Risk Management in Emerging Markets
Centre for the Study of Emerging Markets Series Series Editor: Dr Sima Motamen-Samadian The Centre for the Study of Emerging Markets (CSEM) Series provides a forum for assessing various aspects of emerging markets. The series includes the latest theoretical and empirical studies from both academics and practitioners in relation to the economies and financial markets of emerging markets. These cover a wide range of subjects, including stock markets and their efficiency in emerging markets, forecasting models and their level of accuracy in emerging markets, dynamic models and their application in emerging markets, sovereign debt and its implications, exchange rate regimes and their merits, risk management in emerging markets, derivative markets and hedging decisions in emerging markets, governance and risk in emerging markets, etc. The series will be one of the main sources of reference on emerging markets, both within and outside those markets, for academics, national and international agencies, and financial institutions. Titles include Sima Motamen-Samadian (editor) CAPITAL FLOWS AND FOREIGN DIRECT INVESTMENTS IN EMERGING MARKETS DYNAMIC MODELS AND THEIR APPLICATIONS IN EMERGING MARKETS RISK MANAGEMENT IN EMERGING MARKETS GOVERNANCE AND RISK IN EMERGING AND GLOBAL MARKETS Also by Sima Motamen-Samadian INTERNATIONAL DEBT AND CENTRAL BANKING IN THE 1980s (edited with Z. Res) EMERGING MARKETS Past and Present Experiences, and Future Prospects (edited with C. Garido)
Centre for the Study of Emerging Markets Series Series Standing Order ISBN 1–4039–9521–4 You can receive future titles in this series as they are published by placing a standing order. Please contact your bookseller or, in case of difficulty, write to us at the address below with your name and address, the title of the series and the ISBN quoted above. Customer Services Department, Macmillan Distribution Ltd, Houndmills, Basingstoke, Hampshire RG21 6XS, England
Risk Management in Emerging Markets Edited by
Sima Motamen-Samadian
© Selection and editorial matter © Sima Motamen-Samadian 2005 Individual chapters © contributors 2005 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No paragraph of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, 90 Tottenham Court Road, London W1T 4LP. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this work in accordance with the Copyright, Designs and Patents Act 1988. First published in 2005 by PALGRAVE MACMILLAN Houndmills, Basingstoke, Hampshire RG21 6XS and 175 Fifth Avenue, New York, N.Y. 10010 Companies and representatives throughout the world. PALGRAVE MACMILLAN is the global academic imprint of the Palgrave Macmillan division of St. Martin’s Press, LLC and of Palgrave Macmillan Ltd. Macmillan® is a registered trademark in the United States, United Kingdom and other countries. Palgrave is a registered trademark in the European Union and other countries. ISBN-13: 978–1–4039–9153–9 ISBN-10: 1–4039–9153–7 This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. A catalogue record for this book is available from the British Library. Library of Congress Cataloging-in-Publication Data Risk management in emerging markets / [edited] by Sima Motamen-Samadian. p. cm.—(Centre for the Study of Emerging Markets series) Includes bibliographical references and index. ISBN 1–4039–9153–7 1. Risk management—Developing countries. 2. Securities—Developing countries. 3. Asset-liability management—Developing countries. I. Motamen-Samadian, Sima. II. Series. HG5993.R57 2005 2005047311 332.67 3 091724—dc22 Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham and Eastbourne
Contents
List of Figures and Tables
vii
Preface
xi
Acknowledgements
xiii
Notes on the Contributors
xv
1
Introduction Sima Motamen-Samadian
1
2
Risky Production and Hedging in Emerging Markets Octave Jokung
5
3
An Analytical Study of Option Greeks on Derivative Markets in India Devendra G. Kodwani
17
Global Asset Allocation: Risk and Return Trade-off on Emerging Stockmarkets Mohamed Derrabi and Michel Leseure
35
Random Walk in Emerging Markets: A Case Study of the Karachi Stock Exchange Orla Gough and Ali Malik
57
Insiders’ Market Timing and Real Activity: Evidence From an Emerging Market Tomasz Piotr Wisniewski
71
4
5
6
v
vi Contents
7
8
Trading Risk Management: Practical Applications to Emerging Markets Mazin A.M. Al Janabi Value at Risk: Does it Work in Emerging Markets? Chuntao Yu, Bob Davidson and Mohamed Nurullah
Index
91
137
165
List of Figures and Tables Figures 3.1 3.2 3.3 3.4 3.5 4.1 6.1 8.1 8.2
Daily average turnover in the derivatives segment of the National Stock Exchange, India 18 Growth in the derivatives segment of the National Stock Exchange, India 18 Actual and theoretical premiums on Nifty index calls 25 Implied volatility on index call option expiring 30 August 2001, K = 1,180 30 Implied volatility on index call option expiring 28 March 2002, K = 1,180 30 The efficient frontier based on the weekly indices of emerging markets and developed markets 46 Orthogonized impulse responses of variables to shocks in equations 82 Selecting models for normal market risk 159 Selecting models for abnormal market risk 160
Tables 3.1 3.2 3.3 3.4 3.5 3.6 4.1 4.2 4.3
Historical volatility estimates of the underlying index 91-day Government of India treasury bill yields Expiry dates and exercise prices on Nifty Index options used in this study Option Greeks: equations used for calculations on call options Call premiums, volatilities and option Greeks on nifty index call options Delta estimates and the likelihood of index call options being in-the-money on expiration Analysis of the stock exchange markets of the sample Coefficients of correlation between stockmarkets Application of the global market model to emerging and industrialized markets vii
22 23 23 24 26 32 42 45 50
viii List of Figures and Tables
5.1 5.2 5.3 6.1 6.2 6.3 7.1 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 8.1 8.2
Results of the Dickey–Fuller unit-root test Results of the autocorrelation test Results of the day-of-the-week effect Descriptive statistics Granger causality tests Three-variable innovation accounting Quantitative analysis data: daily volatility, beta, skewness and kurtosis Quantitative analysis data: annual volatility, beta, skewness and kurtosis Quantitative analysis data: exact correlation matrix Quantitative analysis data: correlation 1 matrix Quantitative analysis data: correlation 0 matrix Equity trading risk management report (analysis of case 1) Equity trading risk management report (analysis of case 2) Equity trading risk management report (analysis of case 3) Equity trading risk management report (analysis of case 4) Equity trading risk management report (analysis of case 5) Equity trading risk management report (analysis of case 6) Equity trading risk management report (analysis of case 7) Equity trading risk management report (analysis of case 8) Equity trading risk management report (VaR limits-settings, case 1) Equity trading risk management report (VaR limits-settings, case 2) Equity trading risk management report (VaR limits-settings, case 3) Equity trading risk management report (VaR limits-settings, case 4) A brief summary of current market-risk measures An overview of the three main VaR methodologies
66 66 67 75 80 84 113 114 115 116 117 119 121 122 123 124 125 127 128 129 130 131 132 138 140
List of Figures and Tables ix
8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11 8.12 8.13 8.14 8.15 8.16 8.17 8.18
A summary of the key strengths and weaknesses of three VaR methodologies Responses to question 1 Responses to question 2 Responses to question 3 Responses to question 4 Responses to question 5 Responses to question 6 Statistical results of questions 7 and 8 Statistical results of Q9 Responses to question 10 Responses to question 11 Responses to question 12 Responses to question 13 Responses to question 14 Responses to question 15 Responses to question 16
141 143 143 144 145 146 146 147 149 150 151 151 152 152 153 153
This page intentionally left blank
Preface The eight studies presented in this volume are put together to provide a new insight into the design of risk-management models in emerging markets. The objective is to identify the specific characteristics of emerging markets and specify the most appropriate methods of risk management that suits those markets. The chapters report on empirical studies carried out on a number of countries in Asia, Eastern Europe, North Africa and other emerging markets in various continents. They present the latest findings that are important for better understanding of the nature of risks in those markets, and useful to all those involved in decision-making for investment in emerging markets. Chapter 2 looks at hedging decisions in the presence of price and political risks in emerging markets. Chapter 3 examines the volatility of the index and security-based options in India’s derivative market. Chapter 4 is about asset allocation in both emerging and developed economies, and the extent to which inclusion of emerging markets in a portfolio can affect the overall risk and return of the portfolio. Chapter 5 tests the efficiency of the Karachi Stock Exchange, and Chapter 6 discusses the extent of aggregate insider trading in the Polish stockmarket and the way it affects the market and the economy. Chapter 7 analyses the problems of trading risk management in emerging markets and focuses on the Moroccan stockmarket. Chapter 9 evaluates the applicability of Value at Risk models in emerging economies and offers some new ideas on how the model can be improved to fit emerging markets. Overall the book provides a good coverage of the latest findings about risk management in a diverse range of emerging markets across the world. SIMA MOTAMEN-SAMADIAN
xi
This page intentionally left blank
Acknowledgements This volume is a collection of some of the papers that were presented at the International Conference on Emerging Markets and Global Risk Management in June 2004 in London. The conference was organized by the Centre for the Study of Emerging Markets (CSEM) at the Westminster Business School. My special thanks go to all the contributors for their timely delivery of the chapters, and to my family and in particular my husband Vahab Samadian for his continuous support through the period when I was working on the book. SIMA MOTAMEN-SAMADIAN
xiii
This page intentionally left blank
Notes on the Contributors Mazin A.M. Al Janabi is an Associate Professor of Finance and Banking at the School of Business Administration, Al Akhawayn University, Ifrane (AUI), Morocco. Bob Davidson is a Principal Lecturer in Finance in the Division of Risk, Caledonian Business School, United Kingdom. Mohammed Derrabi is an Associate Professor at the School of Business Administration, Al-Akhawayn University in Ifrane, Morocco. Orla Gough is Chair of the Department of Finance and Business Law at the Westminster Business School, University of Westminster, United Kingdom. Octave Jokung is Associate Professor at the Edhec Business School, Lille, France. Devendra G. Kodwani is a Lecturer in Finance at the Open University Business School, Milton Keynes, United Kingdom. Michel Leseure is a Lecturer in Technology and Operations Management, Aston Business School, Aston University, United Kingdom. Ali Malik is a Visiting Lecturer at the Westminster Business School, University of Westminster, United Kingdom. Sima Motamen-Samadian is Director of the Centre for the Study of Emerging Markets and a Principal Lecturer in Economics at the Westminster Business School, University of Westminster, United Kingdom. Mohamed Nurullah is a Lecturer in the Division of Risk, Caledonian Business School, United Kingdom. Tomasz P. Winsiewski is a Senior Lecturer in Finance at the Department of Finance, Auckland University of Technology, New Zealand. Chuntao Yu is a Consultant at Price Waterhouse Coopers, China.
xv
This page intentionally left blank
1 Introduction Sima Motamen-Samadian
The finance community is increasingly demanding a systematic approach to risk management. Most of the risk management models developed so far tend to be more suitable to markets and institutions in developed economies, and the growth of emerging markets in recent years has raised the need for a reexamination of the existing models and development of a new set of models that take into account the specific features of emerging economies. In this respect this book is designed to provide an understanding of the type of risks that investors might face in various emerging markets and the usefulness of some of the existing models in assessing the trade-off between risks and returns in those markets. The eight studies included here provide a valuable insight into the type of risks that investors might face in emerging markets and the appropriate methods of risk management that should be used. These include background risks such as human, capital and political risks that are non-diversifiable and common in emerging markets, as well as production risks, derivative risks, asset allocation and return tradeoff risks. They also include studies of risks associated with information efficiency of the stockmarkets and the Value at Risk (VaR) in emerging markets. The countries examined include India, Pakistan, Poland and Morocco, as well as a study that examines risks of portfolios that include 16 emerging and 14 developed markets. In Chapter 2 that follows this Introduction, Octave Jokung develops a model in which he tries to incorporate two types of risks; the background risk or non-diversifiable risk such as political risk that investors more commonly experience in emerging markets, and a price risk 1
2 Introduction
that is diversifiable. The author analyses the behaviour of investors with respect to optimal production and hedging in the presence of the above two risks, and shows that when price risk is independent of background risk, the investor’s willingness to hedge decreases if the forward market exhibits contango and increases in the case of backwardation. Jokung also shows that when price risk and background risk are independent, wealthier individuals with decreasing risk-aversion tend to invest a larger proportion of their initial wealth in the forward contract. In the case of dependence of price and background risk, however, the investor over or under-hedges when the dependence is positive or negative respectively. In Chapter 3, Kodwani examines the volatility of index and securitybased options in India. The derivative market was first introduced in India in 2001 and has since grown significantly. The market plays an important role in helping investors to diversify stockmarket risks and hence deserves careful examination. Kodwani provides a valuable insight into the depth and maturity of the Indian derivatives market. Chapter 4 is about global asset allocation and the risk/return trade-offs for investors who invest in both emerging and developed economies. Here Derabi and Leseure use the database of International Finance Corporations (IFC) that include 16 emerging markets and 14 developed markets, and apply it to the Markowitz model and assess the extent by which inclusion of emerging markets affects the risk and return of the portfolio. Their results confirm the expectation that inclusion of emerging markets will increase the return of the portfolio but also add to its risks. They also use the Capital asset pricing model (CAPM) and find that the specific risk associated with emerging markets is higher than that of developed markets. Overall, Derabi and Leseure argue that investment in emerging markets can only grow if the emerging market authorities make more efforts to improve the operational and informational efficiency of those markets. Gough and Malik in Chapter 5 move our attention to Pakistan and examine the extent of the efficiency of that market. They apply the Dickey Fuller and autocorrelation tests to the data from the Karachi Stock Exchange (KSE), and find that the KSE does not satisfy the criteria of weak-form efficiency. In other words, it is possible for investors to use past information to earn abnormal returns in that market. Moreover, the test for the day of the week effect reveals a negative return on Friday which is contrary to the existing evidence.
Sima Motamen-Samadian 3
In Chapter 6, Wisniewski examines the Polish stockmarket for the presence of aggregate insider trading, and finds widespread evidence of such practices. The author tries to determine whether insider trading can be used to predict stockmarket returns and future real activity. He uses a trivariate VAR model of real market returns, growth in industrial production and aggregate insider trading, and shows that insiders seem to observe unexpected changes in cash flows to their own companies prior to their public disclosure, but cannot distinguish whether these are due to firm-specific or economy-wide factors. In this respect Wisniewski’s results are consistent with the hypothesis propounded by Seyhun (1992) who came to the same conclusion. Wisniewski also finds that aggregated insider transactions appear to contribute to growth in industrial production and real returns on an equally weighted market portfolio. Moreover, his results imply that informed agents have relatively short average investment horizons arising from the absence of short-swing profit restrictions in Polish law. Chapter 7 provides an insight into the problems of trading risk management and their application to emerging markets with a specific focus on the Moroccan stockmarket. Here Al Janabi first highlights the important role of emerging markets in recent years in portfolio diversification for both institutional and retail investors. He argues that emerging markets have enormous advantages for market participants despite the fact that they are characterized as illiquid, segmented and politically unstable. Indeed it is their individual differences from developed markets that create unique opportunities. The high expected returns are of course embedded with risks, but these can be managed through specific risk-management techniques. Al Janabi focuses on trading-risk management and uses real-world examples and practical reports of equity trading-risk management in the Moroccan stockmarket to show the proper use of VaR and stresstesting (scenario analysis) methods. The study can provide a valuable insight for financial entities, regulators and policy-makers in settingup their trading-risk management objectives to match the specific needs of emerging-markets. Finally, Chapter 8 once again questions the applicability of the VaR models to emerging markets, and finds some interesting results that can be of interest to both fund managers and regulators. Here Yu, Davidson and Nurullah use different types of models to calculate Value at Risk and provide an extensive literature review, considering
4 Introduction
each of them in turn. In doing so they try to identify the strengths and weaknesses of each model and their suitability to calculate VaR in given situations in emerging markets. They point out that in the past a great deal of time and effort could have been wasted by organizations trying to reduce risk, when in fact no risks existed. There were also cases were institutions failed to target risks because they were not identified by the models used. To address the above problem and to identify the most appropriate VaR models, Yu, Davidson and Nurullah surveyed a panel of senior management ‘experts’, some of whom were based in emerging markets, to incorporate the practitioner view into the research and to ensure that no problems were mistakenly included/excluded. The panel also helped to identify the optimal usage for each model and the problems related to each in a given situation. This involved a cross-sectional analysis of the survey that revealed the need to hierarchically rank the risks being faced by the organization. All these helped to develop the most appropriate model for emerging markets. The study shows that while not all models work well in all situations, some tend to deliver better results in some markets. Moreover, it might be better to use a combination of models to get an optimal result in specific circumstances.
2 Risky Production and Hedging in Emerging Markets Octave Jokung
Introduction Several financial decisions are made in the presence of more than one single source of risk. Among those risks, the class of non-tradable risks or background risks is preeminent. This chapter addresses the question of whether an increase in initial wealth leads an individual to hedge in the forward market in the presence of a background risk. We also analyse how far the behaviour of the decision-maker is affected by the presence and the modification of a non-tradable risk like nondiversifiable income, human capital, political risk, non-marketable assets, informational asymmetries and irreplaceable commodities. This is the case when investing in emerging markets, where investors face two sources of risk: economic risk and background risk. Investors in emerging markets face non-tradable risks which add background to their investment, and when this background risk increases the investors’ willingness to hold risky assets must decrease. Limiting our investigation to a two-period model, we assume that the individual has the choice to cover more or less the risky asset and to determine the optimal level of his production. This chapter addresses mainly two questions. What are the optimal decisions on production? What is the optimal hedging decision? We consider two risks in the problem: the random price of the risky asset and the background risk. Multi-risk problems have only been studied recently; Kihlstrom, Romer and Williams (1981). Pratt and Zeckhauser (1987) and Pratt (1988) are important contributions. Kimball (1993) defines, in an expected utility framework, the property of standard 5
6 Risky Production and Hedging in Emerging Markets
risk-aversion which implies that the presence of some background risk reduces the optimal investment in a risky security with an independent return. Eeckhoudt, Gollier and Schlesinger (1996) study the effect of independent exogenous risk on optimal risk-taking behaviour towards an endogenous risk. They examine conditions on preferences under which some changes in the distribution of the background wealth entail more risk-averse behaviour towards endogenous risk. Eeckhoudt and Kimball (1992) study how a background risk affects optimal demand for insurance against an independent or dependent insurable risk. Eeckhoudt, Gollier and Schlesinger (1996), Gollier and Pratt (1996), Kimball (1993) and Pratt and Zeckhauser (1987) among others show that, under fairly general conditions on the utility function, investors reduce their holding of risky assets when facing an increase in the background risk. Elmendorf and Kimball (1999) use two instruments, the saving and the risky investment, to consider the problem of how to save more or/and to reduce the exposure to background risk. They show that an increase in the variance of permanent income shocks leads to a reduction in both the optimal portfolio allocation to risky investment and the consumption-labour income ratio for any utility function that exhibits decreasing absolute risk-aversion and decreasing absolute prudence in the sense of Kimball (1990). Dor and Jokung (2003) use an inter-temporal framework where the values of the two control variables are jointly and interdependently determined, and they show that the fraction of initial wealth invested in the risky asset is larger for wealthier individuals with decreasing risk-aversion when labour income risk and asset return risk are independent. Our work here is in the same vein. We first point out the separation property as the optimal production level is independent of the price risk. We then analyse the optimality of full coverage in the case of an unbiased or biased forward market. We recover the ‘full hedge theorem’, and the results obtained by Adam-Müller (1993), Briys, Crouhy and Schlesinger (1993) and Franke, Stapelton and Subrahmanyam (1998). We then analyse the behaviour of investors in the case of an increase in initial wealth. Finally, we tackle the dependence case in two steps: firstly with a linear dependence, and secondly with a general dependence but for the case of an unbiased forward market. The chapter is organized as follows. The next section presents the model and the separation property. We then analyse the effect of
Octave Jokung 7
the background risk on the optimal coverage, before presenting the wealth effect on hedging decision. The general case of dependence with an unbiased market is then analysed, and the special case of linear dependence with both biased and unbiased markets. The last section concludes the chapter.
The model Let us consider a two-period model where a risk-averse individual bears a risky asset coming from his production and a riskless asset. The endowment of the risky asset is given by q units of the risky asset whose price denoted by p˜ is risky. The price risk is tradable in a competitive forward market where f is the current forward price for delivery of one unit of the risky asset at maturity. Let us denote by c(q) the cost function which is increasing and convex. The final wealth is given by equation (2.1): ˜ = a − c(q) + qp˜ + F(f − p˜ ) + ε˜ w
(2.1)
where a is the amount of sure wealth, F is the quantity of the risky asset sold forward, and ε˜ is the additive zero-mean background risk modeling the specific risk of the emerging market. The individual maximizes his expected utility function: Max q,F
˜ Eu(w)
(2.2)
The operator E is the expectation operator with respect to the joint distribution of p˜ and ε˜ . The problem yields the following first-order conditions: ˜ =0 E(f − p˜ )u (w)
(2.3)
˜ =0 E(c (q) − p˜ )u (w)
(2.4)
and
The last two equations become: f =
˜ Ep˜ u (w) ˜ Eu (w)
(2.5)
8 Risky Production and Hedging in Emerging Markets
and c (q) =
˜ Ep˜ u (w) ˜ Eu (w)
(2.6)
Therefore c (q) = f and we get the following proposition: Proposition 1 If a forward market exits, then the optimal production rule is given by the maximization of fq − c(q). The optimal production decision is independent of the investor’s preferences. This proposition states that the production decision does not depend on the hedging decision. The two decisions are separated. Therefore the separation property holds. We can also remark that the optimal production is independent of the price of the risky asset. The production decision is also independent of the investor’s preferences because we don’t need the shape of the utility function in order to determine the level of production.
Optimality of full coverage in the presence of an independent background risk In this section, we analyse the optimal hedging decision. The use of forward contracts enables an individual to cover tradable risk, namely the price risk. We are going to analyse the position of the quantity of the risky asset sold forward with respect to the initial exposure. To do so, we must evaluate the first-order condition with F = q. Evaluating this condition given optimal hedging with full coverage gives: dEu(w) = E (f − p˜ )u (a − c(q) + qf + ε˜ ) dF F=q
(2.7)
This condition is equal to: dEu(w) = E f − p˜ E u (a − c(q) + qf + ε˜ ) dF F=q
(2.8)
And the optimality of full coverage depends on the sign of E(f − p˜ ) = f − E(p˜ ) which is the risk premium in the forward market. At this stage we need some definitions concerning forward markets and forward positions. The forward market is said to be unbiased if
Octave Jokung 9
the risk premium is zero. If the risk premium is negative, the forward market exhibits backwardation. If the risk premium is positive, the forward market exhibits contango. The situation where the individual sells more in the forward market than his endowment is called overhedging. The opposite situation is called under-hedging.
An unbiased forward market Let us first consider the case of an unbiased forward market; that is, the expected price coincides with the forward price: E(p˜ ) = f . The first-order condition evaluated at F = q becomes: dEu(w) = E[f − p˜ ]E[u (a − c(q) + qf + ε˜ )] = 0 (2.9) dF F=q Therefore, the optimal hedging decision is full coverage and F ∗ = q. The presence of an independent background risk has no effect on the hedging decision. Recall that in absence of the background risk the optimal solution is also full coverage. This means that in an unbiased market the presence of background risk does not modify the attitude of the individual! We recover the ‘full hedge theorem’ which states that full coverage is optimal with an unbiased forward market.
A biased forward market In the case of a biased forward market, we tackle the optimality of full coverage with two hypotheses. In the first case, the forward market exhibits backwardation which means that the expected price is greater than the forward price (E(p˜ ) < f ). The first-order condition leads to: dEu(w) = E[f − p˜ ]E[u (a − c(q) + qf + ε˜ )] < 0 (2.10) dF F=q and over-hedging is optimal: F ∗ > q. In the second case, the forward market exhibits contango which means that the expected price is less than the forward price (E(p˜ ) > f ) and the first-order condition becomes: dEu(w) = E[f − p˜ ]E[u (a − c(q) + qf + ε˜ )] > 0 (2.11) dF F=q
which means under-hedging is optimal: F ∗ < q. Finally, we recover the results obtained when there is no background risk. Nevertheless, there is a magnitude effect when the market
10 Risky Production and Hedging in Emerging Markets
is biased. Following Kihlstrom, Romer and Williams (1981) let us define the following function: ˆ u(x) = Eε˜ [u(x + ε˜ )]
(2.12)
where Eε is the expectation operator with respect to the marginal distribution of ε˜ . uˆ is the so-called derived utility function which takes into account the presence of the background risk. To point out the magnitude effect, we are going to use the fact that in portfolio theory, the demand for a risky asset is lower (in absolute value) in the absence of background risk because the utility function uˆ is more risk-averse (or concave) than the utility function u when u exhibits decreasing riskaversion; therefore, inducing a lower risky portfolio. This property follows directly from the standard result that increased risk-aversion leads to a decrease in the optimal risky investment (Arrow, 1971). To do so, we must show that our framework can be related to portfolio choice by rewriting the final wealth as follows: ˜ = a − c(q) + qp˜ + F(f − p˜ ) + ε˜ = (a − c(q) + qf ) + (F − q)(f − p˜ ) + ε˜ w = (a − c(q) + qf ) + (q − F)(p˜ − f ) + ε˜ = w + AR˜ + ε˜
(2.13)
Therefore q − F and p˜ − f play the roles of the demand for the risky asset and the risky return respectively. a − c(q) + qf = w represents the initial wealth in portfolio theory. ˆ The following program gives the hedging decision with u: Max F
ˆ − c(q) + qf + (q − F)(p˜ − f )) Eu(a
(2.14)
the optimal solution of which, Fuˆ , is also the solution of our initial program, namely: Max F
Eu(a − c(q) + qf + (q − F)(p˜ − f ) + ε˜ )
(2.15)
The use of the derived utility function permits us to replace a problem with two sources of risk by a problem with a single source of risk. Let Fu be the optimal coverage with utility function u and without the background risk. It is the solution of the following program: Max F
Eu(a − c(q) + qf + (q − F)(p˜ − f ))
(2.16)
In order to point out the magnitude effect we must compare Fuˆ and Fu which are respectively the demand for coverage in the presence of background risk and without background risk. Arrow (1971) showed
Octave Jokung 11
that the more risk-averse individual would demand less risky asset in ˜ > 0 implies Au ≥ Auˆ > 0 and E(R) ˜ < 0 implies absolute value: E(R) Au ≤ Auˆ < 0. Applying the previous result coming from the area of portfolio theory to our framework gives: E(p˜ ) > f ⇒ Fu ≤ Fuˆ < q
and E(p˜ ) < f ⇒ q < Fuˆ ≤ Fu
Finally, we can say that when the forward market exhibits backwardation, the background risk increases the coverage and the result is reversed when the forward market exhibits contango. In the case of backwardation, the presence of background risk induces an increase in the willingness to take coverage, which is not the case when the forward market exhibits contango. We can group these results in the same proposition: Proposition 2 When the forward market is unbiased, the optimal decision is full coverage. When the forward market exhibits backwardation, under-hedging is optimal and this result is reversed when the forward market exhibits contango. The background risk induces an increase in the demand for coverage in the case of backwardation and a decrease in the case of contango. The presence of the independent background risk induces a more riskaverse attitude; the investor becomes more risk-averse knowing that he faces a background risk.
The wealth effect on optimal hedging with independent price and background risks Totally differentiating the first-order condition with respect to a and F, and solving this equation, the effect of a change in initial income on optimal hedging is given by: ˜ dF E(f − p˜ )u (w) =− da ˜ E(f − p˜ )2 u (w)
(2.17)
and the denominator is negative due to the second-order condition of the optimization problem (the investor is risk-averse). ˜ Assuming risk-aversion, then dF/da has the sign of E(f − p˜ )u (w). The final wealth can then be rewritten as: ˜ = a − c(q) + qf + (q − F)(p˜ − f ) + ε˜ w
(2.18)
12 Risky Production and Hedging in Emerging Markets
˜ has to be studied. HowWith this notation, the sign of E(f − p˜ )u (w) ever, because of the independence between p˜ and ε˜ , we can use the ˜ is exactly the one derived function uˆ and the sign of E(f − p˜ )u (w) ˜ of E(f − p˜ )uˆ (w).
An unbiased forward market ˜ In this case, the optimal coverage is full hedging and E(f − p˜ )uˆ (w) becomes: E(f − p˜ )uˆ (a + qf + (q − q)(p˜ − f )) = E(f − p˜ )uˆ (a + qf ) = E[(f − p˜ )]uˆ (a + qf ) = 0
(2.19)
There is no wealth effect. This result is obvious because in the case of an unbiased market, the investor always takes full coverage.
A biased forward market Assume that the forward market exhibits contango. Therefore, the quantity of a risky asset sold forward is less than the initial exposure. The assumption of decreasing absolute risk-aversion of the utility ˆ function u, implies having decreasing absolute risk-aversion with u. This implication is shown by Kihlstrohm, Romer and Williams (1981) and Kimball (1993). Jokung (2004) showed that in the general case of dependence between the two sources of risk, under decreasing absolute risk-aversion, the quantity of the risky asset sold forward in excess (F − q) decreases in absolute value when initial wealth increases. We then have the following proposition: Proposition 3 Under decreasing risk-aversion and in the presence of an independent background risk, the quantity of a risky asset sold forward decreases in absolute value when initial wealth increases if the forward market is biased. Otherwise, optimal hedging is independent of wealth.
The effect of a dependent background risk on the hedging decision Up to now it has been assumed that the individual is faced with two independent sources of risk which are fixed. Let us assume that the background risk is related to the price risk as follows: ε˜ = α + β p˜ .
Octave Jokung 13
The final wealth becomes: ˜ = a − c(q) + qf + (q − F)(p˜ − f ) + α + β p˜ w
(2.20)
which can be rewritten as: ˜ = [a − c(q) + qf + fF + α] + [p˜ (q − F + β)] w
(2.21)
The first part is certain, whereas the second part is uncertain in general. However, the second part can be certain with a special choice of the level of coverage. Proposition 4 When the background risk and the price risk are related thanks to a linear relationship, then the individual fully hedges his final wealth if and only if the hedge position is given by the endowment of the risky asset minus the beta of the background risk with respect to the price risk. The former result depends on the assumption concerning the relationship between the two sources of risk. We assume a perfect correlation between the price risk and the background risk. This is acceptable in emerging markets because the background risk corresponds to the economic risk and therefore it is strongly related to the price of the risky asset. Another implication of this situation is that the optimal hedging policy depends on the sign of the beta of the background risk with respect to the price risk. That is, if the beta is positive (respectively negative) under-hedging (respectively over-hedging) is optimal. In practice, the investor regresses the background risk with respect to the price risk and obtains the beta. Thanks to the sign of beta, the investor knows if he has to over-hedge or under-hedge. Let us relax the assumption regarding the dependence between the background risk and the price risk and recall that the first-order condition for the hedging decision is given by: ˜ =0 E(f − p˜ )u (w)
(2.22)
Which becomes with the definition of the covariance: E[f − p˜ ] =
˜ Cov(f − p˜ , u (w)) ˜ Eu (w)
(2.23)
The sign of the risk premium depends solely on the sign of the covariance because the marginal utility is always positive. We focus on an
14 Risky Production and Hedging in Emerging Markets
unbiased forward market; that is, the premium is zero. The first-order condition then becomes: ˜ Cov(f − p˜ , u (w)) =0 ˜ Eu (w)
(2.24)
In order to sign the position in the forward market, we need to evaluate the covariance with the full coverage: ˜ F=q = Cov(f − p˜ , u (a + fF − c(q) + ε˜ )) Cov(f − p˜ , u w) (2.25) Or equivalently:
˜ F=q = −Cov(p˜ , u (a + fF − c(q) + ε˜ )) Cov(f − p˜ , u w)
(2.26)
First case: the relationship between the two sources of risk is positive. With this assumption, the covariance evaluated at F = q will be negative due to the fact that marginal utility decreases with wealth. Thus, over-hedging is optimal. Second case: the relationship between the two sources of risk is negative. The covariance evaluated at F = q will be positive for the same reason as above, and under-hedging is optimal. Proposition 5 Under an unbiased forward market, over-hedging (respectively under-hedging) is optimal when the dependence between the background risk and the price risk is positive (respectively negative). In the case of dependence between the two sources of risk, the ‘full hedge theorem’ is no longer valid because the investor under-hedges or over-hedges depending on the sign of the dependence.
Conclusion In this chapter we have generalized the ‘full hedge theorem’ to the case of two independent sources of risk, and have recovered the separation property concerning the optimal production decision. We have pointed out the fact that the quantity of a risky asset sold optimally forward is unambiguously larger for wealthier individuals when the price risk and background risk are independent. Therefore, the investor’s willingness to cover the risky asset increases. We have showed that the sign of the risk premium is given by the beta when the background risk is a linear function of the price risk. Furthermore, under an
Octave Jokung 15
unbiased market, the ‘full hedge theorem’ is no longer valid when the two sources of risk are dependent. And with a positive dependence the position in the forward market depends on the sign of the dependence.
References Adam-Müller, A.F.A. (1997) ‘Export and Hedging Decisions under Revenue and Exchange Rate Risk: A Note’, European Economic Review, no. 41, pp. 1421–6. Adam-Müller, A.F.A. (2000) ‘Hedging Price Risk when Real Wealth Matters’, Journal of International Money and Finance, no. 19, pp. 549–60. Arrow, K.J. (1971) ‘Exposition of the Theory of Choice under Uncertainty’, in K.J. Arrow, Essays in the Theory of Risk Bearing (Amesterdam: Elsevier). Briys, E., Crouhy, M. and Schlesinger, H. (1993) ‘Optimal Hedging in a Futures Market with Background Noise and Basis Risk’, European Economic Review, no. 37, pp. 949–60. Briys, E. and Schlesinger, H. (1993) ‘Optimal Hedging when Preferences are State Dependent’, Journal of Futures Markets, no. 13, pp. 441–51. Doherty, N. and Schlesinger, H. (1993) ‘Optimal Insurance in Incomplete Markets’, Journal of Political Economy, vol. 91, pp. 1045–54. Dor, E. and Jokung, O. (2005) ‘Expected or non Expected Utility and the Optimal Choice of Saving and Endogenous Capital Risk’, in Changing Models G. Giappichelli (ed.) (Torino: Kluwer), forthcoming. Eeckhoudt, L. and Kimball, M. (1992) ‘Background Risk, Prudence and Insurance Demand’, in G. Dionne (ed.), Contributions to Insurance Economics (Dordrecht: Kluwer), pp. 239–55. Eeckhoudt, L., Gollier, C. and Schlesinger, H. (1996) ‘Changes in Background Risk and Risk Taking Behaviour’, Econometrica, vol. 64, no. 3, pp. 683–89. Elmendorf, D. and Kimball, M. (2000) ‘Taxation of Labor Income and the Demand for Risky Assets’, International Economic Review, vol. 41, no. 3, pp. 801–33. Franke, G., Stapelton, R.C. and Subrahmanyam, M.G. (1998) ‘Who Buys and Who Sells Options: The Role of Options in a Economy with Background Risk’, Journal of Economic Theory, vol. 82, pp. 89–109. Gollier, C. and Pratt, J. (1996) ‘Risk Vulnerability and the Tempering Effect of Background Risk’, Econometrica, vol. 64, no. 5, pp. 1109–23. Jokung, O. (2002) ‘The Effects of Background Risk on Optimal Portfolios’, in I. Hasan and W.C. Hunter (eds), Research in Banking and Finance, vol. 2 (Amesterdam: Elsevier), pp. 123–47. Jokung, O. (2004) ‘Risky Assets, and Hedging in Emerging Markets’, Economics and Financial Modelling, Summer. Kihlstrom, R., Romer, D. and Williams, S. (1981) ‘Risk Aversion with Random Initial Wealth’, Econometrica, vol. 49, pp. 911–20. Kimball, M. (1990) ‘Precautionary Savings in the Small and in the Large’, Econometrica, vol. 58, no. 1, pp. 53–73.
16 Risky Production and Hedging in Emerging Markets
Kimball, M. (1993) ‘Standard Risk Aversion’, Econometrica, vol. 61, no. 3, pp. 589–611. Mayers, D. and Smith, C.W. (1983) ‘The Interdependence of Individual Portfolio Decisions and the Demand for Insurance’, Journal of Political Economy, 91, no. 2, pp. 304–11. Pratt, J. and Zeckhauser, R. (1987) ‘Proper Risk Aversion’, Econometrica, vol. 55, no. 1, pp. 143–54. Pratt, J. (1988) ‘Aversion to One Risk in the Presence of Others’, Journal of Risk and Uncertainty, vol. 1, pp. 396–413.
3 An Analytical Study of Option Greeks on Derivative Markets in India Devendra G. Kodwani
Introduction: growth of derivatives markets in India Derivatives markets in India are in a nascent stage at present. The National Stock Exchange of India (NSE) commenced trading in derivatives with index futures on 12 June 2000. The futures contracts on the NSE are based on S&P (Standard and Poor’s) CNX Nifty (National index of fifty shares); and options and futures on stocks were introduced in July and November 2001 respectively. Before derivatives were introduced in Indian financial markets, there was a localized solution to the need for short-term holdings of securities for the purpose of hedging or speculating. This system was known as badla transaction which essentially involved borrowing securities for a settlement period and squaring up short positions. That system was far from transparent and transaction costs (carry forward or backwardation charges) could sometimes be very high. The introduction of derivatives has enhanced the scope for all investors to participate in the stockmarket at lower transaction costs. More importantly, the introduction of derivatives goes a long way towards enabling institutional investors to provide better avenues for portfolio risk management. The popularity of derivatives is evident from the significant growth rates observed in the turnover in the derivatives segment since they were introduced. The monthly number of contracts on futures and options (F&O) rose from 35,000 in June 2001 to 6.5 million contracts in April 2004. In value terms, the F&O contracts moved from monthly values of Rs 0.37 billion to Rs 110 billion. Total F&O turnover as a percentage of the cash market turnover was 17
4
3
-0
-0
ar
ec D
M
3
03 p-
Se
Ju n
-0
3
2
ar
-0
-0
D
ec
p-
Se
M
2
02
2
-0
Ju n
1
-0
-0
ar M
D
ec
p-
Se
Ju n
-0
1
180 160 140 120 100 80 60 40 20 0 01
Average turnover (Rs bn.)
18 Option Greeks on Derivative Markets in India
Figure 3.1 Daily average turnover in the derivatives segment of the National Stock Exchange, India
4
M
ar
-0
3 -0
03
ec D
p-
Se
-0
3
3 -0 ar
Ju n
2 M
-0
02
ec D
-0
Se
p-
2
2 -0 ar
M
Ju n
1 -0
p-
ec D
Se
-0 Ju n Figure 3.2 India
01
8,000 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0 1
No. of contracts (000s)
Source: Based on the data of National Stock Exchange, Mumbai, India, www.nseindia.com
Growth in the derivatives segment of the National Stock Exchange,
Source: Based on the data of National Stock Exchange, Mumbai, India, www.nseindia.com
218 per cent in April 2004. Figures 3.1 and 3.2 depict the growth in derivatives markets. In addition to the NSE, derivatives have now been introduced on the Mumbai Stock Exchange, which is much older than the NSE. Options contracts on the Mumbai Stock Exchange are of the American type. This study examines in an exploratory way the implied volatility and delta estimates on the index options on a sample representing the first four years of derivative trading. The delta, which is
Devendra G. Kodwani 19
first partial derivative of the option price to the value of underlying asset, has three common uses in practice (Strong, 2000). First, the delta indicates change in the option price from a small change in the price or value of underlying asset, which in present study is the index value. Second, the delta estimate can be used as hedge ratio, i.e. the number of contracts required on the underlying asset to mimic the returns of the option. For example, a call delta of 0.7 on a share would mean that the call option will act like 0.7 of the share. The third use of delta is that it is crude measure of likelihood of an option to be in the money at the time of expiration. Later in the chapter the delta estimates found on the index options traded on NSE are discussed. There is hardly any previous research available along these lines for the Indian derivatives market. ‘Volatility smiles’ are a well-known phenomenon in the derivatives literature, and here we investigate the behaviour of index options and the implied volatility in them. Our study also estimates the delta for European Index call options. Again there is very little empirical literature on this subject in the context of India. Overall there have been few studies on the Indian derivatives market. Srivastava Yadav and Jain (2002) investigated the stock index futures market in India from the efficiency angle, while Shenbagaraman (2003) examined the effect of options and futures trading on the volatility of stockmarkets. The Indian derivatives market with its very short history has not attracted the attention of many financial economists so far, and our study is one of the few efforts in this context. We also examine the pricing efficiency of this nascent market and find encouraging results. It is found that actual premiums closely track the values obtained from the Black and Scholes pricing model, thereby providing support for efficient pricing of the derivatives on this market during the study period; although conclusive evidence requires more rigorous analysis of pricing of derivatives on wider data from stock options as well as futures contracts. The rest of the chapter is organized as follows: in the following section our methodology and data are described including the description of some technical features of the index options on the Indian National Stock Exchange; we then discuss the results on the pricing of options, implied volatility analysis and option Greeks; in particular, delta estimates obtained in this study. A final section concludes.
20 Option Greeks on Derivative Markets in India
Methodology and data The theoretical premiums on index options have been calculated using the following Black and Scholes option pricing model (BSOPM) for European calls. The underlying index is the S&P CNX Nifty of the National Stock Exchange, India. This is a popular benchmark index on the NSE. C = SN(d1 ) − Ke−rt N(d2 ) √ S σ2 d1 = ln + r+ t/σ t K 2 √ d2 = d1 − σ t
(3.1) (3.2) (3.3)
where C is the price of a call option; S is the price of the underlying asset; K is the strike price of the option; r is the rate of interest; t the time to expiration; σ is the volatility of the underlying asset; N(d1 ) and N(d2 ) are two integrals of the standard normal density; and ln represents the natural logarithm of a number.
Some features of index option contracts on the NSE in India • Contract size. The permitted lot size of S&P CNX Nifty options contracts is 200 and multiples thereof, and price steps in respect of these options is Rs 0.05. S&P CNX Nifty options contracts have a maximum of a three-month trading cycle – the near month (one), the next month (two), and the far month (three). On expiry of the near month contract, new contracts are introduced at new strike prices for both call and put options, on the trading day following the expiry of the near-month contract. The new contracts are introduced for a three-month duration. • Expiry day. S&P CNX Nifty options contracts expire on the last Thursday of the expiry month. If the last Thursday is a trading holiday, the contracts expire on the previous trading day. • Strike price intervals. The Exchange provides a minimum of five strike prices for every option type (that is, call or put) during the trading month. At any time, there are two contracts in-the-money (ITM), two contracts out-of-the-money (OTM) and one contract at-the-money (ATM). The strike price interval is 10. New contracts with new strike prices for existing expiration dates are introduced for trading on the next working day based on the
Devendra G. Kodwani 21
previous day’s close Nifty values, as and when required. In order to decide upon the at-the-money strike price, the Nifty closing value is rounded off to the nearest 10. The in-the-money strike price and the out-of-the-money strike price are based on the at-the-money strike price interval.
Historical volatility Historical volatility for each of the contracts analysed here is estimated from the daily lognormal returns calculated on the underlying index. The estimation period is taken as 90 days immediately preceding the launch of each contract. The variance so calculated is converted to annual volatility by multiplying the daily volatility by 250. The historical volatility estimates of the underlying index are given in Table 3.1. The risk-free interest rate is taken as the implied annualized yield on 91-day Treasury bills issued by the central government of India reported for the week immediately preceding the launch of each contract. Where trading did not take place on the first day of a contract becoming effective, the interest rates on Treasury bills auctioned closest to the first day of trading are used in calculating the option premiums. Interest rates have been falling in India over the past few years, and the rates used in the calculation are shown in Table 3.2. For each contract, four or five different contracts for call options have been analysed with different exercise prices (K) as shown in Table 3.3. All contracts are European calls on the underlying S&P CNX Nifty index. A total of 33 contracts have been analysed spanning four years since the inception year of index options in India.
Implied volatility The implied volatility for each of the contracts was calculated using the Goal Seek procedure in Excel software, keeping everything else in the equation constant and solving for volatility given the actual option premium for that day. The closing index value for each day was taken as the current index value. Although the closing value of the index represents the average of values prevailing in the last half-hour of trading on the market, it may still not capture the entire spectrum of the intra-day volatility observed during the day. To that extent the historical volatility as well as the estimated option premiums may
22 Option Greeks on Derivative Markets in India Table 3.1 Historical volatility estimates of the underlying index Based on natural log of returns Expiry date
Estimation period
30 Aug. 2001 22 Jan. 2001 to 6 June 2001 27 Dec. 2001 24 May 2001 to 28 Sep. 2001 28 Mar. 2002 14 Aug. 2001 to 27 Dec. 2001 27 Jun. 2002 18 Dec. 2001 to 25 Apr. 2002 30 Jan. 2003 24 Jun. 2002 to 31 Oct. 2002 24 Apr. 2003 20 Sep. 2002 to 30 Jan. 2003 26 Feb. 2004 24 Jul. 2003 to 27 Nov. 2003 29 Apr. 2004 24 Sep. 2003 to 29 Jan. 2004
Based on absolute returns
annual annual daily annual annual daily σ 2 (%) σ 2 (%) σ (%) σ 2 (%) σ 2 (%) σ (%) 0.04
9.89
31.45
0.04
9.76
31.25
0.02
5.51
23.48
0.02
5.36
23.14
0.02
6.23
24.96
0.02
6.08
24.67
0.01
3.60
18.98
0.01
3.58
18.92
0.01
1.92
13.86
0.01
1.90
13.77
0.01
1.81
13.46
0.01
1.80
13.42
0.02
5.71
23.89
0.02
5.67
23.81
0.02
6.09
24.68
0.02
6.06
24.63
provide less than the most accurate theoretical value of the option. However, the extent of this error may be indirectly gauged from the difference in the historical and implied volatilities. The daily implied volatility was not calculated for the entire period of the contract as in some cases there was no trading in particular options. Delta has been found N(d1 ) while solving equations (3.1) and (3.2) for calculating theoretical values of the calls using the BSOPM. Delta is the partial derivative of the call premium with respect to the stock price, and is a useful indicator in hedging. It indicates how many options are necessary for hedging exposure in the underlying asset.
Devendra G. Kodwani 23
Table 3.2 91-day Government of India treasury bill yields
Date of T bill issue
Implicit yield at cut-off price (%)
Reported date
Used for contract expiring on
25 May 2001 21 Sep. 2001 14 Dec. 2001 12 Apr. 2001 18 Oct. 2002 17 Jan. 2003 14 Nov. 2003 16 Jan. 2004
7.7472 7.2076 6.7521 5.8842 5.7207 5.3917 4.4080 4.2446
02 Jun. 2001 29 Sep. 2001 22 Dec. 2001 20 Apr. 2002 26 Oct. 2002 25 Jan. 2003 22 Nov. 2003 24 Jan. 2004
30 Aug. 2001 27 Dec. 2001 28 Mar. 2002 27 Jun. 2002 30 Jan. 2003 24 Apr. 2003 26 Feb. 2004 29 Apr. 2004
Source: Reserve Bank of India, Government of India, www.rbi.org.in
Table 3.3 Expiry dates and exercise prices on Nifty Index options used in this study Expiry date 30 Aug. 2001 27 Dec. 2001 28 Mar. 2002 27 Jun. 2002 30 Jan. 2003 24 Apr. 2003 26 Feb. 2004 29 Apr. 2004
Exercise prices (K) 1,020 960 980 1,010 930 960 1,580 1,670
Source: National Stock www.nseindia.com
1,100 1,000 1,100 1,080 1,000 1,020 1,680 1,750
1,140 1,060 1,180 1,140 1,070 1,060 1,780 1,850
Exchange,
1,180 1,100 1,220 1,180 1,120 1,090 1,880 1,940
Mumbai,
2,000
India,
The delta and other Greeks are calculated using the equations given in Table 3.4. Theta measures the sensitivity of a call option to the time remaining to expiration. Theta for a long call would be negative as, everything else being constant, the call option loses its value as it approaches the expiration day. For a short call the passage of time adds value to the call value, and therefore theta would carry a positive sign.
24 Option Greeks on Derivative Markets in India
Table 3.4 Option Greeks: equations used for calculations on call options Measure
Equation used 2
Theta
c = −
Gamma
c =
Sσ e−.5(d1 ) − rKe−rt N(d2 ) √ 2 2π t
(3.4)
2
Vega Rho
e−.5(d1 ) √ Sσ 2π t √ 2 S te−0.5(d1 ) vega = √ 2π ρc = Kte−rt N(d2 )
(3.5) (3.6) (3.7)
Source: Strong (2002).
Discussion of results Actual and theoretical call premiums The actual value of CNX Nifty call options was taken as the closing value for a trading day reported by the NSE. The corresponding theoretical value for each day was calculated for each contract using the BSOPM. One way to compare the two is to plot the actual against the estimated option premiums, and see how closely the actual values track the estimated values. Another way is to find the average closing premium for all the trading days and compare that with the average theoretical premium calculated. Here it is found that in most cases the theoretical premiums are more than the average call premiums, but overall there is a high level of correlation between them as depicted in Figures 3.3a–d. The summary of theoretical and actual call premiums is given in Tables 3.5a–d. The pricing efficiency of the markets is fairly good, noting that the actual and theoretical values are not too far apart. Hence this study shows that there is fairly good evidence of the applicability of the BSOPM. The differences in the two values may be primarily due to the historical or implied volatilities. In this case the historical values are based on the lognormal returns on the index during the 90 trading days preceding the launch of the contracts by the Exchange. However, for estimation purposes the closing values of the index have been taken. Therefore, the intra-day volatility is not
Devendra G. Kodwani 25
(a)
(b) 250 60 50 Call premiums
Call premiums
200 150 100
40 30 20
50
10 0
23 / 25 03/ / 2 27 03/ 004 / 2 29 03/ 004 / 2 31 03/ 004 / 2 02 03/ 004 / 2 04 04/ 004 / 2 06 04/ 004 / 2 08 04/ 004 / 2 10 04/ 004 / 2 12 04/ 004 / 2 14 04/ 004 / 2 16 04/ 004 / 2 18 04/ 004 / 2 20 04/ 004 / 2 22 04/ 004 / 2 24 04/ 004 / 2 26 04/ 004 / 2 28 04/ 004 /0 20 4/ 04 20 04
12 /0 3/ 20 04 19 /0 3/ 20 04 26 /0 3/ 20 04 02 /0 4/ 20 04 09 /0 4/ 20 04 16 /0 4/ 20 04 23 /0 4/ 20 04
0
(c)
(d) 160 140
50 45
120 Call premiums
Call premiums
40 35 30 25 20
100 80 60 40
15 10
20
5
04 26
/0
4/
20
04 20
04 19
/0
4/
20
04 12
/0
4/
20
04 /0
4/
20 05
3/ /0 29
22
/0
3/
20 3/ /0 15
20
04
04
04
20 4/ /0 22
15
/0
4/
20 08
/0
4/
4/ /0 01
20
04
04
04
20
04 25
/0
3/
20
04 18
/0
3/
20
04
20 11
/0
3/
20 3/ /0 04
04
0
0
Figure 3.3 Actual and theoretical premiums on Nifty index calls (a) Expiry: 29 April 2004, K = 1, 670 (b) Expiry: 29 April 2004, K = 1, 850 (c) Expiry: 29 April 2004, K = 1, 940 (d) Expiry: 29 April 2004, K = 1, 750 • denotes actual closing premiums; denotes theoretical premiums from the BSOPM
captured in this analysis. In many cases trading did not take place if the exercise prices were too far away on either side of the prevailing values of the index. Incorporation of dividend yields in the model might further improve the convergence between theoretical and actual prices. Another reason for the differences in the theoretical values could be due to implied volatility which is discussed in the following sub-section.
Implied volatility on index options Implied volatility on the initial contracts that were introduced in India from June 2000, showed the expected smile for near-the-money contracts rather than deep out-of-money or deep in-the-money contracts.
26
Table 3.5 Call premiums, volatilities and option Greeks on nifty index call options (a) Contracts expiring 30 August 2001 Strike price Avg. value of underlying index Historical volatility used in BSOPM
27 December 2001
1,020
1,100
1,140
1,180
960
1,000
1,060
1,100
1084·41
1084·41
1084·41
1084·41
1017·43
1017·43
1017·43
1017·43
0·099
0·099
0·099
0·099
0·055
0·055
0·055
0·055
Avg. closing price of option
49·48
21·23
11·02
6·20
72·17
77·32
17·28
17·22
Avg. theoretical calculated option price
57·70
35·26
22·05
13·53
76·87
63·62
29·62
7·22
Avg. delta
0·830
0·409
0·295
0·194
0·781
0·641
0·383
0·200
Avg. implied volatility
0·027
0·038
0·057
0·077
0·146
0·078
0·068
0·109
−0·768
−0·797
−0·574
−0·406
−0·340
−0·382
−0·504
−0·380
Avg. gamma
0·004
0·004
0·003
0·002
0·003
0·003
0·005
0·004
Avg. rho
0·137
0·314
0·282
0·232
0·181
0·206
0·200
0·153
Avg. vega
0·073
0·120
0·089
0·063
0·133
0·104
0·057
Avg. theta
Avg. leverage ratio
15·60
37·53
53·54
70·78
11·25
15·20
34·77
0·030 63·35
(b) Contracts expiring 28 March 2002 Strike price Avg. value of underlying index Historical volatility used in BSOPM
27 June 2002
30 January 2003
1,100
1,180
1,220
1,010
1,140
1,080
1,070
1,120
1123·17
1123·17
1123·17
1074·96
1074·96
1074·96
1049·29
1049·29
0·062
0·062
0·062
0·036
0·036
0·036
0·019
0·019
Avg. closing price of option
61·99
14·49
4·79
55·39
4·85
9·90
22·22
3·99
Avg. theoretical calculated option price
63·95
17·36
6·85
71·01
5·64
21·59
15·47
2·53
Avg. delta
0·690
0·302
0·144
0·912
0·146
0·460
0·455
0·116
Avg. implied volatility
0·270*
0·060
0·091
0·152
0·104
0·041
0·031
0·076 −0·153
−0·539
−0·556
−0·364
−0·297
−0·213
−0·510
−0·311
Avg. gamma
0·004
0·004
0·003
0·003
0·003
0·008
0·008
0·004
Avg. rho
0·277
0·275
0·200
0·103
0·152
0·234
0·221
0·133
Avg. vega
0·146
0·071
0·040
0·148
0·038
0·092
0·086
0·025
Avg. theta
Avg. leverage ratio
19·16
50·23
86·02
18·56
78·89
62·53
61·79
99·21
27
∗ Implied volatility is based on only 17 observations in this contract. If the implied volatility figures for the last three days before the expiration day are not considered, the avg. implied volatility is 5.7%, close to the 6.2% historical volatility.
Continued
28
Table 3.5 (c)
Contracts expiring 24 April 2003 Strike price Avg. value of underlying index
26 February 2004
960
1,020
1,060
1,780
1,880
2,000
1019·67
1019·67
1019·67
1874·03
1874·03
1874·03
Historical volatility used in BSOPM
0·018
0·018
0·018
Avg. closing price of option
8·73
9·90
3·10
124·80
37·11
15·21
66·18
24·20
9·36
113·08
49·53
13·94
Avg. theoretical calculated option price
0·0571
0·0571
0·0571
Avg. delta
0·858
0·522
0·260
0·812
0·469
0·174
Avg. implied volatility
0·083
0·070
0·049
0·144
0·121
0·167
−0·218
−0·267
−0·184
−0·847
−1·018
−0·555
Avg. gamma
0·004
0·007
0·005
0·003
0·003
0·002
Avg. rho
0·059
0·214
0·211
0·298
0·428
0·307
Avg. vega
0·191
0·140
0·080
0·240
0·165
Avg. theta
Avg. leverage ratio
78·83
77·46
102·59
17·39
37·28
0·072 66·83
(d) Contracts expiring 29 April 2004 Strike price Avg. value of underlying index Historical volatility used in BSOPM
1,670
1,750
1,850
1,940
1824·17
1824·17
1824·17
1824·17
0·0609
0·0609
0·0609
0·0609
Avg. closing price of option
163·23
82·23
28·04
12·62
Avg. theoretical calculated option price
168·95
104·75
46·10
19·07
Avg. delta
0·886
0·747
0·454
0·211
Avg. implied volatility
0·127
0·080
0·078
0·096
−0·497
−0·721
−1·004
−0·641
Avg. gamma
0·001
0·002
0·003
0·002
Avg. rho
0·250
0·397
0·487
0·384
Avg. vega
0·324
0·276
0·186
Avg. theta
Avg. leverage ratio
11·63
18·48
36·38
0·107 56·34
29
30 Option Greeks on Derivative Markets in India
Implied volatility (%)
70 60 50 40
Historical volatility = 9.89%
30 20 10
11
04
/0 6
/2 0
/0 01 6 18 /20 /0 01 6 25 /20 /0 01 6 02 /20 /0 01 7 09 /20 /0 01 7 16 /20 /0 01 7 23 /20 /0 01 7 30 /20 /0 01 7 06 /20 /0 01 8 13 /20 /0 01 8 20 /20 /0 01 8 27 /20 /0 01 8/ 20 01
0
Figure 3.4 Implied volatility on index call option expiring 30 August 2001, K = 1,180
Implied volatality (%)
25 20 Historical volatility = 6.23% 15 10 5
18
/0 2 21 /20 /0 02 2 24 /20 /0 02 2 27 /20 /0 02 2 02 /20 /0 02 3 05 /20 /0 02 3 08 /20 /0 02 3 11 /20 /0 02 3 14 /20 /0 02 3 17 /20 /0 02 3 20 /20 /0 02 3 23 /20 /0 02 3 26 /20 /0 02 3/ 20 02
0
Figure 3.5 Implied volatility on index call option expiring 28 March 2002, K = 1,180
Implied volatility on two contracts, one expiring in August 2001 and the other in March 2002, the initial year of the index options trading on the NSE, show the different patterns as depicted in Figures 3.4 and 3.5. In many cases when the options were deep in or out of the money, the implied volatility solutions were not found, yielding either zero or meaningless values. Where the implied estimates were around the historical values, they were found to be greater than the historical values
Devendra G. Kodwani 31
of volatility. This could be due to the estimation taken for historical values which are calculated from the lognormal returns for 90 days prior to the day when each of the contracts was launched. This was done with a view to matching the estimation period with the life of the contracts. However, the listing and trading of options could itself be affecting the implied volatility, although evidence on this in the literature is not conclusive (Manaster and Rendleman, 1982). It is also observed that the U shape of the implied volatility shows a steep upwards movement during the last few days before expiry of contract. To increase the depths of the markets, stock exchange authorities launched at least four exercisable contracts expiring on a particular date, with exercise prices fixed in such a manner that one or two of the options may be in-the-money. An expected phenomenon in the first few years of emerging markets is evidenced by a lack of trading across the option contracts. It is found that in most of the contracts in the first month of an option’s life trading is quite low or negligible except in the case of contracts whose exercise prices were close to the ongoing values of the underlying index during the life of the contract. The implied volatility in almost all contracts starts with very high levels and goes up towards the end of their lives creating a pattern of a smile.
Delta analysis The delta statistic, a byproduct from the BSOPM, is considered to be useful information by portfolio managers who use derivatives for hedging or speculating. The mathematical definition of delta is: c = ∂C . A partial derivative of the call premium, C, with respect to the ∂S underlying asset value, S, it measures the sensitivity of call premiums to changes in the value of the underlying asset. The delta is also a crude measure of the likelihood that a particular option will be inthe-money at expiration (Strong, 2002). The BSOPM provides an estimate of delta in the form of N(d1 ). Estimates of N(d1 ) and N(d2 ) have been given for a sample of contracts analysed in this study (see Table 3.6). For deep in the money or out of the money contracts deltas appear to be higher than those where closing index values hovered around the exercise price.
32 Option Greeks on Derivative Markets in India
Table 3.6 Delta estimates and the likelihood of index call options being in-the-money on expiration
Expiry date
Strike price
Average N(d1 ) delta
Average N(d2 ) likelihood of option being in-the-money
30 Aug. 2001 30 Aug. 2002 30 Aug. 2003 30 Aug. 2004 27 Dec. 2001 27 Dec. 2002 27 Dec. 2003 27 Dec. 2004 27 Jun. 2002 27 Jun. 2003 27 Jun. 2004 27 Jun. 2005 28 Mar. 2002 28 Mar. 2003 28 Mar. 2004 28 Mar. 2005 30 Jan. 2003 30 Jan. 2004 30 Jan. 2005 30 Jan. 2006
1,020 1,100 1,140 1,180 960 1,000 1,060 1,100 1,010 1,080 1,140 1,180 1,220 1,180 1,100 980 930 1,000 1,070 1,120
0.802705 0.442265 0.295299 0.193606 0.781375 0.658444 0.382925 0.200243 0.91157 0.459573 0.146454 0.060909 0.159225 0.301794 0.690 0.947733 0.962305 0.833316 0.454792 0.116391
0.77726 0.409177 0.266119 0.170033 0.765248 0.640592 0.36632 0.187911 0.9047 0.444514 0.136978 0.055594 0.143573 0.279913 0.667 0.93912 0.959284 0.82668 0.444249 0.110238
Conclusion We have provided an empirical analysis of option pricing, implied volatility and delta of index options introduced recently into Indian financial markets. The results indicate that markets have been showing remarkable efficiency in the pricing of index options, although this conclusion is based on only a limited sample and time period. Considering that derivatives have become very popular with players in Indian financial markets, it is not surprising to find that premiums on options with exercise prices closer to that of the underlying index, are better priced. It is also found while comparing the premiums estimated from the BSOPM with the daily closing prices (on the days when there was trading in the particular contacts) that the
Devendra G. Kodwani 33
differences are not very large. The implication is that for Indian markets (the NSE) the historical volatility based on the preceding 90 days is reflected in the option premiums to a noticeable extent. Volatility smiles are quite distinctly exhibited in some cases, but in most cases implied volatility estimates do not show smooth patterns. One limitation of these findings is that in calculating the implied volatility, closing option premiums are used. Although the closing values are averages of the trades in the last hour or so, they may not capture the entire intra-day price movements, thereby reducing the accuracy of implied volatility estimates. Most values of delta are closer to one, indicating a high level of sensitivity of index options to the variance of the underlying index. The results on other Greeks are not analysed fully here, and are being examined in the context of a wider analysis being carried out by the author, but they show expected values in most cases.
References Manaster, S. and Rendleman, R.J. (1982) ‘Option Prices as Predictors of Equilibrium Stock Prices’, Journal of Finance, vol. 37, pp. 1043–57. Shenbagaraman, P. (2003) ‘Do Futures and Options Trading Increase Stock Market Volatility?’, Working Paper no. 22, National Stock Exchange, Mumbai. Srivastava, S., Yadav, S. and Jain, P.K. (2002) ‘Early Efficiency Signals from the Stock Index Futures Market in India’, Paper presented at the 15th Australasian Finance and Banking Conference, 16–18 Dec. 2002. Srivastava, S. (2003) ‘Informational Content of Trading Volume and Open Interest–An Empirical Study of the Stock Options Market in India’, Working Paper no. 29, National Stock Exchange, Mumbai. Strong, R. (2002) Derivatives: an Introduction (Singapore: Thomson Western).
This page intentionally left blank
4 Global Asset Allocation: Risk and Return Trade-off on Emerging Stockmarkets Mohamed Derrabi and Michel Leseure∗
Introduction The allocational efficiency of financial markets is one of the important conditions for economic growth. The way financial resources are allocated determines the cost, the risk, the return and thus the added value created by investments. These are particularly important when considering emerging economies because of problems and barriers to capital accumulation and its allocation to the most promising activities. In developing countries, bank debt is the most preferred and sometimes the unique formal source for financing. The emergence of stock exchange markets has been restricted for several reasons: small size of the businesses, a lack of sufficient savings, and difficulties associated with the accumulation of capital due mainly to the lack of sufficient country-funds. As a result, the stock exchange markets have for a long time been neglected. In the early 1990s, emerging stockmarkets regained access to both domestic and foreign capital. Capital flows to emerging markets increased dramatically thanks to the emergence of other types of capital flows: portfolio flows (fixed income and equity) and foreign direct investment. This could not have happened without these countries
∗
We have benefited from discussions with Michel Levasseur, Robert Cobbaut and Eric Ghysels. We appreciate the comments of Stijn Claessene, Geert Bekeart, Bruno Biais, Alain François Heude, Eric de Bodt, Rolland Gillet and George Gallais Hamono. All remaining errors are ours. (Corresponding author e-mail:
[email protected])
35
36 Risk and Return on Emerging Stockmarkets
embarking on deep structural changes and reforms. Indeed, to sustain their economic growth in particular, and enhance the efficiency of their financial systems in general, the developing countries have been driven to track the international movement of reforming capital markets. The reforms of the financial and banking systems were crucial at that stage. These changes were beneficial and their impact on the stock exchange market was important. Also, the flow of an important portion of international investment to emerging markets contributed to the dynamism of those markets. Since the second half of the last decade, stock exchange markets in developing countries has been driven ahead thanks to foreign investments in these markets. Indeed, from 1989 to 1999 the total portfolio of the emerging markets (bonds, certificates of deposits, commercial paper and stocks) grew considerably by 12 times to reach $85.8 billion. Investment in stocks was predominant in the portfolio, representing 25 per cent of the total. This growth was also aided by the suppression of barriers that had been hampering domestic investment, as well as the increases in foreign investment (Claessens and Rhee, 1994). Emerging markets began to liberalize their economies by reducing restrictions concerning foreign proprieties and the transfer of hard currencies,1 among other factors. The empirical results show that the emerging markets ensured their integration into the world market (Bekaeret, 1995) although this integration was not completely achieved due to several reasons. These dramatic changes raise a number of intriguing questions: • What are the advantages from investing in these markets? • To what extent are emerging markets integrated in the global market? • What are the barriers to investing in emerging markets? • What are the risks associated with investing in emerging markets? This chapter attempts to answer the above questions, and to compare findings to others results obtained for developed markets. The next section discusses the heterogeneity of emerging markets, and suggests a typological evolution of these markets. Advantages from investing in emerging markets are then discussed, followed by a study of measures of integration, systematic risks and specific risks in emerging markets as well as in developed markets. A final section presents conclusions.
Mohamed Derrabi and Michel Leseure 37
Emerging markets: a heterogeneous group Emerging markets are rarely accurately defined as they encompass an array of markets. The IFC (International Finance Corporation) suggests some attributes to qualify a market as emerging: • The market must be located in a developing country. • A high degree of contribution of the stock exchange market in the financial system that can be measured by the market capitalization ratio (value of all listed financial assets in the exchange market divided by GDP). • A high degree of dynamism of the market that can be measured by its liquidity. • The market must be attractive for domestic investors as well as for foreign ones. These qualification guidelines correspond clearly to the typical problems faced by most of these stock exchange markets: small stockmarket size, fragmentation and segmentation of the market, aggressive growth of performance indexes, concentration of market capitalization in the hands of a few dominant enterprises, weaknesses in functioning and regulation, liquidity problems, and a general unwillingness of the public to invest in these markets. Therefore, these markets cannot be presented as a group of homogeneous markets. important disparities among them exist at the microstructure level (markets organization) as well as the activity volume level (see Bekaert and Harvey, 2002). Some stock exchange markets are older, others are in an embryonic stage, and yet others are comparable to markets in developed countries. An analysis of the evolution of some emerging markets reveals four distinct stages: the embryonic phase, a phase of low activity, an active phase, and a maturity phase (see Derrabi, 2000). 1 Embryonic phase. This phase is characterized by a ‘primitive’ quotation system (auction organized once a day with 2, 3, 4, 5 days of quotations a week), market operations not yet processed automatically, and a low volume of transactions. Other characteristics are irregularities in reporting quoted values, rudimentary regulations,
38 Risk and Return on Emerging Stockmarkets
and neglect of the exchange market both by investors and financial authorities. 2 Low activity phase. Numerous countries have liberalized their economies as suggested by the IMF (International Monetary Fund). The first step typically is towards alleviating budgetary deficits and remedying the problem of overindebtness. One of the suggestions of the IMF towards liberalization is a privatization programme of government-owned companies. This privatization leads to a new dynamism in the market and constitutes the starting point for the second phase in the evolution of emerging markets. Indeed, privatization allows a number of economic operators to discover the stock exchange markets as a source of financing and investing. Privatization is usually followed by an increase in market capitalization, new quotations in the stock exchange market, the development of performance indices, the emergence of a great potential for speculation and the realization of capital gains, and a great volatility in stocks’ prices. In order to sustain this uptake in activity, authorities start to follow developments in the market and initiate regulatory reforms of the organization of the stock exchange market. 3 Active phase. It takes continuous reforms to improve efficiency of markets and the disclosure of information related to stock exchange market activities. These reforms ultimately lead to the active phase, typified by the transition to computer-processing of quotations, and in particular continuous quoting. This phase is also characterized by its openness to foreign capital, and by a growing interest of foreign investors in quoted financial assets. An increase of the liquidity of the market and an increase in the number of intervening parties in the market can be observed. 4 Maturity phase. This is characterized by the end of the stream of regulatory reforms. The stock exchange market has become similar to markets in developed countries both in terms of its operating system and its activities. At this stage, the stock exchange market has gained the confidence of foreign investors seeking an international portfolio. The market begins to be integrated into the international financial market. However, despite being at a mature stage, it is worth stressing that emerging markets are very sensitive to random shocks, in contrast to markets in developed countries. The Asian crisis of 1997 is a demonstration of this sensitivity.
Mohamed Derrabi and Michel Leseure 39
Investing in emerging markets: the potential for diversification Investing in emerging markets encompasses several activities: financial assets can be bought directly from the stockmarket by national or foreign investors; investment may be made via ‘country fund’ financial assets possessed by some deposit institutions (American Depository Receipts (ADRs) and Global Depository Receipts); or financial assets may be sold directly abroad. Emerging markets are characterized by the predominance of individuals holding financial assets, in contrast to developed markets in which market investment is highly institutionalized. Direct investment in shares is the favoured investment in emerging markets. For instance, of a total investment of $101.1 billions in emerging markets between 1989 and 1993, more than half of it was invested in shares. The decrease in interest rates at the beginning of the decade and the suppression of some barriers were behind this trend. A growing trend of investing in emerging markets has resulted in an increase in the volume of activities of these markets. In 1996, market capitalization of shares in emerging markets reached 13 per cent of the total market capitalization worldwide, against 2.5 per cent 10 years earlier. According to estimation by the Federation Internationale des Bourses de Valeurs (International Federation of Stock Exchanges) (FIBV), this proportion will reach 20 per cent in the year 2000. The advantages of investing in emerging markets for an investor depend on the return/risk ratio. In order to estimate this ratio it is advisable to take into consideration the specific behaviour of emerging markets, the risk characteristics of investing in these markets, and the organization of these markets as well as the particularities of their regulations. The importance of the potential returns was behind the growing trend towards investing in emerging markets. Indeed, the best records were registered during 1996 in terms of the evolution of the market indices (Venezuela 23 per cent, Hungary 120.4 per cent and Turkey 47 per cent against 19.1 per cent for the New York Stock Exchange (NYSE). In this section we discuss the advantages of investing in emerging markets. The first advantage is the potential returns as mentioned above, while a second advantage is a reduction of risk by
40 Risk and Return on Emerging Stockmarkets
diversification. According to Divecha, Drach and Stefek (1992), Harvey (1995a) and Wilcox (1992), investors can take advantage of diversification through emerging markets because of the low correlation between returns in emerging markets and those of developed markets. Bekaert and Urias (1996) measure the diversification benefits from emerging equity markets using data on closed-end funds (country and regional funds) and American Depository Receipts (ADRs). They find that investors give up a substantial part of the diversification benefits of investing in foreign markets when they do so by holding closed-end funds. De Roon, Nijman and Werker (2001) and Li, Sarker and Wang (2003) take the transactions costs that investors in emerging markets face directly into account when measuring diversification benefits. They find that the diversification benefits of investing in emerging markets are eliminated when transactions costs and, in particular, short-sale constraints are considered. In this section we examine the existence of diversification benefits to investing in emerging markets. The research reported here measures the correlation between returns of emerging markets and developed markets, and analyses the return/risk ratio using the Markowitz algorithm.
Database Data on the returns of the exchange markets of our sample originated from the DATASTREAM database and the Emerging Markets Database of the IFC. These two databases include indices calculated by Morgan Stanley & Co (MSCI). The indices are weekly indices that take into account the reinvestment of dividends (‘index-returns’). The analysis period was chosen to minimize the bias resulting from structural changes, and for the majority of the markets the period begins at 1 January 1997, a date that coincides with the achievement of financial reforms for these markets. When data are missing from a market, the period begins when the data are available. The end period is 15 November 1999. Returns are calculated by the difference of logarithms of weekly indices of each market. If Rj,t denotes the market return j in period t and MSj,t represents the weekly index of the exchange market j in period t, then the weekly returns of that stock exchange market are: Rj,t = ln(MSj,t ) − ln(MSj,t−1 )
(4.1)
Mohamed Derrabi and Michel Leseure 41
MSCI indices correspond to at least 60 per cent of market capitalization of the market considered. They are constructed following the concept of Laspeyers2 and consider the most liquid assets. The choice of markets for the study was based on the following criteria: • To be representative of all regions of the world (Europe – east and west; Asia, Africa, the Middle East, North America and Latin America). • The market must be attractive to foreign investors. • Availability of the data on MSCI indices during the study period. The goals, as defined above, are (1) to study the correlation between returns in the emerging markets two by two, and (2) to study the returns of emerging markets and developed markets. A low correlation means a great diversification3 opportunity by investing in both countries.
Descriptive analysis of returns Table 4.1 presents the stock exchange markets included in the sample, the period of analysis used for each market, the arithmetic mean of weekly returns, their standard deviation, and the coefficients of skewness and kurtosis, which are centred around zero for normal distributions. The analysis of the results shows that the distribution of returns from the emerging markets differs from a normal distribution,4 and the degree of deviation from a normal distribution is significant. Standard deviation measures the volatility of markets, which is important in emerging markets. It fluctuates from 0.018 in Morocco5 to 0.073 in Turkey, against 0.017 in the United States and 0.04 in Hong Kong. Thus, returns are highly volatile in emerging markets. The work of Harvey (1995a), which is based on monthly returns, confirms this finding; for example, Brazil registered a total return of 64.3 per cent in 1996 and a return of −21 per cent during the sixth month of the year 1995. Turkey registered a return of 47.7 per cent in 1995 (30 June) against a return of − 50 per cent in 1994. For these reasons, the mean of returns in emerging markets and developed markets are usually similar, and sometimes higher in emerging markets.
42
Table 4.1 Analysis of the stock exchange markets of the sample Markets Argentina Brazil Israel Colombia Philippines Taiwan South Africa Thailand Turkey Korea Malaysia Mexico Hungary Indonesia Morocco
No. points
Period
254 254 254 254 254 254 254 254 254 254 254 254 152 254 100
Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 99–Nov. 01
Mean (%) 0·187 0·458 0·025 0·169 −0·048 0·261 0·251 −0·549 0·416 −0·345 −0·168 0·012 0·691 −0·107 0·633
Std. dev (%)
Variance (%)
Skewness
Kurtosis
4·680 6·387 3·230 3·499 3·892 4·019 3·035 5·148 7·325 3·645 4·082 5·712 4·751 4·642 1·816
0·219 0·408 0·104 0·122 0·152 0·162 0·092 0·265 0·537 0·133 0·167 0·326 0·225 0·215 0·033
−0·52881211 −0·50430155 −0·03536622 0·79400569 −0·67818958 −0·65395427 0·30612901 −0·26080593 −0·46059281 −0·64399862 −1·13895297 −1·36747992 0·57386172 −0·44595674 −0·26621351
1·90580261 2·15959811 0·5456658 3·37683845 3·42423916 3·25173855 3·06129497 2·67264247 2·25312352 1·83104193 4·36343718 8·31552219 4·30593443 8·27952004 2·20041847
Czech Republic Italy Germany Hong Kong Australia Netherlands Singapore Spain UK USA Japan France Belgium Canada Denmark
152 254 254 254 255 254 254 254 254 254 254 254 254 254 254
Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01 Jan. 97–Nov. 01
−0·218 0·267 0·304 0·203 0·164 0·389 0·088 0·355 0·261 0·326 0·011 0·194 0·241 0·228 0·337
3·117 3·500 2·118 4·095 2·264 1·921 2·621 2·675 1·786 1·733 3·097 2·212 1·955 2·018 2·053
0·097 0·123 0·045 0·168 0·051 0·037 0·069 0·072 0·032 0·030 0·096 0·049 0·038 0·041 0·042
−0·84593636 1·83031893 0·04858652 0·34060752 −0·01938407 0·69298524 −0·86987257 3·70581126 −0·24750316 1·89263427 0·13029049 0·58766961 −0·41310361 2·92291641 −0·12472761 0·31088722 −0·13879475 −0·17302826 −0·22217943 2·75909758 −0·10666808 2·51603317 −0·07012326 0·07986155 0·11786432 0·16455162 −0·4896822 2·20840287 0·36513278 0·93583852
43
44 Risk and Return on Emerging Stockmarkets
Diversification analysis Table 4.2 shows the coefficients of correlation between markets. Generally, they are very weak; in almost all cases less than 0.1 and zero in some cases (Korea, Colombia) and negative in others (Mexico, Argentina). Also, the correlation between emerging markets and developed markets is low, typically less than 0.1 per cent except for some cases. The highest correlations can be observed between Indonesia and Hong Kong, Malaysia and Singapore and for Indonesia and Singapore. The low correlation between stock exchange markets can be attributed to geographic factors, time zones and opening hours, and the results confirm these hypotheses (that is, high correlation between markets located in the same geographical area: Argentina/Brazil, Singapore/Indonesia, and Hong Kong/Indonesia). Examples of low correlation are found between markets located in distant geographical areas (Korea/Colombia, Malaysia/Argentina, Taiwan /Colombia). These low correlation coefficients between returns in stock exchange markets are an indicator of the potential diversification benefits in emerging markets. Mullin (1993) affirms that the low correlation can be attributed to the low frequency of transactions in emerging markets, while Harvey (1995a) rejects Mullin’s hypothesis. Our findings confirm Harvey’s results, as the research reported here excludes low-frequency transactions (work done weekly and selection of most liquid shares). Furthermore, Harvey (1995a, 2000) shows that even if emerging markets are highly volatile, a well-diversified portfolio of financial assets weakly correlated can reduce the overall portfolio volatility. Other works have reached similar results: Divecha, Drach and Stefek (1992), Harvey (1995a, 1995b) and Wilcox (1992). Figure 4.1 shows the efficient frontier based on the weekly indices of emerging markets and developed markets. This efficient frontier is constructed using the Markowitz algorithm adjusted for the main constraint in emerging markets as suggested by De Roon, Nijman and Werker (2001) and Li, Sarker and Wang (2003) that is, the short-sale constraint. The combination of financial assets from two types of markets in a portfolio gives more interesting returns/risk ratio than a portfolio of stocks of one type of market. In an attempt to assess the advantage of diversification from investing in emerging markets, one can construct a second efficient
Table 4.2 Coefficients of correlation between stockmarkets Argentina Brazil Colombia Korea Hungary Indonesia Israel Morocco Malaysia Mexico Philippines Czech Republic S Africa Taiwan Thailand Turkey 1 0·443 0·058 0·028 0·177 0·01 0·243 −0·132 0·008 −0·037 0·15 −0·093 0·212 0·042 0·051 −0·032 0·051 0·014 0·155 −0·011 0·115 0·162 0·131 −0·01 0·004 0·089 0·144 0·082 0·082 −0·028
1 0·022 −0·027 0·213 −0·026 0·096 −0·1 −0·026 −0·016 0·097 0·029 0·02 −0·027 0·085 −0·014 −0·04 −0·012 0·041 0·033 0·045 0·114 0·056 −0·07 −0·026 0·008 0·003 −0·01 −0·025 −0·019
1 0 0·146 0·035 0·111 0·142 −0·01 0·061 0·106 0·236 0·036 0·006 0·117 −0·029 0·021 0·06 0·029 −0·032 0·091 0·045 0·006 −0·008 0·061 0·067 0·064 −0·055 0·081 0·005
1 0·027 1 0·161 0·22 0·006 0·207 0·094 −0·024 0·237 0·134 0·156 0·172 0·019 0·232 0·217 0·218 0·127 0·329 0·12 0·032 0·051 0·112 0·048 0·229 0·123 0·112 0·157 0·251 0·145 0·18 0·168 0·154 0·094 0·166 0·1 0·052 0·093 0·109 0·23 0·1 0·077 0·191 0·1 0·051 0·152 0·229 0·218 0·153 0·143 0·091 0·11 0·18
1 0·052 −0·067 0·591 0·194 −0·008 0·089 0·054 −0·012 0·142 0·133 0·122 0·21 0·121 0·194 0·042 0·144 0·168 0·408 0·037 0·114 0·213 0·425 0·141 0·257
1 0·081 0·073 0·018 0·158 0·054 0·115 0·046 0·089 −0·006 0·073 −0·14 0·069 0·045 0·083 0·106 0·219 0·08 0·063 0·094 0·13 0·054 0·154 0·083
1 −0·067 −0·068 −0·124 0·183 0·067 0·163 0·012 0·037 −0·117 −0·14 0·034 −0·228 0·053 −0·013 0·12 −0·152 0·082 −0·093 −0·075 −0·064 0·06 −0·144
1 0·207 0·197 0·157 0·149 0·072 0·126 0·166 0·296 0·296 0·278 0·205 0·2 0·2 0·243 0·551 0·119 0·236 0·296 0·661 0·232 0·231
1 0·081 −0·062 −0·042 0·037 0·163 0·075 0·14 0·28 0·153 0·351 −0·031 0·287 0·198 0·328 0·161 0·063 0·189 0·244 0·23 0·398
1 0·24 0·144 0·137 0·129 0·16 0·049 0·085 0·056 0·028 −0·036 0·003 0·046 0·142 0·1 0·021 0·062 0·089 −0·028 −0·003
1 0·109 0·061 0·222 0·139 0·114 0·039 0·161 −0·037 0·284 0·114 0·13 −0·03 0·069 −0·033 0·206 0·054 0·0191 −0·052
1 0·137 1 0·129 0·166 0·16 0·016 0·155 0·08 0·003 0·074 0·128 0·118 0·039 0·038 0·171 0·051 0·025 0·091 0·085 0·202 0·094 0·117 −0·128 0·018 0·073 0·037 0·177 0·132 0·16 0·106 0·09 0·065 0·026 −0·019
1 0·16 0·02 0·12 0·06 0·1 −0·04 0·06 0·04 0·14 −0·06 0·02 0·11 0·06 0·08 0·13
1 0·11 0·162 −0·009 0·098 0·004 0·072 0·083 0·175 −0·127 0·048 0·163 0·168 0·078 0·111
45
Argentina Brazil Colombia Korea Hungary Indonesia Israel Morocco Malaysia Mexico Philippines Czech Rep. S Africa Taiwan Thailand Turkey Germany Australia Belgium Canada Denmark Spain France Hong Kong Italy Japan Netherlands Singapore UK USA
46 Risk and Return on Emerging Stockmarkets
Efficient frontier
0.80% 0.60%
Return
0.40% 0.20% 0.00% 1.00% –0.20%
1.20%
1.40%
1.60%
1.80%
2.00%
2.20%
Risk –0.40% –0.60% Emerging market & developed markets
Developed markets
Figure 4.1 The efficient frontier based on the weekly indices of emerging markets and developed markets
frontier for developed markets only. The international frontier largely dominates the developed market frontier, and so the portfolio including financial assets from both markets (emerging and developed) is more efficient. Therefore, investing in emerging markets encompasses at least two advantages: • The advantage of high returns in emerging markets during positive trends of stock prices. • The advantage of risk diversification by spreading investments over markets only weakly correlated. On the otherhand, and contrary to these advantages and in addition to the high volatility risk mentioned above, investing in emerging markets is risky because of the nature of their organization and because of the high political risk. These risks are real obstacles to investing in such markets.
Emerging markets investment: barriers and risks Barriers to investment in emerging markets Emerging markets in developing countries present numerous obstacles6 to investment, either by nationals as well as foreigners.
Mohamed Derrabi and Michel Leseure 47
These obstacles can be direct and stem from economic market regulations, or indirect such as the lack of economic and financial expansion of the stock exchange market. Monetary and market politics, investment regulations, the lack of reliable infrastructure for the development of a market, a lack of appropriate strategies of informing investors, and the lack of personnel specialized in portfolio management, are factors that delete potential investors from investing in these markets. The economies of emerging markets are depending on the overall economic situation, since these are normally export-oriented economies, and this dependence, combined to a low national income, has a negative impact on the process of capital accumulation and on the capacity of saving. It should be mentioned that other factors such as a huge inequality of revenues, the inflation rate, the different procedures used to share market information, the lack of papers and the lack of measures protecting clients, are handicaps toward the accumulation of saving that would normally be invested in stock exchange markets. The unfavourable character of investing in emerging markets is amplified by the unwillingness of economic actors to invest in these markets. In the majority of developing countries, businesses are family-owned, which is related to a lack of information about stock exchange markets and by the fact that alternative types of investment exist and are better-known. Furthermore, other barriers prevent the integration of emerging market into the global economy; for example foreign investors not being allowed to intervene in some sectors or to acquire proprieties. Also, restrictions on transfers of funds can affect investment in emerging markets. Similarly, taxes, dividend payout regulation, and the level of capital gains are other factors that can enhance or discourage investment in these markets (Demirgüç-Kunt and Huizingal, 1992). The risks of liquidity and political instabilities, specific to emerging markets, can also discourage foreign investors (Bekaert and Harvey (1995)). Chuchan (1992) stressed the problem of information diffusion as a dissuasive factor for foreigners to invest in emerging markets, as these markets are confronted by problem of asymmetric information on the economic and financial situation of local firms that is introduced in the country’s stock market.
48 Risk and Return on Emerging Stockmarkets
Emerging markets investment: risk analysis Harvey (1995a) studied five risk factors related to investing in emerging markets: systematic risk related to the global market; risk associated to the instability of oil prices; risk associated to the growth of industrial production; exchange rate risks; and risk related to inflation. These five factors are considered even though a majority of emerging markets are not significantly exposed to all these factors. For example, of over 20 emerging markets studied, only one presented a beta coefficient (covariance with global portfolio) higher than 1, indicating a low integration of the emerging market in the global market (Claessens 1995). Several studies have been devoted to the analysis of the integration of emerging markets, and Bekaert (1995), Buckberg (1996), Bekaert and Harvey (2002), Korajczyk (1996) among others, have pointed out the difficulty of measuring the degree of integration of emerging markets. Claessens and Rhee (1994) presented some measurement methods. The first consisted of modelling barriers to investment, analysing the impact on the equilibrium model of financial assets and then testing the model. The other method, widely used, consists of interpreting the relation between one market and the global portfolio using an equilibrium model (see Bekaert, 1995, and Harvey, 1995). Bekaert (1995) pointed out that the betas of emerging markets have increased considerably, indicating a high integration of these markets. The objective of this section is to measure the integration level of emerging markets into the global market, to measure the risk associated with an investment in these markets, and to benchmark results against those obtained with developed markets. Using the Sharpe market model, the risks in a portfolio are divided as follows: • systematic risk, measured by the beta coefficient;7 • market risk;8 and • specific risk, represented by the variance of the portfolio return over the market return. If a general index representing the global market is available, one can use the market model at an international level to measure the integration and the risks of emerging markets. The global index MSCI (MSW) can be used for this purpose. Although the emerging markets
Mohamed Derrabi and Michel Leseure 49
are less represented in this index, it is commonly used in these markets as a benchmark. Hence, we propose to use it in this study as the portfolio of the global market. The index consists of a large number of financial assets of developed markets and emerging markets with different weights, which are determined in proportion to their size. The model is: Rj,t = αj + βj MSWt + εj,t
(4.2)
where Rj ,t is the weekly return of the market j; MSW is the return of the global market, measured as the difference of logarithms in respect of the MSCI; and βj indicates the relationship between fluctuation of market j and the global market. It measures the sensitivity of the market j to the variations of global market. This model is used for two reasons. First, to estimate the integration degree of each market in the global market, and second to estimate the systematic risk over these markets. The higher the β coefficient, the higher the integration of the market in the global market, and hence the more systematic risks are present. αj + εj,t is the specific risk, reflecting the effect of specific factors of the market j on Rj,t ; the variance of the error εj,t is used to measure the specific risk of each market. The model indicates that the fluctuations of returns of market j are divided into fluctuations that have an impact on the global market, and the fluctuation of the market subject to the study. Indeed, we have: σ 2 (Rj,t ) = β 2 σ 2 (MSW) + σ 2 (εj,t )
(4.3)
The first part of this equation is the systematic risk of the market j. The second part is the non-systematic risk. To assess the importance of each, we calculate the coefficient of determination of the regression model. This coefficient measures the percentage of the variations of returns of market j explained by factors that affect the market as a whole. The coefficient of determination is higher (lower) if εj,t are less important than the mean, i.e. specific risk of the market j is less (more) important in comparison to other factors affecting the global market as a whole. Table 4.3 presents the results obtained from the application of the model of the global market on the emerging and industrialized
50
Table 4.3 Application of the global market model to emerging and industrialized markets Beta
R2
Systematic risk
Specific risk
0·585∗∗∗∗ 0·498∗∗∗∗ 0·502∗∗∗∗ 0·615∗∗∗∗ 0·346∗∗∗ 0·562∗∗∗∗ 0·596∗∗∗∗ 0·525∗∗∗∗ 0·290∗∗∗ 0·695∗∗∗ 0·690∗∗∗∗ 0·524∗∗∗∗ 0·592∗∗∗∗ 0·765∗∗∗∗
0·342 0·248 0·252 0·378 0·12 0·316 0·356 0·276 0·084 0·483 0·476 0·272 0·351 0·586
0·785 0·570 0·579 0·870 0·275 0·726 0·817 0·634 0·193 1·111 1·095 0·631 0·806 1·346
2·935 3·949 2·87 2·54 3·713 4·874 4·874 12·18 11·27 4·973 1·942 5·003 2·08 1·239
Emerging Argentina 0·059∗ Brazil −0·011 Colombia 0·043 Korea 0·214∗∗ Hungary 0·204∗∗ Indonesia 0·313∗∗∗ Israel 0·15∗∗ Morocco −0·178∗ Malaysia 0·422∗∗∗∗ Mexico 0·386∗∗∗∗ Philippines 0·044 Czech Republic 0·028 South Africa 0·075∗ Taiwan 0·041 Thailand 0·135∗∗ Turkey 0·143∗∗
0·004 0·001 0·002 0·046 0·042 0·098 0·023 0·032 0·178 0·149 0·002 0·001 0·006 0·002 0·018 0·021
0·008 0·0002 0·0042 0·1053 0·0957 0·2254 0·0517 0·0729 0·409 0·342 0·0044 0·0018 0·0129 0·0038 0·0419 0·047
21·91 40·95 12·66 27·64 21·51 19·26 10·22 3·193 11·5 13·42 15·04 9·691 9·656 16·16 25·65 52·64
Markets Developed Germany Australia Belgium Canada Denmark Spain France Hong Kong Italy Japan Netherlands Singapore UK USA
∗ t -statistic significant at 90%; ∗∗ t -statistic significant at 95%; ∗ ∗ ∗ t -statistic significant at 97.5%; ∗ ∗ ∗∗ t -statistic significant at 99%; The systematic risk is calculated by b2∗ (sm )2 , where (sm )2 is the world market variance (MSW). The specific risk is calculated by (se )2 .
Mohamed Derrabi and Michel Leseure 51
markets. These results display the beta coefficient, the coefficient of determination, systematic risk and specific risk for each market. The coefficient of determination, which measures the variations of returns of markets explained by the factors of the market, is low for the emerging markets. It shows the unfitness of the model to characterize the return in emerging markets (see also Harvey, 1995b), the weak integration of the emerging markets in the global market, and the existence of more advantages of diversification in the emerging markets. This coefficient is higher for industrialized markets, fluctuating between 0.12 per cent for Denmark and 0.586 per cent for the USA. In contrast, for the emerging markets the highest figure is for Malaysia at 0.178 per cent and the lowest is 0.001 for the Czech Republic. The beta coefficient, which measures the integration of markets in the global market, is more significant for the industrialized markets and less significant for emerging markets. The beta coefficient fluctuates between 0.290 in Italy and 0.765 in the USA, with a 97 per cent degree of significance. It fluctuates between −0.78 in Morocco and 0.422 in Malaysia for emerging markets, with a lower degree of significance. From this comparison, it is clear that the industrialized markets are more integrated in the global market than emerging markets. It follows that systematic risk is more important in the industrialized markets. However, specific risk is more important for the emerging markets; for the industrialized markets it fluctuates between 0.0001239 in the USA and 0.001218 in Hong Kong, and between 0.0003193 in Morocco and 0.005264 in Turkey. In all markets, the specific risk is greater than the systematic risk except for the USA. From the above, four conclusions can be drawn concerning the stock exchange market for emerging as well as industrialized markets: • Emerging markets are only weakly integrated into the global market; the existence of direct and indirect barriers to investment is the principal reason for this. • Systematic risk is more important in industrialized markets due to the higher integration of these markets in the global market. • Specific risk is more important in emerging markets. This is due mainly to the organization of these markets, to the specificity of
52 Risk and Return on Emerging Stockmarkets
their economical, political and financial environments, and to the significance of volatility and the lack of liquidity in these markets. • There are several advantages to diversifying investments through emerging markets, because specific risk can be diversified.
Conclusion The later 1980s and the beginning of the 1990s witnessed changes in the international financial landscape. More funds were allocated to investment in stock exchange markets, and more investments were channelled to emerging markets. These latter had been neglected in international portfolios for years. Investing in emerging markets encompasses significant benefits for diversification. However, the existence of barriers to investment and the importance of specific risks in these markets limit investment in these markets and, hence, limit into integration of these markets into the global market. Recently, authorities of a large number of emerging markets have understood that the abolishment of political and economic barriers, barriers to enter or to exit exchange markets, tax barriers, and the repatriation of hard currency, are beneficial to their economies. The creation of financial environments propitious to investment are sine qua non conditions for a deeper integration of financial markets into the global economy. The growth of investment in emerging markets then depends on their operational and informational efficiency. In this chapter we have presented the characteristics of emerging markets and the typological evolution of these markets. Numerous stock exchange markets satisfy these criteria. However, these markets are heterogeneous, and important disparities exist among these markets in their microstructures, particularly the systems of transformation of orders into transactions and in the volumes traded. The low correlation between returns from the two types of markets can also contribute to the reduction of volatility by holding well-diversified portfolios, including financial assets quoted in emerging markets. Comparative studies of efficient frontiers of emerging markets and industrialized markets confirm this notion. In addition to advantages from diversification, emerging markets offer the benefit of high potential returns. However, everything else being equal, the emerging markets are riskier. A comparison between
Mohamed Derrabi and Michel Leseure 53
emerging markets and industrialized markets over the degree of integration, specific risks and systematic risk, leads us to conclude that emerging markets are only weakly integrated into the global market, with a greater specific risk. On the other hand, industrialized markets present a greater systematic risk.
Notes 1. Less restriction on foreign investment in emerging markets. In some markets (Brazil, Colombia, Morocco, India, Korea and Taiwan) barriers don’t exist any more. 2. The Laspeyers index compares, over time, values of fixed portfolios that correspond to qualities retained in the calculation of a base year. In this index, prices are weighted by quantities and evaluated for predefined dates. It is a basis of comparison for all periods for which the index is calculated. 3. An advantage of diversification is the possibility of increasing the return adjusted to the risk for investors by spreading the investment over large financial assets that have a low or negative correlation. 4. We will discuss a detailed normality hypothesis later in the chapter. 5. The feebleness of volatility on the Moroccan markets can be explained by the limitation of daily price variations to 3 per cent. 6. See Bekaert (1995) and the report of the IFC for an analysis of barriers to investment in emerging markets. 7. Measures the sensitivity of a portfolio’s returns to the variation returns of the market. It is calculated using the correlation between the return of the portfolio and the return of the market. 8. Market risk can be estimated using the variance of market returns.
References Agrawal, A. and Tandon, K. (1994) ‘Anomalies or Illusions? Evidence from Stock Markets in Eighteen Countries’, Journal of International Money and Finance, vol. 13, pp. 83–106. Asprem, J. and Mads, M. (1989) ‘Stock Prices, Asset Portfolios and Macroeconomic Variables in Ten European Countries’, Journal of Banking and Finance, vol. 13, pp. 589–612. Balaban, E. (1995a) ‘Day of the Week Effects: New Evidence from an Emerging Stock Market’, Applied Economics Letters, vol. 2, pp. 139–43. Balaban, E. (1995b) ‘A Preliminary Note on the Relationship among International Stock Market’ unpublished paper, The Central Bank of the Republic of Turkey: Research Department. Bekaert, G. (1995) ‘Market Integration and Investment Barriers in Emerging Equity Markets’, The World Bank Economic Review, vol. 9, pp. 75–107. Bekaert, G. and Harvey, C. (1995) ‘Time-Varying World Market Integration’, Journal of Finance, vol. 50, pp. 403–44.
54 Risk and Return on Emerging Stockmarkets
Bekaert, G. and Harvey, C. (2000) ‘Foreign Speculators and Emerging Equity Markets’, Journal of Finance, vol. 55(2), pp. 565–613. Bekaert, G. and Harvey, C. (2002) ‘Research in Emerging Markets Finance: Looking to the Future’, Emerging Markets Review, vol. 3(4), pp. 129–48. Bekaert, G. and Urias, M.S. (1996) ‘Diversification, Integration and Emerging Market Close-End Funds’, National Bureau of Economic Research (NBER) Working paper, 4990. Buckberg, E. (1996) ‘Institutional Investors and Asset Pricing In Emerging Markets’, Working paper WP/96/2, International Monetary Fund. Chen Nai-Fu, Roll, R. and Ross, S.A. (1986) ‘Economic Forces and the Stock Market’, Journal of Business, vol. 59(3), pp. 383–403. Chuhan, P. (1992) ‘Sources of Portfolio Investment in Emerging Markets’, World Bank Working Paper, International Economics Department, World Bank, Washington DC. Claessens, S. (1995) ‘The Emergence of Equity Investment in Developing Countries: Overview’, World Bank Economic Review, vol. 9(1), pp. 25–36. Claessens, S. and Gooptu, S. (eds) (1993) Portfolio Investments in Developing Countries, World Bank Discussion paper no. 228. Claessens, S. and Rhee, M.W. (1994) ‘The Effects of Barriers on Equity Investment in Developing Countries’, World Bank Policy Research Paper no. 1263. Demirgüç-Kunt, A. and Levine, R. (1996) ‘Stock Market Development and Financial Intermediaries: Stylized Facts’, The World Bank Economic Review, vol. 10(2), pp. 291–321. Demirgüç-Kunt, A. and Huizinga, H. (1999) ‘Market Discipline and Financial Safety Net design’, CEPR Discussion paper, 2311. Derrabi, M., de Bodt, E. and Cobbaut, R. (2000), ‘Microstructure Changes and Stock Price Behaviour, Evidence from casablanca stock exchange’, NASDAQNotre Dame Microstructure Conference, Paris. De Roon, F.A., Nijman, T.E. and Werker, B.J.M. (2001) ‘Testing for MeanVariance Spanning with Short Sale Constraints and Transaction Costs: The Case of Emerging Markets’, Journal of Finance, vol. 56, pp. 723–44. Dickey, D.A. and Fuller, W.A. (1979) ‘Distribution of the Estimators for Autoregressive Time Series with a Unit Root’, Journal of the American Statistical Association, vol. 74, pp. 427–31. Divecha, A.B., Drach, J. and Stefek, D. (1992) ‘Emerging Markets: A Quantitative Perspective’, Journal of Portfolio Management, vol. 19(1), pp. 41–50. Errunza, V.R. (1994) ‘Emerging Markets: Some New Concept’, Journal of Portfolio Management (Spring), pp. 82–7. Fama, E.F. (1965) ‘The Behaviour of Stock Market Prices’, Journal of Business, vol. 28, pp. 34–105. Fama, E.F. (1970) ‘Efficient Capital Markets: A Review of Theory and Empirical Work’, Journal of Finance, vol. 25, pp. 383–417. Fama, E.F. (1991) ‘Efficient Capital Markets: II’, Journal of Finance, vol. 46, pp. 1575–617. Frankel, J. and Schmukler, S. (1996) ‘Crisis, Contagion and Country Funds: Effects on East Asia and Latin America’, Pacific Basin Working Paper Series, no. PB96-04.
Mohamed Derrabi and Michel Leseure 55
French, K. (1980) ‘Stock Returns and the Weekend Effect’, Journal of Financial Economics, vol. 8, pp. 55–70. Gibbons, M. and Hess, P. (1981) ‘Day of the Week Effects and Asset Returns’, Journal of Business, vol. 54, pp. 579–96. Granger, C.W.J. (1969) ‘Investigating Causal Relationships by Econometric Models and Cross-Spectral Methods’, Econometrica, vol. 37, pp. 424–38. Harvey, C.R. (1995a) ‘Predictable Risk and Returns in Emerging Markets’, Review of Financial Studies (Fall), vol. 8(3), pp. 773–816. Harvey, C.R. (1995b) ‘The Risk Exposure of Emerging Equity Markets’, World Bank Economic Review, vol. 9, pp. 19–50. Harvey, C.R. (2000) ‘The Drivers of Expected Returns in International Markets’, Emerging Market Quaterly (Fall), pp. 32–49. Hauser, S., Marcus, M. and Yaari, U. (1994) ‘Investing in Emerging Markets: Is It Worthwhile Hedging Foreign Exchange Risk?’, Journal of Portfolio Management (Spring), pp. 76-81. Jaffe, J. and Westerfield, R. (1985) ‘The Week-End Effect in Common Stock Returns: The International Evidence’, Journal of Finance, vol. 40, pp. 433–54. Kaminsky, G. and Schmukler, S. (1998) ‘What Triggers Market Jitters? A Chronicle of the Asian Crisis’, World Bank mimeo. Keane, S. (1993) ‘Emerging Markets – The Relevance of Efficient Market Theory’, The Chartered Association of Certified Accountants (ACCA) Technical and Research (T&R) Committee Occasional Research Papers no. 15. Korajczyk, R. (1996) ‘A Measure of Stock Market Integration for Developed and Emerging Markets’, World Bank Economic Review, vol. 10, pp. 267–90. Krueger, A.O., Schiff, M.W. and Valdes, A. (1992) The Political Economy of Agricultural Price Intervention in Latin America (San Francisco, CA: ICS Press). Li, K., Sarker, A. and Wang, Z. (2003) ‘Diversification Benefits of Emerging Markets Subject to Portfolio Constraints’, Journal of Empirical Finance, vol. 10, (1–2), pp. 57–80. Mullin, J. (1993) ‘Emerging Equity Markets in the Global Economy’, Federal Reserve Bank of New York Quarterly Review, vol. 18, pp. 54–83. Roll, R. (1983) ‘The Turn-of-the Year Effect and the Return Premia of Small Firms’, Journal of Portfolio Management (Winter), pp. 18–28. Roll, R. (1992) ‘Industrial Structure and the Comparative Behaviour of International Stock Market Indices’, Journal of Finance, vol. 47, pp. 3–42. Sachs, J., Tornell, A. and Velasco, A. (1996) ‘Financial Crises in Emerging Markets: The Lessons from 1995’, Brooking Papers on Economic Activity, vol. 1, pp. 147–215. Wilcox, J.W. (1992) ‘Global Investing in Emerging Markets’, Financial Analysts Journal, vol. 48, pp. 15–19.
This page intentionally left blank
5 Random Walk in Emerging Markets: A Case Study of the Karachi Stock Exchange Orla Gough and Ali Malik
Introduction The efficient capital market hypothesis has been one of the dominant themes in the academic literature since the 1960s. It was generally believed that securities markets were extremely efficient in reflecting information about individual stocks and about the stockmarket as a whole. When we refer to the efficient capital market hypothesis, we mean that security prices fully reflect all available information (Elton and Gruber, 2003). Three forms of market efficiency have been suggested subject to different information sets (Roberts, 1959; Fama, 1970). Under the weak form of the efficient market hypothesis (EMH), stock prices are assumed to reflect any information that may be contained in the past history of the stock price. Under the semi-strong form all publicly available information is presumed to be reflected in the securities prices. Finally, the strong form takes the theory of market efficiency to the ultimate extreme. It claims all information is reflected in stock prices, including private or insider information as well as that which is publicly available. There exists a strong measure of consensus on the validity of the weak and semi-strong forms of the EMH with respect to capital markets in developed countries such as the USA, Britain and Japan (Dickinson and Muragu, 1994). This view is supported by the tremendous amount of research evidence (for a comprehensive review see Fama, 1970, 1991, and Keane, 1983). The primary role of a capital market is to allocate the scarce economic resources to various available productive uses. The efficient market results in optimal utilization of resources that underpins the 57
58 Random Walk and the Karachi Stock Exchange
investment activities and ultimately strengthens the overall economy of the country. It can be argued that an inefficient market impedes economic growth but an efficient market attracts local and foreign investment, the latter being vital to the economic development of developing and undeveloped countries. This study extends the evidence of stockmarket efficiency from the Karachi Stock Exchange in Pakistan. In this chapter we investigate whether the Karachi capital market could be classified as weak-form efficient under the efficient market hypothesis. The chapter is organized as follows: the next section introduces the concept of the random walk. This is followed by a brief literature review on the weak form of market efficiency and various market anomalies; a profile of the Karachi Stock Exchange; an explanation of the data and methodology; analysis of the results of our statistical analysis; and in the final section we draw conclusions from our findings.
Random walk Fama (1998) expanded the definition of weak-form efficiency, changing the earlier classification ‘weak-form tests’ to the more general category ‘tests for return predictability’. Tests for the weak form of efficiency have their origin in what has come to be known as random walk (RW) theory. The latter derives it name from a series of market studies carried out by researchers, mainly in the 1950s and 1960s, which indicated that successive returns are independent and that the returns are identically distributed over time (Keane, 1986). If share prices follow a random walk, the implication is that share prices only move in response to the disclosure of new information that is relevant to their value. If there is evidence of non-random movements in share prices, this implies that the market is being inefficient because share prices are moving for reasons other than the disclosure of information (Lumby, 1994). The RW tests address questions such as: (a) do prices over time have sufficient co-relation to allow investors to predict future price movements by studying trends, and (b) can trading strategies based on price movements provide opportunities for abnormal profit? The first group of tests have consistently demonstrated that the pattern of share price movements substantially follows a random walk and that price
Orla Gough and Ali Malik 59
changes are independent of prior movements. The second group of tests are focused on the effectiveness of using certain specific trading rules designed to exploit possible systematic patterns in share price movements. LeRoy (1989) argues that resources used on securities analysis are unproductive given the random walk concept. Some commentators claim that the RW hypothesis suggests that the market has no memory, implying that market pricing decisions are not affected by past price levels. The RW model suggests that unexploited patterns in securities’ prices cannot persist because for them to do so would imply that investors are irrationally passing-up profit opportunities, and secondly that investors are nonetheless irrationally wasting their money year after year employing useless securities analysts.
Literature review on weak-form efficiency Weak-form efficiency depends on statistical investigation of the time series of prices. If there is substantial dependence in price changes, this suggests that it may be possible to earn excess returns by using a simple trading rule. If the subsequent price changes are found independent of each other, this suggests that the market is informational efficient. The first reported study examining market prices was by Bachelier (1900), who found that commodity price changes on the French Bourse followed a random walk. Working (1934) and Kendall (1953) also found evidence that changes in market prices were random on both the United States and United Kingdom exchanges. Robert (1959) compared the movement of the Dow Jones industrial average, and found that the random-walk process produced patterns that were very similar to those of the Dow Jones index thereby suggesting that stock price movements may be random. Osborne (1962), a contemporary of Roberts and working along similar lines, Osborne’s showed that share price changes are random in nature and that past price changes have no predictive value. Robert and Osborne’s studies were followed by a substantial amount of statistical research into share prices. The essence of the modern theory of efficient markets is often attributed to Fama for an excellent review see Fama (1965, 1970, 1991), whose major early study (Fama, 1965) tested for serial correlation for the 30 firms comprising the Dow Jones industrial average during the five years ending in
60 Random Walk and the Karachi Stock Exchange
1962. He found a very small amount of serial correlation, but this was not statistically significantly different from zero. Fama also used a runs test to examine for statistical independence in price movements. Again, he found some slight dependence, but it was very small and not significantly different from zero. In the UK, Brealey (1970), Cunningham (1973) and Dryden (1969) were among the earlier researchers, Brealey (1970) and Dryden (1969) used serial correlation tests and runs tests. However, Cunningham (1973) who used serial correlation tests only, found that there was no evidence of significant dependencies in price changes, thus providing evidence that the British stockmarket was weak-form efficient. Rosenberg and Rudd (1982) found a lack of serial correlation in the total returns of securities with respect to each of the major components of a security’s return. A security’s total return is composed of two elements, the return that is common to all securities and the return that is specific to the individual security. Rosenberg and Rudd (1982) found a positive serial correlation for the common component and a correspondingly negative correlation for the specific component, resulting in an increased predictability of the total returns. Studies on emerging markets fall into two categories: a first group of studies conclude that stockmarkets in emerging countries are weakform efficient, while a second group concludes otherwise. Among the first group is Barnes (1986) who provided evidence that the Kuala Lumpur Stock Exchange is weak-form efficient; the evidence was then extended by Chan, Gup and Pan (1992) on major Asian markets. This group also includes Dickinson and Muragu (1994) on the Nairobi Stock Exchange; Chan, Gup and Pan (1997) on the stockmarkets of 18 developing countries; Blasco, Rio and Santamar’ia (1997) who examined evidence on the Madrid Stock Exchange; and Ojah and Karena (1999) who analysed the four Latin American stockmarkets. Among the second group, Cheung, Wong and Ho (1993) tested the stockmarkets of Korea and Taiwan; Clasessens, Dasgupta and Glen (1995) tested the 19 emerging stockmarkets; Khababa (1998) examined the Saudi financial market; while Poshakwale (1996) worked on the efficiency of the Indian stockmarket. Another type of test for weak-form efficiency is to examine whether excess returns can be earned from following mechanical investment strategies. These strategies give objective signals for buying and selling securities and, because they are so explicit, they can be easily tested.
Orla Gough and Ali Malik 61
Many of these strategies are based on rules that earn excess returns on one set of historical data and are therefore claimed as being successful, and proof that the market is not efficient. However, the real test should be whether a strategy could consistently earn excess returns in future. Researchers have found that abnormal returns could be earned by using certain filter techniques, but found that the profit disappeared after taking into account transaction costs (see for example Alexander, 1961; Fama, 1965; Fama and Blume, 1966; Jensen and Bennington, 1970).
The Karachi Stock Exchange The Karachi Stock Exchange (KSE) is the oldest and biggest of the three stock exchanges in Pakistan. It came into existence on 18 September 1947, just one month after the independence of the country, and was later converted and registered as a company limited by guarantee on 10 March 1949. Initially, only five companies were listed with a paid-up capital of 37 million rupees. As at June 2004, there were 673 companies listed on the KSE with a total listed capital of Rs 368,846 millions. Total market capitalization is over Rs 1,459,206 millions. The stockmarket and the corporate sector are regulated by the provision of (1) the Companies Ordinance 1984, and (2) the Securities and Exchange Ordinance 1969. The regulatory agency of the stockmarket and the corporate sector is the Corporate Law Authority (CLA) which is a division of the Ministry of Finance. In 1991 the secondary market was open to foreign investors on an equal basis with local participants. This measure, along with the policy of privatization, has resulted in the rapid growth of the market since 1991. The market is expected to achieve enormous growth in the next five years due to the liberalization and deregulation policies of the government.
Data and methodology We used the stock exchange index, KSE 100, for the period from 1993 to 2002. Daily KSE index data was obtained from Datastream as well as from the record room of the Karachi Stock Exchange. The prices are studied in their logarithmic form using the model: ln Pt − ln Pt−1 = ut
(5.1)
62 Random Walk and the Karachi Stock Exchange
where Pt is the price of a stock adjusted for capital changes at time t; and ut is the first difference in log prices from time t − 1 to time t. The following testing methodology has been used to investigate the random walk hypothesis for the Karachi Stock Exchange.
Runs tests The runs test or non-parametric test examines the pattern presented by the residuals to conclude whether residuals are random or not. We define our hypotheses as: Hypothesis H0 Successive residuals are random and hence cannot be used to forecast future residuals. Hypothesis H1 Successive residuals are non-random and hence can be used to forecast future residuals. A run is defined as an uninterrupted sequence of one symbol such as + or −. The length of a run is the number of elements, here residuals, in it. Now the question is whether calculated runs in our sample are too many or too few as compared with the number of runs expected in a strictly random sequence of same observations. If there are too many runs in our sample, it suggests a negative serial correlation, and vice versa. To prove the existence of autocorrelation statistically, let: n = total number of residuals n1 = number of + residuals n2 = number of − residuals k = number of runs If the hypothesis of randomness is sustainable, we should expect k to lie between [E(k) ± 1.96σ k ] with 95% confidence. The decision rule is that we do not reject the null hypothesis of randomness with 95% confidence if [E(k) − 1.96σ k ≤ k ≤ E(k) + 1.96σ k ]
(5.2)
where E(k) =
2n1 n2 n1 + n2 + 1
Variance = σ 2 k =
2n1 n2 (2n1 n2 − n1 − n2 ) (n1 + n2 )2 (n1 + n2 − 1)
and σk = square root of σ 2 k.
(5.3) (5.4)
Orla Gough and Ali Malik 63
Dickey–Fuller (DF) unit root test The second test used to describe the return behaviour is what is popularly known as the unit-root test of stationarity. Consider the following model: Yt = Yt−1 + ut
(5.5)
where ut is the stochastic error that follows the classical assumption, zero mean and constant variance σ 2 and is non-autocorrelated. The unit-root problem occurs when the coefficient of Yt−1 is actually equal to 1. We rewrite it as: Yt = ρYt−1 + ut
(5.6)
and if we find that ρ = 1, then variable Yt has a unit root. A time series that has a unit root is known as a random walk, and a random walk is an example of a non-stationary time series. Therefore, if the stock prices are non stationary, they follow a random walk. Equation (5.6) is expressed in an alternative form as: Yt = (ρ − 1)Yt−1 + ut
(5.7)
Yt = δYt−1 + ut
(5.8)
or
where δ = (ρ − 1). We define our null as H0 : δ = 0; against the alternative as H1 : δ = 0. If δ is equal to 0, we can write equation (5.8) as: Yt = (Yt − Yt−1 ) = ut
(5.9)
Equation (5.9) says that the first differences of a random-walk time series are a stationary time series because, by assumption, ut is purely random. After running the regression, we examine whether δ = 0 on the basis of the t statistic whose critical values have been calculated and tabulated by Dickey and Fuller (1979). The Dickey–Fuller test is applied to regressions run in the following forms: Yt = δYt−1 + ut Yt = β1 + δYt−1 + ut Yt = β1 + β2 t + δYt−1 + ut
64 Random Walk and the Karachi Stock Exchange
where t is the trend variable. The difference between the last two equations and the first equation lies in the inclusion of the constant (intercept) and the trend term. If the error term ut is autocorrelated, one modifies the last equation as follows: Yt = β1 + β2 t + δYt−1 + αi
Yt−i + ut
(5.10)
The above model is called the Augmented Dickey–Fuller (ADF) test.
The autocorrelation test We also used the coefficient of autocorrelation test to measure the correlation between members of series of observations in our time-series data. The classical linear regression model assumes that autocorrelation does not exist in the disturbances ui . Symbolically: E(ui uj ) = 0
i = j
(5.11)
Symbolically, if autocorrelation exists: E(ui uj ) = 0 i = j
(5.12)
While disturbances ui are generated as follows: ui = ρut−1 + εt
−1 < ρ < 1
(5.13)
where ρ is the coefficient of autocorrelation of lag 1 with the following standard ordinary least-squares (OLS) assumptions: E(ut ) = 0 Var(ut ) = σ 2
(5.14)
By definition, the coefficient of correlation between ut and ut−1 is: ρ=
E(ut ut−1 ) Var(ut−1 )
(5.15)
The coefficient of autocorrelation is related to standard error and the t-ratio. This simple autocorrelation test is used to examine the null of no serial correlation against the alternative of serial correlation in the time series of KSE data.
Orla Gough and Ali Malik 65
Results 1 By solving the above statistics for our sample using the runs test, we get n1 = 27 and n2 = 26, while n = 53, and: E(k) = 27.4905 σk = 3.6035
(5.16)
The 95% confidence interval is: [27.4905 ± 1.96(3.6035)] = (34.6151, 20.4276)
(5.17)
Our number of runs is 53 and clearly falls outside this interval; therefore, we can reject the null with 95% confidence that the observed sequence of residuals is random.
2 The Dickey–Fuller unit-root test was conducted for the 12th order of ADF regression. However, irrespective of the order of the augmentation chosen for the ADF test, the absolute values of the ADF statistics are all well-above the 95% critical value of the test, that is −2.8631. The results suggest that the test statistics approach closer to the critical value as we use the higher-order ADF regression. The other different model selection criteria suggest that the correct order is likely to be between ADF (3) and ADF (4) with the Schwarz Bayesian criterion, as well as with the Akaike Information criteria. The results of ADF regression are presented in Table 5.1.
3 The autocorrelation test results are presented in Table 5.2 and are tabulated as order, autocorrelation coefficient, standard error and Box–Pierce Q and Ljung–Box Q∗ statistics. The figures in square brackets refer to the probability of falsely rejecting the null hypothesis of no serial correlation. The results present clear evidence of serial correlation in KSE index data. The first and fourth-order autocorrelation coefficients 0.1592 and 0.1200 are large relative to their standard errors (t-ratios for these coefficients are 8.19 and 5.94) which are well-above the critical value of the standard normal distribution at the 5% level which is 1.96. All other coefficients are not statistically significant.
66
Table 5.1 Results of the Dickey–Fuller unit-root test (sample period 1993–2002)
DF DF (1) DF (2) DF (3) DF (4) DF (5) DF (6) DF (7) DF (8) DF (9) DF (10) DF (11) DF (12)
Test statistics
AIC
SBC
−43.6769 −31.1035 −25.0252 −20.3996 −18.6888 −17.2459 −16.7943 −15.7831 −14.7518 −13.6377 −13.1018 −12.5036 −12.1043
8069.0 8073.8 8077.6 8089.0 8088.3 8087.7 8087.7 8086.8 8086.4 8087.4 8086.4 8085.7 8084.7
8063.1 8065.0 8065.9 8074.3 8070.6 8067.1 8064.1 8060.4 8057.1 8055.0 8051.1 8047.5 8043.6
Notes: 2,636 observations were used in all the ADF regressions. The 95% critical value for the ADF statistics was −2.8631. AIC = Akaike Information criterion; SBC = Schwarz Bayesian criterion.
Table 5.2 Results of the autocorrelation test [Variable PAKISTAN Sample from 1 to 2,649]
Order 1 2 3 4 5 6 7 8 9 10 11 12
Autocorrelation coefficient ·15924 ·089954 ·082233 ·12006 ·053825 ·045107 −·306908 ·024970 ·033762 ·051010 ·020489 ·028376
Standard error .019429 .019916 .020069 .020196 .020463 .020517 .020554 .020554 .020565 .020586 .020634 .020642
Box–Pierce statistic 67·1745 88·6093 106·5226 144·7041 152·3785 157·7682 157·7684 159·4201 162·4397 169·3325 170·4446 172·5776
[.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000]
Ljung–Box statistic 67·2506 88·7178 106·6650 144·9330 152·6278 158·0338 158·0340 159·6920 162·7241 169·6483 170·7658 172·9101
[.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000] [.000]
Orla Gough and Ali Malik 67
Table 5.3 Results of the day-of-the week effect Day
Coefficients
Monday Tuesday Wednesday Thursday Friday
−·8720003 ·0016066 ·0012886 ·0011067 −·8696300
t-ratio [prob] −1·5008 2·7651 2·2177 1·9040 −1·4966
[.134] [.006] [.027] [.057] [.135]
As an additional measure, we also tested for the day-of-the-week effect. The Monday anomaly favours a positive return on Friday and a negative return on Monday (see French, 1980; Pettengil, 1989; Athanassakos and Robinson, 1994; and Kamara, 1997). We found negative coefficients for both Monday and Friday and positive coefficients for all other days of the week. The results, presented in Table 5.3 are tabulated as days, coefficients, t-ratio [probability]. A possible explanation for our findings may be the fact that there used to be a ‘Friday Holiday’ since 1996, followed closely after by a ‘Sunday Holiday’. Moreover, Friday is a religious day and is subject to less trading activities than other days of the week.
Conclusion Informational efficiency is considered vital to the efficiency of a capital market. The optimal allocation of scarce economic resources cannot be ensured unless the capital market is information-efficient. This chapter has focused on investigating whether the behaviours of the price series in the Karachi stockmarket were consistent with the weak form of the EMH. The runs test, the Dickey–Fuller test and autocorrelation tests were employed on daily and weekly KSE 100 index data to empirically examine whether successive price changes are random. The results contradict the weak form of the EMH, suggesting that the KSE market is informationally inefficient. An interesting finding was a negative return on Fridays that is contrary to the existing evidence. The findings have obvious implications. The securities prices at the KSE do not reflect accurate information and may hence lead to incorrect portfolio decisions by private and foreign investors.
68 Random Walk and the Karachi Stock Exchange
References Alexander, S. (1961) ‘Price Movements in Speculative Markets: Trends or Random Walks?’, in P.H. Cootner (1964) The Random Character of Stock Market Prices, 1964, Cambridge, MA: MIT Press, pp. 199–218. Athanassakos, G. and Robinson, M.J. (1994) ‘The Day-of-the-Week Anomaly: The Toronto Stock Exchange Experience’, Journal of Business Finance and Accounting, vol. 21(6) (September), pp. 833–56. Bachelier, L. (1900) ‘Theorie de la Speculation’, Gauthier-Villars, Paris, reprinted in English (A.J. Bones, trans) in P.H. Cootner (ed.), The Random Character of Stock Market Prices (Cambridge, MA: MIT Press, 1964), pp. 17–78. Barnes, P. (1986) ‘Thin Trading and Stock Market Efficiency: A Case of the Kuala Lumpur Stock Exchange’, Journal of Business Finance and Accounting, vol. 13(4), (Winter), pp. 609–17. Blasco, N., Rio, C.D. and Santarmari’ia, R. (1997) ‘The Random Walk Hypothesis in the Spanish Stock Market: 1980–1992’, Journal of Business Finance and Accounting, 24(5) ( January) pp. 667–84. Brealey, R.A. (1970), ‘The Distribution and Independence of Successive Rates of Return from the British Equity Market’, Journal of Business Finance, vol. 2, pp. 29–40. Chan, K.C., Gup, B.E. and Pan, M. (1992) ‘An Empirical Analysis of Stock Prices in Major Asian Markets and United States’, The Financial Review, vol. 27(2), (May), pp. 287–307. Chan, K.C., Gup, B.E. and Pan, M. (1997) ‘International Stock Market Efficiency and Integration: A Study of Fifteen Nations’, Journal of Business Finance and Accounting, 24(6) (July), pp. 803–13. Cheung, Y., Wong, K. and Ho, Y. (1993), ‘The Pricing of Risky Assets in Two Emerging Asian Markets – Korea and Taiwan’, Applied Financial Economics, vol. 3, pp. 315–24. Claessens, S., Dasgupta, S. and Glen, J. (1995) ‘Return Behaviour in emerging Stock Markets’, The World Bank Economic Review, vol. 9(1), pp. 131–51. Cunningham, S.W., ‘The predictability of British stock market prices’, Applied Statistics, vol. 22, pp. 215–31. Dickey, D.A. and Fuller, W.A. (1979) ‘Distribution of the Estimators for Autoregressive Time Series with a Unity Root’, Journal of the American Statistical Association, vol. 74, pp. 427–31. Dickinson, J.P. and Muragu, K. (1994) ‘Market Efficiency in Developing Countries: A Case Study of the Nairobi Stock Exchange’, Journal of Business Finance and Accounting, 21(1) ( January), pp. 133–50. Dryden, M. (1969) ‘A Source of Bias in Filter Tests on Share Prices’, Journal of Business (July), pp. 321–25. Elton, E.J. and Gruber, M.J. (2003) Modern Portfolio Theory and Investment Analysis (New York: John Wiley), pp. 406–48. Emory, C.W. and Cooper, D.R. (1991) Business Research Methods (Burr Ridge, IL: Irwin). Fama, E. (1965) ‘The Behaviour of Stock Market Prices’, Journal of Business, vol. 38, pp. 34–105.
Orla Gough and Ali Malik 69
Fama, E. (1970) ‘Efficient Capital Markets: A Review of Theory and Empirical Work’, Journal of Finance, vol. 25(2) (May), pp. 383–417. Fama, E. (1976) ‘Efficient Capital Markets: Reply’, Journal of Finance (October), 3(4), pp. 143–5. Fama, E. (1991) ‘Efficient Capital Markets II’, Journal of Finance, vol. XLVI(5) (December). pp 1575–617. Fama, E. and Blume, M. (1966) ‘Filter Rules and Stock Market Trading’, Journal of Business, vol. 39(1) (January), pp. 226–41. Firth, M. and Keane, S.M. (1986) Issues in Finance (Oxford: Philip Allan), pp. 1–43. French, K. (1980) ‘Stock Returns and the Weekend Effect’, Journal of Financial Economics, vol. 8(1), pp. 55–69. Gill, J. and Johnson, P. (1997) Research Methods for Managers, 2nd edn (London: Paul Chapman). Jensen, M. and Bennington, G. (1970) ‘Random Walk and Technical Theories: Some Additional Evidence’, Journal of Finance, vol. XXV(2), pp. 469–82. Kamara, A. (1997) ‘New Evidence on the Monday Seasonal in Stock Return’, Journal of Business, vol. 70, pp. 63–84. Keane, S. (1983) Stock Market Efficiency: Theory, Evidence, Implications (Oxford: Philip Alan, reprinted 1985). Keane, S. (1986) ‘The Efficient Market Hypothesis on Trial’, Financial Analyst Journal, (March–April), pp. 58–63. Kendall, M.G. (1953) ‘The Analysis of Economic Time-Series, Part I: Prices’, in P.H. Cootner (ed.), The Random Character of Stock Market Prices, 1964, Cambridge, MA: MIT Press, pp. 85–99. Khababa, N. (1998) ‘Behaviour of Stock Prices in the Saudi Arabian Financial Market: Empirical Research Findings’, Journal of Financial Management and Analysis, vol. 11(1) (January–June), pp. 48–55. LeRoy, S.F. (1989) ‘Efficient Capital Markets and Martingales’, Journal of Economic Literature, vol. XXVII (December), pp. 1583-621. Lumby, S. (1994) Investment Appraisal and Financial Decisions, 5th edn (London: Chapman Hall). Osborne, M.F.M. (1962) ‘Periodic Structure in the Brownian Motion in the Stock Prices’, Operation Research, vol. 10 (May/June), pp. 345–79. Ojah, K. and Karemera, D. (1999) ‘Random Walks and Market Efficiency Tests of Latin American Emerging Equity Markets: A Revisit’, The Financial Review, vol. 34, pp. 57–72. Pettengill, G.N. (1989) ‘Holiday Closings and Security Returns’, Journal of Financial Research (Spring), pp. 57–67. Poshakwale, S. (1996), ‘Evidence on Weak Form of Effeciency and Day of the Week Effect in the Indian Stock Market’, Finance India, vol. 10, pp. 605–16. Robert, H.V. (1959) ‘Statistical Versus Clinical Prediction of the Stock Market’, Working Paper, University of Chicago.
70 Random Walk and the Karachi Stock Exchange
Rosenberg, B. and Rudd, A. (1982) ‘Factor-Related and Specific Returns on Common Stocks: Serial Correlation and Market Inefficiency’, Journal of Finance, vol. 37, pp. 543–54. Working, H. (1934) ‘A Random Difference Series for Use in the Analysis of Time Series’, Journal of the American Statistical Association, vol. 29, pp. 11–24. www.kse.com.pk (official website of Karachi Stock Exchange).
6 Insiders’ Market Timing and Real Activity: Evidence From an Emerging Market Tomasz Piotr Wisniewski ∗
Introduction Earlier research on insider trading has documented unequivocally that officers, directors and controlling shareholders are in possession of valuable private information and exploit it profitably in security trading.1 It is widely believed that the apparent informational asymmetry arises from the foreknowledge of public disclosures. Consequently, a number of studies have investigated the intensity of insider trading prior to corporate events, such as takeover bids (Seyhun, 1990), dividend and earnings announcements (John and Lang, 1991; Ke, Huddart and Petroni, 2003), stock repurchases (Lee, Mikkelson and Partch, 1992), or bankruptcies (Seyhun and Bradley, 1997). However, as argued by Seyhun (1988a, 1992), not all of the mispricing observed by insiders has to be firm-specific. Insiders are best positioned to recognize unanticipated changes in cash flows to their own firms that signal either a shift in the competitiveness of their ∗
The author wishes to thank participants of the International Conference on Emerging Markets and Global Risk Management organized by the Westminster Business School, the 8th Meeting of the New Zealand Finance Colloquium in Hamilton, the 3rd Annual Conference of the Research Centre on Modern Europe at the Wilfrid Laurier University in Canada and the research seminar at the European University Viadrina Frankfurt (Oder) provided useful remarks and comments. The suggestions made by Martin T. Bohl, Alireza Tourani-Rad, Aaron Gilbert, Shauna Selvarajah and Dobromir Tzotchev are also gratefully acknowledged. The author retains sole responsibility for all remaining errors.
71
72 Insiders’ Market Timing and Real Activity
companies or fluctuations in general business conditions. Nevertheless, all considered, they are not able to assess the relative contribution of these factors ex ante. Only after the macroeconomic development is widely recognized do prices of all assets adjust accordingly. This, in turn, could explain the positive correlations between the lagged values of aggregate insider-trading indices, current market returns and real activity variables found in Seyhun (1988a, 1992). The signal identification problem presented here is essentially analogous to that of Lucas (1973, 1975), but used in another context. Although the literature tends to conform to the conjecture that disclosed insider transactions carry information about future market movements, disagreement about their predictive power remains. Using single-equation modelling methods, Seyhun (1992) concludes that up to 60 per cent of variation in 12-month-ahead excess stock returns can be forecasted using the previous 12-month aggregate insider trading. On the other hand, the results of bivarite causality tests in Chowdhury, Howe and Lin (1993) and Iqbal and Shetty (2002) suggest that the ability of insider transactions to predict subsequent market returns is slight. The causality appeared to be stronger in the opposite direction, indicating that insiders are, in aggregate, contrarian investors. In their comprehensive study, Lakonishok and Lee (2001) arrive at similar conclusions. In light of these conflicting views, the overriding motivation for this chapter is to provide new evidence on the degree of stock return forecastibility and to test the theoretical implications of the cash-flow hypothesis posed in Seyhun (1992). These questions are addressed quantitatively within a trivariate vector autoregressive framework. In particular, causality tests, forecast-error variance decomposition and orthogonalized impulse-response functions are employed to measure the strength of association between growth in industrial production, real market returns and insider-trading activity in Poland. To double-check the results, two aggregate insider-trading indices are constructed; first based on an entire sample of transactions, and a second one which takes into account only the trades of managers. The contribution of this study is threefold. First, the link between insider-trading indices and leading indicators of economic activity found in Seyhun (1992) has not been directly modelled in subsequent studies. This chapter augments the extant evidence, within a model setting which can provide deeper insights into the structure
Tomasz Piotr Wisniewski 73
of dynamic interactions and casual relations. Notably, the inferences based on causality tests in a bivarite vector autoregression, used in the literature, may not be robust to the addition of new variables into the system (Mehra, 1978; Sims, 1980a; and Lee, 1992). Second, the features of the dataset permit the use of actual publication dates. Only after its public disclosure can the information about aggregate insider trading help analysts to forecast market returns and the future state of the economy. Although greatly relevant to the issue at hand, the exact disclosure dates were used previously only in Seyhun (1988a) and proxied with a delay parameter in Seyhun (1992). Third, to the best knowledge of the author, this is the first article to analyse insiders’ ability to time the movements of an emerging market. As suggested by the findings of Bhattacharya et al. (2000) and Bhattacharya and Daouk (2002), the exploitation of confidential information is more evident in markets with lax enforcement of insider-trading sanctions.2 It would be of interest to determine whether the implicit costs of trading, such as the fear of potential indictment, trial or conviction effect the magnitude of predictive power. The remainder of the chapter is organized as follows. The next section describes data sources, construction of variables and sample characteristics, followed by a brief outline of the methodology. Empirical findings on the predictive ability of aggregate insider trading are then presented, followed by a summary and conclusion in the final section.
Data The sample used in this study comprises insider transactions reported to the Securities and Exchange Commission from January 1999 to August 2004, for a total of 68 calendar months.3 The publication date is the date on which information about the trade appeared on the Internet portal Interia. The consumer price index and real industrial production growth time series were taken from the National Bank of Poland archive. Security prices, trading volume and market indices were obtained by courtesy of the Warsaw Stock Exchange. Several filters were applied to the initial insider transaction data. All duplicate records were deleted and the transaction price was assumed to equal the daily closing quote wherever data on it was missing. Trades reflecting the exercise or conversion of managerial options,
74 Insiders’ Market Timing and Real Activity
executive compensation schemes, purchases of stocks in seasoned equity offerings and private transactions were discarded. The overall sample contains 2,663 trades in 215 firms, out of which 1,971 can be attributed to management. Management is defined here as members/chairmen of executive and supervisory boards and close family members of these individuals. The share of registered insider trading in the total value of trade on the Warsaw Stock Exchange amounted to 1.82 per cent. Intuitively, this estimate can be deemed large, especially given the fact that transactions driven by material non-public information are unlikely to be self-reported (Brainbridge, 2000). Furthermore, this proportion is sizeable relative to other markets. In a sample of US companies, Bettis, Coles and Lemmon (2000) found that the average number of shares traded by insiders to the total volume per allowed trading day equalled 0.66 per cent versus 0.21 per cent during blackout periods. Two indicators of insider-trading activity are computed. The first is based on the entire sample of transactions, whereas the second takes into account only transactions by management.4 A cognate way of aggregation can be found in Seyhun (1988b) and Lakonishock and Lee (2001). Aggregate insider-trading indices are defined as follows: nt AITAt
= i=1 nt
di pi,t Vi
i=1
pi,t Vi
nt mi di pi,t Vi AITM = i=1 nt t i=1 mi pi,t Vi
(6.1) (6.2)
where di equals 1 for purchases and −1 for sales, mi takes the value of 1 if the insider is a manager and zero otherwise, pi,t is the transaction price and Vi is the volume of trade. The total number of insider transactions in a month t is denoted by nt . The two remaining variables employed in the vector autoregression model are the real stock returns (rrEW), calculated as the continuously compounded return on the equally weighted market index deflated by the changes in CPI,5 and growth of industrial production (dIP).6 Industrial production has been chosen as a proxy for real activity, as it is the only aggregate data series available on a monthly basis. Descriptive statistics of all the variables included in the sample are displayed in Table 6.1.
Table 6.1
Descriptive statistics Cross-sectional correlation
Mean
Standard deviation
−0.0055
0.0644
0.0592
0.0683
0.2973 (0.0699) –
AIT A
−0.3439
0.5046
–
−0.0077 (0.9634) −0.0652 (0.0699) –
AIT M
−0.3310
0.4958
–
–
Variable rrEW dIP
dIP
AITA
AITM −0.0354 (0.8329) −0.1904 (0.2522) 0.7354 (0.0000) –
Serial correlation r1 0.2391 (0.1483) 0.7119 (0.0000) 0.1743 (0.2953) 0.2350 (0.1555)
r2 0.0638 (0.7035) 0.7097 (0.0000) −0.0158 (0.9250) 0.1677 (0.3142)
r3 0.0633 (0.7058) 0.6781 (0.0000) −0.0935 (0.5766) −0.0554 (0.7411)
The p-values are shown in parentheses. rrEW = real return on the equally weighted market portfolio; dIP = percentage change in the real industrial production; AIT A = aggregate insider trading index (all transactions); AIT M = aggregate insider trading index (managers’ transactions only). rτ is the serial correlation coefficient at lag τ .
75
76 Insiders’ Market Timing and Real Activity
The first part of the period under consideration was characterized by a moderate economic growth. The revival came in early 2003 with the industrial production growth peaking around Poland’s EU accession date. These macroeconomic trends were accompanied by falling real stockmarket prices, with a monthly mean decrease of over .5 per cent. Corporate insiders were, on average, net sellers. Inspection of the distribution of the two aggregate insider-trading indices shows that the behaviour of managers and large shareholders could have been much alike. The subtle and statistically insignificant difference is that managers exhibited stronger persistence in their investment strategies as reflected by the higher first- and second-order autocorrelation coefficients. The correlation patterns shed some more light on the data. Growth in industrial production, which can be viewed as a source of systematic investment risk, is positively correlated with real stock returns. Chen, Roll and Ross (1986) find a similar, although stronger, link using US data. Neither the stockmarket movements nor economic performance exhibits an empirically robust contemporaneous association with the aggregate insider trading. The theoretical background, however, did not provide any clear guidelines as to the direction and strength of this instantaneous relation. By construction, AIT A and AIT M covary strongly in the same direction. Among all variables, dIP exhibits the strongest serial dependence, which can be ascribed to the annual indexing.
Methodology The analysis of the interaction between aggregate insider trading, changes in industrial production and the real market index is embedded in a vector-autoregressive framework developed by Sims (1980b). Since all variables are treated as endogenous and no a priori restrictions are imposed, the vector autoregression model constitutes a flexible approximation of this unknown economic structure. Suppose that yt , a 3 × 1 vector, is a covariance stationary process governed by a pth-order vector autoregression: yt = 1 yt−1 + . . . + p yt−p + xt + ε t ≡ (L)yt + xt + ε t
(6.3)
where yt = (rrEWt , dIPt , AITt ); AIT t could be either or AITtM ; s and are matrices of coefficients to be estimated; xt is a matrix of AITtA
Tomasz Piotr Wisniewski 77
exogenous variables which includes a vector of ones and a time trend; p is the lag length; and ε t is a column vector of forecast errors of the best linear predictor of yt with mean 0 and variance ε . Inferences about causality are made in the spirit of Granger (1969). In particular, the null hypothesis that all p lags of the considered variable do not Granger-cause the dependent variable is tested with the F-test based on the sum of squared residuals from the restricted and unrestricted regressions. In this context, the F-test is preferred to asymptotic block-exogeneity tests, as these are likely to have inadequate empirical sizes. Tracing the cross-equation feedbacks through the inspection of parameters of the estimated equation system (6.3) can be a rather involving and laborious task. An alternative approach is to use the multiplier analysis or innovative accounting technique based on the system’s moving average representation. Given invertibility of the vector autoregression, the MA(∞) representation can be obtained by recursive substitution of the right-hand side of equation (6.3) as follows: yt = (L)xt + ε t + 1 εt−1 + 2 εt−2 + . . . ≡ (L)xt + (L)ε t
(6.4)
where (L) = (IN − 1 L − . . . p Lp )−1 with N = dim(yt ) = 3, and s is the 3 × 3 coefficient matrix of dynamic responses. The operators (L) and (L) have to satisfy the following condition: (L) = [IN − (L)]−1 ⇒ [IN − 1 L − . . . − p Lp ] [IN + 1 L + 2 L2 + . . .] = IN
(6.5)
The coefficients of Li , in the resulting lag polynomial (6.5), were set equal to zero for each i, yielding a triangular simultaneous-equation system. Following Hamilton (1994), the MA coefficient matrices are solved recursively. The impulse-response functions could be derived directly from equation (6.4). A serious drawback of this analysis would be that it considers a shock to a single variable in isolation. In practice, however, innovations in different variables are rarely independent and contemporaneous correlation of the error terms is likely to be observed. A procedure that orthogonalizes the innovations would take this covariation into account. The orthogonalizing transformation proceeds as follows. The positive-definite symmetric matrix ε can be uniquely
78 Insiders’ Market Timing and Real Activity
decomposed into GG’ using Choleski factorization: E εt εt = ε = ADA = AD1/2 D1/2 A = GG
(6.6)
where A is a square matrix whose columns are eigenvectors of the sample covariance matrix ε , D is a diagonal matrix of the corresponding eigenvalues, and G is a non-singular, lower triangular matrix with positive elements on the diagonal. A transformed innovation ut is defined as: ut ≡ G−1 εt
(6.7)
where ut is a matrix of uncorrelated components with mean 0 and variance IN . Substituting equation (6.7) into equation (6.4) and taking a partial derivative with respect to variable’s i innovation yields: ∂yt+s = s gi ∂ui,t
(6.8)
where ∂yt+s /∂ui,t is the orthogonized response of y to a one standard deviation increase in ui,t at lag s, and g i denotes the i-th column of G. The orthogonized responses of the rrEW and dIP variables are subsequently cumulated which simplifies the interpretation the results. Although the task of cumulating market returns is straightforward, the cumulative response of the real industrial production had to be computed using a recursive method due to the annual indexing. Lastly, the responses in the aggregate insider-trading index are not summed over time, because the cumulative figure would have little intuitive content. In addition to the impulse response analysis, the moving average representation (6.4) can also be used to allocate the forecast variance of each element in y to different sources of shocks, as measured by the elements of u. The error of the optimal s-step ahead forecast is: yt+s − yˆ t+s =
s−1 i=0
i εt+s−i =
s−1 i=0
i GG−1 ε t+s−i =
s−1
i ut+s−i
(6.9)
i=0
Denoting the mn-th element of i by θmn,i , the proportion of the s-step ahead forecast error variance in yj accounted for by innovations
Tomasz Piotr Wisniewski 79
in yn is: s−1 i=0
2 θjn,i
s−1 N
2 θjn,i
(6.10)
i=0 n=1
The forecast errors of a firm casual prior in the Granger sense are mostly accounted for by its own innovations rather than by the shocks in other variables in the system.
Empirical results As the order p of the data generation process described in (6.3) is unknown, the Schwarz minimum bias criterion (Schwarz, 1978; Rissanen, 1978) is used to determine it. Relatively to Akaike (1973, 1974) and Hannan and Quinn (1980), the Schwarz criterion has been shown to choose the correct autoregressive order more often and lead to a smaller forecasting error in finite samples (Lütkepohl, 1985). The Schwarz criterion indicates that the estimated optimal lag length in our model is 9, which translates into 39 degrees of freedom per each equation in the system. The results of Granger causality tests are presented in Table (6.2). A strong casual relation between the real market returns and first difference in log industrial production is found running in the direction from the former to the latter. The conclusion that swings in the stockmarket approximately capture changes in expectations of future productivity coheres with the bulk of previous literature.7 The hypothesis that the response of the stockmarket to the information about the real economy does not occur at lags is rejected only in the first model specification, and even then merely at the 10% significance level. This result does not necessarily contradict the semi-strong form of market efficiency, as the statistical data on economic activity is typically disseminated with some delay. Insider trading foreruns both the dIP and rrEW. This evidence lends credence to the cash-flow hypothesis of Seyhun (1992), for it suggests that economy-wide factors contribute to insiders’ ability to predict future market returns. Even after the information about insider transactions becomes publicly available, it is still useful to forecast future market equilibrium returns and changes in industrial production. The conclusion reached is robust, in that the statistical tests confirm
80 Insiders’ Market Timing and Real Activity
Table 6.2 Granger causality tests Dependent variable rrEW
rrEW dIP AIT A
dIP
Panel A model with all transactions 1.3199 5.2155∗∗∗ 2.0724∗ 4.6322∗∗∗ 2.3659∗∗ 2.0957∗∗
AIT A
0.8899 0.9940 2.2226∗∗
Dependent variable rrEW
rrEW dIP AIT M
dIP
AIT M
Panel B model with transactions of managers 1.0887 1.5167 4.7762∗∗∗ 1.7286 1.4151 4.6639∗∗∗ 2.5053∗∗ 1.9571∗ 1.9144∗
F -test against the null hypothesis that 9 lags of the left column variable do not Granger-cause the dependent variable. rrEW = real return on the equally weighted market portfolio; dIP = percentage change in the real industrial production; AIT A = aggregate insider-trading index (all transactions); AIT E = aggregate insider-trading index (managers’ transactions only). ∗ Significance at 10% level; ∗∗ significance at 5% level; ∗∗∗ significance at 1% level.
its validity regardless of the definition of aggregate insider trading. Although large shareholders were slightly outperformed by managers in their market-timing ability, the data on their trades can serve to formulate more accurate predictions about future macroeconomic developments. This may suggest that managers are more likely to engage in a profitable strategy of exploiting temporary deviations of stock prices from fundamentals, whereas substantial block-holders tend to invest whenever the cash-flow projections are optimistic. Aggregate insider trading can be viewed as a casual prior, for its history influences all remaining variables in the system, but itself is not Granger-caused by any factors. In particular, the null hypothesis of non-causality from the equally weighed portfolio returns is not rejected at the conventional significance levels. Consequently, the finding of Cowdhury, Howe and Lin (1993) and Iqbal and Shetty (2002) that insiders buy after stock-price decreases and sell after stock-price increases is not strongly reflected in the Polish data. Had insiders acted that way their gains would not have been much larger. The positive
Tomasz Piotr Wisniewski 81
autocorrelation coefficients of the monthly real returns series reported in Table 6.1 indicate that the profits from a negative feedback trading strategy could not have been impressive. The transmission of shocks within the system is traced by means of multiplier analysis. Figure 6.1 reports the results. To conserve space, only the simulated orthogonalized dynamic responses of the vector autoregression model with AIT A are plotted. Broadly speaking, the impulse response functions of the model specified with AIT M are quite similar. Consistent with the cash-flow hypothesis, a positive innovation in the insider-trading variable induces a rise in asset prices and the index of industrial production. The real change in the stockmarket index provoked by the shock amounts to 4.84 per cent after one year and 4.99 per cent after 18 months. When only the transactions of management are considered, the changes are more pronounced and amount to 5.59 per cent and 5.45 per cent, respectively. However, the positive reaction is confined mostly to the first year, which can be considered a relatively short period compared to other studies. This finding can most probably be attributed to differences in regulation. Short-swing profit restrictions, such as Section 16(b) of the Securities and Exchange Act of 1934 in the USA, are absent in the Polish Law on the Public Trading of Securities of 1997 and rules thereunder. Thus, the expected average holding period should be lower, as insiders will be more inclined to reap the short-run speculative gains. Previous studies concluded that US insiders are contrarian investors. Chowdhury, Howe and Lin (1993) suggest that the negative effect of past market returns on subsequent insider transactions could be ascribed to noise trading, which drives market prices away from fundamentals (see Black, 1986). If securities are priced efficiently, significant market movements can signal deviations from the intrinsic values and the mispricing is recognized by insiders. Figure 6.1g shows the response of AIT A to a one standard deviation shock in stock returns. The picture does not support the notion that insiders tend to invest in accordance with the contrarian investment model. Instead, it reinforces the results of Granger causality tests reported in Table 6.2, and earlier findings of Wisniewski and Bohl (2005) for the Polish market. One explanation of this apparent dissimilarity between Polish insiders and their US peers could rest on the different profitability of negative feedback strategies in both markets.
82
(a)
(b)
0.1 0.08 0.06 0.04 0.02 0 2
(c)
6
8 10 12 14 16 18 20 22 24 (d)
0.06 0.05 0.04 0.03 0.02 0.01 0
2
4
6
8 10 12 14 16 18 20 22 24
2
4
6
8 10 12 14 16 18 20 22 24
2
4
6
8 10 12 14 16 18 20 22 24
2
4
6
8 10 12 14 16 18 20 22 24
0.02 0.01 0 0.01
2 (e)
4
0.02 0.01 0 – 0.02 – 0.02 – 0.03 – 0.04
4
6
8 10 12 14 16 18 20 22 24 (f)
0.04
0.02
0.02
0.01
0 0
– 0.02 – 0.04
– 0.01 2
(g)
4
6
8 10 12 14 16 18 20 22 24 (h)
0.1 0.05 0
0
– 0.05
– 0.05
– 0.1 – 0.15
(i)
0.1 0.05
– 0.1 2
4
6
8 10 12 14 16 18 20 22 24
2
4
6
8 10 12 14 16 18 20 22 24
0.4 0.3 0.2 0.1 0 – 0.1
Figure 6.1 Orthogonized impulse responses of variables to shocks in equations (a) Cumulative response of rrEW to a one standard deviation shock in rrEW; (b) Cumulative response of rrEW to a one standard deviation shock in dIP; (c) Cumulative response of rrEW to a one standard deviation shock in AIT(A); (d) Cumulative response of real industrial production growth to a one standard deviation shock in rrEW; (e) Cumulative response of real industrial production growth to a one standard deviation shock in dIP; (f) Cumulative response of real industrial production growth to a one standard deviation shock in AIT(A); (g) Response of AIT(A) to a one standard deviation shock in rrEW; (h) Response of AIT(A) to a one standard deviation shock in dIP; (i) Response of AIT(A) to a one standard deviation shock in AIT(A)
Tomasz Piotr Wisniewski 83
Industrial production experiences a considerable increase seven months after the asset return innovation and eight months after the shock in the aggregate insider trading. This substantiates the view that both rrEW and dIP can be regarded as leading indicators of economic activity. Furthermore, these indicators are largely complementary, in the sense that the inclusion of one of them into the system does not completely eliminate the predictive power of the other. As can be seen from Figure 6.1d, industrial production in the second year following the one standard deviation jump in asset returns is, on average, 0.77 per cent higher than in would have been had the jump not occurred. Similarly, a positive one standard deviation shock in AIT A implies that the average next year’s production level will be 0.7 per cent higher. The industrial production growth auto-response pattern depicted in Figure 6.1e reveals that real activity tends to move in a cyclical manner. Periods of prosperity are followed by spells of economic downturn. This may be a reflection of the central bank’s policy aimed at alleviating inflationary pressures in an overheated economy. Following a beneficial productivity shock, the asset prices experience an ephemeral phase of increase, likely to be induced by the delay in dissemination of statistical data. However, the investors’ sentiment reverses after one year when the initially favourable macroeconomic climate deteriorates. The innovation accounting data in Table 6.3 indicates that the real industrial production growth is the most endogenous variable in the system, in that its own innovations account for the smallest proportion of its variance. Approximately 20 per cent of the variability in rrEW innovations is associated with shocks in insider trading. This fraction can be considered relatively large, especially compared to the result of Chowdhury, Howe and Lin (1993). A higher proportion of 24-month forecast-error variance of real market returns is attributable to innovations in aggregate insider trading than to shocks in industrial production. Finally, inspection of both panels in Table 6.3 shows that the quality of signals about future real activity received by managers and principal shareholders does not differ radically. Broadly speaking, the results of forecast error-variance decomposition reassuringly confirm the findings of Granger causality tests and impulse-response analysis.
84 Insiders’ Market Timing and Real Activity
Table 6.3 Three-variable innovation accounting Variables explained
by innovations in rrEW
dIP
AIT A
Panel A model with all transactions rrEW 66.47 13.86 dIP 40.39 40.04 AIT A 28.87 8.09
19.69 19.57 63.05
Panel B model with transactions of managers rrEW 64.54 14.88 dIP 34.46 42.49 AIT M 24.56 10.69
20.57 23.05 64.75
Note: Percentages of 24-month forecast-error variance of the left-column variables accounted for by innovations in the top-row variables. rrEW = real return on the equally weighted market portfolio; dIP = percentage change in real industrial production; AIT A = aggregate insider-trading index (all transactions); AIT E = aggregate insider-trading index (managers’ transactions only).
The model was subjected to numerous diagnostic checks. First, the adjusted multivariate portmanteau statistic of Hosking (1980) and the tests for heteroscedasticity of White (1980) indicated that residuals from the vector autoregression are independently and identically distributed. The Kolgomorov–Smirnoff tests could not reject the null of residuals normality. Second, since the results can be sensitive to the ordering of variables for orthogonalization, all order combinations have been tried and no impact on the estimates was observed. Third, another method of aggregating insider transactions was used. It can be argued that the measures defined in equations (6.1) and (6.2) are likely to be influenced by several large trades, and an alternative index based on the proportion of purchases in the total number of transactions was constructed. Nevertheless, this model specification did not produce any significant improvement over the initial fit. Lastly, a real return on the value-weighted all-share index WIG was substituted for the rrEW variable. The predictive power was slightly attenuated, which is consistent with insiders being more active in small companies.
Tomasz Piotr Wisniewski 85
Concluding remarks Insider dealing appears to be widespread on the Polish stockmarket and a more vigorous enforcement regime is needed to temper the exploitation of privileged information. Just the insider transactions that were reported to the Securities and Exchange Commission generated nearly 2 per cent of the total trading value on the Warsaw Stock Exchange. This chapter has addressed the question of whether these transactions, in total, can be used to predict stockmarket returns and future real activity. The results presented here corroborate the conclusions of Seyhun (1988a, 1992), who hypothesized that insiders observe unexpected changes in cash flows to their own companies prior to public disclosure, but are unable to discern whether these changes are due to firm-specific or economy-wide factors. However, once the shift in general business conditions is realized the prices of assets adjust accordingly. In line with this reasoning, insider trades appear to Grangercause growth in industrial production and real returns on an equally weighted market portfolio. This result is robust to various model specifications and the choice of aggregation method. At a horizon of 12 months, a one standard deviation shock in an artificially constructed measure of aggregate insider dealing induces a real rise in stockmarket prices of 4.84 per cent, which tends to be accompanied by increased productivity. Thus, the prescience of macroeconomic development contributes to the predictive power of insider trading. Furthermore, corporate managers did perform slightly better than principal shareholders in timing the market movements. The reaction of variables in the vector autoregression to innovation in insider trading is confined only to the medium term. This finding implies that informed agents have relatively short average investment horizons arising from the absence of short-swing profit restrictions in Polish law. The information contained in self-reported trades does not degenerate immediately after the trades become common knowledge, which could be attributed to the strict disclosure deadline set by the Securities and Exchange Commission (24 hours following the insider transaction). Lastly, our study has not documented that the influence of past stockmarket returns on insider sales and purchases is statistically
86 Insiders’ Market Timing and Real Activity
significant, which is in contrast to the findings of Chowdhury, Howe and Lin (1993) and Iqbal and Shetty (2002) for the US market.
Notes 1. See Lorie and Niederhoffer (1968), Jaffe (1974), Finnerty (1976), Givoly and Palmon (1985), Seyhun (1986), Rozeff and Zaman (1988), Lin and Howe (1990), Lakonishok and Lee (2001) and Del Brio, Miguel and Perote (2002). 2. Since the establishment of the Warsaw Stock Exchange in 1991, the Polish Securities and Exchange Commission forwarded to the public prosecutor 57 notifications of potential breaches of article 176 of the Act on Public Trading in Securities (disclosing and using confidential information). Nevertheless, up to the day of writing, merely one sentence has been passed. 3. An inspection of the database reveals that January 1999 marks the month when insiders started to report their trades regularly. Individuals subject to mandatory disclosure requirements are defined by law as: members/chairmen of executive and supervisory bodies, next of kin of these individuals, holders of over 5% of company shares or 10% of the total number of votes at the general meeting of shareholders (Ministry Decree 2001 Dz.U. Nr 139, poz. 1569, The Law on the Public Trading of Securities §147). Although insiders are legally obliged to file a report to the Securities and Exchange Commission within 24 hours following their transaction, this rule was found to be frequently violated in the sample. 4. For the differences in informativeness of managers’ and large shareholders’ trades see Seyhun (1986), Lin and Howe (1990), Seyhun (2000), Lakonishock and Lee (2001). 5. An equally weighted index is preferred to its value-weighted counterpart, since insider trading has been shown to be more profitable in small firms (Seyhun, 2000). As data on an equally weighted market portfolio are not available in any database, it had to be computed by the author. All of the shares quoted on the main and parallel markets were taken into consideration. 6. To avoid seasonalities in industrial production, the dIP series is indexed to the production in the same month of the previous year. 7. See, for instance, Fama (1981), Schwert (1990), Lee (1992), Choi, Hauser and Kopecky (1999) and Phelps and Zoega (2001).
References Akaike, H. (1973) ‘Information Theory and an Extension of the Maximum Likelihood Principle’, in B.N. Petrov and F. Csáki (eds), The 2nd International Symposium on Information Theory (Budapest: Akadémiai Kiadó), pp. 267–81. Akaike, H. (1974) ‘A New Look at the Statistical Model Identification’, IEEE Transactions on Automatic Control, vol. 19, pp. 716–23.
Tomasz Piotr Wisniewski 87
Bettis, J.C., Coles, J.L. and Lemmon, M.L. (2000) ‘Corporate Policies Restricting Trading by Insiders’, Journal of Financial Economics, vol. 57, pp. 191–220. Bhattacharya, U., Daouk, H., Jorgenson, B. and Kehr, C. (2000) ‘When an Event is not an Event: the Curious Case of an Emerging Market’, Journal of Financial Economics, vol. 55, pp. 69–101. Bhattacharya, U. and Daouk, H. (2002) ‘The World Price of Insider Trading’, Journal of Finance, vol. 57, pp. 75–108. Black, F. (1986) ‘Noise’, Journal of Finance, vol. 41, pp. 529–43. Bainbridge, S.M. (2000) ‘Insider Trading: An Overview’, in Encyclopedia of Law and Economics, Vol. III, pp. 772–812. Chen, N., Roll, R. and Ross, S. (1986) ‘Economic Forces and the Stock Market’, Journal of Business, vol. 59, pp. 383–403. Choi, J.J., Hauser, S. and Kopecky, K.J. (1999) ‘Does the Stock Market Predict Real Activity? Time Series Evidence from the G-7 Countries’, Journal of Banking and Finance, 23, pp. 1771–92. Chowdhury, M., Howe, J.S. and Lin, J. (1993) ‘The Relation between Aggregate Insider Transactions and Stock Market Returns’, Journal of Financial and Quantitative Analysis, vol. 28, pp. 431–7. Del Brio, E.B., Miguel, A. and Perote, J. (2002) ‘An Investigation of Insider Trading Profits in the Spanish Stock Market’, Quarterly Review of Economics and Finance, vol. 42, pp. 73–94. Fama, E.F. (1981) ‘Stock Returns, Real Activity, Inflation, and Money’, American Economic Review, vol. 71, pp. 545–65. Finnerty, J.E. (1976) ‘Insiders and Market Efficiency’, Journal of Finance, vol. 31, pp. 1141–8. Givoly, D. and Palmon, D. (1985) ‘Insider Trading and the Exploitation of Inside Information: Some Empirical Evidence’, Journal of Business, vol. 58, pp. 69–87. Granger, C.W.J. (1969) ‘Investigating Casual Relations by Econometric Models and Cross Spectral Methods’, Econometrica, vol. 37, pp. 424–38. Hamilton, J.D. (1994) Time Series Analysis (Princeton: Princeton University Press) pp. 257–349. Hannan, E.J. and Quinn, B.G. (1979) ‘The Determination of the Order of Autoregression’, Journal of the Royal Statistical Society, vol. 41, pp. 190–5. Hosking, J.R.M. (1980) ‘The Multivariate Portmanteau Statistic’, Journal of the American Statistical Association, vol. 75, pp. 602–8. Iqbal, Z. and Shetty, S. (2002) ‘An Investigation of Causality between Insider Transactions and Stock Returns’, Quarterly Review of Economics and Finance, vol. 42, pp. 41–57. Jaffe, J. (1974) ‘Special Information and Insider Trading’, Journal of Business, vol. 47, pp. 410–28. John, K. and Lang, L. (1991) ‘Strategic Insider Trading around Dividend Announcements: Theory and Evidence’, Journal of Finance, vol. 46, pp. 1361–89.
88 Insiders’ Market Timing and Real Activity
Ke, B., Huddart, S. and Petroni, K. (2003) ‘What Insiders Know about Future Earnings and How They Use it: Evidence From Insider Trades’, Journal of Accounting and Economics, vol. 35, pp. 315–46. Lakonishok, J. and Lee, I. (2001) ‘Are Insider Trades Informative?’, Review of Financial Studies, vol. 14, pp. 79–111. Lee, B. (1992) ‘Causal Relationships Among Stock Returns, Interest Rates, Real Activity, and Inflation’, Journal of Finance, vol. 47, pp. 1591–603. Lee, D.S., Mikkelson, W.H. and Partch, M.M. (1992) ‘Managers’ Trading Around Stock Repurchases’, Journal of Finance, vol. 47, pp. 1947–61. Lin, J. and Howe, J. (1990) ‘Insider Trading in the OTC Market’, Journal of Business, vol. 45, pp. 1273–84. Lorie, J.H. and Niederhoffer, V. (1968) ‘Predictive and Statistical Properties of Insider Trading’, Journal of Law and Economics, vol. 11, pp. 35–51. Lucas, R.E., Jr. (1973) ‘Some International Evidence on Output-Inflation Tradeoffs’, American Economic Review, vol. 63, pp. 326–34. Lucas, R.E., Jr. (1975) ‘An Equilibrium Model of the Business Cycle’, Journal of Political Economy, vol. 83, pp. 1113–44. Lütkepohl, H. (1985) ‘Comparison of Criteria for Estimating the Order of a Vector Autoregressive Process’, Journal of Time Series Analysis, vol. 6, pp. 35–52. Mehra, Y.P. (1978) ‘Is Money Exogenous in Money-Demand Equations’, Journal of Political Economy, vol. 86, pp. 211–28. Phelps, E. and Zoega, G. (2001) ‘Structural Booms. Productivity Expectations and Asset Valuations’, Economic Policy, vol. 32, pp. 83–126. Rissanen, J. (1978) ‘Modeling by Shortest Data Description’, Automatica, vol. 14, pp. 465–71. Rozeff, M.S. and Zaman, M.A. (1988) ‘Market Efficiency and Insider Trading: New Evidence’, Journal of Business, vol. 61, pp. 25–44. Schwarz, G. (1978) ‘Estimating the Dimension of a Model’, Annals of Statistics, vol. 6, pp. 461–4. Schwert, G.W. (1990) ‘Stock Returns and Real Activity: A Century of Evidence’, Journal of Finance, vol. 45, pp. 1237–57. Seyhun, N.H. (1986) ‘Insiders’ Profits, Costs of Trading, and Market Efficiency’, Journal of Financial Economics, vol. 16, pp. 189–212. Seyhun, N.H. (1988a) ‘The Information Content of Aggregate Insider Trading’, Journal of Business, vol. 61, pp. 1–24. Seyhun, N.H. (1988b) ‘The January Effect and Aggregate Insider Trading’, Journal of Finance, vol. 43, pp. 129–41. Seyhun, N.H. (1990) ‘Do Bidder Managers Knowingly Pay too Much?’, Journal of Business, vol. 63, pp. 439–64. Seyhun, N.H. (1992) ‘Why Does Aggregate Insider Trading Predict Future Stock Returns?’, Quarterly Journal of Economics, vol. 107, pp. 1303–31. Seyhun, N.H. and Bradley, M. (1997) ‘Corporate Bankruptcy and Insider Trading’, Journal of Business, vol. 70, pp. 189–216. Seyhun, N.H. (2000) Investment Intelligence from Insider Trading (Cambridge, MA: MIT Press).
Tomasz Piotr Wisniewski 89
Sims, C.A. (1980a) ‘Comparison of Interwar and Postwar Business Cycles: Monetarism Reconsidered’, American Economic Review, vol. 70, pp. 250–7. Sims, C.A. (1980b) ‘Macroeconomics and Reality’, Econometrica, vol. 48, pp. 1–46. White, H. (1980) ‘Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroscedasticity’, Econometrica, vol. 48, pp. 817–38. Wisniewski, T.P. and Bohl, M. (2005) ‘The Information of Registered Insider Trading under Lax Law Enforcement’, forthcoming in the International Review of Law and Economics.
This page intentionally left blank
7 Trading Risk Management: Practical Applications to Emerging Markets Mazin A.M. Al Janabi
Introduction The global deregulation of financial markets has created new investment opportunities, which in turn require the development of new instruments, regulations and efficient risk-management policies/procedures to cope with increased risks. Nonetheless, many disastrous financial crises have hit several financial and non-financial corporations in emerging economies; even so, the developments and innovations in cash-markets instruments and derivative products are on a continuous growth path. Emerging countries and markets, since the early 1990s, have started to play an important role in standardized and over-the-counter (OTC) derivatives and cash-markets. Yet while emerging-market countries share some similarities in development patterns, it is often their individual differences that create unique opportunities and risks that may be addressed through derivative structures and sound risk-management practices. Trading of financial instruments – stocks (equities), bonds (fixed income instruments), derivative products, and structured products, and so on – has been on uninterrupted expansion in emerging economies. Although these markets are characterized in general as illiquid, volatile and politically unstable, the potential of high expected rewards are tremendous and so are the vast unforeseen risks. The role of trading risk management and its proper implementations are essential factors in the success of emerging-markets’ financial trading activities. The techniques of modern risk management are widely regarded as new, perplexing and difficult to implement, the use of which is 91
92 Trading Risk Management
the province of ‘rocket scientists’ and mathematics PhDs armed with powerful computers. The previous statement is partly true! Modern financial risk-management techniques are certainly not new; they have existed for a long time and one can trace their roots to elementary concepts in statistics such as the normal distribution (bell-shaped curve) and probability theory. What is new in modern financial risk management is the adaptation of these techniques to new financial instruments such as complex derivative products and structured instruments. But risk management is not all about mathematics and formulas! In fact, complex mathematical concepts, formulas and approaches can themselves lead to new categories of risk–for instance model risk. The risk-management process must include all accompanying elements such as legal, operations, settlements, regulations and accounting. All these elements are essential for the proper identification, measurements, control and management of all categories of risks. To have a choice between a certain loss and a speculation with cash-markets or derivative instruments, organizational objectives and decisions should be set utilizing modern financial risk-measurement tools to estimate worst-case scenarios. Thereafter, the level of the measured risk should be compared with the organization’s risk appetite, with the objectives to ascertain if the risk falls within its risk limits, and also to reveal if there is a sufficient economic capital cushion to withstand unforeseen surprises. What is most needed is a better understanding of the trading riskmanagement process. This can be accomplished by establishing a number of institutional changes that will help reduce the uncertainties in the trading of securities. In the rapidly changing and increasingly integrated financial markets, better management and closer supervision of the trading positions being taken (and their trading units) will better ward against hidden risks than formal regulations that focus on particular instruments, markets or participants. Naturally this has to be accompanied with a clearer legal environment, risk management and accounting standards, in addition to greater disclosure of trading transactions. In the 1950s Harry Markowitz (1959) described the theoretical framework for modern portfolio theory and the creation of efficient portfolios. The solution to the Markowitz theoretical models revolves around the portfolio weights, or the percentage of asset allocated to
Mazin A.M. Al Janabi 93
be invested in each instrument. William Sharpe (1963) developed the single-index model, which relates returns on each security to the returns on a common index – a broad market index of common stock returns such as the S&P 500 is generally used for this purpose. The concepts of Value at Risk (VaR) and other advanced risk-management techniques are in fact not new, and are based – with some modifications – on modern portfolio theory. Thanks to J.P. Morgan, RiskMetrics™ (1994) document, the concept of VaR and other modern risk-management techniques and procedures were popularized. Since then the VaR concept has become widespread and several specific applications adapted to credit risk management and mutual funds investments. The objectives of this chapter are to provide practical and robust guidelines, procedures and measurements of trading risk (frequently it can be called market risk or price risk) for emerging-markets’ equity-trading portfolios, and also to assist these countries in the establishment of sound risk-management practices within a prudential framework of rules and policies. To this end, the parameters required for the construction of appropriate and simplified Value at Risk and stress-testing methods are defined in this work and refined to the specific applications of these methods to emerging markets. The theoretical mathematical/analytical models that are developed herein are based on a matrix-algebra approach. The latter approach can in fact simplify the programming process in Excel™ worksheets and can also permit easy incorporation of short selling of assets in the equity trading process. Moreover, a simplified approach for the incorporation of illiquid asset, in daily trading risk-management practices, is defined and is appropriately integrated into the VaR and stress-testing models. Trading risk-management models, which are developed in this work, were applied to the Moroccan stockmarket. Databases of daily stocks’ prices and MADEX index levels were all downloaded from the Moroccan stock exchange website. Several case studies were carried out with the objectives of calculating VaR numbers under various possible scenarios in addition to the inception of a practical framework for the establishment of VaR limits-setting. The different scenarios were performed, first, with distinct asset allocation percentages, second by studying the effects of liquidity of trading assets (unwinding period of assets), and finally by taking into account the possibilities of
94 Trading Risk Management
short-selling in daily trading operations. Furthermore, several tests of abnormal (asymmetric) distributions of returns were performed, and to this end various tests of skewness and kurtosis were implemented. This was followed by a study of daily and annual volatilities along with the calculations of betas of the sample stocks.
Predominant characteristics of emerging markets During the 1990s, certain developing countries liberalized their economics and unlocked their financial markets, thereby gaining promotion to the status of emerging-market economies. These emerging markets were attractive to international investors – mainly international portfolio investors – principally due to their expected growth potential, and hence huge amounts of private capital have been flowing to these countries. Nevertheless, foreign money has valuable as well as bad aspects. Emerging markets must have the means to absorb these funds productively. Emerging markets have a series of characteristics that require different risk-management approaches than those of developed countries. Some of these characteristics affect risk-measurement methodologies, while others affect the implementation process, as follow: 1 The instruments traded are in many cases insufficient for the establishment of adequate benchmarks for the valuation of certain transactions. The lack of quotation of long-term government fixedrate bonds, for instance, complicates the setting of interest rates on long-term loans to corporations and also the valuation (mark-tomarket) of previously granted loans. Consequently, this leads to a lack of benchmarks for risk estimation (that is, the probability of a loss of a certain value of the portfolio). On the other hand, low trading volumes (illiquid markets) and missing historical data for many financial instruments create doubts regarding the validity of the quoted prices. 2 These markets are also characterized by frequent government interventions in the financial markets to stabilize the short-term impact of a current event. Measurement of risk on the short end will divulge a relatively risk-stable market and it might give the wrong message – since the impact of the current crisis is diluted with government interventions. However, once risk is measured on a long horizon of
Mazin A.M. Al Janabi 95
time, one should not be surprised to find big swings in the level of risk. In these situations, risk calculations require estimations of the potential fall in financial markets in case of a potential crisis or event. Techniques such as stress-testing and event risks under severe market conditions (assuming all financial assets are correlated positively) are more adequate in this case than market-risk measures – such as Value at Risk (VaR) – which assumes normal distributions of financial assets’ returns. 3 Finally, emerging countries are characterized by lower investment in information technology systems that can handle front office (trading desks), middle office (risk management and legal/regulatory) and back office (settlement and accounting). Besides, they lack personnel with technical backgrounds in economics and finance, such as traders, quantitative analysts and risk managers.
Trading risk management and regulations The demands for risk-management instruments and processes by emerging countries are now large and will continue to grow throughout the twenty-first century. This will put more burdens on banking regulators and supervisors to ensure the safety and soundness of their respective financial systems through effective regulations. This, of course, is a difficult task, as evidenced by the variety of financial crises over the last century. Looking forward the task is likely to get harder, given the likely increase in the complexity of financial instruments and the magnitude of cross-border financial flows. Originating and demanding compliance of rational ‘rules-of-the-game’ is a challenging endeavour for any financial/banking regulator.
Emerging markets and prudential regulations The regulatory challenge is even more difficult and more important to meet in emerging-market economies for several reasons: • The market structure of banking and financial activities is concentrated in just a few major institutions, on which the stability of the whole macro-economy depends. Local financial markets are often characterized as thin, illiquid, lacking information technology infrastructures, and severely volatile, making it even more
96 Trading Risk Management
difficult for local institutions to manage their risks effectively. In some countries, banks and other financial intermediarys’ functions are conducted and interlined with other corporate entity shareholders, creating severe moral-hazard problems. • The political structure and government policies of ensuring stability of the financial system are weak and less-developed, and these markets are characterized with frequent government interventions to stabilize the short-term impact of current events. Banks and other financial institutions may have a high degree of political influence in their countries, but only a limited understanding and acceptance of the needs for independent regulation and supervision. • Financial sophistication for the valuations of complex instruments and reporting of exposure are weak and less stringent than in advanced economies. Additionally, accounting standards vary widely from market to market. Financial-entity management and regulatory-body supervisors are less trained in advanced methods for the identification, measurement, management and control of financial risks. • The lack of adequate historical and current databases for most of these countries’ macroeconomic variables can complicate the logistics of an effective and integral risk-management procedure. Little real progress can be made without good databases and it will take considerable effort to assemble them. Risk-management systems are expensive to create and to run without adequate current and historical databases of most of the markets’ main indicators. Regulations for the trading of securities are essential requirements, and the lack of reinforcement of these requirements had led to many financial crises within the financial industry. These regulations have to include several items such as accounting, the legal environment, risk management, operations, pricing and valuation, and settlement processes. Local regulators and financial entities should give special emphasis to regulating ‘operational-risk’. The complexity and fast pace of developments in the cash and derivatives markets put particular pressure on the control function to keep up. If it does not keep up, the financial entity will be susceptible to both inadvertent errors and internal
Mazin A.M. Al Janabi 97
fraud. Admittedly, there is no substitute for the human element in any risk-management process. Financial institutions must first strive to hire responsible traders who understand and analyse what they are doing, as well as to recognize that many traders (if left alone without adequate supervision), have the propensity and incentive to break the rules. Thus, at this stage, management must have capable monitoring and enforcement of trading rules. At the very least, this requires autonomous reporting of a trader’s activity, and an eagerness to question the causes of extraordinarily profitable activity and to inquire why a trader is betting with high-risk instruments or whether he is above the risk-appetite limits of the institution.
Trading in securities and risk management Trading consists of proprietary positions in financial instruments which are held for resale (available for sale) and/or which are taken on by the financial entity with the intention of benefiting from actual and/or expected differences between purchase and sale prices, or from other price variations (such as spread differentials). Trading risk is defined as the risk that the trading income will decrease due to an adverse price change in the traded financial instrument. Trading risk management is a unit within global risk management that is responsible for monitoring all risks related to ‘proprietary trading’ of the financial institution. Trading risk deals with risk within a short-term time-horizon positioning, where all trading positions are marked-to-market and risk is re-evaluated on a daily basis and performance is measured via daily profit and loss and impacts immediately the financial institution’s income statement. The trading risk management unit is responsible for the measurement and management of several categories of risks, which include: • Market risk: the risk of loss due to changes in risk factors (for example rises or falls in prices, rates or indices). Since such losses occur when an adverse price movement causes a decrease in the mark-to-market value of a position, market risk can also be referred to as ‘position risk’. In essence, market risk measurement/management involves observing the sensitivity of the market value of a portfolio to changes in financial markets, and then determining whether exposure to market changes is within acceptable limits. The key to measuring market risk is the concept
98 Trading Risk Management
of Value at Risk (VaR). This measure is moving into mainstream finance with startling speed and becoming the industry standard for measuring market risk. With this method it is important to measure market risks over an entire portfolio of instruments; otherwise, the measurement will ignore the correlations’ benefits between individual positions and consequently overstate the overall market risk. VaR techniques not only make sure that positions are truly diversified and with adequate capital on hand; they also tell us when we are leaving money on the table by being unnecessarily cautious. • Event risk: the risk of a loss due to extreme changes in risk factors caused by an unforeseen political, economic or other event, which affects the value of the financial contract. The question that may arise is how to manage risks related to abnormal market events? Stress-testing and scenario-analysis are the most common approaches to simulate the effect of unusual market movements, and these approaches are an important part of market risk management since they help users to develop plans for coping with such situations, especially for emerging-markets. • Issuer risk: the risk of a loss due to a change in an issuer’s credit rating or the market’s perception of the issuer’s credit, which results in a reduction in the value of the issuer’s debt or equity trading asset. • Counterparty risk: the risk of suffering a loss due to a counterparty’s inability to perform under the terms of the financial contracts. This risk is also commonly called credit risk because it relates to the creditworthiness of the counterparty; and can arise due to changes in the counterparty’s ability, intention or legal or regulatory obligation to honour a financial contract, as well as the risk that documentation is not adequate to enforce the counterparty’s obligation. Assessment of the magnitude of counterparty risk seeks to address the following questions: (1) How much does the financial entity stand to lose if the counterparty defaults during the exposure period? (2) What is the likelihood that the counterparty will default during the exposure period? The key components of counterparty
Mazin A.M. Al Janabi 99
risk are the exposure to counterparty default, the probability of default, and recovery rates. • Country risk: the risk of inconvertibility of a currency due to government actions curtailing the counterparty’s ability to meet its obligations. • Liquidity risk: the risk that the financial institution will not have sufficient funding at any period (funding risk), or the risks associated with the market liquidity of financial assets (which is considered as part of market risk). • Legal and compliance risk: the risk of a loss as a result of changes in the legal or regulatory environment or insufficient documentation. • System and operational risk: the risk of a loss due to internal and/or external operational or systems related to inefficiencies or problems (for example trade processing errors).
Implementation of internal controls Implementing internal controls requires organizing for risk management, establishing policies and procedures, and communicating these policies and procedures. It is also necessary to decide on how extensive a risk-management system is required to meet a firm’s level of activity and potential exposures. Establishing internal risk-management controls can include – but is not limited to – the following: • Setting up the organizational structure. The foundation for establishing an effective control system is an organizational structure that ensures adequate supervision of risk-management activities, appropriate segregation of duties between departments, and proper reporting of positions and relevant risks. In developing organizational structures, senior management should identify which areas or departments will be involved in the risk-management programme, document their duties, and define the roles and responsibilities of each of those areas/departments. • Originating policies and procedures. Prior to generating trading/hedging activities and risk-management programmes, each department should undertake and document policies and procedures related to the entity’s risk-management activities. Senior management should approve these policies and procedures. In
100 Trading Risk Management
some situations, senior management may want to designate a risk-management committee made up of senior officers, which commands and approves all policies and procedures. • Communicating the risk-management policy. For internal controls to be as effective as possible, senior managers and the involved departments should communicate risk-management policies and strategies clearly and consistently. The risk-management strategy should be reasonably detailed – often risk-management policies are too general, a policy might say that the entity may hedge, but does not define precisely what hedging entails or identify what hedging instruments the entity is allowed to trade or to use. Risk-management strategies should be reviewed and approved by senior management and the board of directors, and periodic meetings with senior management should be held to review the risk-management strategy. Senior management and the board should approve major changes to risk-management policies. Further, the risk-management policy should establish when the treasury department requires senior management approval to use new derivatives/structured products to manage risk. • Further considerations. In many organizations there is a natural tension between the treasury department, which takes positions and manages risk (for instance, on interest rate, foreign exchange and equities), and the controller which reports the accounting results of the treasury department’s trading/hedging activities. The treasury department takes positions, measures exposures, hedges relevant risk and reports performance based on its own reports and records, while the controller is left with the task of reconciling trades with exposures and making entries into the general ledger. One of the common sources of tension is that financial statements do not reflect the gains that the treasury department personnel are reporting. To evade or lessen this natural strain, it is important that treasury and accounting staff work together to agree on performance-evaluation benchmarks for the entity’s firm-wide risk-management programme.
A framework for the calculation of value at risk using the variance/covariance method After a series of big losses in some of the world’s largest financial and non-financial institutions, the vital importance of a systematic
Mazin A.M. Al Janabi 101
approach to trading risk management, control and reporting was established. One of the key concepts of modern risk management is Value at Risk (VaR), which represents an attempt to quantify, with a specified confidence interval, the maximum potential loss for a given trading position over a short period under ‘normal’ market conditions. Calculating VaR figures is progressively becoming a standard procedure for risk management worldwide. In 1994, VaR techniques were mainly used by a handful of international banks and brokerage firms to manage risks on their trading desks, but now the use of VaR is expanding in other dimensions. Currently, it is being used for risks other than market movements, such as the risk of default, and it is also being used by more market participants other than traders, including mutual fund managers and even chief financial offices (CFOs) of non-financial entities. What is new with the development of the VaR method is the quantification of firm-wide, cross-product risk exposures and the extensive use of modern statistical techniques and concepts in the risk-measurement process. These new developments make the VaR approach more powerful than conventional approaches. One of the primary benefits of VaR analysis as opposed to other riskassessment tools is that it can measure the price (market) risks across all types of markets and then distil them down to a single number. This enables those who manage or oversee portfolios containing, for example, both fixed income and equity securities, to examine the price risk in all positions simultaneously because the same methodologies are used across all markets. The Bank for International Settlements (BIS), the Basle Committee and many central banks now set capital-adequacy requirements for market and other kinds of risks in terms of a bank’s own VaR estimates. This means that institutions should be able to allocate their capital to the most profitable business areas on a risk-adjusted basis. To be acceptable to regulators for the purposes of allocating capital, banks’ VaR models (internal models) must meet certain qualitative and quantitative standards. Fundamentally, qualitative standards relate to the institution’s risk-management function as a whole. They demand autonomous validation of the models by both internal and external auditors; effective control over inputs, databases and modification of models; independence of the risk-management function from business lines; full integration of the model into risk management; and, most important, director and senior management
102 Trading Risk Management
supervision of the risk-management process and relevant procedures/ policies.
What is value at risk (VaR)? VaR is a method of assessing market risk that uses standard statistical techniques routinely used in other technical fields. Formally, VaR measures the worst expected loss over a given time interval under normal market conditions at a given confidence level. Consequently, VaR alerts you to the maximum loss that your portfolio (investment or trading portfolios) could experience so you can evaluate such a loss’s potential on your business. The standard deviation of the daily rate of return is used as an estimation of the potential loss or gain the firm may incur. Assuming the return of a financial product follows a normal distribution, linear pay-off profile and a direct relationship between the underlying product and income, the VaR is a measure of the standard deviation of the income for a certain confidence level. In reality, the VaR is a forecast of the standard deviation. Although the method relies on several assumptions, it has gained wide acceptance for the quantification of financial risks. As a result of the generalization of this method, capital allocations for trading activities tend to be calculated and adjusted with the VaR method. A bank might say that the ‘daily’ VaR of its trading portfolio is $1 million at the 99 per cent confidence level. This means there is only 1 chance in a 100 (or one day in every 100 trading days), under normal market conditions, for a loss greater than $1 million to occur. This single number summarizes the bank’s exposure to market risk as well as the probability of an adverse move. Equally important, it measures risk using the same units as the bank’s bottom line – dollars. Shareholders and managers can then decide whether they feel comfortable with this level of risk. If the answer is no, the process that led to the computation of VaR can be used to decide where to trim the risk. While this method is powerful for day-to-day risk-management, it is no substitute for the wider risk-management process of analysing crash scenarios and keeping control of operational and legal risks. In order for this method to perform properly, accurate trading positions should be gathered and, correspondingly, a historical database of these positions should be built. Once the position data are centralized, the overall risk has to be calculated by aggregating the risks from
Mazin A.M. Al Janabi 103
individual contracts across the whole portfolio. This is done by working out the effect of moves in individual ‘risk factors’ (for example equities, money-market instruments, foreign exchange rates) across the portfolio, which may involve large currencies and, with each currency, different asset classes. VaR is worked out from the relationships between the individual risk factors and the effect on the portfolio of moves in each risk factor.
Implementation of VaR using the variance/covariance method So far, there is no industry consensus on the best method for calculating VaR. As with any statistical model, VaR depends on certain assumptions, and the choice of which method of calculation is used is normally dictated by the user’s aversion to unrealistic or over-simplistic assumptions. There are three popular methods: the ‘variance/covariance’ method (also known as the ‘correlation’ or ‘parametric’ method), the ‘historical simulation’ method and the ‘Monte-Carlo simulation’ method. Each of these methods has its own set of assumptions and each is a simplification of reality. The variance/covariance method is the simplest in terms of application to financial practices and computer time consumption. This method assumes that the returns on risk factors are ‘normally distributed’ and the correlations between risk factors are constant. For risk-management purposes, using the normal distribution assumption is generally considered to be acceptable. Deviation from normality usually does not significantly alter the results of the VaR calculations under normal market conditions. Within this method a bellshaped curve (Gaussian distribution) is essentially assumed and it also assumes that extreme price swings, such as market crashes, occur too rarely to contribute to an accurate picture of the likelihood of future events. To calculate VaR using the variance/covariance method, the volatility of each risk factor is extracted from a predefined historical observation period. The potential effect of each component of the portfolio on the overall portfolio value is then worked out. These effects are then aggregated across the whole portfolio using the correlations between the risk factors (which are, again, extracted from the historical observation period) to give the overall VaR value of the portfolio with a given confidence level.
104 Trading Risk Management
Many financial institutions have chosen a confidence interval of 95 per cent (or 97.5 per cent if we only look at the loss side [onetailed]) to calculate VaR. The 97.5 per cent interval means that once every 40 trading days a loss larger than indicated is expected to occur. Some banks use a 99 per cent (one-tailed) confidence interval, which would theoretically lead to larger loss once every 100 trading days. However, due to fat tails of the probability distribution, such a loss will occur more often. Some financial institutions feel that the use of a 99 per cent confidence interval would place too much trust on the statistical model and, hence, some confidence level should be assigned to the ‘art-side’ of the risk-measurement process. Really, the choice of the confidence level also depends on its use. If the resulting VaRs are directly used for the choice of a capital cushion, then the choice of the confidence level is crucial, as it should reflect the degree of risk-aversion of the firm and the cost of a loss of exceeding the calculated VaR numbers. The higher the risk-aversion or the greater the costs, implies that a big amount of capital should be set aside to cover possible losses, and this consequently will lead to a higher confidence level. In contrast, if VaR numbers are only used to provide a firm-wide yardstick to compare risks among different portfolios and markets, then the choice of confidence level is not that relevant. A simplified calculation process of the estimation of VaR risk factors (using variance/covariance method) for single and multiple asset positions will now be illustrated. VaR of a single asset position VARi = α ∗ Value of positioni in dollars ∗ σi where α is the confidence level and σi is the standard deviation (volatility) of the security that constitutes the single position. The value of the position is the amount of investment in dollars, of instrument i. VaR of multiple assets positions In this case, VaR takes correlations fully into account and the formula is much more complicated as follows. One can first consider the simplest possible case, that of two securities or assets: σAB = wA σA2 + wB σB2 + 2ρAB wA wB σA σB (7.1)
Mazin A.M. Al Janabi 105
This formula is the natural product of portfolio theory as proposed by Markowitz’s (1959) model of portfolio management for a twosecurities (A and B) portfolio, where σA and σB are the standard deviation of each security, ρ is the correlation factor between the returns of the two securities, and wA , wB are the percentage weights of each security in the portfolio. Since VaR is presented in terms of dollars (and not in percentage terms as given in portfolio theory), the weights in the above formula are cancelled out and are replaced with the individual VaRs: VARAB =
VAR2A + VAR2B + 2ρAB VARA VARB
(7.2)
This two-security special case can be generalized to n securities. Combining assets with less than perfect positive correlation can reduce portfolio VaR, and furthermore, the smaller the positive correlation the better the diversification and the lower is the VaR. Portfolio VaR is a function of each individual security’s risk and the correlation between the returns on the individual securities: VARP =
|VAR| ∗ |ρ| ∗ |VAR|T
(7.3)
This formula is a general one for the calculation of VaR for any portfolio regardless of the number of securities. It should be noted that this formula is presented in terms of matrix algebra – a useful form to avoid mathematical complexity, as more and more securities are added. This approach can in fact simplify the programming process in Excel™ worksheets and can also permit easy incorporation of short-selling in the trading risk-management process. This means, in order to calculate the VaR (of a portfolio of any number of securities), one first needs to create a matrix of the individual VaR positions, a transpose matrix (indicated above by the letter ‘T ’ on the top of the matrix) of the individual VaR positions, and finally a matrix of all correlation factors. Multiplying the three matrices and then taking the square root of the result, yields the VARP of any portfolio with n number of securities. This simple number summarizes the portfolio’s exposure to market risk. Investors and senior managers can then decide whether they feel at ease with this level of risk. If the answer is no, then the process that led to the estimation of VaR can be used to decide where to reduce redundant risk. For instance, the
106 Trading Risk Management
riskiest securities can be sold, or one can use derivative securities such as futures and options to hedge the undesirable risk.
A model for the incorporation of liquidity of traded assets The choice of the ‘time-horizon’ or number of days to liquidate (unwind) a position is a very important factor and has a big impact on VaR numbers, and it depends upon the objectives of the portfolio and the liquidity of its positions. For a bank’s trading portfolio invested in highly liquid currencies, a one-day horizon may be acceptable. For an investment manager with a monthly re-balancing and reporting focus, a 30-day period may be more appropriate. Ideally, the holding period should correspond to the longest period for orderly portfolio liquidation. In fact, if one assumes a normal distribution, then we can convert the VaR horizon parameter from daily to any t-day horizon. The variance of a t-day return should be t times the variance of a 1-day return or σ 2 = f (t). Thus, in terms of standard deviation (or volatility), σ = f (t 1/2 ) and the daily VaR number can be adjusted for any horizon as: VAR(t-day) = VAR(1-day) ∗
√ t
(7.4)
The above formula was proposed and used by J.P. Morgan in their earlier RiskMetrics™ method (1994). Unfortunately, the latter approach does not consider real-life trading situations where traders can liquidate (or re-balance) small portions of their trading portfolios on a daily basis. To this end, a practical framework of a procedure/methodology within a simplified mathematical approach is proposed below with the purpose of incorporating and calculating illiquid assets’ daily VaR, as follows. In order to take into account the full illiquidity of assets (that is, the required unwinding period to liquidate an asset) we can define the following: t = number of liquidation days (t-day to liquidate the entire asset fully); and σ 2 = overnight (daily) variance.
Mazin A.M. Al Janabi 107
A linear liquidation procedure of the asset is assumed (that is, selling equal parts of each asset every day till the last trading day, t, where the entire asset is sold), and hence the following can be achieved: t 1 2 3 1 σ2 = f or σ 2 = f (1 + 2 + 3 + · · · + t) + + + ··· + t t t t t (7.5) From infinite series in mathematics the following relationship can be obtained: t(t + 1) (1 + 2 + 3 + · · · + t) = 2 1 (1 + 2 + 3 + · · · + t) σ2 = f t Consequently: 1 (1 + 2 + 3 + · · · + t) σ =f t
or σ 2 = f
t +1 2
or
σ =f
t +1 2
(7.6)
(7.7)
The final result is of course a function of time and not the square root of time as employed by some financial market participants based on the RiskMetrics™ methodologies. The above approach can also be used to calculate the VaR for any time horizon. In order to perform the calculation of VaR under illiquid market conditions, we can define the following: VAR = Value at Risk under liquid market conditions: VARadj = VAR ∗
t +1 2
(7.8)
And the latter equation indicates the following: VARadj VAR
(7.9)
Consequently, the difference (VARadj − VAR) should be equal to the residual market risk due to the illiquidity of any asset under illiquid markets conditions. The number of days required to liquidate a position (of course, depending on the type of security) can be obtained from the various publications dealing with ‘capital markets’ and can be compared with the assessments of the individual traders of each trading desk.
108 Trading Risk Management
Hence, one can create some simple statistics of the volume which can be liquidated on a daily basis of each instrument (fixed income, equities, FX, derivatives, and so on) and the necessary total number of days to unwind the whole volume.
Major limitations of the VaR method • Value at Risk is now one of the essential tools of risk management but it is not the whole story. Its purpose is to give an estimate of losses over a short period under ‘normal market’ conditions. It is not going to tell us what might happen during a market crash. VaR estimations cannot be taken as gospel, since they are typically based on historical patterns that are not always a good guide to the future – especially in times of turmoil. For that reason most financial entities supplement the analysis of VaR with other tools such as stress-testing and scenario analysis to grasp a better picture of hidden unexpected events. • The main assumption underpinning VaR, which is also one of the concept’s main drawbacks, is that the distribution of future price (or rate) changes will be similar to that of past price variations. That is, the potential portfolio loss calculations for VaR are worked out using distributions or parameters from historic price data in the observation period. • The assumption that asset returns are normally distributed may underestimate potential risk due to ‘fat tails’ in the distribution of returns. For this reason it will be useful to check the validity of the normality assumption on different assets through other parameters such as skewness (a measure of asymmetry) and kurtosis (a measure of flatness/peakedness) and to carry out scenario analysis to fully understand the impact of extreme moves. The VaR methodology is more appropriate for measuring the risk of cash instruments (with linear payoffs) such as equities and bonds. In dealing with complex instruments (with non-linear payoffs) such as derivatives, the method might not give reasonable answers and may mis-state non-linear risks as in the case of options contracts. • Correlation assumptions in emerging markets must be taken seriously, as correlation can break down or even change signs. Correlation assumptions can be either explicit or implicit. In some ways
Mazin A.M. Al Janabi 109
the implicit assumptions are more dangerous in that they are more easily overlooked. A typical implicit assumption is a correlation of either zero or one. For example, some emerging-market currencies are pegged to the US dollar and one can assume the correlation is very close to one. This is not really a statistical fact, but rather reflects a policy decision that could change abruptly. • Value at Risk does not calculate standard deviations, but rather estimates what they may be in the immediate future. The impact of market volatility and how to forecast its effects is a crucial issue for emerging markets that are characterized by a low level of liquidity. The estimation of statistical parameters such as the volatility of a stock requires a time series of market data. This can be particularly troublesome in markets in which the underlying stock experiences only sporadic bursts of trading volume. While techniques have been developed to account for this, the net result is that a lack of liquidity reduces confidence in the forecasted volatility, which is an essential tool for the pricing of options contracts.
Stress-testing and scenario analysis Previous examples of the limitations of the VaR method that we have cited are intended to make the point that traditional risk measures and the VaR method do not provide a complete risk profile. There may be hidden assumptions that fail to hold true, including transactions levels and correlations. Explicit assumptions can also turn out to be drastically amiss in the event of a sudden change in market conditions. It becomes critical to amplify one’s risk profile by considering the effect of alternative risk scenarios including currency devaluation, defaults and flights-to-quality. Since it is difficult to anticipate every possible such scenario, it is useful to create realistic meaningful ones and to examine trading positions under each one, beside bearing in mind that hindsight is of course useful through learning from other’s misfortunes. As described earlier, the VaR method is only one approach to measuring market risk and is mainly concerned with maximum expected losses under normal market conditions. For prudent risk management and as an extra management tool, firms should augment VaR analysis with stress-testing and scenario procedures. The VaR methodology gives a probabilistic measure of loss that may be exceeded,
110 Trading Risk Management
say, 2.5 per cent of the time. From a risk-management perspective, however, it is desirable to have an estimate of potential losses under severely adverse conditions where statistical tools do not apply. Stress-testing estimates the impact of unusual and severe events on a financial position’s value, and should be reported on a daily basis as part of the risk-reporting process. For emerging-market countries with extreme volatility, the usage of stress-testing should be highly emphasized and a full description of the process should be included in any policy and procedure manual. Stress-testing usually takes the form of subjectively specifying scenarios of interest to assess changes in the value of the portfolio, and it can involve examining the effect of past large market moves on today’s portfolio. The advantage of this method is that it may cover situations that are completely absent from historical data and therefore forces management to consider events that might otherwise have been ignored. Albeit that stress-testing may provide a better idea for potential losses under worst-case events, like the devaluation of an emergingmarket’s currency, political upheavals and so on, it gives little indication of the prospect of such extreme events. It also handles correlations very flimsily, which can be an indispensable component of risk in a portfolio of securities. However, it can be a very robust tool when used to complement the statistical VaR analysis. Subsequent to exploring the bulk of value distribution through VaR methodology, stress-testing might provide key insights by drawing a few situations from the furthest tails.
Equity trading risk management The market risk of a trading position is the risk of experiencing unexpected changes in the value of the position due to unexpected changes in the market variables or factors that affect the valuation of the position. Such market factors may be the level of equity markets or individual equity prices. These market variables that affect the value of a trading position are customarily called market-risk factors. Specifically, a trading-risk manager is interested in the likelihood of unexpected losses (rather that gains) and their magnitude for a trading position over a given time horizon. The interest in possible future losses in a trading operation is obvious as every trading house has only
Mazin A.M. Al Janabi 111
limited capital. In order to continue operating as a going concern even in the most adverse conditions, the allocated trading capital must be able to absorb the ‘maximum’ loss at any given time. The important parameters in market-risk assessment are: • The composition of the trading position and its mark-to-market (MTM) value. • The impact of changes in the market factors on its MTM value. • The time horizon for liquidation of trading positions. • The magnitude of the maximum loss that may be experienced and the likelihood of such a loss. In this study, databases of the most liquid stocks traded in the Moroccan financial markets were gathered from the Casablanca Stock Exchange (CSE). These stocks – 11 in total – are the constituents of a local index, the MADEX (Moroccan Most Active Shares Index, which is a capitalization weighted index). These most active 11 stocks are the following (their respective industrial sectors are included in parentheses): SAMIR (Oil Refinery), MANAGEM (Mining), ONA (Conglomerate), SONASID (Steel), LAFARGE (Cement), WAFA ASSURANCE (Insurance), SNI (Conglomerate), HOLCIM (Maroc) (Cement), BCM (Banking), BMCE (Banking), WAFABANK (Banking). Historical databases of daily prices were downloaded from the Moroccan Stock Exchange (Bourse de Casablanca) website (www.casablancabourse.com). These databases were for almost two years of daily prices and were essential elements for carrying out this research and for the construction of trading risk-management parameters and risk limits. In the process of analysing the data, the daily returns of the 11 sample stocks as well as the daily returns of the MADEX index were first calculated. These daily returns are the fact essential ingredients for the calculation of standard deviations, correlation matrices, stock betas, skewness and kurtosis for all the sample stocks and their relationship vis-à-vis the MADEX index. A software package was contrived for the purpose of creating a trading portfolio of these stocks and consequently to carry out all Value at Risk (VaR) and scenario-analysis calculations and results.
112 Trading Risk Management
The approach used in building-up the software package was based on matrix algebra and multiplication of matrices. In fact, the Excel™ package has many built-in functions that enable multiplication of matrices and other useful mathematical/statistical calculations. The data analysis tool pack of the Excel™ package is extremely useful for the creation of variance/covariance and correlation matrices. Furthermore, built-in functions such as MMULT and TRANSPOSE allow the multiplication and transposition of matrices. Other useful functions are STDEV, SLOPE, SKEW and KURT for the calculations of standard deviation, beta, skewness and kurtosis respectively. The analysis of data and discussion of most of the relevant findings and results of this research will now be discussed.
Analysis of volatility, beta, skewness and kurtosis In this section, the analysis of the particular risk for each stock (daily and annual volatility), the stock’s returns relationship with respect to the MADEX index, and finally a test of normality (symmetry) are performed on the sample stocks and the market index. Before discussing the relevant findings of this study, a brief outline of the most appropriate concepts used in this work is as follows: • Volatility is measured by the standard deviation of the daily returns of the market index and stocks under consideration. The standard deviation is a measure of how widely values are dispersed from the average value (the mean). • Beta measures the systematic risk of each stock with respect to the market index (the MADEX index in this case), or in other words the relative sensitivity of the stock vis-à-vis the market index. It is the slope of the regression line of the stock’s daily returns versus the index’s daily returns. • Skewness characterizes the degree of asymmetry of a distribution around its mean. Positive skewness indicates a distribution with an asymmetric tail extending towards more positive values, whereas negative skewness indicates a distribution with an asymmetric tail extending towards more negative values. A normal distribution, for instance, has a skewness of zero. • Kurtosis characterizes the relative peakedness or flatness of a distribution compared with the normal distribution. A large kurtosis figure (>3) indicates a relatively peaked distribution, whilst a small
Mazin A.M. Al Janabi 113
figure (<3) indicates a relatively flat distribution. The normal distribution, for example, has a kurtosis of three. Table 7.1 illustrates the daily volatility of each of the sample stocks under normal market and severe (crisis) market conditions. Crisismarket volatilities were calculated by multiplying the normal volatilities by a factor of five. From the table one can see that the stock with the highest volatility is WAFA ASSURANCE whereas BMCE has the lowest volatility. Annualized volatilities are shown in Table 7.2, obtained by adjusting (multiplication) the daily volatilities with the square root of 260 – assuming that there are 260 trading days in the calendar year. An interesting outcome of the study of betas (systematic risk) is the manner in which the results varied across the sample stocks as indicated in Table 7.1. MANAGEM appears to have the highest beta (1.56) vis-à-vis the MADEX index (that is, the highest systematic risk), and BMCE seems to have the lowest beta (0.20). Moreover, WAFABANK (with a beta of 0.97) is the best candidate of the entire sample that appears to move very closely with respect to the market index. In another study, measurements of skewness and kurtosis were achieved on the sample stocks and the MADEX index. The results Table 7.1 Quantitative analysis data: daily volatility, beta, skewness and kurtosis
Stocks SAMIR MANAGEM ONA SONASID LAFARGE WAFA ASSUR. SNI HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
Daily volatility Daily volatility (normal market) (crisis market) (%) (%) Beta Skewness Kurtosis 2·13 1·96 1·30 1·22 1·15 2·57 1·34 1·16 1·33 0·71 1·38 0·77
10·65 9·79 6·48 6·08 5·73 12·85 6·70 5·80 6·63 3·57 6·90 3·83
1·44 1·56 1·31 0·77 0·30 1·34 1·16 0·49 0·91 0·20 0·97 1·00
−0·37 −0·35 0·17 −1·97 −0·06 −0·01 −0·11 0·17 −0·23 0·35 0·85 0·35
6·06 4·18 6·08 30·92 8·93 2·58 5·01 6·51 6·10 6·19 4·96 3·26
114 Trading Risk Management
Table 7.2 Quantitative analysis data: annual volatility, beta, skewness and kurtosis
Stocks
Annual volatility Annual volatility (normal market) (crisis market) (%) (%) Beta Skewness Kurtosis
SAMIR MANAGEM ONA SONASID LAFARGE WAFA ASSUR. SNI HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
34·35 31·58 20·88 19·60 18·48 41·43 21·59 18·71 21·37 11·51 22·26 12·36
171·76 157·89 104·41 97·99 92·38 207·14 107·96 93·54 106·85 57·56 111·29 61·78
1·44 1·56 1·31 0·77 0·30 1·34 1·16 0·49 0·91 0·20 0·97 1·00
−0·37 −0·35 0·17 −1·97 −0·06 −0·01 −0·11 0·17 −0·23 0·35 0·85 0·35
6·06 4·18 6·08 30·92 8·93 2·58 5·01 6·51 6·10 6·19 4·96 3·26
are also shown in Table 7.1 where it can be seen that in general all stocks show asymmetric behaviour (both positive and negative values). Moreover, kurtosis studies show similar patterns of abnormality (that is, peaked/flat distributions). At the upper extreme, SONASID shows the greatest negative skewness (−1.97) which is combined with a very high kurtosis – peakedness – of 30.92. One stock, WAFA ASSURANCE, shows a close relationship to normality (skewness of −0.01 and kurtosis of 2.58). Likewise, the MADEX index also indicates some signs of normality with 0.35 and 3.26 of skewness and kurtosis respectively. An interesting outcome of this study suggests the necessity of combining Value at Risk calculations – which assume normal distributions of returns – with other methods such as stress-testing and scenario analysis to get a detailed picture of other remaining risks that cannot be captured with the simple assumption of normality.
Correlation matrices Three matrices of correlation were created, namely for correlations of 1, 0 and exact correlations. The objectives were to establish the necessary quantitative infrastructures for the advanced risk-management analysis that will follow shortly. The correlation matrices are shown in Tables 7.3–7.5 for the three correlation cases. These matrices were the essential elements along
Table 7.3 Quantitative analysis data: exact correlation matrix
SAMIR MANAGEM ONA SONASID LAFARGE WAFA ASSUR. SNI HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
SAMIR (%)
MANAGEM (%)
ONA (%)
SONASID (%)
LAFARGE (%)
WAFA ASSUR. (%)
SNI (%)
100·0 33·3 27·9 31·5 17·2 23·2 22·4
100·0 44·6 29·2 16·6 27·9 31·8
100·0 30·6 10·8 22·6 41·6
100·0 16·6 21·8 36·0
100·0 7·8 13·0
100·0 26·7
100·0
15·4 10·2 6·3 18·4 51·6
16·9 16·5 7·4 34·1 60·9
17·0 18·1 6·4 28·8 77·7
6·9 15·4 10·5 22·4 48·8
1·0 4·8 0·1 8·7 19·8
13·2 19·2 −5·7 38·5 39·9
15·1 20·2 10·2 32·3 66·1
HOLCIM (Maroc) (%)
BCM (%)
BMCE (%)
WAFABANK (%)
100·0 8·6 −0·1 19·8 32·2
100·0 11·5 23·5 52·5
100·0 4·0 21·1
100·0 53·7
MADEX (%)
100·0
115
116
Table 7.4 Quantitative analysis data: correlation 1 matrix
SAMIR MANAGEM ONA SONASID LAFARGE WAFA ASSUR. SNI HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
SAMIR (%)
MANAGEM (%)
ONA (%)
SONASID (%)
LAFARGE (%)
WAFA ASSUR. (%)
SNI (%)
HOLCIM (Maroc) (%)
BCM (%)
BMCE (%)
WAFABANK (%)
100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0 100·0
100·0 100·0 100·0 100·0
100·0 100·0 100·0
100·0 100·0
MADEX (%)
100·0
Table 7.5 Quantitative analysis data: correlation 0 matrix
SAMIR MANAGEM ONA SONASID LAFARGE WAFA ASSUR. SNI HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
SAMIR (%)
MANAGEM (%)
ONA (%)
SONASID (%)
LAFARGE (%)
WAFA ASSUR. (%)
SNI (%)
HOLCIM (Maroc) (%)
BCM (%)
BMCE (%)
WAFABANK (%)
100·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0 0·0
100·0 0·0 0·0 0·0
100·0 0·0 0·0
100·0 0·0
MADEX (%)
100·0
117
118 Trading Risk Management
with the volatilities matrices for the creation of Value at Risk (VaR) and stress-testing calculations for equity trading risk management processes and procedures.
Equity trading risk-management reports The mathematical parameters required for the analysis of trading risk management of proprietary portfolios were described earlier. In fact, the same theoretical framework developed earlier can be adapted for the calculation of trading-risk portfolios with equity proprietary positions. Our matrix-algebra approach once compared with the widely used ‘single-index-model’ approach should yield the same results for trading-risk factors. This is true if one can assume that the residual risk factors – in the single-index model approach – have no relationships (are uncorrelated) at all times – that is, correlation factors of zero. In this section, several case studies were carried out to emphasize the importance of trading risk-management reports for daily risk-taking practices. In the calculations reported here, the effects of different portfolio combinations, various liquidation periods (unwinding horizons of stocks holdings) and short-selling of stocks are all investigated. Table 7.6 illustrates a practical sample report for the coverage of equity trading risk-management activities of a hypothetical equity portfolio consisting of the most active stocks in the Moroccan Stock Exchange. In this first case study, the total portfolio value was Dh22 millions with an equal asset allocation of 9.1 per cent and a liquidity horizon of one trading day – that is, one day to unwind all trading positions. This table illustrates the effects of stress-testing (VaR under severe market conditions) and different correlation factors on daily VaR calculations. The VaR report also depicts the overnight (daily) volatilities that were calculated as the volatility of the percentage price changes (daily returns) of these stocks and their respective betas. The VaR results were calculated under normal and severe market conditions by taking into account different correlation factors (+1, 0 and exact correlations between the various risk factors). Correlation +1 assumes 100% positive relationships between all risk factors (risk positions) at all times, whereas the 0-correlation case states that there is no relationship between all positions at all times. The last
Table 7.6 Equity trading risk management report (analysis of case 1) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
Daily Volatility (%) Beta
SAMIR
$2,000,000
9.1
2.13
1.44
MANAGEM
$2,000,000
9.1
1.96
1.56
ONA
$2,000,000
9.1
1.30
1.31
SONASID
$2,000,000
9.1
1.22
0.77
LAFARGE
$2,000,000
9.1
1.15
0.30
WAFA ASSUR.
$2,000,000
9.1
2.57
1.34
SNI
$2,000,000
9.1
1.34
1.16
HOLCIM (Maroc)
$2,000,000
9.1
1.16
0.49
BCM
$2,000,000
9.1
1.33
0.91
BMCE
$2,000,000
9.1
0.71
0.20
WAFABANK
$2,000,000
9.1
1.38
0.97
0.0
0.77
1.00
MADEX
–
100 Total value in Dirhams
$22,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 348,812
649,333
206,894
1.59%
2.95%
0.94%
The Effects of Diversification $300,521
86.16%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 1,744,060
3,246,666
1,034,470
7.93%
14.76%
4.70%
The Effects of Diversification $1,502,606
86.16%
0.948
119
Beta of the Overall Portfolio
120 Trading Risk Management
correlation case takes into account the real correlations between all positions and was calculated via the variance/covariance matrix. As one might expect, the case with correlation +1 gives the highest VaR numbers (Dh649,333 and Dh3,246,666), owing to the fact that under these circumstances the total VaR of the portfolio is the weighted average of the individual VaR of each trading position. It is essential to include various correlation factors in any stress-testing exercise, based on the fact that current trends in correlations may break down with severe market movements caused by unexpected financial or political crises. The degree of diversification of this hypothetical trading portfolio can also be displayed as the difference in the value of the two greatest VaRs; that is, the VaR of the correlation +1 case versus the VaR of the exact correlation case (Dh300,521 or 86.16% for the normal market condition case). The beta of this portfolio is also indicated in this report as 0.948. Since the variations in daily VaR are mainly related to the ways in which the assets are allocated in addition to the liquidation period of the assets and the effects of short-selling, it is instructive to examine the way in which the VaR figures are influenced by changes in such parameters. Table 7.7 illustrates the changes in VaR numbers when the liquidation period was increased to 10 trading days for all stocks within the portfolio. In Table 7.8 it was assumed that the first six assets should be liquidated in five trading days, whereas the remaining five assets can be liquidated in 10 trading days respectively. The effects of short-selling (albeit short-selling is currently not permitted in the Moroccan Stock Exchange) are depicted in Table 7.9. One of the interesting results of this study is the way in which the VaR numbers have decreased. This behaviour might be explained by the way in which the overall portfolio is funded – in other words, some of the long positions have been funded with the short-selling of other stocks, consequently leading to a reduction in the overall risk. In fact one of the best advantages of calculating VaR within the matrix-algebra framework is the ability to incorporate the effects of short-selling without tedious mathematical analysis. Table 7.10 also shows the effects of short-selling with a liquidation horizon of 10 trading days, and Table 7.11 illustrates the same effects of short-selling but here with the assumption that the first six assets should be liquidated in five trading days and the remaining five assets liquidated in 10 trading days respectively.
Table 7.7 Equity trading risk management report (analysis of case 2) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
Daily Volatility (%) Beta
SAMIR
$2,000,000
9.1
2.13
1.44
MANAGEM
$2,000,000
9.1
1.96
1.56
ONA
$2,000,000
9.1
1.30
1.31
SONASID
$2,000,000
9.1
1.22
0.77
LAFARGE
$2,000,000
9.1
1.15
0.30
WAFA ASSUR.
$2,000,000
9.1
2.57
1.34
SNI
$2,000,000
9.1
1.34
1.16
HOLCIM (Maroc)
$2,000,000
9.1
1.16
0.49
BCM
$2,000,000
9.1
1.33
0.91
BMCE
$2,000,000
9.1
0.71
0.20
WAFABANK
$2,000,000
9.1
1.38
0.97
0.0
0.77
1.00
MADEX
–
100 Total value in Dirhams
$22,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 818,037
1,522,821
485,209
3.72%
6.92%
2.21%
The Effects of Diversification $704,785
86.16%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 4,090,183
7,614,106
2,426,046
18.59%
34.61%
11.03%
The Effects of Diversification $3,523,923
86.16%
0.948
121
Beta of the Overall Portfolio
122
Table 7.8 Equity trading risk management report (analysis of case 3) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
Daily Volatility (%) Beta
SAMIR
$2,000,000
9.1
2.13
1.44
MANAGEM
$2,000,000
9.1
1.96
1.56
ONA
$2,000,000
9.1
1.30
1.31
SONASID
$2,000,000
9.1
1.22
0.77
LAFARGE
$2,000,000
9.1
1.15
0.30
WAFA ASSUR.
$2,000,000
9.1
2.57
1.34
SNI
$2,000,000
9.1
1.34
1.16
HOLCIM (Maroc)
$2,000,000
9.1
1.16
0.49
BCM
$2,000,000
9.1
1.33
0.91
BMCE
$2,000,000
9.1
0.71
0.20
WAFABANK
$2,000,000
9.1
1.38
0.97
0.0
0.77
1.00
MADEX
–
100 Total value in Dirhams
$22,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 672,263
1,269,849
397,063
3.06%
5.77%
1.80%
The Effects of Diversification $597,586
88.89%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 3,361,313
6,349,244
1,985,315
15.28%
28.86%
9.02%
The Effects of Diversification $2,987,931
88.89%
Beta of the Overall Portfolio 0.948
Table 7.9 Equity trading risk management report (analysis of case 4) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
SAMIR
$(2,000,000)
MANAGEM
$2,000,000
Daily Volatility (%) Beta
−33.3
2.13
1.44
33.3
1.96
1.56
−33.3
1.30
1.31
SONASID
$2,000,000
33.3
1.22
0.77
LAFARGE
$2,000,000
33.3
1.15
0.30
ONA
WAFA ASSUR. SNI
$(2,000,000)
$2,000,000 $(2,000,000)
33.3
2.57
1.34
−33.3
1.34
1.16
HOLCIM (Maroc)
$2,000,000
33.3
1.16
0.49
BCM
$2,000,000
33.3
1.33
0.91
BMCE
$(2,000,000)
−33.3
0.71
0.20
33.3
1.38
0.97
0.0
0.77
1.00
WAFABANK MADEX
$2,000,000 –
100 Total value in Dirhams
$6,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 201,341
211,045
206,894
3.36%
3.52%
3.45%
The Effects of Diversification $9,704
4.82%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 1,006,704
1,055,225
1,034,470
16.78%
17.59%
17.24%
The Effects of Diversification $48,521
4.82%
0.742
123
Beta of the Overall Portfolio
124
Table 7.10 Equity trading risk management report (analysis of case 5) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
SAMIR
$(2,000,000)
MANAGEM
$2,000,000
Daily Volatility (%) Beta
−33.3
2.13
1.44
33.3
1.96
1.56
−33.3
1.30
1.31
SONASID
$2,000,000
33.3
1.22
0.77
LAFARGE
$2,000,000
33.3
1.15
0.30
ONA
WAFA ASSUR. SNI
$(2,000,000)
$2,000,000 $(2,000,000)
33.3
2.57
1.34
−33.3
1.34
1.16
HOLCIM (Maroc)
$2,000,000
33.3
1.16
0.49
BCM
$2,000,000
33.3
1.33
0.91
BMCE
$(2,000,000)
−33.3
0.71
0.20
33.3
1.38
0.97
0.0
0.77
1.00
WAFABANK MADEX
$2,000,000 –
100 Total value in Dirhams
$6,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 472,186
494,944
485,209
7.87%
8.25%
8.09%
The Effects of Diversification $22,758
4.82%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 2,360,931
2,474,722
2,426,046
39.35%
41.25%
40.43%
The Effects of Diversification $113,791
4.82%
Beta of the Overall Portfolio 0.742
Table 7.11 Equity trading risk management report (analysis of case 6) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
SAMIR
$(2,000,000)
MANAGEM
$2,000,000
Daily Volatility (%) Beta
−33.3
2.13
1.44
33.3
1.96
1.56
−33.3
1.30
1.31
SONASID
$2,000,000
33.3
1.22
0.77
LAFARGE
$2,000,000
33.3
1.15
0.30
ONA
WAFA ASSUR. SNI
$(2,000,000)
$2,000,000 $(2,000,000)
33.3
2.57
1.34
−33.3
1.34
1.16
HOLCIM (Maroc)
$2,000,000
33.3
1.16
0.49
BCM
$2,000,000
33.3
1.33
0.91
BMCE
$(2,000,000)
−33.3
0.71
0.20
33.3
1.38
0.97
0.0
0.77
1.00
WAFABANK MADEX
$2,000,000 –
100 Total value in Dirhams
$6,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 392,067
410,003
397,063
6.53%
6.83%
6.62%
The Effects of Diversification $17,936
4.57%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 1,960,335
2,050,016
1,985,315
32.67%
34.17%
33.09%
The Effects of Diversification $89,681
4.57%
0.742
125
Beta of the Overall Portfolio
126 Trading Risk Management
Finally, with the purpose of demonstrating the effects of dissimilar asset allocations, two more case analyses were carried out as illustrated in Tables 7.12 and 7.13. In both cases the unwinding period was held for one trading day. While Table 7.12 shows long investments with different asset allocations, Table 7.13 examines the effects of shortselling with various degrees of asset allocations.
VaR limits-setting for equity trading risk management In another study, different VaR calculations have been examined in order to set up procedures for the establishment of VaR trading limits and adequate policies for handling situations in which trading desks (units) are above the authorized VaR limits. Such VaR limits-procedures and methodologies must be analysed and approved by the board of directors of the financial entity. All trading desks need to have such limits of VaR as practical guidelines and also as a strict policy for their daily risk-takings. Any excess of VaR beyond the ratified limits must be reported to top management by the trading risk management unit. Moreover, traders need to give full and justified explanations of the reasons of why their overnight VaRs may be beyond the approved limits. Tables 7.14–7.17 represent several case studies for the setting of VaR limits. In these studies the effects of various asset allocations with and without short-selling were investigated for the purpose of setting adequate VaR limits. Further, in all cases a liquidation horizon of one trading day was assumed, and for the sake of simplification of the analysis, a volume trading limit of Dh110,000,000 was assumed as a constraint – that is, the financial entity must keep a maximum market value of stocks of no more than Dh110,000,000 between long and short positions. While in Table 7.14 equal asset allocations of 9.1 per cent were assumed, in Table 7.15 all equity trading was concentrated in one stock that had the highest daily return volatility – WAFA ASSURANCE in this case. Case studies 3 and 4 are illustrated in Tables 7.16 and 7.17 respectively. In case study 3, unequal asset allocation percentages were assumed by allocating more trading positions to the riskier stocks – that is, stocks with the highest overnight (daily) volatilities; while in Table 7.17 the effect of short-selling of some of the sample stocks was also examined by short-selling stocks with the lowest daily volatilities.
Table 7.12 Equity trading risk management report (analysis of case 7) Asset allocation and value at risk (VaR) report
Stocks
Market Value % Stocks
Daily Volatility (%) Beta
SAMIR
$5,000,000
12.5
2.13
1.44
MANAGEM
$5,000,000
12.5
1.96
1.56
ONA
$5,000,000
12.5
1.30
1.31
SONASID
$5,000,000
12.5
1.22
0.77
LAFARGE
$5,000,000
12.5
1.15
0.30
WAFA ASSUR.
$5,000,000
12.5
2.57
1.34
SNI
$2,000,000
5.0
1.34
1.16
HOLCIM (Maroc)
$2,000,000
5.0
1.16
0.49
BCM
$2,000,000
5.0
1.33
0.91
BMCE
$2,000,000
5.0
0.71
0.20
WAFABANK
$2,000,000
5.0
1.38
0.97
0.0
0.77
1.00
MADEX
–
100 Total value in Dirhams
$40,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 722,380
1,268,194
454,004
1.81%
3.17%
1.14%
The Effects of Diversification $545,814
75.56%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 3,611,899
6,340,969
2,270,021
9.03%
15.85%
5.68%
The Effects of Diversification $2,729,069
75.56%
1.025
127
Beta of the Overall Portfolio
128
Table 7.13 Equity trading risk management report (analysis of case 8) Asset allocation and value at risk (VaR) report
Market Value
SAMIR
$5,000,000
25.0
2.13
1.44
MANAGEM
$5,000,000
25.0
1.96
1.56
ONA
$5,000,000
25.0
1.30
1.31
SONASID
$5,000,000
25.0
1.22
0.77
LAFARGE
$5,000,000
25.0
1.15
0.30
WAFA ASSUR.
25.0
2.57
1.34
SNI
$(2,000,000)
−10.0
1.34
1.16
HOLCIM (Maroc)
$(2,000,000)
−10.0
1.16
0.49
BCM
$(2,000,000)
−10.0
1.33
0.91
BMCE
$(2,000,000)
−10.0
0.71
0.20
WAFABANK
$(2,000,000)
−10.0
1.38
0.97
0.0
0.77
1.00
MADEX
$5,000,000
% Stocks
Daily Volatility (%) Beta
Stocks
–
100 Total value in Dirhams
$20,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 580,870
794,675
454,004
2.90%
3.97%
2.27%
The Effects of Diversification $213,805
36.81%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 2,904,348
3,973,374
2,270,021
14.52%
19.87%
11.35%
The Effects of Diversification $1,069,027
36.81%
Beta of the Overall Portfolio 1.307
Table 7.14 Equity trading risk management report (VaR limits-settings, case 1) Asset allocation and value at risk (VaR) report
% Stocks
Daily Volatility (%) Beta
Stocks
Market Value
SAMIR
$10,000,000
9.1
2.13
1.44
MANAGEM
$10,000,000
9.1
1.96
1.56
ONA
$10,000,000
9.1
1.30
1.31
SONASID
$10,000,000
9.1
1.22
0.77
LAFARGE
$10,000,000
9.1
1.15
0.30
WAFA ASSUR.
$10,000,000
9.1
2.57
1.34
SNI
$10,000,000
9.1
1.34
1.16
HOLCIM (Maroc)
$10,000,000
9.1
1.16
0.49
BCM
$10,000,000
9.1
1.33
0.91
BMCE
$10,000,000
9.1
0.71
0.20
WAFABANK
$10,000,000
9.1
1.38
0.97
0.0
0.77
1.00
MADEX
–
100 Total value in Dirhams
$110,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 1,744,060
3,246,666
1,034,470
1.59%
2.95%
0.94%
The Effects of Diversification $1,502,606
86.16%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 8,720,300
16,233,329
5,172,348
7.93%
14.76%
4.70%
The Effects of Diversification $7,513,029
86.16%
0.948
129
Beta of the Overall Portfolio
130
Table 7.15 Equity trading risk management report (VaR limits-settings, case 2) Asset allocation and value at risk (VaR) report
Market Value
SAMIR
–
0.0
2.13
1.44
MANAGEM
–
0.0
1.96
1.56
ONA
–
0.0
1.30
1.31
SONASID
–
0.0
1.22
0.77
LAFARGE
–
0.0
1.15
0.30
WAFA ASSUR.
$110,000,000
% Stocks
Daily Volatility (%) Beta
Stocks
100.0
2.57
1.34
SNI
–
0.0
1.34
1.16
HOLCIM (Maroc)
–
0.0
1.16
0.49
BCM
–
0.0
1.33
0.91
BMCE
–
0.0
0.71
0.20
WAFABANK
–
0.0
1.38
0.97
MADEX
–
0.0
0.77
1.00
100 Total value in Dirhams
$110,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 5,652,330
5,652,330
5,652,330
5.14%
5.14%
5.14%
The Effects of Diversification –
0.00%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 28,261,649
28,261,649
28,261,649
25.69%
25.69%
25.69%
The Effects of Diversification –
0.00%
Beta of the Overall Portfolio 1.336
Table 7.16 Equity trading risk management report (VaR limits-settings, case 3) Asset allocation and value at risk (VaR) report Daily Volatility (%) Beta
Stocks
Market Value
% Stocks
SAMIR
$20,000,000
18.2
2.13
1.44
MANAGEM
$20,000,000
18.2
1.96
1.56
ONA
$10,000,000
9.1
1.30
1.31
SONASID
–
0.0
1.22
0.77
LAFARGE
–
0.0
1.15
0.30
WAFA ASSUR.
$20,000,000
18.2
2.57
1.34
SNI
$10,000,000
9.1
1.34
1.16
–
0.0
1.16
0.49
$10,000,000
9.1
1.33
0.91
–
0.0
0.71
0.20
$20,000,000
18.2
1.38
0.97
0.0
0.77
1.00
HOLCIM (Maroc) BCM BMCE WAFABANK MADEX
–
100 Total value in Dirhams
$110,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 2,549,538
4,007,239
1,705,842
2.32%
3.64%
1.55%
The Effects of Diversification $1,457,701
57.18%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 12,747,691
20,036,197
8,529,211
11.59%
18.21%
7.75%
The Effects of Diversification $7,288,506
57.18%
1.270
131
Beta of the Overall Portfolio
132
Table 7.17 Equity trading risk management report (VaR limits-settings, case 4) Asset allocation and value at risk (VaR) report Daily Volatility (%) Beta
Stocks
Market Value
% Stocks
SAMIR
$20,000,000
18.2
2.13
1.44
MANAGEM
$20,000,000
18.2
1.96
1.56
ONA
$40,000,000
36.4
1.30
1.31
SONASID
$(10,000,000)
−9.1
1.22
0.77
LAFARGE
$(10,000,000)
−9.1
1.15
0.30
WAFA ASSUR.
$20,000,000
18.2
2.57
1.34
SNI
$10,000,000
9.1
1.34
1.16
$(10,000,000)
−9.1
1.16
0.49
BCM
$20,000,000
18.2
1.33
0.91
BMCE
$(10,000,000)
−9.1
0.71
0.20
$20,000,000
18.2
1.38
0.97
0.0
0.77
1.00
HOLCIM (Maroc)
WAFABANK MADEX
–
100 Total value in Dirhams
$110,000,000
Daily Value At Risk (VaR) In Dirhams Normal Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 2,963,535
4,202,274
2,076,751
2.69%
3.82%
1.89%
The Effects of Diversification $1,238,739
41.80%
Daily Value At Risk (VaR) in Dirhams Severe (Crisis) Market Conditions Correlation = Exact Correlation = 1 Correlation = 0 14,817,676
21,011,370
10,383,756
13.47%
19.10%
9.44%
The Effects of Diversification $6,193,694
41.80%
Beta of the Overall Portfolio 1.551
Mazin A.M. Al Janabi 133
A summary of the VaR results for the four case studies is as follows: Daily trading VaR in dirhams (normal market conditions) with correlations Exact +1 0 Case study 1 1,744,060 3,246,666 1,034,470 Case study 2 5,652,330 5,652,330 5,652,330 Case study 3 2,549,538 4,007,239 1,705,842 Case study 4 2,963,535 4,202,274 2,076,751 Daily trading VaR in dirhams (severe market conditions) with correlations Exact +1 0 Case study 1 8,720,300 16,233,329 5,172,348 Case study 2 28,261,649 28,261,649 28,261,649 Case study 3 12,747,691 20,036,197 8,529,211 Case study 4 14,817,676 21,011,370 10,383,756 As one might expect, the highest VaR number calculated so far is for case study 2, when all the trading budget was put in a single stock with the highest overnight volatility. This is clearly due to the fact that there was no diversification benefit to be taken into account. The principal effect of diversification on VaR limits-setting seems to be through case study 1, that is with an equal asset allocation percentage of 9.1 per cent. As a conclusion of this study, the board of directors can set the maximum daily trading VaR limits as follows: • Approved VaR limit = Dh5,652,330 (under normal market conditions). • Approved VaR limit = Dh28,261,649 (under severe or crisis market conditions).
Concluding remarks There are many methods and ways to identify, to measure and to control trading risk, and trading-risk managers have the task of selecting the one that best suits their needs. In fact, there is no right or wrong way to measure/manage trading risk; it all depends on each entity’s
134 Trading Risk Management
objectives, lines of business, risk appetite and the availability of funds for investment in trading risk-management projects. Regardless of the methodology chosen, the most important factors to consider are the establishment of sound risk practices, policies and standards, and consistency in the implementation process across all lines of businesses and risks. The trading risk-management unit is responsible for monitoring all risks related to proprietary trading. These risks include market risk, event risk, issuer risk and liquidity risk as well as trading-related country, legal and operational risks. The market risk of a trading position is the risk of experiencing changes in the value of the position due to unexpected changes in the market variables or factors that affect the valuation of the position, such as the level of equity markets or individual equity prices. These market variables that affect the value of a trading position are referred to as market-risk factors. The Value at Risk (VaR) concept is the focus for establishing information on position risks. VaR gives the maximum acceptable potential loss arising from the impact of adverse price changes or changes in other market risk factors (for example volatilities) on open positions over a given time horizon. The maximum acceptable loss is usually calculated with a 97.5 per cent confidence interval (2σ ) and with a standard time horizon of one day. In addition, VaR limits are usually set on overnight risk exposures. Under special conditions, when changes in market risk factors are normally distributed, the VaR can be calculated using the variance/covariance approach. For VaR limits-setting and daily trading risk-measurement purposes these assumptions are made for the sake of simplifying the calculation process. However, for an emerging-market environment, one needs to supplement the variance/covariance approach with other analyses such as stress-testing and simulation analysis. This is done with the objective of estimating the impact of assumptions that are made under the VaR approach. Likewise, the effects of illiquidity of trading assets in emerging markets must be dealt with more wisely and should be brought within the VaR framework. In this chapter a simplified and practical method for calculating a portfolio’s trading risk has been presented and analysed. A matrix-algebra approach was used to derive the necessary mathematical/quantitative trading risk-management methods. This approach
Mazin A.M. Al Janabi 135
has several advantages owing to the fact that it can facilitate the programming process in Excel™ worksheets and can also permit easy incorporation of short-selling of trading assets into the equity trading process. The effects of illiquidity of trading assets are also incorporated into the VaR quantitative approach. For this purpose a simplified and practical model for the measurement of the effects of illiquid assets (unwinding of trading assets), in daily trading risk-management practices, has been defined and appropriately integrated into the VaR and stress-testing models. Trading risk-management models, as developed in this work, were applied to the Moroccan stockmarket, and our analyses were carried out for the most active shares on the Moroccan Stock Exchange and the local MADEX index. To this end, databases of daily stock prices and the MADEX index levels were all downloaded from the Moroccan Stock Exchange website. Several case studies were carried out with the objectives of calculating VaR numbers under various scenarios. The different scenarios were performed with distinct asset allocation percentages in addition to analysing the effects of liquidity of trading assets (unwinding horizon period of assets) and the possibilities of short-selling. The analyses performed include volatility, skewness and kurtosis tests along with beta calculations. Our results suggest that in almost all tests there are clear asymmetric behaviours in the distribution of returns of the sample stocks. In the calculations reported herein, the MADEX index and one stock (WAFA ASSURANCE) have shown close signs of normality. Equity trading risk-management reports have been illustrated for several case studies using the matrix-algebra approach. In all these case studies, VaR equity trading risk reports with different asset allocation percentages, short-selling and unwinding periods (the effects of illiquidity of assets) have been shown. Our matrix-algebra approach compared with the widely used ‘single-index model’ approach should yield the same results of trading risk factors. This is only true, however, if one can assume that the residual risk factors – in the single-index model approach – have no relationships at all times. VaR limits-setting is an important issue as part of the daily riskmanagement process. To this end, a procedure was developed to illustrate a practical approach for the setting of VaR limits for an
136 Trading Risk Management
equity-trading unit. In all case studies, the volume limit was assumed to be constant and was used as a constraint for the establishment of VaR limits.
References Al Janabi, Mazin A.M. (2001) ‘Integral Risk Management and Analysis of Derivative Products’, paper presented at Conference organized by the Institute of International Research, June 2001, Mexico City. Al Janabi, Mazin A.M. (2003) ‘Risk Management in Emerging Markets’, paper presented at Global Wealth 2003: The 2nd International Finance and Business Forum and Exhibition, Casablanca, Morocco, 2–5 December. Al Janabi, Mazin A.M. (2003) Formulation of Successful Derivatives Products in Emerging Markets (Morocco: Al Akhawayn University Press). Black, F., and Scholes, M. (1973) ‘The Pricing of Options and Corporate Liabilities’, Journal of Political Economy, vol. 81 (May–June), pp. 637–54. Blommestein, H. (2000) ‘The Changing Nature of Risk and the Challenges to Sound Risk Management in the New Global Financial Landscape’, Financial Market Trends, no. 75 (March). Boyle, P. and Feidhlim, B. (2001) Derivatives: The Tools that Changed Finance. (London: Risk Books). casablanca-bourse.com (Casablanca Stock Exchange Website). Group of Thirty (1993) Derivatives: Practices and Principles, Global Derivatives Study Group, Washington, DC. Holder, M. Tomas III, M. and Robert W. (1999) ‘Winners and Losers: Recent Competition among Futures Exchanges for Equivalent Financial Contract Markets’, Derivatives Quarterly (Winter). Jorion, P. (1997) Value at Risk (Burr Ridge, Illinois: Irwin Professional Publishing). Kimball, R. (2000) ‘Failures in Risk Management’, New England Economic Review (January/February). Markowitz, H. (1959) Portfolio Selection: Efficient Diversification of Investments (New York: John Wiley). Morgan Guaranty Trust Company (1994) Risk Metrics – Technical Document (New York: Morgan Guaranty Trust Company, Global Research). Rahl, L. and Esseghaier Z. (2000) ‘Measuring Financial Risk in the 21st Century’, Bank Accounting and Finance (Spring). Sharpe, W. (1963) ‘A Simplified Model for Portfolio Analysis’, Management Science, no. 9, pp. 277–93.
8 Value at Risk: Does it Work in Emerging Markets? Chuntao Yu, Bob Davidson and Mohamed Nurullah
Introduction Estimating and controlling risks has become a vital topic in financial institutions (Bouchaud, 2000; Mulvey et al., 1996). All financial intermediaries (FIs) entail the assumption, management and pricing of risk (Cornett and Saunders, 1999). Risk facing financial intermediations, in this broader sense, encompasses credit risk, market risk, liquidity risk, gearing risk, solvency risk, operational risk and sovereign risk (Bessis, 2001; Santemero and Babbel, 1996; Sinkey, 2002; Thygerson, 1995). Among these different risk issues and categories, market risk is a central risk faced by financial institutions (Bessis, 2001; Cornett and Saunders, 1999; Heffernan, 1996; Santomero, 1997), and there are a number of models that are currently employed by various financial institutions to measure it; Table 8.1 gives a brief outline of the measurement of market risk. Among the market-risk measurement methods, Value at Risk (VaR) has been adopted as the central measure of market risk by financial institutions (BIS, 2001). The VaR approach was strongly recommended in the July 1993 study by the Group of Thirty, which has become a blueprint for managing all financial risk. In 1996, the BCBS issued an amendment to the Capital Accord of July 1988 to use VaR for measuring market risk to set capital requirements for banks (BCBS, 1996). Since then, many national regulatory authorities have adopted the BCBS recommendations. For example, the European Union’s Capital Adequacy Directive makes the VaR of the market risk in a bank’s trading book part of the calculation of their capital reserve requirement. 137
138 Value at Risk in Emerging Markets
Table 8.1 A brief summary of current market-risk measures Methods
Brief description
Major models utilized
Value at Risk
Measures potential Parametric method losses in normal markets Non-parametric
Stresstesting
Estimates potential Mechanical stress losses in abnormal tests markets Scenario analysis
Extreme value theory
Calculates the Relatively market risk value emerging on the behaviour technique of the ‘tails’ (i.e., under extreme conditions and low potential values) of probability distributions
Back testing
Verifies the accuracy of the model-based approach
No common criteria have been set so far
Variance–covariance VaR Historical simulation of VaR Monte Carlo simulation of VaR Sensitivity test Maximum loss optimization Historical scenarios Hypothetical scenarios EV–VaR Other EVT-based methods (future research area)
BFTL back test ...
Spurred by regulators’ proposals, VaR has become a standard measure of financial market risk that is increasingly used by other financial and even non-financial firms as well (Berkowitz and O’Brien, 2001; Culp and Neves, 1997; Wong et al., 2002). Since Chase Manhattan Bank ran its first VaR a decade ago, a number of leading financial institutions (for example HSBC, Lehman Bros, Swiss Re, UBS AG, Royal Bank of Scotland and Barclays) have developed VaR models to aggregate and monitor their market risk. Moreover, VaR has also gained strong support as an industry standard in various forms of
Chuntao Yu, Bob Davidson and Mohamed Nurullah 139
academic literature (for example Christoffersen et al., 2001; Jorion, 2001; Heffernan, 1996; Santomero, 1997). Proponents of VaR suggest speaking a common language in VaR with respect to market risk. Formally, a VaR = VaRt,T is defined as the upper bound of the onesided confidence interval: Pr[P(T ) ≤ VaR] = 1 − c
(8.1)
where c is the confidence level and P(T ) = Pt (T ) is the relative change (return) in the portfolio value covering the time horizon T : Pt (T ) = P(t + T ) − P(t)
(8.2)
where P(t) = log S(t); S(t) is the portfolio value at t; the time period is [t, T ], with T − t = T , and t is the current time. Since the introduction of the simplest VaR models a little over 10 years ago, the range of techniques used to obtain VaR estimates has expanded both in number and in complexity. Yet, so far, there is no industry consensus on the best method for calculating VaR (Engel and Gizycki, 1999; Sarma et al., 2001). As with any statistical model, VaR depends on certain assumptions (Härdle et al., 2002), and it is commonly regarded that there are three main approaches to calculating VaR (see for example Chew, 1996; CMRA, 1996; Laubsch, 1999; Stambaugh, 1996; Zecher and Putnam, 2002): variance–covariance approaches (also known as the correlation method, parametric or analytic VaR), the historical simulation and the Monte Carlo simulation. An overview of the three VaR methodologies as well as a summary of their key strengths and weaknesses are shown in Tables 8.2 and 8.3 respectively. Each VaR method has its own set of assumptions and each is a simplification of reality (see Table 8.2). All methods have their own strengths and weaknesses, and together the differences between the approaches predictably results in different risk perspectives. The differences in common VaRs emphasize the fact that no single set of parameters, data, assumptions and methodology is accepted as the ‘correct’ approach. A good example has been given by Beder (Beder, 1995). His study found that depending on the selection of the time horizon, data-base and correlation assumptions across instrument/ asset classes, the same model may produce widely divergent VaR views
140 Value at Risk in Emerging Markets
Table 8.2 An overview of the three main VaR methodologies Brief description
Calculations involved
Variance– covariance
Estimates VaR with equation that specifies parameters such as volatility, correlation, delta and gamma
Volatility and correlation matrix; matrix algebra to arrive at VaR
Historical simulation
Estimates VaR by reliving history; takes actual historical rates and revalues positions for each change in the market
Historical data-set; simulate portfolio using historical returns as actual return distribution
Methodology
Monte Carlo Estimates VaR by simulation simulating random scenarios and revaluing positions in the portfolio
Applications Accurate for linear derivatives; loses precision for non-linear derivatives
Appropriate for all types of instruments, linear Volatility and and non-linear correlation matrix; Monte Carlo simulation to generate portfolio return distribution
for the sample portfolio. Therefore, the key is not the model itself, but the right selection of an appropriate method and its suitability to financial institutions’ risk objectives.
Research methodology A cross-sectional global survey method using an e-mail-based questionnaire was selected for the present research. A panel of highlyauthoritative experts participated in the project, all with many years experience in market-risk activities and with publications in first-class journals. The level of the respondents’ qualifications was carefully examined according to a further criterion that specialists who did not have expertise covering all the models involved in the project were rejected. Respondents were further chosen through two selection criteria. The first was industry selection; the respondents include regulators
Chuntao Yu, Bob Davidson and Mohamed Nurullah 141
Table 8.3 A summary of the key strengths and weaknesses of three VaR methodologies Methodology Variance– covariance
Advantage
Historical simulation
Monte Carlo simulation
Relatively easy to understand Fast and simple calculation Fully captures non-linear risks No need to make distributional assumptions. Ease of communications with senior management
Fully captures non-linear risks Accommodates any statistical assumptions about risk factors
Disadvantage
Fat-tails problem Less accurate for non-linear portfolios Relies on historical data and may be difficult to sample far back Incorporates the fat-tails problem (only if historical data-set includes tail events) Somewhat computationally intensive and time-consuming Sampling risk if not-adequate simulations Quantified fat-tailed risk (only if market scenarios are generated from the appropriate distributions) Computationally intensive and time-consuming
from central banks, market-risk practitioners from financial institutions, and university scholars. The second selection criterion was based on location; questionnaires were sent to Europe, Asia, Oceania, the Middle East and America. The questionnaire contains a mixture of 16 questions with predefined answers as well as some where the respondents are free to say whatever they wish. Hague (1993) defines this type of questionnaire as ‘semi-structured’, which is a more flexible tool to find out certain answers. The questions have been grouped into three main sections. Section A enquires about the effectiveness of the existing market-risk measurement models; section B asks the respondents to rate their selection criteria and indicate the most urgent future search area; and in Section C respondents are asked the extent of their agreement with the model combinations.
142 Value at Risk in Emerging Markets
The design of the survey questionnaire also utilises a 5-level rating scheme. For example, level 1 represents ‘the least relevant factors for selecting’ and 5 ‘the most important’. The reason for adopting this scale is to give some degree of flexibility of choice to respondents to reflect the intensity of their views (Hague, 1993). Another main benefit of adopting a 5-level rating is to facilitate the eventual process of tabulating the results for statistical analysis (Silverman, 1993). After suitable pilot testing, the revised and final version of the survey questionnaire was distributed one-by-one through e-mail to all the 16 proposed participants on 7 July 2003. A return date was also given so that respondents would be encouraged to return the completed survey within a reasonable time frame (Saunders et al., 2003). A total of 11 respondents eventually returned their answers. Of these questionnaires, all were correctly completed. The response rate was 68.75 per cent. This response rate is arguably high compared with the literature on international risk surveys, where a 10 per cent response rate is typically considered sufficient. For example, Alexander Stenhouse & Co.’s (1991) ‘Global Risk Financing and Insurance Survey’, had a response rate of only 8.8 per cent. However, in the present study, because the intended sample was small to begin with, a high response rate was essential.
Analysis and results This part contains a detailed discussion of the overall data analysis. Descriptive statistics including chi-square tests and means were utilized. Such methods may be considered appropriate in view of the small sample obtained from the survey (Siegel and Castellan, 1988). The quantitative results presented in this evaluation were addressed relevant to each question, from the three main sections in the questionnaire. Section A deals with model usage which contains six questions; section B deals with model selection which contains three questions; and section C deals with model combinations which contains seven questions. Respondents were asked to indicate the level of their agreement/disagreement according to the 5-point scale.
Model usage (questionnaire section A) Q1: The use of Value at Risk (VaR) or other market-risk models must be embedded into a well-developed risk-management
Chuntao Yu, Bob Davidson and Mohamed Nurullah 143
Table 8.4 Responses to question 1 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 27.64
degrees of freedom = 4
0 0 0 2 9 Significant @ 1%
Table 8.5 Responses to question 2 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 34.91
degrees of freedom = 4
0 0 0 1 10 Significant @ 1%
infrastructure – including policies and procedures, systems and welldefined senior management responsibilities. As Table 8.4 shows, all the experts agree with the statement, an obviously statistically-significant result. The success of the market risk measures in financial institutions has been due in large part to its risk infrastructure. During the 1990s, some well-publicized financial losses raised the issue of whether the methodology of risk management could be transported to those corporate governance environments effectively. For example, Barings failed in 1995 and VaR systems failed to inform top management to shut down rogue trader Nick Leeson’s position. A VaR measurement system was not in place at Barings because the bank had lost its control and made risk models impossible to implement (Culp et al., 1997). Q2: Training and education is important for ensuring that the market risk models are understood at all levels of the risk team. This second question was intended to gather estimates of the importance of communication and education. The overwhelming majority of participants clicked the check box of ‘Agree Strongly’. This
144 Value at Risk in Emerging Markets
Table 8.6 Responses to question 3 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 26.73
degrees of freedom = 4
9 1 1 0 0 Significant @ 1%
result points to a factor that has been largely omitted by the existing modelling literature. In the real world, close team cooperation is crucial to continued success (BIS, 1996), and decisions on risk-taking are often made by groups. Yet, the current risk modelling techniques have become more and more complex, for example the calculation of VaR is a very challenging statistical exercise (Engle and Maganelli, 1999; Marshall and Siegel, 1996). If the risk team cannot understand the complex models, they must rely on experts’ understanding, and the whole group may become dependent on their expertise. This may be the greatest risk of all (Bulter, 1999; Waring and Glendon, 1998). Thus, any good risk-modelling programme should be supported by effective training and education. As recommended by the Basle Committee on Banking Supervision (BCBS, 1994, p. 7), ‘Risk measurement standards should be understood by relevant personnel at all levels of the institution – from individual traders to the board of directors’. Q3: There is a ‘super’ model that can precisely measure all kinds of portfolio and market risk. Here the results illustrate a very good confirmation of the argument that one best universal method to measure market risk in all circumstances does not exist, as suggested in the introductory section. Q4: VaR-based reports (including Extreme Value VaR (EV-VaR)) are sufficient to inform the hedging of market risk. The experts’ answer to Q4 do not present a statistically significant result across the five categories. However, combining the five categories into three basic categories, the statistical outcome becomes weakly significant at the 5 per cent confidence level even though seven of the experts disagreed with the Q4 statement. This implies that most
Chuntao Yu, Bob Davidson and Mohamed Nurullah 145
Table 8.7 Responses to question 4 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 6.73
degrees of freedom = 4
3 5 1 2 0 Not significant
Combining the above data into 3 categories Disagree Neither Agree Nor Disagree Agree Chi square (χ 2 ) = 7.82
degrees of freedom = 2
7 1 2
Significant @ 5%
experts do not agree with the strategy of solely using the VaR report to hedge market risk for portfolios. In other words, while VaR has been widely adopted as a risk standard because of its merits compared to more traditional risk measures, it is also true that the VaR method is not, by any means a flawless paragon and the model itself suffers from a number of weaknesses as previous analysis has demonstrated. Q5: VaR (including EV-VaR) is useful only to certain types of companies in particular circumstances (e.g, where the consolidated values of exposures across a variety of activities are at issue). Q6: Financial institutions need to devise their own models (perhaps based on some standard approach). For Q5 and Q6, both the five- and three-category chi-square results are not statistically significant. It is, therefore, hard to conclude on the experts’ comments, and further research for these two aspects is required.
Model selection (questionnaire section B) To analyse the determinants of selecting market risk models, the participants in the survey were presented with two interrelated questions. First, 11 potential criteria were proposed in Q7 and the participants were asked to single out the three they considered the most important for selecting market-risk models. Next, in Q8, the participants
146 Value at Risk in Emerging Markets
Table 8.8 Responses to question 5 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 4.91
degrees of freedom = 4
2 1 1 5 2 Not Significant
Combining the above data in 3 categories Disagree Neither Agree Nor Disagree Agree Chi square (χ 2 ) = 5.09
degrees of freedom = 2
3 1 7
Not Significant
Table 8.9 Responses to question 6 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 6.73
degrees of freedom = 4
1 0 3 5 2 Not Significant
Combining the above data in 3 categories Disagree Neither Agree Nor Disagree Agree Chi square (χ 2 ) = 5.09
degrees of freedom = 2
1 3 7
Not Significant
were asked, out of the same 11 determinants, to classify each using a 5-point scale where 1 corresponded to less important and 5 to major importance. The importance of the determinants was assessed by computing the mean of the respective variable (cf. Table 8.10). Analysis of the determinants’ mean for each item presents a clear picture. Table 8.10 summarizes the results.
Chuntao Yu, Bob Davidson and Mohamed Nurullah 147
Table 8.10 Statistical results of questions 7 and 8 The three most important criteria∗∗ Rank 1 2 3 4 5 6 7 7
9 10
11
Selection determinants The data available The purpose of the analysis The composition of the portfolio The ease of communicability (e.g, report to management) The IT/system resources available Satisfy regulatory standards The expense that will be occurred Integrating the analysis into enterprise-wide risk management The accuracy required The speed at which the analysis has to be conducted The degree to which subjective judgment is involved
Mean∗
First
Second
Third
In total
4.55 4.37 4.18
2 3 3
3 1 4
3 2 0
8 6 7
3.82
1
0
1
2
3.73
0
0
2
2
3.64 3.45
1 1
2 0
0 2
3 3
3.45
0
0
2
2
3.37 3.19
2 0
0 0
0 0
2 0
2.91
0
0
0
0
Note: ∗ denotes the mean of the scale that ranged from 1 (minor importance) to 5 (major importance); ∗∗ denotes the number of respondents that chose it as one of the three most important criteria.
Three determinants were rated well above all the others with their mean above the average value of 4. Among them, the ‘the data available’ was unquestionably the top answer – it recorded the highest mean in the 5-point scale and was chosen as one of the three most important criteria by seven respondents. The second most important determinant in terms of market model selection is to decide ‘the purpose of the analysis’. This was considered as one of the top three criteria by six of the respondents. Three experts in particular believed it should be the most important selection criterion. ‘The composition of the portfolio’ was also among the top three determinants; with a mean value of 4.18, it was chosen as one of the three major criteria by seven of the participants.
148 Value at Risk in Emerging Markets
A fourth determinant that attracted attention is the variable ‘the ease of communicability’. With a mean value of 3.82, it was singled out as one of the three major criteria by two experts. This should not be a surprise since it corresponds with market-risk models needing to be understood by the whole risk team. In contrast, the remaining items had low overall ratings, although they were seen as the main rationale by a substantial number of respondents. These determinants are strong for specialist criteria – they were very important for some financial institutions but not generally important in the overall population. A special determinant that deserves notice is the variable ‘the accuracy required’. Before the survey this, especially, was expected to be one of the most important selection criteria. Yet, the actual level of the statistical mean of this variable was rather low though two experts checked it as the most important. This result requires further exploration. Previous analysis has showed that market measurement systems cannot flawlessly predict cash flow, earnings or capital. As a result, such systems should be treated as only a forecasting tool (FDIC, 1998). Accordingly, it is not surprising to see the low mean of this determinant in Q8 and the experts’ non-agreement of Q4. It is also worth noting that two experts recommended that liquidity risk, that is ‘Incorporation of differences in assets liquidity when measuring market risk’ (returned questionnaire no. 8), should also be a factor influencing model selection. Though the two respondents do not rank the factor into the top three, it still deserves further exploration. The recommendation suggests that a choice should be made between risk measures that change slowly with time, capturing low-frequency risk changes but not high-frequency changes, and risk measures that are more reactive to current information but jump around to reflect short-term variations in risk. Q9 addresses the issue of future research, and the importance of different future research areas was ranked reported in Table 8.11. The results in Table 8.11 suggest a number of interesting recommendations for future research. With the support of the statistical mean, ‘creation of new methods beyond VaR’, ‘stress-testing techniques’, and ‘selection and appropriate use of existing models’ were identified as the three most urgent areas. Quite revealing was that the ‘creation of new methods beyond VaR’ was, overall, the most
Chuntao Yu, Bob Davidson and Mohamed Nurullah 149
Table 8.11 Statistical results of Q9 Rank
Future research area
1 2 3 4 5 6 7 8 9
Creation of new methods beyond VaR Stress-testing techniques Selection and appropriate use of existing models Expected tail loss (ETL)/ conditional VaR/ tail VaR Back-testing techniques Extreme value VaR (EV-VaR) Improvement of the existing Value-at-Risk (VaR) models Development of new VaR approaches New methods of extreme value theory
Mean∗ 4.09 4.19 4.0 3.64 3.8 3.55 3.37 3.27 3.0
∗ Mean of the scale that ranged from 1 (minor importance) to 5 (major importance)
important area. As could be expected from previous analysis, the current practice of adopting VaR as an industry standard is not appropriate, and, to a further extent, it may be a dangerous application since VaR suffers too many weaknesses. The result of ‘stress-testing techniques’ being the second most urgent area confirms the argument that the methodologies of stress-testing are still regarded as being in their infancy, as previous analysis has justified. What seems to us most significant is the third most urgent determinant; our respondents reported a high level of importance for ‘selection and appropriate use of existing models’. This precisely supports and confirms the motivation of our research as shown in the previous section. Although Dickson and Giglierano’s (1986) studies are not within the discipline of market-risk modelling, their conclusions still indirectly support the background of the research in a way that poor model choices can lead to ‘sinking the boat’ or ‘missing the boat’. In fact, inappropriate risk models have allowed some extraordinary failures to occur during the past 10 years, leading to ‘model risk’ (Allport, 2000). In this way, the present study is of significant importance in terms of a comparison with the existing literature.
Model combination (questionnaire section C) Q10: It is advisable that a basic VaR (including EV-VaR) approach be supplemented and combined with other market-risk measurement methods.
150 Value at Risk in Emerging Markets
Table 8.12 Responses to question 10 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 18.55
degrees of freedom = 4
0 0 0 4 7 Significant @ 1%
All of the experts agree with the statement. Again, the finding implies that the use of VaR should not dominate financial risk management as the only industry standard. This is because individual VaR methods present cons as well as pros. Therefore, it may be valuable to combine them together and allow them to compensate each other. By adopting multiple methods, market-risk measurement may be better captured than solely relying on benchmark models. Secondly, dependence on a single risk measure has a problem in disregarding important information on portfolio risk (BCBS, 1994). To capture the information disregarded by VaR, it is essential to supplement it with other market-risk techniques. Nevertheless, different methods of computing market risk may generate widely varying results (to be discussed in a later section). The use of multi-methods was explored in Q11. Q11: The problem that different models yield different results for an identical portfolio can be resolved through: The responses to Q11 show that ‘relying on one key model’s outcome, supplementing by the others’ is the only item that was agreed on by the experts. Put in contrast, ‘adopting the model that yields the least prudent result’ has not been regarded as an appropriate adoption since seven of the participants disagreed strongly. The above results showed that the the most appropriate strategy in assessing the market risk is to adopt multiple models. Returning to the issue discussed in Q10, it provides an opportunity to reduce the negatives while holding the positives. Previous analysis has disclosed that extreme value theory (EVT) is a relatively new and emerging technique and lacks an undisputed
Chuntao Yu, Bob Davidson and Mohamed Nurullah 151
Table 8.13 Responses to question 11 Items
DS∗ D∗ N∗ A∗ AS∗ Chi-square Significant? Result
Relying on one key model’s outcome, supplementing it by the others
1
2
0
7
1
14.00
@1%
Agree
Adopt the model that yields the most prudent result
0
2
2
4
3
4.00
Not
Null
Using an average value
4
4
3
0
0
7.64
Not
Null
Adopting the model that yields the least prudent result
7
1
2
1
0
14.00
@1%
Disagree
∗ DS: disagree strongly; D: disagree slightly; N: neither agree nor disagree; A: agree slightly; AS: agree strongly.
Table 8.14 Responses to question 12 Items Parametric VaR Historical simulation VaR Monte Carlo VaR
DS∗
D∗
N∗
A∗
AS∗
Chi-square
0 0
0 1
4 4
4 4
3 2
0
0
3
7
1
Significant?
Result
7.64 5.82
Not Not
Null Null
15.82
@1%
Agree
∗ DS: disagree strongly; D: disagree slightly; N: neither agree nor disagree; A: agree slightly; AS: agree strongly.
taxonomy. Q12, Q13 and Q14 were dedicated to explore the appropriate adoption of the model. Q12: It is a good idea to use extreme value theory (EVT) in conjunction with the following models. The results reported in this table confirmed the strategy of combining Monte Carlo VaR and EV-VaR as a good application, as recommended from the previous section. The core procedure of Monte Carlo approaches is to run repeated random simulations, each of which
152 Value at Risk in Emerging Markets
is based on assumed stochastic processes. That is to say, the principles of Monte Carlo simulation can be applied to any appropriate stochastic assumptions. There is no reason why a Monte Carlo simulation cannot be carried out using random variables drawn from an EV distribution. In this way, one can get the combined benefits of both the preciseness of the Monte Carlo method, and the more accurate computation of fat-tail effects presented by EVT. Q13: We should let EVT ‘replace’ the following stress-testing techniques: Q14: We should let EVT be ‘combined’ with the following stress test: The majority of experts disagree with the question asked, implying that EVT should not totally substitute stress-testing. Yet, since the responses show no clear statistical significance, further research on combining EVT with stress-testing techniques is required. Q15: A hybrid method proposed by Boudoukh (Boudoukh et al., 1998), which combines the historical-simulation VaR with the RiskMetrics method, will be better than any single application of either model. Table 8.15 Responses to question 13 Items Sensitivity test Maximum loss optimization Scenario analysis
DS∗ D∗ N∗ A∗ AS∗ Chi-square Significant? Result 6 4
4 2
1 2
0 3
0 0
13.09 4.00
@5% Not
Disagree Null
7
2
1
1
0
14.00
@1%
Disagree
∗ DS: disagree strongly; D: disagree slightly; N: neither agree nor disagree; A: agree slightly; AS: agree strongly.
Table 8.16 Responses to question 14 Items Sensitivity test Maximum loss optimization Scenario analysis
DS∗
D∗
N∗
A∗
AS∗
Chi-square
0 0
0 3
3 4
4 3
4 1
0
2
2
4
3
Significant?
Result
7.64 4.91
Not Not
Null Null
4.00
Not
Null
∗ DS: disagree strongly; D: disagree slightly; N: neither agree nor disagree; A: agree slightly; AS: agree strongly.
Chuntao Yu, Bob Davidson and Mohamed Nurullah 153
Table 8.17 Responses to question 15 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 3.09
degrees of freedom = 4
1 3 4 2 1 Not significant
Combining the above data in 3 categories Disagree Neither Agree Nor Disagree Agree Chi square (χ 2 ) = 0.184
degrees of freedom = 2
4 4 3
Not significant
Table 8.18 Responses to question 16 Categories
Number of respondents
Disagree Strongly Disagree Slightly Neither Agree Nor Disagree Agree Slightly Agree Strongly Chi square (χ 2 ) = 6.73
degrees of freedom = 4
0 3 5 1 2 Not significant
Combining the above data in 3 categories Disagree Neither Agree Nor Disagree Agree Chi square (χ 2 ) = 0.73
degrees of freedom = 2
3 5 3
Not significant
Q16: Combining Monte Carlo and historical-simulation VaR, as proposed in the RiskMetrics Technical Document (3rd ed.) will be better than any single application of either method. With nearly half of the experts neither agreeing nor disagreeing with the statement, Q15 and Q16 gained the most indistinct data in the questionnaire. In fact, the two questions are intended to explore whether the combinations of the multiple models is better
154 Value at Risk in Emerging Markets
than a single adoption, but this is clearly not easy to answer even by our highly-authoritative experts. A combined approach–sometimes referred to as ‘Structured Monte Carlo’–would be first to generate consistent sets of scenarios based on historical volatilities and correlations, or the implied volatilities and correlations observed from the options market. Yet, the RiskMetrics Technical Document did not empirically evaluate this hybrid method. Different from the RiskMetrics Technical Document, Boudoukh et al. (1998) believes that their hybrid approach could combine some of the benefits of extant methods for VaR estimation, such as giving less weight to distant observations (as in the parametric VaR), and account for the empirical tails-skewness and other properties of the observed data series (as in the HS method). Nyluhd’s study (cited in Cuthbertson and Nitzsche, 2001) showed that the mixture of a normal distribution model combined with the RiskMetrics technique performs significantly better than the standard RiskMetrics technique within the context of risk management for univariate equity positions. Yet, it should be noted that both the two new combined approaches may also suffer from the problems of computational complexity, data intensity and programming difficulty. Again, the successful use of a hybrid method needs careful implementation and cautionary evaluation. Maybe these rationales led our respondents to ‘neither agree nor disagree’.
Discussion and recommendation Drawing upon the data and previous analysis, it should be pointed out that selecting market-risk models may be highly complex in practice since there are many competing determinants on how to choose the best model. For example, if a small company prefers a low cost in market-risk assessment, they should not simultaneously demand high accuracy, as fulfilling that goal may be too costly. Again, it is unwise to seek a method of computing VaR to be both accurate and available on a time basis, since there is likely to be an inherent trade-off between these objectives since more rapid methods tend to be less accurate (Pritsker, 1996). It may be argued that the choice of market-risk measures can be examined from a hierarchical perspective. Financial institutions would first structure various selection criteria into a multi-level hierarchy, with managers identifying the
Chuntao Yu, Bob Davidson and Mohamed Nurullah 155
appropriate selection criteria ranked by those factors influencing their choice. The more appropriate criteria frameworks are used, the better the fit, and the best choice will be determined by which dimensions the risk-manager considers most important. Therefore, it should be recommended that knowing the factors that are central considerations in the model choices are of critical important to market-risk practitioners. Secondly, the actual situation and resources available restricts the freedom of selection. For instance, in some cases not enough historical data can be utilized to easily communicate with senior management through a historical simulation VaR report. Therefore, a premise of best selection and use of risk measures should ensure that maximum available resources can be guaranteed. In particular, a strong database should be available, along with intensive IT backing, risk education and training, all of which within the discipline of a well-developed risk infrastructure recommended by the project’s participants. Thirdly, when choosing the criteria framework, financial institutions are further faced with deciding the number of determinants that should be taken into consideration. This is important because as more determinants are included, the fit will improve. However, at some point adding more criteria will result in over-fitting, the fit will be better for the observed portfolio, but, on the other hand, will result in worse predictive performance. Even worse, there may be no answer to the cases where risk managers use a plethora of determinants to choose market-risk measures. Thus, the variance for selecting is a sum of two components: fit and feasibility. As more criteria are included, the fit increases, but the feasibility decreases. When a financial institution limits the number of determinants used for selection, the feasibility increases, yet the fitness decreases. Thus it is necessary to find the right trade-off between reducing variance and enlarging fitness. Fourthly, the approach of model selection involves a dilemma between picking a single best model and using multiple measures. Philosophically, this might arise because one feels that there is a single true best model, and the goal of selection is to identify that model. In some areas of application this approach may be justified, but in other areas one may doubt that an analyst could ever find an exact best model. For example, one may never find the best fit model to include all of the determinants in the selection framework. However,
156 Value at Risk in Emerging Markets
model selection is still required in order to find the closest approximation to the best model. For practical purposes, a financial institution may not need an exact-fit model, but can find a model or models that they feel are sufficiently close to the best model. For example a financial institution may seek a model that can be easily reported to the senior manager and that simultaneously captures the fat-tailed problem. In this case there will be no single best-fit model, but rather a combination of two: one a historical simulation VaR which can be easily understood, the other a EV-VaR which is dedicated for the fattailed distribution. Sometimes, those model combinations that could be successful lead to the creation of a hybrid new method, such as demonstrated in Boudoukh et al. (1998) and the RiskMetrics Technical Document. Even worse, if the combination plan fails or it is hard to find any single fitted model, risk managers have to face the arduous task of developing their own models. For financial giants who have a strong research capacity, they still have an alternative choice between devising new or adopting existing models, even if custom fitted models could be provided. This is the issue that Q6 tried to address in the questionnaire. An obvious issue, then, is the question of whether the model combination will produce a better result than the single-method initiative. Previous data analysis demonstrated that VaR should be best supplemented and combined with other market-risk measures. Yet, experts are controversial in their views that combination might outperform the single-model application. This is not a surprise as the existing literature not only lacks adequate research to this end, but also creates a troubling paradox. The brief review already discussed in the data analysis of Q15 and Q16 demonstrates that undertaking model combination is very challenging. Although the necessity and benefits of pursuing the multi-model scheme is sometimes unquestionable, it should be noted that there are a number of problems such as the computation itself and data intensity, demanding great effort, and the difficulties in finding a counterbalance between multiple outcomes. These difficulties together with the uncertainty as to whether its use can outperform the single-model method precludes many risk managers from attempting the multi-model initiative. It may be argued that only through careful use and successful solution of these problems will the multivariate model framework be a better application.
Chuntao Yu, Bob Davidson and Mohamed Nurullah 157
Another issue that must be addressed for the multiple-model approach is how to utilize the multi-models’ outcomes, to a further extent, to solve the problem that different models yield different results for an identical portfolio – as put forward in the questionnaire. Information from the survey shows the aggregate amount of affirmative answers to the strategy ‘Relying on one key model’s outcome, supplementing by the others’ makes this the normal option for most companies. When compared with the extensive literature, the survey presented the first answer to the conundrum. Yet it may not be adequate. It will be necessary to further decide about the key and supplementing models, the degree by which the assistant model should be utilized, as well as when and where it should be used. Thus it may be recommended that if a multi-model is to be used then a careful designed plan should be employed. Last but not least, this thesis suggests that the selection framework should be fluid rather than deterministic. It has been argued that financial institutions should define the selection framework in and according to a given environment. It is very clear (see Waring and Glendon, 1998) that the environment facing organizations is continuously undergoing change, the stimulus for change coming from the extra-organizational environment in terms of such forces as technology change, economic and market factors, and changes in the corporate risk culture. These factors require frequent alignment of the model-selection framework on a time basis. Take the emergence of the euro as an example. For a risk manager’s favourite historical simulation, this may not be good news as the manager may need to change their risk model since no historical data may be available, particularly since the euro first emerged only in 1998. In fact, adapting to the environment change is itself a determinant for selection, as recommended by two respondents. Adapting these arguments, we suggest that the selection framework should be adjusted regularly, contingent on the type and level of environment uncertainty. Drawing from previous analysis and the survey data, the foregoing section presented a framework for selecting and finding models with a best-fit probability as summarized in Figures 4.1 and 4.2. The data synthesis presented in these two figures is appealing for several reasons. First, it is based on the opinion of authoritative experts. Second, although it may not lead to a best fit for some financial institutions in detailed circumstances, it provides a universal selection scheme,
158 Value at Risk in Emerging Markets
since defining one’s own selection framework may be a difficult task as argued earlier. At its least, it demonstrates and presents an example of how a selection framework may be constructed. With the importance of its qualitative nature, there is a strong rationale to view ‘purpose of analysis’ as the first level in the hierarchy of the selection framework. Guided by this criterion, market-risk models have been modelled in two ways. One is to model the normal market, mainly including various VaR methods and EVT (Figure 8.1). The other is to set a model or models to calculate a baseline against which an extreme event occurred – that is, for the abnormal market (Figure 8.2). For the normal market, the second selection determinant disclosed by the survey is ‘the composition of the portfolio’. From here marketrisk computations are segregated into two basic types of models, for linear and for non-linear portfolios. The next question to ask is whether enough historical data can be utilized for a historical simulation of VaR if unfortunately not, a financial institution will lose the chance of utilising this easiest communication tool for reporting VaR to its senior management. The rest of the selection process will then focus on the assumption of a statistical distribution. Special attention is given to non-linear portfolios as variance–covariance VaR should not be included because of its inability to capture non-linear derivatives such as options. The last two key factors determining which approach should be selected in the normal market framework are given by the response to two questions. Firstly, can the portfolio accept the simplifying assumption of normality – is it reasonable to assume the market movements follow a normal distribution? If so, variance–covariance VaR should be best utilized because of its rapidity and simplicity of implementation. Second, will the extreme-value distributions, such as student-t, be embedded into the statistical assumption? If yes, EVVaR will be the best choice. If the answer is no to the two types of distributions, risk managers will have very limited choice, to only include the Monte Carlo simulation VaR. Abnormal-market risk measures can also be viewed as two major categories – linear and non-linear. For linear abnormal market risk, the model-selection process becomes very straightforward. The factorpush stress-testing method will be the best choice as it enjoys both simplicity and strictness of fit. Non-linear model selection differs
Chuntao Yu, Bob Davidson and Mohamed Nurullah 159
Purpose of analysis For abnormal market (see Figure 8.2) Market risk of normal market
Composition of the portfolio?
Linear
No
Back-testing techniques
Is there enough historical data that can be utilized?
Non-linear
Is there enough historical data that can be utilized?
No Yes
Yes Will management understand advanced statistics? No
HS-VaR* Yes
HS-VaR
Will the extreme event distribution be utilized?
Yes Will assumption of normality be accepted? No Will the extreme event distribution be utilized?
No
Will management understand advanced statistics?
No Yes
Yes EV-VaR*
HS-VaR* (if enough historical data) MC-VaR* (if not enough historical data)
VC-VaR*
Yes * VC-VaR: variance–covariance VaR EV-VaR*
No HS-VaR* (if enough historical data) MC-VaR* (if not enough historical data)
Figure 8.1
* HS-VaR: historical–simulation VaR * MC-VaR: Monte Carlo VaR * EV-VaR: extreme–value VaR
Selecting models for normal market risk
dramatically from the linear group in resource commitment. Previous analysis showed that Maximum Loss Organisation is the more thorough technique than scenario analysis, yet a strong IT backing is required before the decision can be made. If not, the remaining selection will be decided between the historical scenarios and the hypothetical scenarios. In case of no adequate historical data, there
160 Value at Risk in Emerging Markets
Purpose of analysis For normal market (see Figure 8.1) Market risk of abnormal market
Back-testing techniques
Linear
Factor push analysis
Composition of the portfolio?
Non-linear
Have computationally intensive backing?
No
Are there appropriate historical events data that can be utilized?
Yes
Yes
Maximum loss optimization
No Hypothetical scenario analysis
Historical scenario analysis, supplementing with hypothetical scenarios
Figure 8.2
Selecting models for abnormal market risk
will be no other choice but hypothetical scenario analysis. However, if the answer to the question is yes, risk managers should strike an appropriate balance between historical and some form of hypothetical scenarios. The benefits that will be gained from the combination have already been discussed in the previous section. We should especially emphasize that back-testing should accompany any market-risk measurement system, as the process of testing the model’s efficiency against reality. As mentioned earlier, this usually consists of a comparison of the actual losses for a given horizon that are observed by the institution, versus the predicted maximum negative returns (the VaR estimates).
Chuntao Yu, Bob Davidson and Mohamed Nurullah 161
References Allport, G.N. (2000) ‘Managing Risk Organization-Wide’, CA Magazine, vol. 133 (7), pp. 37–8. Bank for International Settlements (BIS) (2001) A Survey of Stress Tests and Current Practices at Major Financial Institutions, Report by a Task Force established by the Committee on Global Financial Systems of the Central Banks of the Group of Ten countries. Bank of International Settlement (1996) Supervising Global Financial Markets May 1996: Submission to the G-7 Lyon Summit of June 1996 (Basle: BIS). Basle Committee on Banking Supervision (BCBS) (1994) Risk Management Guidelines for Derivatives (Basle: BIS). Beder, T.S. (1995) ‘VaR: Seductive but Dangerous’, Financial Analysts Journal (September–October), pp. 12–23. Berkowitz, J. and O’Brien, J. (2002) ‘How Accurate are Value-at-Risk Models at Commercial Banks?’ The Journal of Finance, vol. LVII(3), pp. 1093–111. Bessis, J. (2001) Risk Management in Banking, 2nd edn (Chichester: John Wiley & Sons). Bouchaud, J.P. (2000) ‘Elements for a Theory of Financial Risks’, Physica A., vol. 285, pp. 18–28. Boudoukh, J., Richardson, M. and Whitelaw, R. (1998) ‘The Best of Both Worlds: A Hybrid Approach to Calculating Value at Risk’, International Journal of Theoretical and Applied Finance (May), pp. 64–7. Bulter, C. (1999) Mastering Value at Risk: A Step-by-step Guide to Understanding and Applying VaR (London: Financial Times and Pitman Publishing). Capital Markets Risk Advisers (CMRA) (1996) ‘Risk Standards Overview’. Chew, D. (1996) Managing Derivative Risks: The Use and Abuse of Leverage (Chichester: Wiley). Christoffersen, P., Hahn, J. and Inoue, A. (2001) ‘Testing and Comparing Valueat-Risk Measures’, Journal of Empirical Finance, vol. 8, pp. 325–42. Cornett, M.M. and Saunders, A. (1999) Fundamentals of Financial Institutions Management (Singapore: McGraw-Hill). Culp, C., Miller, M.H. and Neves, A.M.P. (1997) ‘Value-at-Risk: Uses and Abuses’, Journal of Applied Corporate Finance, vol. 10 (Winter), pp. 26–38. Cuthbertson, K. and Nitzsche, D. (2001) Financial Engineering: Derivatives and Risk Management (West Sussex: John Wiley & Sons). Dickson, P.R. and Giglierano, J.J. (1986) ‘Missing the Boat and Sinking the Boat: A Conceptual Model of Entrepreneurial Risk’, Journal of Marketing, vol. 50(3), pp. 58–70. Dowd, K. (1998) Beyond Value at Risk: The New Science of Risk Management (West Sussex: John Wiley & Sons). Dowd, K. (2002 ) An Introduction to Market Risk Measurement (West Sussex: John Wiley & Sons). Engel, J. and Gizycki, M. (1999) ‘Conservatism, Accuracy and Efficiency: Comparing Value-at-Risk Models’, Working Paper, Australian Prudential Regulation Authority.
162 Value at Risk in Emerging Markets
Engle, R.F. and Manganelli, S. (1999) ‘CAViaR: Conditional Value at Risk By Quantile Regression’, National Bureau of Economic Research Working Paper, Cambridge, no. 7341. Ferguson, G.A. and Takane, Y. (1989) Statistical Analysis in Psychology and Education, 6th edn (London: McGraw-Hill). Hague, P. (1993) Questionnaire Design (London: Kogan Page). Härdle, W., Kleinow, T. and Stahl, G. (2002) ‘Applied Quantitative Finance. Theory and Computational Tools’, Working Paper, All but Value-at-Risk Website (online), available from: http://www.gloriamundi.org (Accessed 8 March 2003). Heffernn, S. (1996) Modern Banking in Theory and Practice (Chichester: John Wiley & Sons). J.P. Morgan/Reuters (1995) RiskMetrics – Technical Document, 3rd edn (New York: Morgan Guaranty Trust Company of New York). Jorion, P. (2001) Value at Risk: The New Benchmark for Managing Financial Risk, 2nd edn (New York: McGraw-Hill). Laubsch, A.J. (1999) Risk Management – A Practical Guide, 1st edn (New York: Risk Metrics Group). Marshall, C. and Siegel, M. (1996) ‘Value at Risk: Implementing a Risk Measurement Standard’, Working Paper of the Wharton Financial Institutions Centre, The Warton School, University of Pennsylvania. Mulvey, J.M., Rosenbaum, D.P. and Shetty, B. (1996) ‘Strategic Financial Risk Management and Operations Research’, European Journal of Operational Research, vol. 97, pp. 1–16. Prisker, M. (1995) ‘Evaluating Value at Risk Methodologies: Accuracy Versus Computational Time’, Working Paper of the Wharton School, University of Pennsylvania. Santomero, A.M. (1997) ‘Commercial Bank Risk Management: An Analysis of the Process’, Journal of Financial Services Research, vol. 12(2/3), pp. 83–115. Santomero, A.M. and Babbel, D. (1996) Financial Markets, Instruments and Institutions (Illinois: Irwin). Sarma, M. (2001) ‘Testing EVT-based VaR Measures’, Working Paper, Indira Ghandi Institute for Development Research. Saunders, M., Lewis, P. and Thornhill, A. (2003) Research Methods for Business Students, 3rd edn (Essex: Prentice Hall/Pearson Education). Siegel, S. and Castellan, J.R. (1988) Nonparametric Statistics for the Behavioural Sciences (London: McGraw-Hill). Silverman, D. (1993) Interpreting Qualitative Data (London: Sage). Sinkey, J.F. (2002) Commercial Bank Financial Management – in the FinancialService Industry, 6th ed. (New Jersey: Prentice Hall). Stambaugh, F. (1996) ‘Risk and Value at Risk’, European Management Journal, vol. 14(6), pp. 612–21. Stenhouse, Alexander & Co. (1991) ‘Global Risk Financing & Insurance Survey’, Alexander Stenhouse & Co Research. Thygerson, K. (1995) Management of Financial Institutions (New York: Harper Collins).
Chuntao Yu, Bob Davidson and Mohamed Nurullah 163
Waring, A. and Glendon, A.L. (1998) Managing Risk (London: International Thomson Business Press). Wong, M.C.S., Cheng, W.Y. and Wong, C.Y.P. (2002) ‘Market Risk Management of Banks: Implications from the Accuracy of Value-at-Risk Forecasts’, Journal of Forecasting (January). Zecker, P.D. and Putnam, B.H. (2002) ‘A Short Note on the Forecasts Concept of Risk Management and VaR for Asset Management Firms’, Investor Analytics Institute.
This page intentionally left blank
Index
Akaike, 79, 86 Information Criteria, 65, 66 American Depository Receipt (ADR), 39, 40 Argentina, 42, 44, 45, 50 Augmented Dickey–Fuller (ADF), 64, 65, 66 unit root test 65 Australia, 43, 45, 50 autocorrelation, 2, 62, 64–7, 76, 81 autoregression, 73, 74, 76, 77, 81, 84, 85, 87
coefficient, 44, 65–75, 76, 81 matrices, 111, 114–17, 140–64 method, 39 serial, 60, 62, 64, 65 tests, 60 credit risk, 93, 98, 137 Czech Republic, 43, 45, 50, 51 Delta, 18, 19, 22, 23, 26, 31–3, 140 Denmark, 43, 45, 50 derivatives, 2, 17–19, 31, 33, 91, 96, 100, 108, 136, 140, 158, 161 Dickey–Fuller (DF), 2, 63, 65 unit root test, 63, 65, 66 see also ADF dynamic, 73, 77, 81
back testing, 149, 159, 160 banking, 15, 33, 36, 53, 87, 95, 111, 144, 161, 162 Belgium, 43, 45, 50 Black and Scholes, 19, 20 Brazil, 41, 42, 44, 45, 50, 53 business cycle, 88, 89
eigenvalues, 78 eigenvector, 78 emerging economies, 1, 3, 91 emerging market economies, 94, 95 euro, 157 exchange rates, 103 extreme value, 144, 149–51, 158
call options, 19, 21, 24, 26, 32 premiums, 24–6, 31 Canada, 45, 50 CAPM, 2 cash flow, 3, 71, 72, 79, 80, 85, 148 Choleski factorization, 78 Colombia, 42, 44, 45, 50 confidence interval, 65, 101, 104, 134, 139 continuous, 38, 74, 91, 157 correlation, 2, 13, 24, 40, 41, 44, 45, 52, 53, 59, 60, 62, 64–7, 70, 72, 75–7, 81, 98, 103–5, 108–11, 118–20, 125, 127–33, 154
financial risk management, 92, 150, 162, 164 forecast, 62, 72, 73, 77, 79, 83, 84, 109, 148, 163 France, 43, 45, 50 F-test, 77, 80 gamma, 24, 26–9, 140 Gaussian distribution, 103 Germany, 43, 45, 50
165
166 Index
Granger, 55, 77, 87 causality test, 79–81, 83 cause, 77, 80, 85 sense, 79 growth, 1, 3, 17, 18, 35–7, 48, 52, 58, 61, 72–4, 76, 82, 83, 85, 91, 94 Hong Kong, 43–5, 50 Hungary, 42, 44, 45, 50 hypothesis, 3, 44, 53, 57–9, 62, 65, 68, 69, 72, 77, 79–81 implied volatility, 21, 22, 25, 27, 30–3 index options, 21, 23, 25, 30, 32, 33 India, 1, 2, 17–25, 30, 32, 33, 53, 60, 69 Indonesia, 42–5, 50 inflation, 47, 48, 83, 87, 88 Insider trading, 71–89 see also trading institutional, 3 change, 92 investors, 17, 54 integration, 36, 47–9, 51–5, 68, 101 interest rate, 21, 39, 88, 94, 100 International Finance Corporation (IFC), 37, 40, 53 Israel, 42, 45, 50 Italy, 43, 45, 50 Japan, 43, 45 Karachi Stock Exchange (KSE), 2, 57, 58, 60–2, 64, 65, 67, 68, 70 Korea, 42, 44, 45, 50 lag, 64, 72, 75 Latin America, 41, 54, 55, 60, 69 likelihood, 19, 31, 32, 86, 98, 103, 110, 111 linear, 6, 7, 13, 14, 64, 77, 102, 107, 108, 140, 158–60
MADEX index, 111 market risk, 2, 48, 53, 93, 95, 97–9, 101, 102, 105, 107, 109, 110, 111, 134, 137–45, 148–50, 154–6, 158, 161, 163 Malaysia, 44, 45, 50 Mexico, 42–4, 45, 50, 136 Monte Carlo, 103 simulation, 103, 139, 140, 141, 152, 153, 159 Morgan Stanley & Co. Index (MSCI), 40, 41, 48, 49 multiplier, 77, 81 Netherlands, 43, 45, 50 non-linear, 160 derivatives, 140, 158, 164 models, 158 payoffs, 108 portfolios, 141, 158 risk, 108, 141 non-singular, 78 option Greeks, 2, 17–31 index, 21, 23, 25, 30, 32, 33 pricing, 20, 32 orthogonized, 78, 82 ordinary least squares (OLS), 64 optimal hedging, 5, 8, 9, 11–13, 15 Philippines, 42, 45, 50 polynomial, 77 portfolio, 1–3, 6, 10, 11, 15–7, 31, 35, 36, 38, 44, 46–9, 52–5, 80, 84–6, 92–4, 97, 98, 101–6, 108, 111, 118–32, 134, 136, 139–41, 144, 145, 147, 150, 155, 157, 160 quantitative finance, 162 random walk, 57–70 risk modelling, 144, 149 premium, 8, 9, 13, 14
Index 167
risk – continued specific, 2, 3, 7, 36, 48–53 systematic, 36, 48, 49, 51–3, 112, 113 serial correlation, 59, 60, 62, 64, 65, 70, 75 dependence, 76 simulation, 134, 140, 141 historical, 103, 138, 139, 152, 153, 155–8 Monte Carlo, 103, 138, 139, 152, 153, 155–8 Singapore, 43–5, 50 Spain, 43, 45, 50 South Africa, 42, 44, 45, 50 stress testing, 93, 95, 98, 108–10, 114, 118, 120, 134, 135, 148, 149, 152, 158 structural changes, 36, 40, 88 Taiwan, 42, 44, 45, 50 Thailand, 42, 45, 50
trading, 59, 67–9, 71–3 insider, 71–89 risk, 91–136 volume, 73 triangular matrix, 78 simultaneous equation, 77 Turkey, 42, 45, 50 UK, 43, 45, 50 USA, 43, 45, 50 valuation, 88, 94, 96, 100, 109, 110, 134, 142, 154 value at risk, 1, 3, 93, 95, 98, 100, 102, 107–9, 111, 114, 118, 121–5, 127, 132, 134, 136–62 Venezuela, 39 volatility, 41, 44, 46, 52, 53, 103, 104, 106, 109, 110, 112, 113, 114, 118, 119, 121, 125–33, 135, 140, 156, 164 World Bank, 53–5, 68