CVs Daniel HERLEMONT - Yats.com

Daniel HERLEMONT
chemin du guerrier,
31450 Deyme FRANCE
���� +33 (0) 6 10 48 02 99
mailto:dherlemont@yats.com
born: 1955
language: French, English
EDUCATION :
1993-1994: MBA - CPA (Centre de Perfectionnement aux Affaires, Groupe Lagardère)
1979-1981: Graduated from "Ecole Supérieure d'Electricité"
1976-1979: Graduated from Ecole Polytechnique (X76)
EXPERIENCE (summary) :
1998 - ... Professor of Finance at Mathematical Finance School of Léonard de Vinci, Paris La Défense
Risk Manager for a Hedge Fund
Lecturer in Business Schools and Universities for Master in Finance: Toulouse Business
School and University, Poland, …
YATS: consulting services and solutions for e-business applications, including:
• a powerful automated trading platform and systems, for real time execution, simulation
and training, featuring statistical arbitrage and artificial intelligence based strategies,
• advanced risk management and Optimization tools for Hedge Funds
• Information systems for on line trading, front, middle and back office applications,
development, integration and operations of advanced search engines, integration of real
time quotes and news feeds, decision aid tools ...
Main references :
2002-... : teaching, risk management for Hedge Funds, proprietary trading, ...
1999-2002: Société Générale (FIMATEX): developing & hosting news feeds, search engines &
online tools (warrants, opinions, profiling, ...)
1999-2001: WARGNY: internet architecture, development of online decision aids,
1998-1999: creation of yats.com, ranked as one of the most attractive financial site.
1987-1998 ASTRIUM (ex MATRA MARCONI SPACE) the Europe’s largest space company, Toulouse
• 1998-1994 Deputy Manager of Ground Systems & Network Unit (300 engineers), including
diversification in satellites applications: telematics, environment monitoring, Air Traffic
Management, telemedicine, .....
• 1991-1994: Manager of the "Technologies & Innovation" division (60 engineers)
• 1987-1991: Manager of the Control Centre division (50 engineers).
1981-1987
THOMSON-CSF / Central Research Labs: development of operating system and object oriented
middleware components RISC microprocessors (within a Joint-Venture with MIPS inc., USA).
YATS SARL, 440 653 285 RCS Toulouse , chemin du guerrier 31450 DEYME email : dherlemont@yats.com

Risk Management Philosophy
By Daniel HERLEMONT, YATS Finances & Technologies
Risk Model: Implementing Safety First Principle
Risk Management may be the most important element of an
investment process. Risk Management is the key to surviving
during inevitable, dramatic market moves.
Market risk is the risk of losses due to the financial market
variables (such as interest rates, foreign currency, equities,
and commodities) adverse
prices fluctuations. It is
primarily measured with
Value at Risk or VaR.
VaR is a statistical
measure of downside risk
that is simple to
implement. VaR
measures the maximum
loss for a given confidence interval.
VaR model is primarily based on Risk Management
Standards such as RiskMetrics and Risk Management
Guides (from JP Morgan) as well as Basle II
recommendations.
VaR, however, is not the only risk measure that YATS uses.
VaR is complemented by Stress Testing which identifies
potential losses under extreme market conditions. YATS
Market Risk model also implements Extreme Values theory
to provide more robust estimates of VaR. Other risk
measures are extensively used such as Maximum Drawdown
analysis, Conditional VaR, Risk Adjusted Performance
Measures such as Omega measures, and Stutzer Index. The
Stutzer Index is a consistent generalization of the Sharpe
Ratio for fat tails returns and prudent investor, implementing
a Safety First Principle. The Stutzer Index has been recently
adopted by Morningstar.
Risk measures and preferences will be used to determine
trading limits, position sizing and stop losses as well as
optimal leverage to meet risk management constraints such
as VaR and Maximum Drawdown objectives while
maximizing the expected growth rate. More precisely, the
leverage is adjusted to maximize the expected growth rates
under the constraints that a VaR or Maximum Drawdown
objective will be satisfied with a predefined high probability
level. When this probability is set to 1, then the policy is the
same as implementing strict portfolio insurance. However,
strict guarantee may be too restrictive. In a more general
setting, with a probability, of say 95%, the trader will have
more flexibility to capture opportunities in specific market
states.
YATS, Daniel HERLEMONT, Tel +33 (0) 6 10 48 02 99 email: dherlemont@yats.com
YATS Risk Model: the actual implementation
YATS Risk Management duties generally include:
• Modeling traded assets, historical distributions, extreme
events and, most importantly, the dynamic strategy
• Defining the most suited Volatility models to implement a
dynamic VaR that may depend on the different market
regimes; typically, this model will dictate to reduce
positions when volatility increases dramatically
• Back testing the monthly VaR to meet a typical objective of
a 95% VaR not exceeding 4% of the capital
• Implementing Conditional VaR to avoid some VaR
shortcomings (see hereafter)
• Implementing a Safety First Principle at strategy level by
optimizing the Stutzer Index,
• Implementing overall optimal allocation and leverage under
the VaR and Drawdown constraints
• Performing extensive scenario and stress testing
The positions are monitored on a daily basis and mark to market VaR
is provided as well as other parameters computed such as VaR
exceptions. Regular and comprehensive Risk reports are directly sent
to clients.
Market
Insight and
objectives
scénarii
Trader
Risk Manager
Optimal Leverage
Stop loss and
Position sizing
Trading
Positions
Monitoring & Control
safety rules enforcements

Return
-0.6 -0.4 -0.2 0.0 0.2
Risk Management Philosophy
By Daniel HERLEMONT, YATS Finances & Technologies
Value at Risk and Beyond
While there is no doubt that VaR is a useful generic measure
of risk exposure, VaR has a number of limitations that are
particularly problematic for hedge-fund investments. In
addition, VaR tells noting about actual losses… Other Risk
measures, such as the Conditional VaR or CVaR, have not
the shortcoming of the VaR. The CVaR measures the
expected loss conditional on the loss being greater than or
equal to the VaR. By using the CVaR, we are able to capture
the left-tail risk of those hedge fund strategies that have short
put option-like exposures. The VaR might be blind on actual
losses. In addition the CVaR can be used in optimization
programs, which is not possible with VaR or may lead to
risky positions if the investment process is tailored to trade
at the VaR boundaries under fat tailed returns.
Nearly all hedge fund strategies show fat tails (negatively
skewed return distributions with positive excess kurtosis).
The efficient frontier is changed when using modified VaR
rather than VaR as the risk constraint. Application of the
mean-variance framework in the case of some hedge fund
strategies can result in underestimation of tail risk by as
much as 50%.
Leverage will determine the ongoing viability of the
program.
One can build a business or investment program around
positive expected value opportunities, but the particular
leverage level and hedging strategy chosen will determine
the ongoing viability of the program. There are well known
examples of statistically valid strategies resulting in billions
of dollars of losses (Metallgesellschaft in 1993, Long Term
Capital Management in 1998…). Obviously, the leverage
level and hedging strategies chosen were flawed. When
choosing too much leverage, an investor sometimes faces the
problem of possibly being able to win the “game”, but
consistently running out of time.
In a multi-periods
setting, the typical
growth of the wealth
does not linearly depend
on the leverage. Rather,
the dependency can be
approximated by the so
called Kelly parabola.
There exists an optimal
leverage that will lead
the best growth.
0.0 0.5 1.0 1.5 Exceeding this optimal
Leverage
leverage may lead to
dramatic losses.
Misunderstanding the leverage effects explains the high level
of mortality in the Hedge Fund industry (15%-20% default
rate per year)
Risk Adjusted Performance Measure: a Safety First Principle
The Sharpe ratio is certainly one of the most popular Risk Adjusted
Return Measure. However, several authors have pointed out the
shortcomings of using the Sharpe ratio. As a matter of fact, one can
define an optimal strategy maximizing the Sharpe ratio in complete
and arbitrage-free markets. This strategy mainly consists in selling
put options far out of the money. This returns distribution of this
strategy has a truncated right tail and an elongated left tail. It is well
known that this type of strategy may lead to rare but dramatic losses.
Using the Sharpe ratio to evaluate risk premia strategies will create
the same type of problems as with option sellers. Hence, by
undertaking a maximum Sharpe ratio strategy, an investor may be
accepting very risky investments in exchange for improving the mean
or variance of the investment. The problem with this trade-off is that
investors are risk averse; they most certainly have a preference for
upside risk and an aversion to downside risk.
The Stutzer Index does not suffer the Sharpe Ratio shortcomings.
The main concern for investors is the probability of underperforming
a benchmark on average. The Stutzer index rewards those portfolios
that have a lower likelihood of underperforming a specified
benchmark on average. This measure penalizes negative skewness
and high kurtosis (for given levels of mean returns and variance).
The Stutzer index is reduced to the Sharpe Ratio when return
distributions are normally distributed. Equivalently, we can use the
Hodges Generalized Sharpe Ratio that has the same properties as the
Stutzer Index. The higher it is, the better is the portfolio.
Unveiling Dependencies
An important selling point for alternative investments is their
typically low correlations with traditional market indexes such as the
S&P500. However, correlations are linear measures of association
that ignore certain nonlinear relations often present among hedgefund
investments. Nonlinearities should be incorporated into any risk
analysis of hedge fund investments. Rather than using conventional
linear correlation measures, we are using non linear regressions as
well as non conventional measures like relative entropy that is much
more robust a measure based on information theory.
Incentive Contracts and Hedge Fund Management
Hedge fund managers typically receive a fraction of the fund’s return
in excess of the high-water mark. The high-water mark for each
investor is the maximum share value since his or her investment in
the fund. These performance fees generally range from 15% to 25%
of the new profits earned each year. Option-like characteristics of
incentive fees clearly could induce risk-taking behavior when the
asset value is below the high-water mark. This “option like” payoff is
taken into account in overall integrated risk model so that the risk
behavior of the investment manager is under control and restricted to
meet investors’ objective rather than the manager‘s objective only.
YATS, Daniel HERLEMONT, Tel +33 (0) 6 10 48 02 99 email: dherlemont@yats.com

YATS-RAPT : Real-time Automated Profitable Trading
YATS-RAPT offers professional solutions to develop, test and
operate automated trading systems for futures and stock
markets.
The YATS-RAPT plateform is a result of more than 15 years
endeavor in computational finance, development and actual
trading on futures and stock markets.
YATS- RAPT runs on any
environment : WINDOWS, LINUX, ...
Actual operation on going, with more
than 50 orders / day.
100 000 of source lines code.
Commercial offering
Trading Systèm
YATS components
JAVA, JAVASCRIPT, DDE .NET bridge
Interactive Brokers TWS,
FIX protocol,
Web brokers interface
• Automated orders with Interactive Brokers , FIX interface, WEB
broker interface.
• Dynamic tables, real time charts
• User defined strategies in JAVASCRIPT,
• Errors controls, automatic re-connection,
• Tick by tick Recording, order logs, portfolios,
• Operating modes: actual trading, live test, training, debug,
simulation, backtest, studies ...
• High fidelity simulator: on recorded or simulated ticks, variable
speeds, simulate order book using full tick data (bid/ask/last), as well as
order priorities (bid/ask/last sizes) and transmission delays, ...
• Trading Systems Debugger including, event breakpoint (order status,
tick data, ...), step by step mode.
• Trading Systems Backtester including parameters optimizations
based on user defined objective functions (reward/risk)
• Support for execution of parallel multiple strategies.
• More than 100 Indicators : technical analysis, many math, logical
functions on intraday data, in tick time, or business time, aggregation of
ticks data in regular time series, user defined functions/indicators in
JAVASCRIPT.
• Volatility models, covariances, using high/low values, ACD-CARGH
modeling,
• Risk, money management : Sharpe ratio, Sterling ratio, VaR, utility
functions,...
• Portfolio management: CAPM, optimization of user defined utility
function, Kelly criteria, CRP (Constant Rebalanced Portfolio), ....
• Options pricers :, classic or exotic options
• Numerical analysis: global optimizations (simulated annealing), genetic
algorithms, Kalman filtering, wavelets, ...
• Monte Carlo Simulations, (MCMC)
• Probability & Statistics, significance testing : t-stat, khi2, Kolmogorov
Smirnov, Watson, Jarque Berra, runs test, linear and non linear
simple/multiple regressions, maximum likelihood, bootstrap, time series
analysis: ARIMA, VARMA, stochastic processes and related
distributions (min/max, first passage time, ...),
• Variable length Markov Chains
• Machine learning, data mining: classification, segmentation, non
linear regressions (KNN, ...), regression trees, dynamic programming,
reinforcement learning, universal predictions & portfolios
• predefined and fully parameterizable Trading Systems
momentum, trend following, mean reversion, breakout,
long/short positions ....
• Stat arb trading systems : based on cointegration
VAR/VECM & Kalman filter
• Market Making algorithms
• "Model Based Trading"
data → model → Monte Carlo Simulation of market
model → Optimal Trading System
WEB interface can be used for modeling and backtests.
orders
quotes actual simulated
actual actual trading (live) test, training
recorded or
simulated
Real time charts
Operating Modes
YATS SARL, 440 653 285 RCS Toulouse , chemin du guerrier 31450 DEYME email : dherlemont@yats.com
training, debug, simulation,
backtesting,
Trading Automata state machine

Computational Finance
Success in the markets will depend on having superior models and algorithms ..
.
During the last 15 years, I studied intensively all aspects of
research in finance with the objective to implement superior
trading algorithms, focusing on the most promising areas:
• Behavioral Finance : deviation from the efficient market
hypothesis stems from behavioral biases and models based
on herding, loss aversion, information flow, sector
rotation, etc ... such behaviors is the source of repeated,
robust and persistent patterns explaining most of market
anomalies, market bubbles and crashes.
• Computational finance, defined as the application of
modern techniques in computer science, mathematics and
statistics to solve complex problems in finance.
• Model based trading consists in modeling markets data
and finding optimal trading systems by Monte Carlo
simulation on model generated data.
Even if most subjects are still active research efforts, they are
integrated within the YATS environment and used to trade
actual markets.
Keywords : Data Mining and Knowledge Discovery, Natural
Language Processing and Information Extraction, Machine
Learning, Pattern Recognition, Genetic and Evolutionary
Computing, Simulation and Monte Carlo.
More specifically, I studied and implemented online
algorithms
• to perform online and real time arbitrage, using
multivariate cointegration and filtering (Kalman filter)
• Non parametric techniques for pattern recognition, like
Variable Length Markov chains
• Reinforcement learning algorithms (Q-learning, TDlearning):
• Wavelets, to identify long term dependencies,
• Universal Online prediction techniques (quite similar to
compression algorithm), universal portfolios, ...
• Methodology to avoid data snooping and overfitting
Some other subjects: I studied and implemented: probabilities,
approximations (Edgeworth, ..) extreme value theory (Cramer,
...), Time Series Analysis (AR, MA, ARIMA, ..), non
parametric methods (gaussian kernels, KNN, LWR, ...),
bayesian learning, chaos (BDS tests), fractals (Hurst
exponents, ...), Markov chains, information theory (relative
entropy), estimation procedure: maximum likelihood, global
optimization (simulated annealing, ...), game theory (Minority
Games, ...), stochastic processes, including first hitting time,
distribution of minimum and maximum, volatility modeling
(GARCH, ..) and estimation using highs and lows, higher
moment estimates, kurtosis/herding, skews /crashes, intraday
modeling : intensity, ACD models, cointegration, Dickey-
Fuller, VAR/VECM, Kalman filter, utility functions, CAPM,
Arbitrage Pricing Theory, VaR, optimal portfolio and
leverage, options pricing (including exotic options), dynamic
programming & optimal control, ...
Natural Language Processing (NLP) can be used as a
support tool. For example, YATS search engines can be
used to screen internet and derive opinions on stocks,
market sentiment, detect rumors like merger, summarize
news, count occurrences of stocks in the press, in forums,
etc .... and to exploit information flow delays or well
known behavioral biases, like popular blueships
experienced more severe crashes than less popular stocks.
Some selected references / authors:
• Jean-Philippe Bouchaud, Marc Potters, 2000, Theory
of Financial Risks, Cambridge University Press,
• Merton, Robert C., 1992, Continuous Time Finance.
• W. Feller, 1997, An introduction to probability theory
and its applications, Vol I & II, second edition - John
Wiley and sons, New York
• Karatzas and Shreve, 1988, Brownian motion and
stochastic calculus, Springer Verlag
• Andrew W. Lo and A. Craig A Non-Random Walk
Down Wall Street
• Grinold, Active Portfolio Management,
• Ralph Vince, 1990, Portfolio Management Formulas,
• Journal Of Finance, Quantitative Finance,
• Behavioral finance literature: R. Thaler, D.
Kahneman, Odean, Chen, M. Rabin, R. Shiller, ...
• and many other authors and references that is not
possible to list here ...
Articles/Authors:
• Rama Cont : stylized facts, herding models, ...
• JP Bouchaud: numerous articles on risk management,
derivatives, market modeling
• Didier Sornette one of the best and most productive
author in "econophysics", specially on crahs
modeling, extreme events modeling, portfolio
management using higher order moments,
• Farmer, Lux, for agent modeling,
• other "econophysics" authors: Maslov, Baviera,
Mantegna, ...
• Ait-Sahalia
• Thomas Cover (Information theory, universal
portfolio)
• Alexander (cointegration), Hasbrouck (intraday
modeling), .
• learning to trade: Moody, Dempster, Jones
In French:
• Denis Bosq, Jean Pierre Lecoutre, 1992, Analyse et
prévision des séries chronologiques, méthodes
paramétriques et non paramétriques, Masson,
• Cours Polytechnique et DEA Probabilités et Finance
Paris VI (El Karoui)
• Rochet, J-C. et G. Demange, 1992, Méthodes
Mathématiques de la Finance, Economica,
• Autres auteurs: Jean Jacques Laffont, Christian
Gourieroux, ...
Mastering C++, JAVA, R-project, Latex, VBA/Excel, ...
YATS SARL, 440 653 285 RCS Toulouse , chemin du guerrier 31450 DEYME email : dherlemont@yats.com

TECHNOLOGIES
Developed up to now (June 2003).
more than 300 000 source lines codes
in JAVA, mastering all JAVA packages, including:
• user interface (SWING),
• graphics and imaging (JAVA2D),
• data bases Interfaces (JDBC),
• network and I/O, RMI, new io package
• security, SSL programming.
• naming
• XML
J2EE:
• Servlets and JSP.
• Enterprise JavaBeans (EJB)
• Java Message Service (JMS)
• Java Mail
• Management / JMX
• Security (JAAS)
Mastering related open source products:
• JAKARTA / Apache suites : TOMCAT (Web
server), JAMES (SMTP, POP3, NNTP server),
TURBINE (servlet framework), JETSPEED
(Enterprise Information Portal)
• Scripting facilities (BSF/ Javascript/JPython)
• XML: Apache (Xerces, COCOON, ...),
• JBOSS, a J2EE,server, featuring JMX, JMS,
• ORION : a full server J2EE, basis of ORACLE
9iAS
• OpenJMS, OPENSYMPHONY,
• ....
Other products : Websphere, Weblogic,
Methods & Tools:
• Object orientation: UML, design patterns
• "`Extreme Programming"', iterative development
with very short life cycles, numerous testing and
frequent refactoring, resulting in high quality
software.
• IDE: JBuilder.
• CVS, for team development and configuration
management.
• JUnit, for unit testing
• ANT, for automated tasks
• Obfuscators
Data bases: Postgresql, MySQL, Microsoft SQL Server,
ORACLE.
Mastering DHTML et JAVASCRIPT, XML/XSL, C++,
C#, .NET, CORBA, ...
Brokerage systems are complex and critical systems, with
stringent requirements on time response, data
management, reliability, security, man machine
interfaces. I developed complementary products, to fulfil
such requirements and improve productivity:
• Servlets compression and caching technology that
may boost some internet based applications by a
factor better than 1000.
• STAY a persistent layer for JAVA, (nearly) compliant
with ODMG 3.0 and JDO (Java Data Object) for
managing high performance, distributed and large
databases. STAY can be used with any Relational
Database.
• YPSO: Yats Persistent Store Objects, to provide high
performance persistent collections (lists, hashmaps,
trees, mapped files..).
• Events server : a robust and high performance
alternative to JMS, based on raw sockets.
• Numerous reusable libraries: scheduling (java cron),
automated FTP, text processing, XML, graphics,
math, statistics, ....
• Real time kernel primitives.
Intelligent parsing and Natural Language Processing for
targeted search engines: opinions on stocks.
Assessments of JINI, Autonomous agents, AOP (Aspect
Oriented Programming), JSP/Servlets frameworks.
Mathematical and Machine learning libraries: numerous
libraries have been developed, wrapping existing open
source libraries, porting some C++ algorithms, and, if
any, redeveloping from scratch :
• math: functional framework, to provide automatic
calculus (composition, integral, derivatives, etc ... )
• global optimisation package (including BFGS,
simulated annealing, GA, combined optimisers)
• statistics (significance testing, p-values, estimations
by maximum likelihood, linear and non-linear
simple/multiple regressions, ...), discrete and
continuous time series, ...
• porting some Numerical Recipes from C to in JAVA
• Markov chain
• Data structures : kdtree, multi-arrays, persistent
collections, ...
• Machine learning algorithms, data mining, ...
LINUX Administration & security: to comply to high
security levels required in brokerage systems: masking
ports, disabling ICMP, using IPCHAIN + NMAP &
NESSUS for testing, SSH2 tunnelling for all activities on
the server (FTP, telnet, CVS, Data Bases accesses, … ),
incoming IP control, cryptography.
YATS SARL, 440 653 285 RCS Toulouse , chemin du guerrier 31450 DEYME email : dherlemont@yats.com

1998-...: YATS e-FINANCE
Search engines, integration of news feeds:
• integrating, more than 30 different sources and 10
000 news per day, with heterogeneous content and
format (HTML, XML, email, ftp pull, ftp push, etc ...
),
• including : REUTERS, DowJones NewsWire,
CompanyNews, PRLine, AFP, ... internet sites:
Firstinvest, boursier.com, smallcaps, Newsinvest,
Square Finance, Trading Central, CercleFinance,
CacTrading, Le Revenu, …) specific sources,
• merging into a single and homogenous feed, including
codes, name, date/hour, URL, text, categories (news,
opinion, country, sector, ....), etc
• dispatching in real time in various formats :
WEB/HTML, WAP/WML, email, ftp, ... with real
time alerts via e-mail, mobile,
• integrating with online payment and billing system.
• Front Office Information System Architectures
interfacing with PATIO of ACTIO/FINANCE and
ATOS brokerage plateforms.
• Automated opinions on stocks, analyzing informal
opinion on the WEB using natural language
processing to derive quantitative opinions, consensus
and divergences.
Automated consensus user interface
Extension to automatic market sentiment, events
detection (merger, ...), ...
• Real time Quotes Feed integration (GL Trade,
Reuters): for France, Europe and US markets.
• Derivatives :
search engine on warrants : sources included Société
Générale, Citibank, BNP, Crédit Lyonnais,
Commerzbank, Dresdner,
pricer with methods based on Black&Scholes,
binomial, and GARCH volatility models
News integration and dispatching for FIMATEX
All components are developed by YATS, including search autonomous
agents, data bases, mini sites for Dowjones, SG Faxes, AOF, Reuters,
opinions pages, specific extractions in XML, ...
Worked continuously 24/7 for more that 4 years
without any interruption of service.
• Expert Market place (specification and design)
Interactive and Collaborative portal for finance
community (using Apache/Jetspeed).
• Mutual fund search engine (on S&P Micropal and
FININFO/Euro Performance data)
• Fundamentals Screening with Jacques CHAHINE
Finance, FININFO, COFISEM,
• Asset allocation and customer profiling tool
• Interactive technical analysis charting .
Main References :
FIMATEX from December 1998 to mid-2003
FIDEURAM- WARGNY : from June 1999 to mid 2002.
YATS SARL, 440 653 285 RCS Toulouse , chemin du guerrier 31450 DEYME email : dherlemont@yats.com