Dynamic Effects of Monetary Policy Shocks in Malawi* - African ...

University of Cape TownSchool of economicsDynamic Effects of Monetary Policy Shocks in Malawi Harold P.E. Ngalawa*AbstractThis paper sets out to investigate the process through which monetary policy affects economic activity in Malawi.Using innovation accounting in a structural vector autoregressive model, it is established that monetaryauthorities in Malawi employ hybrid operating procedures and pursue both price stability and high growth andemployment objectives. Two operating targets of monetary policy are identified, viz., bank rate and reservemoney, and it is demonstrated that the former is a more effective measure of monetary policy than the latter.The study also illustrates that bank lending, exchange rates and aggregate money supply contain importantadditional information in the transmission process of monetary policy shocks in Malawi. In addition, it is shownthat the floatation of the Malawi Kwacha in February 1994 had considerable effects on the country’s monetarytransmission process. In the post-1994 period, the role of exchange rates became more conspicuous than beforealthough its impact was weakened; and the importance of aggregate money supply and bank lending intransmitting monetary policy impulses was enhanced. Overall, the monetary transmission process evolved from aweak, blurred process to a somewhat strong, less ambiguous mechanism.Paper Presented at the 14 th Annual Conference of the African Econometric Society,8-10 July 2009, Abuja, Nigeria This paper is part of a PhD Thesis in progress. I thank Professor Nicola Viegi, my supervisor, for his usefulcomments. Any errors are my responsibility, though.*University of Cape Town, School of Economics, Private Bag, Rondebosch 7701, South Africa. Email:hngalawa@yahoo.co.uk / Harold.Ngalawa@uct.ac.za

1.0. INTRODUCTIONWhile it is generally agreed that monetary policy can significantly affect both real economicactivity and prices in the short run and only prices in the long run, considerable debate remainsabout how monetary policy shocks are transmitted. Views differ in the emphasis placed onmoney, credit, interest rates, exchange rates, asset prices and the role of commercial banks andother financial institutions (Taylor, 1995). The differences are prevalent even in individualindustrialized countries where the topic has been a subject of research for many years. Indeveloping countries, the process is even more uncertain (Kamin, Turner, & Van't dack, 1998).In spite of the prominence given to monetary policy, the transmission process in a typicaldeveloping country is not well understood (Montiel, 1991). One of the typical developingcountries, Malawi, is no exception.Monetary policy plays a very important role in the management of the Malawi economy. Asoutlined in the Reserve Bank of Malawi (RBM) Act of 1989, one of the principal objectives ofthe central bank is to influence money supply, credit availability, interest rates and exchangerates in order to ultimately promote economic growth, employment and price stability (MalawiGovernment, 1989). Achieving this objective clearly requires an understanding of the processthrough which monetary policy affects economic activity. There is, however, no study that weare aware of that has quantitatively measured the transmission process of monetary policy inMalawi. This study, therefore, contributes to the literature by filling this gap. The study isolatesmonetary policy autonomous disturbances from other shocks, quantifies their dynamic behaviourand measures the consequent macroeconomic implications using a structural vectorautoregressive model (SVAR) with short run restrictions. Within the same framework, the studyalso assesses how the country‟s monetary policy transmission process was altered by RBM‟smigration from direct to indirect tools of monetary control in the late 1980s and 1990s.Since Sims‟ (1980) pioneering work, vector autoregression models (VARs) are consideredbenchmarks in econometric modelling of monetary policy transmission (Borys & Horvath,2007). While natural experiments would be ideal, the real world does not provide for this optionand SVARs are the only other place where experiments can be performed (Christiano,Eichenbaum, & Evans, 1998). SVAR experiments aimed at measuring the effect of monetarypolicy on economic activity have traditionally involved setting apart monetary policy shocks andtracking the response of macroeconomic variables to the monetary policy impulses.Most studies that have applied SVARs to study the dynamic behaviour of monetary policyshocks have used developed market economies as case studies (Karame & Olmedo, 2002;Bernanke & Mihov, 1998; Sims & Zha, 1998; Bernanke & Mihov, 1996; Piffanelli, 2001). Inrecent years, there has been growing attention on emerging market economies of Latin America,Asia and Easten Europe (Borys & Horvath, 2007; Vonnak, 2005; Disyatat & Vongsinsirikul,2003; Dabla-Norris & Floerkemeier, 2006) and on Australia (Berkelmans, 2005; Brischetto &Voss, 1999). In contrast, quasi-emerging market economies (QEMEs), particularly in subsaharanAfrica, have attracted very little attention.2 | P a g e

Among the few studies undertaken on QEMEs, Mutoti (2006) employed a cointegrated structuralVAR identified with short run and long run restrictions to model the monetary transmissionprocess of post-liberalisation Zambia. Using monthly data for the period 1992-2003, he foundout that much of the output volatility in the country is attributable to aggregate supply shocks,with IS shocks turning to be the most important demand factors for the business cycle. He alsonoted that money demand shocks modestly boost short run output. He further established that inthe short run, changes in consumer prices are accounted for by aggregate supply, money demandand exchange rate shocks whilst at longer horizons, they are mainly underlined by aggregatesupply shocks and modestly by foreign price shocks. The study concluded that a very smallproportion of output variance is due to monetary policy shocks.In a study of Kenya, Maturu (2007) used a structural VAR to argue that interest rate andexchange rate channels are unambigously important channels of monetary policy transmission inthe country besides the traditional money channel. On this basis, he pointed out that there ispotential for signalling monetary policy using the REPO interest rate. In another study of Kenya,Cheng (2006) also used a structural VAR to examine the impact of monetary policy shocks onoutput, prices and the nominal effective exchange rate for the country during the period 1977-2005. Cheng found out that an exogenous increase in the short term interest rate is followed by adecline in prices and an appreciation of the nominal exchange rate, but has insignificant impacton output. He further showed that variations in short term interest rates account for significantfluctuations in the nominal exchange rate and prices, while accounting little for outputfluctuations.In the case of Malawi, there are two theoretical studies, one by Bolnick (1991) and another byPhiri (2002), and no empirical analysis on the country‟s monetary transmission process. TheBolnick study was carried out at the time RBM was converting from direct to indirect tools ofmonetary management. The study investigated how the changing conditions would weaken oralter links in the country‟s monetary transmission process. Unfortunately, Bolnick was unable topredict the emergence of a reasonably competitive financial sector within the foreseeable futureand consequently concluded erroneously that after RBM‟s implementation of indirect monetarycontrols, the transmission of monetary policy shocks in Malawi would be stage-managed by thecentral bank through informal consultations with commercial bank managers. Phiri‟s study, onthe other hand, is not very useful as it falls short of going beyond a theoretical exposition oftextbook transmission channels of monetary policy. For instance, he presents the other assetprice effects channel operating through an efficient stock market as one of Malawi‟s monetarytransmission channels. However, with only eight listed companies at the time of the study, it iscommon knowledge that the Malawi Stock Exchange (MSE) was still is in its infancy to act as aconduit of monetary transmission. As at the end of June 2009, the MSE had 15 listed companiesand was still classified as „infant.‟The rest of the paper is organised as follows. Section 2 is an overview of monetary policy inMalawi since independence in 1964. A methodological framework characterising structuralVARs, identification of the structural shocks, data sources, variable definitions and measurement3 | P a g e

of variables is presented in Section 3. Estimation results and inferences are discussed in Section4. Section 5 presents a summary, conclusion and policy recommendations.2.0. OVERVIEW OF MONETARY POLICY IN MALAWI SINCE 1964The conduct of monetary policy in Malawi since independence can be outlined in three broadlydistinct monetary policy regimes viz., period of financial repression (1964-86), period offinancial reforms (1987-1994) and period of financial liberalisation (post-1994). Atindependence in 1964, the formal banking system which the country adopted from the colonialgovernment was perceived to be primarily interested in serving the needs of an expatriatecommunity, to have little interest in direct lending to local entrepreneurs, and to imposeunreasonably high charges on routine banking services (Gondwe, 2001). To get rid of thesedistortions, direct controls on credit and interest rates were imposed. The agricultural sector, inparticular, was accorded preferential lending rates and quota credit allocations in line withgovernment policy to promote agricultural production. Besides these controls, government alsoadopted a fixed exchange rate system and imposed price ceilings on selected commodities.In the late 1970s, a hostile external environment forced the economy into a deep recession,which persisted through the 1980s. Intensifications of civil war in neighbouring Mozambique, aconsequent flooding of refugees into the country and disruption of a cost effective route to thesea ports of Beira and Nacala; the 1979 oil crisis; and drought in 1980 were some of the factorsthat triggered the recession. The failure of the economy to adjust to these shocks revealedstructural weaknesses in the design of the country‟s macroeconomic framework. Governmentwas forced, therefore, to implement a policy change from the mid 1980s to the 1990s, movingaway from direct to indirect tools of monetary control, among others. A phased financialliberalisation program targeted at enhancing competition and efficiency in the financial sectorwas adopted.The reforms commenced with partial deregulation of lending rates in July 1987 and deposit ratesin April 1988. The partial deregulation allowed commercial banks to determine their ownlending and deposit rates but not to effect any adjustment without prior consultation with thecentral bank. Credit ceilings were abolished in 1988. In January 1990, the authorities announcedthe abolition of preferential lending rates to the agricultural sector. Complete deregulation of theinterest rates occurred in May 1990.The reform program also overhauled the legal and regulatory framework of the banking system,which involved revision of the RBM Act of 1964 and Banking Act of 1965 in May 1989 andDecember 1989, respectively. While the central bank was previously supervising commercialbanks only, the revised Banking Act extended its coverage to include non-bank financialinstitutions (NBFIs), a function that was previously in the hands of the Treasury. In addition,inspection of the financial institutions was broadened to include adherence to prudentialrequirements besides compliance to exchange control regulations.4 | P a g e

Interest Rate (Percent)In line with the revised RBM Act, the central bank introduced two new instruments of monetarypolicy, namely liquidity reserve requirement (LRR) and discount window facility. The discountwindow facility led to the introduction of the bank rate, which has since become a very powerfulindicator of monetary policy. A change in the bank rate is usually followed by near instantaneouscorresponding changes in both lending and deposit rates. Average yields on governmentsecurities also follow the same direction (see Figure 2.1).FIGURE 2.1: Interest Rates in Malawi (2002-2007)60Bank RateBase Rate50Savings Rate91 day TB Rate407-day Call30201002002:6 2003:2 2003:10 2004:6 2005:2 2005:10 2006:6 2007:2 2007:10Source: Reserve Bank of MalawiThe country‟s financial reforms reached near-completion with the floatation of the MalawiKwacha on February 7, 1994. Subsequently, the monetary authorities removed exchange controlregulations, allowed for the establishment of foreign exchange bureaux, introduced foreigncurrency denominated accounts, established a forward foreign exchange market and started thetrading of foreign exchange options and currency swaps. Eight new commercial banks (one ofwhich has since been liquidated) entered the commercial banking sector since 1994, changing thestructure of the market from a duopoly to a fairly competitive sector. The country‟s first discounthouse entered the financial sector in 1998 followed by a second one in 2002.The official position of RBM is that monetary policy in the country utilises a quantitativeoperating target known as reserve money, defined as the sum of currency in circulation, vaultcash and bank deposits with the central bank (Banda, 2004). On the sources side, reserve moneyis defined as net domestic assets (NDA) plus net foreign assets (NFA). In this framework, theforeign exchange market operates on the supply side of reserve money through the effect offoreign exchange transactions on NFA. Banda argues that the rationale for reserve moneytargeting by the central bank is to balance supply and demand conditions of the monetaryaggregate in the money market so as to achieve price stability.As an instrument target of monetary policy, reserve money is used to reflect open marketoperations (OMO) and foreign exchange transactions. In Malawi‟s OMO, the central bankundertakes weekly trading of RBM Bills and Malawi Government Treasury Bills (TBs) in acompetitive bidding system. The former are short-term financial instruments issued by the5 | P a g e

central bank for monetary policy purposes while the latter are short-term financial instrumentsissued by the Government through RBM to borrow funds from the general public for thefinancing of current fiscal deficits and maturing government debt. In addition to the regulartrading, RBM also sales extra securities on an ad hoc basis to discount houses and commercialbanks, referred to as „tap sales‟.While the monetary policy framework in Malawi is officially designated as reserve moneytargeting with OMO playing an important role, the system operates as if the central bank alsotargets short term interest rates through adjustments in the bank rate. To avoid prejudice, thisstudy assumes the central bank targets both the bank rate and reserve money and goes on toempirically determine which of the two is a more effective measure of monetary policy.3.0. METHODOLOGY3.1. Structural VAR FrameworkTo draw the SVAR mainframe, suppose Malawi‟s monetary transmission process is described bya dynamic system whose structural form equation is given by:where is an invertible matrix describing contemporaneous relations among thevariables; is an vector of endogenous variables such that ; isa vector of constants; is an matrix of coefficients of lagged endogenous variables; is an matrix whose non-zero off–diagonal elements allow fordirect effects of some shocks on more than one endogenous variable in the system; and areuncorrelated or orthogonal white-noise structural disturbances i.e. the covariance matrix of isan identity matrix =1. Equation (3.1) can be rewritten in compact form as:where is an finite order matrix polynomial in the lag operator L.At the centre of analysis in similar vector autoregression models are monetary policy shocks andtheir dynamic behaviour in the system. Following Vonnak (2005), we define monetary policyshocks as unexpected deviations from the systematic behaviour of monetary policy. Vonnakexplains changes in monetary policy operating targets as mostly endogenous reactions ofmonetary policy to other types of shocks from the economy given the monetary policy feedbackrule. Tracking down developments in monetary policy goals following a change in an operatingtarget only provides a reflection of the consequences of the shock, which among others, triggeredthe change in the operating target. Evidently, it is important to isolate autonomous disturbancesemanating from monetary policy shocks from other types of shocks, which is carried out inSVARs.6 | P a g e(3.1)(3.2)

3.2. Identification of Structural ShocksThe SVAR presented in the primitive system of equations (3.1) and (3.2) cannot be estimateddirectly due to the feedback inherent in a VAR process (Enders, 2004). Nonetheless, theinformation in the system can be recovered by estimating a reduced form VAR implicit in (3.1)and (3.2). Pre-multiplying equation (3.1) by yields a reduced form VAR of order p, which instandard matrix form is written as:where ; ; and is an vector of error terms assumedto have zero means, constant variances and to be serially uncorrelated with all the right hand sidevariables as well as their own lagged values though they may be contemporaneously correlatedacross equations. The variance-covariance matrix of the regression residuals in equation (3.3) isdefined as. Given the estimates of the reduced form VAR in equation (3.3), thestructural economic shocks are separated from the estimated reduced form residuals by imposingrestrictions on the parameters of matrices and in equation (3.4):which derives from equation (3.3). The orthogonality assumption of the structural innovationsi.e. =1, and the constant variance-covariance matrix of the reduced-form equationresiduals i.e. impose identifying restrictions on and as presented in equation(3.5):Since matrices and are both , a total of unknown elements can be identifiedupon which restrictions are imposed by equation (3.5). To identify A and B,therefore, at least or additional restrictions are required. Theserestrictions can be imposed in a number of ways. One approach is to use Sims‟ (1980) recursivefactorisation based on a Cholesky decomposition of matrix . The approach assumes thatelements of matrix are recursively related and are, therefore, lower triangular. The implicationof this relationship is that identification of the structural shocks is dependent on the ordering ofvariables, with the most endogenous variable ordered last (Favero, 2001). Thus, with a givenordering, the first variable has no contemporaneous relationships with all other variables in themodel, indicating that its reduced form shock is identical to its structural shock; the secondvariable has contemporaneous interactions only with its own and the first structural shock; thethird variable is contemporaneously affected by its own and the first two structural shocks; andso on. In this framework, the system is just (exactly) identified.While there are many models that are consistent with the recursiveness assumption, the approachis nonetheless controversial (Christiano, Eichenbaum, & Evans, 1998). The assumptionsrationalising the ordering of variables are often different in different studies using the same(3.3)(3.4)(3.5)7 | P a g e

variables; and since estimation-results in a VAR identified by Cholesky factorisation differ withordering of variables, these studies tend to be incomparable. Changing the order changes theVAR equations, coefficients and residuals, and there are n! recursive VARs, representing allpossible orderings (Stock & Watson, 2001). The validity of Cholesky factorisation is alsoquestioned in cases where a simultaneity problem among monetary or macroeconomic variablesexists. Following the apparent shortfalls in the approach, many authors have adopted alternativeapproaches to the identification of structural shock (Sims & Zha, 1998; Bernanke & Mihov,1995; Leeper, Sims, & Zha, 1996; Sims, 1986; Bernanke, 1986).More recent literature has used structural factorisation, an approach which uses relevanteconomic theory to impose restrictions on the elements of matrices and (Sims & Zha, 1998;Bernanke & Mihov, 1995; Sims, 1986; Bernanke, 1986). This study adopts a similar approach.The underlying structural model is identified by assuming orthogonality of the structuraldisturbances, ; imposing that macroeconomic variables do not simultaneously react tomonetary variables, while the simultaneous feedback in the reverse direction is allowed for; andimposing restrictions on the monetary block of the model reflecting the operational proceduresimplemented by the monetary policy-maker (Favero, 2001, p. 166).Seven variables are included in the SVAR namely, output , consumer price level ,commercial bank loans , exchange rates , aggregate money supply , bank rateand reserve money . Output and consumer prices enter the SVAR as policy goals;bank rate and reserve money as operating targets; and commercial bank loans, exchange ratesand monetary aggregates as intermediate targets. The structural shocks in equation (3.4) areidentified according to the following scheme:(3.6)8 | P a g e

The non-zero coefficients and in matrices and , respectively, show that any residual jin matrices and , in that order, has an instantaneous impact on variable i. The first twoequations suggest that output and consumer prices are sluggish in responding to shocks tomonetary variables in the economy. This scheme is based on the observation that most types ofreal economic activity may respond only with a lag to monetary variables because of inherentinertia and planning delays (Karame & Olmedo, 2002). Proposed by Bernanke and Mihov(1995), the validity of this argument has been supported by a number studies (Cheng, 2006;Berkelmans, 2005; Vonnak, 2005; Karame & Olmedo, 2002).Commercial bank loans are postulated to be contemporaneously affected by all variables in thesystem. Blundell-Wignall and Gizycki (1992) argue that expectations of future activity form animportant determinant of credit demand. Assuming current output, price level, exchange rates,interest rates, and money supply give some indication of what is expected in the future(Berkelmans, 2005) and that economic agents are indeed forward looking, bank lending mayrespond contemporaneously to all variables in the system.Modelling contemporaneous responses of exchange rates to other variables in an SVAR isrelatively standard across studies. Becklemans (2005) uses a real trade-weighted exchange rateindex in a study of Australia and assumes that the index responds instantaneously to all variablesin the system. In a study of Kenya, Cheng (2006) employs a nominal effective exchange rate andmaintains that the exchange rate responds contemporaneously to all variables in the SVAR.Similarly, Borys and Horvath (2007) in a study of the Czech Republic and Piffanelli (2001) in astudy of Germany assume all variables in the system affect exchange rates instantaneously. Noneof these studies, however, attempts to explicitly rationalise the assumption. While the exchangerate may respond contemporaneously to changes in the level of output and consumer prices inthe case of Malawi, there is no reason to believe that it will also respond contemporaneously tomonetary variables given the lack of depth in the financial sector. This study, therefore, takes adeparture from these studies by postulating that the exchange rate responds with a lag to interestrates, bank loans and monetary aggregates.Some studies include variables that specifically capture external price shocks, most common ofwhich are price shocks. Oil price disturbances are usually singled out among the oil price shocks.Other studies also incorporate international financial market interest rate shocks and the FederalReserve Bank Funds Rate has been widely used as a proxy for these shocks. This study,however, does not explicitly model the external shocks for three reasons. First, Malawi was notexposed to any major external shocks during the study period. Second, it is expected that anydisturbances in the external sector will be ably captured by the exchange rate variable. Third andfinally, the complete SVAR analysed in this study has seven variables, which is already large bySVAR standards. Increasing the number of variables without proper justification, therefore, islikely to decrease the power of the model without making meaningful additions.The fifth equation is a standard money demand function. The equation postulates that moneydemand behaviour in the country makes aggregate money supply respond contemporaneously tochanges in consumer prices, output and interest rates but not to other variables in the system,9 | P a g e

akin to Sims and Zha (1998). The last two equations constitute the monetary policy feedbackrule. While Malawi‟s official position is that it targets reserve money, there is reason to believethat the monetary authorities also target short term interest rates. The study, therefore, assumesthat the country employs hybrid operating procedures, with the bank rate and reserve money asoperating targets. In this framework, both interest rates and reserves are expected to containinformation about monetary policy (Bernanke, 1996). The country‟s effective operating target,accordingly, is determined empirically.The monetary policy feedback rule is drawn on the assumption that information delays impedepolicymakers to react within the present to economic activity and price level developments(Karame & Olmedo, 2002). Both the bank rate and reserve money, therefore, do not respondcontemporaneously to output and consumer prices. The bank rate, specifically, respondscontemporaneously to exchange rates only. While exchange rate data is available real-time, dataon other variables including bank loans and monetary aggregates is usually available to themonetary authorities with a lag. Reserve money, on the other hand, is assumed to respondcontemporaneously to all monetary variables because by its definition, this information isinherent in the monetary aggregate.3.3. AnalysisAnalysis of the SVAR is carried out in three modular experiments. First, a generic modelcomprising the country‟s monetary policy goals and operating targets is estimated. Output andprice level enter the model as policy goals while bank rate and reserve money go in as operatingtargets. The rationale for estimating the generic model is to establish how the two monetarypolicy goals respond to each of the operating targets and to find out if monetary authorities reactto changes in the policy goals. In addition, the estimated generic model is used to determinewhich of the two monetary policy operating-targets has a greater impact on the policy goals. Atthe second level of analysis, bank lending, exchange rates and M2, representing three differenttransmission processes, are separately appended to the generic model and estimated. FollowingDisyatat and Vongsinsirikul (2003) and Morsink and Bayoumi (2001), two sets of impulseresponses are calculated in each case: one with the variable of interest endogenised and the otherwith the variable exogenised. The latter procedure generates a VAR identical to the former,except that it blocks off any responses within the VAR that passes through the variable ofinterest (Disyatat and Vongsinsirikul, Ibid). The two sets of impulse responses are latercompared. The size of the difference in the impulse responses is an indicator of the level ofinformation contained in the variable of interest associated with a particular transmissionchannel. Large differences denote more information in the variable of interest and suggestgreater importance of the related transmission channel. At the third and final level of analysis, allvariables found to hold important information in the country‟s monetary transmission process arepooled and a composite SVAR is estimated. A general identification scheme based on short runrestrictions developed in system of equations (3.6) is used for identifying structural shocks ineach of the models. Analysis in the study is carried out only for the short run to conform with thesubject matter under investigation. Since economists generally agree that monetary policy affects10 | P a g e

only the price level in the long run, there is little value in extending investigation of the monetarytransmission process to cover the long run.3.4. Data, Data Sources and Measurement of VariablesThe study employs monthly time series data for the period 1988:1 to 2005:12. The starting datehas been chosen in conformity with the period when monetary authorities in Malawi migratedfrom using direct measures of monetary control to indirect measures. The cut-off date, on theother hand, corresponds to the date when the latest data on all variables of interest was available.Major sources of data include RBM, National Statistical Office (NSO) (Malawi), MeteorologicalDepartment (Malawi) and the University of Malawi.Bank rate is defined as the rate at which the central bank provides short term loans tocommercial banks and discount houses in its function as a lender of last resort. The variableenters the SVAR as an instrument target of monetary policy. Reserve money is alsoemployed as an instrument target of monetary policy in the SVAR. Components of areidentified as total cash reserves held by the central bank, vault cash in commercial banks andcurrency held by the non-bank public. The variable captures commercial bank loans andadvances and it enters the SVAR as an intermediate target of monetary policy. Similarly,exchange rate ( ) enters the SVAR as an intermediate target of monetary policy. Middlenominal exchange rates of the Malawi Kwacha vis-à-vis the United States Dollar (USD) are usedas a proxy for . Aggregate money supply is measured by the sum of currency in circulation,demand deposits and time deposits (M2). The variable also enters the SVAR as an intermediatetarget of monetary policy.Consumer prices ( ) are measured by the all items national composite consumer price indexwith base year 2000. The variable enters the SVAR as a monetary policy goal. A measure ofoutput ( ) enters the SVAR as a monetary policy goal as well. GDP data (used as a proxy for) for Malawi, however, is only available in annual frequency. This presents a case forinterpolation. Several studies have used interpolated monthly GDP series in SVARs. Amongthem, Cheng (2006) used monthly production data of key sectors in Kenya to interpolate thecountry‟s annual GDP to monthly frequency; and Borys and Horvath (2007) used the quadraticmatchaverage procedure to interpolate GDP from quarterly to monthly frequency in CzechRepublic. This study employs the Friedman method of interpolating time series by related seriesto obtain the required monthly GDP series from annual data (see appendix A for a detailedoutline of the GDP interpolation).All variables, with the exception of interest rates, are expressed in logarithms. They are alsoseasonally adjusted using TRAMO (Time Series Regression with Autoregressive MovingAverage (ARIMA) Noise, Missing Observations, and Outliers) and SEATS (Signal Extraction inARIMA Time Series) with a forecast horizon of 12 months. While the former performsestimation, forecasting, and interpolation of regression models with missing observations andARIMA errors, in the presence of possibly several types of outliers, the latter performs anARIMA-based decomposition of an observed time series into unobserved components11 | P a g e

(Quantitative Micro Software, LLC, 2005). Seasonal adjustment removes cyclical seasonalmovements that are common in time series observed at quarterly and monthly frequency. Theunderlying trend component of the series, however, is retained after the data has been seasonallyadjusted.The variables are further subjected to a test for stationarity, which reveals that they are all I(1)(See Table B1 in Appendix B). The study, however, proceeds with estimation of the SVAR inlevels consistent with standard practice anchored on the canonical paper of Sims, Stock andWatson (1990). The Sims, Stock and Watson paper demonstrates in part that the commonpractice of attempting to transform models to stationary form by difference or cointegrationoperators whenever it appears likely that the data are integrated is unnecessary because statisticsof interest often have distributions that are unaffected by nonstationarity, which suggests thathypotheses can be tested without first transforming to stationary regressors. The issue, accordingto the study, is not whether the data are integrated, but rather whether the estimated coefficientsor test statistics of interest have a distribution which is nonstandard if in fact the regressors areintegrated.The Sims, Stock and Watson (1990) findings have been generally accepted and widely adoptedin the structural VAR literature. In a study of the German Bundesbank, Bernanke and Mihov(1997, p. 1037) maintain that „we include the output, price, and reserves measures in levelsdespite their nonstationarity, as has become standard practice in VAR studies.‟ They furtherpoint out that the levels specification yields consistent estimates whether cointegration exists ornot, whereas a differences specification is inconsistent if some variables are cointegrated. Thepreference of VARs in levels, according to Kim and Roubini (2000) and Becklelmans (2005),can be explained, at least in part, by a reluctance to impose possibly incorrect restrictions on themodel. Kim and Roubini stress that if false restrictions are imposed, the resulting inferenceswould be incorrect as well. Other studies that have followed this approach of estimatingstructural VARs in levels even when the variables are I(1) include Piffanelli (2001), Dungey andPagan (2000), Kim (1999), Brischetto and Voss (1999), Bernanke and Mihov (1998),Ramaswamy and Sloek (1998), and Sims (1992), among many others.Some studies, however, have opted to transform non-stationary data prior to estimating structuralVARs. A common element in a large number of these studies is the focus on prevalentrelationships in the variables of interest beyond the short run. A standard approach in this case isto explicitly model I(1) variables and co-integrating relationships present in the data by imposingcointegrating restrictions on a VAR in levels. Rationalising the approach, Haug et al (2005)argue that for the long run, a vector error correction model (VECM) estimation produces moreprecise and efficient parameter estimates than a VAR in levels. Haug et al nonetheless concedethat for the short run, the VAR parameters are estimated consistently by least-squares if the VARis estimated in levels without imposing co-integrating restrictions present in the data. Davidson(1998) and Johansen (1988) further add that when supplemented with cointegration analysis, theVAR technique allows for a rigorous modelling of the long-run relationship of non-stationaryvariables. Among many others, some studies that have used cointegration analysis to identify12 | P a g e

long-run relationships in a linear cointegrating model with I(1) variables include Garratt et al(2003), Ehrmann (1998), Lutkepohl and Wolters (1998) and King et al (1991).While the debate on whether to transform models to stationary form by difference orcointegration operators when dealing with I(1) variables appears to lean towards the Sims, Stockand Watson (1990) conclusion, there are other authors that appeal to the traditional approach oftransforming the data to stationary regressors prior to estimation regardless of whether the pointof focus is long run or short run relationships (see, for example, Enders (2004)). To illustrate thatresults obtained from the two methodologies are not diametrically opposite to each other, we alsoestimate the composite model using a cointegrated SVAR (see Appendix B for details).Appendix B demonstrates that while there may be some differences, as expected, estimationresults from the cointegrated SVAR are on the whole similar to what we obtain from estimationin levels.Malawi had credit ceilings until 1988, interest rate controls until 1990 and a fixed exchange ratepeg until 1994. Accordingly, free movement of monetary variables was not allowed for until inthe mid 1990s (See Figures 3.1-3.6). While the monetary authorities have targeted reserve moneyas a way of influencing M2, the two variables do not appear to move together (see Figure 3.1).There is some close correlation, though, between M2 on the one hand and commercial banklending (see Figure 3.2) and GDP (see Figure 3.3), on the other. There appears to be no obviousrelationship between movements in the exchange rate and the bank rate (see Figure 3.4). Theexchange rate and CPI show very high correlation (see Figure 3.5) while the bank rate andcommercial bank lending display an inverse relationship (see Figure 3.6).4.0. ESTIMATES AND INFERENCES4.1. Generic ModelInvestigation of the monetary transmission process commences with a simple four variablegeneric model. The vector of endogenous variables included in the model is presented inequation (4.1):Following the identification scheme in system of equations (3.6), the equation separatingstructural economic shocks from the estimated reduced form residuals for the generic model ispresented as:(4.1)(4.2)13 | P a g e

Selection of the optimal lag length is guided by established criteria 1 . Akaike and Hannan-QuinnInformation Criteria (AIC and HIC, respectively) suggest a lag length of order 42 whileSchwartz Information Criterion (SIC) suggests a lag length of order two. The problem with theformer is that it uses up all degrees of freedom and renders the estimated VAR unstable with 83inverse roots of the characteristic autoregressive (AR) polynomial lying outside the unit circle.The latter, on the other hand, performs well in a robustness check and it is adopted. At thechosen lag length (of order two), all the eight inverse roots of the characteristic AR polynomialhave modulus less than one and lie inside the unit circle, indicating that the estimated VAR isstationary or stable (see Table C1 in Appendix C). A VAR lag exclusion Wald test also revealsthat all endogenous variables in the model are jointly significant at each lag length for allequations collectively. Separately, all equations are significant at lag length of order one while atlag length of order two, only consumer price and output equations are insignificant (see Table C2in Appendix C). Since the SIC imposes a harsher penalty for adding more regressors than AIC,the criterion is deemed more appropriate for determining lag length in a VAR.Departures from established criteria for the determination of optimal lag length are notuncommon in the literature. Piffanelli (2001) argued that 12 lags are appropriate with monthlydata to avoid seasonality problems. She, however, included six lags in a study of Germanywithout rationalisation. Becklelmans (2005) used a lag length of order three in a study ofAustralia „because it provides reasonable dynamics without shortening the estimation sample toomuch.‟ Vonnak (2005) used three lags in a study of Hungary „because they were enough toproduce uncorrelated residuals, based on the evidence of the multivariate-LM test.‟ Usingquarterly data in a study of Thailand, Disyatat and Vongsinsirikul (2003) settled for two lagswhile the optimal lag length under various criteria appeared to be one because they felt onequarter is too short a period to capture the dynamics of a system. This study, however, finds tojustification to depart from the established criteria.Before making inferences on the structural shocks in the model, we analyse correlations betweenmovements in the bank rate and reserve money and their corresponding recovered structuralshocks to ascertain if the monetary policy shocks are reasonable. Figures 4.1 and 4.2 presentplots of recovered bank rate and reserve money structural innovations against month-on-monthgrowth rates of the bank rate and the logarithm of reserve money, respectively, with therecovered structural shocks plotted on the primary vertical axis and the monetary policyoperating targets on the secondary vertical axis. Positive structural innovations of the bank rateand negative structural innovations of reserve money are associated with monetary policytightening while negative structural innovations of the bank rate and positive structuralinnovations of reserve money are associated with monetary policy loosening.The charts reveal that there is some correlation in the movements of the monetary policyoperating targets and their respective recovered innovations. The correlations are, however, morepronounced between the bank rate and its recovered structural shocks compared to reservemoney and its recovered structural shocks. Reliability of the structural shocks is also ascertainedby assessing the efficiency of the structural coefficients estimated in the SVAR. Table C3 in1 This approach is applied in all subsequent models.15 | P a g e

Percent(RM-Innovations)Percent (BR-Innovations)Appendix C shows that all structural estimates of the coefficients in matrices A and B of thegeneric model have standard errors that are less than one, implying that the coefficients areefficient. Accordingly, the measured structural shocks can be relied upon as being a truereflection of reality.Figure 4.1: Bank Rate Movements and Related Recovered Structural Innovations15.010.05.00.0-5.0-10.0-15.0BR-Innovations1988:3 1989:12 1991:9 1993:6 1995:3 1996:12 1998:9 2000:6 2002:3 2003:12 2005:920.010.00.0-10.0Log(BR)Figure 4.2: Reserve Money Movements and Related Recovered Structural Innovations10.50.19.58.5-0.17.5RM-Innovations6.5Log(RM)-0.35.51988:6 1990:12 1993:6 1995:12 1998:6 2000:12 2003:6 2005:12PercentageLog(RM)Next we analyse the response of the central bank to shocks in the policy goals. Figure 4.3presents impulse responses of the bank rate and reserve money to structural one standarddeviation innovations in output and consumer prices over a five-year horizon. Impulse responsesof output and consumer prices to own shocks are also presented in the same figure. The timescale measured on the primary horizontal axis is in months and the dashed lines are analyticconfidence intervals obtained from variance-covariance matrices after the final iteration. Anoutput shock corresponding to an unanticipated 11 percent increase in output and a supply shockequivalent to an unexpected 2.2 percent rise in consumer prices trigger significant responses bythe central bank, illustrating that monetary authorities in Malawi are concerned with bothinflation and economic growth in line with the RBM Act of 1989. The bank responds to theoutput shock by loosening monetary policy through a decrease in the bank rate to further buoythe output growth. In response to the supply shock, monetary policy is tightened by raising thebank rate to arrest the increase in the consumer prices. The central bank‟s response with regardto reserve money, however, is surprising. Following the sudden increase in output, reservemoney declines while the unexpected rise in consumer prices triggers an increase in reservemoney.16 | P a g e

FIGURE 4.3: Impulse Responses of Bank Rate and Reserve Money: The Generic ModelResponse of BR to GYResponse of BR to CPResponse of GY to GY33.1222.0811.0400.00-1-1-.04-210 20 30 40 50 60-210 20 30 40 50 60-.0810 20 30 40 50 60Response of RM to GYResponse of RM to CPResponse of CP to CP.12.12.06.08.08.04.04.04.02.00.00.00-.02-.04-.04-.04-.0810 20 30 40 50 60-.0810 20 30 40 50 60-.0610 20 30 40 50 60To analyse how monetary policy goals are affected by shocks to the operating targets, we plotimpulse responses of output and consumer prices to structural one standard deviation shocks inthe bank rate and reserve money. Figure 4.4 reveals that a monetary policy shock correspondingto an unanticipated increase in the bank rate of about 2.2 percent leads to a decline in output,which bottoms after 5 months at 1.4 percent below baseline. The price level, however, respondsto the monetary tightening with an increase, insignificant though, which is maintained even afterfive years. This finding, referred to as the „price puzzle,‟ is common in the literature. Some of thestudies that have reported the puzzle include Weitong (2007), Kugler et. al. (2004), Disyatat andVongsinsirikul (2003), Piffanelli (2001), Mihira and Sugihara (2000), Clarida and Gertler(1996), Bernanke and Mihov (1996) and Sims (1992).Several explanations to the price puzzle have been suggested. Disyatat and Vongsinsirikul(2003) argue that the failure to include a rich enough specification of the information available topolicy makers is what causes the puzzle to show up. They maintain that if policy makers are ableto observe variables that contain useful information about future prices, but those variables areleft out of the model, a monetary tightening may be associated with higher prices because theypartly reflect systematic policy responses to information indicating that inflation is on the way.Empirical evidence, however, does not support the Disyatat-Vongsinsirikul hypothesis. In astudy of Germany, Sims (1992) added a number of variables, including commodity prices andexchange rates in his system of equations to control for unanticipated future inflation after he hadencountered the price puzzle. However, the perverse price response persisted. Piffanelli (2001)argues that the price puzzle may occur if an incorrect operating target is used in the analysis. Inher study of Germany, the price puzzle appears when the call rate is used and it disappears whenthe Lombard rate is used. A similar finding is reported by Bernanke and Mihov (1996).17 | P a g e

Figure 4.4 also shows that an expansionary monetary shock equivalent to a 7.6 percent suddenincrease in reserve money causes an increase in output, peaking at 1.4 percent above baselineafter 15 months. The price puzzle, however, disappears with reserve money as an operatingtarget. Consumer prices respond to the unexpected increase in reserve money with an increasewhich peaks at 0.4 percent above baseline after 10 months. Overall, shocks to either of themonetary policy operating targets attract significant output responses and insignificant consumerprice responses, suggesting that monetary factors may not be primary determinants of inflation inMalawi. This finding is supported by the preponderant weight of food costs (58.1 percent) in therepresentative basket of commodities used for measuring national consumer price indices, whichindicates that structural rigidities in food production may be more a important cause of inflationthan monetary variables.FIGURE 4.4: Impulse Responses of Output and Consumer Prices: The Generic ModelResponse of GY to BRResponse of GY to RMResponse of BR to BR.12.123.08.082.04.041.00.000-.04-.04-1-.0810 20 30 40 50 60-.0810 20 30 40 50 60-210 20 30 40 50 60Response of CP to BRResponse of CP to RMResponse of RM to RM.06.06.12.04.04.08.02.00.02.00.04-.02-.02.00-.04-.04-.04-.0610 20 30 40 50 60-.0610 20 30 40 50 60-.0810 20 30 40 50 60To determine the relative importance of each structural innovation in explaining fluctuations ofthe variables in the generic model, Table 4.1 presents variance decompositions for each variablein the model over a five-year forecast horizon. Given the two policy goals, fluctuations in boththe bank rate and reserve money are dominated by consumer prices, connoting that the principalobjective of monetary policy in the country is price stability. While shocks to consumer pricesaccount for 13.1 percent of the bank rate fluctuations after 6 months, 18.6 percent after a yearand 21.9 percent after 2 years, output shocks account for 8.2 percent of the bank rate fluctuationsafter 6 months, 14.7 percent after a year and 17.4 percent after 2 years. Shocks to consumerprices also account for 3.4 percent of the reserve money fluctuations after 6 months, 10.7 percentafter a year and 25.5 percent after 2 years while shocks to output account for 4.8 percent of thereserve money fluctuations after 6 month, 9.8 percent after a year and 16.3 percent after twoyears.18 | P a g e

TABLE 4.1: Variance Decomposition for the Generic ModelVariance Decomposition of GYMonth GY CP BR RM1 100 0 0 06 92.3109 2.26977 4.85201 0.5672812 76.8961 11.7495 7.83193 3.5224824 55.1723 30.9392 5.84306 8.0454536 45.2397 42.9679 4.38645 7.4059848 40.2602 49.4123 4.79446 5.5330560 37.3931 52.581 5.96782 4.0581Variance Decomposition of CPMonth GY CP BR RM1 0.00429 99.9957 0 06 8.83843 89.5588 0.70214 0.9006512 17.2802 79.9114 1.4836 1.3247224 23.9426 71.2309 3.73211 1.0944536 26.3566 67.0872 5.87991 0.6763348 27.408 64.4648 7.65085 0.4763560 27.9006 62.5823 9.06388 0.45322Variance Decomposition of BRMonth GY CP BR RM1 0 0 100 06 8.22009 13.0522 76.6691 2.0586212 14.6781 18.6407 60.9944 5.686724 17.4205 21.865 50.6677 10.046836 17.7836 22.7513 47.3985 12.066648 17.8461 23.0723 46.0195 13.062260 17.8509 23.1937 45.3665 13.5889Variance Decomposition of RMMonth GY CP BR RM1 0 0 0.62107 99.37896 4.82833 3.42472 0.25099 91.49612 9.83185 10.6703 0.17049 79.327324 16.3189 25.4689 0.50491 57.707436 20.3133 36.1984 1.78088 41.707448 22.8209 42.7248 3.50981 30.944560 24.3806 46.4778 5.22628 23.915419 | P a g e

Table 4.1 also reveals that the difference in the proportion of fluctuations in output attributedseparately to the bank rate and reserve money is not pronounced. The bank rate, however,accounts for a notably larger proportion of the fluctuations in consumer prices than reservemoney. On the whole, the preliminary indication is that the bank rate is a more effective tool ofmonetary policy than reserve money. While bank rate shocks account for 4.8 percent of thefluctuations in output after 6 months, 7.8 percent after a year, 5.8 percent after 2 years and 6percent after five years, reserve money shocks account for 0.6 percent of the output fluctuationsafter 6 months, 3.5 percent after a year, 8 percent after 2 years and 4.1 percent after five years.This shows that interest rate shocks account for a larger proportion of the fluctuations in outputup to a year and, thereafter, reserve money shocks are attributed for most of the variations inoutput. Reserve money shocks account for only 0.9 percent of the fluctuations in consumerprices after 6 months, 1.1 percent after two years and 0.5 percent after 4 years while bank rateshocks account for 0.7 percent of the fluctuations in consumer prices after 6 months, 3.7 percentafter 2 years and 7.7 percent after 4 years, illustrating that the bank rate accounts for most of theconsumer price variations given the two operating targets.4.2. Channels of Monetary TransmissionIn order to unfold the monetary transmission process, analysis moves away from the genericmodel to examination of more specific transmission channels. Three channels are considered,viz., bank lending, exchange rates and money effect, with particular attention to measuring theimportance of each channel in the transmission process.4.2.1. Bank Lending ModelThe bank lending model is a component of the credit channel of monetary transmission. Theunderlying argument in the credit channel is that asymmetric information and costly enforcementof contracts create agency problems in financial markets (Bernanke & Gertler, 1995). Twomechanisms explain this process: balance sheet and bank lending. The balance sheet modeloperates through the net worth of business firms (Mishkin, 1995) and transmission occurs eitherthrough equity prices or interest rates and firms‟ cash-flows. These affect the severity of adverseselection and moral hazard problems, which in turn impact on lending, investment and output.Owing to data constraints, the balance sheet channel is not pursued further in this study.The bank lending model, on the other hand, operates through quantity rather than price of credit.A monetary policy shock is assumed to be transmitted through changes in bank reserves, the totalamount of available bank credit, and bank lending. The channel presumes that firms facinginformational frictions in financial markets rely on bank loans for external finance because it isprohibitively expensive for them to issue securities in the open market (Disyatat &Vongsinsirikul, 2003). A decline in available bank credit, therefore, adversely affectsinvestments and output. Appending commercial bank loans to equation (4.1) transforms thegeneric model to a bank lending model and the corresponding vector of endogenous variablesbecomes:20 | P a g e(4.3)

The SVAR under investigation in equation (4.3) comprises five variables, which include output,consumer prices, bank loans, bank rate and reserve money. In line with the identification schemein system of equations (3.6), the bank lending model is identified according to the followingscheme:(4.4)Figure 4.5 presents impulse responses of output, consumer prices and bank loans to innovationsin the bank rate, reserve money and bank lending. The figure shows that a bank rate shockequivalent to an unanticipated increase in the bank rate of about 2.2 percent causes bank lendingto decline, bottoming at 2 percent below baseline after 18 months. This response is significantbetween 6 and 24 months. A reserve money shock, on the other hand, corresponding to a 7.2percent sudden increase in reserve money leads to an increase in bank loans, peaking at 1.5percent above baseline after 3 years. This response, however, is not significant. An unexpected5.5 percent rise in bank loans, on the other hand, leads to an increase in both output andconsumer prices.FIGURE 4.5: Impulse Responses for the Bank Lending ModelResponse of BL to BLResponse of BL to BRResponse of BL to RM.08.08.08.04.04.04.00.00.00-.04-.04-.04-.0810 20 30 40 50 60-.0810 20 30 40 50 60-.0810 20 30 40 50 60Response of GY to BLResponse of CP to BLResponse of BL to CP.06.08.12.04.04.08.04.02.00.00.00-.02-.04-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0810 20 30 40 50 6021 | P a g e

To determine the importance of the bank lending channel, impulse responses of consumer pricesand output to bank rate and reserve money shocks are plotted in each case with two scenarios:endogenous and exogenous bank lending. The case of exogenous bank lending blocks offresponses that pass through bank loans while the case of endogenous bank loans allows banklending to transmit the monetary policy shocks. Figures 4.6 shows that in all the four instances,there is a considerable difference in the size of impulse responses when bank lending isexogenous and when it is endogenous. This provides preliminary evidence that bank lendingcontains important additional information in the country‟s monetary transmission process. In linewith theoretical expectations, output decreases following a sudden increase in the bank rate andincreases following an unexpected increase in reserve money while consumer prices go up inresponse to an unanticipated increase in reserve money. The response of consumer prices to anunexpected rise in the bank rate continues to show the price puzzle, dissipating faster thoughwhen bank lending is endogenous.FIGURE 4.6: Response of Output and Consumer prices to Bank Rate and Reserve MoneyShocks with Endogenous and Exogenous Bank Lending0.030.020.010.00-0.01-0.02-0.03Response of GY to BR-Shock1 7 13 19 25 31 37 43 49 550.020.020.010.010.00-0.01-0.01Response of GY to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56BL-EndogenousBL-ExogenousBL-EndogenousBL-Exogenous0.030.030.020.020.010.010.00Response of CP to BR-Shock1 7 13 19 25 31 37 43 49 55BL-EndogenousBL-Exogenous0.0080.0060.0040.0020.000-0.002-0.004-0.006-0.008-0.010Response of CP to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56BL-EndogenousBL-Exogenous22 | P a g e

4.2.2. Exchange Rate ModelTaylor (1995), Obstefield and Gertler (1995) and others have drawn attention to monetary policyoperating through exchange rates and net exports. Monetary policy can influence the exchangerate through interest rates, direct intervention in the foreign exchange market or inflationaryexpectations. The changes in the exchange rate, in turn, affect aggregate demand through the costof imported goods, the cost of production and investment, international competitiveness andfirms‟ balance sheets in the case of high-liability dollarisation (Dabla-Norris and Floerkemeier,2005). We investigate the channel by appending the exchange rate variable , to the genericmodel. The vector of endogenous variables in the exchange rate model is, accordingly, presentedas follows:The five variables in the model are output, consumer prices, exchange rates, bank rate andreserve money. In line with system of equations (3.6), the model is identified according to thefollowing scheme:(4.5)(4.6)Figure 4.7 presents impulse responses of exchange rates to own, bank rate and reserve moneyshocks and responses of output and consumer prices to innovations in exchange rates. Amonetary tightening corresponding to an unexpected 2.2 percent increase in the bank rate causesthe domestic currency to appreciate, moving 1.5 percent below baseline after 3 years. Theresponse, however, is insignificant. Contrary to theoretical expectations, the exchange rateresponds to a reserve money shock equivalent to a 7.6 percent sudden increase in reserve moneywith an appreciation, moving 1 percent below baseline after a year. This response is alsoinsignificant. An exchange rate shock equivalent to a depreciation of the domestic currency by5.5 percent, however, attracts significant responses in both consumer prices and output.Consumer prices rise, peaking at 4 percent above baseline after 3 years while output declines inthe first 4 months and rises thereafter, peaking at 4.3 percent above baseline after 4 years.In spite of the weak responses of exchange rates to innovations in monetary policy operatingtargets, Figure 4.8 demonstrates that impulse responses of output and consumer prices to bankrate and reserve money shocks are reasonably different when exchange rates are exogenous towhen they are endogenous, indicating that inclusion of the exchange rate provides importantadditional information to the monetary transmission process.23 | P a g e

FIGURE 4.7: Impulse Responses for the Exchange Rate ModelResponse of XR to XRResponse of XR to BRResponse of XR to RM.08.08.08.04.04.04.00.00.00-.04-.04-.04-.0810 20 30 40 50 60-.0810 20 30 40 50 60-.0810 20 30 40 50 60Response of GY to XR.08Response of CP to XR.08Response of XR to CP.12.06.04.08.04.04.02.00.00.00-.02-.04-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0810 20 30 40 50 60FIGURE 4.8: Responses of Output and Consumer prices to Bank Rate and Reserve MoneyShocks with Endogenous and Exogenous Exchange Rates0.010.00-0.01-0.02-0.03Response of GY to BR-Shock1 7 13 19 25 31 37 43 49 55XR-EndogenousXR-Exogenous0.020.020.010.010.00-0.01-0.01Response of GY to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56XR-EndogenousXR-Exogenous0.010.010.00-0.01-0.01-0.02Response of CP to BR-Shock1 7 13 19 25 31 37 43 49 55XR-EndogenousXR-Exogenous0.0040.0020.000-0.002-0.004-0.006-0.008-0.010Response of CP to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56XR-EndogenousXR-Exogenous24 | P a g e

4.2.3. The Money Effect ModelAn alternative channel of monetary transmission is the monetarist view. The channel downplaysthe role of interest rates and liquid asset adjustment in the transmission mechanism, reducing theprocess to a direct link between changes in aggregate money supply and absorption (Bolnick,1991). According to this view, prices and output respond to monetary impulses becausehouseholds and businesses fail to anticipate or perceive correctly all of the future implications ofpast and current actions (Meltzer, 1995). These misperceptions occur primarily because of theexistence of a time lag between observing the impulses and being able to distinguish betweenpermanent and transitory impulses and real and nominal shocks. A monetary shock, therefore,brings a wedge between money supply and money demand, which triggers adjustments inportfolio holdings that in turn alter spending decisions. We use aggregate money supply (M2) asan indicator of the money effect. Appending to the generic model, the vector of endogenousvariables in the money effect model is presented as:where the five variables in the model are output, consumer prices, M2, bank rate and reservemoney. Following the identification scheme in system of equations (3.6), the model is identifiedas:(4.7)(4.8)Figure 4.9 presents impulse responses of M2 to own, bank rate and reserve money shocks andresponses of output and consumer prices to M2 shocks. A monetary tightening equivalent to a2.2 percent unexpected increase in the bank rate leads to a significant increase in M2. A reservemoney shock corresponding to a sudden 7.2 percent increase in reserve money, however, triggersno response in M2. A possible explanation for this occurrence is the dominance of commercialbanks in the trading of government securities. A sudden change in reserve money arising fromOMO transactions changes bank reserves proportionately without significantly affectingcurrency and, term and demand deposits, except for the interest component in maturingsecurities. Accordingly, aggregate money supply is insignificantly affected by the reserve moneyshock.Both output and consumer prices respond significantly to unexpected changes in M2. Anunanticipated 6.1 percent increase in M2 is followed by a rise in output, which peaks at 2.4percent above baseline after 10 months and is significant up to 2 years. Consumer prices respondto the monetary expansion with an initial price increase, peaking at 0.6 percent above baselineafter 8 months. The response is significant up to five months.25 | P a g e

Figure 4.9:Responses to Structural One Standard Deviation Innovations: The MoneyEffect ModelResponse of M2 to M2Response of M2 to BRResponse of M2 to RM.08.08.08.06.06.06.04.04.04.02.02.02.00.00.00-.02-.02-.02-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60Response of GY to M2Response of CP to M2Response of M2 to CP.06.08.12.04.06.08.02.04.04.00.00-.02.02.00-.02-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60To determine the importance of the money effect model, Figure 4.10 presents impulse responsesof consumer prices and output to bank rate and reserve money shocks with two scenarios:endogenous and exogenous M2. The figure confirms that M2 contains important additionalinformation in the monetary transmission process, which is more pronounced in the responses ofoutput to bank rate shocks and consumer prices to reserve money shocks.FIGURE 4.10:Response of Output and Consumer prices to Bank Rate and ReserveMoney Shocks with Endogenous and Exogenous M20.030.020.010.00-0.01-0.02-0.03Response of GY to BR-Shock1 7 13 19 25 31 37 43 49 55M2-EndogenousM2-Exogenous0.010.000.000.000.000.000.000.00Response of GY to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56M2-EndogenousM2-Exogenous0.030.030.020.020.010.010.00Response of CP to BR-Shock1 7 13 19 25 31 37 43 49 55M2-EndogenousM2-Exogenous0.0030.0020.0010.000-0.001-0.002Response of CP to RM-Shock1 6 11 16 21 26 31 36 41 46 51 56M2-EndogenousM2-Exogenous26 | P a g e

4.3. The Composite ModelPreliminary indications from the preceding section suggest that bank lending, exchange rate andmoney effect channels contain important additional information for the country‟s monetarytransmission process. Putting everything together, a composite model of monetary transmissionin Malawi can be drawn with the following vector of endogenous variables:which is identified according to system of equations (3.6). Impulse responses for theconsolidated model over a five year period are presented in Figure D1 in Appendix D. The figureillustrates that bank lending and money effect are important channels of monetary transmissionin Malawi but the transmission process is somewhat weak. Among the three intermediate policytargets, none responds significantly to reserve money shocks while only bank lending and M2respond significantly to bank rate shocks, although the M2 response is only marginallysignificant. Bank lending responds to a 2.2 percent sudden increase in the bank rate with adecline, bottoming at 1.7 percent below baseline after 2 years. The response is significantbetween 12 and 30 months. M2 responds to the shock with an instantaneous decline of 0.8percent, before rising in the next 6 months and declining thereafter. The response is marginallysignificant between 16 and 24 months.Output responds significantly to unexpected changes in both bank lending and M2. Anunanticipated 5.7 percent increase in bank lending causes output to rise, peaking at 1.3 percentabove baseline after 15 months. A sudden 5.8 percent increase in M2 also causes output to rise,peaking at 1.6 percent above baseline after 5 months. Consumer prices, however, respondinsignificantly to shocks emanating from both bank lending and M2, consistent with earlierfindings.The money effect channel is strengthened by significant responses of M2 to bank lending andexchange rate shocks. In contrast, the exchange rate channel is not well established. Exchangerates respond insignificantly to all monetary variables in the model but they prompt significantresponses in both output and consumer prices. Thus, while there is no evidence that they aredriven by monetary policy shocks, exchange rates are an important determinant of output andconsumer prices. On this basis, it is probable that exchange rates are exogenously determined inthe model. To ascertain this claim, the composite model is re-estimated with the exchange ratetreated as an exogenous variable and similar results are obtained (see Figure D2 in Appendix D).Historical events suggest that financial sector operations during the pre-1994 period wereconsiderably different in comparison to the post-1994 period. The country had credit ceilingsuntil 1988, direct interest rate controls until May 1990 and a fixed exchange rate peg untilFebruary 1994. In the post-1994 period, numerous financial innovations emerged, the number ofcommercial banks increased considerably and the financial sector become reasonablycompetitive. Cognisant that the impact of financial sector operations on economic activity mayhave also changed in the post-1994 period, the composite model (with endogenous exchange27 | P a g e(4.9)

ates) is re-estimated with the sample period truncated, starting instead from 1994:03. Thetruncation date is chosen to separate the periods of fixed exchange rate peg (pre-1994:03) andfloating exchange rates (post-1994:02). Impulse responses for the model with a truncated sampleare presented in Figure 4.11. While the patterns are broadly similar to the full sample patterns,there are notable differences as well.FIGURE 4.11: Impulse Responses – Composite Model with Truncated SampleResponse of GY to GYResponse of GY to CPResponse of GY to BLResponse of GY to XRResponse of GY to M2Response of GY to BRResponse of GY to RM.15.15.15.15.15.15.15.10.10.10.10.10.10.10.05.05.05.05.05.05.05.00.00.00.00.00.00.00-.05-.05-.05-.05-.05-.05-.0510 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of CP to GYResponse of CP to CPResponse of CP to BLResponse of CP to XRResponse of CP to M2Response of CP to BRResponse of CP to RM.04.04.04.04.04.04.04.02.02.02.02.02.02.02.00.00.00.00.00.00.00-.02-.02-.02-.02-.02-.02-.02-.04-.04-.04-.04-.04-.04-.0410 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of BL to GYResponse of BL to CPResponse of BL to BLResponse of BL to XRResponse of BL to M2Response of BL to BRResponse of BL to RM.08.08.08.08.08.08.08.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.04-.04-.04-.04-.04-.04-.04-.08-.08-.08-.08-.08-.08-.0810 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of XR to GYResponse of XR to CPResponse of XR to BLResponse of XR to XRResponse of XR to M2Response of XR to BRResponse of XR to RM.08.08.08.08.08.08.08.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.04-.04-.04-.04-.04-.04-.04-.08-.08-.08-.08-.08-.08-.0810 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of M2 to GYResponse of M2 to CPResponse of M2 to BLResponse of M2 to XRResponse of M2 to M2Response of M2 to BRResponse of M2 to RM.08.08.08.08.08.08.08.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.04-.04-.04-.04-.04-.04-.0410 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of BR to GYResponse of BR to CPResponse of BR to BLResponse of BR to XRResponse of BR to M2Response of BR to BRResponse of BR to RM444444422222220000000-2-2-2-2-2-2-2-4-4-4-4-4-4-410 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60Response of RM to GYResponse of RM to CPResponse of RM to BLResponse of RM to XRResponse of RM to M2Response of RM to BRResponse of RM to RM.1.1.1.1.1.1.1.0.0.0.0.0.0.0-.1-.1-.1-.1-.1-.1-.110 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 6010 20 30 40 50 60First, the response of exchange rates to unexpected changes in the bank rate is significant in thetruncated sample. This is not surprising since the exchange rate was flexible during the entire28 | P a g e

post-1994 period, which allowed the Malawi Kwacha to respond freely to monetary variables.The response of output to exchange rate shocks, while still significant, is now less pronouncedcompared to the full sample. Thus, the impact of exchange rate shocks on policy goals is weakerin the truncated model although the exchange rate as a monetary policy transmission channel isnow apparent. Second, the significant response of M2 to bank rate shocks is more pronounced inthe truncated sample. This underlines the importance of monetary policy in the flexible exchangerate regime. Third, the significant output response to unexpected changes in bank lending is morepronounced in the truncated model, highlighting the importance of bank lending as a standalonechannel of monetary transmission.To determine the proportion of fluctuations in a given variable that is caused by different shocks,variance decompositions of each variable in the composite model with a truncated sample arecomputed at forecast horizons of 1 to 5 years (see Table 4.2). The table shows that besides ownshocks, output fluctuations are largely attributed to M2 up to about a year, exchange rates atabout 2 years and bank lending from about 3 years and beyond. Collectively, bank lending,exchange rates and M2 account for 8.12 percent of the fluctuations in output after a year, 19.4percent after 2 years, 28 percent after three years and 36.9 percent after 5 years. Excluding ownshocks, variations in consumer prices are mostly accounted for by exchange rates up to about 3years and by bank lending thereafter. M2 accounts for less than 1 percent of the fluctuations inconsumer prices across the forecast horizon, implying that shocks in aggregate money supply arenot responsible for inflation in Malawi. Consistent with earlier findings, consumer prices accountfor a larger proportion of the fluctuations in both bank rate and reserve money fluctuations, giventhe two policy goals, reconfirming that the primary goal of monetary policy in Malawi is pricestability, though the output goal is also pursued.4.4. Robustness CheckWhile all models are subjected to robustness checks, we report here results of the estimatedcomposite model from the truncated sample only. Structural estimates of the coefficients inmatrices A and B of the model are presented in Table C4 of Appendix C. The table shows thatnearly all coefficients in the model have standard errors with values of less than one, implyingthat they are efficient and hence form a solid basis for measuring monetary policy shocks. Inaddition, 12 of the 17 structural coefficients have correct signs. Inverse roots of the characteristicAR polynomial for the determination of stability (stationarity) of the model are reported in TableC5 of Appendix C. The table shows that all inverse roots of the characteristic AR polynomialhave modulus less than one and they lie inside the unit circle, indicating that at the chosen laglength (of order three), the estimated model is stationary or stable. Finally, serial correlation testresults reported in Table C6 of Appendix C show no evidence of serious serial correlation in themodel. Thus, the composite model with a truncated sample is robust and its inferences can berelied upon.29 | P a g e

TABLE 4.2: Variance Decomposition – Composite Model with Truncated SampleResponse of GYMonth GY CP BL XR M2 BR RM1 100 0 0 0 0 0 06 88.9972 0.40382 0.01772 2.39005 4.9986 0.40715 2.7854912 78.4499 3.68215 0.05778 2.5118 5.55366 1.20002 8.5447224 59.7092 7.7282 4.33016 9.01098 6.05287 5.96018 7.2084736 47.5531 7.56146 12.5306 10.5965 5.46594 10.5178 5.7746548 40.6831 6.76495 18.4448 9.79906 5.58786 13.3136 5.4066760 35.8408 5.97162 21.9176 8.78839 6.14028 15.0565 6.28477Response of CPMonth GY CP BL XR M2 BR RM1 0.4792 99.5208 0 0 0 0 06 0.12692 78.5446 3.14765 15.6286 0.50971 1.46601 0.5765512 1.4681 51.49 7.17394 29.3745 0.17985 4.62699 5.686624 3.96805 29.8407 16.7095 28.74 0.10506 8.76834 11.868436 3.71715 24.468 23.9787 25.109 0.10531 11.3679 11.253948 3.26477 21.9025 28.5072 22.8361 0.37441 13.2559 9.8591660 2.97764 19.9325 31.0985 20.9767 0.97151 14.6828 9.36032Response of BLMonth GY CP BL XR M2 BR RM1 2.25829 0.19857 90.2722 0.01399 0.03068 0.09299 7.13336 5.52293 2.53294 80.576 1.51271 1.13583 0.84956 7.8700612 4.10393 1.78157 78.8282 2.4896 2.00075 5.57694 5.2190424 3.19664 1.44509 72.3613 1.79589 3.34287 10.8788 6.9794536 3.62075 1.64873 61.5861 1.62123 5.65369 12.488 13.381648 3.91385 1.78059 53.247 1.45903 7.60075 13.209 18.789860 4.03579 1.85327 47.5869 1.31464 8.99576 13.7526 22.4611Response of XRMonth GY CP BL XR M2 BR RM1 0.41431 14.0718 0 85.5139 0 0 06 4.96644 15.4216 6.55759 66.815 0.13971 0.89385 5.2057412 6.5994 12.2794 13.7981 51.0453 0.27603 3.86841 12.133324 6.18497 10.1437 22.9385 39.6472 0.31664 7.09144 13.677636 5.62408 9.50432 26.9406 36.2663 0.45737 8.68215 12.525248 5.29833 8.92229 28.815 33.9814 0.92018 9.88385 12.17960 5.07369 8.37084 29.7261 31.9085 1.56142 10.8526 12.506930 | P a g e

TABLE 4.2 (Cont.): Variance Decomposition – Composite Model with Truncated SampleResponse of M2Month GY CP BL XR M2 BR RM1 1.92636 0.28743 0 0.00352 95.4028 2.3799 06 0.6548 1.81596 7.52457 1.21604 81.1561 1.71303 5.9194812 1.19103 7.08723 14.4188 3.06702 60.4949 3.13342 10.607624 0.85575 6.11429 25.9479 6.9906 42.1039 10.7904 7.1971636 0.73459 4.81951 32.2545 6.07727 34.0857 15.022 7.0064748 0.93304 3.92695 34.1127 5.01824 29.888 16.9727 9.1483960 1.21924 3.33834 34.0998 4.20601 27.2372 17.9386 11.9608Response of BRMonth GY CP BL XR M2 BR RM1 0.00072 0.02431 0 0.14773 0 99.8273 06 0.16952 25.4735 0.62922 15.7219 7.20235 47.9679 2.8356512 1.80112 24.6725 2.04384 23.2621 6.99854 22.3638 18.858124 5.09523 17.5419 4.5163 20.9502 7.25953 11.4689 33.16836 5.39437 16.3351 4.95889 19.036 7.83212 9.84913 36.594448 5.42576 16.0741 4.86045 18.4364 8.1086 9.40063 37.69460 5.45687 15.9262 4.75147 18.1548 8.26378 9.21899 38.2279Response of RMMonth GY CP BL XR M2 BR RM1 1.19511 1.22385 20.7566 3.12052 11.4659 0.84125 61.39686 1.09737 1.12111 17.1994 1.78105 18.0151 0.76545 60.020512 1.06856 1.80043 14.4123 3.72633 18.6833 2.18612 58.12324 0.97493 2.30316 14.8494 7.21981 18.143 7.90619 48.603636 0.84538 2.10466 19.0743 6.91603 16.9618 11.9046 42.193248 0.89279 1.81678 22.1946 6.09373 16.2624 14.1667 38.57360 1.05659 1.59219 24.002 5.32603 15.8866 15.5339 36.60285.0. SUMMARY, CONCLUSION AND POLICY IMPLICATIONSThis study set out to investigate the process through which monetary policy affects consumerprices and output in Malawi. Using innovation accounting in a structural vector autoregressivemodel, it is established that contrary to the official position that monetary policy in the countrytargets reserve money only, monetary authorities in Malawi also target short term interest rates.Effectively, the country employs hybrid operating procedures and it is demonstrated that thebank rate is a more effective measure of monetary policy compared to reserve money. In linewith Part III, Section 4(d) of the RBM Act of 1989, it is also established that the monetary31 | P a g e

authorities pursue both price stability and high growth and employment objectives. It is furthershown that price stability is the principal objective of monetary policy in the country. With theexception of exchange rate shocks, however, consumer prices respond weakly to monetaryimpulses suggesting that inflation in Malawi may not be predominated by monetary factors. Thefact that food costs have a preponderant weight (58.1 percent) in the all items national compositeconsumer price index reveals that structural rigidities in food production may be more importantdeterminants of inflation than monetary considerations.The study also illustrates that bank lending, exchange rates and aggregate money supply containimportant additional information in the transmission process of monetary policy shocks inMalawi. Besides own shocks, output fluctuations are largely attributed to M2 up to about a year,exchange rates at about 2 years and bank lending from about 3 years and beyond. Excluding ownshocks, variations in consumer prices are mostly accounted for by exchange rates up to about 3years and bank lending thereafter. M2 accounts for less than 1 percent of the consumer pricefluctuations across the five-year forecast horizon.Truncating the study period to include only the flexible exchange rate period (post-1994) revealsa number of interesting issues. First, the role of the exchange rate becomes more conspicuousalthough its impact on economic activity is weakened. Second, the importance of aggregatemoney supply and bank lending in transmitting monetary policy impulses is enhanced. It isconcluded, therefore, that with the floatation of the Malawi Kwacha in 1994, the monetarytransmission process evolved from a weak, blurred process to a somewhat strong, lessambiguous mechanism, consistent with theoretical expectations.Several policy issues emerge from the findings of this study. First, monetary authorities inMalawi need to consider moving away from using hybrid operating procedures to a scalarindicator of policy. The recommendation is to abandon the use of reserve money as a measure ofmonetary policy in favour of the more effective bank rate. RBM may, nonetheless, retain OMOas a vehicle for government borrowing on the domestic market through TBs and sterilising theimpact of this undertaking through RBM Bills.Second, the evidence that RBM also targets economic growth and employment objectivesbesides price stability suggests that the bank may be overloaded and hence unable to achieve itstargeted objectives due to inefficiencies arising from the multiple objectives. We suggest that thebank should narrow down its focus to concentrate on price stability and this should be reflectedin the RBM Act which governs its operations. In the literature, there is near consensus that toachieve the maximum attainable level of sustainable economic growth, the main objective ofmonetary policy should be price stability. Accordingly, we do not expect the governmentobjective of promoting economic growth and employment to be compromised with the centralbank‟s a focus on a single objective, namely price stability.32 | P a g e

REFERENCESBanda, W. (2004). Exchange Rate Management Under Monetary Targeting. Dealers Associationof Malawi lakeshore Conference (pp. 1-21). Mangochi: DEAMA.Becklelmans, L. (2005). Credit and Monetary Policy: An Australian SVAR. Reserve Bank ofAustralia Research Discussion Paper 2005-06 , 2005 (September), 1-29.Bernanke, B. (1986). Alternative Explanations of the Money-Income Correlation. NBERWorking Paper Series (NBER Working Paper Number 1842), 1-62.Bernanke, B. (1996). What Does the Bundesbank Target? NBER Working Paper Series (WP5764), 1-42.Bernanke, B., & Gertler, M. (1995). Inside the Black Box: The Credit Channel of MonetaryPolicy Transmission. Journal of Economic Perspectives , 9 (4), 27-48.Bernanke, B., & Mihov, I. (1995, May). Measuring Monetary Policy. National Bureau ofEconomic Research Working Paper WP-97-7 .Bernanke, B., & Mihov, I. (1998). Measuring Monetary Policy. The Quaterly Journal ofEconomics , 113 (3), 869-902.Bernanke, B., & Mihov, I. (1996). What Does the Bundesbank Target? NBER Working PaperSeries (Working Paper No. 5764), 1-42.Blundell-Wignall, A., & Gizycki, M. (1992). Credit Supply and DEmand and the AustralianEconomy. Reserve Bank of Australia Discussion Papers (Discussion paper No. 9208).Bolnick, B. (1991). Weak Links in the Monetary Transmission Mechanism: The Case of Malawi.ODI/HIID Conference on Improving Monetary policy in Africa and Asia (pp. 1-27). Harvard:ODI/HIID.Borys, M. M., & Horvath, R. (2007). The Effects of Monetary Policy in the Czech republic: AnEmpirical Study. CERGE-EI Working Paper Series (339), 1-26.Brischetto, A., & Voss, G. (1999). A Structural Vector Autoregression Model of MonetaryPolicy in Australia. Reserve Bank of Australia Research Discussion Paper Series (Discussionpaper No. 1999-11).Cheng, K. C. (2006). A VAR Analysis of Kenya's Monetary Policy Transmission Mechanism:How Does the Central Bank's REPO Rate Affect the Economy? IMF Working Paper(WP/06/300), 1-26.Chipeta, C., & Mkandawire, M. (1991). The Informal Financial Sector and MacroeconomicAdjustment in Malawi. AERC Research paper Series (Research Paper 4), 1-58.Christiano, L., Eichenbaum, M., & Evans, C. (1998). Monetary Policy Shocks: What have weLearned and to What End? NBER Working Paper Series (Working Paper Number 6400), 1-77.Clarida, R., & Gertler, M. (1996). How the Bundesbank Conducts Monetary Policy. NBERWorking Paper Series (Working Paper No. 5581), 1-50.33 | P a g e

Dabla-Norris, E., & Floerkemeier, H. (2006). Transmission Mechanisms of Monetary Policy inArmenia: Evidence from VAR Analysis. IMF Working Paper WP/06/248 , 1-26.Disyatat, P., & Vongsinsirikul, P. (2003). Monetary Policy and the Transmission Mechanism inThailand. Journal of Asian Economies , 2003 (14), 389-418.Dungey, M., & Pagan, A. (2000). A Sructural VAR Model of the Australian Economy.Economic Record , 76 (235), 321-342.Enders, W. (2004). Applied Econometric Time Series. Hoboken: John Wiley & Sons, Inc.Favero, C. A. (2001). Applied Macroeconometrics. New York: Oxford University Press Inc.Gondwe, S. (2001). The Impact of Liberalisation Policies on Commercial Bank Behaviour andFinancial Savings in Malawi. Zomba: Unpublished MA Thesis.Kamin, S., Turner, P., & Van't dack, J. (1998). The Transmission of Monetary Policy. BIS PolicyPapers , January 1998 (3).Karame, F., & Olmedo, A. (2002, February). The Assymetric Effects of Monetary Policy shocks:A Non-Linear Structural VAR Approach. Unpublished Manuscript , 1-28.Kim, S., & Roubini, N. (2000). Exchange Rate Anomalies in the Industrial Countries: A Solutionwith a Structural VAR Approach. A Journal of Monetary Economics , 45 (3), 561-586.Kugler, P., Jordan, T., Lenz, C., & Savios, M. (2004). The Analysis of Forward LookingMonetary Policy in a SVAR Framework. (Unpublished Manuscript), 1-8.Leeper, E. M., Sims, C. A., & Zha, T. (1996). What Doed Monetary policy Do. BrookingsPapers on Economic Activity , 1996 (2), 1-78.Malawi Government. (1989). Reserve Bank of Malawi Act. Laws of Malawi , 44 (02), 1-19.Maturu, B. (2007). Channels of Monetary Policy Transmission in Kenya. UnpublishedManuscript , 1-25.Meltzer, A. (1995). Money, Credit and (Other) Transmission Processes: A MonetaristPerspective. The Journal of Economic Perspectives , 19 (4), 49-72.Mihira, T., & Sugihara, S. (2000). A Structural VAR Analysis of the Monetary Policy in Japan.Economic Research Institute Discussion Paper Series (Discussion Paper No. 94), 1-28.Mishkin, F. (1995). Symposium on the Monetary Transmission Mechanism. Journal ofEconomic Perspectives , 9 (4), 3-10.Montiel, P. (1991). The Transmission Mechanism for Monetary Policy in Developing Countries.IMF Staff Papers , 38 (1), 83-108.Morsink, J., & Bayoumi, T. (2001). A Peek Inside the Black Box: The Monetary TransmissionMechanism in Japan. IMF Staff Papers , 48 (1), 22-57.Mutoti, N. (2006). Monetary Policy Transmission in Zambia. Bank of Zambia Working PaperSeries (WP/06/2006), 1-22.34 | P a g e

Obstefield, M., & Gertler, M. (1995). The Mirage of Fixed Exhange Rates. Journal of EconomicPerspectives , 9 (4), 73-96.Phiri, G. (2002). The Monetary Transmission Mechanism in Malawi. Reserve Bank of MalawiWorking Paper Series (2002).Piffanelli, S. (2001). The Instrument of Monetary Policy in Germany. A Structural VARApproach. United Nationa Department of Economic and Social Affairs (DESA) Working paperSeries (DESA Discussion Paper No. 19), 1-6.Setlhare, L. (2004). Bank of Botswana's Reaction Function: Modelling Botswana's MonetaryPolicy Strategy. The South African Journal of economics , 72 (2), 384-406.Sims. (1986). Are Forecasting Models Usable for Policy Analysis? Federal Reserve Bank ofMinneapolis Quarterly Review , 10 (1), 2-16.Sims, C. A., & Zha, T. A. (1998). Does Monetary Policy Generate Recessions? Federal ReserveBank of Atlanta Working Paper Series , Federal Reserve Bank of Atlanta Working Paper 98-12(July 1998), 1-61.Sims, C. (1992). Interpreting the Macroeconomic Time Series Facts: The Effects of MonetaryPolicy Time. European Economic Review , 36 (1992), 975-1011.Sims, C. (1980). Macroeconomics and Reality. Econometrica , 48 (January 1980), 1-49.Sims, C., & Zha, T. (1998). Does Monetary Policy Generate Recessions? Federal Reserve Bankof Atlanta Working Paper Series (Working Paper 98-12), 1-38.Sims, C., Stock, J., & Watson, M. (1990). Inference in Linear Time series Models with SomeUnit Roots. Econometrica , 58 (1), 113-144.Stock, J., & Watson, M. (2001, July). Vector Autoregressions. Unpublished Manuscript , 1-28.Taylor, J. (1995). The Monetary Transmission Mechanism: An Empirical Framework. Journal ofEconomic Perspectives , 9 (4), 11-26.Taylor, J. (1995). The Monetary Transmission Mechanism: An Empirical Framework. Journal ofEconomic Perspectives , 9 (4), 11-26.Vonnak, B. (2005). Estimating the Effects of Hangarian Monetary Policy Within a StructuralVAR Framework. Magyar Nemzeti Bank Working Paper 2005/1 , 1-37.Weitong, R. (2007). Economic Fluctuations and Japanese Monetary Policy: Two EmpiricalStudies with SVAR Approach. Unpublished Manuscript , 1-64.35 | P a g e

APPENDIX AInterpolation of GDP using Related Series 2While this study sets out to use monthly GDP time series, among other variables, the series is notavailable in monthly frequency. For this reason, we employ the Friedman method ofinterpolating time series by related series to obtain the required monthly GDP series from annualdata. The method employs log-linear interpolations of a vector of variables X t , which areavailable in both annual and monthly frequency, that explain the variable of interest , tocompute actual errors in the trend interpolation for the elements of X t . These errors are then usedto adjust the linear trend interpolation for by the weighted individual error in trendinterpolation for each regressor, where the weights are given by the respective coefficients on theX jt variable in the annual regression of X t on .To illustrate the mechanics of the approach, suppose variable is available at annual but notmonthly frequency and a vector of related variables is available at both annual and monthlyfrequency. The criteria for the selection of variables is that the intra-yearly movements ofare highly correlated with the intra-yearly movements of . This may be based on nonquantitativeconsiderations; on observed high correlation but for a different time period otherthan that earmarked for the interpolation; on co-movements between and for different timeunits for the same period; or on co-movements between series other than and butanalogous to them for the same time units and for the same period. We further assume that thevector does not contain any variables that will be included in the analysis of the study. Theinterpolation process is carried out in the following four steps:(1) Using annual observations only, compute the mean monthly growth for and . Use this toderive log-linear interpolations for both and . The linear interpolations are given by:, where and (A1)2 This methodology is adopted from Friedman, M. (1962). “The Interpolation of Time Series by Related Series”.Journal of the American Statistical Association, Vol. 57, No. 300 (Dec. 1962), pp. 729-757.i | P a g e

, where and (A2)(2) Using the annual observations only, run the regression:(A3)(3) From the actual monthly values of the variables and their interpolated values ,compute the actual error in the trend interpolation for the elements of, expressed as:(A4)(4) This error term is then used to adjust the simple linear trend interpolation for by theweighted individual error in trend interpolation for each regressor where the weights aregiven by the coefficient on the variable in the annual regression estimated in (A3). Thefinal interpolation of is:(A5)Clearly, this method results in every twelfth observation being exactly correct (since the error inthe interpolation of the variables is identically equal to zero for each variable), so that theinterpolated series always returns to the known annual observations for the series. In addition, itprovides a robust and transparent way of weighting the error term.In the interpolation of GDP for Malawi, we used actual errors in the trend interpolation oftobacco exports to adjust the linear trend interpolation for GDP. The choice of tobacco exports isbased on the fact that agriculture is the mainstay of the Malawi economy, accounting for about36 percent (2007 estimate) of the country‟s GDP and nearly 80 percent (2006 estimate) of thecountry‟s exports; employing an estimated 84.5% of the labour force; and responsible for about82.5% of foreign exchange earnings (Malawi, 2004). In addition, the economy is monoculture,with tobacco accounting for 60% of total export earnings followed by tea and sugar, contributingii | P a g e

GDPMalawi Kwacha (millions)about 10% each. It is, therefore, expected that tobacco exports and GDP will be highly correlatedas shown in Figure A1.The annual time series for the regression of tobacco exports on GDP covered the period 1968-2006 and the regression results are presented in Table A1. Both variables were I(1) and,accordingly, an error correction model was estimated. ERROR is the error correction term. Thecoefficient of tobacco exports, 20.83184, was used as a weight to compute weighted individualerrors in the trend interpolation of tobacco exports, which in turn are used for adjusting the lineartrend interpolation of GDP.Figure A2 shows actual and interpolated tobacco exports, the difference of which is weightedagainst the calculated coefficient of tobacco exports (20.83184) to adjust the linear trendinterpolation of GDP. The interpolated monthly GDP series is expected to be highly correlatedwith the actual monthly tobacco exports series as shown in Figure A3.Figure A1: GDP and Tobacco Exports for Malawi (1968-2006)500,000450,000400,000350,000300,000250,000200,000150,000100,00050,000GDPTobacco Exports60,00050,00040,00030,00020,00010,000Tobacco ExportsMalawi Kwacha (millions)-1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005-Source: NSO (Malawi)iii | P a g e

TABLE A1: Regression Results of GDP on Tobacco ExportsDependent Variable: GDPCoefficient Std. Error t-Statistic Prob.TOBACCO(-6) 20.83184 1.649441 12.86171 0.0000ERROR(-4) -0.820164 0.283190 2.896157 0.0077R-squared 0.674937Adjusted R-squared 0.635929S.E. of regression 14849.58Durbin-Watson stat 1.413473Figure A2: Tobacco Export (Actual) and GDP (Interpolated)8,0007,0006,0005,0004,0003,0002,0001,0000Tobacco Exports (Actual)Tobacco Exports (Interpolated)1988:1 1990:3 1992:5 1994:7 1996:9 1998:11 2001:1 2003:3 2005:5Source: NSO (Malawi)Figure A3: Tobacco Exports (Actual) and GDP Interpolated (1988:1-2005:12)80007000600050004000Tobacco Exports (Actual)GDP (Interpolated)30002000100001988:1 1990:3 1992:5 1994:7 1996:9 1998:11 2001:1 2003:3 2005:5400000350000300000250000200000150000100000500000Source: NSO (Malawi)iv | P a g e

APPENDIX BB-1.0. Data PropertiesCointegrated Structural VAR ApproachAnalysis of the cointegrated SVAR is carried out on the truncated sample only (1994:03-2005:12) covering the period when Malawi adopted indirect tools of monetary control. Using theAugmented Dickey-Fuller Test, it is established that all variables are integrated of order 1 at 1percent except for reserve money, which is integrated of order 1 at 5% (see table B1). Thisprovides a basis to test for cointegrating relations, defined as deviations from steady staterelations. In line with the exploratory data analysis carried out in Section 3.4 (Data, Data Sourcesand Measurement of Variables), we allow for each endogenous variable a linear deterministictrend with intercept and trend in the cointegrating equation and no trend in the VAR. A laglength of 3 is chosen in conformity with the standard information criteria (AIC, SIC and HIC)while ensuring stability of the system. At this lag length, the system is stationary or stable withall 21 inverse roots of the characteristic autoregressive polynomial showing modulus less thanone and lying inside the unit circle (see Table B2). Results of the Johansen Cointegration Testreveal that the system has a reduced rank 3 . While a trace suggests that there are threecointegrating relations, a maximum eigenvalue test indicates that there is only one cointegratingrelation (see Table B3). The study adopts results of the trace test since it is known that the test ismore robust to both skewness and excess kurtosis in the residuals than the maximal eigenvaluetest (see Harris & Sollis (2003); Cheung & Lai (1993)).3 Applying the Johansen approach to cointegration of multivariate systems, three possible cases are expected in therelationship between the rank of matrix and its characteristic roots. First is a case of full rank i.e. rank(number of variables). All rows (columns) are linearly independent (i.e. all variables are I(0)) and all characteristicroots are inside the unit circle with modulus less than 1. The system is effectively stationary and the levels of thevariables have stationary means. The appropriate modelling strategy, therefore, is to estimate the standard VAR inlevels. Second is a case of zero rank i.e. rank . In this event, , all rows are linearly dependent,there are no cointegrating variables and the system is non-stationary. In this case, is an matrix of zerosand the appropriate model is a VAR in first differences involving no long run elements as given by:The third case is one of reduced rank i.e. rank. The system is nonstationary but there are rcointegrating relations among the variables. The cointegrating relation is determined by and .v | P a g e

B-2.0. Cointegrated SVAR AnalysisEstimation results from the cointegrated structural VAR are generally similar to those from thelevels estimation carried in line with Sims, Stock and Watson (1990). Understandably, a numberof differences also show up. Among the differences, impulse responses from the cointegratedSVAR die off very quickly compared to those from the estimation in levels. In order to retainclear visual images, the forecast horizon is reduced from 60 months in the levels estimation to12-month in the cointegrated SVAR.The cointegrated SVAR confirms the finding in the levels estimation that monetary policy inMalawi employs hybrid operating procedures, with the bank rate and reserve money as operatingtools. Both the bank rate and reserve money respond significantly to shocks in the threeintermediate targets of monetary policy namely exchange rates, aggregate money supply andbank lending (see Figure B1), revealing that the central bank is concerned with movements in thethree targets and to achieve desired levels in these targets, the two policy tools are used.Consistent with the levels estimation, the cointegrated SVAR also shows that the exchange rateand money effect are important channels of monetary transmission in the country, though theimpact is not as pronounced as in the levels estimation. The effect of bank lending in themonetary transmission process, however, is insignificant in the cointegrated SVAR, whichcontradicts the finding in the levels estimation.The observed differences from the two estimation approaches are not unexpected. An importantsource of these differences is the imposition of what may be possibly incorrect cointegratingrestrictions in the process of estimating the cointegrated VAR. Kim and Roubini (2000) andBecklelmans (2005) argue that this is usually the case in cointegrated VARs with the implicationthat the resulting inferences are often incorrect as well. In an attempt to circumvent the problem,some studies opt for a simple differences specification (see, for example, Weitong (2007);Boivin & Giannoni (2002); Kasa & Popper (1997); Kugler et al (2004); Karame & Olmedo(2002); Mihira & Sugihara (2000)). The approach, however, is not persuasive as it yieldsinconsistent estimates if some variables are cointegrated (Bernanke & Mihov, 1997).vi | P a g e

Table B1: Stationarity ResultsVariable ADF p-value(t-statistic)BR 0.0000(-14.65723)XR 0.0000(-8.460660)CP 0.0000(-10.48257)GY 0.0000(-11.31887)M2 0.0000(-5.590931)RM 0.0000(-4.023323)BL 0.0000(-18.11330)Test Critical Values1% level: -3.4608845% level: -2.87486810% level: -2.5739511% level: -4.0015165% level: -3.43096310% level: -3.1391141% level: -4.0032265% level: -3.43178910% level: -3.1396011% level: -4.0036755% level: -3.43200510% level: -3.1397281% level: -4.0036755% level: -3.43200510% level: -3.1397281% level: -4.0041325% level: -3.43222610% level: -3.1398581% level: -4.0015165% level: -3.43096310% level: -3.139114Order ofIntegrationLevel ofSignificanceI(1) 1%I(1) 1%I(1) 1%I(1) 1%I(1) 1%I(1) 5%I(1) 1%vii | P a g e

Table B2: Roots of Characteristic PolynomialEndogenous variables: GY, CP, BL, XR, M2, BR, RMExogenous variables: CLag specification: 1 3RootModulus0.998043 0.9980430.919903 - 0.059154i 0.9218030.919903 + 0.059154i 0.9218030.791814 - 0.440799i 0.9062410.791814 + 0.440799i 0.9062410.879170 - 0.038116i 0.8799960.879170 + 0.038116i 0.8799960.664155 - 0.163656i 0.6840220.664155 + 0.163656i 0.684022-0.329850 + 0.519439i 0.615320-0.329850 - 0.519439i 0.6153200.114515 + 0.532327i 0.5445050.114515 - 0.532327i 0.544505-0.484589 0.484589-0.156873 - 0.418284i 0.446734-0.156873 + 0.418284i 0.446734-0.435546 0.4355460.035243 - 0.367722i 0.3694070.035243 + 0.367722i 0.3694070.028758 - 0.121287i 0.1246490.028758 + 0.121287i 0.124649No root lies outside the unit circle.VAR satisfies the stability condition.viii | P a g e

Table B3: Johansen Cointegration Test ResultsSeries: GY, CP, BL, XR, M2, BR, RMLags interval (in first differences): 1 to 2Unrestricted Cointegration Rank Test (Trace)Hypothesized Trace 0.05No. of CE(s) Eigenvalue Statistic Critical Value Prob.**None * 0.427296 203.5590 150.5585 0.0000At most 1 * 0.221686 126.0823 117.7082 0.0132At most 2 * 0.189775 91.24529 88.80380 0.0329At most 3 0.165744 61.99359 63.87610 0.0713At most 4 0.130145 36.80464 42.91525 0.1784At most 5 0.093521 17.42402 25.87211 0.3839At most 6 0.026799 3.775896 12.51798 0.7743* denotes rejection of the hypothesis at the 0.05 level**MacKinnon-Haug-Michelis (1999) p-valuesUnrestricted Cointegration Rank Test (Maximum Eigenvalue)Hypothesized Max-Eigen 0.05No. of CE(s) Eigenvalue Statistic Critical Value Prob.**None * 0.427296 77.47678 50.59985 0.0000At most 1 0.221686 34.83697 44.49720 0.3745At most 2 0.189775 29.25170 38.33101 0.3723At most 3 0.165744 25.18895 32.11832 0.2755At most 4 0.130145 19.38062 25.82321 0.2804At most 5 0.093521 13.64812 19.38704 0.2786At most 6 0.026799 3.775896 12.51798 0.7743* denotes rejection of the hypothesis at the 0.05 level**MacKinnon-Haug-Michelis (1999) p-valuesix | P a g e

FIGURE B1: Impulse Responses of the Composite Model (1994:03 – 2005:12)Response of D(GY) to D(GY)Response of D(GY) to D(CP)Response of D(GY) to D(BL)Response of D(GY) to D(XR)Response of D(GY) to D(M2)Response of D(GY)) to D(BR)Response of D(GY) to D(RM).08.08.08.08.08.08.08.06.06.06.06.06.06.06.04.04.04.04.04.04.04.02.02.02.02.02.02.02.00.00.00.00.00.00.00-.02-.02-.02-.02-.02-.02-.02-.04-.04-.04-.04-.04-.04-.04-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12-.061 2 3 4 5 6 7 8 9 10 11 12Response of D(CP) to D(GY)Response of D(CP) to D(CP)Response of D(CP) to D(BL)Response of D(CP) to D(XR)Response of D(CP) to D(M2)Response of D(CP) to D(BR)Response of D(CP) to D(RM).03.03.03.03.03.03.03.02.02.02.02.02.02.02.01.01.01.01.01.01.01.00.00.00.00.00.00.00-.01-.01-.01-.01-.01-.01-.01-.02-.02-.02-.02-.02-.02-.02-.03-.03-.03-.03-.03-.03-.03-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12Response of D(BL) to D(GY)Response of D(BL) to D(CP)Response of D(BL) to D(BL)Response of D(BL) to D(XR)Response of D(BL) to D(M2)Response of D(BL) to D(BR)Response of D(BL) to D(RM).08.08.08.08.08.08.08.06.06.06.06.06.06.06.04.04.04.04.04.04.04.02.02.02.02.02.02.02.00.00.00.00.00.00.00-.02-.02-.02-.02-.02-.02-.02-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12Response of D(XR) to D(GY)Response of D(XR) to D(CP)Response of D(XR) to D(BL)Response of D(XR) to D(XR)Response of D(XR) to D(M2)Response of D(XR) to D(BR)Response of D(XR) to D(RM).05.05.05.05.05.05.05.04.04.04.04.04.04.04.03.03.03.03.03.03.03.02.02.02.02.02.02.02.01.01.01.01.01.01.01.00.00.00.00.00.00.00-.01-.01-.01-.01-.01-.01-.01-.02-.02-.02-.02-.02-.02-.02-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12-.031 2 3 4 5 6 7 8 9 10 11 12Response of D(M2) to D(GY)Response of D(M2) to D(CP)Response of D(M2) to D(BL)Response of D(M2) to D(XR)Response of D(M2) to D(M2)Response of D(M2) to D(BR)Response of D(M2) to D(RM).08.08.08.08.08.08.08.06.06.06.06.06.06.06.04.04.04.04.04.04.04.02.02.02.02.02.02.02.00.00.00.00.00.00.00-.02-.02-.02-.02-.02-.02-.02-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12-.041 2 3 4 5 6 7 8 9 10 11 12Response of D(BR) to D(GY)Response of D(BR) to D(CP)Response of D(BR) to D(BL)Response of D(BR) to D(XR)Response of D(BR) to D(M2)Response of D(BR) to D(BR)Response of D(BR) to D(RM)3333333222222211111110000000-1-1-1-1-1-1-1-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12-21 2 3 4 5 6 7 8 9 10 11 12Response of D(RM) to D(GY)Response of D(RM) to D(CP)Response of D(RM) to D(BL)Response of D(RM) to D(XR)Response of D(RM) to D(M2)Response of D(RM) to D(BR)Response of D(RM) to D(RM).08.08.08.08.08.08.08.06.06.06.06.06.06.06.04.04.04.04.04.04.04.02.02.02.02.02.02.02.00.00.00.00.00.00.00-.02-.02-.02-.02-.02-.02-.02-.04-.04-.04-.04-.04-.04-.04-.06-.06-.06-.06-.06-.06-.06-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12-.081 2 3 4 5 6 7 8 9 10 11 12x | P a g e

APPENDIX CTABLE C1: Roots of Characteristic Polynomial – Generic ModelEndogenous variables: LOG(GY), LOG(CP), BR, LOG(RM)Exogenous variables: CLag specification: 1 2RootModulus0.995863 0.9958630.977973 0.9779730.880114 0.8801140.734039 0.734039-0.248588 0.248588-0.210701 - 0.052382i 0.217115-0.210701 + 0.052382i 0.2171150.099467 0.099467No root lies outside the unit circle.VAR satisfies the stability condition.xi | P a g e

APPENDIX CTABLE C2: VAR Lag Exclusion Wald Test Results – Generic ModelSample: 1988M01 2005M12Included observations: 214Chi-squared test statistics for lag exclusion:Numbers in [ ] are p-valuesLOG(GY) LOG(CP) BR LOG(RM) JointLag 1 42.14412 224.7095 205.1882 127.8966 566.2024[ 1.56e-08] [ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000]Lag 2 18.10525 5.583003 12.14915 8.687186 42.42655[ 0.001177] [ 0.232529] [ 0.016276] [ 0.069412] [ 0.000341]df 4 4 4 4 16xii | P a g e

APPENDIX CTABLE C3: Structural VAR Estimates - Generic VAR ModelSample (adjusted): 1988M03 2005M12Included observations: 214 after adjustmentsCoefficient Std. Error z-Statistic Prob.a 21 0.001328 0.013865 0.095757 0.9237a 43 0.002753 0.002381 1.156456 0.2475b 11 0.108091 0.005225 20.68816 0.0000b 22 0.021923 0.001060 20.68816 0.0000b 33 2.163468 0.104575 20.68816 0.0000b 44 0.075347 0.003642 20.68816 0.0000xiii | P a g e

xiv | P a g eAPPENDIX CTABLE C4: Structural VAR Estimates - Composite Model with TruncatedSampleEstimated A matrix:1.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000-0.007633 1.000000 0.000000 0.000000 0.000000 0.000000 0.0000000.087871 -0.208193 1.000000 -0.111557 0.111252 -0.001226 -0.266192-0.013354 -1.187919 0.000000 1.000000 0.000000 0.000000 0.000000-0.061896 0.221568 0.000000 0.000000 1.000000 0.003680 0.0000000.000000 0.000000 0.000000 2.178026 0.000000 1.000000 0.0000000.000000 0.000000 0.614035 0.333893 -0.504922 0.000756 1.000000Estimated B matrix:0.126568 0.000000 0.000000 0.000000 0.000000 0.000000 0.0000000.000000 0.013922 0.000000 0.000000 0.000000 0.000000 0.0000000.000000 0.000000 0.067558 0.000000 0.000000 0.000000 0.0000000.000000 0.000000 0.000000 0.040770 0.000000 0.000000 0.0000000.000000 0.000000 0.000000 0.000000 0.053785 0.000000 0.0000000.000000 0.000000 0.000000 0.000000 0.000000 2.308357 0.0000000.000000 0.000000 0.000000 0.000000 0.000000 0.000000 0.071345Coefficient Std. Error z-Statistic Prob. Expected Sign Right Sign?a 21 -0.007633 0.009231 -0.826883 0.4083 - a 31 0.087871 0.052475 1.674542 0.0940 + a 32 -0.013354 0.027097 -0.492834 0.6221 + a 34 -0.061896 0.035747 -1.731512 0.0834 + a 35 -0.208193 0.444451 -0.468427 0.6395 - a 36 -1.187919 0.245746 -4.833932 0.0000 - a 37 0.221568 0.324233 0.683360 0.4944 - a 41 0.614035 0.623890 0.984204 0.3250 - a 42 -0.111557 0.240466 -0.463921 0.6427 - a 51 2.178026 4.393726 0.495713 0.6201 + a 52 0.333893 0.136933 2.438360 0.0148 + a 56 0.111252 0.298593 0.372586 0.7095 - a 64 -0.504922 0.110127 -4.584901 0.0000 + a 73 -0.001226 0.002593 -0.472719 0.6364 - a 74 0.003680 0.001954 1.883492 0.0596 + a 75 0.000756 0.002670 0.283020 0.7772 + a 76 -0.266192 0.567014 -0.469462 0.6387 - b 11 -0.126568 0.007510 -16.85230 0.0000b 22 0.013922 0.000826 16.85230 0.0000b 33 -0.067558 0.020611 -3.277832 0.0010b 44 0.040770 0.002419 16.85230 0.0000b 55 -0.053785 0.003192 -16.85230 0.0000b 66 -2.308357 0.136976 -16.85230 0.0000b 77 0.071345 0.011029 6.468919 0.000011

APPENDIX CTABLE C5: Roots of Characteristic Polynomial – Composite Model withTruncated SampleEndogenous variables: LOG(GY) LOG(CP) LOG(BL) LOG(XR) LOG(M2) BR LOG(RM)Exogenous variables: CLag specification: 1 3RootModulus0.999573 0.9995730.962881 0.9628810.925938 0.9259380.876079 - 0.119740i 0.8842240.876079 + 0.119740i 0.8842240.708600 - 0.412825i 0.8200850.708600 + 0.412825i 0.8200850.631851 0.6318510.129862 + 0.527902i 0.5436400.129862 - 0.527902i 0.543640-0.201258 - 0.445031i 0.488423-0.201258 + 0.445031i 0.488423-0.450605 - 0.092084i 0.459918-0.450605 + 0.092084i 0.4599180.143127 - 0.353651i 0.3815160.143127 + 0.353651i 0.3815160.245836 + 0.163825i 0.2954220.245836 - 0.163825i 0.295422-0.283407 0.283407-0.160308 - 0.189921i 0.248534-0.160308 + 0.189921i 0.248534No root lies outside the unit circle.VAR satisfies the stability condition.xv | P a g e

APPENDIX CTABLE C6: VAR Residuals Cross Correlations Ordered by Lags - CompositeModel with Truncated SampleLOG(GY) LOG(CP) LOG(BL) LOG(XR) LOG(M2) BR LOG(RM)LOG(GY) 1.000000 0.069224 -0.144704 0.064367 0.134520 0.024400 0.121499LOG(CP) 0.069224 1.000000 0.056692 0.378680 -0.066714 0.132476 -0.063637LOG(BL) -0.144704 0.056692 1.000000 0.019721 -0.009806 0.031459 -0.245548LOG(XR) 0.064367 0.378680 0.019721 1.000000 0.091350 -0.041563 -0.165204LOG(M2) 0.134520 -0.066714 -0.009806 0.091350 1.000000 -0.158261 0.348787BR 0.024400 0.132476 0.031459 -0.041563 -0.158261 1.000000 -0.086392LOG(RM) 0.121499 -0.063637 -0.245548 -0.165204 0.348787 -0.086392 1.000000LOG(GY(-1)) 0.008743 0.002966 -0.001933 -2.50E-05 -0.004844 0.016853 -0.009941LOG(CP(-1)) -0.023744 -0.017763 -0.008357 0.053354 -0.033778 -0.065503 -0.011302LOG(BL(-1)) 0.005911 0.000295 -0.016009 0.010127 0.010937 0.004852 -0.031909LOG(XR(-1)) -0.045191 0.019798 -0.065500 0.092454 -0.056949 -0.060321 -0.059860LOG(M2(-1)) 0.052810 0.034580 0.007631 -0.033807 0.018313 0.039954 0.020228BR(-1) 0.029365 0.007467 0.013761 -0.010101 0.050410 -0.012668 0.036919LOG(RM(-1)) 0.049911 -0.025703 0.007086 0.002395 0.027717 -0.019197 -0.002836LOG(GY(-2)) -0.007885 0.019435 0.000788 -0.034051 0.039586 0.028968 0.012532LOG(CP(-2)) 0.033423 -0.007264 -0.017436 0.002848 0.013484 -0.046282 0.038073LOG(BL(-2)) 0.008579 0.006912 -0.005305 0.036103 0.025270 0.016077 -0.024647LOG(XR(-2)) 0.019074 -0.004778 0.012461 -0.056126 -9.18E-07 0.030916 0.079219LOG(M2(-2)) 0.026778 0.040005 0.049794 -0.045713 -0.043304 0.033677 0.026769BR(-2) 0.024817 0.009807 0.015989 0.061723 0.048378 -0.019987 0.004665LOG(RM(-2)) 6.47E-05 -0.010625 0.028686 -0.027806 0.038956 0.038571 -0.000262LOG(GY(-3)) -0.091180 0.084452 -0.023172 0.031069 0.049548 0.104300 -0.003561LOG(CP(-3)) -0.106593 -0.004962 -0.023112 -0.037222 -0.082600 0.168213 -0.125354LOG(BL(-3)) 0.192291 -0.030054 0.009405 -0.019671 -0.116271 0.113202 -0.129837LOG(XR(-3)) -0.148414 0.019636 -0.033146 0.079315 -0.213402 -0.020024 -0.109427LOG(M2(-3)) -0.022098 -0.198734 0.010798 -0.135172 -0.047220 -0.023919 -0.075066BR(-3) 0.001780 0.001169 -0.069966 0.069139 -0.006857 0.014891 -0.003920LOG(RM(-3)) 0.014891 -0.073304 0.052525 -0.035020 0.063196 0.004471 -0.018601LOG(GY(-4)) 0.114072 0.048954 0.051874 0.045787 0.044634 0.061699 0.079103LOG(CP(-4)) -0.164509 -0.015610 0.089113 -0.136795 -0.179274 0.072764 -0.048616LOG(BL(-4)) -0.031417 0.013103 0.041416 0.170622 -0.055561 0.038832 -0.010366LOG(XR(-4)) 0.041875 0.022343 -0.119559 -0.040072 -0.088987 -0.048405 0.058287LOG(M2(-4)) -0.018958 0.032305 -0.056632 -0.035057 0.088090 0.092828 -0.007974BR(-4) 0.132110 0.028919 -0.064699 0.065915 -0.027534 -0.028486 -0.044774LOG(RM(-4)) -0.062529 -0.075334 0.026283 -0.143604 -0.112719 0.074884 -0.138152LOG(GY(-5)) 0.092360 -0.125757 0.011022 -0.113104 0.057710 -0.121379 -0.033253LOG(CP(-5)) -0.064873 -0.088590 -0.029373 0.060724 -0.148179 -0.074205 0.006488LOG(BL(-5)) -0.003306 0.172869 -0.073482 0.084305 0.005919 0.175992 0.033072LOG(XR(-5)) -0.019394 -0.142525 -0.042438 0.063656 -0.126065 -0.127581 0.017574LOG(M2(-5)) -0.005944 0.062198 0.140324 0.069899 -0.053679 0.060263 -0.030331BR(-5) -0.099954 -0.023893 0.113303 -0.087117 -0.044847 -0.111560 0.037655LOG(RM(-5)) 0.011224 -0.066209 -0.009160 -0.030644 0.147512 -0.042332 -0.040673xvi | P a g e

APPENDIX DFIGURE D1:Impulse Responses - the Composite Model (Full Sample)with Endogenous Exchange rates.15Response of GY to GY.15Response of GY to CP.15Response of GY to BL.15Response of GY to XR.15Response of GY to M2.15Response of GY to BR.15Response of GY to RM.10.10.10.10.10.10.10.05.05.05.05.05.05.05.00.00.00.00.00.00.00-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60.08Response of CP to GY.08Response of CP to CP.08Response of CP to BL.08Response of CP to XR.08Response of CP to M2.08Response of CP to BR.08Response of CP to RM.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60Response of BL to GYResponse of BL to CPResponse of BL to BLResponse of BL to XRResponse of BL to M2Response of BL to BRResponse of BL to RM.10.10.10.10.10.10.10.05.05.05.05.05.05.05.00.00.00.00.00.00.00-.05-.05-.05-.05-.05-.05-.05-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60.08Response of XR to GY.08Response of XR to CP.08Response of XR to BL.08Response of XR to XR.08Response of XR to M2.08Response of XR to BR.08Response of XR to RM.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60Response of M2 to GYResponse of M2 to CPResponse of M2 to BLResponse of M2 to XRResponse of M2 to M2Response of M2 to BRResponse of M2 to RM.08.08.08.08.08.08.08.04.04.04.04.04.04.04.00.00.00.00.00.00.00-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60Response of BR to GYResponse of BR to CPResponse of BR to BLResponse of BR to XRResponse of BR to M2Response of BR to BRResponse of BR to RM444444422222220000000-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60.1Response of RM to GY.1Response of RM to CP.1Response of RM to BL.1Response of RM to XR.1Response of RM to M2.1Response of RM to BR.1Response of RM to RM.0.0.0.0.0.0.0-.1-.1-.1-.1-.1-.1-.1-.210 20 30 40 50 60-.210 20 30 40 50 60-.210 20 30 40 50 60-.210 20 30 40 50 60-.210 20 30 40 50 60-.210 20 30 40 50 60-.210 20 30 40 50 60xvii | P a g e

APPENDIX DFIGURE D2:Impulse Responses - Composite Model (Full Sample) withExogenous Exchange rates.15Response of GY to GY.15Response of GY to CP.15Response of GY to BL.15Response of GY to M2.15Response of GY to BR.15Response of GY to RM.10.10.10.10.10.10.05.05.05.05.05.05.00.00.00.00.00.00-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60-.0510 20 30 40 50 60Response of CP to GYResponse of CP to CPResponse of CP to BLResponse of CP to M2Response of CP to BRResponse of CP to RM.03.03.03.03.03.03.02.02.02.02.02.02.01.01.01.01.01.01.00.00.00.00.00.00-.0110 20 30 40 50 60-.0110 20 30 40 50 60-.0110 20 30 40 50 60-.0110 20 30 40 50 60-.0110 20 30 40 50 60-.0110 20 30 40 50 60.10Response of BL to GY.10Response of BL to CP.10Response of BL to BL.10Response of BL to M2.10Response of BL to BR.10Response of BL to RM.05.05.05.05.05.05.00.00.00.00.00.00-.05-.05-.05-.05-.05-.05-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60-.1010 20 30 40 50 60Response of M2 to GYResponse of M2 to CPResponse of M2 to BLResponse of M2 to M2Response of M2 to BRResponse of M2 to RM.08.08.08.08.08.08.04.04.04.04.04.04.00.00.00.00.00.00-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 60-.0410 20 30 40 50 603Response of BR to GY3Response of BR to CP3Response of BR to BL3Response of BR to M23Response of BR to BR3Response of BR to RM222222111111000000-1-1-1-1-1-1-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60-210 20 30 40 50 60Response of RM to GYResponse of RM to CPResponse of RM to BLResponse of RM to M2Response of RM to BRResponse of RM to RM.12.12.12.12.12.12.08.08.08.08.08.08.04.04.04.04.04.04.00.00.00.00.00.00-.04-.04-.04-.04-.04-.04-.08-.08-.08-.08-.08-.08-.1210 20 30 40 50 60-.1210 20 30 40 50 60-.1210 20 30 40 50 60-.1210 20 30 40 50 60-.1210 20 30 40 50 60-.1210 20 30 40 50 60xviii | P a g e