vol. 46 No. 11, pp. 1485-1496.
Camerer, C. and Ho, T.H. (1994), “Violations of Betweenness Axiom and Nonlinearity in Probability”, Journal of
Risk and Uncertainty, Vol.8 No. 2, pp. 167-196.
Devers, C., Wiseman, R. and Holmes, M. (2007), “The effects of endowment and loss aversion in managerial stock
option valuation”, Academy of Management Journal, Vol. 50 No. 1, pp. 191-208.
Dittmann, I. and Maug, E. (2007), “Lower salaries and no options? On the optimal structure of Executive pay”, The
Journal of Finance, Vol. 62 No. 1, pp. 303-343.
Hall, B. and Murphy, K. (2000), “Optimal Exercise prices for executive stock options”, American Economic Review,
Vol. 90 No. 2, pp. 209-214.
Hall, B. and Murphy, K. (2002), “Stock options for undiversified executives”, Journal of Accounting and
Economics, Vol. 33 No. 2, pp. 3-42.
Hallock, K. and Olson, G. (2006), “The value of stock options to non-executive employees”, Working Paper, No.
11950, National Bureau of Economic Research, Cambridge.
Heath, C., Huddart, S. and Lang, M. (1999), “Psychological factors and stock option exercise”, Quarterly Journal of
Economics, Vol. 114 No. 2, pp. 601-628.
Henderson, V. (2005), The impact of the market portfolio on the valuation, incentives and optimality of executive
stock options, Quantitative Finance, Vol. 5 No. 1, pp. 35-47.
Hodge, F. Rajgopal, S. and Shevlin, T. (2006), “How do managers value stock options and restricted stock ?”,
Working Paper, University of Washington.
Holmstrom, B.R. and Milgrom, P.R. (1987), “Aggregation and linearity in provision of intertemporal incentives”,
Econometrica, Vol. 55 No. 2, pp. 303-328.
Huddart, S. and Lang, M. (1996), “Employee stock option exercises: an empirical analysis”, Journal of Accounting
and Economics, Vol. 21 No. 1, pp. 5-43.
Jensen, M. and Murphy, K.J. (1990), “Performance pay and top-management incentives”, Journal of Political
Economy, Vol. 98 No. 2, pp. 225-264.
Johnson, S.A. and Tian, Y.S. (2000), “The value and incentive effects of non-traditional executive stock option
plans”, Journal of Financial Economics, Vol. 57 No. 1, pp. 3-34.
Kahneman, D. and Tversky, A. (1979), “Prospect Theory: An analysis of decision under risk”, Econometrica, Vol.
47 No. 2, pp. 263-292.
Lambert, R., Larcker, D. and Verrecchia, R. (1991), “Portfolio considerations in valuing Executive Compensation”,
Journal of Accounting Research, Vol. 29 No. 1, pp. 129-149.
Lambert, R. and Larcker, D. (2001), “How do employees value (often incorrectly) their stock options”, available at:
knowledge@wharton.
Sawers, K., Wright, A. and Zamora, V. (2006), “Loss aversion, stock-based compensation and managerial riskseeking
behavior”, paper presented at the AAA 2007 Management Accounting Section Meeting, available at:
http://ssrn.com/ abstract=864224
Thaler, R. (1999), “Mental Accounting Matters”, Journal of Behavioral Decision Making, Vol. 12 No. 3, pp. 183-
206.
Tversky, A. and Kahneman, D. (1992), “Advances in Prospect Theory: Cumulative representation of uncertainty”,
Journal of Risk and Uncertainty, Vol. 5 No. 4, pp. 297-323.
Wu, G. and Gonzalez, R. (1996), “Curvature of the Probability Weighting function”, Management Science, Vol. 42
No. 12, pp. 1676-1690.
508

509

EMERGING MARKETS
510

511

COMOVEMENTS IN THE VOLATILITY OF EMERGING EUROPEAN STOCK MARKETS
Radu Lupu, Institute for Economic Forecasting, Romania
Iulia Lupu, Center for Financial and Monetary Research ”Victor Slavescu”, Romania
Email: radu.a.lupu@gmail.com, iulia.lupu@gmail.com
Abstract. The analysis of the comovements of stock market returns was approached with many modeling techniques ranging from the
simple and GARCH style dynamic conditional correlation to multivariate GARCH and studies of the bivariate distribution. The quest
for the analysis of the now standardized concept of international contagion made room for the employment of all these techniques. Our
paper focuses on the analysis of the comovements in the volatilities of the returns of stock market indices from the most important
developed and emerging European countries, using different forms of computation for different frequencies, starting from intra-day 5minute
returns to weekly returns (data used from Bloomberg). After a brief characterization of the distribution of returns and a
reconfirmation of the stylized facts for the European emerging markets we focus on the clustering effect of volatilities, in the attempt
to identify the moments when a new cluster is formed, i.e. when the volatilities change their size (from small to big or from big to
small). The analysis of these events for the respective countries intends to reveal the mechanism of international information
transmission. The paper also fits a jump-diffusion process, along the lines of Maheu and McCurdy (2007) adjusted for the series of
volatilities, where the Poisson process characterizes the time until a change in the volatility cluster occurs.
Keywords: comovement, returns’ volatility, European emerging stock markets
JEL classification:C39, G15
1 Introduction
Even if the process of financial globalization has not followed a linear trend over time, and the exact timing of
financial liberalization remains somewhat controversial, there is a broad consensus among economists that capital
markets are much moreintegrated today than they were 30 years ago.
Over the years, many papers have contributed to the very important debate onthe interaction across international
stock markets, looking at volatility spillovers, correlation breakdowns, trends in correlation patterns. The integration
of the financial markets wore down muchof the gains from international diversification which rely on low
correlations across internationalstock markets. The intensity of the comovements and spillovers effects driven by the
financial integration may increase the risk of global financial instability.
Most studies into comovements in stock markets have focused on developed economies. Lately there has been a
growing body of empirical research on emerging capital markets, partly in response to the diversifying activities of
multinational enterprises in these markets, and as a result of the growing interest shown by private and large
institutional investors seeking to diversify their portfolios in international capital markets. International differences
in the institutional framework of emerging market economies may playan important role on the magnitude of shocks
transmitted across countries.
When one compares US with European stock markets, that have different trading hours, we can observe the
international transmission mechanism. When New York stock market opens its business day, many things already
happened on the European stock market. Similarly, European brokers take into account how New York market
ended. These equity markets are linked through trade and investment and because of that, any change about
economic fundamentals in one region can have implications for the other one. In the same time, both foreign
exchange markets and national stock markets share a number offacts for which a satisfactory explanation is still
missing in standard theoriesof financial markets.
In this paper the analysis of the comovements of stock market returns was approached with many modeling
techniques ranging from the simple and GARCH style dynamic conditional correlation to multivariate GARCH and
studies of the bivariate distribution. The paper investigates the behavior of stock market indexes in two respective
ways: on one hand, we study the properties of the jumps (outliers) for the high-frequency stock market returns when
we look at many European indexes and, on the other hand, we try to capture the behavior of the co-movement of the
volatilities of these indexes at the same frequency.
512

This paper is organized as follows. In Section 2, is presented the literature review. In Section 3, a dataoverview
and methodology are provided and in Section 4 the empirical results are discussed. Finally, Section 5 concludes.
2 Literature review
Economists have been studyingwhy there is propagation of volatility from one market to another, since
alongtime.Grubel (1968) is a most sited paper representing the start of researches in the field of stock market returns
comovement. Since that, the analysis of the benefit of international portfolio diversification and of the stock market
synchronization received a special attention in international finance.
Bekaert and Harvey (1995), Forbes and Rigobon (1999) are only some important paper that investigate the
cross-country linkages between stock markets. As a matter of fact, a growing body of literature has emerged more
recently on the issue of international stock prices comovement (King et al., (1994), Lin et al. (1994), Longin and
Solnik (1995, 2001), Karolyi and Stulz (1996), Forbes and Rigobon(2002), Brooks and Del Negro (2005, 2006)). In
particular, most of those studies have found that the comovement of stock returnsis not constant over time. Evidence
of increasing international comovement of stock returns since the mid-90s among the major developed countries,
was found by Brooks and Del Negro (2004) and Kizys and Pierdzioch (2009).
The determinants of the cross-country financial interconnections are split in trade intensity factors (Chinn and
Forbes, 2004), financial development factors (Dellas and Hess, 2005), business cycle synchronization factors (Walti,
2005), and geographical variables (Flavin et al., 2002). The divergence of the results and main conclusions can be
partly explained by the high degree of heterogeneity 1 in the empirical approaches adopted by the literature.
However, the comovement analysis should also take into account the distinction between the short and long-term
investor (Candelon et al. (2008)).
The comovement of stock returns was evaluated usually through the correlation coefficient while the evolving
properties have been investigated either through a rolling window correlation coefficient as in the study of Brooks
and Del Negro (2004) or by considering nonoverlapping sample periods like in the paper of King and Wadhwani
(1990) and Lin et al. (1994).Morana andBeltratti (2008) found strong linkagesacross European, the US and the
Pacific Basin stock markets, involving comovements in prices, returnsand volatility over the period 1973–2004.
Their results show that the heterogeneity between Europe and the US has steadily reduced, these markets being
currently strongly integrated.
Applying a new technique, the wavelet analysis, that allow for time and frequency simultaneous
characterization, Rua and Nunes (2009) focused on Germany, Japan, UK and US stock markets over the last four
decades. A noteworthy finding of their paper is that the strength of the comovement ofinternational stock returns
depends on the frequency. The authors argue that the comovement between markets is stronger at thelower
frequencies suggesting that the benefits from international diversification may be relatively less important in the
long-term than in the short-term and that the strength of the comovement in thetime-frequency space varies across
countries as well as across sectors.Taking into account that the United States was the crises epicenter, Didier and all
(2010) analyzed the factors driving the comovement between US and 83 countries stock returns, differentiating the
periods before and after the collapse of Lehman Brother. The authors have argued that there is evidence of a “wakeup
call” or “demonstrationeffect” in the first stage of the crisiswhere investorsbecame aware that certain
vulnerabilities present in the US context could put other economies atrisk, because countrieswith vulnerable banking
and corporate sectors exhibited higher comovement with the US market, the main transmission channel being the
financial one.
Thestock exchanges from Central and Eastern European countriesperform in a quite different manner as
compared to developed markets. First studies that specified these differences are Barry, Peavy III and Rodriguez
(1998), Harvey (1995), Divecha, Drach and Stefek (1992), as well as Bekaert et al. (1998). The literature evidenced
a number of empirical regularities: high volatility, low correlations with developed markets and within emerging
markets, high long-horizon returns, and more variability in the predictability power as compared to the returns of the
1 Heterogeneity is observed in the sample of included countries (developed vs. developing countries), the nature of the econometric approach
(cross-sectional vs. time-series), the measurement of market comovement, and the nature and measurement of explanatory factors.
513

stocks traded in the developed markets. It is also well evidenced that emerging markets are more likely to experience
shocks induced by regulatory changes, exchange rate devaluations, and political crises.
Pajuste (2002) observes that Central and Eastern European capital markets are quite different in terms of their
correlations with European Union capital markets. While the Czech Republic, Hungary and Poland display higher
correlations among them and with the European Union market, Romania and Slovenia show inexistent or even
negative correlation with the European Union capital market. A stock market convergence of Central andEastern
European (CEE) countries to the rest of Europe was studied furthermore by Harrison and Moore (2009), using
threeapproaches to obtain time-varying estimates of thecomovement between returns: realized correlation analysis,
rolling unit root tests, andrecursive cointegration tests.The results suggest that there is arelatively weak correlation
between stock markets in CEE countries and those in Europe, with the link between the exchanges strengthening
since 2002.Analysis in this area are realized as well by Horobet and Lupu (2009) and Lupu and Lupu (2009)
showing the properties of these correlations with different techniques – cointegration and Granger causality tests on
one hand and dynamic conditional correlations performed at the burst of the actual crisis on the other hand.
Harrison, Lupu and Lupu (2010) identified in their paper the statistical properties of the Central and Eastern
European stock market dynamics. The paper focuses on the stock market indices of ten emerging countries from the
Central and Eastern European region – Slovenia, Slovak Republic, Estonia, Latvia, Lithuania, Bulgaria, Czech
Republic, Romania, Hungary and Poland –over the 1994-2006 period and present evidence of stationarity for the
returns of these indices and identified some common characteristics of these markets taken as a whole.
The international transmission of stock returns and volatility was investigated by Lin, Engle and Ito (1994)
usingintradaily data that define Tokyo and New York markets. They argue that information revealed duringthe
trading hours of one market has a globalimpact on the returns of the other market and that the interdependencein
returns and volatilities is generally bi-directional.
As stipulated in the literature (Soydemir (2000)), emerging markets respond morequickly to shocks originating
in their own market than from foreign market disturbances and the emerging market economies thathave opened
their markets to achieve greater financial integration are more prone to externalshocks.
Current research has documented the importance of jump dynamics in combinationwith autoregressive volatility
for modeling returns. Jorion (1988), Andersenet al. (2002), Chib et al. (2002), Eraker et al.(2003), Chernov et al.
(2003), and Maheuand McCurdy (2004) are only a few examples. Jumps provide auseful addition to
stochasticvolatility models by explaining occasional, large abrupt moves in financialmarkets,accounting for
neglected structure, but they are generally not used to capture volatility clustering. Maheu and McCurdy
(2007)proposed a new discrete-time model of returns in which jumps capture persistence in the conditional variance
and higher-order moments and the evaluation focuses on the dynamics of theconditional distribution of returns using
density and variance forecasts. The empirical results indicate that the heterogeneous jump model effectively captures
volatilitypersistence through jump clustering and that the jump-size variance isheteroskedastic and increasing in
volatile markets.
3 Data and methodology
The data that we used consists of five-minute stock market index returns from some of the developed European
markets as well as the Eastern markets: DAX (Germany), CAC (France), UKX (UK), IBEX (Spain), SMI
(Switzerland), FTSEMIB (Italy), PSI20 (Portugal) ISEQ (Ireland), ATX (Austria), WIG (Poland), PX (Czech
Republic), BUX (Hungary), BET (Romania) and SBITOP (Slovenia). The period we took into account was from the
3rd of August 2010 until the 10th of February 2011.
The trading sessions are different in the countries in our analysis (some start at 8:00 hours, local time, others
start at 8:30 and they tend to stop at different moments) and this is why, since we are interested in studying the comovement
of these returns, we had to build a database that identifies the moments in time when all the indexes were
traded. Another issue was that the high frequency returns tend to have a small size and at the turn of the day we may
find higher values for the returns. This is why we decided to take out of the sample the returns that were recorded at
the change of the day (the returns from the value of the index at the end of the day to the value of the index at the
514

eginning of the next day). Therefore, our returns are not presumed to show any jumps (outliers) caused by the
accumulation of information between trading sessions.
.015
.010
.005
.000
-.005
-.010
-.015
-.020
-.025
SBITOP
BET
BUX
PX
WIG
ATX
ISEQ
PSI20
FTSEMIB
SMI
IBEX
UKX
CAC
DAX
Figure 1. The distribution of the five-minute returns: medians, means and outliers
In terms of methodology, we started the analysis with the identification of outliers in sense of finding the
moments when the returns were outside a 95% confidence interval for each of the stock index. Next, we simply
made an analysis of the simultaneity of the jumps for the common sample of all the indexes. The results of this
analysis are provided in the next section.
The following step was to use a Dynamic Conditional Correlation GARCH-like model (DCC) to fit the changes
in the correlations of the returns for the five-minute frequency. The dynamical correlations as well as the GARCH
estimates of the volatilities were then used to characterize the co-movement of the high-frequency returns.
The specification of the DCC model starts from the GARCH-like specification
2
� � � � �R
R
2
� �� ,
which makes the correlations between returns iand j to be:
�
ij,
t�1
�
ij,
t�1
i,
t j,
t ij,
t
� � �R
i,
t
R
j,
t
� ��
2
ij,
t
2 2
2 2
( � � �Ri,
t � �� i,
t )( � � �R
j,
t � �� j,
t
Standardizing each return by its dynamic standard deviation, we get
z
, �1 � i t
R
�
We notice the conditional covariance of news equals the conditional correlation of the raw returns
i,
t�1
i,
t�1
�R R �
��
�
� � �
R ��
R � E
�
E � ��
��
t z z
�
��
��
��
i,
t�1
j,
t�1
t i,
t �1
j,
t�1
ij,
t�1
i,
t�1
j,
t�1
� Et
�
� � , �1
��
��
�
ij t
�
�
i,
t�1
� j,
t�1
� i,
t�1�
j,
t�1
� i,
t�1�
j,
t�1
515

Thus modelling the conditional correlation of the raw returns is equivalent to modelling the conditional
covariance of the standardized returns. We can consider GARCH(1,1) type specifications of the form
�z z � � �� � �q �
q � � ��
� �
ij,
t�
1 ij i,
t j,
t ij ij,
t ij .
In estimating the dynamic conditional correlation models suggested above, we can rely on the quasi maximum
likelihood estimation (QMLE) method.
The next step in our analysis consisted in the calibration of a VAR model on the volatilities of the highfrequency
returns, which were proxied by the squared returns. The specification of the VAR model on volatilities
was constructed to allow for 1 lag dependence among the 14 variables taken into account. The results are provided
in the next section as well.
The last step in our analysis consisted in the calibration of a jump process in the specification of Maheu and
McCurdy (2007). The specification is presented in the following lines.
rt – returns for the period t = 1,…,T
μ – the mean of the returns when there is no rare event (no jump)
Jt – indicates the occurrence of the jump (Jt = 1 means that we will observe a jump at moment t; if Jt = 0 there is
no jump at moment t)
� t – the probability to have a jump at moment t i.e. Pr(Jt = 1)
� t – the size of a possible jump at moment t
� J – the mean for the jump-size variable
2
� J ,t – the variance of the jump-size variable
2
Xt-1 – the absolute value of rt-1 which allows the variance � to be positive.
The estimation of the model was realized through the Markov Chain Monte Carlo using a Gibbs algorithm. In
the analysis realized by Maheu and McCurdy (2007) the vector of parameters θ = { � ,
J ,t
2
� , J
where � ={ 0 �
,
� 1 � �
} and ={ 0 ,
� 1 } is augmented with the unobserved state vectors � ={
� 1,
…, T
times J = {J1, …, JT} and the jump sizes � = { 1 �
, …,
� T }.
� , � , � },
� }, jump
We compute the conditional densities for all the parameters and run 5000 simulations with draws from these
densities by the following algorithm:
1. sample � | θ-μ,i-1 � � i-1, Ji-1, i-1, r
2.
2 2
sample � | θ-�
i-1, � �
i-1, Ji-1, i-1, r
516

�
3. sample J �
| θ- J i-1 ,� � i-1, Ji-1, i-1, r
4. sample � | θ- � i-1,� i-1, Ji-1, � i-1, r
( , � )
5. sample blocks
� t �( t, � ) |θ- i-1,� � i-1, Ji-1, i-1, r
6. sample � |θ- � i-1,� i, Ji-1, � i-1, r
7. sample � | θ i, � i, J i-1, r
8. sample J | θ i, � i, � i, r
9. goto 1
where r = {r1, …, rT} and i is the iteration number.
4 Results
In terms of co-movements, the most interesting issues observed in our analysis deal with the fact that, in case we
take into account only the values that are outside a very conventional 95% confidence interval (two sigmas away
from the mean in both directions) we notice that for this high frequency data these events are happening usually
simultaneously for all the indexes in our database. In case we consider these to be jumps, then our dynamics show a
close co-movement of the stock markets in Europe.
On average, we found that, out of the 14 indexes that we took into account, for a 5-minute frequency, we have
about 10.15 of them happening in the same time. We consider this to be a proof of co-movement, showing the fact
that usually, when unusual returns are realized in the stock markets, they have the tendency to be realized in the
mean time for all the stock markets, showing that the information is moving in a fast manner around Europe and also
showing that the large movements are probably the cause of important information affecting the whole environment,
since many of them are reacting in a five-minute interval.
In order to produce a better characterization of the structure of the outliers (jumps) and, in the same time, reveal
the properties of the co-movements, we organized the data to compute the number of situations in which we had
simultaneous outliers out of the whole sample of indexes taken into account.
1 2 3 4 5 6 7 8 9 10 11 12 13 14
52.43
%
6.60% 4.86% 4.17% 4.17% 4.17% 1.39% 4.86% 3.82% 3.13% 2.78% 1.04% 5.90% 0.69%
-8.15E- -1.02E- 1.80E- -1.39E- 1.47E- 2.68E- -2.37E- -2.50E- 1.23E- 3.61E- 7.20E- -1.08E- 4.45E- 2.79E-
05 03 03 03 03 03 03 03 03 03 03 02 04 03
Table1: The percentage of simultaneous outliers out of all the situations when we experienced outliers in the sample
of our stock market index returns
As we can notice from the above table, out of a number of 288 moments when outliers were recorded (from a
sample of 2413 records at the five-minute frequency), a high number (52.43%) happened in isolated situations, i.e.
only one stock index experienced the outlier, but we notice that almost half of the outliers were experienced in at
least 2 situations, with a lot of them happening in 8 and mostly 13 stock markets in the same time. Important
evidence is also the fact that, when taking into account the mean values of the outliers, the ones that are isolated
(only one index exhibiting a jump) are the smallest, while for the others we see higher absolute values. This can be
considered to be a proof of the fact that usually the big jumps tend to spread, while the ones that are not so big tend
to be local and, on the other hand, this difference in size also might explain the bigger number of isolated jumps,
since a smaller size means that they are closer to the mean and, hence, they tend to have a higher frequency (higher
rate of realization).
517

1 2 3 4 5 6 7 8 9 10 11 12 13 14
Germany 3 11 21 8 25 58 75 71 91 89 88 100 100 100
France 2 11 14 8 25 50 75 71 91 100 100 100 100 100
UK 2 0 21 33 50 58 50 57 91 89 75 100 100 100
Spain 2 11 14 25 25 25 75 71 73 67 75 100 100 100
Switzerland 3 32 14 33 33 50 75 64 73 89 88 100 88 100
Italy 1 21 21 33 42 8 75 64 64 78 88 100 100 100
Portugal 5 11 21 17 17 42 50 50 64 67 100 100 100 100
Ireland 1 16 43 50 25 33 50 50 4 67 50 67 94 100
Austria 2 5 7 33 67 50 50 57 55 89 88 67 100 100
Poland 7 0 36 42 42 58 50 50 64 56 75 67 100 100
Czech
Republic 3 21 7 17 50 50 75 50 73 89 88 100 100 100
Hungary 3 21 29 50 42 42 0 43 45 56 100 100 100 100
Romania 12 11 36 25 17 33 0 71 27 44 75 100 88 100
Slovenia 54 32 14 25 42 42 0 29 27 22 13 0 29 100
Table 2:The frequency of outliers for each stock market index (in percentage)
The previous table shows the proportion of outliers in which each of indexes was involved out of the number of
all the outliers identified in the sample. Hence, we can say that, out of all the situations in which we had only one
outlier in the whole sample, which was not accompanied in the same moment by another outlier, there were only 3
of these situations in which Germany was involved (in only 3 of the cases Germany had an isolated outlier).
Accordingly, out of all the situations in which we had 7 simultaneous outliers, we notice that Poland had an outlier
in 50 of the cases.
The countries that tend to have isolated outliers were those countries for which we see big numbers in the left
side of the table and relatively small numbers in the right side of the table. Slovenia is the only one with such a
situation. However, the vast majority of the countries tend to show a higher proportion of outliers that were
happening in the same time. This is important evidence on stable co-movement of the European stock markets,
showing that relevant information has the power to provide important shocks at a regional level.
The next step was to compute dynamic conditional correlations using the GARCH specification mentioned in
the previous section. The means of the correlations for each pair are produced in the following table.
Germany
France 0.86
Germany France UK Spain Switzerland Italy Portugal
UK 0.78 0.79
Spain 0.70 0.77 0.68
Switzerland 0.74 0.72 0.70 0.64
Italy 0.75 0.81 0.71 0.78 0.66
Portugal 0.50 0.54 0.51 0.57 0.49 0.54
Ireland 0.34 0.34 0.33 0.29 0.29 0.31 0.27
Austria 0.44 0.42 0.43 0.40 0.41 0.43 0.38
Poland 0.50 0.45 0.45 0.40 0.40 0.41 0.35
Czech 0.41 0.40 0.39 0.35 0.34 0.37 0.34
518

Germany
France
UK
Spain
Republic
Switzerland
Italy
Portugal
Ireland
Hungary 0.31 0.29 0.31 0.24 0.27 0.25 0.26
Romania 0.10 0.00 0.10 -0.70 0.07 0.08 0.04
Slovenia 0.00 0.00 0.01 0.01 0.03 -0.01 -0.02
Austria 0.26
Ireland Austria Poland
Poland 0.18 0.33
Czech
Republic 0.22 0.48 0.33
Hungary 0.18 0.24 0.28 0.27
Romania 0.04 0.14 0.05 0.07 0.08
Czech
Republic Hungary Romania Slovenia
Slovenia -0.02 0.02 0.02 0.02 -0.03 -0.01
Table 3:The mean values of the correlations computed with the DCC model for five-minute returns
We can notice that the higher values of the correlations are present in the upper part of the table, which means
that the index returns tend to be correlated mostly among the developed countries and less correlated with the
Eastern European ones. The latter also show that they are not so much correlated neither among themselves, which
is evidence of their particular independence on the Western European stock markets.
The next analysis in our work dealt with the construction of a VAR for the volatilities of the stock index returns.
Due to lack of space, the final results of the estimation are not provided here but we mention that we estimated a
VAR up to the second lag for all the series in our database. Only very few coefficients proved to be significant:
DAX, CAC, UKX, SMI and FTSEMIB seem to be dependent on the first lag of WIG, PSI20 is dependent on the
second lag of DAX and the second lag of WIG, ATX is dependent on the first lag of SMI, WIG is dependent on the
first lag of DAX and BUX is dependent on the second lag of FTSEMIB and its first lag.
The last part of our analysis consisted in the calibration of the Maheu and McCurdy (2007). Our aim was to find
the coefficients for each of the returns in our sample, and then to analyze the parameters of the jump densities by
comparing them among the returns in our database. Thus, the jump densities were introduced into another VAR
analysis and we tried to see if the densities are providing any kind of dependence. We are not providing here the
results, but we mention that the estimation did not provide any significant parameters.
5 Conclusions
This paper uses high frequency stock market index returns to check for the co-movements in the European region.
Using 14 indexes for a period of about one year, the common dynamics of the five-minute returns were analyzed
using many tools: the properties of the distributions of each series, the jumps defined as outliers of the 95%
519

confidence interval and their simultaneity, the dynamics of the correlations, the relationships among volatilities and
then the relationships among the densities for a jump-diffusion model.
Our study revealed the fact that there is significant dependence of the returns on the movement of the other
returns in the sample, especially when we took into account the extreme values. These “jumps” tend to be
simultaneous at the regional level and they are evidence of the fact that the information is spread fast at the
European capital markets.
6 Acknowledgements
This research is part of the CNCSIS Young Research Teams type research project with the title “Bayesian
Estimation of the Structural Macroeconomic and Financial Relationships in Romania: Implications for Asset
Pricing”, contract number 25/5.08.2010, director Dr. PetreCaraiani, Institute for Economic Forecasting, Romanian
Academy.
7 References
Andersen, T. G., Benzoni, L. & Lund, J. (2002).An empirical investigation ofcontinuous-time equity return
models.Journal of Finance, no. 62, pp. 1239–1284.
Barry, C. B., Peavy III, J. W. & Rodriguez, M. (1998).Performance Characteristics of Emerging Capital
Markets.Financial Analysts Journal, vol. 54, no. 1, pp. 72-80.
Beine, M. &Candelon, B. (2011).Liberalisation and stock market co-movement betweenemerging economies.
Quantitative Finance. Vol. 11, no. 2, pp. 299-312.
Bekaert, G., Erb, C. B., Harvey, C. R. &Viskanta, T.E. (1998). Distributional Characteristics of Emerging Market
Returns and Allocation.Journal of Portfolio Management, vol. 24, no. 2, pp. 102-116.
Brooks, R., Del Negro, M. (2004). The rise in comovement across national stock markets: market integration or IT
bubble?.Journal of Empirical Finance, no. 11, pp.659–680.
Brooks, R., Del Negro, M. (2005). Country versus region effects in international stock returns. Journal of Portfolio
Management, pp. 67–72,Summer, 2005.
Brooks, R., Del Negro, M. (2006).Firm-level evidence on international stock market comovement.Review of
Finance, no. 10, pp. 69–98.
Candelon, B., Piplack, J., Straetmans, S. (2008). On measuring synchronization of bulls and bears: the case of East
Asia.Journal of Banking and Finance, no. 32, pp. 1022–1035.
Chernov, M., Gallant, R. A., Ghysels, E. &Tauchen, G. (2003).Alternative models forstock price dynamics. Journal
of Econometrics, no. 116, pp. 225–257.
Chib, S., Nardari, F. &Shephard, N. (2002).Markov chain Monte Carlo methods for stochastic volatility models,
Journal of Econometrics, no. 108, pp. 281–316.
Didier, T., Love, I., &Peria, M. S. M. (2010).What Explains Stock Markets’ Vulnerabilityto the 2007–2008
Crisis?.The World Bank Policy Research Working Paper 5224
Divecha, A.B., Drach, J. &Stefek, D.(1992).Emerging Markets: A Quantitative Perspective.Journal of Portfolio
Management, 19, 1, pp. 41-50.
Eraker, B., Johannes, M. S. & Polson, N. G. (2003).The impact of jumps in volatilityand returns, Journal of
Finance, vol. 58, no. 3, pp. 1269–1300.
Forbes, K., Rigobon, R.(2002). No contagion, only interdependence: measuring stock market comovements. Journal
of Finance, no. 57, pp. 2223–2261.
Grubel, H.(1968). Internationally diversified portfolios: welfare gains and capital flows.American Economic Review,
vol. 58, no. 5, pp. 01299–1314.
520

Harrison B., Lupu, R. &Lupu, I. (2010), Statistical Properties of the CEE Stock Market Dynamics.A Panel Data
Analysis.The Romanian Economic Journal, Year XIII, no. 37, pp. 41-54.
Harrison, B. & Moore, W. (2009).Stock Market Comovement in the European Union andTransition
Countries.Financial Studies, vol. 13, no. 3, pp. 124-151.
Harvey, C.R (1995). Predictable Risk and Returns in Emerging Markets.Review of Financial Studies, vol. 8, no. 3,
pp. 773-816.
Horobet, A. &Lupu, R. (2009). Are Capital Markets Integrated? A Test of Information Transmission within the
European Union.Romanian Journal of Forecasting, no.2, Vol. 10.
Jorion, P. (1988). On jump processes in the foreign exchange and stock markets.Review of Financial Studies, vol. 1,
no. 4, pp. 427–445.
Karolyi, G.A., Stulz, R.M. (1996). Why do markets move together? An investigation of U.S.–Japan stock return
comovements.Journal of Finance, vol. 51, no. 3, pp. 951–986.
King, M., Sentana, E., Sushil, W. (1994). Volatility and links between national stock markets. Econometrica, vol.
62, no. 4, pp. 901–933.
King, M., Wadhwani, S. (1990).Transmission of volatility between stock markets.Review of Financial Studies, vol.
3, no. 1, pp. 5–33.
Kizys, R., Pierdzioch, C.(2009). Changes in the international comovement of stock returns and asymmetric
macroeconomic shocks. Journal of International Financial Markets, Institutions and Money, vol. 19, no. 2, pp.
289–305.
Lin, W.-L., Engle, R. F. & Ito, T. (1994).Do Bulls and Bears Move Across Borders? International Transmission of
Stock Returns andVolatility, The Review of Financial Studies, vol. 7, no. 3, pp. 507-538.
Lin,W.-L., Engle, R., Ito, T. (1994). Do bulls and bears move across borders? International transmission of stock
returns and volatility.Review of Financial Studies, vol. 7, no. 3, pp. 507–538.
Longin, F., Solnik, B. (1995). Is the correlation in international equity returns constant: 1960–1990?.Journal of
International Money and Finance, vo. 14.no. 1, pp. 3–26.
Longin, F., Solnik, B. (2001).Extreme correlation of international equity markets.Journal of Finance,vol. 56, no. 2,
pp. 649–676.
Lupu, R. &Lupu, I. (2009). Contagion across Central and Eastern European Stock Markets: A Dynamic Conditional
Correlation Test.Economic Computation and Economic Cybernetics Studies and Research, Vol. 43, No. 4, pp.
173-186.
Maheu, J. M. & McCurdy, T. H. (2004). News arrival, jump dynamics, and volatilitycomponents for individual
stock returns.Journal of Finance, vol. 59, no. 2.
Maheu, J. M. & McCurdy, T. H. (2007).Modeling foreign exchange rates with jumps.Working Papers tecipa-279,
University of Toronto, Department of Economics.
Morana, C. &Beltratti, A. (2008).Comovements in International Stock Markets.Journal of International Financial
Markets, Institutions and Money, vol. 18, no. 1, pp. 31-4.
Pajuste, A. (2002). Corporate Governance and Stock Market Performance in Central and Eastern Europe: A Study of
Nine Countries’,Stockholm School of Economics. Working paper, Available at SSRN:
http://ssrn.com/abstract=310419 or doi:10.2139/ssrn.310419
Rua, A. &Nunes, l. C. (2009). International comovement of stock market returns: A wavelet analysis. Journal of
Empirical Finance, no. 16, pp. 632–639.
Soydemir, G. (2000). International Transmission Mechanismof Stock Market Movements: Evidencefrom Emerging
Equity Markets. Journal of Forecasting, no. 19, pp. 149-176.
521

OWNERSHIP STRUCTURE, CASH CONSTRAINTS AND INVESTMENT BEHAVIOUR IN RUSSIAN
FIRMS 1
Tullio Buccellato (a) , Gian Fazio (b) and Yulia Rodionova (c)
February 2011
30 April 2010
Abstract. In this paper, we investigate to what extent Russian firms are liquidity constrained in their
investment behaviour and how ownership structure changes the relationship between internal funds and
investment decisions of these firms. We estimate a structural financial accelerator model of investment and
first test the hypothesis that Russian firms are cash constrained. We conduct random effects estimation on a
large representative panel dataset of 8637 firms in the European part of Russia, using their balance sheet
information over the period 2000-2004. Our results confirm that firms are liquidity constrained when the
ownership structure is omitted from the econometric specifications. With regards to the ownership structure
and degree of concentration, we find that state owned companies and private individuals/families are less
cash constrained, independently of whether their ownership structure is concentrated. No significant impact
is found for banks and financial companies and institutions.
JEL codes: F1, F2, G24, G32, 03.
Keywords: entrepreneurial firm, emerging market, financing choices, sales growth, nnovation, manager's
characteristics.
(a) Dipartimento di Economia e Istituzioni, Univesita’ di Roma Tor Vergata.
(b) School of Slavonic and East European Studies, University College London.
(c) Department of Accounting and Finance, Leicester Business School, De Montfort University
1 We are grateful to the participants of the 10th EACES Conference in Moscow, CICM conference at London
Metropolitan University, research seminar at Leicester Business School at DMU, and to Panagiotis Andrikopoulos,
Ashley Carreras, Tomila Lankina, Fred Mear,and to Pasquale Scaramozzino for helpful comments and suggestions
522

MORTGAGE HOUSING CREDITING IN RUSSIA: CRISIS OVERCOMING AND FURTHER
DEVELOPMENT PERSPECTIVES
Liudmila Guzikova, Saint Petersburg State Polytechnic University
E-mail: guzikova@mail.ru
Abstract. Russian system of mortgage housing crediting (SMHC) has been created on the basis of the American two-level model
uniting the mortgage loan market and the mortgage based securities market. In the period till 2008 Russian SMHC developed actively.
In the course of its development the important role was played by factors, characteristic for emerging economy. Especially should be
noticed that social expectations connected with SMHC were initially overestimated. Financial and economic crisis has drawn attention
to national SMHC and to revealing the ways and perspectives of its further development. Overcoming of the crisis consequences and
further SMHC functioning demand to work out the concept of SMHC operated complex balanced development. The basis of this
concept is the system of development principles realized in the offered balanced dynamic model, coordinating social expectations,
solvent demand, real estate supply, investment potential of financial institutions and population and capacity of mortgage securities
market, and in the uniform system of SMHC performance indicators.
Keywords: Mortgage, Mortgage Market
JEL classification: G210
1 Introduction
The national system of mortgage housing crediting (SMHC) in Russia has been created on the basis of the American
two-level model uniting mortgage loan market and mortgage securities market. By this time the circle of its
participants had been determined, the standard-legal base regulating their mutual relations created, the market
infrastructure generated.
SMHC realizes only one of the ways of housing maintenance possible in the market economy. However, at the
governmental level SMHC is considered to be the main instrument of dealing with the large scale and extremely
sharp housing problem. Such point of view is reflected in the government programs providing grants to separate
categories of the population for the purpose of their involvement in SMHC.
Financial and economic crisis has aggravated problems and disproportions of the national SMHC and has drawn
attention to its condition and further development perspectives.
2 Drawing lessons from the crisis and substantiation of SMHC postcrisis development model
2.1 Pre-crisis development and mechanisms of Russian SMHC crisis
Since 2005 volumes of mortgage housing crediting have grown high rates (figure 1), however, the volume of debt
under mortgage housing credits did not exceed 2.6 % with respect to GNP (for comparison: in pre-crisis 2006 in the
USA the volume of mortgage housing credits made up 76 % with respect to GNP, in Denmark – 101 %, in Great
Britain – 83 %).
Figure 1. Debt volume on housing mortgage loans in dynamics, million rbl.
523

Since 2007 the financial and economic crisis which has arisen in the market of mortgage lending has started in
the USA. Due to globalization of economy and the financial markets virtually all economically developed and
developing countries, including Russia, were involved in the crisis.
During the pre-crisis period processes and tendencies similar to those on American mortgage market took place
in Russia:
� agiotage rise of real estate prices;
� aspiration to use housing as an investment asset (Minz, 2007);
� liberalization of banks requirements to borrower quality, caused by their aspiration to strengthen position in the
new market (Deljagin, 2008).
However, the national SMHC hasn't played an appreciable role in the development of non-stationary processes
in the Russian economy. Due to small scale and backwardness of its separate elements, in particular, the mortgage
securities market, the SMHC became rather a victim of the crisis, than its active conductor. We notice that the
maximum annual volume of the mortgage loans was reached in 2008 and enabled to buy no more than 10.5 % of the
housing constructed the same year at the current prices of that period and taking into account an initial payment of
30 % of housing cost. The fall of the Russian mortgage market took place in 2009.
2.2 Features of Russian SMCH development
The major factors determining the character of the national SMCH development are the low incomes of the most
population and high regional and social differentiation of the population by incomes.Now for a family consisting of
two persons with their incomes equal to the bottom of the most highly remunerative cluster (monthly income of
25,000 rbl. per person), indicated by the state statistics, only the limited set of credit conditions is accessible: interest
rates not above 12 % for long crediting terms over 15 years. The volume of this cluster makes up no more than 10 %
of the whole population.
Drawing on the statistical data for 81 regions of Russia on 10/1/2010 the estimation of correlations between the
characteristics reflecting SMHC conditions (table 1) was conducted.
Parameters analyzed Correlation factor
Per capita housing and housing construction volume 0.2654
Per capita housing and income 0.1878
Per capita housing construction volume and income 0.1482
Housing price in the secondary market and per capita income 0.7880
Housing price in the primary market and per capita income 0.7436
Volume of housing construction and population 0.3108
Population and total volume of mortgage loan debt 0.8763
Urban population and total volume of mortgage loan debt 0.9414
Agricultural population and total volume of mortgage loan debt 0.1337
Per capita housing and volume of mortgage loan debt 0.0042
Volume of housing construction and total volume of mortgage loan debt 0.2321
Per capita volume of mortgage loan debt and income 0.5480
Table 1: Correlation between regional characteristics of the population, housing construction and mortgage debt
From the analysis results it is possible to draw the following conclusions:
� the building branch as a whole is not focused on the housing problem solution, which is shown in low
correlations of housing construction volume with the level of housing for population (social aspect) and with the
existing income level (commercial aspect);
524

� the significance of incomes of the population in providing housing is rather small, the existing level of housing
was not reached by market methods;
� price level in the housing market significantly, but not absolutely, positively correlates with the income level;
� the volume of the mortgage loans significantly positively correlates with the population quantity, with the
overwhelming majority of mortgage borrowers being city dwellers;
� The volume of the mortgage loans is not connected with housing level and has low positive correlation with the
income level.
Correlation analysis between the indicators of incomes and housing level, the prices of the housing market and
volumes of the general and past-due debts on mortgage housing credit (table 2) was carried out drawing on the data
concerning 15 regions, which were leaders on the volumes of the past-due debts on the mortgage loans given out in
roubles and foreign currency. The results of this analysis can be summarized as follows:
1. The level of housing expensiveness, presented as the ratio of square meter cost to per capita income, has
low negative correlation with the volume of the past-due debt on mortgage loans in roubles and is virtually
not connected with the volume of the past-due debt in currency;
2. Per capita general mortgage debt on loans given out in roubles has low negative correlation with the per
capita volume of past-due mortgage loan debt in roubles;
3. Per capita general mortgage debt on loans given out in foreign currency has strong positive correlation with
per capita volume of past-due mortgage loan debt in currency.
Correlation factor
Parameters analyzed
for credits in for credits in
roubles currency
Level of housing expensiveness and volume of past-due mortgage debt -0.3213 0.0037
Per capita mortgage debt and past-due mortgage debt -0.1635 0.9428
Per city dweller mortgage debt and past-due mortgage debt -0.0937 0.9388
Table 2: Correlation between characteristics of population, housing market and mortgage loan market of in the regions having the highest level
of past-due mortgage debt
Thus, it is possible to ascertain the presence of disproportions between housing construction volumes, level of
incomes, housing and mortgage crediting features. The working out and consecutive realization of the SMHC
operated complex balanced development concept are demanded for the elimination of the revealed disproportions.
2.3 Current SMHC conditions
Post-crisis national SMHC restoration occurs fairly quickly. In 2010 the Russian banks provided mortgage loans for
the sum of 378.9 billion rbl., that is 2.5 times more than the volume of the mortgage loans given out in 2009 (152.5
billion rbl.) but 42 % less, than in 2008 (655.8 billion rbl.). In total, the borrowers received about 298 thousand
mortgage loans in 2010 that is only 10 % less in comparison with the pre-crisis indicators. The emergence of the
new SMHC participants on the market became a sign of overcoming the crisis: in 2010 there were 628 banks in the
market of mortgage credits, whereas in 2009 there were 584, while in 2008 there were 602 operating banks. The
share of the past-due debt in total amount of mortgage housing credits was 3.61 % in 2010 confirming rather
rigorous bank requirements to mortgage borrowers and, partly, the effectiveness of measures on re-structuring of the
past-due mortgage debt.
The basic tendencies of post-crisis SMHC development are the following:
1. increase of the banks number giving out mortgage housing credits. There were 523 SMHC participants
giving out credits by the end of 2010, with 105 participants serving the credits given out earlier.
2. further growth of the absolute and relative indicators characterizing the volumes of mortgage housing
credits given out. In 2010, the banks gave out 301433 mortgage housing credits for a total sum of 380.1
billion rbl., making up 10.4 % of the total consumer credit amount. The average credit size has slightly
increased and made up 1.25 million rbl. The quantity of the mortgage loans given out has increased 2.3
times in comparison with the previous year;
525

3. growth of the mortgage housing credit share given out in national currency (roubles). In 2010 the number
of rouble credits equaled 298213 for the total sum of 354.6 billion rbl., vs 3220 credits given out in foreign
currency. In comparison with 2009, the share of rouble credits increased by 2.2 % and made up 95.9 %;
4. further decrease of credit interest rates and liberalization of credit granting conditions. The average interest
rate on rouble credits decreased by 1.2 %, on currency credits by 1.7 % and made up 13.1 % and 11 %
accordingly on 1/1/2011. The average terms of the rouble credits given out were reduced by 1 month and
made up 196 months, whereas on credits in foreign currency the terms grew up by 15 months, constituting
155 months;
5. steady tendency of increase of the past-due debt volume and its share in total mortgage debt amount. In
comparison with 2009, the past-due debt in rouble credits increased by 27.2 %, and those in foreign
currency credits – by 44.5 %. The share of the past-due debt in total amount made up 2.5 % and 3.7 %
accordingly. Past-due debt with the delay over 180 days constitutes the greatest share in the total volume of
past-due debt.
In 2010 the total volume of refinancing of mortgage loans made up 64.6 billion rbl. that corresponds 17 % of
the total volume of the mortgage loans given out that year, but the further issue of mortgage securities was not
supposed. For comparison we notice that in 2009 the refinancing volume made up 65.4 billion rbl. and was equal to
43 % of total volume.
Crediting condition liberalization at the time when real incomes decrease reflects the banks aspiration to
increase mortgage lending volume and represents the dangerous tendency, capable to accelerate past-due debt
growth. The government has developed past-due debt re-structuring programs, which are realized by Mortgage
Housing Credits Re-structuring Agency (MHCRA), and continues granting some groups of population for
participation in SMHC. It shows that the lessons of the crisis are not adopted properly.
2.4 Principles of the SMHC development
For providing steady SMHC functioning it is necessary to work out principles constituting the basis of the SMHC
operated complex balanced development concept. These principles should be adopted by all SMHC participants.
Principles of the national SMHC development are united in four groups:
1. development management principles;
2. balanced development principles;
3. development dynamics principles;
4. complex development principles.
2.4.1 Development management principles
Mortgage market crisis development and steps made for its overcoming in the USA, Western European countries
and Russia confirmed the necessity of the SMHC governmental management. The government should carry out the
following functions:
1. ideological function. The government should realize this function working out the general concept of the
national SMHC development. There is the project called «Long-term strategy of mortgage lending
development in the Russian Federation» in Russia. The objectives of the specified strategy are the
following:
� carrying out the common position of all mortgage market participants on long-term perspectives, benchmarks,
principles of the mortgage market and other forms of housing financing development;
� determination of the main long-term objectives and government policy tasks on mortgage market and other
forms of housing financing development for the period till 2030;
� determination of the basic measures and actions for tasks decision, including priority strategic actions, in view
on strategic objectives achievement;
2. legislative function. The national SMHC creation became possible in Russia after adoption of the Federal
laws № 2972-1 from 5/29/1992 «Concerning collateral» and №102 from 7/16/1998 «Concerning mortgage
(real estate collateral)». In the framework of the legislative function the government should adopt acts
providing strategic SMHC development perspective.
526

3. standard-regulating function. In the framework of this function the government carries out tactical and
operational SMHC administration and provides flexible response to circumstances and operating changes
generated by external and internal factors;
4. plan-indicative function. This function should be realized by formation of the indicator system reflecting
the SMHC development, and by working out forecasts and indicative plans;
5. control-analytical function. The government should realize this function to reveal the SMHC development
negative tendencies timely, to analyze the reasons of deviations from indicative plans and to carry out
necessary correcting measures. Central bank of the Russian Federation should also provide this function for
the banks participating in the SMHC;
6. information-analytical function. Market mechanism functioning is most effective under the information
transparency and publicity. Within the SMHC framework information-analytical function can be carried
out by the independent analytical companies and the self-adjustable organizations;
7. protection of the rights and legitimate interests of the SMHC participants. This function can be realized by
working out and adoption the legislative and regulatory acts concerning to the SMHC participants conflicts
and by their executive mechanism creation. The SMHC represents an open system as its participants are
free in decision-making to participate its functioning or not. Entrance and exit barriers establishment is one
of the ways of the SMHC reliability maintenance. Corresponding requirements working out and their
observance control can be assigned to the self-adjustable organizations within the SMHC (for example,
Association of the Russian banks – ARB, National association of the mortgage market participants –
NAMMP).
In the SMHC development management the following principles should be realized:
� functional importance of the government as the SMHC management subject;
� objective management;
� combination of strategic, tactical and operational administration;
� scientific validity of all level administrative decisions;
� competence of administrative decisions;
� management flexibility and alternatives consideration;
� centralized management and self-regulation combination;
� portfolio management approach.
2.4.2 Balanced development principles
The government should provide the balanced social and economic development. The SMHC development objectives
should fit the general social and economic development objectives, and the aims of the separate SMHC participants
categories should be coordinated. The priority of the social purposes and corresponding results is characteristic for
the government at federal and regional levels. Commercial objectives are priority for individual SMHC participants,
for example, creditor banks and insurers.
The government should support balance of the SMHC and other social and economic national projects, of the
SMHC and other housing providing systems (Polterovich, 2007). The SMHC should also act as the gear balancing
supply and demand in the housing real estate market (the Mortgage as a financing source, 2007).
The national SMHC is the developing system, which absolute and structural characteristics change under
impact of external and internal, operated and uncontrollable factors. It should be flexible and variable enough to
respond new requirements and possibilities, to develop new positive tendencies, to answer to inquiries of its
participants adequately. At the same time it should be stable enough to exclude faults of its functioning, to provide
performance predictability and to retain the trust of its actual and potential participants. The balance between
stability and variability of the SMHC structure and parameters and its separate subsystems and elements appreciably
depends on the rate and scale of innovations within the SMHC framework.
The balanced development assumes observance of certain structural proportions between the SMHC and its
subsystems characteristics. The balanced development principle is diverse in its displays and ways of realization and
includes the followings:
� balance of the SMHC strategic, tactical and operative regulation;
527

� balance of differential and integrated management approaches;
� balance of general social and economic development objectives and the objectives of the SMHC development;
� balance of the SMHC and its environment;
� balance of social and commercial objectives within the SMHC framework;
� balance between the SMHC levels;
� internal balance between the SMHC elements distinguished on the basis of various classification criteria;
� balance of stability and variability;
� balance between a competition and cooperation;
� balance of risk and utility (profitability);
� resource balance between the SMHC subsystems and elements;
� balance of publicity and confidentiality;
� balance of personified (individual) and standardized approaches.
2.4.3 Development dynamics principles
Dynamics of the SMHC development should submit to the following principles:
� dynamic stability, i.e. observance of listed above balance parities in dynamics;
� non-stationarity factors monitoring;
� non-stationarity factors avoiding and elimination;
� minimization of non-stationary processes caused by external factors negative consequences.
2.4.4 Complex development principles
The SMHC complex development should be provided with observance of the following principles:
� application of mentioned above management, balanced development and development dynamics principles to
the SMHC in whole and all its subsystems development;
� plurality of subsystems determination criteria;
� application of management, balanced development and development dynamics principles at all stages of
management cycle, including account, analysis, forecasting and planning, plan execution and control;
� openness;
� sharing experiences and mutual training.
2.5 SMHC development model
The SMHC development model can be constructed in general on the base of relations caused by the SMHC
designation and its elements functionality. Small life period of the national SMHC in combination with nonstationarity
of its development during this time doesn't allow econometric methods application to determine the
SMHC development model parameters. Model parameters are to be specified by computing experiments.
Mortgage loans market increment is defined by its formerly reached volume, accessible financial resources and
housing construction volume. Volume of financial resources in turn depends on population incomes and refinancing
level provided by mortgage securities issue. We notice that foreign loans constitute the significant part of long-term
credit resources of the Russian banks.
Mortgage securities market increment depends on formerly reached refinancing level and on mortgage
securities demand which in turn depends on population incomes. Excessive mortgage securities demand can
stimulate crediting standards liberalization and/or mortgage securities interest rate decrease as a result enlarging
mortgage securities risk.
Housing construction activity of building branch is financed both by loans raised from SMHC and by another
ways of housing acquisition expense. Alternative ways of housing acquisition expense are acquisition using budget
facilities, acquisition using own facilities entirely, acquisition using loans not raised through the SMHC and amount
of initial payments. It is obvious that the listed channels capacity is connected with population incomes.
528

Population income change is, in our opinion, the significative characteristic of the SMHC development. We
consider that feedback chain «crediting volume growth – construction volume growth – allied industries growth–
population incomes growth» has duration comparable to mortgage lending terms. However, it is also necessary to
consider incomes formed out of the SMHC.
The dynamic SMHC development model can be presented in the form of linear non-uniform differential
equations system with constant factors as follows:
dV
dt
� a11V
� a12
I � a13
R � a
dR
dt
� a 21V
� a 22 I � a 23 R
dG
dt
� a31V
� a32
I
dI
dt
� a 41V
� a 42 I � F ( t )
14
G � Z ( t )
where V –given out mortgage housing credits volume,
R – mortgage based securities issue volume,
G –housing construction volume,
I –population incomes level,
the disturbance terms:
Z (t) –long-term credit resources volume,
F (t) – independent of the SMHC income growth component,
аij – factors reflecting current SMHC characteristics impact on its development.
The disturbance terms and factors are the operated parameters which may be changed by standard restriction
establishment or indirect governmental impact.
It is supposed to analyze and test the SMHC development dynamic model presented in the form of linear nonuniform
differential equations system by the following directions:
1. definition of the system development trajectory passing the point corresponding to the current SMHC state
at preset factor values and disturbance functions, i.e. Cauchy problem solving. It will allow to make the
strategic forecast of SMHC development and to reveal necessity of operating means application;
2. search of the particular solution moving the system from the current (initial) condition to the new
(demanded) condition within the limited time at preset factor values and disturbance functions, i.e.
boundary problem solving;
3. definition of unknown functions and parameters of the system taking into account additional conditions
imposed on its parameters, i.e. characteristic value problem solving.
2.6 Uniform SMHC performance indicator system
Practical observance of formulated above principles and the SMHC participant objectives achievement, in our
opinion, characterize the degree of the SMHC development and its performance.
The uniform SMHC performance indicator system should cover the SMHC on a vertical, allowing to estimate
performance at federal, regional and individual levels, and across, allowing to estimate performance of separate
participant categories. It should provide spatiotemporal comparability of indicators and their balance at every level
(table 3).
529

Calculation and
Indicator group
analysis level
1. Potential indicators (mortgage lending availability) Federal
Regional
2. Indicators of macroeconomic and regional economic performance Federal
Regional
3. Social efficiency indicators Federal
Regional
4. Mortgage portfolio indicators
4.1. Mortgage portfolio structural indicators Federal
Regional
Individual
4.2. Mortgage portfolio profitability indicators Federal
Regional
Individual
4.3. Refinancing indicators Federal
Regional
Individual
4.4. Collateral Indicators Federal
Regional
Individual
Table 3: Structure of the SMHC performance uniform indicator system
The first three groups of indicators characterize the SMHC performance from the borrower point of view and
display government objectives achievement, the fourth group of indicators characterize bank objectives achievement
- demanded quality of mortgage portfolio (Kozlovskaya et al., 2008). Mortgage portfolio quality is determined by
combination of profitability and risk determining in turn portfolio refinancing possibility (Oseledets, 2006). The risk
level is concerned not only as borrower solvency loss probability but also as collateral insufficiency probability to
repay the loan and interest.
3 Summary
The financial crisis sharpened the question of the Russian national SMHC further development. The SMHC
problems and disproportions elimination and its moving to stable development trajectory demand to revise the point
of view concerning to which the SMHC is the main instrument for large scale and extremely sharp housing problem
overcoming.
The SMHC performance depends on the adequacy of the social expectations connected with it to its real
capacity. In our opinion, only those groups of population should participate the SMHC as borrowers, whose income
level and its stability in combination with available savings allow to make initial payment for the obtained housing
independently, to carry out credit obligations in due time and in full amount and to bear further burden of the
obtained housing maintenance.
Low incomes of great part of the population sustain the SMHC quantitative growth, but do not counteract to its
qualitative improvement on the base of the suggested operated complex balanced development concept using the
principles formulated in the paper. The government should regulate the SMHC functioning by market methods
providing conditions for its proportional and systematic development. The main and indispensable condition is
providing population incomes growth. The dynamic SMHC development model and the SMHC development
uniform indicator system suggested in the paper can be used as planning, forecasting and performance evaluation
tools. Thus the probability of the Russian SMHC crisis will be minimized.
530

4 References
Deliagin M.G. Ipotechniy krizis ne rassosiotsia (Делягин М.Г. Ипотечный кризис не рассосется. Банковское
дело, 2008, №2. – C.20-23)
Ipoteka kak istochnik finansirovaniya stroitel’noy deyatel’nosi (Ипотека как источник финансирования
строительной деятельности. Региональные аспекты инновационной инвестиционной деятельности,
Под ред. А.А. Румянцева. – СПб. ИРЭ РАНН, 2001)
Kozlovskaya E.A., Guzikova L.A., Savrukova E.N. Povishenie effektivnosti bankovskogo ipotechnogo
kreditovaniya (Козловская Э.А., Гузикова Л.А., Саврукова Е.Н. Повышение эффективности банковского
ипотечного кредитования в России. Научно-технические ведомости СПбГПУ. Том 1, Экономические
науки. Санкт-Петербург, СПбГПУ. 2008. №1.– С.175-185)
Mints V.O. O faktorakh dinamiki tsen na zhiluyu nedvizhimost’ (Минц В. О факторах динамики цен на жилую
недвижимость. Вопросы экономики. 2007. №2. – C. 111-121)
Oseledets V.M. Otsenka pokazateley effektivnosti ipotechnogo bankovskogo kreditovaniya (Оселедец В.М.
Оценка показателей эффективности ипотечного банковского кредитования. Сибирская финансовая
школа. 2006. № 4. – С. 46-49)
Polterovich V.M., Starkov O.Y. Formirovanie ipoteki v dogoniaiuschikh ekonomikakh: problema transplantatsii
institutov (Полтерович В.М., Старков О.Ю. Формирование ипотеки в догоняющих экономиках: проблема
трансплантации институтов. - М: Наука, 2007. – 196 с.)
531

FOREIGN INVESTORS’ INFLUENCE TOWARDS SMALL STOCK EXCHANGES BOOM AND BUST:
MACEDONIAN STOCK EXCHANGE CASE
Dimche Lazarevski PhD, University American College Skopje, FYRO Macedonia
Email: lazarevski@uacs.edu.mk, www.uacs.edu.mk
Abstract. This paper aims to answer the question if and how much the foreign investors influence the boom and bust of small stock
exchanges. It examines the impact of the foreign investors’ turnover towards small stock exchange turnover, particularly the
Macedonian Stock Exchange. Based on the Macedonian Stock Exchange data for the period of January 2006 to July 2009, I find
strong evidence that for a small and open stock exchange such as the Macedonian Stock Exchange, foreign investors substantially
contributes to the Stock Exchange boom and bust.
Keywords: Stock Exchange, Foreign Investors, Turnover, Linear Regression
JEL classification: C35, G01, G12, N24, O16
1 Introduction
Global Financial Crisis has big impact on capital markets, especially stock exchanges. For those young and
underdeveloped stock exchanges, such as the Macedonian Stock Exchange, which exists from 1995, the period from
the beginning of 2005 until the middle of 2009 has been the most remarkable period that will be remembered for
generations. The value of the basic Macedonian Stock Exchange index MBI10 in January 2005 was 1.000. At its
peak, at the end of August 2007 the value rose to 10.057 or 905% growth for only 2 years and 8 months, and now, in
April 2011 decreased to only 2.500, or 75% less than the highest value. This phenomenon evokes many questions,
but maybe the most important one is “if and to what extent foreign investors were the reason for the Macedonian
Stock Exchange boom and bust.
The impact of foreign investors towards the big Stock Exchange was immense. For example, their withdrawal
of US$35bn from the Russian stock market through early September 2007, forced the Russian Central Bank to
defend the ruble, sending reserves down 4.2%. Russian Stock Exchange Index for the following 12 months lost
almost 80% of its value. Foreign Investors withdrawals decreased more than 50% from Dow Jones Industrial
Average.
Underdevelopment of the Macedonian Stock Exchange did not challenge the scientific and professional public
to analyze the reasons for its fast boom and bust. Therefore, this study addresses these issue by: (1) analyzing
Macedonian Stock Exchange development since its beginnings in 1995, in order to determine the level of its
development and exposure to influences from abroad, (2) investigating Macedonian Stock Exchange index MBI 10
since its beginning at the end of 2004 in order to establish the reasons for the fast growth and reduction (3)
measuring Foreign Investors impact on Macedonian Stock Exchange using linear regression on the percentage
change in Foreign Investors and Macedonian Stock Exchange turnover (Risteski and Tevdovski, 2008 and Field,
2005).
2 Data and Model
2.1 Macedonian Stock Exchange development
The story of the Macedonian Stock Exchange began on September 13, 1995. This is the official date of the
establishment of the first organized securities exchange in the history of the FYRO Macedonia. The conclusion of
transactions was carried out on the "stock exchange floor" in the trading hall through the method of continuous
auction and the model of "order-driven market" until 2001.
In December 1999 the Securities Law was amended, which set the basis for dematerialization of securities and
introduced the centralized record keeping of shares ownership in the FYRO Macedonia. In 2001, the trading on the
532

stock exchange floor stopped and the electronic trading of securities started using the new "Bourse Electronic
System of Trading" (BEST).
The first official stock exchange index - the Macedonian Bourse Index (MBI) was promoted. The long-awaited
process of complete dematerialization of shares in more than 670 companies in the country and the operation of the
Central Securities Depository had begun. In 2002, price limitations were introduced for the listed securities on the
Official Market from +/-10% of their last official average price, which contributed to the high volatility of the
Macedonian Stock Exchange shares in the following years.
In 2004 the listing criteria on the Official Market were raised to a higher level, the Unofficial Market was split
into two parts - Market for Publicly Held Companies and a Free Market. The criteria for shares listing were
toughened, and new continuous obligations for reporting were introduced (submitting of a three-month cumulative
unaudited Income Statements, publication of a dividend calendar and notification of large owners). Publication of
the new Exchange index was announced (MBI-10).
A software application called SEI-NET was introduced in 2005, as the official and only way of delivering
information from the listed companies to the Exchange. The new Securities Law was passed, with which a further
harmonization of the regulation in the securities industry with the EU Directives and the principles of IOSCO
(International Organization of Securities Commissions) was made.
The main characteristic of the year 2006 was entrance of capital inflow from regional institutional investors into
FYRO Macedonian capital market. Two private pension funds appear at the domestic market as new institutional
investors. In 2006, state owned company for electricity supply and distribution “ESM” was privatized by selling
certain percentage of their capital to foreign investor. During the year, the state presence on the securities market
was considerable, due to the selling state owned capital in many joint stock companies among which the most
significant was Makedonski Telekom AD Skopje and few more listed companies.
Year 2007 was the most successful since the foundation of MSE back in 1995. The MSE realized record
turnover of 41.7 billion € and MBI 10 index achieved its biggest value breaking the barrier of 10.000 index points,
or 905% growth for 2 years and 8 months. As a comparison, Dow Jones Industrial Average for the same period
(January 2005 – August 2007) rose only 23,87% (from 10.783,01 to 13.357.74), and Standard & Poor’s rose only
21,62% (from 1.211,92 to 1473,99).
The global financial crisis from the middle of 2007 has its impact on the Macedonian Stock Exchange as well,
having in mind its openness and the immense exposure to foreign investors. Foreign institutional investors in order
to cover their liquidity needs, raised from investor’s requests for stepping out of their investment funds, were forced
to sell their stock positions especially those in FYRO Macedonia where highest returns were made.
2.2 MBI 10 Index and Foreign Investors’ turnover
The initial value of MBI 10 index was established at 1.000 on December 30, 2004. For only one year (2005), it grew
to 2.292,04 or 129,2% (Figure 1, peak 1). Additional 61,54% the MBI 10 Index grew in 2006 to value of 3.702,54
(peak 2). Only for the first quarter in 2007, the index grew for additional 48,06% to value of 5.481,92 (peak 3). On
the end of June 2007, the index reaches the value of 6.917,51 or continuing the growth with additional 26,29% just
for the second quarter (peak 4). Macedonian Stock Exchange and it’s MBI 10 Index reaches its peak on August 31,
2007 when the value of the index stop at 10.057,77 (peak 5), which was 45,4% growth for only 2 months, 171,65%
growth for those eight months in 2007 or 905,78% growth for only 2 years and 8 months when the index was first
introduced and measured.
On the end of 2007, MBI 10 starts declining for the first time for 16,61% for the fourth quarter and 23,04%
from its peak in the end of August (peak 6). Decline of 13,35% was evidenced for the first quarter of 2008 and
additional 27,16% for the second quarter of 2008, or 51,42% for only 10 months – from the peak (peak 7 and 8).
Third quarter of 2008 finished with decline of 8,95% and the fourth quarter with additional 52,88% or 79,16%
decline from the peak 16 months ago. MBI 10 index reach its bottom on March 10, 2009 when the value of the
index was only 1.598,5 or 84,11% decline from the highest peak in August 2007 (peak 9). Second quarter of 2009
533

finished with growth of 32,89% compared to the first quarter value or value of 2.532,43 (peak 10). This value has
not change until today (April 29, 2011) with only small fluctuations.
Source: Macedonian Stock Exchange data, atuhors’ own creation
Figure 1. Macedonian Stock Exchange Index MBI 10
Foreign Investors appear on the Macedonian Stock Exchange at the beginning of 2005. From Figure 2, we can
see that first more significant turnover they realized at the beginning of 2006 when they had 4,8 billion MKD
turnover 1 - period when the first more significant growth in MBI 10 can be noticed from Figure 1. Year 2006 was
finished with 11 billion MKD turnover or 282% growth from the previous year, which result with the additional
growth in MBI 10 Index of 61,54% on annual basis. In 2007 we can find the most significant activity of foreign
investors since the total turnover in this year was 26,2 billion MKD or increase of 137,34% from the previous year
(from peak 2 to peak 6 in Figure 1). In this year MBI 10 reach its peak on August 31, with growth of 171,65% only
for those 8 months from that year.
Source: Macedonian Stock Exchange Data, Authors’ own creation
1 Foreign exchange rate for 1 € = 61,3 MKD
P1
P2
P3
P4
P5
Figure 2. Foreign Investors’ turnover
534
P6
P7
P8
P9
P10

After the first decline in the forth quarter of 2007, foreign investors’ turnover starts declining significantly, thus
sending stocks into a tailspin. This can be explain partially with the decline in shares prices but mainly with the
worries over a possible recession in the US. Since then, foreign investors decrease their turnover to minimum
historical levels, investing their money in more developed stock exchanges where the first signs that the global
financial crisis has finished were noticed through the positive rise in the value of their Indexes. For example from
the bottom at the value of 6.626,94 on March 6, 2009 the index Dow Jones Industrial Average rose to 12.810,54 on
April 30, 2011 or 93,31% 2 ; S&P 500 Index from the minimum value of 683,38 on March 6, 2009 rose to 1.363,61
on April 30,2011 or 99,54%.
According to Sumanjeet and Paliwal (2010), the increase in the volume of foreign institutional investment
inflows in recent time has led to concerns regarding the volatility of these flows, threat of capital flight, its impact on
the stock markets and influence of changes in regulatory regimes. They revealed that any problem related to foreign
institutional investors is the problem of management and that India should develop new tools to manage foreign
institutional investors effectively and efficiently. This can be good direction towards which the Macedonian Stock
Exchange and the security market regulators should be acting in the next period if immense volatility wants to be
eliminated and more stable security market established. This can be crucial for the future preservation of domestic
investors, their investments and participation in the security market.
2.3 Linear regression
In determining the influence of foreign investors on the Macedonian Stock Exchange, quantitative analysis has been
used. In the analysis, applied is simple linear regression to assess the impact of the percent change in the foreign
investors’ turnover (independent variable) on the percent change in Macedonian Stock Exchange turnover
(dependent variable). The elaborated period includes monthly data from January 2006 to July 2009 and covers 42
observations. For the analysis, data from the Macedonian Stock Exchange have been used (Table 2).
The theoretically substantiated result, for a small and open economy such as FYRO Macedonia, is that the
higher Foreign Investors’ turnover, substantially contributes to higher turnover on the Macedonian Stock Exchange.
The obtained results indicate high positive linear correlation of 95.9% between the observed variables.
Multiple R 0,959
R Square 0,919
Adjusted R Square 0,917
Standard Error 0,893
Observations 42
Source: Authors’ own calculations
Table 1: Regression Statistics
Adjusted coefficient of determination that is also high, shows that the share of the explained variability in the
total variability was 91.7%, or it can be interpreted that variations or changes that cause the change of the
Macedonian Stock Exchange turnover is 91.7% explained with the changes caused in the Foreign Investors’
turnover, while the rest 8.3% is caused by variations of domestic investors’ turnover.
2 http://www.google.com/finance?client=ob&q=INDEXDJX:DJI
535

Macedonian
Stock Exchange
Turnover 3
% change of
Macedonian Stock
Exchange Turnover
Foreign investors’
turnover 4
% change of Foreign
investors’ turnover
January 2006 729.821.735 n/a 378.193.623 n/a
February 2006 577.553.739 -20,86% 209.652.007 -44,56%
March 2006 11.686.914.615 1923,52% 4.816.177.513 2197,22%
April 2006 675.779.595 -94,22% 408.103.297 -91,53%
May 2006 1.444.799.993 113,80% 695.526.717 70,43%
June 2006 6.973.598.598 382,67% 592.755.881 -14,78%
July 2006 805.369.545 -88,45% 374.738.449 -36,78%
August 2006 1.633.634.810 102,84% 563.604.009 50,40%
September 2006 1.080.810.050 -33,84% 441.510.905 -21,66%
October 2006 2.247.645.453 107,96% 415.140.115 -5,97%
November 2006 1.191.008.537 -47,01% 548.340.330 32,09%
December 2006 1.970.997.263 65,49% 1.620.751.049 195,57%
January 2007 4.223.183.976 114,27% 3.074.477.935 89,69%
February 2007 1.939.990.193 -54,06% 1.079.798.541 -64,88%
March 2007 3.113.169.854 60,47% 1.484.982.020 37,52%
April 2007 4.726.054.924 51,81% 2.268.978.969 52,80%
May 2007 4.288.787.998 -9,25% 2.289.783.912 0,92%
June 2007 1.682.118.805 -60,78% 982.862.018 -57,08%
July 2007 2.862.587.167 70,18% 3.008.865.371 206,13%
August 2007 3.484.656.539 21,73% 1.406.755.845 -53,25%
September 2007 3.633.652.176 4,28% 1.803.018.210 28,17%
October 2007 6.660.933.629 83,31% 5.188.867.297 187,79%
November 2007 3.690.422.068 -44,60% 3.071.538.287 -40,81%
December 2007 1.396.763.119 -62,15% 600.049.436 -80,46%
January 2008 799.744.209 -42,74% 410.988.549 -31,51%
February 2008 1.343.493.855 67,99% 901.081.329 119,25%
March 2008 1.039.593.255 -22,62% 617.414.434 -31,48%
April 2008 1.122.998.859 8,02% 896.826.889 45,26%
May 2008 550.902.297 -50,94% 308.009.474 -65,66%
June 2008 817.860.186 48,46% 536.025.566 74,03%
July 2008 833.487.050 1,91% 697.878.707 30,20%
August 2008 1.738.052.930 108,53% 1.800.101.420 157,94%
September 2008 732.285.137 -57,87% 363.799.256 -79,79%
October 2008 1.248.960.065 70,56% 778.351.913 113,95%
November 2008 482.533.487 -61,37% 165.605.493 -78,72%
December 2008 1.664.049.515 244,86% 475.252.541 186,98%
January 2009 350.616.520 -78,93% 384.731.507 -19,05%
February 2009 354.127.668 1,00% 284.187.454 -26,13%
March 2009 367.193.897 3,69% 156.461.320 -44,94%
April 2009 1.519.038.214 313,69% 233.628.077 49,32%
May 2009 856.912.383 -43,59% 354.504.653 51,74%
June 2009 570.652.166 -33,41% 103.002.716 -70,94%
July 2009 658.728.115 15,43% 163.232.827 58,47%
Source: Macedonian Stock Exchange Data, Authors’ own creation
Table 2: Macedonian Stock Exchange Data
3 Macedonian Stock Exchange Turnover represent the total value of bought or sold securities for that month
4 Foreign Investors’ turnover represent the total value of both bought and sold securities for that month by the foreign investors
536

Source: Authors’ own calculations
Figure 3. Line Fit Plot
The standard error of regression is small, amounting to 0.893, and shows us the absolute deviation of the
empirical data from the regression analysis of the sample. The regression equation of the sample is as follows:
'
y � b � b x � 0,
105 � 0,
858x
0
1
i
i
Stretch of the Y-axis or value of the dependent variable when the independent variable X = 0 is the value of
regression parameter b 0 � 0,
105.
Regression parameter 1 0,
858 � b indicates the change in the dependent variable Y
when the independent variable X will increase by one of its unit. In this case, it suggests that if the Foreign
Investors’ turnover were increased by 1%, then the Stock Exchange turnover on average would increase by 0.858%.
Estimated value of this ratio with the risk of error of 5% will range from 0.777% to 0.940%. Testing the regression
coefficient b1
� 0,
858 gives p value of 0.000. This value indicates the smallest level of significance for which the
zero hypothesis H 0 : �1
� 0 can be dismissed and the alternative hypothesis H1 : �1
� 0 accepted. In our case, the
alternative hypothesis is accepted, i.e. the regression coefficient b1
� 0,
858 also statistically is confirmed that is
different from 0, or that the independent variable truly affects the dependent variable. Residual Plot Figure shows
the standardized residuals of the regression on Y-axis and the predicted standardized values of the regression on
X-axis.
Source: Authors’ own calculations
Figure 4. Residual Plot
537

Source: Authors’ own calculations
Figure 5. Normal probability plot
From the previously elaborated, it can be concluded that the assumption of full linearity is fulfilled, but the
same it cannot be entirely declared for the assumption of homoscedasticity or equality of residuals.
All this statistically confirms the theoretical expectation that Macedonian Stock Exchange turnover positively
depends on the Foreign Investors’ turnover.
3 Summary
From the analysis of the Macedonian Stock Exchange development since its beginnings in 1995, having in mind the
small history of existence, it can be concluded that the Stock Exchange is still underdeveloped and too much opened
and exposed to influences from abroad. Macedonian Stock Exchange Index MBI 10 for only 2 years and 8 months
grew 905%, and for the following 18 months was reduced for 84%, which is an evidence for very high volatility.
Investigating the reasons, we can conclude that despite the global financial crises, direct reason for the fast growth
and reduction of the MBI 10 Index was the foreign investors’ turnover.
The theoretically substantiated result, for a small and open economy such as FYRO Macedonia, is that the
higher Foreign Investors’ turnover, substantially contributes to higher turnover on the Macedonian Stock Exchange.
The obtained results from the regression indicate high positive linear correlation of 95.9% between the observed
variables.
Both theoretically and empirically, this study proved that foreign investors substantially contribute to the
Macedonian Stock Exchange boom and bust. In order to be protected from future immense volatility, FYRO
Macedonian security market regulators need to establish more sound rolls that will enable effective and efficient
management of the stock exchange. This should provide stability and eliminate the negative influence coming from
the openness and exposure of the Macedonian Stock Exchange to foreign investors.
4 Acknowledgements
University American College Skopje supported this Research paper. I appreciate the effort of Marjan Petreski for his
critical review and Aleksandar Ajevski for his assistance during data collection.
538

5 References
Field, A. (2005). Discovering Statistics Using SPSS. Second edition, Sage publications
http://www.mse.org.mk/ReportDetail.aspx
http://www.google.com/finance
Risteski, S., Tevdovski, D. (2008). Statistics for Business and Economics, Economic Faculty - Skopje
Sumanjeet, S., Paliwal, M. (2010). Liberalization of foreign institutional investments (FIIS) in India: Magnitude,
impact assessment, policy initiatives and issues. Global Journal of International Business Research, 2010, Vol.
3. No. 3, pp 22-41
539

GENDER BIAS IN HIRING, ACCESS TO FINANCE AND FIRM PERFORMANCE: EVIDENCE FROM
INTERNATIONAL DATA *
Nigar Hashimzade
School of Economics, University of Reading, United Kingdom
Yulia Rodionova †
Leicester Business School, De Montfort University, United Kingdom
October 18, 2010
Abstract. We analyze the effect upon the firm's performance of the interaction between an employer and
an employee, with the possibility of gender-based bias in hiring. In the presence of bias the characteristics
of a successful employer-employee match may be independent of the employee's productivity. In a model
with imperfect information and signalling we make some predictions about the impact of the gender of the
owner and the top manager on the firm's performance and on the degree of discrimination in access to
finance. Further, using firm-level data from twenty-six countries of the former Soviet bloc from the
Business Environment and Enterprise Performance Survey in 2009, we explore the effect of the gender of
the owner and the manager on a firm's performance, both directly and through access to finance, to test the
model's hypotheses. We find that a hiring policy based on individual characteristics other than productivity
(after controlling for various factors such as industry, country and possible constraints on access to external
finance) has a negative effect on firm's performance. Our empirical findings are consistent with the
predictions of the model. In particular, we find that firms with female top managers perform better than the
firms with male top managers; this effect is mitigated if the firm is owned by a female, or if it operates in
the industry where female-style leadership is valued. The effect is also less pronounced for those female top
managers who started working in the same industry under the Soviet system - before 1992. If the
unobserved heterogeneity in the firms' performance is caused by the difference in the entrepreneurial skills
of both the owner and the top manager, our findings suggest, in line with the model predictions, that the
average skills of female owners (managers) are lower (higher) than those of their male counterparts. We
also find an evidence of spillovers of discrimination in the labour market to statistical discrimination in
access to finance. Our results suggest that gender bias in hiring may lead to an inefficient allocation of
physical and human resources.
Keywords: discrimination, entrepreneurship, finance, gender, job sorting, signalling, small and medium
enterprises
* This version: October 15, 2010. Based on a previous version by Hashimzade et al. (2010).
† y Corresponding author. Email: yrodionova@dmu.ac.uk . The authors are very grateful to Mark Casson, Susan Marlow and
Natalia Vershinina as well as participants of the EMFT conference in Milas, Turkey, June 2010 and WIEM2010 conference for
helpful comments and suggestions.
540

INTELLECTUAL CAPITAL DIMENSIONS AND QUOTED COMPANIES IN TEHRAN EXCHANGE,
BASED ON BOZBURA MODEL
Rasoul Abdi
Department of accounting, Islamic Azad University of Bonab Iran
Abdi_rasool@yahoo.com
Nader Rezaei
Department of accounting, Islamic Azad University of Bonab Iran
naderrezaeimiyandoab@gmail.com
Yagoub Amirdalire Bonab
Department of accounting, Islamic Azad University of Bonab Iran
amirdalir@yahoo.com
Abstract:The purpose of present study is the presentation of a proper definition of intellectual capital, its components, and the
relationship between intellectual capital and market value of quoted companies in Tehran exchange based on Bozbura model. Mostly,
the components of intellectual capital were defined in three dimensions of human capital, relation capital and structural capital. The
researchers tried to represent the effects of intellectual capital on the market value of quoted companies in Tehran exchange. This
study addressed four hypotheses based on the nature and significance of the study. Based on the results of data analysis, there was a
significant relationship between human capital and market value of quoted companies in Tehran exchange. Also, structural capital was
significantly related to human capital and relation capital.
Keywords: Market value, intellectual capital, human capital, structural capital, relation capital, and Tehran exchange.
Introduction
Intellectual capital (IC) information gained importance because it is seen as an integral part of firms’ value-creating
processes (Bukh, 2003). Nowadays, intangible assets such as staff’s skills, strategic and process quality, software,
patents, brands, supplier and customer relationship provide a great involvement to the successes in many corporate,
so as considered as valuable assets. These assets are delivering a fast-growing contribution to corporate
competitiveness and usually are classified as intellectual capital (IC) (Hofman, 2005). The IC is defined as
intellectual resources that have been “formalized, captured, and leveraged” to produce higher value assets (Prusak,
1998). There are many models and classifications on intellectual resources in literature. Most of them could be
termed as the Sveiby-Stewart-Edvinsson model (Bukh et al., 2001). The model consists of human capital (HC),
structural capital (SC), and relationship capital (RC). According to Sveiby (1997), the HC involves capacity to act in
wide variety of situation to create both tangible and intangible assets. Stewart (1997) emphasized that the primary
purpose of a HC is innovation of new products and services or of improving in business process.Although the
importance of intellectual capital (IC) has increased greatly in the last two decades (Serenko and Bontis, 2004),
many organizations are still struggling with better management of IC due to measurement difficulties (Dzinkowski,
2000). Many authors have argued that IC, which represents the stock of assets generally not recorded on the balance
sheet, has become one of the primary sources of competitive advantage of a firm (Bontis, 1996; 1998, 2001;
Edvinsson and Malone, 1997; Roos et al., 1998; Stewart, 1997; Sveiby, 1997). Given the remarkable shift in the
underlying production factors of a business within the new knowledge economy (Drucker, 1993), it is important for
firms to be aware of the elements of IC that could lead to value creation. Knowledge management and its related
area emphasize that, achievement of the permanent competition advantage, in concept of global modern economy,
is depended on the organization's capacities and abilities, and proper usage of organization's knowledge based
sources. It must be mentioned that, all of the organization's sources are not in a same importance level. The
structures of organization's properties are in change. In the past the objective properties of organizations were more
significant than the subjective properties of organizations but nowadays the subjective properties are more
important.The modern economy which is formed of knowledge and information has the effective role on intellectual
capital in research and business branch. The main purpose of the present study is the presentation of a proper
definition to intellectual capital, its components, and the relationship between intellectual capital and market value
of quoted companies in Tehran exchange, based on Bozbura model . In this area many researches have been done
yet, such as (Anvari Rostami & Rostami, 2003), (Bontis, 2002), (Kujansivu & Lonnqvist, 2008), (Mouritsen, 2001),
541

(Lynn, 2000), and some articles presented in Copenhangen city of Denmark by European Accounting
Association(EAA).
Accounting for intellectual capital
Although the writers including Reich (1991) and Stewart (1991) had previously identified the growing importance
of intellectual capital as a source of long-term value creation for organizations, it was in the mid-1990s that interest
in it began to escalate. At that time a number of popular texts on the subject were published, including Brooking
(1996), Edvinsson and Malone (1997), Roos et al. (1997), Stewart (1997) and Sveiby (1997), together with
Edvinsson’s seminal 1997 paper on Skandia AFS’s pioneering work in intellectual capital management. From a
specifically accounting perspective, the emergence of intellectual capital as a key organizational resource raised the
question of how to report it alongside other assets in financial statements. This was not a new problem, however, as
there have long been key organizational assets for which it has not been possible to report values. The goodwill built
up by a business could only be included within the stock of intangible assets when it was acquired by another
business, subsequently to be amortised or more commonly written of at the time of acquisition. Specific intangible
assets such as brands, themselves later to be included within the designation of intellectual capital, were subject to
the same provisions. Attempts to account for people”, or more specifically their experience, expertise, creativity,
organizational commitment, etc., currently designated as the human capital component of intellectual capital, had all
but been abandoned after first human asset, then human resource accounting had slipped down the research agenda
in the late 1970s.
The problem of accounting for intellectual capital was clearly one that could not be ignored. Intellectual
capital’s growing importance was underlined by the growing disparity between the book values of many businesses,
determined in accordance with prevailing financial accounting and reporting provisions, and the market values of
these same entities. The extent of what Edvinsson 1997) termed the “hidden value” of organizations was frequently
greatest in the knowledge-based industries, precisely the places where intellectual capital is most critical to longterm
business performance. This situation quickly became more apparent with the rise of the dot.com companies in
the late 1990s, although post-Enron concerns about the veracity of financial statements and the general downturn in
the global economy following the events of September 2001 have seen a reversal in this upward trajectory. The
worry was, and remains, that disparities of this sort have the capacity to disrupt the works of the capital market. In
order to ensure that this did not occur, some means had to be identified for reporting intellectual capital to the
market. Coupled with this, the growing tendency to link executive remuneration to share price meant that the
accountancy profession was under great pressure to demonstrate the true value of the business in its financial
statements If it had been possible to identify some simple means of extending the established accounting calculus to
incorporate intellectual capital, the (on-going) debate about accounting for intangible assets would have already
provided clear indications on how to proceed. It had not, which meant that the accountancy profession was not well
placed to deliver reliable information of the sort many stakeholders might, not unreasonably, expect of it. In
Edvinsson (1997) the dilemma facing the accountancy profession is clearly visible.
Edvinsson argues that a preferable approach is one that provides information on the success with which an
organization’s management has grown the stock of intellectual capital. Such information would, of necessity, be of a
more prospective nature, and thereby of greater relevance to stakeholders interested in the sustained value creation
capacity of the business. The Navigator was designed to provide information on the human, customer, process and
renewal and development foci of the organization, in addition to financial information, and to do so using a set of
indicators that would also convey how the organization viewed its own value creation strategy. Initially, Skandia
reported such information in the form of brief supplements to its financial statements, quickly developing a more
comprehensive approach in the late 1990s (Mouritsen et al., 2001a). In retrospect, it is possible to view the Balanced
Scorecard as providing an alternative approach to the Navigator (Kaplan and Norton, 1992, 1993, 1996), while
Sveiby’s Intangible Assets Monitor offers a third approach (Sveiby, 1997). Lev (2001) has subsequently developed a
Value Chain Scoreboard that makes use of a possible nine different information foci.
As if to emphasise the difference between reporting the success (or otherwise) of intellectual capital
management from traditional financial reporting, the various scorecard approaches normally commended the
extensive use of non-financial indicators. These are selected for their relevance to the task in hand, an idea more
commonly associated with managerial accounting than financial ccounting. In the last analysis, however, scorecards
542

affirmed the value of employing a quantitative approach to reporting, something that was soon to be challenged as
researchers sought to develop a second wave of intellectual capital reporting frameworks.
In the vanguard of this development were researchers associated with the Danish Agency for Trade and
Industry, which funded a programme beginning in 1998 and continuing to date. What these researchers commend is
a narrative approach to intellectual capital accounting and reporting, using what are termed as Intellectual Capital
Statements. The basis for such statements is the organization’s knowledge narrative (subsequently, termed the
management narrative), from which is adduced a number of key management challenges. These in turn inform a set
of management challenges for which relevant indicators are identified. The whole information set (including
financial statements) is then reported to stakeholders on a regular basis, using a range of representational forms inter
alia extensive narratives. A similar approach was also commended in the final report of the Meritum research group,
the Intellectual Capital Report combining three elements: the vision of the firm; the summary of intangible resources
and activities; and the system of indicators (Meritum, 2002). When discussing the indicators to be used in the Value
Chain Scoreboard, Lev (2001, p. 115) comments that they “should satisfy three criteria to ensure maximal
usefulness.”. The three criteria are: quantitative in nature; standardized (or easily standardized) to permit inter-firm
comparison; and empirically linked to corporate value. This observation is at odds with the trajectory implicit in the
move to narratives. In the case of narratives, the objective would seem to be that of developing an intellectual capital
account that adds value by virtue of the richness of its information content and reflexive nature. A strong divergence
of approach seems to be occurring. Whilst the European model is moving in a strongly qualitative direction, the
North American model is seeking to retain as much of a “hard number emphasis as possible. This is evident in the
case of the Value Creation Index (Cap Gemini, 2000; Low, 2000) in which an organisation’s scores in respect of
nine intangible value drivers are weighted in an attempt to determine an index that might be compared with like
organisations. Lev himself has also been active in developing a Knowledge Capital Earnings methodology (Gu and
Lev, 2001; Stewart, 2001; Tayles et al., 2002). In common with the earlier Calculated Intangibles Value metric
(Dzinkowski, 1999), this approach is designed to generate market-friendly intangibles valuations incorporating
reliable and robust financial information.
Origins of the Study
This part of study is about the concept and components of intellectual capital based on presented model. Until now,
the definitions of IC have been discordant. In recent years, driven by necessity, many individuals and groups from
different disciplines have tried to agreeon a standard definition for IC (Edvinsson and Malone, 1997). Initially,
Edvinsson and Malone’s (1997) and Stewart’s (1997) research helped to bring the term to the forefront. Edvinsson
and Malone (1997, p. 358) defined IC simply as “knowledge that can be converted into value.” Stewart (1997, p. x)
broadened the definition to IC as “intellectual material – knowledge, information, intellectual property, experience –
that can be put to use to create wealth” by developing competitive advantage in an organization. When intellectual
material is formalized and utilized effectively, it can create wealth by producing a higher value asset, called IC. The
intellectual capital is a part of the total capital which is knowledge based and belongs to company. Based on this
definition, intellectual capital includes the knowledge transferred to intellectual possession, intellectual assets of a
company, and the final results of the transferring process. The standard definition of moral ownership is included
some ownership rights such as patent, trade mark and copy right. These assets are the single form of intellectual
capital which is appropriate for accounting aims (Anvari Rostami & Rostami, 2003).
Intellectual capital consists of knowledge, information, intellectual assets and experiments used for wealth
gathering. It includes subjective group abilities or key knowledge as a group (Bontise, 2000).
Intellectual capital includes various intangible sources which are valuable to an organization (Kujansivu &
Lonnqvist, 2008).
Mostly, in all models of the intellectual capital assessment which have designed, its components are defined in three
dimensions of human capital, relation capital and structural capital.
543

Human Capital
The human capital, which can be defined as the first dimension of intellectual capital, is the abilities, skills, and the
proficiency of organization's personnel. The human capital enjoys thought and involves all the abilities and skills of
the organization personnel (Lynn, 2000). The combination of Sveiby and Stewart definitions of HC if given by
Edvinsson and Malone (1997) who defined it as combinations of knowledge, skill, innovativeness, and ability of the
company’s individual employees . In other words, the human capital is consisted of general and professional
knowledge of personnel, leadership abilities, risk and the problems solving. The main purpose of human capital is
goods innovation, services and business importance and the improvement of business process (Mouritsen, 2001).
The most important indexes of human capital are the professional competences of personnel, the experiments, the
knowledge, the number of company personnel with the earlier related knowledge, and the accurate distribution of
responsibilities and named as the individual capability in the model of “Intangible Asset Monitor” (Sveiby, 1997),
and in the “Balance Score Card” model, it is named as the learning and development dimension (Kaplan and Norton,
1999). Therefore, when forming our model, one dimension of this model should be saved for the human capital .
Relation Capital
Relation capital as a second dimension of intellectual capital is the representation of company's relations to the out
world. It includes out organization's dependence such as customer's loyalty, company's famous, and company's
relations with the provider of its sources. On considering the relationship between company and outside, it is
observed that there are other factors which are more effective than customers. So the producer relations and society
relations must be defined (Bozbura, 2004). Relational capital comprises the knowledge embedded in all the
relationships an organization develops, whether it is with customers, competitors, suppliers, trade ssociations or
government bodies (Bontis, 1999). One of the main categories of relational capital is usually referred to as customer
capital and denotes the “market orientation” of the organization. There is no consensus on a definition of “market
orientation” (Bontis et al., 2001), but Kohli and Jaworski (1990) defined it as the organization-wide degree of
market intelligence generation, dissemination, and action based on the current and future needs of customers.
Relation capital includes the company's possessions grant, the company's relations with people, and organization
related to customers, customer's preservation or lose rate, market share, and also net profitably per each customer
(Anvari Rostami & Rostami, 2003).
Structural Capital
Structural capital is the third dimension of intellectual capital which includes capacities to percept of market needs
such as patent, knowledge based structures, organizational processes and cultures. This dimension of organizational
asset sometimes is named intellectual asset, ultra structure asset, innovation capital or process capital (Bozbura,
2004) The Structural dimension is defined in the intellectual capital as the Structural capital. The Structural capital is
the sum of all assets that make the creative ability of the organization possible. The mission of the firm, its vision, its
basic values, strategies, working systems, and in-firm processes can be counted among these assets. Structural
capital is one of the foundation stones of creating learning organizations. Even though the employees possess
adequate or high capabilities, an organizational structure that is made up of weak rules and systems and which
cannot turn these capabilities into a value, prevents the firm from having a high performance. In contrast, a strong
Structural capital structure creates a supporting environment to its workers and thus leads to workers’ risk taking
after their failures. Besides, it leads to the decrease of the total cost and to the increase of the firm’s profit and
productivity. Therefore, the Structural capital is a vital structure for organizations and in an organizational level; it
has a critical importance for the realization of measuring the intellectual capital (Bontis, 1998, 1999, 2001,
2003).Many factors are defined among the models that are made in order to measure the Structural capital. Visible
assets such as the patents of the firm, copyrights, databases, computer programs and intangible assets such as the
methods related to business management, company strategies, the culture of the company is among these factors.
The high investments of technology or the high number of computer and programs in a firm us not feature, which
adds a plus value to a firm. In order for these to make a contribution to the company, the workers in the firm should
have the abilities to use these systems to interpret the results, to make them knowledge and to use them in the
relations (Fitz-enz, 2001). As long as they are not put to use, the existence of systems that possess and transmit
knowledge, which is the foundation stone of the Structural capital, is not means of adding value. Therefore, it would
544

e wrong to claim that the Structural capital has a direct and linear relationship with the performance of the
company.
Development of research hypotheses
The set of hypotheses in this study that explores the relationship between human, structural and relational capital
and market value of quoted companies in Tehran exchange based on Bozbura model . Using a questionnaire for data
collection, Bontis (1998) found a significant relationship between each element of IC (human, organizational, and
relational) and firm market value. Several financial measures such as profit, profit growth, sales growth and ROI
were used as indicators of firm performance. He also found significant associations between IC elements. In another
study that used the same questionnaire, Bontis et al. (2000) studied the inter relationship of the components of IC
and the relationship between structural capital and firm performance.
Therefore, we can our hypotheses related to the human, structural and relational capital as thus
H1. There is a significant relationship between human capital and market value of quoted companies in Tehran
exchange.
H2 .There is a significant relationship between relation capital and market value of quoted companies in Tehran
exchange.
H3 .There is a significant relationship between structural capital and human capital.
H4.There is a significant relationship between structural capital and relation capital
According to above discussion, For these hypotheses, we can define our model of research as shown in Figure 1
_
_
Methodology
Positive relation
H: hypothesis
Human capital
H3
Structural capital
H4
Relation capital
Figure 1. Theoretical framework of research hypotheses
Company's
market
value
The analysis of the data was on the basis of descriptive statistics, which allows the researchers to describe the
sample data. The qualitive research model is the former of structural theory for quantitive analysis (Sieber, 1973).
545
H1
H2

From the time point of view, present study is a segmental one which discusses on managers, nurses, and experts
points of view of 2010.
The population of the present study was consisted of all the quoted companies 442 companies which were in several
industries- on Tehran exchange. Because of the company's span and their dispersal in Iran, there was not an access
for considering all of the population, so the population was considered by using statistic methods. The research
population was a normal distribution. It is used COCHRAN formula for sample capacity determining by supposing
below:
By supposing the hypothesis verification proportion as (P) and the hypothesis deficiency proportion as (q), in equal
situation of 50%, error estimate (d =10%)10% and distance assurance of 95% (t= z =1/96), the content sample of
research was determined on the base of Cochran formula:
N: population sample extent
n: selected sample extent
Finally, with the selected sample content of 78 companies at error estimate of 10% and coefficient assurance of
95%, it was clamed that the selected sample had the all characteristics of a population. So, the results were
generalized to the total of population.
It was adjusted to a questionnaire for hypotheses assuring. This questionnaire consisted of 33 questions which
included five choice according to Likert's scale. There were 10 questions on human capital area, 11 question on
relative capital and 12 questions on structural capital. It must be mentioned that all of the questions were determined
according to below scales:
1. The scales were according to organizational structures of Iranian companies.
2. There was not any industrial prejudice on questionnaire.
3. The introduced scales were qualitive one which could be discussed with the quantitive dimension.
4. Questions were significant. Because of the fact that the problem of the present study was the sufficient
unfamiliarity of managers and experts with the concepts of intellectual capital, the scales, and the definition
of intellectual capital, the scales, definition of intellectual capital, and its components were submitted them,
too.
Human Capital Scales
Bontis has defined some dimensions of human capital such as the satisfaction of the personnel, underwriter
company, motivation, staff preserving, leadership and management, knowledge producing, knowledge distributing,
learning, knowledge assembling and the period assigned for personnel instructions (Bontis, 2002).Also, in other
study done by Miller, there have been defined some dimensions such as the workers industrial knowledge, the
workers learning expense, and the workers high level education such as M.A and P.H.D (Miller, 1999). On
designing of the questionnaire, some scales were regarded such as the personnel informing on various information,
encouragement of group work, personnel innovation and risk, the personnel ideal level of general skills, and the
importance of investment on instructions.
Relation Capital Scales
2
N(
t �(
p.
q))
n � 2 2
N(
d ) � t �(
p.
q)
Relative capital is one of the most important dimensions of the intellectual capital which includes the relationship
between company's parts on value chain. It is clear that the fundamental scales of relation capital depend on
customer and market. So, besides the shareholders who are the important part of the company, the producers and
society must be defined in relation capital. In question designing of relation capital there were regarded some scales
546

such as customer loyalty, customer satisfaction, sale's volume of permanent customers, the number of customer's
complaints, the extent of customer's information using in company, customer based information and the market
requirements.
Structural Capital Scale
Structural capital is the whole of the assets which make possible the organization creatorship ability. In the research
done on Canadian industry, there were some scales which regarded on question designing such as earned income in
lieu of R&D expenditure, access to information bases, the presentation manner of new product, creatorship support,
management efficiency of company's information systems and financial results created in organization, access to
unlimited information, the information system of MIS management, investing on R&D, information bases updating,
idea's expanding leadership and new products and productivity increase.
It is used the 15 th edition of SPSS software and Cronbach’s for determining the question validity. Cronbach’s α
coefficient was between 0 and 1. The 0 was the demonstrator of invalidity, and the 1was the demonstrator of the
questions validity. The resultant which its validity was low, it was not be valid. However, the high validity degree
was not the guarantee of appropriate measuring usage. According to Nunnally, the α coefficient of question's
validity must be more than 0.7. The results of human capital scales, intellectual capital scales and structural capital
scales analyzed according to Cronbach’s α test. Human capital alpha coefficient was 0.873, the structural capital α
coefficient was 0.843 and the relative capital alpha coefficient was 0.823. Because all of these amounts are more
than 0.7, it was concluded that the research questions were valid.
Discussion
To prove hypothesis of the model a multiple regression equation is created. For the H1 and H2, an equation can be
written as follows:
y � B0
� B1
X 1 � B2
X 2
In this formula Y was the demonstrator of company's market value, B0 was axis intersection, B1 was the human
capital inclination, X1was the human capital average, B2 was relative capital inclination and X2 was the relative
capital average. The analysis of present study was done based on X2 of 15 th edition of SPSS software. The results
were as follow:
In discussion about the first hypothesis: there is a significant relationship between human capital and market value
of quoted companies in Tehran exchange, the analysis of this hypothesis showed that sig = 0.021< 0.05, so the
rejection probability of 0 H was rejected and H1 hypothesis was accepted in the 95% assurance.
Table I.Test of hypothesis first
Residual
-252.0
-26.0
278.0
Chi-Square Test
Frequencies
Expected N Observed N
260.0 8
260.0 234
260.0 538
780
Test Statistics
option
544.092
2
.021
Chi-Square(a)
df
Asymp. Sig.
3
4
5
Total
a 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency is 260.0.
547

In discussion about the second hypothesis: there is a significant relationship between relation capital and market
value of quoted companies in Tehran exchange, according to the results of data analysis sig= 0.009< 0.05, so the
rejection probability of 0 H was rejected and H1 hypothesis was accepted in the 95% assurance.
Table II. Test of hypothesis second
Chi-Square Test
Frequencies
Test Statistics
a 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency is 286.0.
After the first and second hypotheses testing, it was computed human capital inclination, B 1 = 0.63, relation capital
B = 0.74, and B 0 = -2.818. Finally the below formula was presented:
inclination, 2
Table III
Residual
-277.0
-72.0
349.0
Standardized
Coefficients
Beta
4.251
5.243
Expected N
286.0
286.0
286.0
option
712.287
2
.009
Std. Error
Unstandardized
Coefficients
.000
.000
.000
B
-2.818
1
. 63
. 74
y � �2
/ 818 � 0 / 63x
� 0/
74x
(Constant)
human
capital
relative
capital
In discussion about the third hypothesis: there is a significant relationship between intellectual capital and human
capital, according to the results of data analysis sig= 0.019< 0.05, so the rejection probability of H 0 was rejected
and H1 hypothesis was accepted in the 95% assurance.
548
Observed N
9
214
635
858
Chi-Square(a)
df
Asymp. Sig.
2
3
4
5
Total
Mod
e
1

Table IV. Test of hypothesis third
Chi-Square Test
Frequencies
Test Statistics
a 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency is 156.0.
In discussion about the case of last hypothesis: there is a significant relationship between intellectual capital and
relation capital, according to the results of data analysis sig= 0.015< 0.05, so the rejection probability of H 0 was
rejected and H1 hypothesis was accepted in the 95% assurance.
Table V. Test of hypothesis last
Conclusion
Residual
-140.0
-32.0
172.0
Residual
-138.0
-24.0
162.0
Expected N
156.0
156.0
156.0
option
321.846
2
.019
Chi-Square Test
Frequencies
Observed N
16
124
328
468
Chi-Square(a)
df
Asymp. Sig.
Expected N Observed N
156.0 18
156.0 132
156.0 318
468
Test Statistics
option
294.000
2
.015
Chi-Square(a)
df
Asymp. Sig.
3
4
5
Total
3
4
5
Total
a 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency is 156.0.
The results of this study showed that the presented research model is meaningful in Iran industry. Based on the
results of data analysis, it was concluded that the main results of the present study was the significant relationship
between human capital and the relation capital of quoted companies in Tehran exchange with company's market
value. Also, the structural capital of these companies is significantly related to human capital and relation capital.
549

References
Anvari Rostami & Rostami, (2003) "apprising the companies intellectual capital measurement and evaluation
models "the Iranian accounting and auditing review,winter,vol,10.number34.pp51-75
Bontis,N,Keow ,w.c.and Richardson,s. (2000), "Intellectual capital and business performance in Malaysian
industries" . Journal of Intellectual capital,Vol.1No.1,pp.85-100
Bontis,N(2003), "Intellectual capital disclosure in Canadian corporations", Journal of Human Resource Costing and
Accounting,Vol.7 Nos 1/2, pp.9-20.
Bozbura,F.T.(2004), "Measurment and application of intellectual capital in Turkey". The Learning Organization,
Vol.11 NO.4/5,PP.357-367.
Bontis, N. (2001), “Assessing knowledge assets: a review of the models used to measure intellectual capital”,
International Journal of Management Reviews, Vol. 3 No. 1, pp. 41-60.
Bontis, N. (2004), “National intellectual capital index: a United Nations initiative for the Arab region”, Journal of
Intellectual Capital, Vol. 5 No. 1, pp. 13-39.
Benefit, Deutsche Bank Research, Frankfurt am Main, available at: www. dbresearch.com .
Bukh, P.N. (2003), “The relevance of intellectual capital disclosure: a paradox?”, Accounting, Auditing &
Accountability Journal, Vol. 6 No. 1, pp. 49-56
Bukh, P.H., Larsen, H.T. and Mouritsen, J. (2001), “Constructing intellectual capital statements”, Scandinavian
Journal of Management, Vol. 17, pp. 87-108
Bontis, N. (1999), “Managing organisational knowledge by diagnosing intellectual capital: framing and advancing
the state of the field”, International of Journal of Technology Management, Vol. 18 No. 5, pp. 433-62
Dzinkowski, R. (2000), “The measurement and management of intellectual capital: an introduction”, Management
Accounting, Vol. 78 No. 2, pp. 32-6.
Edvinsson, L. and Malone, M.S. (1997), Intellectual Capital: Realizing Your Company’s True Value by Finding its
Hidden Brainpower, HarperBusiness, New York, NY.
Hofman, J. (2005), Value Intangibles! Intangible Capital can and must be Valued Owners and Valuers alike will
Kujansivu,P.and Lonnqvist,A.(2008), "Business process management as a tool for intellectual capital
managerment". Knowledge and Process Management,Vol.15 NO.3,PP159-169.
Lynn,B.E.(2000), "Intellectual capital: unearthing hidden value by managing intellectual assets". Ivey Business
Journal, Vol. 64.N.3,PP.48-52.
Mouritsen,J,Larsen,H.T,Bukh,P.N.D.(2001), "Intellectual Capital and the capable Firm": Narrating,Visualizing and
Numbering for Managing Knowledge. Accounting Organizations and society,Vol.26,pp.735-762.
Mouritsen,J(2003), "Overview Intellectual capital and the capital market": the circulability of intellectual
capital,Accounting,Auditing and Accountability Journal, Vol.16 NO.1PP.18-30
Miller,M.,DuPont,B.,Fera,V.,Jeffrey,R.,Mahon,B.,Payer,B.andStarr,A.(1999),“Measuring And reporting intellectual
capital–from adiverse Canadian industry perspective , “ International Symposium Measuring and Reporting
Intellectual Capital, 9-10 June
Meritum (2002), Proyecto Meritum: Guidelines for Managing and Reporting on Intangibles, Madrid.
Prusak, L. (1998), Working Knowledge: How Organizations Manage What They Know, Harvard Business School
Press, Cambridge, MA
Sieber,S.D.(1973),“Theintegrationoffieldworkandsurveymethods”, American Journal of Sociology, Vol. 78 No. 6,
pp. 59 – 1335 .
Stewart, T.A. (1997), Intellectual Capital, Nicholas Brealey, London.
Sveiby, K. (1997), The New Organizational Wealth: Managing and Measuring Knowledge based Assets, Barret-
Kohler, San Francisco, CA.
Sveiby, K.E. (1997), The New Organisational Wealth: Managing and Measuring Knowledge Based Assets, Berret-
Koehler, San Francisco, CA.
550

551

EMPIRICAL FINANCE
552

553

ANALYZING THE LING BETWEEN U.S. CREDIT DEFAULT SWAP SPREADS AND MARKET RISK:
A 3-D COPULA FRAMEWORK
Hayette Gatfaoui, Rouen Business School, France
Email: hgt@rouenbs.fr, hgatfaoui@gmail.com
Abstract. The recent mortgage subprime credit crisis, which burst in mid-2007, shed light on the heavy interconnections between
credit markets and stock markets. Such turmoil generated a huge questioning of the prevailing Basel 2 requirements with respect to
risk assessment. In this lens, we focus on the credit default swap market through Markit CDS indexes, and to study its linkages with
the U.S. stock market. The linkages under investigation are envisioned at two levels among which a directional price channel and a
volatility price channel. Furthermore, we assess the simultaneous interaction between the CDS market and the U.S. stock market
through a three dimensional copula setting. Therefore, we account simultaneously for the asymmetric dependence structures as well as
the differences between the dependence structures of CDS spreads with both market price and market volatility. Our study is relevant
for risk monitoring and risk management prospects such as value-at-risk implementations (and other related scenario analysis).
Keywords: Credit risk · Dependence measures · Market risk · Multivariate copulas· Risk management · Tail dependence
JEL classification: C16 · C32 · D81
1 Introduction
The recent mortgage subprime crisis as well as the resulting global financial crisis shed light on the weaknesses and
required enhancements of the prevailing risk management practices. Among the most important enforcements,
liquidity concerns, counterparty credit risk, the correlation between various risks and model stress testing as well as
related scenario analysis have been highlighted by the Basel Committee on Banking Supervision under the Basel 3
framework. With regard to liquidity, various liquidity measures are proposed at both the level of financial assets and
the bank level (i.e. in- and off-balance sheet prospects). On the correlation viewpoint, the risk of correlation between
risks (e.g. impacts of liquidity risk on market risk and vice versa) refers to the linkages between asset classes, and
between banks/financial institutions among others. On the stress testing and scenario viewpoints, the mitigation of
potential model risk and measurement errors is targeted.
Under the Basel 3 setting, we focus on the correlation risk between credit default swap (CDS) spreads and
market risk components (Norden and Weber, 2009). Specifically, CDS spreads represent a credit risk proxy whereas
market risk is envisioned with respect to two dimensions, namely a market price risk and a market volatility risk
(Dupuis et al., 2009; Gatfaoui, 2010; Scheicher, 2009; Vaz de Melo Mendes and Kolev, 2008). The market price
risk illustrates the impact of the global market trend (i.e. common general link within the stock market) whereas
market volatility risk represents the magnitude of global market moves (i.e. volatility feedback, liquidity concerns).
We study the asymmetric linkages between CDS spreads and both market price and market volatility risks through a
three-dimension copula methodology.
The previous setting targets a sound assessment of credit risk in the light of the stock market’s influence with
respect to the curse of dimensionality, namely the trade-off between the number of parameters, the problem’s
dimension and the sample size. The corresponding multivariate dependence structures exhibit a negative link
between CDS spreads and market price risk, and a positive link between CDS spreads and market volatility risk.
Taking into account simultaneously the dependence of CDS spreads relative to both market price and market
volatility channels allows for a better assessment of the correlation risk between the credit and the stock market.
2 Data and stylized features
We introduce the data under consideration and corresponding statistical patterns.
554

2.1 Data
We consider two categories of data among which U.S. stock market indexes and credit default swap data focusing
on both North America and emerging markets. Our daily data consist of closing quotes extracted from Reuters, and
ranging at most from September 28 th 2007 to March 24 th 2010, namely 618 observations per data series. With
regard to the first category of data, we consider the logarithmic returns of the Standard & Poor’s 500 stock market
index in basis points and the level of the CBOE implied volatility index. Specifically, those two indexes are
considered to be a proxy of the two complementary dimensions of market risk, namely the market price risk and the
market volatility risk (see Gatfaoui, 2010). With regard to the second category of data, we consider the spreads of
Markit credit default swap indexes, or equivalently, the spreads of credit derivatives indexes that we name Markit
CDX spreads. Those CDX indexes are split into two groups among which one set of spreads focuses on reference
entities domiciled in North America and the other one relates to reference entities domiciled in emerging markets
(see table 1). In particular, the CDXEM index focuses on sovereign entities whereas the CDXED relates to corporate
and sovereign entities. Moreover, the crossover index accounts for potential rating divergences between Standard &
Poor’s and Moody’s rating agencies. Furthermore, the CDX spreads 1 under consideration are expressed in basis
point and consist of the mid-market quotes on individual issuers. Incidentally, CDXEM spread data range from
February 1 st 2008 to March 24 th 2010, namely 538 observations per data series.
2.2 Stylized features
CDS label Detail about reference entities and indices
CDXEM Emerging Market
CDXED Emerging Market Diversified
CDXHY North America Investment Grade High Yield
CDXHB North America Investment Grade High Yield and B-rated
CDXBB North America Investment Grade High Yield and BB-rated
CDXIG North America Investment Grade
CDXIV North America Investment Grade High Volatility
CDXXO North America Crossover
SP500 Standard & Poor’s 500 stock index
VIX CBOE Implied Volatility Index
Table 1: Markit CDS indexes and stock market indices
We focus on the link prevailing between CDX spreads changes on one side, and changes in both SP500 returns as
well as VIX level. For this prospect, we first control for an existing link based on the non parametric Kendall and
Spearman correlation coefficients (see table 2). 2
Spread
Kendall correlation with
Spearman correlation with
SP500 VIX SP500 VIX
CDXEM -0.2676 0.4200 -0.3634 0.5644
CDXED -0.0916 0.2191 -0.1329 0.3114
CDXHY -0.2423 0.4077 -0.3369 0.5627
CDXHB -0.1382 0.3038 -0.1949 0.4246
CDXBB -0.1409 0.2861 -0.1999 0.4042
CDXIG -0.2887 0.4196 -0.4072 0.5779
1
The spreads are computed against corresponding LIBOR rates. The reader is invited to consult Markit Corporation’s website at
http://www.markit.com for further information.
2
CDS spread changes as well as market indexes exhibit a skewness and a positive excess kurtosis, underlining then their asymmetric behavior
over time.
555

CDXIV -0.2107 0.3484 -0.2975 0.4887
CDXXO -0.1772 0.2749 -0.2584 0.3912
Table 2: Kendall and Spearman correlations between CDX spread changes and changes in both SP500 and VIX
The obtained correlation estimates emphasize the significance of the correlation between the CDS market and the
U.S. stock market. As expected, the link between CDS spreads and market price is negative whereas the link
between CDS spread and market volatility is positive. Such a pattern illustrates the well-known volatility feedback
pattern, which was formerly introduced by Black (1976).
Focusing on the dependencies between CDX spreads and market indexes, we then investigate graphically the
existence of such links. For this prospect, we plot the CDX spread changes against changes in SP500 index on one
side, and changes in VIX implied volatility index on the other side (see figure 1 for example). The plots exhibit
clearly linkages between CDX spreads and both market price and market volatility. However, such linkages are
asymmetric. Moreover, we notice clearly differences between the dependence structure of CDX spread changes with
respect to SP500 changes, and the dependence structure with respect to VIX changes.
Figure 1: Dependence structures of CDXHY spreads with both SP500 and VIX indexes
As a result, there exist negative linkages between CDX spread changes and SP500 return changes, and positive
linkages between CDX spread changes and VIX changes. Such linkages reveal to be asymmetric and the two types
of dependence structures of CDX spread changes relative to SP500 return changes and VIX changes respectively
exhibit noticeable differences.
3 A Multivariate Copula Application
The previous stylized facts advocate the use of an appropriate statistical tool to handle simultaneously the
dependence structure between the CDS market and the two components of market risk, namely market price and
market volatility risks. For this purpose, we introduce the three-dimension copulas under consideration, the
corresponding data fitting process and the selection criterion of the best copula model.
3.1 Copulas
Copulas are a useful tool to model multivariate dependence structures (Cherubini et al., 2004; Durrleman et al.,
2000; Embrechts et al., 2003; Genest et al., 1995; Joe, 1997; McNeil et al., 2005; Nelsen, 1999; Sklar, 1973). They
present the advantage of not necessarily having to determine the distribution function of each of the variable under
consideration. Hence, it is possible to specify the global dependence structure without knowing the marginals (i.e.
univariate distribution function) of each variable under consideration. As a consequence, the corresponding model
risk is minimized. As an example, figure 2 plots the empirical copula function which describes the bivariate
dependence structure of CDXHY spread changes with respect to SP500 changes on one side, and VIX changes on
the other side. The observed empirical behavior can easily be linked to the theoretical behavior of some well-known
copula representations (Cherubini et al., 2004; Joe, 1997; Nelsen, 1999).
556

We focus on the three-dimension copula representations of the dependence structures of CDX spread changes
and the changes in the two market risk channels. Under such a setting, we face the well-known curse of
dimensionality, which represents the trade-off between the dimension of our setting (i.e. a three-dimension setting),
the number of parameters for each considered copula representation, and finally the number of available data points.
Given that statistics often advocate parsimonious models, we’ll focus on a specific set of Archimedean and Elliptical
copulas (see table 3). In particular, the Frank and Gaussian copulas exhibit non tail dependence, namely no link
between the variable’s extreme values. However, the Student T copula exhibits a symmetric left- and right-tail
dependence. Differently, the remaining copulas exhibit asymmetric tail dependences. In particular, the Clayton
copula exhibits a lower tail dependence whereas the Gumbel copula exhibits an upper tail dependence.
Figure 2: Empirical bivariate copula functions of CDXHY spread changes
Copula
Clayton
Attribute Parameters
Frank
Gumbel
Archimedean Correlation �
Gaussian
Student T
Elliptical
Correlation matrix
Degree of freedom �, Correlation matrix
Table 3: Three-dimension copulas and characteristics
Each of the three dimensions of our multivariate copula framework relates respectively to CDX spread changes,
SP500 return changes and finally VIX changes from one day to another. This way, the relationship between CDX
spreads and the two market dimensions are simultaneously accounted for. For any positive correlation parameter �
and u1, u2, u3 in [0,1], the Clayton copula writes:
C
�u , u , u ; �
1
2
3 � �
1
� � �
��
��
��
u � u � u � 2
For any correlation matrix � and u1, u2, u3 in [0,1], the Gaussian copula writes:
C
1
� � � � �
� �
� t �1
u , u , u ; � � exp � � � � 1�
1
2
3
�
1
1
2
where � and � -1 are a three-dimension matrix and its inverse respectively, |�| is the determinant of the correlation
matrix, � is the vector of the inverse standard univariate Gaussian cumulative distribution function applied to each
element u1, u2, u3, and finally � t is the transposed vector of �. We also introduce a three-dimension vector 1 of unit
numbers.
557
2
�
�
1
1
2
3

For any positive correlation parameter � and u1, u2, u3 in [0,1], the Frank copula writes:
C
��
��u1
��u2
��u3
1
� � �e �1��
e �1��
e �1���
u1,
u2,
u3;
� � � ln�1
�
�
��
2
� ��
�e �1�
�� For any positive correlation parameter � and u1, u2, u3 in [0,1], the Gumbel copula writes:
C
�
�
�
1
� �
� � �
1
2
3 �
�
�
�
�u , u , u ; � � � exp � �� lnu
� � �� lnu
� � �� lnu
�
1
2
3
For any correlation matrix �, degree of freedom � and u1, u2, u3 in [0,1], the Student T copula writes:
�� � 3��
��
��
� 1 t �1
�
��
����
��
�1
� � � � �
1 � 2 � �
�
� �
� � 2 ��
�
C u1,
u2,
u3;
�,
� � 1
�
2
3
�
3
2
� �� � 3��
��
� � �n
�
���
��
��
����
1�
��
� � 2 ��
� 2 � n�1
� � �
where � and � -1 are a three-dimension matrix and its inverse respectively, |�| is the determinant of the correlation
matrix, � is the Gamma function, � is the vector (�1, �2, �3) of the inverse univariate Student 3 cumulative distribution
function applied to each element u1, u2, u3, and finally � t is the transposed vector of �.
3.2 Estimation and selection
We first estimate the copula parameters while running a maximum likelihood estimation methodology (MLE).
However, we correct for possible parameters’ uncertainty while applying a parametric bootstrapping technique in
order to conform to the related MLE asymptotics (i.e. bootstrap MLE; Chen and Fan, 2006; Chernick, 1999;
Davison and Hinkley, 2006; Efron, 1979; Simon, 1997; Varian, 2005). The parametric bootstrap, which is also a
resampling method, allows for assigning an accuracy measure to parameter estimates. Indeed, parameter uncertainty
usually yields the under- or overestimation of model parameters. Correcting for uncertainty and sticking to MLE
assumptions allow therefore for getting more accurate estimates and therefore sounder risk assessment. Then, our
selection process of the most appropriate copula representation relies on the information criterion principle (i.e.
selection tool). In particular, we consider the Akaike, Schwarz and Hannan-Quinn information criteria. Those
information criteria encompass two components, which are the forecast error committed by the model and number
of estimated unconstrained parameters (Akaike, 1974; Lütkepohl, 2006; Hannan and Quinn, 1979; Schwarz, 1978).
The model selection rule requires minimizing the information criterion. By doing so, the selection process targets an
accurate and parsimonious model (i.e. reducing the potential errors and misestimation problems).
The negative Kendall correlation between CDX spreads and SP500 return changes is incompatible with the
Clayton copula representation. Moreover, the obtained parameter estimates for Frank copula are also incompatible
with the corresponding theoretical specification. As a result, we display only the chosen information criteria for the
remaining copulas (see tables 4, 5 and 6). Amongst the range of representations under consideration, the best copula
or the optimal three-dimension copula estimation is that one which minimizes at least one (when not all) the
information criteria previously mentioned, namely Akaike, Schwarz and Hannan-Quinn information criteria.
According to tables 4 to 6, the optimal copula representation consists of the Student T copula for all CDX spreads
under consideration, which implies a symmetric tail dependence of CDX spreads with respect to market risk
channels.
3 This is a Student distribution with � degree(s) of freedom.
558
3
� �3
�
2
�1
�
2

Spread
Information criterion
Akaike Schwarz Hannan-Quinn
CDXEM 2.00747665 6.28599811 3.67664928
CDXED 2.00650408 6.42486904 3.72035251
CDXHY 2.00650411 6.42486907 3.72035253
CDXHB 2.00650411 6.42486907 3.72035254
CDXBB 2.00650407 6.42486903 3.72035250
CDXIG 2.00650409 6.42486905 3.72035252
CDXIV 2.00650410 6.42486906 3.72035253
CDXXO 2.00650415 6.42486911 3.72035258
Spread
Table 4: Information criteria for Gumbel copula estimation
Information criterion
Akaike Schwarz Hannan-Quinn
CDXEM -496.82 -484.01 -491.84
CDXED -409.29 -396.05 -404.16
CDXHY -592.34 -579.11 -587.22
CDXHB -472.05 -458.82 -466.93
CDXBB -444.55 -431.31 -439.43
CDXIG -616.53 -603.30 -611.41
CDXIV -517.10 -503.86 -511.98
CDXXO -429.38 -416.15 -424.26
Spread
Table 5: Information criteria for the Gaussian copula estimation
Information criterion
Akaike Schwarz Hannan-Quinn
CDXEM -709.90 -692.83 -703.27
CDXED -505.90 -488.27 -505.90
CDXHY -704.44 -686.81 -704.44
CDXHB -578.60 -560.97 -578.60
CDXBB -517.75 -500.11 -517.75
CDXIG -735.21 -717.57 -735.21
CDXIV -604.00 -586.37 -604.00
CDXXO -486.81 -469.17 -486.81
Table 6: Information criteria for the Student T copula estimation
Further, table 7 displays the corresponding Student T parameter estimates, namely the elements of the
correlation matrix � and the number � of degrees of freedom.
559

Spread
Correlation with Correlation between Degree of
SP500 VIX SP500 and VIX freedom
CDXEM -0.1185 0.3448 -0.6216 3
CDXED 0.0116 0.2182 -0.6072 4
CDXHY -0.2580 0.5142 -0.6223 4
CDXHB -0.1185 0.3448 -0.6216 3
CDXBB -0.0589 0.2280 -0.5775 4
CDXIG -0.3466 0.5648 -0.6061 3
CDXIV -0.2893 0.4166 -0.5967 3
CDXXO -0.1655 0.2379 -0.6068 5
Table 7: Parameter estimates of the three-dimension Student T copula
Apart from CDXED spreads, results conform to empirical facts so that:
1. the correlation between the changes in CDX spreads and SP500 returns is negative,
2. the correlation between the changes in CDX spreads and VIX levels is positive,
3. the correlation between the changes in SP500 returns and VIX levels is negative.
The positive correlation between the changes in CDXED spreads and SP500 returns is probably due to the curse
of dimensionality concern. Indeed, we have more than 100 data points less as compared to the other CDX spread
time series. We hope to solve this issue while completing our time series as soon as possible. Finally, the obtained
correlation matrix elements are slightly different from the previous Kendall correlation estimates. Indeed, the
average differences between the copula-based correlation and the Kendall counterparts are 0.0267 and 0.0237 with
respect to SP500 returns and VIX levels. In the same way, the average absolute differences between those two types
of correlation estimates are 0.0648 and 0.0665 with respect to SP500 returns and VIX levels.
4 Conclusion
In this paper, we focused on the dependence structures between CDX spread changes on one side, and changes in
both SP500 returns and VIX index on the other side. We empirically exhibited the asymmetric nature of each
bivariate dependence structure, namely de dependence structures between CDX spreads and SP500 returns and
between the CDX spreads and VIX index. By the way, we also emphasized the differences between those two types
of bivariate dependence structures, which we handled simultaneously within a three-dimension copula analysis.
Balancing the curse of dimensionality with a parsimonious modeling framework, we selected three Archimedean
copulas and two classic elliptical copulas in order to test for various tail dependencies.
The estimation process and the selective information criterion statistics exhibited the Student T dependence
structure as an optimal three-dimension copula representation. Therefore, we have to cope with symmetric tail
dependencies between CDX spreads and the two market risk channels above-mentioned. Additionally, we are
therefore able to realize a more accurate and global credit risk scenario analysis in the light of both market trend and
market volatility levels. Hence, the three-dimension copula framework is a useful tool for risk
monitoring/management and risk reporting prospects under Basel 3. Moreover, the natural extension of our study
consists of a scenario analysis describing the impact of the market risk channels on the evolution of CDS spreads.
Such a framework is useful for value-at-risk or even stressed value-at-risk implementations as well as related
scenario analysis.
560

5 References
Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control
19(6), 716–723.
Black, F. (1976). Studies of stock price volatility changes. Proceedings of the 1976 Meetingsof the American
Statistical Association, Business and Economical Statistics Section, 177–181.
Chen, X., & Fan, Y. (2006). Estimation of copula-based semiparametric time series models. Journal of
Econometrics 130(2), 307–335.
Chernick, M. R. (1999). Bootstrap Methods, A practitioner's guide. Wiley Series in Probability and Statistics.
Cherubini, U., Luciano, E., & Vecchiato, W. (2004 ). Copula Methods in Finance. Chichester: Wiley.
Davison, A. C., & Hinkley, D. (2006). Bootstrap Methods and their Application. 8 th edition, Cambridge: Cambridge
Series in Statistical and Probabilistic Mathematics.
Dupuis, D., Jacquier, E., Papageorgiou, N., & Rémillard, B. (2009). Empirical evidence on the dependence of credit
default swaps and equity prices. Journal of Futures Market 29(8), 695-712.
Durrleman, V., Nikeghbali, A., & Roncalli, T. (2000 ). Which copula is the right one? Technical Report,
Operational Research Group of Crédit Lyonnais, Paris.
Efron, B. (1979). Bootstrap Methods: Another Look at the Jackknife. Annals of Statistics 7(1), 1-26.
Embrechts, P., Lindskog, F., & McNeil, A.J. (2003 ). Modeling dependence with copulas and applications to risk
management. In S. Rachev (Ed.), Handbook of Heavy Tailed Distributions in Finance (pp. 329–384). North-
Holland: Elsevier.
Gatfaoui, H. (2010). Investigating the dependence structure between credit default swap spreads and the U.S.
financial market. Annals of Finance, 6(4), 511-535.
Genest, C., Ghoudi, K., & Rivets, L.-P. (1995 ). A semi-parametric estimation procedure of dependence parameters
in multivariate families of distributions. Biometrika 82(3), 543–552.
Hannan, E. J., & B. G. Quinn (1979). The determination of the order of an autoregression. Journal of the Royal
Statistical Society B 41, 190–195.
Joe, H. (1997). Multivariate Models and Dependence Concepts. Monographs on Statistics and Applied Probability,
Vol.73. London: Chapmann & Hall.
Lütkepohl, H. (2006). Palgrave Handbook of Econometrics. Volume 1: Econometric Theory, Chapter Vector
Autoregressive Models, pp. 477–510. Houndmills: Palgrave Macmillan.
McNeil, A., Frey, R., & Embrechts, P. (2005 ). Quantitative Risk Management: Concepts, Techniques and Tools.
Princeton: Princeton University Press.
Nelsen, R.B. (1999 ). An Introduction to Copulas. Lectures Notes in Statistics, 139. New York: Springer.
Norden, L., & Weber, M. (2009). The co-movement of credit default swap, bond and stock markets: An empirical
analysis. European Financial Management 15(3), 529-562.
Scheicher, M. (2009). The correlation of a firm’s credit spread with its stock price: Evidence form credit default
swaps. In G. N. Gregoriou (Ed.), Stock Market Volatility (Chap. 21, pp. 405-419). London: Chapman &
Hall/CRC Finance.
Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics 6(2), 461–464.
Simon, J. L. (1997). Resampling: The New Statistics. Resampling Stats.
Sklar, A. (1973 ). Random variables, joint distribution functions and copulas. Kybernetika 9(6), 449–460.
Varian, H. (2005). Bootstrap tutorial. Mathematica Journal 9(4), 768-775.
Vaz de Melo Mendes, B., & Kolev, N. (2008). How long memory in volatility affects true dependence structure.
International Review of Financial Analysis 17(5), 1070-1086.
561

MODELING OF LINKAGES BETWEEN STOCK MARKETS INCLUDING THE EXCHANGE RATE
DYNAMICS
Malgorzata Doman, Associate Professor, Poznan University of Economics, Department of Applied Mathematics
Al. Niepodleglosci 10, Poznań, Poland, E-mail:malgorzata.doman@ue.poznan.pl
Ryszard Doman, Associate Professor, Adam Mickiewicz University in Poznan, Faculty of Mathematics and Computer Science,
Umultowska 87, 61-614 Poznan, Poland, E-mail: rydoman@amu.edu.pl
Abstract. The analysis of linkages between national stock markets is usually based on models describing dependencies between returns
on stocks or indices. Some papers concerning the subject present results for pure quotations and the other consider the stock
market data denominated in one currency (usually in the US dollar). In the paper, we address the question of how the introducing of
the exchange rate dynamics into a model affects the dependence analysis. We apply and compare two different ways of tackling the
problem. The first one consists in denominating the analyzed quotations in one currency (USD or euro). The second deals with a direct
introducing the exchange rate into a model for the dependence structure. Our analysis is based on the return series on selected stock
indices from the period 1995-2010. To describe the dependence structure, we apply dynamic copulas. Such approach allows us to separate
the dynamics of dependence from the volatility dynamics.
1. Introduction
The knowledge about linkages between stock markets is of importance in risk management and building investment
strategies. Moreover, it is crucial for understanding the nature of global financial market. So, it is quite natural that
there exist many papers dealing with this problem. Most of them belong to the contagion literature. The most popular
approach here is to denominate the indices (or other stock market quotations) in local currencies (Eun and Shin
1989, Koutmos 1992, Theodossiou and Lee 1993, Wong et al. 2004). The next popular choice is denomination in the
US dollar (e.g. Karolyi and Stulz 1996, Rodriguez 2007)). There exist analyses performed both in a local currency
and the US dollar (e.g. Lee et al. 2001). Chen and Poon (2007) use local currency for indices in the case of developed
markets and for emerging market they use US dollar denominated indices. Veiga and McAleer (2004) remarked
that the use of the US dollar as a common currency is a complicating factor. This is because in such situation
the US market is always included in the empirical analysis. Changes in the US dollar are largely influenced by
changes in US fundamentals, which also drive financial returns. Thus, it is likely that some of the co-movements
observed among returns in different markets expressed in a common currency are caused by changes in the fundamentals
driving the US dollar exchange rate. However, their results (Veiga and McAleer 2004) based on quite extensive
analysis of the sensitivity of spillover effects on denomination show that the denomination has no significant
impact on the results.
In the paper, we ask how introducing the exchange rate dynamics influences the dynamics of linkages between
stock indices. We consider dependencies between the S&P500 index and two European indices, the DAX and the
WIG20 (the main index of the Warsaw Stock Exchange). The analysis of linkages is performed by means of a DCCcopula
model. We estimate dynamic copula correlations between the indices denominated in local currencies and in
chosen alternative currencies. The aim of the presented investigation is to analyze the sensitivity of the dynamic
copula correlation estimates to the denomination of indices in alternative currencies. In the case of the S&P500 and
the DAX the considered currencies are the US dollar and the euro. The analysis for the S&P500 and the WIG20
includes denomination in the US dollar, the euro and the Polish zloty. Moreover, for both pairs of the indices we
calculate the dynamic copula correlations based on a three-dimensional DCC-copula model estimated jointly for the
indices denominated in local currencies and the corresponding exchange rate (USD/EUR in the case of SP500-DAX
and USD/PLN for SP500-WIG20).
2. DCC-copula models
Modeling the dependencies between financial returns is a difficult task because of special properties of these series.
Typical return series usually exhibit conditional heteroskedasticity, different types of asymmetries, and structural
breaks which strongly influence estimation results for models of the dependence structure. Moreover, the dynamics
of dependencies significantly changes in time. For example, it is well documented in many studies that dependence
between returns on different assets is usually stronger in bear markets than in bull markets (Ang and Bekaert 2002,
Ang and Chen 2002, Patton 2004). This example of asymmetric dependence in financial markets is of great impor-
562

tance for portfolio choice and risk management. The main problem connected with this phenomenon is, however,
that from the theoretical point of view, the mentioned asymmetry cannot be produced by a statistical model for the
returns that assumes an elliptical multivariate conditional distribution, and thus applying the linear correlation is not
justified. An alternative concept that allows for modeling the dependence in general situation is copula. Roughly
speaking, a d-dimensional copula is a mapping : [ 0,
1]
� [ 0,
1]
d
C from the unit hypercube into the unit interval
which is a distribution function with standard uniform marginal distributions.
Assume that X � X , �,
X ) is a d-dimensional random vector with joint distribution F and marginal distri-
butions i
( 1 d
F , i , d , � 1 � . Then, by a theorem by Sklar (1959), F can be written as
F x , �, x ) � C(
F ( x ), �,
F ( x )) . (1)
( 1 d
1 1 d d
The function C is unique if Fi are continuous. Otherwise, C is uniquely given by
C(
u , � , u
( �1)
( �1)
) � F(
F ( u ), �,
F ( u )) , (2)
1 d
1 1 d d
�1
for u i �[
0,
1]
, where Fi ( u)
� inf{ x : Fi
( x)
� ui}
. In that case, C is called the copula of F or of X. Since the marginals
and the dependence structure can be separated, it makes sense to interpret C as the dependence structure of
the vector X. We refer to Patton (2009) and references therein for an overview of financial time series applications
of copulas. There one can also find more information about advantages and limitations of copula-based modeling.
�
The simplest copula is defined by C ( u1
, �, ud
) � u1
��
� ud
, and it corresponds to independence of marginal
�
distributions. The next two important examples are C u , �, u ) � min( u , �,
u ) , and, in the two-dimensional
( 1 d
1 d
�
case, C ( u , u ) � max( u � u �1,
0)
. The first corresponds to comonotonicity or perfect dependence (one variable
i
j
i
j
can be transformed almost surely into another by means of an increasing map), and the second, to countermonotonicity
or perfect negative dependence of the variables X i and X j (one variable can be transformed almost surely
into another by means of a decreasing map). In the empirical part of this paper we will use the Student t copula. It is
defined as follows:
St
d �1
�1
C�
, R ( u1,
�, u d ) � t�
, R ( t�
( u1
), �,
t�
( u d )) , (3)
t denotes the d-dimensional Student’s t distribution with � degrees of freedom and correlation matrix R,
where � , R
and t� stands for 1-dimensional Student’s t distribution with � degrees of freedom. In the bivariate case we will use
the notation
St
C� , � , where � stands for correlation coefficient.
The density associated to an absolutely continuous copula C is a function c defined by
c
� C(
u , �,
u )
d
( u1,
�,
ud
) � 1 d
�u1��u
d
. (4)
For an absolutely continuous random vector, the copula density c is related to its joint density function h by the
following canonical representation:
f x , � x ) � c(
F ( x ), �,
F ( x )) f ( x ) � f ( x ) , (5)
( 1 d
1 1 d d 1 1 d d
where d F F , , 1 � are the marginal distributions, and f 1, , f d
� are the marginal density functions.
In the case of non-elliptical distributions, measures of dependence that are more appropriate than the linear correlation
coefficient are provided by two important copula-based tools known as Kendall’s tau and Spearman’s rho
(Embrechts et al. 2002). Since the dynamics of Kendall’s tau can be easily derived for the results presented in this
paper, we recall suitable definitions. If ( X , Y ) is a random vector and )
~
( X
~
, Y is an independent copy of ( X , Y ) then
Kendall’s tau for ( X , Y ) is defined as
� ( X , Y ) � P{(
X � X
~
)( Y �Y
~
) � 0}
� P{(
X � X
~
)( Y � Y
~
) � 0}.
(6)
Thus Kendall’s tau for ( X , Y ) is the probability of concordance minus the probability of discordance. If
( X , Y ) is a vector of continuous random variables with copula C, then
For the Student t copula
� ( X , Y ) � 4��
C(
u,
v)
dC(
u,
v)
�1
. (7)
2
[ 0,
1]
2
�
St
C� , � , Kendall’s tau equals arcsin( �)
563
.

A very important concept connected with copula, relevant to dependence in extreme values, is tail dependence
(Nelsen 2006). If X and Y are random variables with distribution functions F and G then the coefficient of upper tail
dependence is defined as follows
�1
�1
� lim P(
Y � G ( q)
| X � F ( q))
, (8)
U � �
q�1
provided a limit � �[
0,
1]
exists. Analogously, the coefficient of lower tail dependence is defined as
U
�1
�1
� lim P(
Y � G ( q)
| X � F ( q))
, (9)
L � �
q�0
provided that a limit �L �[
0,
1]
exists. If � U �(
0,
1]
( � L �(
0,
1]
), then X and Y are said to exhibit upper (lower) tail
dependence. Upper (lower) tail dependence quantifies the likelihood to observe a large (low) value of
large (low) value of X. The coefficients of tail dependence depend only on the copula C of X and Y:
Y given a
C(
q,
q)
� L � lim � ,
q�
0 q
Cˆ
( q,
q)
� U � lim �
q�0
q
(10)
where Cˆ ( u,
v)
� u � v �1�
C(
1�
u,
1�
v)
. For the Student t copula
dence are both equal to 2 1�� ( � �1)(
1�
�)
/( 1�
�)
�
t (see McNeil et al. 2005).
� �
St
C� , � , the coefficients of upper and lower depen-
Introduced by Patton (2004), the notion of conditional copula allows to apply copulas to modeling the joint distribution
of t r conditional on information set �t �1
, where r � ( 1,
, , , ) �
t r t � rd
t is a d-dimensional vector of financial
returns. In this paper we consider the following general conditional copula model
r ~ F ( � | � ), � , r | � ~ F ( � | � ) , (11)
1, t | � t�1
1,
t t�1
d , t t�1
d , t t�1
t | � t�1
~ Ft
( � | � t �1
)
t ( rt
| � t�1
) � Ct
( F1,
t ( r1,
t | � t�1
), , Fd
, t ( rd
, t | � t�1
) | � t�1
r , (12)
F � )
(13)
where the set t � includes the up to time t information on the returns on both considered financial assets, and Ct is
the conditional copula linking the marginal conditional distributions. Further, we assume that
r � μ � y , μ E r | � ) , (14)
t t t
y i,
t � i,
t�
i,
t
t � ( t t�1
2
i,
t � var( ri
, t | �t
�1
� , � ) , (15)
� ~ iid Skew_
t(
0,
1,
� , � ) , (16)
i,
t
i i
where Skew _ t(
0,
1,
�,
�)
denotes the standardized skewed Student t distribution with � � 2 degrees of freedom, and
skewness coefficient � 0
r , , i � 1, �,
d , we fit
� (Lambert and Laurent 2001). To the marginal return series i t
ARMA-GARCH models with skewed Student’s t distributions for the 1-dimensional innovations.
When modeling the joint conditional distribution, the evolution of the conditional copula Ct has to be specified.
Usually (Patton 2004, 2006), the functional form of the conditional copula is fixed, but its parameters evolve
through time. In this paper, we follow that approach and apply the DCC model proposed by Engle (2002), extended
to Student t copulas. Thus in our DCC-t-copula model we assume that the conditional copula Ct is a Student t Copu-
la
t
C� , R t
such that
�1
2
�1
2
�diag�Q ��
Q �diag�Q ��
R t � t t
t , (17)
1 1
1
~ ~
Q t � ( 1 � � � � ) Q � � y � t�
y t�
� � Q t�
, (18)
where � � 0 , � � 0 , � � � � 1 , ~ yi
, t
�1
� t�
( y i,
t ) , i � 1, �,
d , and Q is the unconditional covariance matrix of t y~ .
3. The Data
In the paper we present results of analysis concerning the dependencies between the S&P500 and two European
indices, the DAX and the WIG20 (the main index of the Warsaw Stock Exchange). We choose the indices from two
neighbor countries which differ in the level of development. As it was mentioned in Introduction, the very common
approach in stock market linkages analysis is to investigate the indices of developed markets in local currency, and
that from emerging market, denominated in an alternative currency (mostly the US dollar). Our dataset contains the
quotations of the considered indices and the exchange rates EUR/ USD and USD/PLN. The quotation series were
obtained from the service Stooq. The period under scrutiny is from January 3, 1995 to December 11, 2009.
564

Since the patterns of non-trading days in national stock markets differ, for the purpose of modeling dependencies
the dates of observations for each pair of indices were checked and observations not corresponding to ones in
the other index quotation series were removed. The time series under scrutiny are percentage logarithmic daily returns
calculated by the formula
rt � 100(ln Pt
�ln
Pt
) , (19)
where Pt denotes the closing index value on day t .
The descriptive statistics of the analyzed return series are presented in Table 1. In Tables 2-3 we show insample
estimates of the unconditional correlations.
Table 1. Descriptive statistics of the analyzed return series
Index Mean Maximum Minimum Stand. Dev. Skewness Kurtosis
S&P500 0.0238 10.957 -9.4695 1.289 -0.1783 10.9195
S&P500 in EUR 0.0190 9.5946 -8.7688 1.4663 -0.2152 7.0311
S&P500 in PLN 0.0282 10.054 -9.1886 1.4320 -0.0718 8.1544
DAX 0.0276 10.797 -9.791 1.5877 -0.0635 10.9230
DAX in USD 0.0325 13.5020 -9,4710 1.6741 0.0517 8.5868
WIG20 0.0302 13.709 -14.161 1.9544 -0.1548 6.7066
WIG20 in USD 0.0262 14.995 -19.463 2.2717 -0.2531 8.2124
WIG20 in EUR 0.0212 16.368 -17.481 2.3220 -0.1743 8.6857
Table 2. S&P500 and DAX. Estimates of the unconditional correlation of the returns
S&P500 S&P500 in EUR
DAX 0.5590 0.5309
DAX in USD 0.5227
Table 3. S&P500 and WIG20. Estimates of the unconditional correlation of the returns
4. Empirical analysis of the stock market linkages
S&P500 S&P500 in EUR S&P500 in PLN
WIG20 0.2682 ----- 0.1212
WIG20 in USD 0.2824 -----
WIG20 in EUR ----- 0.2749
The course of presented analysis is as follows. We investigate the dependencies between the returns for two pairs of
indices: S&P500-DAX and S&P500-WIG20. In each case we model the dynamic copula correlations by means of
the DCC-t-copula model described in section 3. The pair S&P500-DAX is considered in local currencies, in the US
dollar, and in the euro. For S&P500-WIG20 we additionally take into account denomination in the Polish zloty.
Moreover, we estimate jointly the dynamic copula correlations for triples of returns: S&P500-DAX-EUR/USD and
S&P500-WIG20-USD/PLN. The advantage of copula models is that they allow to separate the dependence dynamics
from the volatility dynamics. The feedback between these two features causes many problems in traditional analyses
based on multivariate volatility models
The DCC-t-copula models are estimated using a two-step maximum likelihood approach. The first step includes
fitting a GARCH model to each return series (Laurent 2009). The types of fitted models differ depending on currency
used to denominate an index (Table 4). Next, the GARCH standardized residuals are transformed by means of
their theoretical cumulative distribution functions to obtain the series of data uniformly distributed on [0,1]. In the
second step the DCC-t-copula models are fitted to the transformed series. Thus we follow the method of inference
functions for margins (Joe and Xu 1996).
First observation coming from Table 4 is that the conditional mean an volatility dynamics is sensitive to denomination.
The DCC-t-copula parameter estimates are presented in Table 5. It is worth to notice that in the case of the
pair S&P500-WIG20 denominated in local currencies the dynamics of conditional copula correlations is very week.
Table 4. Types of fitted ARMA-GARCH models
Return
series
S&P500
S&P500
in EUR
S&P500
in PLN
DAX
DAX
in USD
WIG20
WIG20
in USD
WIG20
in EUR
ARMA (1,1) (0,2) (1,1) (2,2) (1,0) (0,1) (2,0) (0,0)
GARCH GJR-GARCH(1,2) GJR-GARCH(1,2) GARCH(1,1) FIAPARCH(1,1) GARCH(1,1) FIAPARCH(1,1) GJR-GARCH(1,2) FIGARCH(1,1)
Error
distribution
Skewed Student Skewed Student Skewed Student Skewed Student Skewed Student Student Student Student
565

Table 5. Parameter estimates for the fitted DCC-t-copula model (standard errors in parentheses)
S&P500 and DAX S&P500 and WIG20
Parameter
in local
S&P500-DAX in local
in EUR in USD in EUR in USD
currencies -USD/EUR currencies
in PLN
S&P500-DAX
-USD/EUR
�
0.0146
(0.003
0.0204 0.0185
(0.006) (0.003)
0.0175
(0.003)
0.0075
(0.008)
0.0109 0.0102
(0.004) (0.004)
0.0072
(0.002)
0.0125
(0.003)
�
0.9841
(0.004)
0.9750 0.9805
(0.008) (0.004)
0.9789
(0.004)
0.9900
(0.016)
0.9836 0.9878
(0.007) (0.005)
0.9871
(0.005)
0.9821
(0.005)
�
14.4538
(3.819)
13.189415.5688
(3.582) (4.665)
12.5504
(1.843)
14.6137
(3.710)
16.187914.9618
(4.616) (3.911)
11.3950
(2.286)
16.6111
(3.004)
Figure 1 shows a comparison of the dynamic copula correlations for the pair S&P500-DAX obtained in all considered
cases. The dynamics of the correlations is quite strong. The strongest dependencies are observed in the years
2001-2004 and 2008-2009. The values of the correlations calculated for the indices denominated in local currencies,
in the euro, and modeled jointly with the exchange rate EUR/USD are quite closed each other. Only in the case of
the denomination in the US dollar the correlation estimates are clearly lower. The mean levels of the estimated dynamic
copula correlations (Table 6) are significantly different and the highest mean is obtained in the case of the
dependencies between the indices and the exchange rate EUR/PLN modeled jointly. The null hypothesis about equal
means was tested using the Model Confidence Set procedure (Hansen et al. 2003, 2011, Hansen and Lunde 2007)
applied to the set of the dynamic copula correlations series.
Figure 1. SP500 and DAX. Dynamic copula correlations from DCC-t-copula model
The estimates of dynamic copula correlations obtained for the pair S&P500-WIG20 are much lower but show
similar pattern as in the previous case – the dynamics of the conditional copula correlations is strong but it does not
depend significantly on the choice of currency. The only exception concerns the clearly weaker dependencies in the
case of the indices denominated in the Polish zloty. The difference is more visible after Poland joining the EU. The
testing procedure, the same as in the previous considered case, indicated that the mean levels of the estimated dynamic
copula correlations (Table 6) are significantly different.
Table 6. Means of the dynamic copula correlation estimates
S&P500 and DAX WIG20
in local currencies 0.4975 0.2434
in USD 0.4339 0.2345
in EUR 0.4903 0.2711
modeled jointly with the exchange rate 0.4992 0.2455
in PLN 0.1325
566

Figure 2. SP500 and WIG20. Dynamic copula correlations from DCC-t-copula models
In Figures 3-10, we put together the level and volatility of the exchange rates and the differences of the dynamic
copula correlations estimated based on the series of return calculated for the indices denominated in the considered
currencies. There is no clear pattern visible. However, it seems that the dynamic copula correlation estimates are
mostly higher when the local currencies are applied. Introducing the corresponding exchange rate into a DCC-tcopula
model does not change the results significantly (Figure 7). The linkages measured for the indices denominated
in local currencies are usually stronger than in the other cases. The linkages of the indices denominated in the
US dollar seem to be weaker during the financial crises.
Figure 3. EUR/USD Figure 4. EUR/USD. Volatility estimate
Figure 5. S&P500-DAX. Differences between the dynamic copula
correlations calculated for the indices denominated in local currencies
and in the euro
567
Figure 6. S&P500-DAX. Differences between the dynamic copula
correlations calculated for the indices denominated in local currencies
and in the US dollar

Figure 7. S&P500-DAX. Differences between the dynamic copula
correlations calculated for the indices denominated in local currencies
and the dynamic copula correlations from the model fitted to
the triple of return series (S&P500-DAX-EUR/PLN)
Figure 9. S&P500 and DAX. Differences between the dynamic
copula correlations calculated for indices denominated in euro and
the dynamic copula correlations from the model fitted to the triple
of return series (S&P500-DAX-EUR/PLN)
Figure 8. S&P500-DAX. Differences between the dynamic copula
correlations calculated for the indices denominated in the euro and
in the US dollar
Figure 10. S&P500 and DAX. Differences between the dynamic
copula correlations calculated for the indices denominated in the
US dollar and the dynamic copula correlations from the model
fitted to the triple of return series (S&P500-DAX-EUR/USD)
The plots for the pair S&P500-WIG20 are shown in Figures 11-18. The highest mean level of the dynamic copula
correlations occurs in the case of denomination in the euro (Table 6). Generally, the results for the pair S&P-
WIG20 differ from those presented for S&P-DAX. First of all, the correlations calculated for the indices denominated
in local currencies are not systematically higher than the other. Denominating the indices in the Polish zloty
results in the lowest values of the correlation estimates. This is especially visible when one compares this case with
the case of denomination in the euro. In the period 1995-2003 the correlations calculated between the indices denominated
in the US dollar are lower than that in the euro. The situation changes in 2004 when the correlation estimates
for the indices denominated in the US dollar become higher. This effect probably can be connected with Poland’s
accession to the EU.
Figure 11. USD/PLN
568
Figure 12. USD/PLN. Volatility estimates

Figure 13. S&P500-WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in local
currencies and in the euro
Figure 15. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in local
currencies and in the Polish zloty
Figure 17. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in the
euro and the dynamic copula correlations from the model fitted to
the triple of return series (S&P500-WIG20-USD/PLN)
Figure 19. S&P500 and WIG20. S&P500 and WIG20. Differences
between the dynamic copula correlations calculated for the indices
denominated the US dollar and in the euro
569
Figure 14. S&P500-WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in local
currencies and in the US dollar
Figure 16. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in local
currencies and the dynamic copula correlations from the model
fitted to the triple of return series (S&P500-WIG20-USD/PLN)
Figure 18. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in the
US dollar and the dynamic copula correlations from the model
fitted to the triple of return series (S&P500-WIG20-USD/PLN)
Figure 20. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in the
US dollar and in the Polish zloty

Figure 21. S&P500 and WIG20. Differences between the dynamic
copula correlations calculated for the indices denominated in the
Polish zloty and the dynamic copula correlations from the model
fitted to the triple of return series (S&P500-WIG20-USD/PLN)
5. Conclusions
Figure 22. Differences between the dynamic copula correlations
calculated for the indices denominated in the euro and in the
Polish zloty S&P500 and WIG20.
The aim of the presented research was to determine how the dynamics of linkages between stock markets changes
under the impact of introducing the exchange rate dynamics into a model. We considered dependencies between the
S&P500 index and two European indices, the DAX and the WIG20. To analyze the stock indices linkages we use
DCC-t-copula models. The advantage of the applied approach is that it allows to separate the dynamics of linkages
from the volatility dynamics. The presented results are introductory and slightly ambiguous but generally show that
the impact of denomination or introducing the exchange rate directly into the model for dependencies is weak. The
patterns observed for the pairs S&P-DAX and S&P-WIG20 differ from each other. In the case of the DAX the linkages
measured for the indices denominated in local currencies are mostly stronger than the other. The highest mean
level of the dynamic copula correlations appears when the indices are denominated in local currencies and modeled
jointly with the exchange rate EUR/PLN. For the WIG20 index the highest mean level of the correlations appears
when the indices are denominated in the euro. Some impact of Poland joining the EU on the investigated dependencies
can be observed. Nevertheless, our findings seem to support the opinion about lack of significant impact of
denomination on the dynamics of stock indices linkages.
6. References
Ang A., Bekaert G. (2002), International Asset Allocation with Regime Shifts, Review of Financial Studies 15,
1137-1187.
Ang A., Chen J. (2002), Asymmetric correlations of equity portfolios, Journal of Financial Economics 63, 443-494.
Chen S., Poon S.-H. (2007), Modelling International Stock Market Contagion Using Copula and Risk Appetite,
MBS Working Paper Series, SSRN: http://ssrn.com/abstract=1024288
Engle R.F. (2002), Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional
heteroskedasticity models, Journal of Business and Economic Statistics 20, 339-350.
Eun C.S., Shim S. (1989), International Transmission of Stock Market Movements, Journal of Financial and Quantitative
Analysis, 24, 241-256.
Embrechts P., McNeil A., Straumann D. (2002), Correlation and Dependence in Risk Management: Properties and
Pitfalls, in: Risk Management: Value at Risk and Beyond, Cambridge University Press, Cambridge, 176-223.
Hansen P.R., Lunde A., Nason J.M. (2003), Choosing the Best Volatility Models: The Model Confidence Set Approach,
Oxford Bulletin of Economics and Statistics 65, 839-861.
Hansen P.R., Lunde A. (2007), MulCom 1.00. Econometric Toolkit for Multiple Comparisons, Available at:
www.hha.dk/~alunde/MULCOM/MULCOM.HTM
Hansen P.R., Lunde A., Nason J.M. (2011), The Model Confidence Set, Econometrica 79, 453-497.
Joe H., Xu J.J. (1996), The Estimation Method of Inference Functions for Margins for Multivariate Models, Technical
Report no. 166, Department of Statistics, University of British Columbia.
570

Karolyi G.A., Stulz R.M. (1996), Why do markets move together? An investigation of U.S.–Japan stock market
comovements, Journal of Finance 51, 951–986.
Koutmos G. (1992), Asymmetric Volatility and Risk Return Tradeoff in Foreign Stock Markets, Journal of Multinational
Financial Management, 2, 27-43.
Laurent S. (2009), Estimating and forecasting ARCH models using G@RCH TM 6, Timberlake Consultants Ltd, London.
Lambert P., Laurent S. (2001), Modelling financial time series using GARCH-type models with a skewed Student
distribution for the innovations, Institut de Statistique, Université Catholique de Louvain, Discussion Paper
0125.
Lee B., Rui O.M., Wang S.S. (2001), Information Transmission Between NASDAQ and Asian Second Board Market,
Journal of Banking & Finance 28, 1637–1670.
McNeil A.J., Frey A., Embrechts P. (2005), Quantitative Risk Management, Princeton University Press, Princeton.
Nelsen R.B. (2006), An Introduction to Copulas, Springer Science+Business Media, Inc., New York.
Patton A.J. (2004), On the Out-of-Sample Importance of Skewness and Asymmetric Dependence for Asset Allocation,
Journal of Financial Econometrics 2, 130-168.
Patton A.J. (2006), Modelling Asymmetric Exchange Rate Dependence, International Economic Review 47, 527-
556.
Patton A.J. (2009), Copula-Based Models for Financial Time Series, in: T. G. Andersen, R. A. Davies, J.-P. Kreiss,
and T. Mikosch, eds., Handbook of Financial Time Series, Springer, Berlin, pp. 767-785.
Rodriguez J.C. (2007), Measuring financial contagion: A copula approach, Journal of Empirical Finance 14, 401-
423.
Sklar A. (1959), Fonctions de répartition à n dimensions et leurs marges, Publications de l’Institut Statistique de
l’Université de Paris 8, 229-231.
Theodossiou P., Lee U. (1993), Mean and Volatility Spillovers across Major National Stock Markets: Further Empirical
Evidence, Journal of Financial Research, 16, 327-350.
Wong W.-K, Penm J., Terrel R.D, Lim K. (2004), The Relationship Between Stock Markets of Major Developed
Countries and Asian Emerging Markets, Journal of Applied Mathematics And Decision Sciences, 8, 201–218.
Veiga B., McAleer M. (2004), Testing the sensitivity of spillover effects across financial markets, in C. Pahl-Wostl,
S. Schmidt, A.E. Rizzoli and A.J. Jakeman (eds.), Complexity and Integrated Resources Management: Transactions
of the International Conference on Environmental Modelling and Software, Osnabrueck, Germany,
published by iEMSs, Manno, Switzerland, 1523-1529
571

PROPAGATION OF SHOCKS IN GLOBAL STOCK MARKET: IMPULSE RESPONSE ANALYSIS IN A
COPULA FRAMEWORK
Ryszard Doman, Associate Professor, Adam Mickiewicz University in Poznan, Faculty of Mathematics and Computer Science,
Umultowska 87, 61-614 Poznan, Poland, E-mail: rydoman@amu.edu.pl
Malgorzata Doman, Associate Professor, Poznan University of Economics, Department of Applied Mathematics,
Al. Niepodleglosci 10, 60-697 Poznan, E-mail: malgorzata.doman@ue.poznan.pl
Abstract. As the effect of increasing speed of the information spreading and ease in the capital moving, interdependencies between
markets have strengthened during last few decades. Closer linkages between international markets can contribute to a faster transmission
of shocks and thus have a significant impact on decision-making in risk management. Therefore, the understanding of the nature
and the dynamics of international markets dependencies is of great importance in finance as well as in macroeconomic policies. To investigate
the process of shock transmission, we introduce a concept of impulse response function describing the time profile of the effect
of shocks on linkages modeled by means of dynamic copula model and measured using Kendall’s tau and the tail dependence
coefficients. This approach is applied to analyze the persistence of the impact of some important historically observed shocks on the
strength of linkages between the returns on selected stock indices during the crisis 2007-2009.
1. Introduction
Integration of financial markets observed in last decades is accompanied by strengthening of linkages between national
markets and, in particular, can contribute to a faster transmission of shocks and thus have a significant impact
on decision-making in risk management. Therefore, the understanding of the nature and the dynamics of international
markets dependencies is of great importance in finance, as well as in macroeconomic policies. The measures of
dependence usually used in financial markets modeling allow to describe and evaluate the strength of linkages but
are not very useful in determining the direction of impact of shocks on dependencies. The aim of this paper is to
analyze how shocks hitting one market influence the connection between that market and another one. For some
shocks (which could be attributed to one market) this should give us a possibility to determine the direction of linkages
between the markets.
To investigate the process of shock transmission, we introduce a concept of impulse response function describing
the time profile of the effect of shocks on linkages modeled by means of dynamic copula, and measured using
Kendall’s tau and the tail dependence coefficients. This approach is applied to analyze the persistence of the impact
of some important historically observed shocks on the strength of linkages between the returns on selected stock
indices during the crisis 2007-2009.
2. A Review of Impulse Response Functions
A simple linear model useful in modeling multivariate asset returns is the vector autoregressive model VAR(p)
r � � � r ��
� � r � u , (1)
where 0 � is a k-dimensional vector, t
nonsingular covariance matrix u � , and i
t � 0 1 t �1 1 t �1
t
u is a sequence of serially uncorrelated random vectors with mean zero and
�
process has the canonical MA representation
are fixed ( k �k ) coefficient matrices (Tsay 2005). Suppose that the
� �
rt � � � �iu
t�i
,
i�0
�0 � I k . (2)
Then the coefficient matrix �i can be considered as the effect of t u on the future observation rt � i . Because of this,
�i is usually referred to as the impulse response function of r t . If the components of ut are correlated, the interpretations
of the elements of �i is difficult. So one usually applies the Cholesky decomposition P P � �
u � to obtain the
uncorrelated innovations
�1
wt
� P ut
. In such a case, the decomposition (2) takes the form
� �
i�0
r � � � � w , (3)
t
i
572
t�i

where �i � �i
P is called the impulse response function of rt with orthogonal innovations r t .
The generalized impulse response function for not necessary linear time series was proposed by Koop, Pesaran
and Potter (1996). They consider a process of the form
Y F(
Y 1, � , Y ) � AV
, (4)
t � t�
t�
p
where F is a known function, Vt is a k-dimensional vector of IID disturbances, At is a ( k �k ) random matrix which
is a function of { Yt �1, � , Yt
� p}
, and the shocks have zero mean and finite variance. The generalized impulse response
is then defined as
GI n,
� , ) � E(
Y | � , � ) � E(
Y | � ) , (5)
Y ( t �t �1
t�
n t t �1
t �n
t �1
where �t�1 denotes the set containing information used to forecast t Y , and t
t
t
� is an arbitrary current shock. Thus the
generalized impulse response function is the difference between the mean of the response vector conditional on
history and a present shock, and the baseline expectation that only conditions on history. So in contrast to the linear
case, now the shape of the impulse responses depends on the history of the variables and may be different in each
time point. Moreover, if, for example, the impulses represent news arriving in a financial market, positive news may
have quite different effects than negative news.
The next step was done by Hafner and Herwartz (2006) in the framework of multivariate volatility models. For
a multivariate volatility model
r � � � y ,
t t t
t � E(
rt
| �t
�1)
yt
y�
t �t
� H t ) | ( 1 ,
� , (6)
E �
they consider responses of the conditional covariance matrix H t , given by the formula
Vt ( � 0 ) � E(vech(
H t ) | �0
, ��1)
� E(vech(
H t ) | ��1
) , (7)
where vech( ) denotes the operator that stacks the parts of columns of a symmetric matrix that start from the diagonal.
For many classes of multivariate GARCH models there exist explicit formulas for the impulse response functions
defined by (7). Covariance matrix, however, describes dependence properly only for elliptical distributions. In general
case, much better solution is to use copulas.
3. Copulas and Dependence Measures
Let X be an k-dimensional random vector with joint distribution F and marginal distributions F i , i , k , � 1 � . If the
functions Fi are continuous, then by a theorem by Sklar (1959) there exists a unique function C such that the following
decomposition holds:
F( x1,
�, xk
) � C(
F1
( x1),
�,
Fk
( xk
)) .
In the above situation the function C is given by the formula
(8)
�
�
C u , � , u ) � F(
F ( u ), �,
F ( u )) , (9)
( 1 k
1 1 k k
�
where Fi ( ui
) � inf{ xi
: Fi
( xi
) � ui}
for u i �[
0,
1]
. It follows from (9) that the function C can be seen as a multivariate
distribution with the one-dimensional marginal distributions being uniform on the unit interval. On the other
hand, since the marginals and the dependence structure in (9) are separated, it makes sense to interpret C as the dependence
structure of the vector X. The function C is called the copula of X (or F).
In this paper, we deal only with bivariate copulas, though many of the considered issues can be easily extended
to the general multivariate case (Nelsen 2006, Joe 1997). The simplest copula C � describes independent marginal
�
�
distributions and thus is defined by C ( u1
, u2
) � u1u2
. The next important examples are C ( u1,
u2
) � min( u1,
u2
) and
�
C ( u1,
u2
) � max( u1,
u2
�1,
0)
. The first corresponds to comonotonicity or perfect dependence (one variable can be
transformed almost surely into another by means of an increasing map), and the second to countermonotonicity or
perfect negative dependence (one variable can be transformed almost surely into another by means of a decreasing
map). A well known example is also the Gaussian copula defined as
Gauss 1 �1
C ( u1,
u2
) ( ( u1),
( u2
)
�
� � � � � � , (10)
573

where � � denotes the distribution of a standard 2-dimensional normal vector with the linear correlation coefficient
� , and � stands for the standard normal distribution function. In the empirical part of this paper we use the generalized
Clayton (GC) copula called also the BB1 copula (Joe 1997). It is defined by the following formula:
C� , �
GC
( u , u )
1
2
��
� ��
� [(( u1
�1)
� ( u2
�1)
)
� 1/
�
�1]
�1/
�
, (11)
where � � 0 , and � � 1 .
If a copula C has density c then almost everywhere in the interior of the unit square [ 0,
1]
� [ 0,
1]
,
2
� C(
u1,
u2
)
c(
u1,
u2)
� .
�u1�u2
(12)
In the case of continuous random vector, the copula density c is related to joint density function f by the following
canonical representation:
f ( x1,
x2
) � c(
F1(
x1),
F2
( x2
)) f1(
x1)
f 2(
x2
) , (13)
where Fi are the marginal distribution functions, f i are the densities ( i �1,
2 ).
It is well documented (McNeil, Frey and Embrechts 2005) that for non-elliptical distributions the linear correlation
coefficient is inappropriate measure of dependence, and often can be misleading. In that case, other measures
of dependence, based on the notion of concordance provide better alternatives. An example of such measures is
Kendall’s tau. Since it plays a significant role in our approach to measure impulse responses, we recall its definition.
If ( X 1, X 2 ) is a random vector and ( X
~
,
~
1 X 2 ) is an independent copy of ( X 1, X 2 ) , then Kendall’s tau for ( X 1, X 2 )
is defined as
� X , X ) � P{(
X � X
~
)( X � X
~
) � 0}
� P{(
X � X
~
)( X � X
~
) � 0}
. (14)
( 1 2
1 1 2 2
1 2 2 2
Thus, in fact, Kendall’s tau gives probability of concordance minus the probability of discordance. If ( 1, 2 ) X X is a
continuous random vector with copula C then its Kendall’s tau can be calculated by the formula
� X , X ) � 4��
C(
u , u ) dC(
u , u ) 1 .
( 1 2
1 2 1 2 �
2
[ 0,
1]
This means that Kendall’s tau depends only on the linking copula and, in particular, it can be considered to be a
measure of the degree of monotonic dependence between X 1 and X 2 . It can also be shown that �1 ��
( X , Y ) �1
.
Moreover, � ( X,
Y ) �1
is equivalent to comonotonicity, and � ( X , Y ) � �1
means countermonotonicity (Nelsen
Gauss
2
GC
2006). For the Gaussian copula C� , Kendall’s tau equals arcsin( �)
, and for the BB1 copula C� , � , it is known
�
to be equal to 1� 2 (( 2 ��
) � ) .
A very important concept connected with copula, relevant to dependence in extreme values, is tail dependence.
If X 1 and X 2 are random variables with distribution functions F1 and F 2 , then the coefficient of upper tail dependence
is defined as follows
�1
�1
� lim P(
X � F ( q)
| X � F ( q))
, (15)
U � �
q�1
provided a limit � �[
0,
1]
exists. Analogously, the coefficient of lower tail dependence is defined as
L
2
2
1
1
�1
�1
� lim P(
X � F ( q)
| X � F ( q))
, (16)
L � �
q�0
2
2
provided that a limit �L �[
0,
1]
exists. If � U �(
0,
1]
, then 1 X and X 2 are said to exhibit upper tail dependence. Analogously,
if � L �(
0,
1]
, then X 1 and X 2 are said to exhibit lower tail dependence. Upper (lower) tail dependence
quantifies the likelihood to observe a large (low) value of 2 X given a large (low) value of X 1 . The coefficients of
tail dependence depend only on the copula C of X 1 and X 2 :
C(
q,
q)
� L � lim � , q�0
q
(17)
Cˆ
( q,
q)
� U � lim � ,
q�0
q
(18)
where Cˆ ( u,
v)
� u � v �1
� C(
1�
u,
1�
v)
. For the Gaussian copula it holds that � � � 0 , while for the BB1 co-
pula, � � 2
L
�1
( �� )
and �
U
1
� 2 � 2
�
(McNeil, Frey and Embrechts 2005).
574
1
1
U � L

4. Stochastic Dynamic Copula Model
The standard definition of copula has been extended by Patton (2004) to the conditional case. The only complication
involved is that the conditioning set must be the same for both marginal distributions and the copula. In this
paper, however, contrary to Patton’s approach, we do not assume deterministic formulas for copula dynamics but
instead, similarly to Hafner and Manner (2008) and Almeida and Czado (2010), the dynamic copula parameters are
modeled as stationary autoregressive stochastic processes. More specifically, we consider the following general
model for 2-dimensional vector time series r r , r ) � of financial returns:
t � ( 1,
t 2,
t
1,
t | �t �1 1,
t 2,
t | �t �1 2,
t
rt | t �1
~ Ct ( F1,
t ( �),
F2,
t ( �)
| �t
�1;
θt
)
where �t �1
is the information set that includes all realized up to time �1
r ~ F ( �)
, r ~ F ( �)
, (19)
� . (20)
t returns on both considered financial
instruments, Ct is the conditional copula, and θt is some vector of parameters that evolves in time. We restrict
ourselves to two-parameter families of bivariate copulas, which means that θ � , � ) � . Moreover, we assume
t � ( 1,
t 2,
t
2
that θt � h(
γ t ) for some bijective transformation h of Euclidean plane R onto the range of θ t , and
γ � μ � diag � , � )( γ � � μ)
� diag(
� , � ) ε , (21)
t
( 1 2 t 1
1 2
1 � � 2 � .
where ~ ( , 2)
I 0 εt NID , | � 1 | � 1 , | � 2 | � 1 , � 0 , 0
In addition, when it is the case that for the applied copula family the parameters can be expressed by the upper
and lower tail dependence coefficients:
�1 � g1( �L
, �U
) , �2 � g2 ( �L
, �U
) ,
we propose that dynamics enters the model through them:
(22)
� � � ) ( 1�
exp( � )) , � � � ) ( 1�
exp( � )) . (23)
L,
t exp( 1,
t
1,
t
5. Posterior and Predictive Inference for the Copula Model
U , t exp( 2,
t
2,
t
We are going to use the introduced stochastic dynamic copula model to define impulse response functions describing
the impact of past shocks on Kendall’s tau, and lower and upper tail dependence coefficients. Because of the
presence of the latent processes γt and the necessity of prediction with very long horizon by a highly nonlinear
model, we perform a two-stage estimation. In the first step, univariate ARMA-GARCH models are fitted, and the
standardized residuals are transformed by means of the corresponding distribution functions into the series
u � ( 1,
, 2,
) �
t u t u t of uniform variates. In the second step, the series ut is considered as the data in the Bayesian posterior
and predictive inference for the copula model. Here we use the Metropolis within Gibbs sampler (see e.g. Geweke
2005) for the model parameters α � ( μ,
β,
ν)
� , as well as for the latent variables γ � ( γ : t � T � K)
� , and the
F
missing observations u � , u ) : T � t � T � K)
� . In our case, the objective of inference can generally be ex-
(( u1, t 2,
t
pressed as the posterior density of a vector of interest, ω , conditional on the observed data U � u , �,
u } :
�
�
p T
T
T
t
T
t
{ 1 T
( ω | U ) � p(
ω | θ,
U ) p(
θ | U ) dθ
, (24)
where the vector of interest, ω , will stand for future values of Kendall’s tau, and lower and upper tail dependence
coefficients.
The Bayesian approach enables us to calculate the responses basing not only on predictive moments but also on
predictive quantiles, and, in particular, the predictive median. This is because both the mean, E (ω)
, and the quantiles
of the distributions of the components of ω are Bayes actions corresponding to some loss functions.
m
The elements of a Bayesian decision problem (Berger 1985, Geweke 2005) are: an action a� A � R controlled
by the decisionmaker, a loss function L ( a,
ω)
(� 0)
depending on the action and a vector of interest,
and a distribution p ( ω | Y)
. A Bayesian action is any action a � A � R
ω � �
q
� R ,
m
ˆ which minimizes the posterior expected
loss E ( L(
a, ω)
| Y)
� � L(
a,
ω)
p(
ω | Y)
dω
. If
�
L ( a, ω)
� ( a � ω)
�Q(
a � ω)
, where Q is a positive definite matrix,
then the Bayes action aˆ � E(
ω | Y)
. If a � A � R , � � � � R , and L(a,ω) is a linear-linear function given by the
575

� , q �(
0,
1)
, then the Bayes action â is the qth quantile of
formula L( a,
) � ( 1�
q)(
a � �)
I ( ��,
a)
( �)
� q(
� � a)
I ( a,
�)
the posterior distribution of � .
For our purposes, it is important that, under some regularity conditions, Bayes actions can be approximated by
Markov Chain Monte Carlo (MCMC) sampling. More specifically, the following result holds (Geweke 2005, Amemiya
1985). Suppose that in the Markov chain C the sequence
( m) ( m)
( θ , ω ) is ergodic with invariant density
p ( θ | Y)
p(
ω | θ,
Y)
. Let L( a,
ω)
� 0 be a loss function defined on A � � and suppose that the risk function
� �
R ( a) � L(
a,
ω)
p(
θ | Y)
p(
ω | θ,
Y)
dθdω
(25)
� �
has a strict global minimum at
m
aˆ � A � R . Then under some additional regularity conditions, for any � � 0 ,
lim M ��
P(inf
a�A
�1
( a � aˆ
) �(
a � aˆ
) � � | C)
� 0 , where AM is the set of roots of M
M
( m)
�L(
a , ω ) �a
� 0 .
6. MCMC Implementation
M
�m �1
We perform inference for our stochastic dynamic copula model by using the OpenBUGS package (Thomas et al.
2006). This is a freely available software package. The acronym BUGS stands for the initials of the phrase “Bayesian
inference Using Gibbs Sampling”. The user of the package can specify a statistical model of (almost) arbitrary
complexity, by stating the relationships between related variables. The software includes an expert system for choosing
an effective MCMC scheme for each full conditional posterior distribution. The user then controls the execution
of the scheme and can modify it.
As concerns priors, we assume that the model parameters ( �1, �2,
�1,
�2
, �1,
� 2 ) are mutually independent. The
prior distributions are specified as follows:
� � ~ N(
0,
100)
i
i � �i
� 2 1
* *
� � , � ~ beta( 20,
2.5)
i
2
� � i ~ Inverse-gamma ( 2.
5,
0.
025)
In determining the number of iterations to achieve convergence to the stationary distribution, we apply diagnostics
implemented in the CODA (Best, Cowles and Vines 1996) and BOA (Smith 2005) R packages. They include the
Geweke, Gelman-Rubin, Raftery-Lewis, and Heidelberger-Welch diagnostics.
1. A Generalized Impulse Response Function Associated with a Bayesian decision problem
Let p( ω | Y,
θ)
be the posterior density of some vector of interest ω given a vector of unobservables θ and a vec-
tor of observables Y . For a given loss function L let t â and aˆ t�1
denote, respectively, the Bayes action corresponding
to the situation where the set of observables consists of the data U t�1
and a shock ν t , and of the data U t�1
only.
Suppose that the Bayes actions â t and aˆ t�1
are strict global minima of the corresponding risk functions (25). Under
the above assumptions, we define the generalized impulse response function as
GIR a , L,
ω ( ν t , Yt
�1
) � aˆ
t � aˆ
t�1
. (26)
In our investigation concerning the copula based dependence measures, as the vector of interest, ω T �n
, we consider
the future values of Kendall’s tau, and lower and upper tail dependence coefficients.
2. The Data
We investigated dependencies between the daily returns on several selected stock indices. In this paper, however, we
report only the results obtained for S&P500 and DAX, and two chosen shocks. We model the time series of percentage
logarithmic daily returns given by the formula
r (ln P � ln P )
t � 100 t t�1
where Pt is the value of stock index on day t.
The first considered shock is a negative one of September 29, 2008. It corresponds to the following event: “The
US House of Representatives rejected the Bush administration’s $700 billion emergency rescue plan. […] The Dow
576

Jones industrial average lost 777.68 points, its biggest single-day fall ever, easily beating the 684 points it lost on the
first day of trading after the Sept. 11, 2001, terrorist attacks. Crude oil futures closed down $10.52 in their biggest
decline since Jan 17, 1991, when the US opened strategic oil reserves during the first Gulf war”.
The other considered shock, positive one, from October 13, 2008 is connected with the event: “Stock markets
rejoiced after governments worldwide launched multibillion-dollar bailouts to shore up banks, and Britain called for
a new Bretton Woods agreement to reshape the world financial system. The US Central Bank said it would provide
unlimited dollars the European Central Bank, the Bank of England and the Swiss National Bank. Britain committed
£37 billion ($64 billion) to capitalize its big banks. Wall Street rebounded with the biggest stock rally since the
Great Depression. The DJIA rose 936 points to close at 9,387.61, its largest point gain ever and one of its largest
percentage increases” * .
The data used for estimation included in each case 600 observations preceding the shock. Thus in the case of the
first of the described shocks the data cover the period from April 28, 2006 to September 29, 2008, and in the second,
case the period under scrutiny is from May 5, 2006 to October 13, 2008.
3. Empirical Results
In this Section we present the results of the posterior and predictive inference applied to the considered data and
shocks. We fitted to the data a stochastic dynamic copula model described in section 4, based on the BB1 copula
defined by (11). Tables 1-4 contain the posterior summaries for the model parameters. In each case, the inference
was performed for the set of observation until the day preceding the shock, and for the dataset including shock. The
estimation results differ quite clearly after the shock including.
Table 1. September 29, 2008. Posterior summaries for the model parameters before the shock
Parameter Mean Std. Deviation 2.5% Median 97.5%
�1 -1.6600 0.3554 -2.3790 -1.5870 -1.1230
�2 -0.5094 0.1261 -0.7746 -0.5099 -0.2414
�1 0.7713 0.1246 0.4757 0.7926 0.9526
�2 0.7810 0.1157 0.4924 0.8041 0.9393
�1 0.1182 0.0475 0.0616 0.1062 0.2456
� 2
0.1179 0.0410 0.0632 0.1093 0.2228
Results obtained after 500 000 iterations (discarding 500 000 as burnt-in)
Table 2. September 29, 2008. Posterior summaries for the model parameters including the shock
Parameter Mean Std. Deviation 2.5% Median 97.5%
�1 -1.3380 0.3349 -2.0540 -1.3090 -0.7089
�2 -0.6012 0.1600 -0.8975 -0.6077 -0.2639
�1 0.8125 0.1267 0.5108 0.8374 0.9786
�2 0.8058 0.1184 0.5136 0.8323 0.9567
�1 0.1341 0.0656 0.0663 0.1190 0.2947
� 2
0.1286 0.0591 0.0647 0.1131 0.2883
Results obtained after 500 000 iterations (discarding 500 000 as burnt-in)
Table 3. October 13, 2008. Posterior summaries for the model parameters before the shock
* The both shock descriptions are taken from http://timelines.ws/
577

Parameter Mean Std. Deviation 2.5% Median 97.5%
�1 -1.7080 0.4739 -2.4530 -1.8120 -0.6341
�2 -0.5308 0.1677 -0.9146 -0.5121 -0.2453
�1 0.7739 0.1220 0.4849 0.7958 0.9437
�2 0.8023 0.1248 0.4975 0.8278 0.9653
�1 0.1184 0.0488 0.0614 0.1064 0.2513
� 2
0.1320 0.0663 0.0638 0.1137 0.3114
Results obtained after 500 000 iterations (discarding 500 000 as burnt-in)
Table 4. October 13, 2008. Posterior summaries for the model parameters including the shock
Parameter Mean Std. Deviation 2.5% Median 97.5%
�1 -1.4400 0.2417 -1.8650 -1.4550 -0.9399
�2 -0.5947 0.1332 -0.8882 -0.5844 -0.3568
�1 0.7739 0.1337 0.4533 0.7961 0.9570
�2 0.7938 0.1176 0.5123 0.8166 0.9504
�1 0.1203 0.0568 0.0616 0.1056 0.2770
� 2
0.1182 0.0437 0.0617 0.1085 0.2312
Results obtained after 500 000 iterations (discarding 500 000 as burnt-in)
Next, basing on the obtained posterior predictive simulation data, the generalized impulse response functions
were calculated for Kendall’s tau and lower and upper tail coefficients, taking the corresponding predictive mean,
median, and 2.5 and 97.5 percentiles as the Bayes actions. In figures 1-6, we present plots for the medians. To understand
the reaction of linkages between S&500 and DAX, or, in other words, between the American and European
stock markets, on the shock, we analyze the impulse responses for the three measures of dependence.
0.001
0.000
- 0.001
- 0.002
- 0.003
- 0.004
- 0.005
- 0.006
- 0.007
0 10 20 30 40 50 60 70 80 90 100
Figure 1. September 29, 2008. Impulse response function for Kendall’s tau median
578

0 .0 4 5
0 .0 4 4
0 .0 4 3
0 .0 4 2
0 .0 4 1
0 .0 4
0 .0 3 9
0 .0 3 8
0 .0 3 7
0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0
Figure 2. September 29, 2008. Impulse response function for the lower tail dependence coefficient median
- 0 .0 1 5
- 0 .0 1 7
- 0 .0 1 9
- 0 .0 2 1
- 0 .0 2 3
- 0 .0 2 5
- 0 .0 2 7
- 0 .0 2 9
0 2 0 4 0 6 0 8 0 1 0 0
Figure 3. September 29, 2008. Impulse response function for the upper tail dependence coefficient median
The plots presented in figures 1-3 show us the impulse response in the case of a negative shock hitting the USA
economy. The shapes of all three plots are very similar. However, in the case of Kendall’s tau and the upper tail
dependence coefficient, we can observe a negative impact on the linkages, strengthening during first 30 days. For
the lower tail dependence we have a positive impact on connections, weakening in the first period. The complicated
dynamics of the impulse responses is a result of the high non-linearity in the dependence models. These results are
in agreement with our expectations. The shock was connected with the American economy, so it did not touch the
German stock market in comparable high degree. The weakening of the linkages is thus quite understandable. The
increase in lower tail dependence is, however, an effect of the fear connected with the uncertainty in financial markets.
579

0 .0 0 0
-0 .0 0 1
-0 .0 0 2
-0 .0 0 3
-0 .0 0 4
-0 .0 0 5
-0 .0 0 6
-0 .0 0 7
-0 .0 0 8
-0 .0 0 9
0 .0 4 7
0 .0 4 6
0 .0 4 5
0 .0 4 4
0 .0 4 3
0 .0 4 2
0 .0 4 1
0 .0 4
0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0
Figure 4. October 13, 2008. Impulse response function for Kendall’s tau median
0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0
Figure 5. October 13, 2008. Impulse response function for the lower tail dependence coefficient median
-0 .0 0 5
- 0 .0 0 7 5
-0 .0 1
- 0 .0 1 2 5
-0 .0 1 5
- 0 .0 1 7 5
-0 .0 2
- 0 .0 2 2 5
-0 .0 2 5
0 1 0 2 0 3 0 4 0 5 0 6 0 7 0 8 0 9 0 1 0 0
Figure 6. October 13, 2008. Impulse response function for the upper tail dependence coefficient median
The pattern of reactions in the case of the analyzed positive shock is very similar. We can observe a negative
impact on linkages measured by Kendall’s tau and the upper tail dependence coefficient and positive in the case of
lower tail coefficient. The difference is that during the first 10 days the positive impact on lower tail dependence is
strengthening. So, it seems that a positive news results in stronger connection between indices during market panic
days. It should be admitted, however, that the period under scrutiny was rather a turbulent one, and the presented
example can be misleading.
580

4. Conclusions
The understanding of the nature and the dynamics of dependencies between international markets is of great importance
in finance, as well as in macroeconomic policies. To investigate the process of shock transmission, we proposed
a concept of impulse response function describing the time profile of the effect of shocks on linkages modeled
by means of dynamic copula and measured using Kendall’s tau and the tail dependence coefficients. This approach
was applied to analyze the persistence of the impact of chosen important historically observed shocks on the strength
of linkages between the returns on the S&P500 and DAX indices in 2008. Our results show that the introduced
notion can be a useful tool allowing to perform a deeper analysis of the reaction of the dependencies on various
kinds of shocks.
5. References
Almeida C., Czado C. (2010), Efficient Bayesian inference for stochastic time-varying copula models, Preprint,
Chair of Mathematical Statistics, Technische Universität München.
Amemiya T. (1985), Advanced Econometrics, Harvard University Press, Cambridge, MA.
Berger J.O. (1985), Statistical Decision Theory and Bayesian Analysis, Springer, New York.
Best N., Cowles M., Vines K. (1996), CODA: Convergence Diagnostics and Output Analysis Software for Gibbs
Sampling Output, Version 0.30, MRC Biostatistics Unit, Institute of Public Health, Cambridge, UK.
Geweke J. (2005), Contemporary Bayesian Econometrics and Statistics, Wiley, New York.
Hafner C.M., Herwartz H. (2006), Volatility impulse responses for multivariate GARCH models: An exchange rate
illustration, Journal of International Money and Finance 25, 719-740.
Hafner C.M., Manner H. (2008), Dynamic stochastic copula models: Estimation, inference and Applications, ME-
TEOR Research Memorandum RM/08/043, Maastricht University.
Joe H. (1997), Multivariate Models and Dependence Concepts, Chapman & Hall/CRC, New York.
Koop G., Pesaran M.H., Potter S.M. (1996), Impulse response analysis in nonlinear multivariate models, Journal of
Econometrics 74, 119-147.
McNeil A., Frey R., Embrechts P. (2005), Quantitative Risk Management, Princeton University Press, Princeton.
Nelsen R.B. (2006), An Introduction to Copulas, Springer Verlag, 2nd Edition, New York.
Patton, A.J. (2004), On the Out-of-Sample Importance of Skewness and Asymmetric Dependence for Asset Allocation,
Journal of Financial Econometrics 2, 130-168.
Sklar A. (1959), Fonctions de répartition à n dimensions et leurs marges, Publications de l’Institut Statistique de
l’Université de Paris 8, 229-231.
Smith B. (2005), Bayesian Output Analysis Program (BOA) Version 1.1.5, User’s Manual, Technical Report, Department
of Public Health, The University of Iowa.
Thomas A., O’Hara B., Ligges U., Sturtz S. (2006), Making BUGS Open, R News 6, 12-17.
Tsay R. (2005), Analysis of Financial Time Series, 2nd Edition, Wiley-Interscience, Hoboken.
581

STOCK SPLITS AND HERDING
Maria Chiara Iannino, Queen Mary University of London, UK
Email:m.c.iannino@qmul.ac.uk
Abstract. This paper addresses institutional herding in the case of the announcement of a stock split. The analysis consists in
investigating whether companies that announce stock splits exhibit a systematic abnormal level of herding, and whether the intensity
of this phenomenon helps to explain the market reaction to the event. Using data on the trading activity of US institutional investors,
from 1994 to 2005, institutional herding is captured by the correlation between the institutional demand over two consecutive periods.
The results are consistent with a positive and highly significant level of imitative behaviour, particularly between 1998 and 2001. Big
investors tend to herd more on splitting companies in boom markets. Further analyses factor out the effect of fundamentals and
common public information, showing passive strategies and non-intentional herding are affecting more the institutional demand for
nonsplitting companies. We decompose herding into the contributions of several types and the results are congruent with the presence
of informational cascades at the split announcement. In particular, the difference in herding between the two groups of companies is
explained predominantly by proxies for the quality of the information, as the dispersion of beliefs among analysts. Herding on
nonsplitting companies is instead motivated more by characteristic preference. Finally, the imitative behaviour we observe for splitting
stocks has a stabilizing effect on future returns. This result supports the hypothesis that an informational content is included in the
announcement of the event, and the market underreacts to it.
Keywords: Herding, Institutional investors, Stock Splits, Informational Cascades
JEL classification: G11 G14 G20.
1 Introduction
This paper addresses institutional herding in the specific occurrence of a stock split. This event is still a puzzling
phenomenon because of the abnormal market reaction following its announcement and occurrence (among many,
Lakonishok et al., 1986, Ikenberry et al., 2002). The presence of imitative behaviour could, on one hand, exacerbate
suboptimal decisions in the functioning of the markets and to the reaction to the announcements. On the other hand,
a stabilizing herd behaviour would help prices to aggregate more quickly any informational content that is driven by
the event. In the light of previous literature, that evince an informational content on the announcement of stock
splits, we investigate whether companies that announce stock splits exhibit a systematic abnormal level of herding
with respect to the rest of the market.
Using data on quarterly stocks holdings of US institutional investors, 1994 to 2005, we measure institutional
herding as the correlation between trades among financial institutions over two consecutive periods of time (as Sias,
2004). The analysis proceeds in three steps. First, we measure the level of correlation among investors' decisions
both in the overall market and in a subsample of companies that have announced at least one stock split in the
quarter. Then, we propose an analysis of the motivations of this behaviour according to the theoretical literature and
in particular of the motivations behind the difference in herding between spitting and nonsplitting companies.
Finally, we investigate the stabilizing effect of herding on splitting stocks. 1
Our results are consistent with the presence of a significant level of imitative behaviour. It is particularly intense
in moments of crisis for nonsplitting companies, while big investors tend to herd more on splitting companies in
boom markets. In particular, the difference in herding between the two groups of companies is explained
predominantly by proxies for the quality of the information, consistently with a signal hypothesis in the event
announcement. Moreover, such imitative behaviour on splitting stocks has a stabilizing effect on future returns, as
the market underreacts to a positive informational content in the event.
1 It still persists a gap between the empirical literature, verifying both the presence and the causes of herding among investors in real-functioning
markets, and the theoretical developments on the motivations to herd. Because of the lack of data on the private signals and communications
among agents, empirical tests are based on the excess level of correlation with respect to a benchmark of independence (Lakonishok et al., 1992,
Sias, 2004). It is a clear and easy-to-test definition of institutional herding, however, it does not allow to investigate the coordination mechanism
between agents, and therefore, to distinguish among different reasons to herd.
582

The remainder of this paper ensures as follows. Section 2 describes the methodology we use to detect, measure
and motivate herding in our samples. Section 3 describes the data and discusses the main empirical results. Section 4
concludes offering some final remarks.
2 Sample and methodology
We carry out the empirical investigation on quarterly US stock holdings by financial institutions, extracted from the
Thompson Financial database, over a twelve-year period. We consider all types of professional investment
companies and advisors who are asked to fill the 13F form according to the SEC regulations. Information are
completed by market data about the companies, such as stock splits data, prices and capitalization, from the CRSP
daily database, and data about the analysts' forecasts and coverage from the I/B/E/S monthly database, aggregated
per quarter. The overall sample is composed of 1,760 companies, traded by 3,690 investors. We select two
subsamples of splitting and nonsplitting stocks. A splitting stock is a company that has announced at least one split
in the quarter of analysis, according to CRSP. We have 1,602 announced events by 890 companies.
The starting point of our analysis is the estimation of the intertemporal correlation of the institutional demand
(as Sias, 2004). In the presence of herding, a quarter’s trading actions are affected by previous quarter trades. The
potential level of herding in quarter t is measured as the correlation across companies between the standardized
fraction of buyers of stock i in the quarter t, Δi,t, and the analogous proportion in the previous period t-1:
Δi,t = βt Δi,t-1 + εi,t
where Δi,t=(Pi,t-Pt)/σt; Pi,t is the fraction of institutional buyers of stock i at the end of quarter t; Pt and σt are the
mean and standard deviation of the proportions Pi,t across companies. A positive coefficient βt is consistent with
investors following the past aggregate behaviour of all the institutional investors in the market. We first estimate the
betas on the overall sample, and then in each of the two subsamples of splitting and nonsplitting stocks. 2
We perform further analyses on the betas in order to account for the influence of factors other than intentional
herding. In fact, if investors are exposed to similar market conditions, passive trading strategies and correlated
information, they could exhibit clustered, but nonvoluntary behaviour. We factor out the effect of fundamentals and
common public information, regressing the estimated coefficients on the four factors of Carhart (1997):
βt = α + γHML HMLt + γSMB SMBt + γMt RMt + γMOM MOMt + εt
where HMLt, SMBt, RMt and MOMt are the returns on value-weighted zero-investment factors that mimic
portfolios for, respectively, book-to-market, company size, market returns and momentum, in quarter t. The
coefficients of the factors indicate the loadings of the total beta βt ("Sias' beta") that is attributable to fundamentaldriven
clustering. Then, βt = (α + εt) corresponds to a clean measure of intentional herding for the quarter t,
"conditional on the market conditions", or "beta adjusted". As before, we distinguish between splitting stocks and
nonsplitting stocks and regress the previous equation separately in the two samples.
The second part of the analysis aims to investigate the reasons behind the presence of such observed level of
herding and the difference between splitting and nonsplitting stocks. We impose and test specific assumptions for
four theoretical motivations for herding. In particular, we test whether herding on splitting companies is
informational- or characteristics-based. In uncertain informational environment, agents facing decisions rationally
ignore their noisy and imperfect private information, causing informational cascades to arise (Bikhchandani, 1992,
Avery, 1998). Moreover, under the same informational conditions, rational agents might mimic the investment
decisions of other agents in order to maximize their reputation (Scharfstein, 1990, Dasgupta, 2011). Therefore, we
identify market or company conditions proxies for imperfect information (Wermers, 1999, Chan, 2005) as small
market capitalization, high dispersion of analysts' forecasts and low analysts' coverage. Operationally, we regress the
2
In all the analysis, we investigate the difference in herding between splitting and nonsplitting companies either estimating all the models
separately in each sample, or employ model specifications with binary variables δi,t S , that assumes value 1 if the company has announced at least
one stock split in the quarter of interest. We interact the dummy with the lag institutional demand and, in the following models, including also as
many interacted dummies as the number of regressors.
583

institutional demand on its lag, decomposing the total beta between the effect from the information quality proxies,
Xi,t-1 and other unspecified factors:
C
� = �NIH, t �i,
t-1
+
i, t ��c, tX
c, i, t-1�
i, t-1
� �i,
t
c�1
The coefficients ϕc,t are catching the effect of informational-based herding, in the form of informational
cascades or reputational herding; βNIH,t represents the remaining part of the total beta that cannot be attributed to
informational contents, while βIH,t = (βt - βNIH,t) represents the "Informational Beta". 3
Else than informational-based, a correlation of trades can arise because of investors sharing the same trading
strategies. Gompers et al. (2001) consider the impact of three main variables on the institutions’ demand for stocks:
prudence or regulations, liquidity of the stocks and the historical returns pattern. We distinguish characteristicsbased
herding controlling for variables which mirror such stock characteristics: annual cash dividends per quarter
and volatility of the stock, as proxies for prudence; market capitalization, price per share and share turnover, as
liquidity; and returns over the previous year, as historical pattern of returns. Operationally, we regress the
institutional demand on its lag, decomposing the relation between the effect of the characteristics of the company i at
the quarter t-1 and other unspecified factors:
Q
�i, t = �NCH, t �i,
t-1
+ �� c, tZc,
i, t-1�
i, t-1
� �i,
t
q�1
where ψt is the vector of coefficients of the Q company characteristics; βNCH,t is the remaining part of the Sias'
beta that is not attributable to characteristics preference among investors, while βCH,t = (βt - βNCH,t) is the
"Characteristics Beta".
Institutional investors could also herd because there are momentum traders, leading to a positive relation
between the demand for stocks of quarter t and the past returns of the stocks. 4 Freely following Sias (2007), we
decompose the total correlation between the past returns effect and other factors, adding the lag returns interacted
with the lag demand, as:
Δi,t = βNMT,t Δi,t-1 + ρt Ri,t-1 Δi,t-1 + εi,t
βNMT,t is the remaining part of the correlation not explainable by momentum trading, while βMT,t = (βt -βNMT,t) is
the "Momentum beta".
We replicate all the previous analysis on the motivations to herd for splitting and nonsplitting companies, and
we adjust the betas for the Carhart factors.
The third and final step is to test for the effect of herding on the future returns of companies. Past literature
shows a positive relation between institutional demand and same quarter or previous quarter returns, and a weak
positive correlation with future returns (see, among many, Nofsinger, 1999; Grinblatt, 1995; Sias, 2004). A negative
relation between demand and subsequent returns will be consistent with a destabilizing effect on prices due to
herding, particularly of intentional imitative behaviour either irrational or positive feedback-driven. Alternatively,
either an intentional correlation due to informational motivations or a fundamental-driven correlation will bring the
prices closely and quickly towards the true value, as a stabilizing effect (Sias, 2004). Therefore, we regress the
institutional demand on the past quarter, the same period, and on the two consecutive quarter's returns after the
measurement period.
3 Looking more carefully at the distinction between reputation concerns and informational cascades, we test for herding looking separately at
different institutional types and sizes of the managed portfolio, drawing considerations on the correlation of their trades within the same group or
extra group. The results of this analysis are not reported in this version of the paper.
4 Thus, we take into consideration the possibility of a confounding effect in the beta coefficient, which comes from the fact that the past demand
proxies last quarter returns if there is momentum among investors.
584

3 The main results
3.1 Sias’ Beta
Looking at the presence of herding, Table 1 reports the results from the Sias’ model, estimated on the overall
market, the splitting and nonsplitting samples. As we can see, the estimated coefficients in the overall market are all
positive and statistically significant in all quarters of analysis. This result is consistent with the hypothesis of a level
of herding in the trading decisions of institutions. On average, the beta is 0.457 across all quarters, ranging from
0.346 in 2005 to 0.562 in 2000. The phenomenon is particularly intense in moments of crisis, as in the years 1998 to
2001.
Restricting the analysis to splitting stocks, we observe a negligible difference on the beta coefficients with
respect to the nonsplitting sample. On average, the estimated coefficient for herding in the splitting companies is
very close (0.467) to the alternative group, but more volatile across quarters. However, even if the difference in
means is not statistically significant, the median results are still clearly higher (0.471 against 0.442) for the group of
interest.
Sias' models
1) Overall market: Dependent variable: Δ i,t
Variables
Mean
estimated se
t median min max Q1 Q3
Significant
pos. qrts.
Significant
neg. qrts.
Beta Adjusted
mean se
Δi,t-1 0.457 *** 0.008 56.66 0.447 0.346 0.562 0.417 0.498 48 0 0.448 *** 0.0079
2) Splitting companies: Dependent variable: Δ S i,t
Variables
Mean
estimated se
t median min max Q1 Q3
Significant
pos. qrts.
Significant
neg. qrts.
Beta Adjusted
mean se
Δ S i,t-1 0.454 *** 0.031 14.83 0.471 -0.074 0.915 0.336 0.583 36 0 0.461 *** 0.0290
3) Non-splitting companies: Dependent variable: Δ NS i,t
Variables
Mean
estimated se
t median min max Q1 Q3
Significant
pos. qrts.
Significant
neg. qrts.
Beta Adjusted
mean se
Δ NS i,t-1 0.453 *** 0.008 55.36 0.442 0.344 0.559 0.414 0.497 48 0 0.444 *** 0.0080
Test: Beta S = Beta NS t = 0.0426 (0.9662) Beta adj: t = 0.5941 (0.5549)
Table 1. The table reports the summary statistics of the coefficients of the lag institutional demand (Sias' Betas) estimated in each
quarter of analysis, from 1994 to 2005, for the models based on Sias (2004). Institutional demand Δi,t, as the fraction of buyers of stock i
at quarter t, is firstly regressed on its lag Δi,t-1 in the overall sample (Model 1). Model 2 considers the institutional demand computed on
the sample of splitting companies regressed on the lag demand for all stocks. Analogously, Model 3 is applied to the sample of
nonsplitting companies. The reported t-values are computed from the standard error of the estimates series. The numbers of significant
quarters are identified with a 10% of significance level. The last column considers the Beta adjusted, as the lag coefficients, once
controlling for the four factors à la Cahart (1997). We finally report the statistic and the p-value of the tests on difference between
splitting and nonsplitting samples for the betas and the betas adjusted.
Looking at three subperiods of four years each, we see that investors tend to herd slightly more when they trade
on splitting companies in the subperiod from 1994 to 2001, while we observe a higher herding on nonsplitting
companies, even if still not significant, from 2002 onwards (Fig. 1.1). In period of crisis, herding appears to increase
for nonsplitting companies, consistently with the literature (Lakonishok et al., 1996). Splitting stocks are instead
affected by market crises as their frequency decreases, but not in the intensity of the herding phenomenon.
This variation over time, and the negligible average difference between the two groups, motivates additional
analysis. Taking into account the effect of different trading activity among companies, we see that convergence of
behaviour increases with the trading activity on the company, and herding is more likely to occur among
nonsplitting stocks once we take out the effect of thin markets (Fig. 1.2). We also consider the difference in herding
in the two groups, differentiating for characteristics of the investors, such as type and portfolio size (Fig. 1.3 and
1.4).
585

This time pattern could also be caused by market factors. Cleansing the coefficients from common factors, we
are then able to approximately discriminate between intentional and unintentional herding. The last column of the
previous Table 1 reports the average standardized coefficients for the three samples. The factors are determinants of
the herding phenomenon, but the average adjusted betas are still all considerably significant, and continue to
represent almost all of the convergence of behaviour. On average, the estimated adjusted beta is 0.448, as 98% of the
total correlation measured by the average Sias' beta.
These factors are significant determinants of the institutional demand especially for nonsplitting companies.
Consistently with what stated above nonsplitting stocks are more sensible to market conditions. Passive strategies
based on the four factors account significantly in the trading activities of nonsplitting companies, while splitting
stocks might tend to be more actively traded, as they appear to be less affected by unintentional factors, as the
adjusted beta still accounts for 93% of the total correlation (against 85%) with a significantly positive difference in
mean.
Figure 1.1 Average Splitting Dummies estimated per subperiod Figure 1.2 Average Splitting Dummies estimated per Number of Traders.
ts
n
ie
fic
e
c
o
y
m
u
d
g
e
ra
a
v
e
Figure 1.3 Average Splitting Dummies estimated per size of investor portfolio Figure 1.4 Average Splitting Dummies estimated per type of investors
average dummy coefficients
-.1 -.05 0 .05
5
,0
0
0
,0
0
5
,0
-0
0
,1
-0
0,048 0,047
1994-1997 1998-2001 2002-2005
-.099257
.010215
.051335
-0,092
small investors medium investors
big investors
-.15 -.1 -.05 0
average dummy coefficients
average dummy coefficients
-.1 -.05 0 .05
-.029251
-.089904
-.037966
-.059659
.031022
-.099528
.022123
-.154854
>= 10 traders >= 20 traders
>= 50 traders >= 100 traders
banks insurance co.
investment co. indip. advisors
not defined
Figure 1. The following graphs report more detailed investigations on the difference in herding among the two groups. We estimate the
Sias’ model including a dummy variable for split, interacted with the lag institutional demand. The results report the average dummy
coefficients by sub-period, by trading activity of the company, by size of the managed portfolio and by type of investors.
3.2 Motivations to institutional herding
We examining the impact of the four theoretical types of herding on the estimated correlation: informational
cascades, reputational herding, characteristic herding and momentum trading. Table 2 reports the average estimated
coefficients for the variables in all the different models. Table 3 includes the model specifications with the interacted
dummies of splitting. In summary, in the overall market, herding is on average mostly affected by characteristics
such as size, coverage, turnover, price, dividend and past returns. Different considerations can be drawn for the
splitting sample.
586
-.02936

1) Splitting companies
Variables
mean t mean t mean t mean t mean t
Δi,t-1 0.4170 *** 14.63 -0.2413 -0.79 0.2405 1.63 0.4278 *** 13.02 0.2155 0.23
Dispersion*t-1 0.4378 ** 2.01 -2.1512 -0.96
Coverage*t-1 0.1728 1.30 0.3278 0.81
Size*Coverage*t-1 0.4031 0.97 -2.8701 -1.21
Size*i,t-1 -0.1744 -0.43 0.2979 *** 4.82 3.4335 1.23
Price*i,t-1 -0.1120 -1.16 -2.0436 -1.13
Turnover*i,t-1 -0.0514 -0.60 -0.5519 -1.27
StDeviation of returns*i,t-1 0.1340 1.25 3.3859 1.04
Returns*i,t-4 0.0172 0.46 -0.0276 -0.42
Dividends*i,t-1 -0.0218 -0.35 -0.0623 -0.28
Returns*i,t-1 -0.0283 -1.19 -0.0867 -1.12
2) Nonsplitting companies
Average Estimated Coefficients for all Models
Dependent variable: Δ S i,t
Sias' model Informational- based Characteristic- based
Dependent variable: Δ NS i,t
Momentum Unifying model
Variables
Sias' model Informational- based Characteristic- based Momentum Unifying model
mean t mean t mean t mean t mean t
Δi,t-1 0.4749 *** 56.68 0.4155 *** 29.15 0.3721 *** 18.70 0.4660 *** 56.34 0.3759 *** 16.04
Dispersion*t-1 -0.0139 -1.40 0.1378 -1.02
Coverage*t-1 0.1071 *** 10.93 -0.0086 *** 11.13
Size*Coverage*t-1 -0.0801 *** -6.25 0.1010 *** -6.18
Size*i,t-1 0.1467 *** 11.65 0.0935 *** 15.23 -0.0732 *** 11.17
Price*i,t-1 0.0708 *** 7.92 0.0526 *** 6.81
Turnover*i,t-1 0.0392 *** 3.99 0.0082 0.94
StDeviation of returns*i,t-1 -0.0211 -1.19 -0.0336 * -1.64
Returns*i,t-4 0.0035 0.60 0.0056 1.09
Dividends*i,t-1 0.0115 ** 2.09 0.0214 *** 3.87
Returns*i,t-1 0.0530 *** 6.74 0.0522 *** 6.44
Table 2. The table reports the average standardized coefficients of all the variables used in the five models. (1) Sias' model regresses the
institutional demand Δi,t on its lag Δi,t-1 only. (2) Informational-based models regress the institutional demand on its lag and a set of
proxies for the quality of information, such as size, dispersion, coverage and size*coverage at the previous quarter. (3) Characteristicsbased
models regress the institutional demand on its lag and a set of company characteristics, such as size, price, turnover, standard
deviation, returns of stocks and quarterly dividends, measured at the previous quarter. (4) The Momentum models regress the
institutional demand on its lag and the previous year returns. (5) Finally, an unifying model regresses the institutional demand on all the
previous variables. The significance is attributed estimating the t statistics from the time series of the beta estimates. *10%, **5%,
***1% of significance level.
3.2.1 Informational Cascades
The splitting sample presents interesting results, consistent with a predominance of informational-based herding.
The model is well specified and the F tests on all the Informational regressors provide evidence of the significance
of the proxies for informational-based herding in most of the quarters. Moreover, the signs of the estimated
coefficients of the proxies confirm that herding increases as the quality of the information available decreases.
Particularly important is the dispersion of beliefs among analysts. In fact, the dispersion coefficient is on average
positive and significant, explaining the most of the Sias' beta. Higher is the dispersion of beliefs in the quarter
preceding the split, higher is the level of herding.
The difference in the level of herding, between splitting and nonsplitting companies, remaining after addressing
informational content, is significant, consistently with the hypothesis that stock splits announcement convey
information to the market, creating the environment of uncertainty for informational cascades to arise. Also, looking
at the adjusted betas, cleaned from common factors, we substantiate even stronger conclusions, and the test for a
positive difference in mean is significant at 1%.
587

Average Estimated Coefficients for all Models (dummy specifications)
Dependent variable: Δi,t
Overall market with splitting dummies
Variables
Sias' model
mean t
Informational- based Characteristic- based
mean t mean t
Momentum
mean t
Unifying model
mean t
Δi,t-1 0.4743 *** 56.79 0.4116 *** 28.68 0.3671 *** 18.52 0.4655 *** 56.47 0.3709 *** 15.66
Dispersion*t-1 -0.0134 -1.38 -0.0092 -1.07
Coverage*t-1 0.1095 *** 11.05 0.1035 *** 11.31
Size*Coverage*t-1 -0.0814 *** -6.05 -0.0768 *** -6.49
Size*i,t-1 0.1549 *** 11.67 0.1027 *** 17.80 0.1476 *** 12.09
Price*i,t-1 0.0726 *** 8.02 0.0550 *** 6.99
Turnover*i,t-1 0.0406 *** 4.26 0.0096 1.14
StDeviation of returns*i,t-1 -0.0174 -0.98 -0.0305 -1.51
Returns*i,t-4 0.0030 0.51 0.0075 1.51
Dividends*i,t-1 0.0109 ** 1.98 0.0196 *** 3.45
Returns*i,t-1 0.0534 *** 6.72 0.0514 *** 6.22
δi,tΔi,t-1 -0.0009 -0.19 -0.0034 -0.65 -0.0050 -1.03 0.0000 0.00 -0.0014 -0.26
δi,t*Dispersion*t-1 0.0262 *** 2.52 -0.0378 -0.73
δi,t*Coverage*t-1 -0.0090 -0.96 -0.0324 -1.26
δi,t*Size*Coverage*t-1 0.0798 1.30 -0.0989 -0.65
δi,t*Size*i,t-1 -0.0630 -1.04 0.0157 *** 2.74 0.1257 0.89
δi,t*Price*i,t-1 0.0031 0.31 0.0406 1.12
δi,t*Turnover*i,t-1 0.0013 0.19 0.0242 1.18
δi,t*StDeviation of returns*i,t-1 0.0139 * 1.93 0.0065 0.40
δi,t*Returns*i,t-4 0.0041 1.30 0.0055 1.26
δi,t*Dividends*i,t-1 -0.0109 * -1.83 -0.0200 *** -2.61
δi,t*Returns*i,t-1 -0.0124 *** -2.72 -0.0151 *** -2.45
Table 3. The table reports the standardized coefficients of all the variables used in the five models. (1) Sias' model regresses the
institutional demand Δi,t on its lag Δi,t-1 only. (2) Informational-based models regress the institutional demand on the lag demand and a
set of proxies for the quality of information, such as size, dispersion, coverage and size*coverage at the previous quarter. (3)
Characteristics-based models regress the institutional demand on its lag and a set of company characteristics, such as size, price,
turnover, standard deviation, returns of stocks and quarterly dividends, measured at the previous quarter. (4) Momentum models regress
the institutional demand on its lag and the previous year returns. (5) Finally, an unifying model regresses the institutional demand on all
the previous variables. The significance is attributed estimating the t statistics from the time series of the beta estimates. *10%, **5%,
***1% of significance level.
3.2.2 Positive Feedback Herding
Checking for positive feedback herding, we look at first at characteristics-based herding. The remaining betas are
still positive and significant. Moreover, looking at the regressors coefficients, we see that company characteristics
have an important effect on the convergence in the overall market. The F tests on the regressors mainly reject the
null of non-joint significance for all samples. Moreover, the signs of the coefficients are consistent with the
hypothesis that institutions tend to herd on nonsplitting companies, preferring large companies, with high price-pershare,
high turnover and high dividends.
On splitting companies, the results are weaker than the previous informational-based herding. Even if the F tests
show a joint significance of the regressors in explaining the convergence of behaviour, only size has a significant
and positive coefficient, while all the other coefficients are rejecting the hypothesis of characteristics-based
convergence. The difference in means with nonsplitting companies is also not significant.
In the class of positive feedback herding, we also investigate the presence of momentum trading. Momentum
strategies have an effect on the convergence towards stocks, as the coefficients of the past returns are mostly
significant. However, the effect is smaller with respect to the other types, especially for the splitting sample.
588

Looking at the difference between groups, the adjusted betas have also interesting results. The difference in
means between the samples is significant at 1%, therefore, once past returns are accounted for and the effect of
passive strategies is cleared out, investors herd more on splitting companies for reasons other than momentum.
3.3 The stabilizing effect of herding
Finally, we look at the effect of herding on the future performance of the companies. We regress the institutional
demand on the past quarter, the same period, and on the two consecutive quarters’ returns after the measurement
period. Consistently with the literature, we observe a positive relationship between institutional demand and past
quarter and same quarter returns for the overall market and for nonsplitting companies. We do not conclude for any
similar relationship for the splitting companies (Table 4).
Instead, interestingly, we observe a positive and highly significant relation between institutional demand and
returns in the two following quarters for splitting firms (0.3704 and 0.3692 respectively). This is consistent with a
stabilizing effect on prices for splitting companies due to herding. According to Sias (2004), such a positive relation
is further evidence of the presence of informational-based herding.
1) Splitting companies: Dependent variable: Δ S i,t
Variables
Mean
estimated
Stabilizying Effect Models
se t min max
Δi,t-1 0.4057 *** 0.0470 8.63 -0.3454 1.4248 26 0
Rett-1 0.0819 0.2240 0.37 -5.0132 3.2220 9 3
Rett 0.0035 0.2140 0.02 -5.2177 2.8125 8 1
Rett+1 0.3704 *** 0.1760 2.11 -1.8904 3.7678 8 2
Rett+2 0.3692 *** 0.1794 2.06 -3.5798 3.3968 7 1
Rett+4 0.1201 0.1904 0.63 -4.1861 3.9397 4 5
2) Non-splitting companies: Dependent variable: Δ NS i,t
Significant
pos. qrts.
Mean Significant
Variables
se t min max
estimated
pos. qrts.
Significant
neg. qrts.
Significant
neg. qrts.
Δi,t-1 0.4484 *** 0.0081 55.18 0.3633 0.5552 44 0
Rett-1 0.2212 *** 0.0395 5.61 -0.3450 0.8020 23 1
Rett 0.3420 *** 0.0369 9.27 -0.2119 0.8572 31 0
Rett+1 -0.0015 0.0342 -0.04 -0.5669 0.6303 10 6
Rett+2 0.0100 0.0301 0.33 -0.3770 0.4235 5 4
Rett+4 -0.0337 0.0289 -1.17 -0.4689 0.3463 3 5
Table 4. The table reports the summary statistics of the coefficients of the lag institutional demand estimated in each quarter of
analysis, from 1994 to 2005, for four Stabilizing Models. Model (1) regresses the institutional demand (as fraction of buyer of
stock i at quarter t) on the lag institutional demand for the overall market and past quarter, same quarter, following two quarters
and following year returns. Model (2) regresses the institutional demand on its lag, the previous set of investors and a set of
splitting dummies interacted with all the regressors. Models (3) and Model (4) regress the same specification as Model (1) in
only the splitting sample and nonsplitting sample respectively. The t-values reported are computed from the standard errors of
the estimates series. The numbers of significant quarters are identified with a 10% of significance level. *10%, **5%, ***1% of
significance level.
4 Summary
With this empirical paper we aim to contribute to the understanding of stock splits and their market reaction, in the
light of the impact of herding, or correlated trading decisions among institutional investors.
We have found evidence of positive and significant correlation in each quarter of analysis. Also, cleaning for
intentional correlation, we have found that most of the correlation is not attributable to the four factors, namely size,
market return and book-to-market, that proxy for unintentional or spurious herding.
589

Distinguishing between herding in splitting and nonsplitting companies in the overall case, we do not evince a
significant difference on average. However, the difference in correlation decreases over the three four-year
subperiods, going from a positive but not significant average in the subperiod 1994 - 1998, to a negative (but still
not significant) average in 2002-2005.
However, we find a difference in the motivation to herd, being splitting companies more affected by
informational-based herding. The presence of informational-based herding, especially for splitting companies, is
also confirmed by observing the relation between institutional demand and future returns. Moreover, the positive
relation we find between institutional demand for splitting firms and their future returns in the following two
quarters, is consistent with the stabilizing effect of herding. On the contrary, we do not report any significant
relationship between institutional demand and future returns in the nonsplitting sample or in the overall market.
Our results are therefore consistent with the presence of informational content in the split event and to the
underreaction of the market. We should also note however, that this underreaction is affected by trading on herd.
Still a significant part of the correlation among investors and of the difference between the subsamples is not
explained by these four types, suggesting that further studies should be carried out to better understand other
motivations, probably irrational, to the phenomenon. Moreover, further development of this research will focus on
investigating both the change in herding and its impact on the future performance of the company on the days
around the announcement of stock splits.
5 References
Avery, Christopher and Peter Zemsky (1998), ‘Multidimensional Uncertainty and Herd Behaviour in Financial
Markets’, The American Economic Review 88(4), 724-748.
Bikhchandani, Sushil, David Hirshleifer and Ivo Welch (1992), ‘A Theory of Fads, Fashion, Custom, and Cultural
Change as Informational Cascades’, Journal of Political Economy 100(5), 992-1026.
Carhart, Mark M. (1997), ‘On Persistence in Mutual Fund Performance’, The Journal of Finance 52(1), 57.
Chan, Kalok, Chuan-Yang Hwang and Mujtaba G. Mian (2005), ‘Mutual fund herding and Dispersion of Analysts’
Earnings Forecasts’, Working Paper February, 1-41.
Dasgupta, Amil, Andrea Prat and Michela Verardo (2011), ‘The Price Impact Of Institutional Herding’, Review of
Financial Studies 24 (3), 892-925
Gompers, Paul A. and Andrew Metrick (2001), ‘Institutional Investors and Equity Prices’, Quarterly Journal of
Economics 116(1), 229-259.
Grinblatt, Mark, Sheridan Titman and Russ Wermers (1995), ‘Momentum Investment Strategies, Porfolio
Performance, and Herding: A Study of Mutual Fund Herding’, The American Economic Review 85(5), 1088-
1105.
Ikenberry, David L. and Sundaresh Ramnath (2002), ‘Underreaction to Self-Selected News Events: The Case of
Stock Splits’, Review of Financial Studies 15(2), 489-526.
Lakonishok, Josef, Andrei Shleifer and Robert Vishny (1992), ‘The impact of institutional stock prices’, Journal of
Financial Economics 32, 23-43.
Lakonishok, Josef and Theo Vermaelen (1986), ‘Tax- induced trading around ex-dividend’, Journal of Financial
Economics 16, 287-319.
Nofsinger, John R. and Richard W. Sias (1999), ‘Herding and Feedback Trading by Institutional and Individual
Investors’, Journal of Finance 54(6), 2263-2295.
590

Scharfstein, David S. and Jeremy C. Stein (1990), ‘Herd Behaviour and Investment’, The American Economic
Review 8(3), 465-479.
Sias, Richard W. (2004), ‘Institutional Herding’, Review of Financial Studies 17(1), 165-206.
Sias, Richard W. (2007), ‘Reconcilable Differences: Momentum Trading by Institutions’, The Financial Review 42,
1-22.
Wermers, Russ (1999), ‘Mutual Fund Herding and the Impact on Stock Prices’, The Journal of Finance 54(2), 581-
622.
591

TIME-VARYING BETA RISK FOR TRADING STOCKS OF TEHRAN STOCK EXCHANGE IN IRAN:
(A COMPARISON OF ALTERNATIVE MODELING TECHNIQUES)
Majid Mirzaee Ghazani
University of Tehran, Faculty of Economics
E-mail: MajidMirza@ut.ac.ir http://economics.ut.ac.ir/
Abstract. This paper investigates the time-varying behavior of systematic risk of 30 stocks in the form of 10 different sectors that are
trading in Tehran stock exchange (TSE).The daily data is used for this study and the Period is from July 2003 to December 2009.
Three different modeling techniques in addition to the standard constant coefficient model are employed: a bivariate t-GARCH (1, 1)
model, two Kalman filter based approaches as well as a bivariate stochastic volatility model estimated via the efficient Monte Carlo
likelihood technique. A comparison of the different models' ex-ante forecast performances indicates that the random-walk model in
the Kalman filter approach is the preferred model to describe and forecast the time-varying behavior of stocks’ betas in Tehran stock
exchange (TSE).
Keywords: time-varying beta risk; Kalman filter; bivariate t-GARCH; stochastic volatility; efficient Monte Carlo likelihood.
JEL classification: C10; C58; G10; G12; G17.
1 Introduction
Beta is a parameter that is mostly used by financial economists and practitioners to measure an asset’s or portfolio’s
risk. This parameter is a measure of systematic risk, the non-diversifiable portion of the variability in returns in the
Capital Asset Pricing Model (CAPM) which originally developed by Sharpe and Lintner (1965). The main
assumption that is considered for calculation of the betas in this context is the time-invariant assumption or
constancy of beta’s CAPM models. At this approach, betas have been estimated via ordinary least squares (OLS).
However, there are considerable literatures that have focused on the dependency of systematic risk of an asset on
macroeconomic factors and at last have rejected the assumption of beta stability, e.g., Bos and Newbold (1984),
Collins et al. (1987), Brooks, Faff and Lee (1992), Pope and Warrington (1996), Choudhry (2005), Choudhry et al
(2010).
Due to excess kurtosis and leptokurtic nature of financial time series and weak performance of time-invariant
Betas, several studies in recent decades have been deployed to show the time-varying behavior of systematic risk
that some of these studies mentioned above.
In this paper, different techniques have been employed to analyze the time-varying behavior of betas in 30
diverse stocks and in the form of 10 sectors 1 . The first technique that has been utilized to estimate time varying betas
is based on the multivariate general autoregressive conditional heteroskedasticity (M-GARCH) model which has
been proposed by Bollerslev (1990). A GARCH (1,1) model is used to estimate conditional variances for generating
of the series of conditional time-varying betas. This model has been applied in different studies to estimate timevarying
betas, e.g., Giannopoulos (1995), Brooks et al. (1998), Faff et al. (2000), Li (2003), Choudhry and Wu
(2007), Mergner and Bulla (2008), Choudhry et al. (2010).
The second technique which is employed to model the time-varying systematic risk of selected sectors is based
on another type of volatility-based approaches that as known, Stochastic Volatility (SV) models. Unlike the GARCH
models which define the time-varying variance as a deterministic function of past squared innovations and lagged
conditional variances; In the SV models, the variance is modeled as an unobserved component that follows some
stochastic process. SV models are reviewed in, e.g., Harvey et al. (1994), Danielsson (1994), Ghysels et al. (1996),
Shephared (1996), Jacquier et al. (1999), Broto and Ruiz (2004), Asai et al. (2006).
1 - Because some studies indicate the usefulness of investigating betas on sector; instead of stock level, e.g., Yao and Gao (2004).
592

The last method which is utilized in this paper to model time-varying betas is based on State-space approaches.
One of the well known techniques that are used in this approach is known as Kalman Filter (KF). The main
difference between this technique and SV model is latented in the process of estimation. By employing the Kalman
Filter, the time-varying betas can be estimated directly, while in SV model this process should be done indirectly by
utilizing estimated conditional variance series. This paper has focused on two forms of transition equation in KF
approach, which are known as Random Walk (RW) and Mean-Reverting (MR) models and have been applied in
several studies, e.g., Wells (1994), Groenewold and fraser (1999), Mergner and Bulla (2008), He and Kryzanowski
(2008).
The reminder of this paper is organized as follows. The specification of competing techniques discussed in
section 2. The data and descriptive analysis are discussed in section 3 and the empirical results mentioned in section
4. Conclusion and main implication of results are presented in section 5.
2 Methodology
2.1 The unconditional beta in the CAPM
Following the Sharpe (1964) and Lintner (1965), an asset’s unconditional beta in the CAPM can be estimated via
OLS:
j=1, 2, , N
t= 1, 2, , T (1)
Where is the expected rate of return on sector j, is the risk-free rate, is the expected
rate of return on the market portfolio and is calculated as follows:
In this notation, and in order, represent excess returns on sector j and market portfolio.
2.2 Estimation of Conditional Betas by GARCH Model
In traditional CAPM is assumed that return series are independently and identically distributed (IID), while in
practical, there are signs of autocorrelation and volatility clusters that violate this assumption.
One of the main characters of financial time series is the existence of a phenomenon that commonly referred to
as volatility clustering and because of this feature, the volatility process is time-varying. In the meantime the
GARCH models have been proposed to model and forecast time-varying structure of volatilities. Also the GARCHbased
approach to model time-varying beta has been utilized in various studies, e.g., McClain et al. (1996), Lie et al.
(2000), Faff et al. (2000), Brooks et al. (2002), Li (2003), Marti (2006), Choudhry and Wu (2007), Mergner and
Bulla (2008), Choudhry et al (2010). Furthermore, we should declare that multivariate GARCH (M-GARCH)
model, which first was proposed by Bollerslev (1990); plays a key role to modeling and estimating of betas.
According to Bauwens et al. (2006), we can distinguish three approaches for constructing M-GARCH models;
(1) Direct generalization of the univariate GARCH model of Bollerslev (1986). In this approach we have VEC,
BEKK and Factor models; (2) Linear combinations of univariate GARCH models, which we have orthogonal and
latent Factor models; (3) Non-linear combinations of univariate GARCH models; for this category we can mention
Constant Conditional Correlation (CCC) and Dynamic Conditional Correlation (DCC) models along with General
Dynamic Covariance (GDC) and Copula-GARCH models.
In this paper a bivariate version of the multivariate GARCH (M-GARCH) model is applied to calculate timevarying
betas. Also a system of two conditional mean equations has been considered as follows:
593
(2)

Where represents a vector that declares the excess returns of sector j and exposes the
excess return of the market portfolio. Also is a vector of constants and denotes a vector
of innovations that conditioned by the complete information set of . As follows, we consider a general
bivariate GARCH model that is shown in below equation:
Where
E , E (5)
As mentioned earlier in (2.2) section; there are different methods to specify or conditional variance matrix,
that we categorized them in three groups. In this paper, we employ Constant Conditional Correlation (CCC) model
that has been proposed by Bollerslev (1990). By the using of this method, we can reduce the complexity of
computation of general multivariate GARCH (1,1) models. The (CCC) model is defined as:
Where
And R is a systematic positive definite matrix with . In the bivariate case, the number of
parameters will be seven numbers. The calculation of this parameters are performed by OX 6 of Doornik (2009) and
the G@RCH 6 package of Laurent (2009). Due to exhibition of leptokurtosis and skewness of sectors’ excess returns,
which have been revealed in table (1) and also mentioned by Sandman and Koopman (1998) and Mergner and Bulla
(2008); we have chosen a t-GARCH model that‘t’ in this definition, refers to the student’s t-distribution for
innovation in equation (4). By considering of these conditions, we can calculate time-varying beta of t-GARCH
model as follows:
In this equation, is the conditional covariance between sector j and the market portfolio and also
shows conditional market portfolio’s variance.
2.3 Estimation of Conditional Betas by Stochastic Volatility (SV) Model
An alternative way of modeling conditional time-varying betas is offered by Stochastic Volatility (SV) models
which are introduced by Taylor (1982, 1986). This type of models has some features that distinct them from
GARCH-type models (Asai et al., 2006). Some of major differences between these two approaches are mentioned
below:
i) In SV models the volatility process is Random, whereas in GARCH-type models conditional variance of returns
is assumed to be a deterministic function of past returns. (Asai et al, 2006)
ii) An unobserved shock added to return variance of SV models that makes them more flexible than GARCH-type
models. (Kim et al, 1998)
iii) The GARCH-type models are observation-driven while SV models are parameter-driven. (Mergner, 2009)
Because of these properties of SV models, we tried to use them for modeling of time-varying betas.
We can represent a general SV model in mean and variance equation. The mean equation is given by
594
(3)
(4)
(6)
(7)
(8)

t = 1, (9)
Where is the mean adjusted return on asset j. Also the variance equation is given by
In this equation, is known as positive scaling factor. The stochastic process for is modeled as an AR (1)
process as follows:
Persistence parameter is restricted to a positive value less than one to ensure stationarity.
Because the conditional covariance in SV models is latent and should be integrated out from the joint likelihood
function of excess returns and the conditional covariance, parameters in this type of models, cannot be estimated by
a direct standard Maximum Likelihood technique. Therefore, numerous techniques have been developed to solve
this barrier. They include a set of models, from GMM which proposed by Taylor (1986) and quasi-Maximum
Likelihood (QML) that introduced by Harvey et al, (1994) to more efficient methods like Monte Carlo Markov
Chain (MCMC) Which proposed by Jacquier (1994), Monte Carlo Likelihood (MCL) proposed by Danielsson
(1994) and efficient MCL that developed by Sandmann and Koopman (1998).
In this paper we have used the efficient MCL method for estimation of SV model. The calculations for
estimating of time-varying betas have performed by OX 6 of Doornik (2009) and the ssfpack package by Koopman et
al. (1999). Equation (12) in below, represents the formula for estimating of sector’s betas.
Where in this equation is the correlation coefficient between excess returns of sector j and market portfolio.
2.4 Estimation of Conditional Betas by Kalman Filter Method
The last method we employ in this paper to estimate time-varying betas, is one of the state-space models that is
called the Kalman Filter (KF). This method has been applied in different studies to estimate time-varying betas, e.g.,
Bos and Newbold (1984), Faff et al. (2000), Yao and Gao (2004), Mergner and Bulla (2008), Adrian and Franzoni
(2009).
Using of Kalman Filter method has some benefits in contrast to models which mentioned in last sections
(GARCH-type and SV models). The first advantage back to state-space models that allows us to estimate the betas
directly (Mergner, 2009). At second, the Kalman Filter converges quickly and the underlying model has no effect on
the converging process (Yao and Gao, 2004).
The Kalman Filter estimates the conditional beta by using the following structure:
Equation (13) represents the observation or measurement equation of the state-space model. The other equation that
declares the dynamic process of time-varying betas is called transition equation, which has been shown in equation
(14) as follows:
(14)
595
(10)
(11)
(12)
(13)

The transition or state equation can be flexible and be shown as a random walk, random coefficients or a
random mean model. In this study, two forms of transition equation for estimation of time-varying betas have been
used; they include Random Walk (RW) and Mean-Reverting (MR) models, which have been specified as follows:
Random Walk (15)
Mean-Reverting (16)
The Kalman Filter models, have been computed by OX 6 Doornik (2009) and the ssfpack package by Koopman
et al. (1999).
3 Data and Descriptive Analysis
3.1 Data series of Excess Returns
The data that is used in this paper represents the time series of daily excess returns 2 on 30 stocks that formed in 10
diverse sectors. The time range of data is from July 2003 to December 2009. All the information and data about
indices and sectors are gathered from Tehran Stock Exchange (TSE) and also the main index of TSE serves as a
proxy for the market portfolio. The actual excess returns of Main Index and two sectors (pharmaceutical and
machinery and equipments) stated in figure (1) in below.
2 -The excess returns between period t and t-1 for index j are calculated as which in this equation is the
risk-free rate of return based on rate of 1 year Deposit Account that converted to Daily returns.
596

3.2 Descriptive Analysis
Figure 1: Daily excess returns (in %) of Main Index and 2 sectors.
The Descriptive Statistics for the data are presented in Table (1). It’s observable that the highest daily mean excess
return is devoted to Basic Metals (0.13%) and the lowest is seen in Investment sector. Also it’s easily detectable that
the data series of daily excess returns, show high level of kurtosis (leptokurtic). The Jarque-Bera statistic, which
mentioned in last column, strongly satisfies the rejection of normality hypothesis for all return series at the 1%
significance level.
4 Empirical Results
Sector N Mean Std. Dev. Skewness Kurtosis Jarque-Bera
Main Index 1255 -0.0100 0.5999 0.04 11.61 7055
Basic Metals 1255 0.1332 1.2372 6.38 87.81 4.1176e+005
Pharmaceutical 1255 0.0236 0.5364 3.41 38.72 80853
Food & Beverage 1255 0.0181 0.8633 -0.05 46.75 1.1429e+005
Automobiles and Parts 1255 0.0258 0.9113 1.90 41.89 92513
Chemical 1255 0.0714 1.0840 2.61 58.23 1.7875e+005
Investment 1255 0.0073 0.8175 -0.12 17.76 16498
Banks & Credit 1255 0.1305 1.8981 3.26 88.14 4.0852e+005
Homebuilding 1255 0.0875 0.9769 0.40 10.44 5739
Machinery & Equipments 1255 0.0267 1.0147 3.31 48.05 1.2304e+005
Petroleum Products 1255 0.0306 0.8935 4.42 49.31 1.3124e+005
Table 1: Descriptive Statistics of Daily excess returns
4.1 Comparison of Alternative Methods’ Results on Beta Estimation
In this section, we compare results of four diverse methods on estimation of betas; including time-invariant method
(OLS) and three other techniques that are based on time-varying feature of betas. These three techniques that are
mentioned earlier in section 2, are including of two volatility-based models (t-GARCH (1,1) and SV) and a statemodel.
This state-space model is based on Kalman Filter (KF) approach that is investigated, in two forms of random
walk and mean reverting specification.
The results of unconditional and conditional betas have been presented in table (2). The conditional betas have
been shown in mean, because their values varying across the time.
sector * * * *
Automobiles and Parts 1.125 1.174 1.146 1.241 1.191
Banks & Credit 1.085 1.052 1.077 1.068 1.034
Basic Metals 0.945 0.876 0.931 0.958 0.915
Chemical 1.011 1.027 1.029 1.022 0.986
Food & Beverage 0.865 0.907 0.891 0.886 0.876
Homebuilding 1.135 1.118 1.107 1.113 1.124
Investment 1.354 1.312 1.308 1.313 1.293
597

Machinery & Equipments 1.252 1.237 1.206 1.231 1.189
Petroleum Products 0.914 0.927 0.911 0.883 0.954
Pharmaceutical 1.184 1.166 1.147 1.139 1.152
*-The conditional beta series represented in mean
4.2 Measurement of Forecasting Accuracy
Table 2: The unconditional and conditional beta series on different sectors
In an attempt to understand the relative performance of selected models in terms of their forecast accuracy and also
to determine, which one of the mentioned approaches makes the best measurement for estimation of time-varying
betas, we employ two well known criteria to evaluate in-sample forecast performance of alternative methods. These
criteria are mean squared error (MSE) and the mean absolute error (MAE):
And
In these equations, denotes the series of forecasted excess returns for sector j. The results have been
presented in Table (3) and Table (4). On the basis of the both criteria, we can see that calculation of beta by Kalman
filter and through the random walk specification has revealed the best result. Also the value of MAE in KF approach
and in the random walk model has the least quantity among other techniques. Furthermore, as visible in table (3), the
random walk model is superior to other techniques in all sectors except Automobiles and Parts and Pharmaceutical
that in these two sectors; the random walk model has been located in second place with a little difference.
sector
Automobiles and Parts 0.0146 0.0156 0.0152 0.0141 0.0155
Banks & Credit 0.0110 0.0097 0.0086 0.0091 0.0093
Basic Metals 0.0124 0.0117 0.0110 0.0112 0.0113
Chemical 0.0122 0.0119 0.0114 0.0116 0.0120
Food & Beverage 0.0102 0.0095 0.0092 0.0092 0.0093
Homebuilding 0.0119 0.0119 0.0115 0.0117 0.0118
Investment 0.0111 0.0098 0.0078 0.0084 0.0086
Machinery & Equipments 0.0118 0.0098 0.00865 0.0095 0.0097
Petroleum Products 0.0134 0.0129 0.01018 0.0123 0.0128
Pharmaceutical 0.0117 0.0114 0.0113 0.0112 0.0117
Average MAE 0.0120 0.0114 0.0105 0.0108 0.0112
Table 3: In- sample mean absolute error (MAE)
In addition to MAE criterion for investigating of forecasting performance, another benchmark (MSE) is applied
and the results are represented in Table (4). Similar to the previous findings about supremacy of Kalman Filter
technique (in random walk model), we can observe that this method has the least mean square error or MSE among
rival techniques and also it shows the best performance in all sectors except Homebuilding and Machinery &
Equipments.
sector
Automobiles and Parts 0.2845 0.3112 0.2145 0.2731 0.2914
Banks & Credit 0.1863 0.1645 0.1518 0.1712 0.1685
Basic Metals 0.3144 0.2953 0.2641 0.2761 0.2822
Chemical 0.2664 0.2418 0.2218 0.2453 0.2719
Food & Beverage 0.3651 0.3462 0.3387 0.3425 0.3712
Homebuilding 0.1628 0.1586 0.1541 0.1517 0.1612
Investment 0.4132 0.3825 0.2894 0.3466 0.3368
598
(17)
(18)

5 Conclusions
Machinery & Equipments 0.1918 0.2317 0.1894 0.1854 0.2019
Petroleum Products 0.3154 0.2985 0.2616 0.2714 0.2916
Pharmaceutical 0.2371 0.2281 0.2085 0.2114 0.2561
Average MSE 0.2737 0.26584 0.22939 0.24747 0.26328
Table 4: In- sample mean square error (MSE) [
Several empirical evidences suggest that unconditional estimates of systematic risk are not stable over the time. Our
results confirm these findings and show overwhelmingly that the betas are unstable and systematically time-varying.
In this paper, we considered three alternative techniques to model this time variation in conditional betas: (i) a
bivariate t-GARCH (1,1) model (ii) two Kalman Filter based approaches, including of random walk (RW) and mean
reverting (MR) and (iii) a bivariate stochastic volatility (SV) technique.
In our analysis, excess returns were forecasted in-sample, by using the conditional beta series that were
generated by each of these techniques. The comparison of the in-sample forecast accuracy of each conditional beta
technique indicates that time-varying sector betas are best described by the use of Kalman Filter in a random walk
process. This result supports the findings of Mergner and Bulla (2008) that Kalman Filter technique is superior to
other alternative approaches. Results that presented in this study advocates further research in this field, such as
using exogenous factors for explaining of time-varying behavior of systematic risks, applying behavioral finance
approaches and building of new methods by modifying the shortcomings of existing techniques.
6 References
Adrian, T., and F. Franzoni. 2009. Learning about beta: Time-varying factor loadings, expected returns, and the
conditional CAPM. Journal of Empirical Finance, 16(4): 537-556
Asai, M., M, McAleer, and J. Yu. 2006. Multivariate stochastic volatility: A Review. Economic Review, 25 (2-3):
145-175.
Bauwens, L., S., Laurent, and J., Rombouts. 2006. Multivariate GARCH Models: A Survey, Journal of Applied
Econometrics, Vol. 21: 79–109
Bollerslev, T. 1986. Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics 31: 307–27.
———. 1987. A conditionally heteroskedastic time series model for speculative prices and rates of return. Review of
Economics and Statistics 69: 542–7.
———. 1990. Modeling the coherence in short-run nominal exchange rates: A multivariate generalized ARCH
model. Review of Economics and Statistics 72, no. 3: 498–505.
Bollerslev, T., R.F. Engle, and J.M. Wooldridge. 1988. A capital asset pricing model with time-varying covariances.
Journal of Political Economy 96, no. 1: 116–31.
Bos, T. and P. Newbold. 1984. An empirical investigation of the possibility of stochastic systematic risk in the
market model. Journal of Business 57, no. 1: 35–41.
Brooks, R.D., R.W. Faff, J., Lee. 1992. The Form of Time Variation of Systematic Risk: Some Australian Evidence,
Applied Financial Economics, 2:191 – 198.
Brooks, R.D., R.W. Faff, and M.D.McKenzie. 1998. Time-varying beta risk of Australian industry portfolios:
comparison of modeling techniques. Australian Journal of Management 23, no. 1: 1–22.
Choudhry, T. 2005. Time-Varying Beta and the Asian Financial Crisis: Investigating the Malaysian and Taiwanese
Firms. Pacific-Basin Finance Journal, 13: 93-118.
Choudhry, T., and H., Wu. 2007. Time-Varying Beta and Forecasting UK Company Stock Returns: GARCH Models
vs. Kalman Filter Method, University of Southampton Press.
Choudhry, T., Lu, L. and K. Peng. 2010. Time-Varying Beta and the Asian Financial Crisis: Evidence from the
Asian Industrial Sectors. Japan and the World Economy, 22: 228-234.
599

Collins, D.W., J. Ledolter, and J. Rayburn. 1987. Some further evidence on the stochastic properties of systematic
risk. Journal of Business 60, no. 3: 425–48.
Danielsson, J. 1994. Stochastic volatility in asset prices, estimation with simulated maximum likelihood. Journal of
Econometrics 64: 375–400.
Doornik, J.A. (2009). An Object-Oriented Matrix Language Ox 6, London: Timberlake Consultants Press.
Durbin, J., and S.J. Koopman. 2001. Time series analysis by state space methods. Oxford Statistical Science Series.
Oxford: Oxford University Press.
Engle, R.F. 1982. Autoregressive conditional heteroskedasticity with estimates of the variance of UK inflation.
Econometrica 50: 987–1008.
Engle, R.F., and K.F. Kroner. 1995. Multivariate simultaneous generalized ARCH. Economic Theory 11: 122–50.
Faff, R.W., D. Hillier, and J. Hillier. 2000. Time varying beta risk: An analysis of alternative modeling techniques.
Journal of Business Finance and Accounting 27, no. 5: 523–54.
Ghysels, E., A.C. Harvey, and E. Renault. 1996. Stochastic volatility.Vol. 14 of Handbook of Statistics, ed. G.S.
Maddala and C.R. Rao, 128–98. Amsterdam: North Holland.
Giannopoulos, K. 1995. Estimating the time varying components of international stock markets’ risk. European
Journal of Finance 1: 129–64.
Groenewold, N., and P. Fraser. 1999. Time-varying estimates of CAPM betas. Mathematics and Computers in
Simulations 48: 531–9.
Harvey, A.C., E. Ruiz, and N. Shephard. 1994. Multivariate stochastic variance models. Review of Economic Studies
61: 247–64.
He, Z., and L. Kryzanowski. 2008. Dynamic betas for Canadian sector portfolios. International Review of Financial
Analysis 17: 1110–1122.
Jacquier, E., N.G. Polson, and P.E. Rossi. 1994. Bayesian analysis of stochastic volatility models. Journal of
Business and Economic Statistics 12: 371–89.
Kim, S., N. Shephard, and S. Chib (1998). Stochastic volatility: Likelihood inference and comparison with ARCH
models. Review of Economic Studies 65 (3), 361-393.
Koopman, S.J., N. Shephard, and J.A. Doornik. 1999. Statistical algorithms for models in state space using SsfPack
2.2. Econometrics Journal 2: 113–66. http://www.ssfpack.com.
Laurent S. (2009). G@RCH 6, Estimating and Forecasting ARCH Models. London: Timberlake Consultants Press.
Li, X. 2003. On unstable beta risk and its modeling techniques for New Zealand industry portfolios. Working Paper
03.01, Massey University Commerce, Auckland, New Zealand.
Lie, F., R. Brooks, and R. Faff. 2000. Modeling the equity beta risk of Australian financial sector companies.
Australian Economic Papers 39: 301–11.
Lintner, J. 1965. The valuation of risky assets and the selection of risky investments in stock portfolios and capital
budgets. Review of Economics and Statistics 47: 13–37.
Mergner, S., J., Bulla. 2008. Time-varying Beta Risk of Pan-European Industry Portfolios: A Comparison of
Alternative Modeling Techniques, European Journal of Finance, Vol. 14:771–802.
Mergner, S. 2009. Applications of State Space Models in Finance. G ttingen University Press.
Pope, P. and M. Warrington. 1996. Time-varying properties of the market model coefficients. Accounting Research
Journal 9: 5-20
Sandmann, G., and S.J. Koopman. 1998. Estimation of stochastic volatility models through Monte Carlo maximum
Likelihood. Journal of Econometrics 87: 271–301.
Sharpe, W.F. 1964. Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of Finance
19:425–42.
Wells, C. 1994. Variable betas on the Stockholm exchange 1971–1989. Applied Economics 4: 75–92.
Yao, J. and J. Gao. 2004. Computer-intensive time-varying model approach to the systematic risk of Australian
industrial stock returns. Australian Journal of Management 29, no. 1: 121–46.
600