Asia-Pacific Financial Markets, Volume 8: Integration, Innovation and Challenges

ASIA-PACIFIC FINANCIALMARKETS: INTEGRATION,

INNOVATION AND CHALLENGES

INTERNATIONAL FINANCEREVIEW

Series Editor : J. Jay Choi

Recent Volumes:

Volume 1: Asian Financial Crisis: Financial, Structural andInternational DimensionsEdited by J. J. Choi

Volume 2: European Monetary Union and Capital MarketsEdited by J. J. Choi and J. Wrase

Volume 3: Global Risk Management: Financial, Operationaland Insurance StrategiesEdited by J. J. Choi and M. Powers

Volume 4: The Japanese Finance: Corporate Finance andCapital Markets in Changing JapanEdited by J. J. Choi and T. Hiraki

Volume 5: Latin American Financial Markets: Developmentsin Financial InnovationsEdited by H. Arbelaez and R. W. Click

Volume 6: Emerging European Financial Markets:Independence and Integration Post-EnlargementEdited by J. A. Batten and C. Kearney

Volume 7: Value Creation in Multinational EnterpriseEdited by J. Jay Choi and Reid W. Click

INTERNATIONAL FINANCE REVIEW VOLUME 8

ASIA-PACIFIC FINANCIALMARKETS: INTEGRATION,

INNOVATION ANDCHALLENGES

EDITED BY

SUK-JOONG KIMThe University of New South Wales, Australia

MICHAEL D. MCKENZIERMIT University, Australia

Amsterdam – Boston – Heidelberg – London – New York – Oxford – Paris

San Diego – San Francisco – Singapore – Sydney – Tokyo

JAI Press is an imprint of Elsevier

JAI Press is an imprint of Elsevier

Linacre House, Jordan Hill, Oxford OX2 8DP, UK

Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands

525 B Street, Suite 1900, San Diego, CA 92101-4495, USA

First edition 2008

Copyright r 2008 Elsevier Ltd. All rights reserved

No part of this publication may be reproduced, stored in a retrieval system

or transmitted in any form or by any means electronic, mechanical, photocopying,

recording or otherwise without the prior written permission of the publisher

Permissions may be sought directly from Elsevier’s Science & Technology Rights

Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333;

email: [email protected]. Alternatively you can submit your request online by

visiting the Elsevier web site at http://www.elsevier.com/locate/permissions, and selecting

Obtaining permission to use Elsevier material

Notice

No responsibility is assumed by the publisher for any injury and/or damage to persons

or property as a matter of products liability, negligence or otherwise, or from any use

or operation of any methods, products, instructions or ideas contained in the material

herein. Because of rapid advances in the medical sciences, in particular, independent

verification of diagnoses and drug dosages should be made

British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-7623-1471-3

ISSN: 1569-3767 (Series)

Printed and bound in the United Kingdom

08 09 10 11 12 10 9 8 7 6 5 4 3 2 1

For information on all JAI Press publications

visit our website at books.elsevier.com

CONTENTS

LIST OF CONTRIBUTORS ix

EDITORIAL ADVISORY BOARD xiii

ABOUT THE SERIES xv

PART I: AN OVERVIEW

CHAPTER 1 INTRODUCTION TO ASIA-PACIFICFINANCIAL MARKETS: INTEGRATION,INNOVATION AND CHALLENGES

Suk-Joong Kim and Michael D. McKenzie 3

PART II: ASIA-PACIFIC STOCK MARKET INTEGRATION

CHAPTER 2 A NEW APPROACH FOR ESTIMATINGRELATIONSHIPS BETWEEN STOCK MARKETRETURNS: EVIDENCE OF FINANCIALINTEGRATION IN THE SOUTHEAST ASIAN REGION

T. J. Brailsford, T. J. O’Neill and J. Penm 17

CHAPTER 3 CORRELATION DYNAMICS BETWEENASIA-PACIFIC, EU AND US STOCK RETURNS

Stuart Hyde, Don Bredin and Nghia Nguyen 39

CHAPTER 4 CONDITIONAL AUTOCORRELATIONAND STOCK MARKET INTEGRATION IN THEASIA-PACIFIC

Suk-Joong Kim and Michael D. McKenzie 63

v

CHAPTER 5 THE IMPACT OF THE OPENING UP OFTHE B-SHARE MARKETS ON THE INTEGRATIONOF CHINESE STOCK MARKETS

Langnan Chen, Steven Li and Weibin Lin 95

CHAPTER 6 A SINGLE CURRENCY FORASEAN-5: AN EMPIRICAL STUDY OF ECONOMICCONVERGENCE AND SYMMETRY

Zhi Lu Xu, Bert D. Ward and Christopher Gan 117

PART III: BUBBLES AND SPILLOVERS

CHAPTER 7 PERIODICALLY COLLAPSINGBUBBLES IN THE ASIAN EMERGING STOCKMARKETS

Ako Doffou 143

CHAPTER 8 CURRENCY CRISES IN ASIA:A MULTIVARIATE LOGIT APPROACH

Jan P. A. M. Jacobs, Gerard H. Kuper and Lestano 157

CHAPTER 9 EVIDENCE OF BUBBLES IN THEMALAYSIAN STOCK MARKET

Gary J. Rangel and Subramaniam S. Pillay 175

PART IV: STOCK MARKETS

CHAPTER 10 ABNORMAL RETURNS AFTERLARGE STOCK PRICE CHANGES: EVIDENCE FROMASIA-PACIFIC MARKETS

Vu Thang Long Pham, Do Quoc Tho Nguyen andThuy-Duong To

205

CONTENTSvi

CHAPTER 11 PRICE LIMITS IN ASIA-PACIFICFINANCIAL MARKETS: THE CASE OF THESHANGHAI STOCK EXCHANGE

Bert Scholtens and Liu Yao 229

CHAPTER 12 CHINA’S SECURITIES MARKETS:CHALLENGES, INNOVATIONS, AND THE LATESTDEVELOPMENTS

Xinyi Yuan, Wei Fan and Qiang Liu 245

CHAPTER 13 TEMPORAL CAUSALITY OFRETURNS OF INDEX FUTURES AND STOCKMARKETS: EVIDENCE FROM MALAYSIA

Wee Ching Pok 263

CHAPTER 14 PRICE BEHAVIOUR SURROUNDINGBLOCK TRANSACTIONS IN STOCK INDEXFUTURES MARKETS: INTERNATIONAL EVIDENCE

Alex Frino, Jennifer Kruk and Andrew Lepone 289

PART V: CORPORATE FINANCE

CHAPTER 15 THE DETERMINANTS OF CAPITALSTRUCTURE: EVIDENCE FROM VIETNAM

Nahum Biger, Nam V. Nguyen and Quyen X. Hoang 307

CHAPTER 16 SHAREHOLDERS’ VALUE CREATIONAND DESTRUCTION: THE STOCK PRICES’ EFFECTSOF MERGER ANNOUNCEMENT IN JAPAN

Ognjenka Zrilic and Yasuo Hoshino 327

CHAPTER 17 TAKEOVERS AND SHAREHOLDERVALUE CREATION ON THE STOCK EXCHANGEOF THAILAND

David E. Allen and Amporn Soongswang 347

Contents vii

PART VI: FUNDS MANAGEMENT

CHAPTER 18 HERD BEHAVIOUR OF CHINESEMUTUAL FUNDS

Jean Jinghan Chen, Xinrong Xiao and Peng Cheng 373

CHAPTER 19 PERFORMANCE PERSISTENCEOF PENSION FUND MANAGERS: EVIDENCE FROMHONG KONG MANDATORY PROVIDENT FUNDS

Patrick Kuok-Kun Chu 393

CHAPTER 20 FINANCIAL MARKET IMPLICATIONSOF INDIA’S PENSION REFORM

Helene K. Poirson 425

PART VII: BANKING AND DEBT MARKETS

CHAPTER 21 ON THE SAFETY AND SOUNDNESSOF CHINESE BANKS IN THE POST-WTO ERA

Lei Xu and Chien-Ting Lin 447

CHAPTER 22 MARKET DISCIPLINE BY CDHOLDERS: EVIDENCE FROM JAPAN WITHA COMPARISON TO THE US

Ayami Kobayashi 471

CHAPTER 23 WHAT ARE THE NEXT STEPS FORBOND MARKET DEVELOPMENT IN THAILAND?

Jonathan A. Batten and Pongsak Hoontrakul 497

CONTENTSviii

LIST OF CONTRIBUTORS

David E. Allen Edith Cowan University, Australia

Jonathan A. Batten Hong Kong University of Scienceand Technology, Hong Kong andMacquarie University, Australia

Nahum Biger Touro University International, USA

T. J. Brailsford University of Queensland, Australia

Don Bredin University College Dublin, Ireland

Langnan Chen Zhongshan University, China

Jean Jinghan Chen University of Surrey, UK

Peng Cheng University of Surrey, UK

Patrick Kuok-KunChu

University of Macau, Macao SAR, China

Ako Doffou Sacred Heart University, USA

Wei Fan University of Electronic Science andTechnology of China, China

Alex Frino University of Sydney, Australia

Christopher Gan Lincoln University, New Zealand

Quyen X. Hoang Touro University International, USA

Pongsak Hoontrakul Chulalongkorn University, Thailand

Yasuo Hoshino Aichi University and University of Tsukuba,Japan

Stuart Hyde University of Manchester, UK

Jan P. A. M. Jacobs University of Groningen, The Netherlands

ix

Suk-Joong Kim University of New South Wales, Australia

Ayami Kobayashi Tokai Tokyo Securities Co., Ltd., Japan

Jennifer Kruk University of Sydney, Australia

Gerard H. Kuper University of Groningen, The Netherlands

Andrew Lepone University of Sydney, Australia

Lestano Atma Jaya Catholic University, Indonesia

Steven Li University of South Australia, Australia

Weibin Lin Zhongshan University, China

Chien-Ting Lin University of Adelaide, Australia

Qiang Liu University of Electronic Science andTechnology of China, China

Michael D. McKenzie RMIT University, Australia

Nghia Nguyen University of Manchester, UK

Do Quoc Tho Nguyen University of New South Wales, Australia

Nam V. Nguyen National Economics University of Hanoi,Vietnam

T. J. O’Neill The Australian National University,Australia

J. Penm The Australian National University,Australia

Vu Thang Long Pham Osaka University, Japan

Subramaniam S.Pillay

The University of Nottingham (MalaysiaCampus), Malaysia

Helene K. Poirson International Monetary Fund, USA

Wee Ching Pok Universiti Teknologi MARA, Malaysia

Gary J. Rangel Altera Corporation (Malaysia) Sdn. Bhd.,Malaysia

Bert Scholtens University of Groningen, The Netherlands

LIST OF CONTRIBUTORSx

Amporn Soongswang Shinawatra University, Thailand

Thuy-Duong To University of Adelaide, Australia

Bert D. Ward Lincoln University, New Zealand

Xinrong Xiao University of Surrey, UK

Lei Xu University of South Australia, Australia

Zhi Lu Xu SIAM Commercial Bank, Singapore

Liu Yao University of Groningen, The Netherlands

Xinyi Yuan University of Electronic Science andTechnology of China, China

Ognjenka Zrilic University of Tsukuba, Japan

List of Contributors xi

This page intentionally left blank

EDITORIAL ADVISORY BOARD

M. AdlerColumbia University, NY, USA

W. BaileyCornell University, Ithaca,NY, USA

I. CooperLondon Business School, UK

J. DoukasOld Dominion University/European Financial Management,Norfolk, VA, USA

G. DufeyUniversity of Michigan,Ann Arbor, MI, USA

V. ErrunzaMcGill University, Montreal,Que., Canada

R. GrosseThunderbird Business School,Glendale, AZ, USA

Y. HamaoUniversity of Southern California,Los Angeles, CA, USA

C.R. HarveyDuke University, Durham,NC, USA

R. HawkinsGeorgia Institute of Technology,Atlanta, GA, USA

J.E. HodderUniversity of Wisconsin,Madison, WI, USA

M. LeviUniversity of British Columbia,Vancouver, BC, Canada

D. LogueDartmouth College, Hanover,NH, USA

J. LothianFordham University, NY, USA

R. MarstonUniversity of Pennsylvania,Philadelphia, PA, USA

R. RollUniversity of California atLos Angeles, CA, USA

A. SaundersNew York University,NY, USA

R. SweeneyGeorgetown University,Washington, DC, USA

xiii

ABOUT THE SERIES

International Finance Review is an annual book series in the internationalfinance area (broadly defined). The IFR, will publish theoretical, empirical,institutional or policy-oriented articles on multinational business financeand strategies, global capital markets and investments, global risk manage-ment, global corporate finance and institutions, currency markets andinternational financial economics, emerging market finance, or relatedregional or country-specific issues. In general, each volume will have aparticular theme. Those interested in contributing an article or editinga volume should contact the Series Editor, J. Jay Choi at E-mail:[email protected]

xv

PART I:

AN OVERVIEW

CHAPTER 1

INTRODUCTION TO ASIA-PACIFIC

FINANCIAL MARKETS:

INTEGRATION, INNOVATION

AND CHALLENGES

Suk-Joong Kim and Michael D. McKenzie

1. OVERVIEW

Perhaps the most significant development in the global business arena in thepost-war period has been the emergence of the Asia-Pacific rim countries asa significant economic force.

The trend began in the late 1950s, when Japan, aided by the US,underwent a highly successful program of industrial development. TheJapanese economy recorded average annual growth rates of around 10% fornearly four decades and ultimately became the second largest economy inthe world. The emergence of Japan was followed less than a decade later bythe Asian Tiger economies of Hong Kong, Singapore, South Korea andTaiwan (also known as Asia’s four dragons). Their focus on educationalreforms and the pursuit of an export oriented growth strategy, provedhighly successful and they were transformed in a relatively short period oftime from countries with low per capita incomes and a small industrial baseto affluent Newly Industrialized Economies (NIEs).

Asia-Pacific Financial Markets: Integration, Innovation and Challenges

International Finance Review, Volume 8, 3–13

Copyright r 2008 by Elsevier Ltd.

All rights of reproduction in any form reserved

ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00001-5

3

dx.doi.org/10.1016/S1569-3767(07)00001-5.3d

A third wave of Asian economic growth originated from Thailand,Malaysia and Indonesia in the 1980s. Following in the footsteps of the AsianNIEs, these countries chose to follow a similar path for economic and, whilethey are yet to achieve the same levels of wealth, they are not very far behind.The most recent chapter in this story is being written by China, who hasemerged from the chaos of the cultural revolution to take its place as one ofthe most significant players in international trade in a remarkably shortperiod of time.1

It is interesting to note that despite over half a century of significant andsustained economic growth, the financial markets of these high-performingAsian economies have only relatively recently emerged as an important partof the global investment community. While economic growth does have animportant role to play in stock market development, other factors haveproven to be more important in the emergence of these countries’ sharemarkets. In the early 1990s, most developed economies were experiencing arecession and expected returns from investment were low. In an attempt toimprove their return on equity, many international fund managers began toseek out alternative investment opportunities. Coincidentally, over thissame period, many developing countries were liberalizing their capitalmarkets. This gave foreigners unprecedented access to a wide range of newand potentially high yielding investment opportunities (see Harvey, 1998).These factors combined to create a situation in which capital flows to theemerging markets sector increased substantially in a remarkably short periodof time.

To highlight this trend, Fig. 1 presents World Bank data on the netportfolio equity flows2 to the emerging markets sector and the East Asia andPacific region. Prior to 1985, portfolio equity flows were frequently netoutflows and relatively small in magnitude (values of between US$1m andUS$5m are typical). From 1986 to 1993, however, a change took place asemerging equity markets found favor with international fund managers. Overthis period, billions of dollars of capital was invested in local share markets,initially in the Latin America and later in South-East Asia. The averageannual rate of increase in portfolio equity flows to emerging equity marketsover this period was 172% and, at their peak in 1993, indirect investmentaccounted for almost 40% of all foreign investment in the emerging marketssector. The 1994 Mexican Peso crisis temporarily dampened internationalfund managers’ enthusiasm for the emerging markets sector and the flows ofequity to emerging markets abated. A resurgence of capital flows to emergingequity markets occurred in the second half of 1995, although this renewedenthusiasm for the sector was to prove short lived. The 1997–1998 Asian

SUK-JOONG KIM AND MICHAEL D. MCKENZIE4

financial crisis sent currency and share markets crashing and portfolio equityflows were negative for the East Asian region over this period. Post financialcrisis, strong returns in the more traditional avenues for investment divertedfund managers’ attention away from the emerging markets sector. Morerecently, there has been a resurgence of interest in emerging markets, causedin part by the poor performance of the US share market following thedot.com crash in 2000 and the spectacular emergence of the Chineseeconomy.

While the Asia-Pacific financial markets have experienced significantinterest from international investors for almost 20 years, they were generallyconsidered to be an alternative source of investment opportunities in contrastto the more traditional investment venues. Recent events have changed thatview however, and Asia-Pacific financial markets are increasingly perceivedas integrated with the world markets. The first key event that contributed tothis change was the 1997 Asian financial crisis. This series of speculativeattacks, which began with the Thai Baht in July, quickly spread to othercurrencies, including some developed countries’ exchange rates (Volume 1 inthis series is devoted solely to considering various aspects of this event). Thesecond important milestone occurred on February 27, 2007 when the Chinesestock market fell 8.8% leading to significant corrections in other markets,most notably the US, which suffered a 3.5% fall. While there have been largemarket corrections in the Asia-Pacific region before, their impact was largely

-10,000

0

10,000

20,000

30,000

40,000

50,000

60,000

70,000

1970 1980 1990 2000

US

$m

All developing countries

East Asia & Pacific

Fig. 1. Net Portfolio Equity Flows. Source: World Bank – Global Development

Finance (April, 2006).

Introduction to Asia-Pacific Financial Markets 5

confined to the emerging markets sector. This event was significant as for thefirst time, the share markets of developed nations responded to events inthe Asia-Pacific region. While the markets did recover expeditiously, thesignificance of this event cannot be denied and clearly demonstrates that theregion and its markets are an important part of the modern global economy.

This volume of International Finance Review focuses on the Asia-Pacificfinancial markets. A total of 22 original chapters, not published elsewhere,have been selected from a competitive field. The chapters utilize a variety ofmethods, including theoretical, empirical and qualitative. Several chaptersoffer combinations of these different categories and among the empiricalchapters, there are a wide variety of datasets analyzed. While China doesplay a significant part in the analysis of five of the chapters in this volume,which is to be expected, a host of other countries are also considered. Thismakes this volume truly international in its scope. These chapters each serveto contribute to the knowledge on a particular issue related to the financialmarkets within this region and for this volume, three issues have beenspecified: integration, innovation and challenges.

The chapters are divided into seven parts, including this introduction. Part IIformalizes much of the sentiment expressed in this introduction andconsiders the issue of Asia-Pacific market integration from the viewpoints ofstock and foreign exchange markets. This integration theme is extended inPart III, where a series of chapters are presented that focus on markets incrisis and the evidence on bubbles and contagion. Part IV contains aselection of chapters that focus more generally on the stock markets withinthe region and their pricing behavior in contrast to more developed markets.Part V focuses on the firms that are listed on stock markets and presents aseries of chapters that consider issues related to corporate finance of Asianfirms. The last section on stock markets, Part VI, has three chapters thatconsider the performance of the funds management industry in the Asia-Pacific region. The last part of this book moves away from the stock marketto consider the banking and debt markets in Part VII.

2. ASIA-PACIFIC STOCK MARKET INTEGRATION

Part II considers the issue of Asia-Pacific stock market integration. Thediscussion begins with a chapter from Tim Brailsford, T. J. O’Neill andJ. Penm entitled ‘‘A New Approach for Estimating Relationships betweenStock Market Returns: Evidence of Financial Integration in the Southeast


Asian region.’’ In this chapter, the authors use a new data weighting processto examine the relationships between stock market returns for Singapore,Thailand, Malaysia, the Philippines and Indonesia. Financial integrationbetween those ASEAN countries and the larger Asia-Pacific region,including the United States, Japan and China, is also considered. Theirfindings indicate that, after the Asian financial crisis, financial integration hascontinued in most ASEAN countries and between ASEAN countries and thelarger Asia-Pacific region. The only exception is Malaysia, where theprogress of financial integration has been relatively slow.

The second chapter in this section, ‘‘Correlation Dynamics between Asia-Pacific, EU and US Stock Returns’’ by Stuart Hyde, Don Bredin and NghiaNguyen, investigates the correlation dynamics of equity markets in theAsia-Pacific, Europe and the US. Using an asymmetric dynamic condi-tional correlation GARCH model, the authors find significant variation incorrelation between markets through time. Stocks exhibit asymmetries inconditional correlations in addition to conditional volatility. Overall, theirfindings provide evidence that is consistent with increasing global marketintegration.

The third chapter on integration is by Suk-Joong Kim and Michael D.McKenzie and is entitled ‘‘Conditional Autocorrelation and Stock MarketIntegration in the Asia-Pacific.’’ This chapter considers the relationshipbetween stock market autocorrelation and the presence of internationalinvestors and stock market volatility. Drawing from a sample of nine Asia-Pacific stock indices, significant evidence of a relationship between thepresence of international investors and the level of stock marketautocorrelation is found. This evidence is consistent with the view thatinternational investors are positive feedback traders and further testing of themodel suggested that the trading strategy of international investors changedas a result of the Asian currency crisis.

The fourth chapter is by Langnan Chen, Steven Li and Weibin Linentitled ‘‘The Impact of the Opening Up of the B-Share Markets on theIntegration of Chinese Stock Markets.’’ The focus of this chapter is on themarket integration impact of allowing domestic investors to access China’sB-share market. The results reveal that while the Chinese stock marketswere segmented before the event, they were integrated to some extent afterthe opening up of B-share markets.

The final contribution to the discussion on the Asia-Pacific marketintegration is provided by Zhi Xu, Bert Ward and Christopher Gan in theirchapter ‘‘Single Currency for ASEAN-5: An Empirical Study of EconomicConvergence and Symmetry.’’ This chapter is a good complement to the


earlier chapters in this section. The authors argue that the feasibility ofadopting a single currency is limited where the national economies are notconverging, or if the responses of national economies to random shocks areasymmetric. According to these two preconditions for a currency union, thischapter tests the economic convergence properties of the ASEAN-5countries, relative to Japan and the USA. The research findings suggestthat Singapore, Malaysia and Thailand (ASEAN-3) appear to be relativelysuitable for forming an optimum currency area, however, it is not obviouswhether the Yen or US dollar should be adopted. The authors conclude thatthere is some way to go before countries in the ASEAN region will be ableto form a mutually beneficial currency union.

3. BUBBLES AND SPILLOVERS

Following on the heels of the discussion of whether Asia-Pacific markets arebecoming more integrated, Part III focuses on markets in crisis and theevidence on bubbles and contagion.

Ako Doffou provides us with a chapter entitled ‘‘Periodically CollapsingBubbles in the Asian Emerging Stock Markets,’’ which investigates theexistence of periodically collapsing bubbles in the Asian emerging stockmarkets. A momentum threshold autoregressive model is used to analyzebubble driven run-ups in stock prices that are followed by a crash and thefindings for 10 Asian emerging stock markets from 1993 to 2005 refute thebubble hypothesis.

Jan Jacobs, Gerard Kuper and Lestano’s chapter entitled ‘‘CurrencyCrises in Asia: A Multivariate Logit Approach’’ attempts to identify anearly warning system for currency crisis by applying factor analysis to arange of indicators suggested by the literature, that are used as explanatoryvariables in logit models. They find that money growth, national savingsand import growth all have important roles to play in forecasting animpending currency crisis.

Subramaniam Pillay and Gary Rangel provide some country specificevidence in their chapter ‘‘Evidence of Bubbles in the Malaysian StockMarket.’’ In this chapter, the authors test for evidence of price bubbles in theMalaysian stock market and find evidence of stock price bubbles. Althoughthe authors stop short of commenting on whether authorities should attemptto ‘‘prick’’ bubbles, they do argue that transparent information dissemina-tion is important in minimizing the impact of such bubbles.


4. STOCK MARKETS

Part IV contains a selection of chapters that focus more generally on thestock markets within the Asia-Pacific region.

The first chapter in this section comes from Vu Thang Long Pham, DoQuoc Tho Nguyen and Thuy-Duong To and is entitled ‘‘Abnormal Returnsafter Large Stock Price Changes: Evidence from Asia-Pacific Markets.’’Their chapter aims to expand the overreaction literature by examiningwhether the price reversals occur in the short and long term following largeone-day price changes. The results suggest that stock prices tend to reverseover the short-term period after large price changes. Further, in the case oflarge price declines, while profitable trading opportunities do exist, they aretypically small and less than the profit generated by passive funds.

Bert Scholtens and Yao Liu’s chapter ‘‘Price Limits in Asia-PacificFinancial Markets: The Case of the Shanghai Stock Exchange’’ considers theprice behavior of stocks listed on the Shanghai Stock Exchange followingdaily limit moves. These limits are designed to reduce price volatility and theyfind weak evidence for the overreaction on prices on the Shanghai market.

Xinyi Yuan, Wei Fan and Qiang Liu provide a chapter that builds on theinnovation and challenges theme of the book. Their chapter titled ‘‘China’sSecurities Markets: Challenges, Innovations, and the Latest Developments’’considers some recent developments in China’s securities markets, namely,the Share Reform, the warrant market, the innovative listed open-end funds(and exchange-traded funds), corporate bonds with detachable warrant,exchange-traded asset-backed securities, are discussed. The discussionfocuses on unique, innovative features of these products, as compared totheir counterparts available in more mature markets and points to possiblefuture research themes.

Wee Ching Pok’s chapter entitled ‘‘Temporal Causality of Return of IndexFutures and Stock Markets: Evidence from Malaysia’’ investigates theimpact change of the composition of market agents on the timing of thearrival of information in Bursa Malaysia. The price discovery role of futurestrading on the spot market is examined. The results find no evidence of asignificant long-run relationship, however futures are found to lead the spotin the short run. This study suggests that the significant change in thecomposition of market agents could contribute to the variation of the lead-lag relationship.

Alex Frino, Jennifer Kruk and Andrew Lepone build on the previouscontribution with their chapter entitled ‘‘Price Behavior Surrounding BlockTransactions in Stock Index Futures Markets: International Evidence.’’ The


authors examine the price impact of large trades in futures markets across14 stock index futures contracts in 11 different international markets. On thebalance, they find that part of the initial price effect of futures trades istemporary. These initial price effects are partially reversed, implying thatthey incur a liquidity premium. They also find strong evidence that largebuyer and seller initiated trades have positive and negative permanenteffects on prices, implying they convey information. These results are in linewith developed market results, which suggest that traders in futures marketsare informed.

5. CORPORATE FINANCE

Part V focuses on the firms that are listed on stock markets and presentsa series of chapters that consider issues related to corporate finance andAsian firms.

Nahum Biger, Nam Van Nguyen and Quyen Xuan Hoang focus onVietnam in their chapter entitled ‘‘The Determinants of Capital Structure:Evidence from Vietnam.’’ Vietnam is currently in a transitional period as theyattempt to dismantle their centrally controlled economy. This study addressesthe question as to whether financing decisions by Vietnamese firms are similarto those observed in economies characterized by market mechanisms andproperty rights. They find that financial leverage in Vietnamese firms iscorrelated with industry characteristics and increases with firm size andmanagerial ownership and decreases with profitability. In contrast to priorempirical studies, for Vietnamese firms, the firm’s leverage decreases withfixed assets and increases with growth opportunities. Further, corporateincome tax has the negative albeit small effect on firm’s financial leverage.

Ognjenka Zrilic and Yasuo Hoshino provide another country specificchapter in their contribution ‘‘Shareholders’ Value Creation and Destruc-tion: The Stock Prices’ Effects of Merger Announcement in Japan.’’ Theyinvestigate the relative importance of different sources of gains and lossesfor Japanese acquirers in the post-bubble period. The authors find anaverage 1.19% cumulative abnormal return in 3 days surrounding themerger announcement. They empirically test value creation, buying growth,hubris and rescue merger hypotheses on their sample of Japanese domesticmergers. They find that differences in the allocation of financial resourcesmay provide a source of value gains. Moreover, mergers with fast-growingtarget are value enhancing when acquirer has prior ownership in target.Consistent with hubris hypothesis, announcement returns are adversely


related to acquirer’s past performance, implying that well-performingacquirers possibly overestimate the true value of deal and overpay to acquirethe target. Finally, returns are significantly positive for mergers announcedafter 1998, indicating that recent deregulation of financial markets resultedin improvement of conditions for merger activity.

David Allen and A. Soongswang take the focus of our analysis in thissection to Thailand in their chapter entitled ‘‘Takeovers and ShareholderValue Creation on the Stock Exchange of Thailand.’’ This study features ananalysis of the impact of successful takeovers and target and bidding firms’performances both before and after the takeover are investigated. Theresults suggest that Thai takeover effects are positive; enhancing bothsuccessful party firms, offeree and offeror, and shareholder wealth. Theevidence also suggests that the market anticipates positive news prior to thetakeover announcement months.

6. FUNDS MANAGEMENT

The last section on stock markets, Part VI, has three chapters that considerthe performance of the funds management industry in the Asia-Pacificregion.

Jean Chen, Xinrong Xiao and Peng Cheng provide further insights intothe Chinese market in their chapter entitled ‘‘Herd Behaviour of ChineseMutual Funds.’’ This study finds that Chinese mutual funds exhibit bothherding in buying and selling behavior. Compared with their Americancounterparts, the Chinese mutual funds exhibit higher level of herding.While firm-specific factors and momentum investment strategies have beenfound to affect US mutual funds herding behavior, none are influential inthe Chinese case. This raises the interesting question of exactly what are thedeterminants of mutual funds investment in the Chinese markets?

The next chapter is ‘‘Performance Persistence of Pension Fund Managers:Evidence from the Hong Kong Mandatory Provident Funds’’ by PatrickChu and it considers fund managers in the special administrative region ofChina. Patrick examines the performance persistence of Hong KongMandatory Provident Fund (MPF) schemes and finds that the raw returns,traditional Jensen alphas and conditional Jensen alphas in the previous yearpossess predictive power. When the funds are classified into high- and low-volatility samples, the high-volatility funds are found to possess strongerperformance persistence. Finally, neither hot-hand nor cold-hand phenom-ena are found in the equity funds managed by same investment manager.


Outside of China, India has also emerged as an important market. Thesecond chapter in this section is authored by Helene Poirson and is entitled‘‘Financial Market Implications of India’s Pension Reform.’’ This chapterfocuses on India’s planned pension reform, which will set up a properregulatory framework for the pension industry and open up the sector toprivate fund managers. Drawing on international experiences, the chapterhighlights pre-conditions for the reform to kick-start financial development,including: the build-up of critical mass; sufficiently flexible investmentguidelines and regulations, including on investments abroad; and concurrentreforms in capital markets. Given the limited scale of the planned reform, thekey challenge for India is to achieve sufficient critical mass early on. Optionsto address this challenge include granting permission for existing workers toswitch to the new system or outsourcing all or part of the reserves of privatesector provident funds to the new pension fund managers.

7. BANKING AND DEBT MARKETS

The last part of this book moves away from the stock market. Part VIIfocuses on the banking and debt markets.

Lei Xu and Chien-Ting Lin present a chapter entitled ‘‘On the Safety andSoundness of Chinese Banks in the Post-WTO Era.’’ This chapter focuses onChina’s accession to World Trade Organization, which opened its financialmarkets to foreign banks. In addition to foreign banks’ expertise andexperience in modern banking activities, they find that they also appear tohave a number of other advantages over Chinese banks in the traditionalareas of business. It is argued that such competition will lead to a loss ofdeposits and loans from local banks in favor of the foreigners. This presentsa potential problem since Chinese banks are currently burdened with largenon-performing loans and low capital adequacy, the entry of foreign bankswill exert further pressure on the banks’ profitability and solvency. Withoutlarger regular bailouts from the central government, it is argued that Chinacould experience a banking crisis in the post-WTO era.

The next contribution by Ayami Kobayashi focuses on Japan. Thischapter entitled ‘‘Market Discipline by CD Holders: Evidence from Japanwith a Comparison to the US’’ tests whether Certificates of Deposit (CD)reflect market participants perception of banks’ failure probabilities. Thechapter develops reduced-form models that describe how interest rates andthe quantity of CDs may be related to banks’ financial measures. Among


the Japanese CAMEL ratings, this chapter finds that CD interest rates aresensitive to the capital adequacy ratio (CAR) and that CD quantitiesare sensitive to ROA. These results suggest that CD holders in Japan aresensitive to bank risks and impose on banks a form of market discipline inaddition to the explicit regulatory discipline.

The final chapter in this section and this volume focuses on Thailand.Jonathan Batten and Pongsak Hoontrakul present a chapter entitled ‘‘Whatare the Next Steps for Bond Market Development in Thailand?’’ Theobjective of this study is to investigate some of the key empirical features ofthe Thai international bond market, which may impede or enhanceinternational bond issuance. The authors focus upon bond return volatilityand skewness as an impediment to international participation in domesticbond markets. They argue that appropriate government policy should focuson stabilizing the macroeconomic environment rather than enhancingdomestic and regional infrastructure.

NOTES

1. This change in the economic landscape has brought with it a number ofinteresting challenges. Most notably it has put pressure on the institutions that wereestablished after WWII to provide international security (the UN and NATO),international monetary stability (the IMF and the World Bank), and orderly inter-national trade (the GATT and its successor, the WTO, as well as the OEEC, whichlater became the OECD). As King (2006) points out, these institutions wereestablished in an era that no longer exists. In 1950, Asian countries accounted foronly a sixth of world GDP, whereas they currently account for more than a third.This change has meant that it is proving increasingly difficult for these institutions toremain relevant.2. Portfolio equity flows are defined as the sum of country funds, depository

receipts (American or global), and direct purchases of shares by foreign investors.

REFERENCES

Harvey, C. R. (1998). The future of emerging markets. NBER Reporter Online, available at

http://www.nber.org/reporter/

King, M. (2006). Through the looking glass: Reform of the international institutions. Inaugural

Melbourne Centre for Financial Studies International Distinguished Lecture available at

www.melbournecentre.com.au/MCFSDistLectureKing06.pdf


PART II:

ASIA-PACIFIC STOCK MARKET

INTEGRATION

CHAPTER 2

A NEW APPROACH FOR

ESTIMATING RELATIONSHIPS

BETWEEN STOCK MARKET

RETURNS: EVIDENCE OF

FINANCIAL INTEGRATION IN

THE SOUTHEAST ASIAN REGION

T. J. Brailsford, T. J. O’Neill and J. Penm

ABSTRACT

In this chapter we use a new data weighting process to examine the

relationships between stock market returns in major Southeast Asian

nations. Investigation is then directed to financial integration between

those ASEAN countries and the larger Asia-Pacific region.

The findings indicate that, after the Asian financial crisis, financial

integration has continued in most ASEAN countries and between ASEAN

countries and the larger Asia-Pacific region. Such effects can be

accounted for by the forgetting factor technique. This new technique will

provide revenue managers with a decision-making tool to evaluate some

complex underlying relationships which managers cannot comprehend

prima facie.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00002-7

17

dx.doi.org/10.1016/S1569-3767(07)00002-7.3d

1. INTRODUCTION

There have been significant attempts by the Association of South East AsianNations (ASEAN) to achieve greater financial and economic integration,including the agreements of the ASEAN Free Trade Area in Bangkok on15 December 1995, and the ASEAN Economic Community in Bali on 7October 2003. The objective of these agreements is to establish a communityamong the ASEAN region characterised by a single market and productionbase, with free flows of goods, services, investment, labour and capital.

To facilitate the integration of regional goods markets, it is necessary tofirst achieve financial integration. An important issue is to understand theprogress of regional financial integration. Because of various institutionalarrangements still in place in the ASEAN region, the progress of financialintegration is likely to vary significantly among member countries. Inaddition to integration within the ASEAN region, significant interestalso exists in the relationships between individual ASEAN countries andmajor world markets, especially those in the Asia-Pacific region includingthe United States, Japan and China. Recent studies on ASEAN financialintegration include Johnson and Soenen (2002) and Phylaktis andRavazzolo (2002).

Investing through stock markets has gained significant popularity in recentyears. Thus, a measure of regional financial integration is the nature of therelationships or co-movements between regional stock market returns. Amajor issue in empirical testing for the progress of financial integrationrelates to the process of integration itself. As financial integration progresses,it is likely that such a process will lead to gradual evolution in the underlyingrelationships among financial market variables. This issue has beenoverlooked in many previous studies. One conventional method to deal withthis issue is to break the sample into periods and compare the estimationresults between sub-samples. This method, however, is not consideredeffective, as no account is given of the evolution within sub-samples.

In this chapter, we propose to include a forgetting factor in the estimationof regional financial integration. The forgetting factor technique is a dataweighting process that allows the estimation to place greater weight on morerecent observations and less weight on earlier data. In such estimation,the effects on the underlying relationships of evolution generated by thefinancial integration process will be accounted for.

In the latter part of the chapter we present the estimation results, with aforgetting factor, of the interrelationships among stock market returns infive ASEAN countries, namely Singapore, Thailand, Malaysia, the

T. J. BRAILSFORD ET AL.18

Philippines and Indonesia. Using the measurement of linear feedbackdeveloped by Geweke (1982), we estimate the progress of financial inte-gration for each country. Using a similar procedure, we also examine theprocess of financial integration between each ASEAN country and the widerAsia-Pacific region, including the United States, Japan and China.

This chapter is structured in the following way. In Section 2, briefdescriptions are given of the forgetting factor technique and data usedfor estimation. The estimation results for financial integration within theASEAN region are given in Section 3. In Section 4, the estimation offinancial integration between each ASEAN country and the wider Asia-Pacific region is discussed. A summary is provided in Section 5.

2. METHODOLOGY AND DATA

2.1. The Forgetting Factor

As briefly mentioned above, researchers interested in the estimation offinancial integration are often concerned that the coefficients of theirestablished systems may not be constant over the sample because theunderlying relationships could evolute as financial integration progresses.This concern has motivated us to utilise the forgetting factor technique inthe estimation of financial integration.

The forgetting factor is a data weighting process that gives more weight torecent observations and less weight to earlier data. The use of forgettingfactor in time series analysis has attracted considerable interest in recentyears. For example, Goto, Nakamura and Uosaki (1995) used the forgettingfactor in the recursive least squares ladder algorithm for spectral estimationof a nonstationary process. Brailsford, Hyung, Penm and Terrell (2004)utilised a forgetting factor in subset autoregressive modelling of the spotaluminium and nickel prices on the London Metal Exchange. The use of theforgetting factor technique to estimation and simulation of financial marketvariables has been reported by Brailsford, Penm and Terrell (2006).

Consider a vector autoregressive (VAR) model of the following form:

yðtÞ þXp

i¼1

Aiyðt� iÞ ¼ �ðtÞ (1)

y(t) is a n� 1 vector of wide-sense stationary series. �ðtÞ is a n� 1 vector ofindependent and identically distributed random process with Ef�ðtÞg ¼ 0

Evidence of Financial Integration in the Southeast Asian Region 19

and Ef�ðtÞ�0ðt� tÞg ¼ S if t ¼ 0 and ¼ 0 if t40: Ai, i=1, y p are n� n

matrices of coefficients. The observations y(t){t=1, y, T} are available.Let lðtÞ ¼ ½ l1ðtÞ . . . . . . lnðtÞ � denotes a 1� n vector associated with time t.

Following Hannan and Deistler (1988), a strategy for determining the valueof the forgetting factor lðtÞ is as follows:

liðtÞ ¼ dZ�tþ1 if 1 � t � Z and liðtÞ ¼ 1 if Zot � T for i ¼ 1; . . . n (2)

Eq. (2) means that ‘forgetting’ of the past occurs from time Z. Noforgetting is involved from time Z+1 to time T. If l=1 for every t, then weobtain the ordinary least squares solution. If 0olo1, the past is weighteddown geometrically from time Z. In theory, the value of d could be differentbetween liðtÞ (a so-called variable forgetting factor). For simplicity, we onlyconsider the case in which the value of d is constant for liðtÞ (a fixedforgetting factor).

This means that the coefficients in Eq. (1) are estimated to minimise:XT

t¼1

lðtÞ½yðtÞ �Xp

i¼1

Aiyðt� iÞ�½yðtÞ �Xp

i¼1

Aiyðt� iÞ�0 (3)

One important issue relating to the use of the forgetting factor inestimation is how to determine the value of d in applications. Theconventional method is based on arbitrary or personal choices. Brailsford,Penm and Terrell (2002), Brailsford et al. (2006) proposed to determine thevalue of d using the bootstrap. In this study, their recommended methodwas adopted for the determination of the value of d. While Brailsford et al.(2002) also proposed a procedure to determine the value of dynamicforgetting factor for nonstationary systems, we have focused on the use ofa fixed forgetting factor in this study, because stock market returns wouldlikely be stationary series (see below).

2.2. Data

In this study, daily observations of stock market returns over the period1 July 1998 to 23 March 2006 are used (2017 observations). These dataare obtained from Datastream. Following Cavoli et al. (2003), we definestock market returns as the difference of the logarithm of the stock marketindex. For consistency, all of the stock market returns are convertedinto US dollar terms through the daily exchange rates. The stock marketindexes used are Straits Times for Singapore, Bangkok SET for Thailand,


Kuala Lumpur Composite for Malaysia, Philippines SE Composite forthe Philippines, Jakarta SE Composite for Indonesia, New York Dow Jonesfor the United States, Nikkei 225 for Japan and Shenzhen SE B Sharesfor China.

Before proceeding with the estimation, we first examine the unitroot property for each series. The results of the augmented Dickey–Fuller test indicate that the series of stock market returns for each countryis stationary. Using the procedure developed by Penm, Penm andTerrell (1997), the test results indicate that there is no co-integratingrelationship among the above stock market indexes in US dollar terms.For brevity, the test statistics are not presented but will be supplied onrequest.

3. ESTIMATION OF ASEAN FINANCIAL

INTEGRATION

In this section, we present the estimation results of the interrelationshipsbetween ASEAN stock market returns. To illustrate the usefulness of theforgetting factor, we first estimate a system that includes stock marketreturns of the above mentioned five ASEAN countries without a forgettingfactor. We then include a forgetting factor in the system. A comparison isundertaken between the estimation results.

As mentioned in the previous section, the series of individual ASEANstock market returns is stationary and there is no co-integrating relationamong them. Given this, we form the ASEAN stock market returns as aVAR system. The specification of this system is determined using theprocedure developed by Brailsford et al. (2002, 2006).

In Table 1, the determined specification for the system without aforgetting factor is presented. The estimated specification suggests that theinterrelationships among ASEAN stock market returns are weak. Based onthe coefficient estimates and associated t-statistics, the process of financialintegration appears more significant in Thailand and the Philippines than inSingapore, Malaysia and Indonesia.

These results appear consistent with findings of previous studies. Cavoliet al. (2003) is a recent study on the issue. They estimate the extent ofASEAN financial integration and conclude that the results for the post-crisisperiod (2000–2001) indicate a fall in the degree of regional integra-tion. However, Malaysia is excluded from their study, due to the capital


Table 1. The VAR Specification for ASEAN Stock Market Returns without a Forgetting Factor.

dSt

dTt

dMt

dPt

dIt

2666666664

3777777775¼

0:093ð3:52Þ

0:049ð2:47Þ

0 0 0

0:183ð5:58Þ

0 0 0 0:040ð2:07Þ

0 0 0 0 0

0:146ð4:61Þ

0:108ð4:46Þ

0 0:111ð4:39Þ

0

0 0:112ð3:55Þ

0 0 0:136ð5:46Þ

26666666666664

37777777777775

dSt�1

dTt�1

dMt�1

dPt�1

dIt�1

2666666664

3777777775þ

0 0 0 0 0

0:063ð2:20Þ

0 0 0 0

0 0:062ð2:35Þ

0 0 0

0 0 � 0:041ð2:32Þ

�0:070ð2:95Þ

0

0 0 0 0 0

266666666664

377777777775

dSt�2

dTt�2

dMt�2

dPt�2

dIt�2

2666666664

3777777775

þ

0 0 � 0:035ð2:18Þ

0 0

0 0 0 0:068ð2:75Þ

0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

26666666664

37777777775

dSt�3

dTt�3

dMt�3

dPt�3

dIt�3

2666666664

3777777775þ

0 0 0 0 0

0 0 � 0:063ð2:97Þ

0 0

0 0 � 0:121ð4:80Þ

0 0:061ð2:83Þ

0 0 � 0:047ð2:32Þ

0 0:048ð2:82Þ

0 0 0 0 0

266666666664

377777777775

dSt�4

dTt�4

dMt�4

dPt�4

dIt�4

2666666664

3777777775

þ

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

0 0:056ð2:65Þ

0 0 0

0 0 0 0 0

26666666664

37777777775

dSt�5

dTt�5

dMt�5

dPt�5

dIt�5

2666666664

3777777775þ

0 0 0 0:063ð3:32Þ

0

0 0 0 0 0

0:102ð2:93Þ

0 � 0:079ð3:10Þ

0 0

0 0 0 0 0

0:076ð1:98Þ

0 � 0:089ð3:17Þ

0 0

2666666666664

3777777777775

dSt�6

dTt�6

dMt�6

dPt�6

dIt�6

2666666664

3777777775

S denotes stock market index in Singapore (in logarithms), T Thailand, M Malaysia, P the Philippines, I Indonesia and d first difference.

Sample period: 1 July 1988 to 23 March 2006. t-statistics in brackets.

T.J.

BRAIL

SFORD

ET

AL.

22

control policy and a fixed exchange rate. Another related study is Johnsonand Soenen (2002). They examine co-movements between Japan and12 Asian stock market returns from 1988 to 1998. They find indicationsthat stock market returns could be integrated between Malaysia andother ASEAN countries. Given the sample used in their study coversmostly observations before capital controls and a fixed exchange rateregime were introduced in Malaysia, their results are not strictly comparableto ours.

For measuring the extent of integration, we employ the measurementof linear feedback introduced by Geweke (1982). In the case of Singapore,for example, we divide the system into two sub-systems, x ¼ ½S�; y ¼

½T ;M;P; I � where S denotes stock market returns in Singapore, T theThailand, M the Malaysia, P the Philippines and I the Indonesia. Toundertake this measurement, we estimate the linear feedback from y to x,ly!x. Within this framework, a higher ly!x means a higher degree ofregional financial integration. The estimated measures of linear feedback arepresented in Table 2.

Similar conclusions can be drawn from the results of linear feedback.Relatively, Thailand and the Philippines appear to exhibit a higher degree ofregional financial integration, while the process of integration seems weakerin other ASEAN countries. Following Brailsford et al. (2002, 2006), we alsoestimate the confidence intervals for these measures. Except for Thailandand the Philippines, the estimates for Singapore, Malaysia and Indonesia

Table 2. Measures of Integration in ASEAN Stock Markets.

Without a Forgetting Factor With a Forgetting Factor Confidence Interval

Integration within ASEAN

Singapore 0.013 0.036 (0.004 0.045)

Thailand 0.040 0.045 (0.016 0.060)

Malaysia 0.015 0.019 (�0.006 0.025)

Philippines 0.055 0.103 (0.058 0.131)

Indonesia 0.015 0.053 (0.017 0.065)

Integration with Asia-Pacific

Singapore 0.023 0.091 (0.051 0.111)

Thailand 0.013 0.043 (0.013 0.058)

Malaysia 0.010 0.017 (�0.004 0.023)

Philippines 0.015 0.066 (0.031 0.082)

Indonesia 0.022 0.031 (0.005 0.045)

Integration is measured using Geweke (1982) and Brailsford et al. (2002, 2006).


are not statistically significant at the 5 per cent level. For brevity, theconfidence intervals are not presented.

The results for Malaysia appear consistent with a priori expectations.Because various institutional arrangements, including capital controls and afixed exchange rate against the US dollar, were introduced since the Asianfinancial crisis in 1997, it is reasonable to expect that the process of regionalfinancial integration has been slow in Malaysia. However, we find the resultsfor Singapore and Indonesia less convincing.

In an attempt to verify these results, we include a forgetting factor in theestimation. As mentioned earlier, we suspect that the process of regionalfinancial integration may have worked as an external disturbance and led togradual evolution in the interrelationships among stock market returns,especially for Singapore and Indonesia. Given the stationarity property ofstock market returns, we employ a fixed forgetting factor in the estimation.

Table 3 presents the specification of the system of ASEAN stock marketreturns with a forgetting factor. Based on Brailsford et al. (2002, 2006), thevalue of d is determined to be 0.995. Z is set at 2015. Again, the Brailsfordet al. (2002, 2006) procedure is used to determine the VAR specification,and the measures of linear feedback, together with the confidence intervals,are estimated. The results are also presented in Table 2 for the purpose ofcomparison.

The estimation results obtained from the specification includinga forgetting factor are different from those without a forgetting factor.While the results consistently indicate that regional financial integrationhas been insignificant in Malaysia, the measures of linear feedback arenoticeably stronger for Singapore, the Philippines and Indonesia. Inparticular, the measures of linear feedback for Singapore, Thailand, thePhilippines and Indonesia are all statistically significant at the 5 per centlevel. This provides strong evidence that regional financial integration hasbeen progressing in these countries. The measure of financial integration forthe Philippines remains the highest, followed by Indonesia, Thailand andSingapore.

These results appear more consistent with a priori expectations. It isinteresting to find that the measures of financial integration are relativelyhigher for the countries more significantly affected by the Asian financialcrisis. During the crisis the Philippines, Indonesia and Thailand requestedand received assistance from the International Monetary Fund. Asmentioned before, the results for Malaysia appear to be related toinstitutional arrangements, which have been barriers to regional financialintegration in that country.


Table 3. The VAR Specification for ASEAN Stock Market Returns with a Forgetting Factor.

dSt

dTt

dMt

dPt

dIt

2666666664

3777777775¼

0:107ð4:43Þ

0 0 0 0

0:143ð3:73Þ

0 0:167ð3:44Þ

�0:077ð2:44Þ

0

0 0 0:127ð5:14Þ

0 0:037ð2:73Þ

0:097ð3:12Þ

0:071ð3:53Þ

� 0:098ð2:53Þ

0:125ð4:98Þ

0:121ð5:74Þ

0:101ð2:80Þ

0:077ð3:22Þ

0 0 0:098ð3:86Þ

266666666666664

377777777777775

dSt�1

dTt�1

dMt�1

dPt�1

dIt�1

2666666664

3777777775þ

�0:051ð2:18Þ

0 0 0:079ð3:82Þ

0

0 0:065ð2:71Þ

0 0 0

0 0 0 0 0

0 0 0:122ð3:38Þ

�0:064ð2:58Þ

0

0:106ð3:09Þ

0 0 0 � 0:106ð4:26Þ

26666666666664

37777777777775

dSt�2

dTt�2

dMt�2

dPt�2

dIt�2

2666666664

3777777775

þ

0 0 0 0 0

0 � 0:073ð2:98Þ

0 0:098ð3:14Þ

0

0:052ð2:78Þ

� 0:064ð4:81Þ

0 0 0

� 0:084ð3:01Þ

0:118ð5:87Þ

0 0 0

0 0:063ð2:69Þ

0 0:103ð3:50Þ

0

26666666666664

37777777777775

dSt�3

dTt�3

dMt�3

dPt�3

dIt�3

2666666664

3777777775þ

0 0 0 0 � 0:052ð3:10Þ

0 0:055ð2:23Þ

0 0 � 0:074ð2:80Þ

0 0 0 0 0

0 0 0 0 0

0 0 � 0:109ð2:55Þ

0 0

2666666666664

3777777777775

dSt�4

dTt�4

dMt�4

dPt�4

dIt�4

2666666664

3777777775

þ

0 0 0 0 0

� 0:133ð3:80Þ

0:180ð7:27Þ

0 0 0

0 0 0:060ð2:50Þ

0 0

0 0 0 0 0

0 0 0:152ð3:53Þ

0 0

2666666666664

3777777777775

dSt�5

dTt�5

dMt�5

dPt�5

dIt�5

2666666664

3777777775þ

� 0:093ð3:91Þ

0:036ð2:13Þ

0 0:096ð4:71Þ

0

0 0 � 0:123ð2:69Þ

0 � 0:078ð3:09Þ

0 0 0 0 0

� 0:118ð4:33Þ

0:081ð4:08Þ

0 0 0

0 0 0 0 0

266666666664

377777777775

dSt�6

dTt�6

dMt�6

dPt�6

dIt�6

2666666664

3777777775

S denotes stock market index in Singapore (in logarithms), T Thailand, M Malaysia, P the Philippines, I Indonesia and d first difference.

Sample period: 1 July 1988 to 23 March 2006. t-statistics in brackets. The value of the forgetting factor is 0.995.

Evid

ence

of

Fin

an

cial

Integ

ratio

nin

the

So

uth

east

Asia

nR

egio

n25

4. INTEGRATION WITH THE ASIA-PACIFIC

MARKETS

In the section, we again use the forgetting factor technique to investigate theprogress of financial integration between ASEAN and the important Asia-Pacific markets, namely the United States, Japan and China. In recent years,this issue has attracted significant research interest with previous studiesincluding Moosa and Bhatti (1997) and Wong (1995).

A similar approach is undertaken in this investigation. Given the largenumber of variables involved, we adopt a novel approach by estimating theinterrelationships between stock market returns in individual ASEANcountry and those in the United States, Japan and China. Two estimationprocedures are adopted: one is without a forgetting factor and the otherwith a forgetting factor. In each system, the VAR specifications and thevalue of the forgetting factor are determined using the Brailsford et al.(2002, 2006) procedure. For each system, the determined value of theforgetting factor is similar (see Tables 9–13).

Tables 4–8 present the estimation results without a forgetting factor foreach ASEAN country. Again, the linear feedback from movements in stock

Table 4. Interrelationships between Stock Market Returns in theUnited States, Japan, China and Singapore without a Forgetting Factor.

Singapore

dSt

dUSt

dJt

dCt

2666664

3777775 ¼0:101ð3:93Þ

0:112ð4:23Þ

0 0

0:141ð6:15Þ

� 0:073ð2:86Þ

0 0

0:116ð3:65Þ

0 � 0:083ð3:24Þ

0

0:145ð3:51Þ

0 0 0:105ð4:10Þ

26666666664

37777777775

dSt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0 0:089

ð3:35Þ0 0

0 0 0 0

0 0 0 0

0 0 0 0:092ð3:58Þ

266666664

377777775

dSt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775

þ

0 0 0 0

0:048ð2:32Þ

0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dSt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775þ0 0 0:075

ð3:79Þ0

0 0 0 0

0 0 0 0

0 0 0 0:064ð2:49Þ

266666664

377777775

dSt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

S denotes stock market index in Singapore (in logarithms), US the United States, J Japan,

C China and d first difference. Sample period: 1 July 1988 to 23 March 2006. t-statistics in

brackets.


Table 5. Interrelationships between Stock Market Returns in the United States, Japan, China and Thailandwithout a Forgetting Factor.

Thailand

dTt

dUSt

dJt

dCt

2666664

3777775 ¼0:076ð3:05Þ

0 0 0

0:033ð1:92Þ

�0:059ð2:29Þ

0:059ð2:96Þ

0

0 0 � 0:058ð2:33Þ

0

0:093ð2:90Þ

0 0 0:102ð4:00Þ

26666666664

37777777775

dTt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0:050ð1:97Þ

0:117ð3:09Þ

0 0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dTt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0 0:094ð2:53Þ

0 0

0 0 0 0

0 0 0 0

0 0 0 0:093ð3:24Þ

266666664

377777775

dTt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ0 0:077

ð2:08Þ0 0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dTt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775

þ

0 0 0 0

0:034ð2:08Þ

0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dTt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775þ0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:066ð2:55Þ

26666664

37777775dTt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

T denotes stock market index in Thailand (in logarithms), US the United States, J Japan, C China and d first difference. Sample period: 1 July

1988 to 23 March 2006. t-statistics in brackets.

Evid

ence

of

Fin

an

cial

Integ

ratio

nin

the

So

uth

east

Asia

nR

egio

n27

Table 6. Interrelationships between Stock Market Returns in the United States, Japan, China and Malaysiawithout a Forgetting Factor.

Malaysia

dMt

dUSt

dJt

dCt

2666664

3777775 ¼0 0 0 0

� 0:035ð2:17Þ

0 0:071ð3:68Þ

0

0:054ð2:52Þ

0 � 0:059ð2:29Þ

0

0:113ð3:75Þ

0 0 0:090ð3:50Þ

2666666664

3777777775

dMt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0 0 0:087

ð2:92Þ0

0 0 0 0

0 0 0 0

0:061ð2:03Þ

0 0 0

266666664

377777775

dMt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:086ð3:34Þ

26666664

37777775dMt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ� 0:097ð3:88Þ

0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:092ð3:56Þ

266666664

377777775

dMt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775þ0:064ð2:51Þ

0:083ð1:98Þ

� 0:074ð2:37Þ

0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dMt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775

þ

� 0:063ð2:51Þ

0 0 0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dMt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775þ0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:055ð2:13Þ

26666664

37777775dMt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

M denotes stock market index in Malaysia (in logarithms), US the United States, J Japan, C China and d first difference. Sample period:

1 July 1988 to 23 March 2006. t-statistics in brackets.

T.J.

BRAIL

SFORD

ET

AL.

28

Table 7. Interrelationships between Stock Market Returns in the United States, Japan, China and thePhilippines without a Forgetting Factor.

The Philippines

dPt

dUSt

dJt

dCt

2666664

3777775 ¼0:170ð6:71Þ

0 0:063ð2:40Þ

0

0 0:062ð3:24Þ

0 0

0 0 � 0:050ð1:93Þ

0

0:113ð3:12Þ

0 0 0:098ð3:81Þ

26666666664

37777777775

dPt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ� 0:068ð2:68Þ

0:077ð2:26Þ

0 0

0 0 0 0

0 0 0 0

�0:081ð2:24Þ

0 0 0

266666664

377777775

dPt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0 0 0 � 0:036ð2:03Þ

0 0 0 0

0 0 0 0

0 0 0 0:086ð3:35Þ

266666664

377777775

dPt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:092ð3:58Þ

26666664

37777775dPt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775

þ

0 0:104ð3:05Þ

� 0:056ð2:21Þ

0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dPt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775þ0 0 0 0

0 0 0 0

0 0 0 0

0 0 0 0:061ð2:37Þ

26666664

37777775dPt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

P denotes stock market index in the Philippines (in logarithms), US the United States, J Japan, C China and d first difference. Sample period:

1 July 1988 to 23 March 2006. t-statistics in brackets.

Evid

ence

of

Fin

an

cial

Integ

ratio

nin

the

So

uth

east

Asia

nR

egio

n29

market returns in the United States, Japan and China to individual ASEANmarket is measured. For ease of comparison, the results of linear feedbackare also presented in Table 2.

It is important to note that the estimation results without a forgettingfactor indicate that the interrelationships between stock market returns inASEAN and those in the wider Asia-Pacific region are not strong. All theestimated measures of linear feedback are insignificant at the 5 per cent level.

In Tables 9–13, the results for the estimation with a forgetting factor arepresented. Notably, the results with a forgetting factor are significantlydifferent from those obtained without a forgetting factor. The linearfeedback and associated confidence intervals are also presented in Table 2.Except for Malaysia, the estimated measures are statistically significant forSingapore, Thailand, the Philippines and Indonesia.

Again, we find that the estimation results with a forgetting factor aremore consistent with a priori expectations. Singapore is found to berelatively more integrated with the larger Asia-Pacific region, with thehighest estimate of linear feedback. Measured by the linear feedback,Singapore also appears more financially integrated with the Asia-Pacific

Table 8. Interrelationships between Stock Market Returns in theUnited States, Japan, China and Indonesia without a Forgetting Factor.

Indonesia

dIt

dUSt

dJt

dCt

2666664

3777775 ¼0:171ð6:81Þ

0 0 0

0 0 0:068ð3:56Þ

0

0:044ð2:41Þ

0 � 0:056ð2:18Þ

0

0 0 0 0:097ð3:78Þ

26666666664

37777777775

dIt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0 0:106

ð2:29Þ0 � 0:051

ð2:10Þ

0 0 0 0

0 0 0 0

0 0 0 0:085ð3:29Þ

266666664

377777775

dIt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775

þ

0 0 0 0

0 0 0 0

� 0:042ð2:38Þ

0 0 0

0 0 0 0:087ð3:38Þ

266666664

377777775

dIt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775þ0 0 0:140

ð4:03Þ0

0:027ð2:03Þ

0 0 0

0 0 0 0

0 0 0 0:09ð3:72Þ

2666666664

3777777775

dIt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

I denotes stock market index in Indonesia (in logarithms), US the United States, J Japan,


brackets.


Table 9. Interrelationships between Stock Market Returns in the United States, Japan, China and Singaporewith a Forgetting Factor.

Singapore

dSt

dUSt

dJt

dCt

2666664

3777775 ¼0:170ð6:76Þ

0 0 0

0:244ð10:66Þ

� 0:212ð8:63Þ

0:062ð4:21Þ

0

0:286ð7:68Þ

0 0 0

0 0 0 0

266666664

377777775

dSt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ� 0:131ð5:18Þ

0:175ð6:27Þ

0 0

0 0 0 0

� 0:136ð3:86Þ

0 0 � 0:088ð3:77Þ

0:158ð4:24Þ

� 0:139ð3:16Þ

0 0

2666666664

3777777775

dSt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775þ0 0:130

ð4:82Þ0 0:045

ð2:85Þ

0 0 0 0

0 0 � 0:061ð2:71Þ

0

0 0 0:084ð3:52Þ

0

2666666664

3777777775

dSt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775

þ

0 0:106ð3:93Þ

0 0:040ð2:51Þ

0:090ð4:44Þ

0 0 0

0:120ð3:43Þ

0 0 0

0 0 0 �0:054ð2:17Þ

26666666664

37777777775

dSt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775þ0 0 � 0:075

ð4:40Þ0

� 0:048ð2:20Þ

0 � 0:079ð5:18Þ

0

� 0:091ð2:44Þ

0 � 0:092ð3:39Þ

� 0:060ð2:58Þ

0 0 0 � 0:149ð5:95Þ

26666666664

37777777775

dSt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775

þ

� 0:100ð3:80Þ

0 0:089ð5:00Þ

0

0 0 0 0

0:092ð2:38Þ

� 0:152ð3:61Þ

0:107ð4:04Þ

� 0:115ð4:95Þ

0 0 0 0

26666664

37777775dSt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775þ0 � 0:137

ð5:10Þ0 0

0:093ð4:49Þ

0 � 0:112ð7:78Þ

0

0 0 0 0

0:178ð5:09Þ

0 0 0:085ð3:42Þ

2666666664

3777777775

dSt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

S denotes stock market index in Singapore (in logarithms), US the United States, J Japan, C China and d first difference. Sample period:

1 July 1988 to 23 March 2006. t-statistics in brackets. The value of the forgetting factor is 0.99.

Evid

ence

of

Fin

an

cial

Integ

ratio

nin

the

So

uth

east

Asia

nR

egio

n31

Table 10. Interrelationships between Stock Market Returns in the United States, Japan, China and Thailandwith a Forgetting Factor.

Thailand

dTt

dUSt

dJt

dCt

2666664

3777775 ¼0 0:105

ð2:89Þ0 � 0:048

ð1:97Þ

0:142ð5:61Þ

0 0:105ð6:08Þ

0

0 0 0 0

0:100ð3:83Þ

0 0 0

2666666664

3777777775

dTt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0 0:147

ð4:06Þ0 0

0 0 0 0

0 0 0 � 0:048ð2:06Þ

0 0 0 0

26666664

37777775dTt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

� 0:062ð2:52Þ

0:153ð4:20Þ

0 0:077ð3:19Þ

0 0 0 0

0 0 0 0

0 0 0:061ð23:4Þ

0

266666664

377777775

dTt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ0 0 0:082

ð3:34Þ0

0:036ð2:21Þ

0 0 0

0 0:074ð2:13Þ

0 0

0 0 0 0

266666664

377777775

dTt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775

þ

0:117ð4:77Þ

0 0 0

0 0 � 0:066ð3:84Þ

0

0 0 � 0:073ð2:89Þ

0

0 0 � 0:054ð2:08Þ

� 0:081ð3:20Þ

26666666664

37777777775

dTt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775þ0 0 0 0

0:039ð2:28Þ

0 0 0

0:061ð2:42Þ

� 0:093ð2:53Þ

0:069ð2:71Þ

� 0:068ð2:90Þ

0 0 0:073ð2:81Þ

0

2666666664

3777777775

dTt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775

þ

0 0 0 0

0:038ð2:17Þ

0 � 0:069ð4:08Þ

0

0:095ð3:80Þ

0 0 0

0 0 0 0:089ð3:52Þ

2666666664

3777777775

dTt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

T denotes stock market index in Thailand (in logarithms), US the United States, J Japan, C China and d first difference. Sample period: 1 July

1988 to 23 March 2006. t-statistics in brackets. The value of the forgetting factor is 0.995.

T.J.

BRAIL

SFORD

ET

AL.

32

region than with other ASEAN countries (0.091 versus 0.036). In contrast,the Philippines and Indonesia seem more financially integrated with otherASEAN countries than with the wider Asia-Pacific region. Thailand appearswell integrated with both the ASEAN and Asia-Pacific regions. Again,Malaysia is lagging behind in the process of financial integration with thewider Asia-Pacific region.

5. SUMMARY

In this chapter, we demonstrate the usefulness of the forgetting factortechnique in the investigation of financial integration. We use a forgettingfactor to account for evolution in the interrelationships among financial

Table 11. Interrelationships between Stock Market Returns in theUnited States, Japan, China and Malaysia with a Forgetting Factor.

Malaysia

dMt

dUSt

dJt

dCt

2666664

3777775 ¼0:140ð5:42Þ

� 0:075ð3:68Þ

0:038ð2:70Þ

0:026ð2:07Þ

0 � 0:142ð5:60Þ

0:106ð6:10Þ

0

0 0 0 0

0:151ð3:08Þ

0 0 0

2666666664

3777777775

dMt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ0 0:046

ð2:45Þ0 0

0 0 0 0

0 0 0 0

0 0 0 0

26666664

37777775dMt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0 0 0 0

0 0 0 0

0:227ð4:91Þ

0 � 0:065ð2:65Þ

0

0 0 0:058ð2:23Þ

0

266666664

377777775

dMt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ0:068ð2:81Þ

0 0 0

0 0 � 0:058ð3:42Þ

0

0 0 � 0:064ð2:58Þ

0

0 0 0 � 0:080ð3:15Þ

26666666664

37777777775

dMt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775

þ

0 0 0 0

0 0 0 0

0 0 0 � 0:059ð2:54Þ

0 0 0:072ð2:76Þ

0

266666664

377777775

dMt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775þ0 0 0 0

0 0 � 0:058ð3:55Þ

0

0 0 0 0

0 0 0 0:084ð3:32Þ

266666664

377777775

dMt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

M denotes stock market index in Malaysia (in logarithms), US the United States, J Japan,


brackets. The value of the forgetting factor is 0.995.


Table 12. Interrelationships between Stock Market Returns in the United States, Japan, China and thePhilippines with a Forgetting Factor.

The Philippines

dPt

dUSt

dJt

dCt

2666664

3777775 ¼0:203ð7:84Þ

0 0:054ð2:68Þ

�0:064ð3:24Þ

0:045ð2:35Þ

� 0:187ð7:54Þ

0:113ð7:76Þ

0

0:094ð3:00Þ

0 0 0

0 0 0:061ð2:56Þ

� 0:067ð2:67Þ

26666666664

37777777775

dPt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ� 0:056ð2:19Þ

0:097ð2:82Þ

0 0

0:039ð2:29Þ

0 0 0

0 0 0 � 0:090ð3:68Þ

0:169ð5:33Þ

� 0:166ð3:83Þ

0 0

26666666664

37777777775

dPt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0 0:201ð5:77Þ

� 0:046ð2:32Þ

0:045ð2:36Þ

0 0:078ð3:34Þ

0 0

0 0 � 0:065ð2:77Þ

0

� 0:062ð1:98Þ

0 0:098ð3:98Þ

0

26666666664

37777777775

dPt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ� 0:065ð4:40Þ

0 0 � 0:059ð3:06Þ

0 0 0 0

0:138ð3:10Þ

� 0:089ð2:08Þ

0 0

0 0 � 0:080ð3:42Þ

0

2666666664

3777777775

dPt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775

þ

0 0 0 0

� 0:061ð3:46Þ

0 � 0:077ð5:46Þ

0

0:079ð2:43Þ

� 0:118ð2:66Þ

� 0:097ð3:39Þ

�0:082ð3:39Þ

0 0 0 � 0:146ð6:00Þ

2666666664

3777777775

dPt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775þ0 0 0 0

0 0 0 0

0 � 0:205ð4:67Þ

0:124ð5:00Þ

� 0:143ð5:98Þ

0 0:118ð2:84Þ

0 0

266666664

377777775

dPt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775

þ

0 � 0:111ð3:25Þ

0 � 0:058ð2:96Þ

0:049ð2:96Þ

0 � 0:078ð5:49Þ

0

0 � 0:122ð2:73Þ

0:104ð4:04Þ

0

0 0:140ð3:33Þ

0 0:079ð3:16Þ

26666666664

37777777775

dPt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

P denotes stock market index in the Philippines (in logarithms), US the United States, J Japan, C China and d first difference. Sample period:

1 July 1988 to 23 March 2006. t-statistics in brackets. The value of the forgetting factor is 0.99.

T.J.

BRAIL

SFORD

ET

AL.

34

Table 13. Interrelationships between Stock Market Returns in the United States, Japan, China andIndonesia with a Forgetting Factor.

Indonesia

dIt

dUSt

dJt

dCt

2666664

3777775 ¼0:162ð6:34Þ

0 0 0:053ð2:30Þ

0:073ð3:87Þ

� 0:145ð5:72Þ

0:091ð5:16Þ

0

0:093ð3:47Þ

0 0 0

0 0:098ð2:60Þ

0 0

26666666664

37777777775

dIt�1

dUSt�1

dJt�1

dCt�1

2666664

3777775þ� 0:088ð3:53Þ

0:189ð5:45Þ

0 0

0 0 0 0

0 0 0 � 0:065ð2:81Þ

0 0 0 0

26666664

37777775dIt�2

dUSt�2

dJt�2

dCt�2

2666664

3777775

þ

0:072ð2:82Þ

0 0:050ð2:08Þ

0

0 0 0 0

0 0 0 0

0 0 0:062ð2:39Þ

0

266666664

377777775

dIt�3

dUSt�3

dJt�3

dCt�3

2666664

3777775þ0 0 0:058

ð2:47Þ0

0 0 0 0

� 0:076ð3:10Þ

0 0 0:058ð2:51Þ

0:071ð2:60Þ

0 0 0

2666666664

3777777775

dIt�4

dUSt�4

dJt�4

dCt�4

2666664

3777775

þ

0 0 0 0

0 0 � 0:073ð4:22Þ

0

0 0 � 0:089ð3:58Þ

0

� 0:123ð4:50Þ

0 0 � 0:082ð3:24Þ

2666666664

3777777775

dIt�5

dUSt�5

dJt�5

dCt�5

2666664

3777775þ0:054ð2:16Þ

0 0 0

0 0 0 0

0:109ð4:34Þ

0 0 � 0:071ð3:09Þ

0 0 0:083ð3:21Þ

0

2666666664

3777777775

dIt�6

dUSt�6

dJt�6

dCt�6

2666664

3777775

þ

0:078ð3:14Þ

0 0 0

0 0 � 0:060ð3:64Þ

0

0 0 0 0

0 0 0 0:094ð3:72Þ

2666666664

3777777775

dIt�7

dUSt�7

dJt�7

dCt�7

2666664

3777775

I denotes stock market index in Indonesia (in logarithms), US the United States, J Japan, C China and d first difference. Sample period: 1 July

1988 to 23 March 2006. t-statistics in brackets. The value of the forgetting factor is 0.99.

Evid

ence

of

Fin

an

cial

Integ

ratio

nin

the

So

uth

east

Asia

nR

egio

n35

market variables generated by the financial integration process. The resultsare clear. In the two systems we examine, the estimation results using aforgetting factor improve significantly and are more consistent with a priori

expectations.Some other important findings are obtained from the estimations.

Singapore is found to be more financially integrated with the Asia-Pacificregion than with other ASEAN countries, while the Philippines andIndonesia are more financially integrated with the ASEAN region. Thailandis well integrated with both ASEAN countries and the Asia-Pacific markets.In contrast, the process of financial integration has been slow andinsignificant in Malaysia.

There is scope to expand the use of the forgetting factor in the study offinancial integration. In this chapter, we measure the extent of financialintegration by examining the dynamics between stock market returns. Othercommonly used measures of financial integration based on market pricesinclude the purchasing power parity and the uncovered interest rate parityconditions. Other measures, such as savings–investment correlations andconsumption correlations, have also been used to measure the extent offinancial integration. The applicability of the forgetting factor technique tothose measures would also be of interest, but is not part of this study.

REFERENCES

Brailsford, T. J., Hyung, N., Penm, J. H. W., & Terrell, R. D. (2004). The sequential fitting of

subset auto regressions with a forgetting factor. Journal of Economic Research, 9, 29–57.

Brailsford, T. J., Penm, J. H. W., & Terrell, R. D. (2002). Selecting the forgetting factor in

subset autoregressive modelling. Journal of Time Series Analysis, 23, 629–650.

Brailsford, T. J., Penm, J. H. W., & Terrell, R. D. (2006). Kernel bandwidth applications to US

mutual fund and Euro movements. Research in Finance, 23, 81–98.

Cavoli, T., Rajan, R., & Siregar, R. (2003). A survey of financial integration in East Asia: Trends,

issues and implications. Report prepared for the Regional Economic Monitoring Unit of

the Asian Development Bank (January).

Geweke, J. (1982). Measurement of linear dependence and feedback between multiple time

series. Journal of the American Statistical Association, 77(378), 304–313.

Goto, S., Nakamura, M., & Uosaki, K. (1995). On-line spectral estimation of nonstationary

time series based on AR model parameter estimation and order selection with a

forgetting factor. IEEE Transactions on Signal Processing, 43, 1519–1522.

Hannan, E. J., & Deistler, M. (1988). The statistical theory of linear systems. New York: Wiley.

Johnson, R., & Soenen, L. (2002). Asian economic integration and stock market comovement.

The Journal of Financial Research, XXV(1), 141–157.

Moosa, I., & Bhatti, R. (1997). Are Asian markets integrated? International Economic Journal,

11(1), 51–67.


Penm, J. H., Penm, J. H. W., & Terrell, R. D. (1997). On the selection of subset cointegrating

vectors in vector error-correction modelling. Econometric Reviews, 16(3), 281–304.

Phylaktis, K., & Ravazzolo, F. (2002). Measuring financial and economic integration with

equity prices in emerging markets. Journal of International Money and Finance, 21,

879–903.

Wong, J. (1995). China’s economic reform and open-door policy viewed from South East Asia.

ASEAN Economic Bulletin, 11(3), 269–279.


CHAPTER 3

CORRELATION DYNAMICS

BETWEEN ASIA-PACIFIC, EU

AND US STOCK RETURNS

Stuart Hyde, Don Bredin and Nghia Nguyen

ABSTRACT

This chapter investigates the correlation dynamics in the equity markets

of 13 Asia-Pacific countries, Europe and the US using the asymmetric

dynamic conditional correlation GARCH model (AG-DCC-GARCH)

introduced by Cappiello, Engle, and Sheppard (2006). We find significant

variation in correlation between markets through time. Stocks exhibit

asymmetries in conditional correlations in addition to conditional

volatility. Yet asymmetry is less apparent in less integrated markets.

The Asian crisis acts as a structural break, with correlations increasing

markedly between crisis countries during this period though the bear

market in the early 2000s is a more significant event for correlations with

developed markets. Our findings also provide further evidence consistent

with increasing global market integration. The documented asymmetries

and correlation dynamics have important implications for international

portfolio diversification and asset allocation.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00003-9

39

dx.doi.org/10.1016/S1569-3767(07)00003-9.3d

1. INTRODUCTION

Over recent years there has been a large amount of research focussed onlinkages between asset markets in developed economies and emergingmarkets. The level of interaction or interdependence between markets hasimportant consequences in terms of predictability, portfolio diversificationand asset allocation. Theory predicts that gains can be achieved throughinternational portfolio diversification if returns in the different markets arenot perfectly correlated. Policies of deregulation and the liberalisation ofcapital markets, coupled with technological advances, suggest that marketshave become more integrated over time. Increasing levels of integrationsuggests that opportunities for portfolio diversification are reduced.Moreover, evidence from crisis events such as the Asian financial crisissuggests that market comovements lead to contagion and consequentlyhigher correlations reducing diversification opportunities. Understandingand careful estimation of the time varying nature of volatilities, covariancesand correlations is paramount to capture changes in risk and identify thenature of comovement between markets.

Evidence of spillover and volatility transmission from one market toanother is well established (see, inter alia, Engle, Ito, & Lin, 1990; Hamao,Masulis, & Ng, 1990). Further evidence on contagion and financial criseshighlights the impact of events such as the Asian crisis and the Russian crisison other markets across the globe (see, inter alia, Kaminsky & Reinhart,1998; Edwards & Susmel, 2001; Bae, Karolyi, & Stulz, 2003). In addition tothese short-run relationships, there is a body of evidence suggesting capitalmarkets share common trends over the long term (Kasa, 1992; Garrett &Spyrou, 1999). This suggests that for investors with long-term investmenthorizons, the benefits of international portfolio diversification could beoverstated. Despite the existence of such long-run relationships it is unlikelythat the benefits of diversification will be eroded since returns may only reactvery slowly to the trend. Indeed the benefits of diversification are likely toremain and hence accurate measurement of volatilities and correlationsbetween markets is of great importance.

Moreover, it is well established that stock return correlations are notconstant through time. Correlations tend to rise with economic or equitymarket integration (Erb, Harvey, & Viskanta, 1994; Longin & Solnik, 1995;Goetzmann, Li, & Rouwenhorst, 2005). They also tend to decline in bullmarkets and increase during bear markets (Longin & Solnik, 2001; Ang &Bekaert, 2002). Longin and Solnik (1995, 2001) show that correlationsbetween markets increase during periods of high market volatility, with the

STUART HYDE ET AL.40

result that correlations would be higher than average exactly in the momentwhen diversification promises to yield gains. Consequently, such changes incorrelations imply that the benefits to portfolio diversification may be rathermodest during bear markets (Baele, 2005).

In this chapter, we investigate the correlation dynamics across the Asia-Pacific region and with Europe and the US using both local currency andUS dollar returns. Using the recently developed asymmetric generaliseddynamic conditional correlation GARCH model (AG-DCC-GARCH) ofCappiello, Engle, and Sheppard (2006) we examine how conditionalcorrelations evolve over time. The model explicitly captures the asymmetricresponse of conditional volatilities and correlations to negative returns. Wefind evidence of asymmetries in conditional volatilities for local currencyreturns yet this asymmetry disappears in most markets for US dollarreturns. Further the lack of volatility feedback is most visible in countrieswith low correlations with the developed markets of the US and Europe.There are significant asymmetries in conditional correlations. Thesecorrelations evolve through time. Evidence of significant increases incorrelation during the Asian crisis is largely limited to crisis countries.Correlations with the US and Europe do not systematically increase duringthis period, rather they peak during the most recent bear market. Our resultsalso demonstrate that correlations are higher towards the end of the sampleperiod than in the early 1990s indicative of greater market integration.

The remainder of the chapter is organised as follows. Next, we brieflyreview the existing literature investigating asset market linkages in Asia-Pacific markets. In Section 3, we discuss the methodology while Section 4presents the results and analysis. Section 5 offers some concluding remarks.

2. LITERATURE REVIEW

Research into asset market linkages and integration in both developedmarkets and emerging markets has developed over recent years establishingthe nature of these relationships for different assets and markets. As aconsequence of the Asian financial crisis, the majority of studies havefocussed on emerging equity markets in the Pacific Basin (see, inter alia,Phylaktis & Ravazzolo, 2002; Manning, 2002), although there is evidence forother asset markets in the region (e.g., Phylaktis (1999) using real interestrates) and for other emerging economies (e.g., Bekaert & Harvey, 1995, 1997).

It is well understood that markets, developed and emerging, can movetogether over the short run. Janakiramanan and Lamba (1998) and Cha and

Correlation Dynamics 41

Cheung (1998) examine linkages between Asia-Pacific equity markets andthe US using vector autoregression (VAR) models, establishing that the UShas a significant influence on these markets in addition to a number of inter-relationships within the Asia-Pacific region. Further, while such researchestablishes spillovers in mean relationships between markets, there has beenmuch research (initiated by Engle et al., 1990; Hamao et al., 1990)examining the presence of spillovers in volatility. More recent studies offinancial crises and contagion provide further evidence that there issignificant transmission across markets (Kaminsky & Reinhart, 1998; Baeet al., 2003). Consequently, it is well documented that mean and volatilityspillovers occur between asset markets suggesting that events in one marketcan be transmitted to others and that the magnitude of such inter-relationships maybe strengthened during crisis periods. Examining thenature of volatility spillovers from Japan and the US to the Pacific-Basinand the impact of financial liberalisation, Ng (2000) finds that both the USand Japan influence volatility in the Pacific-Basin region. While liberal-isation is likely to be a key event, its influence describes only a smallproportion of the total variation suggesting other intra-region influences areimportant. Similarly, Worthington and Higgs (2004) provide evidence of thetransmission of return and volatility among nine developed and emergingAsia-Pacific markets finding significant spillovers across markets usingmultivariate GARCH models. Kim (2005) investigates linkages betweenadvanced Asia-Pacific markets (Australia, Hong Kong, Japan andSingapore) with the US uncovering contemporaneous return and volatilitylinkages which intensified after the Asian crisis.

In addition to mean and volatility spillovers, there is strong evidence tosuggest that markets display common trends over the long term. A numberof studies have investigated the existence of a long-run equilibriumrelationship between Asia-Pacific stock markets and between these marketsand developed markets (see, inter alia, Chan, Gup, & Pan, 1992; Garrett &Spyrou, 1999; Maish & Maish, 1999; Ghosh, Saidi, & Johnson, 1999;Darrat & Zhong, 2002). However, recently studies have investigated thestability of this long-run relationship. Yang, Kolari, and Sutanto (2004) findno evidence of long-run cointegrating relationships between emergingmarkets and the US prior to the Asian financial crisis, but such relationshipsexist during the crisis period. Further, Yang, Kolari, and Min (2003) examineboth long-run relationships and short-run dynamics around the period ofthe Asian crisis demonstrating that linkages between markets strengthenduring the crisis and that markets have remained more integrated post-crisis.Although, Manning (2002) argues that the convergence of South East Asian


equity markets was abruptly halted and somewhat reversed by the crisis. Thevarious alternative findings suggest these relationships vary thorough timeand are naturally impacted by events such as the Asian crisis.

3. METHODOLOGY

In order to investigate the correlation dynamics between the Asia-Pacificequity markets we employ the asymmetric generalised dynamic conditionalcorrelation GARCH model (AG-DCC-GARCH) of Cappiello et al. (2006).This model is the generalisation of the DCC-MVGARCH model of Engle(2002) to capture the conditional asymmetries in correlation. The DCC-MVGARCH is estimated using a two-stage procedure. In the first stage,univariate GARCH models are fit for each of the asset return series andstandardised residuals (residuals standardised by their estimated standarddeviations) are obtained. The second stage uses the standardised residuals toestimate the coefficients governing dynamic correlation.

Let rt denote a n� 1 vector of return innovations (residuals) at time t,which is assumed to be conditionally normal with mean zero and covariancematrix Ht:

rtjOt�1 � Nð0;HtÞ (1)

where Ot�1 represents the information set at time t�1, and the conditionalcovariance matrix Ht can be decomposed as follows:

Ht ¼ DtRtDt (2)

where Dt ¼ diagffiffiffiffiffihit

p� �is the n� n diagonal matrix of time-varying

standard deviations from univariate GARCH models withffiffiffiffiffihit

pon the ith

diagonal, and Rt is the n� n time-varying correlation matrix, containingconditional correlations. The proposed dynamic correlation structure is:

Rt ¼ diagðQtÞ�1QtdiagðQtÞ

�1 (3)

Qt ¼ ðQ� A0QA� B0QB� G0NGÞ þ A0�t�1�0t�1Aþ B0Qt�1Bþ G0Zt�1Z

0t�1G

(4)

where diagðQtÞ ¼ffiffiffiffiffiffiqiit

p� �is a diagonal matrix containing the square root

of the diagonal elements of Qt, A, B and G are n� n parameter matrices,

�it ¼ rit=ffiffiffiffiffihit

pis the standardised residuals, Q ¼ E½�t�0t� ¼ T�1

PTt¼1 �t�0t is the

unconditional correlation matrix of rt, and N ¼ E½ZtZ0t� ¼ T�1

PTt¼1 ZtZ

0t,


with Zit ¼ I ½�ito0� � �i;t, where I ½�ito0� is the indicator function which takes on

value 1 if �ito0 and 0 otherwise, and ‘‘�’’denotes the Hadamard product.This term will capture the conditional asymmetries in correlations. Thegeneralised DCC (G-DCC) model is a special case of AG-DCC when G=0.It is clear from Eq. (4) that Qt will be positive-definite, if ðQ� A0QA�

B0QB� G0NGÞ is positive definite. The AG-DCC model is estimated usingquasi-maximum likelihood (QMLE).

We can extend this model to allow for structural breaks in mean ofcorrelation equation. For example, a researcher might want to test whethera structural break has occurred in the intercept following the Asian financialcrisis 1997. Let dt be the dummy variable 1 if t � tbreakoT , and 0 otherwise.In this case, Eq. (4) can be extended to:

Qt ¼ ðQ1 � A0Q1A� B0Q1B� G0N1GÞð1� dtÞ

þ ðQ2 � A0Q2A� B0Q2B� G0N2GÞdt

þ A0

�t�1�0t�1Aþ B0Qt�1Bþ G0Zt�1Z

0t�1G ð5Þ

where Q1 ¼ E½�t�0t� for totbreak, and Q2 ¼ E½�t�0t� for t4tbreak; N1 and N2

are analogously defined. Since the model in Eq. (5) nests the model inEq. (4), it can be tested for breaks in mean of correlation process usinglikelihood ratio test with k(k�1)/2 degrees of freedom.

We illustrate the asymmetric response of correlation to joint bad newsand joint good news using news impact surfaces introduced by Kroner andNg (1998). The news impact surface for correlation can be estimated asfollows:

f ð�i�jÞ ¼cij þ ðaiaj þ gigjÞ�i�j þ bibjrijtffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðcii þ ða2i þ g2

i Þ�2i þ b2

i Þðcjj þ ða2j þ g2

j Þ�2j þ b2

j Þ

q �i; �jo0

f ð�i�jÞ ¼cij þ aiaj�i�j þ bibjrijtffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðcii þ ða2i þ g2

i Þ�2i þ b2

i Þðcjj þ a2j �

2j þ b2

j Þ

q �io0; �j40

f ð�i�jÞ ¼cij þ aiaj�i�j þ bibjrijtffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðcii þ a2i �

2i þ b2

i Þðcjj þ ða2j þ g2

j Þ�2j þ b2

j Þ

q �i40; �jo0

f ð�i�jÞ ¼cij þ aiaj�i�j þ bibjrijffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ðcii þ a2i �

2i þ b2

i Þðcjj þ a2j �

2j þ b2

j Þ

q �i; �j40

(6)

where �i and �j are standardised residuals for markets i and j; and cii; cij ; cjj

are the corresponding elements of the constant matrix in correlation


equation; ai, bi are the corresponding elements of matrices A and B; and rijt

is the corresponding element of unconditional correlation matrix Q.Covariance news impact surfaces can also be obtained from the productof correlation surfaces with the appropriate component of the news impactcurves from the univariate EGARCH models.

4. DATA AND EMPIRICAL RESULTS

The data employed in this study are weekly observations on stock returns(continuously compounded returns based on Wednesday-to-Wednesdayclosing prices) from 13 Asia-Pacific equity markets, a European (EU) indexand the US over the period 03/01/1991 to 28/12/2006. We choose to workwith weekly data to alleviate problems associated with non-synchronoustrading resulting from the fact that not all the markets are open duringthe same hours of the day. The specific markets are Australia (ASX AllOrdinaries), China (Shanghai Composite), Hong Kong (Hang Seng), India(BSE National), Indonesia (Jakarta Composite), Japan (Nikkei 225), Korea(KOSPI), Malaysia (Kuala Lumpur Composite), New Zealand (NZ Allshare), Pakistan (Karachi SE 100), Singapore (Straits Times), Taiwan (SEweighted), Thailand (Bangkok SET) and US (S&P500). The EU index is avalue weighted index of returns from France, Germany, Italy and the UK. Allstock indices are expressed in both local currency and US dollars, representingunhedged and hedged returns. All data is obtained from Datastream.

Table 1 presents descriptive statistics for the returns series. Panel A reportsthe summary statistics for local currency returns, while panel B gives the figuresfor US dollar denominated returns. The majority of countries have positivemean returns with only Japan and New Zealand experiencing negative returnsin local currency, while Indonesia, Japan and Thailand have negative returns inUS dollars. All median returns are positive (with the exception of US dollarreturns for Japan). Consistent with previous empirical evidence, most of thereturns are negatively skewed.1 All returns exhibit excess kurtosis and Jarque–Bera tests clearly reject the null of a Gaussian distribution in all cases.

Table 2 reports the unconditional correlations between returns in bothlocal currency and US dollar terms. China and Pakistan have muchlower correlations with the other markets, with means of 0.03 and 0.07,respectively in both local currency and US dollar terms. India has a meancorrelation around 0.15 while all other markets are moderately correlatedwith mean correlations in the range 0.22–0.35.2 As would be expected the


correlations with the US and the EU markets relative to Australia,Hong Kong, Japan, New Zealand and Thailand are quite high. Whilethe correlations for China, India, Indonesia, Malaysia and Pakistanare considerably smaller. Our results also take account of foreign exchangemovements and the impact that this may have on the correlations. The table

Table 1. Summary Statistics.

Series Mean Median Standard Deviation Skewness Kurtosis Jarque–Bera

Panel A: Local currency returns

Australia 0.0783 0.0809 0.7297 �0.2761 4.3363 43.84�

China 0.1550 0.1148 2.5187 2.1910 28.6010 813.36�

Hong Kong 0.0969 0.1736 1.4394 �0.5130 4.7359 56.716�

India 0.1343 0.2783 1.7484 0.0412 6.2278 190.38�

Indonesia 0.0778 0.1643 1.5015 �0.0892 5.3233 114.66�

Japan �0.0170 0.0189 1.2763 �0.0040 4.1727 38.348�

Korea 0.0373 0.0212 1.7721 �0.1142 4.8708 80.917�

Malaysia 0.0400 0.0562 1.5001 0.4260 12.3453 743.54�

New Zealand �0.0670 0.0809 0.9003 �0.0979 6.7182 232.45�

Pakistan 0.1472 0.2019 1.728 �0.3452 5.0762 83.644�

Singapore 0.0603 0.0407 1.2476 0.0069 5.6091 138.47�

Taiwan 0.0308 0.0686 1.6428 �0.1990 4.9570 83.962�

Thailand 0.0070 0.0615 1.7356 0.1723 4.3554 47.030�

Europe 0.0653 0.1244 0.8981 �0.4187 6.1587 152.79�

US 0.0760 0.1292 0.9031 �0.1164 5.1680 102.16�

Panel B: US dollar returns

Australia 0.0796 0.1740 0.9840 �0.3378 3.4653 16.213�

China 0.1336 0.1232 2.6240 1.5121 25.0470 1254.6�

Hong Kong 0.0971 0.1860 1.4474 �0.5121 4.6811 54.485�

India 0.0879 0.1945 1.8118 �0.3037 5.5440 119.20�

Indonesia �0.0028 0.0000 2.4154 �0.8054 13.7600 528.74�

Japan �0.0099 �0.0040 1.4114 0.0929 4.1730 37.996�

Korea 0.0237 0.0000 2.1223 �0.7821 10.0035 378.44�

Malaysia 0.0262 0.0571 1.8264 �0.9958 21.2030 1363.8�

New Zealand 0.0766 0.1825 1.1239 �0.3735 5.6568 120.84�

Pakistan 0.0935 0.1780 1.7420 �0.3583 5.0629 81.921�

Singapore 0.0668 0.0642 1.3284 �0.1541 5.9226 160.09�

Taiwan 0.0210 0.1005 1.7432 �0.2567 4.9055 77.809�

Thailand �0.0111 0.0163 1.9437 0.0655 5.1981 105.66�

Europe 0.0707 0.0943 0.9008 �0.4395 5.3717 95.049�

Note: This table reports summary statistics for weekly (Wednesday-to-Wednesday) stock

returns. The sample period is 02/01/1991–27/12/2006.�Indicates significance at 1%.


Table 2. Correlation Matrix.

Australia China Hong

Kong

India Indonesia Japan Korea Malaysia New

Zealand

Pakistan Singapore Taiwan Thailand Europe US

Australia 1 0.1017 0.4663 0.1751 0.2356 0.3726 0.2726 0.2754 0.5201 0.1151 0.4006 0.2415 0.2710 0.4148 0.5064

China 0.0804 1 0.0537 0.0581 0.0532 0.0203 0.0108 0.0126 0.0506 0.0024 0.0437 0.0035 0.0334 0.0111 �0.0032

Hong Kong 0.4680 0.0384 1 0.1796 0.3102 0.3287 0.4105 0.4029 0.3426 0.0561 0.6094 0.3152 0.3955 0.4152 0.4228

India 0.2120 0.0399 0.1663 1 0.1770 0.1571 0.2135 0.1272 0.1045 0.1559 0.2068 0.1809 0.1805 0.1159 0.1104

Indonesia 0.2466 0.0370 0.3214 0.1273 1 0.1916 0.2434 0.4003 0.1603 0.1239 0.4020 0.2090 0.4382 0.1609 0.1438

Japan 0.3249 0.0184 0.3148 0.1539 0.1653 1 0.3227 0.2141 0.2528 0.0193 0.3592 0.2821 0.1950 0.2842 0.3575

Korea 0.2948 0.0096 0.3940 0.2040 0.2299 0.3432 1 0.2377 0.2154 0.0497 0.3800 0.2820 0.3720 0.2499 0.2800

Malaysia 0.2289 �0.0043 0.3615 0.1253 0.4269 0.1996 0.2074 1 0.2525 0.1244 0.5578 0.2658 0.4379 0.1526 0.2266

New Zealand 0.5626 0.0594 0.3366 0.1212 0.1783 0.2769 0.2497 0.2110 1 0.0082 0.322 0.2056 0.2170 0.3518 0.3726

Pakistan 0.1183 �0.0092 0.0751 0.1578 0.0879 �0.0028 0.0020 0.0991 0.0454 1 0.1215 0.0472 0.1197 �0.0304 0.0470

Singapore 0.4155 0.0434 0.6046 0.2217 0.4215 0.3542 0.3504 0.5173 0.3608 0.1221 1 0.3253 0.5066 0.3389 0.3660

Taiwan 0.2698 �0.0292 0.3397 0.1938 0.2029 0.2656 0.2831 0.2799 0.2351 0.0657 0.3654 1 0.2424 0.2275 0.2346

Thailand 0.3253 0.0283 0.4159 0.2018 0.4506 0.2035 0.3486 0.4450 0.2524 0.1342 0.5561 0.2753 1 0.2416 0.2316

Europe 0.3917 0.0459 0.3841 0.1368 0.1012 0.2683 0.2461 0.1315 0.4002 0.0175 0.3365 0.2118 0.2002 1 0.6619

US 0.4229 0.0019 0.4204 0.1094 0.1120 0.3078 0.2697 0.1666 0.3281 0.0583 0.3437 0.2324 0.2143 0.6036 1

Note: This table reports unconditional correlation coefficients for weekly (Wednesday-to-Wednesday) stock returns. The sample period is 02/

01/1991–27/12/2006. Coefficients below the diagonal are US dollar returns, above the diagonal they are local currency returns.

Co

rrelatio

nD

yn

am

ics47

highlights that estimated correlations are different for local currency and USdollar returns for each market, accounting for currency variations has asignificant although not systematic affect on correlation. This appears toparticularly the case for countries with low correlations with the US andEurope, namely China, India, Indonesia and Pakistan. For example, thecorrelation between Malaysia and the US moves from 0.23 in local currencyto 0.17 in US dollar terms, for a period where the ringgit was pegged to theUS dollar for a number of years during the current sample.

The first stage of the estimation process is to fit univariate GARCHspecifications for each of the 15 return series. To account for possibleasymmetry in conditional volatility we estimate EGARCH models in eachcase. We find evidence of asymmetry in most of the stock markets underinvestigation. It appears that there is very little evidence of asymmetry for alarge number of the emerging markets. In particular markets that have lowcorrelations with the US (and Europe) provide very little evidence ofvolatility feedback, namely China, India and Indonesia. This is the case forboth local and US dollar returns.3,4 Parameter estimates from the univariateEGARCH models are reported in Table 3.

Table 3. Univariate Asymmetric GARCH Models.

Series Panel A: Local Currency Returns Panel B: US Dollar Returns

o a b g o a b g

Australia �0.2074 0.1928 0.9230 �0.1185 �0.0360� 0.0214� 0.7304 �0.1636

China �0.2598 0.3901 0.9747 0.0380� �0.2693 0.4079 0.9747 0.0303�

Hong Kong �0.1350 0.1802 0.9858 �0.0023� �0.1389 0.1849 0.9855 �0.0024�

India �0.2058 0.3936 0.9000 0.0003� �0.0938� 0.3404 0.8510 �0.0411�

Indonesia �0.0866 0.1289 0.9834 �0.0103� �0.1444 0.2150 0.9848 �0.0232�

Japan �0.0909 0.1395 0.9505 �0.0933 �0.1046 0.1602 0.9627 �0.0872

Korea �0.0686 0.0942 0.9921 �0.0467 �0.1093 0.1648 0.9813 �0.0669

Malaysia �0.1941 0.2625 0.9797 �0.0353 �0.1888 0.2599 0.9913 �0.0080�

New Zealand �0.1226 0.1550 0.9833 0.0424 �0.1525 0.2082 0.9424 0.0011�

Pakistan �0.1828 0.4597 0.8267 0.0085� �0.1619 0.4449 0.8223 0.0100�

Singapore �0.1449 0.1940 0.9723 �0.0465 �0.1414 0.1937 0.9745 �0.0405

Taiwan �0.1698 0.2697 0.9485 �0.0306� �0.1521 0.2704 0.9362 �0.0378�

Thailand �0.0766 0.1098 0.9886 �0.0062� �0.0811 0.1217 0.9863 �0.0131�

Europe �0.1672 0.1816 0.9425 �0.0982 �0.2038 0.2348 0.9439 �0.0757

US �0.1441 0.1612 0.9582 �0.1129 �0.1441 0.1612 0.9582 �0.1129

Note: This table reports parameter estimates for the univariate EGARCH models for weekly

(Wednesday–to-Wednesday) stock returns. The sample period is 02/01/1991–27/12/2006.�Indicates parameters insignificant at 5%.


Given the large literature on the Asian crisis and contagion, we considerthe possibility that the crisis period represented a structural break due to thelarge number of markets affected in the Asia-Pacific region and beyond. Toaccount for this, we test for the existence of a structural break in theintercept. We also consider two alternate crisis dates: 02/07/1997 when theThai Baht devalued and the crisis commenced, and 22/10/1997 when therewere devaluations of the Taiwanese dollar and Korean won and a large fallin the Hong Kong equity market, representing the widening of the crisis.Table 4 reports the log-likelihood values from a series of models. Thelikelihood ratio tests reject the null hypothesis of no structural break inmean, indicating that all the models allowing for a mean break significantlyoutperform the non-break models. Similarly, all the asymmetric generalisedDCC models outperform the non-asymmetric models. These results aresupported by the BIC results. Moreover, in both local currency and USdollar models, adopting a break at 22/10/1997 (the widening of the crisis)

Table 4. Log-likelihood Values.

Model Log-likelihood

Value

Number of

Parameters in

the Correlation

Evolution

BIC

Panel A: Local currency returns

DCC �16191.5 105+2 39.644

DCC with mean break at 02/07/1997 �15811.1 210+2 39.578


AG-DCC �15792.6 105+102 39.494

AG-DCC with mean break at 02/07/1997 �15427.3 210+102 39.465


Panel B: US dollar returns

DCC �16130.4 105+2 39.497



AG-DCC �16001.3 105+102 39.994



Note: This table reports log-likelihood values for six estimated DCC GARCH models for both

local currency returns and US dollar returns. DCC is the dynamic conditional correlation, AG-

DCC is asymmetric generalised dynamic conditional correlation. We test for a break due to the

Asian crisis, at 02/07/1997 when the Thai Baht devalued (commencement of the crisis) and

22/10/1997 when the Taiwanese dollar and Korean Won devalued and the Hong Kong stock

market fell (crisis spreads throughout the region).


rather than 02/07/1997 (when the crisis commenced) reduces the BIC,implying that 22/10/1997 is a more preferable break date for the crisis. We,therefore, report our results for the AG-DCC-GARCH model with a meanbreak at 22/10/1997.5

The parameter estimates of the AG-DCC-GARCH model are reported inTable 5. Most parameter coefficients are statistically significant atconventional levels. In all cases except China, India and Indonesia, we findevidence of asymmetries in conditional correlations. The conditionalcorrelations and conditional covariances for local currency returnsestimated from the AG-DCC-GARCH model with a mean break areplotted in Fig. 1 for the correlations and covariances of the 14 markets withthe US and Fig. 2 for the correlations and covariances of the 13 marketswith the EU. While correlations indicate the relationship between tworeturns, the covariance captures the amount of comovement between them.Thus it is possible to determine whether changes in comovement are due to achange in the correlations between markets or simply due to volatility. Fig. 3provides plots of the conditional correlations and conditional covariancesbetween the five markets central to the Asian crisis; Indonesia, Korea,

Table 5. AG-DCC GARCH Models.

Series Panel A: Local Currency Returns Panel B: US Dollar Returns

a2 b2 g2 a2 b2 g2

Australia 0.0030 0.9778 0.0013 0.0022 0.9596 0.0015

China 0.0009� 0.8126 0.0002� 0.0008� 0.6165 0.0001�

Hong Kong 0.0063 0.9489 0.0035 0.0058 0.9242 0.0018

India 0.0011� 0.9358 0.0003� 0.0004� 0.8993 0.0001�

Indonesia 0.0058 0.9416 0.0006� 0.0079 0.9273 0.0004�

Japan 0.0031 0.9559 0.0019 0.0030 0.9486 0.0022

Korea 0.0030 0.9526 0.0028 0.0019 0.9601 0.0028

Malaysia 0.0049 0.9069 0.0015 0.0053 0.9006 0.0004�

New Zealand 0.0003� 0.9176 0.0013 0.0001� 0.8655 0.0002�

Pakistan 0.0002� 0.7437 0.0001� 0.0001� 0.6961 0.0001�

Singapore 0.0077 0.9719 0.0020 0.0080 0.9628 0.0019

Taiwan 0.0006 0.9152 0.0011 0.0005� 0.9167 0.0014

Thailand 0.0073 0.9010 0.0016 0.0066 0.8844 0.0022

Europe 0.0032 0.9658 0.0027 0.0028 0.9598 0.0017

US 0.0020 0.9749 0.0051 0.0020 0.9745 0.0052

Note: This table reports parameter estimates for the AG-DCC GARCH model for weekly

(Wednesday-to-Wednesday) stock returns. The sample period is 02/01/1991–27/12/2006.�Indicates parameters insignificant at 5%.


Fig. 1. Conditional Correlations and Conditional Covariances with US. Conditional Correlations and Covariances for

Local Currency Returns. Shaded Area Corresponds to Asian Crisis Period 02/07/1997–30/12/1998. Line Corresponds to

Break at 22/10/1997.

Co

rrelatio

nD

yn

am

ics51

Fig. 2. Conditional Correlations and Conditional Covariances with EU. Conditional Correlations and Covariances for

Local Currency Returns. Shaded Area Corresponds to Asian Crisis Period 02/07/1997–30/12/1998. Line Corresponds to

Break at 22/10/1997.

STUART

HYDE

ET

AL.

52

Fig. 3. Conditional Correlations and Conditional Covariances between Asian Crisis Countries (Indonesia, Korea, Malaysia,

Taiwan and Thailand). Conditional Correlations and Covariances for Local Currency Returns. Shaded Area Corresponds to

Asian Crisis Period 02/07/1997–30/12/1998. Line Corresponds to Break at 22/10/1997.

Co

rrelatio

nD

yn

am

ics53

Malaysia, Taiwan and Thailand.6 On each plot the break date of 22/10/97 ismarked with a vertical line, while the shaded area corresponds to the Asiancrisis period 02/07/97–30/12/98. There is clear evidence of considerablevariation in correlations and covariances in all cases. Typically the dynamicpattern of correlations is also witnessed in the corresponding covariances,although variation in volatility leads to periods of significantly differentbehaviour. There is evidence of further global market integration towardsthe end of the sample period, since correlations rise while covariances tendto fall as a consequence of decreasing volatility.

Correlations of Asia-Pacific countries with the US and the EU provide noclear pattern across the Asian crisis period. Indeed, consistent with Longinand Solnik (2001) and Ang and Bekaert (2002), analysing the time varyingconditional correlations highlights that correlations with the US and the EUtend to increase and reach a maximum during the recent bear marketbetween 2000 and 2003. Further, correlations tend to be higher post 2001than in the early part of the sample, despite reduced correlations due to thebull market post 2003, suggesting greater equity market integration. This isparticularly the case for newer emerging markets in the region such as Chinaand India, although developed markets such as Japan also witnesssignificantly higher correlations towards the end of the sample.

In contrast to correlations with the US and the EU, Fig. 3 clearly shows alarge increase in correlation among the five Asian crisis countries at theonset of the crisis. In most cases, we witness correlations falling after the endof the crisis, yet correlations levels seem to remain higher than pre-crisislevels. The majority of correlations with Malaysia, Taiwan and Thailand inboth local currency and US dollars, and with Indonesia and Korea in USdollars peak during the crisis period.7 The results show that the Asian crisiscaused a significant increase in intra-regional correlations. However, nosuch impact was witnessed with respect to correlations with the US andEurope.

To investigate further the impact of the observed asymmetries, weexamine the news impact surfaces of Kroner and Ng (1998) for each ofKorea and Thailand with the US (Fig. 4a), with Europe and betweenthemselves and Malaysia (Fig. 4b) for both local currency and US dollarreturns.8 The asymmetry in correlation to joint bad and joint good news isidentifiable in virtually all cases. The correlation news impact surface revealsa much larger response in the negative–negative (�/�) quadrant than in thepositive–positive (+/+) quadrant. Hence the impact observed whennegative shocks (bad news) occur simultaneously in both markets is higherthan for joint positive shocks (good news) for both unhedged local currency


returns and hedged US dollar returns. The effect is strongest for correlationswith the US and Europe, while its presence is virtually undetectable forUS dollar return correlations with Malaysia. This corresponds with therelatively high levels of asymmetry reported in Table 5 for the US and thelack of asymmetry for Malaysia.

The effect of asymmetry becomes even more striking when we examine thecovariance news impact surfaces. Fig. 5a reports the surfaces for each of Koreaand Thailand with the US and Fig. 5b between themselves and Malaysia. Thecombination of the correlation with the two conditional volatilities produces ahuge increase in the �/� quadrant. The increase witnessed in response to jointgood news is typically much lower. There is little evidence of asymmetry in the+/� and �/+ quadrants for covariances with the US and Europe, howeverthese asymmetries are visible in covariances between Asia-Pacific markets.

Fig. 4. (a) Correlation News Impact Surfaces. Conditional Correlation News

Impact Surfaces for Korea and Thailand with the US.


Fig. 4. (b) Correlation News Impact Surfaces. Conditional Correlation News

Impact Surfaces between Korea, Malaysia and Thailand.


5. CONCLUSION

In this chapter, we investigate correlation dynamics between 13 Asia-Pacificstock markets, the EU and the US. Correlations are key to internationalportfolio diversification and asset allocation decisions. While most ofprevious literature on volatility transmission only concentrates on covariancebetween markets, we provide a more comprehensive view showing bothdynamic covariance and dynamic correlation between asset prices acrossmarkets. Using the recently developed asymmetric generalised dynamic

Fig. 5. (a) Covariance News Impact Surfaces. Conditional Covariance News Impact

Surfaces for Korea and Thailand with the US.


Fig. 5. (b) Covariance News Impact Surfaces. Conditional Covariance News Impact

Surfaces between Korea, Malaysia and Thailand.


conditional correlation GARCH model (AG-DCC-GARCH) of Cappielloet al. (2006), we examine how conditional correlations and covariances forboth local currency and hedged US dollar returns evolve over time. Weuncover evidence of wide variation in correlations through time, withconditional correlations deviating significantly from the levels of uncondi-tional correlations. Importantly, we also establish significant asymmetry incorrelations between many markets. Reinforcing the established view thatcorrelations increase in response to bad news, crisis events, bear markets.Although importantly there seems to be little asymmetry in countries that arenot highly correlated with developed markets, suggesting a link betweenlevels of market integration and volatility feedback.

Incorporating a structural break due to the Asian crisis at 22/10/97improves the fit of the estimated model. However, significantly, increases inconditional correlations during the Asian crisis seem to be mainly limited tocrisis countries in the region, correlations involving other markets are notsystematically effected. Although correlations with the US and Europe arerelatively immune to the crisis, they do rise during the bear market in theearly 2000s. In addition we document a general increase in correlations overthe entire sample period indicative of greater global market integration.

Further we demonstrate the asymmetric response of both conditionalcorrelations and covariances to join bad and good news highlighting thatthe impact of crises and bear markets on correlation are furthercompounded by volatility. These findings throw further light on correlationand covariance dynamics between equity markets. These dynamics highlightsubstantial time variation in international portfolio diversification oppor-tunities across the Asia-Pacific, EU and US markets.

NOTES

1. China, Malaysia and Singapore have positively skewed local currency returns,while China, Japan and Thailand have positively skewed US dollar returns.2. The median correlation is (excluding China, India and Pakistan) 0.23 (0.28).3. Indeed, in some cases (China, India, New Zealand and Pakistan) we find the

‘‘good news’’ chasing effect documented in emerging markets, however, the positiveasymmetry coefficient is typically always statistically insignificant.4. Evidence of asymmetry is much weaker for the hedged (US dollar) returns.5. Aside from poorer in-sample performance, qualitatively the results do not

change if a break date of 02/07/97 is adopted.6. We select these as the correlations and covariances to report and discuss, plots

of all 105 local currency and all 105 US dollar correlations and covariances areavailable from the authors on request.


7. The results highlight that correlations with New Zealand peak in virtually everycase during the Asian crisis (22/28 cases).8. Correlation news impact surfaces with respect to Europe are not reported and

are available from authors on request. The results are qualitatively similar to the US.

REFERENCES

Ang, A., & Bekaert, G. (2002). International asset allocation with regime shifts. Review of

Financial Studies, 15, 1137–1187.

Bae, K.-H., Karolyi, G. A., & Stulz, R. M. (2003). A new approach to measuring financial

contagion. Review of Financial Studies, 16, 717–763.

Baele, L. (2005). Volatility spillover effects in European equity markets. Journal of Financial and

Quantitative Analysis, 40, 373–401.

Bekaert, G., & Harvey, C. R. (1995). Time-varying world market integration. Journal of

Finance, 50, 403–444.

Bekaert, G., & Harvey, C. R. (1997). Emerging equity market volatility. Journal of Financial

Economics, 43, 29–77.

Cappiello, L., Engle, R. F., & Sheppard, K. (2006). Asymmetric dynamics in the correlations of

global equity and bond returns. Journal of Financial Econometrics, 4, 537–572.

Cha, B., & Cheung, Y.-L. (1998). The impact of the US and the Japanese equity markets on the

emerging Asia-Pacific equity markets. Asia-Pacific Financial Markets, 5, 191–209.

Chan, K. C., Gup, B. E., & Pan, M. (1992). An empirical analysis of stock prices in major Asian

markets and the United States. Financial Review, 27, 289–307.

Darrat, A. F., & Zhong, M. (2002). Permanent and transitory driving forces in Asian-Pacific

stock markets. Financial Review, 31, 343–363.

Edwards, S., & Susmel, R. (2001). Volatility dependence and contagion in emerging equity

markets. Journal of Development Economics, 66, 505–532.

Engle, R. F. (2002). Dynamic conditional correlation: A simple class of multivariate generalized

autoregressive-conditional heteroskedasticity models. Journal of Business and Economic

Statistics, 20, 339–350.

Engle, R. F., Ito, T., & Lin, W.-L. (1990). Meteor showers or heat waves? Heteroskedastic

intra-daily volatility in the foreign exchange market. Econometrica, 58, 525–542.

Erb, C. B., Harvey, C. R., & Viskanta, T. E. (1994). Forecasting international equity

correlations. Financial Analysts Journal, 50, 32–45.

Garrett, I., & Spyrou, S. (1999). Common stochastic trends in emerging equity markets.

Manchester School, 67, 649–660.

Ghosh, A., Saidi, R., & Johnson, K. H. (1999). Who moves the Asia-Pacific stock markets – US

or Japan? Financial Review, 34, 159–169.

Goetzmann, W. N., Li, L., & Rouwenhorst, K. G. (2005). Long-term global market

correlations. Journal of Business, 78, 1–38.

Hamao, Y., Masulis, R., & Ng, V. (1990). Correlation in price changes and volatility across

international stock markets. Review of Financial Studies, 3, 281–307.

Janakiramanan, S., & Lamba, A. (1998). An empirical investigation of linkages between Pacific-

Basin stock markets. Journal of International Financial Markets, Institutions and Money,

8, 155–173.


Kaminsky, G. L., & Reinhart, C. M. (1998). Financial crises in Asia and Latin America.

American Economic Review, 88, 444–448.

Kasa, K. (1992). Common stochastic trends in international stock markets. Journal of Monetary


Kim, S. (2005). Informational leadership in the advanced Asia-Pacific stock markets: Return,

volatility and volume information spillovers from the US and Japan. Journal of Japanese

and International Economics, 19, 338–365.

Kroner, K. F., & Ng, V. K. (1998). Modelling asymmetric comovements of asset returns.

Review of Financial Studies, 11, 817–844.

Longin, F., & Solnik, B. (1995). Is the correlation in international equity returns constant.

Journal of International Money and Finance, 14, 3–26.

Longin, F., & Solnik, B. (2001). Extreme correlation and international equity markets. Journal

of Finance, 56, 649–676.

Maish, A. M. M., & Maish, R. (1999). Are Asian stock market fluctuations due mainly to intra-

regional contagion effects? Evidence based on Asian emerging markets. Pacific-Basin

Finance Journal, 7, 251–282.

Manning, N. (2002). Common trends and convergence? South East Asian equity markets

1988–1999. Journal of International Money and Finance, 21, 183–202.

Ng, A. (2000). Volatility spillover effects from Japan and the US to the Pacific-Basin. Journal of

International Money and Finance, 19, 207–233.

Phylaktis, K. (1999). Capital market integration in the Pacific Basin region: An impulse

response analysis. Journal of International Money and Finance, 18, 267–287.

Phylaktis, K., & Ravazzolo, F. (2002). Measuring financial and economic integration with

equity prices in emerging markets. Journal of International Money and Finance, 21,

879–903.

Worthington, A., & Higgs, H. (2004). Transmission of equity returns and volatility in Asian

developed and emerging markets: A Multivariate GARCH analysis. International

Journal of Finance and Economics, 9, 71–80.

Yang, J., Kolari, J. W., & Min, I. (2003). Stock market integration and financial crises: The case

of Asia. Applied Financial Economics, 13, 477–486.

Yang, J., Kolari, J. W., & Sutanto, P. W. (2004). On the stability of long-run relationships

between emerging and US stock markets. Journal of Multinational Financial Manage-

ment, 14, 233–248.


CHAPTER 4

CONDITIONAL

AUTOCORRELATION AND

STOCK MARKET INTEGRATION

IN THE ASIA-PACIFIC

Suk-Joong Kim and Michael D. McKenzie

ABSTRACT

This chapter considers the relationship between stock market autocorre-

lation and (i) the presence of international investors which is proxied by

the level of capital market integration and (ii) stock market volatility.

Drawing from a sample of nine Asia-Pacific stock indices, significant

evidence of a relationship between the presence of international investors

and the level of stock market autocorrelation is found. This evidence is

consistent with the view that international investors are positive feedback

traders. Robustness testing of this model suggests that the trading

strategy of international investors changed as a result of the Asian

currency crisis. The evidence for the role of volatility in explaining

autocorrelation is, however, is generally weak and varies across the

sample countries.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00004-0

63

dx.doi.org/10.1016/S1569-3767(07)00004-0.3d

1. INTRODUCTION

The capital flows of international investors have been subject of a great dealof interest in the academic literature. The primary issues revolve aroundhow international investors behave and their impact on the capital marketsin which they invest. A brief survey of the literature reveals an interestingdivergence of opinion. On the one hand, international investors areperceived as a respectable group who provide capital to countries, whichhave a range of investment opportunities but only limited means. Theypossess a superior set of information when compared to local investors andtheir portfolio allocation decisions are based on a sophisticated set ofinvestment strategies which focus on the fundamentals (see Froot &Ramadorai, 2001; Seasholes, 2004). On the other hand, a competing viewcasts international investors as the scourge of the global economy. Underthis view, international investors are thought to pursue positive feedbacktrading strategies which exacerbate trends causing overshooting, excessvolatility and increased market vulnerability (see Dornbusch & Park, 1995;Choe, Kho & Stulz, 1999; Kim & Wei, 2002; Grinblatt & Keloharju 2000;Froot, O’Connell, & Seasholes, 2001). In the extreme, internationalinvestors have been blamed for a number of financial market disasters,such as the 1997 Asian currency crisis (Radelet & Sachs, 1998).

In general terms, investors may pursue either ‘information’ or ‘feedback’trading strategies. The trading behavior of this latter group has been linkedto autocorrelation in asset prices (see Sentana & Wadhwani, 1992).A feedback trader bases the decision to buy, sell, or hold on past pricemovements. Two types of feedback trader can be characterized: ‘positive’(‘negative’) feedback traders systematically follow the strategy of buying(selling) after price rises and selling (buying) after price falls. Thus, positivefeedback traders reinforce price movements such that prices will continuallyovershoot the levels suggested by current publicly available information. Asthe market corrects for this over-reaction in the following trading period,prices tend to move in the opposite direction and so negative autocorrela-tion is induced. The converse situation is true for negative feedback traderswho are thought to induce positive autocorrelation.

Recognizing the existence of both types of traders, it can be argued thatthe sign and strength of any observed return autocorrelation may well reflectthe relative market dominance of one group of feedback traders overanother. Positive (negative) stock return autocorrelation would tend tosuggest negative (positive) feedback traders are the dominant trading groupfor that particular asset. This autocorrelation may vary over time as

SUK-JOONG KIM AND MICHAEL D. McKENZIE64

different trading strategies come into, and go out of, favor with investors(see Farmer, 2000). Information traders are benign in this context however,as they do not follow market trends and so, do not contribute to marketmomentum.

Safvenblad (2000) shows that the return autocorrelation of individualstocks is an important determinant of stock index autocorrelation. Thus, themarket will exhibit a given level of autocorrelation that reflects the amountand type of feedback trading by investors in individual stocks. If we beginby assuming the market is closed to foreigners, then the level ofautocorrelation observed in the market will reflect the trading strategiesemployed by local investors. If foreign traders are granted access to themarket, then types of trading strategy employed by this group may impacton the observed level of autocorrelation. If international investors pursuefeedback trading strategies, ceteris paribus, the collective presence offeedback traders in the market as a whole will have increased. This hasimplications for the level of autocorrelation exhibited by the market. Forexample, if international investors are positive feedback traders, then theirtrading activity will serve to further exacerbate the momentum of markettrends causing an even greater reversal the following day. In this case, lowerand possibly even negative autocorrelation will result. On the other hand, ifinternational investors are negative feedback traders, then their presence inthe local market will add to the negative feedback trading of locals. Greaterprofit taking in a rising market means an increased likelihood of a pricecontinuation the following day and so heightened autocorrelation will beobserved. Where international investors pursue information-based strate-gies, the level of feedback trading will not change. In this case, the presenceof international investors in the local market should have no impact onautocorrelation.1

In this chapter, we have a dual aim of investigating the impact ofinternational investors on local stock market dynamics and the relationshipbetween market volatility and conditional autocorrelation in a number ofemerging Asia-Pacific stock markets. Recovering from the devastation ofthe Asian financial crisis, this region has re-emerged to take a leading role indriving growth in the world economy and stock markets. China inparticular, has played a pivotal role in this process. Thus, investigation ofthese fast growing stock markets before and after the Asian financial crisis isan important addition to knowledge.

Our research findings will be of interest to investors, economists, marketregulators, and government policy makers alike. For example, Stiglitz(2000) called for regulation of international capital flows arguing that

Conditional Autocorrelation and Stock Market Integration 65

developing countries are extremely vulnerable to fluctuations in interna-tional capital flows. We argue that the presence of international investors inthe emerging Asia-Pacific markets will influence the observed level ofautocorrelation if they pursue feedback trading strategies. The nature of therelationship will reflect the type of feedback trading strategy employed. Animportant issue with this type of research relates to how to measure thepresence of foreigners in the local market. The previous literature hasanalyzed datasets, which directly capture information on the trading activityof foreigners and residents. Such datasets are typically highly specializedand not readily available for a wide selection of countries. In this chapter,we adopt a different tact by using a measure of capital market integration toproxy for the presence of foreign investors. An important part of the processof Asia-pacific capital market integration involves the removal of capitalmarket restrictions on the participation of foreigners in domestic stockmarkets.2 As such, increased levels of trading by foreigners will accompanyhigher levels of integration. According to our hypothesis, higher levels ofintegration should significantly impact on the observed level of autocorrela-tion and the direction of this relationship will be a function of the type oftrading strategy employed by international investors.

As for the role of volatility in determining autocorrelations, we argue thatthe presence or lack of feedback traders would have an implication. Asautocorrelation is argued to reflect the activity of feedback traders (seeSentana & Wadhwani, 1992; Black, 1988, 1989) changes in volatilitytherefore have implications for the level of autocorrelation. Where negativereturn autocorrelation exists, volatility increases should serve to heightenthe observed level of autocorrelation. On the other hand, where positiveautocorrelation is evident, a rise in volatility should lessen the level of returnautocorrelation. In support of this theory, a negative relationship betweenvolatility and autocorrelation has been found in the literature (see inter aliaSentana & Wadhwani, 1992; Koutmos, 1997; McKenzie & Faff, 2003) forindividual stocks. In testing the nature of the relationship between volatilityand autocorrelation, the previous literature has failed to recognize thatheightened volatility may result from either an increase or a decrease inprices. In this chapter, we argue this to be an important distinction andinvestigate the disaggregated influence of heightened volatility with eitherpositive or negative returns on conditional autocorrelations.

To test our hypotheses, we specify a conditional measure of autocorrela-tion that is generated using a multivariate generalized ARCH (M-GARCH)model. The autocorrelation term of the covariance equation in this modelhas been augmented to include a measure of market integration and


measures of market volatility as well as other determinants found to be ofimportance in the literature such as daily periodicity, large returns, etc. Thisissue of integration is an important one and a substantial volume ofliterature has been devoted to considering the question of whether capitalmarkets are integrated, in particular for emerging economies (for a surveysee Bekaert & Harvey, 2002, 2003). The evidence suggests that capitalmarkets are imperfectly integrated and that the level of integration changesover time. As such, we specify a time-varying integration parameter adoptedfrom Bekaert and Harvey (1995) in our analyzes. This model is to be appliedto a wide range of emerging market data. Harvey (1995) reports thatemerging markets typically exhibit higher levels of autocorrelationcompared to developed markets. To provide a control sample for theanalysis a number of developed markets are also tested in this framework.The value and volume of transactions in these markets are substantial andthe trading strategies employed by incumbent investors span the fullspectrum of information and momentum-based trading strategies. Thepresence of foreigners is not expected to alter the playing field in anysignificant way and as such, no relationship between autocorrelation and thepresence of foreigners is hypothesized for these developed markets.

The results of our analysis find important evidence of a significantrelationship between the presence of international investors and the level ofAsia-Pacific stock market autocorrelation. Specifically, lower levels ofconditional autocorrelation in returns are associated with the increasedpresence of international investors. This result is consistent with the view theinternational investors are positive feedback traders and is consistent withprevious research. The nature of the relationship however, may change overtime. For example, analysis of our model for post-1997 Asian currency crisisdata suggests that the extent to which positive feedback trading is a featureof the market has diminished and foreign investors either withdrew fromthe market or modified their trading strategies to suit the new regime. As forthe impact of market volatility on the autocorrelations, we find thatvolatility is not as significant a determinant of autocorrelation as haspreviously been found in the individual stock setting. The limited evidenceof a relationship in our sample is more mixed compared to the past literaturewhere higher levels of volatility are typically associated with lower levels ofautocorrelation.

The remainder of the chapter is organized as follows. In the next section,we outline our empirical approach as well as the Markov regime switchingmodels used to generate proxies for market volatility and integration.Section 3 presents the data used in the analysis and discusses the results.


Robustness testing of our results to the 1997 Asian currency crisis is alsoundertaken. Finally, Section 4 presents some concluding comments.

2. BIVARIATE GARCH MODEL ESTIMATES OF

CONDITIONAL AUTOCORRELATION

Empirical estimates reveal that stock return autocorrelation is sampledependent andmay exhibit sign reversals (see Chan, 1993; Knif, Pynnonen, &Luoma, 1996) which suggests that it is appropriate to model auto-correlation as a time-varying process. To this end, Sentana and Wadhwani(1992), Koutmos (1997), and Booth and Koutmos (1998) generatedconditional autocorrelation estimates whose temporal variation was drivensolely by changes to the variance. One weakness of this model is theassumption of a constant covariance, which potentially suppresses animportant source of variation in autocorrelation. In this chapter, conditionalautocorrelation estimates are generated using an M-GARCH model inwhich both the variance and covariance equations are time varying.Estimates of conditional autocorrelation may be generated where thisM-GARCH model is fitted to that returns series (R1,t) as well as its laggedvalues (R2,t).

Specifically, the mean equation for each series is specified with a constantas well as day-of-the-week dummies, i.e.,

R1;t ¼ a1;c þ a1;Lag R1;t�1 þ a1;WRTN WRTNt�1

þXThu

i¼Mon

a1;i DayDumi;t þ e1;t

R2;t ¼ a2;c þ a2;Lag R2;t�1 þ a2;WRTN WRTNt�2

þXThu

i¼Mon

a2;i DayDumi;t�1 þ e2;t

(1)

where R is the continuously compounding return on an index, calculated aslog price relative � 100, WRTN the return to a world market index, andDayDumi,t is the dummy variable capturing daily periodicity where i=Mon,Tue, Wed, and Thu. The error terms (e1,t, e2,t) are assumed to be normallydistributed with a mean of zero and a conditional variance which is modeledas a GARCH process, which has been modified to include a threshold term


and day of the week dummy variables, i.e.,

hR1;t ¼ b1;c þ b1;h h1;t�1 þ be1;1 e21;t�1 þ be1;2 e

21;t�1 I1;t

þ b1;WVLT WVLTt�1 þXThu

i¼Mon

b1;i DayDumi;t

hR2;t ¼ b2;c þ b2;h h2;t�1 þ be2;1 e22;t�1

þ be2;2 e22;t�1 I2;t þ b2;WVLT WVLTt�2 þ

XThui¼Mon

b2;i DayDumi;t�1

ð2Þ

where I1,t is an indicator variable that takes one where e1,t�1o0, and zerootherwise. I2,t is similarly defined for e2,t�1.

3 The threshold term is designedto capture the asymmetric nature of volatility responses to positive andnegative shocks to the market (see Bollerslev, Engle, & Nelson, 1994).WVLTt�1 is the conditional variance generated from a GARCH(1,1) modelof the world index returns.

In addition to the variance equations, the covariance equation also needsto be specified and a conditional specification is adopted in which allelements are time varying, i.e.,

hR1;R2;t ¼ rt

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffihR1;t � hR2;t

p(3)

where rt is the conditional return autocorrelations of an index which isspecified as:

rt ¼ d0 þ d1 rt�1 þd2 ðe1;t�1 e2;t�1Þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

hR1;t�1 � hR2;t�1

p (4)

The focus of this chapter is on identifying the determinants ofautocorrelation and as such, Eq. (4) may be augmented to include anumber of determinant variables, i.e.,

rt ¼ d0 þ d1 rt�1 þd2 ðe1;t�1 e2;t�1Þffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

hR1;t�1 � hR2;t�1

pþ c1 MRP3t�1 þ c12 MRP4t�1 þ c2 AAPt�1

þ c3 AANt�1 þ c4 MarkovIntt�1 þXThu

i¼Mon

ci DayDumi;t

(5)

where MRP3t�1 (MRP4t�1) is the time series of filtered Markov regimeprobabilities of return regime 3 (4) which corresponds to a negative


(positive) return and high return volatility. These terms and their derivationare explained more fully in Section 2.1. A change in autocorrelation from agiven rise in volatility however, is argued to be less where the underlyingcause for the change in volatility is falling prices. Recognizing this potentialasymmetry in the context of the model, suggests that the coefficientassociated with the high volatility/falling market scenario will be less thanthe coefficient estimated for the high volatility/rising market scenario,i.e.,|c1|o|c12|. AAPt�1 (AANt�1) is a dummy variable that takes the value ofone if an above average positive (negative) return is observed. DayDumi,t isdefined as in (1) and (2).

MarkovIntt�1 is the time-varying probability of integration which isgenerated using the approach of Bekaert and Harvey (1995). Section 2.2.provides a detailed explanation of its derivation. We use it as a proxy for thepresence of foreign investors in the individual stock markets, andhypothesize that a negative coefficient suggests a presence of positivefeedback trading in the market. Dominance of foreigners with predomi-nantly positive feedback trading strategies would imply a lower and possiblya negative conditional autocorrelation. As we focus on a selection ofemerging stock markets that have recently liberalized, investigating theextent to which foreigners dominate the market movements, as proxied bythe integration probabilities, would shed light on the nature of tradingpatterns of these foreign investors.

By augmenting the autocorrelation equation in this way, this chapteravoids the two-step estimation procedure that has been previously used in theliterature, with resulting gains in estimation efficiency. Further, the use ofMarkov probabilities to proxy volatility avoids the issue of endogeneity thatoccurs when the proxy and the autocorrelation series are not independent.4

2.1. Markov Regime Shifting Models of Index Return Volatility

The observed volatility clustering in high frequency return series may beexplained by the existence of different regimes with different variances presentin the data generating process. These regimes can be modeled as a pureMarkov switching variance process (see Turner, Starz, & Nelson, 1989; Kim,Nelson, & Startz, 1998). We use the Markov model of Bollen, Gray, andWhaley (2000) to generate the regime probabilities which are interpreted as aproxy for volatility in that series. The return R in period t is defined as:

Rt ¼ mMSP1;t þ et; et � Nð0;s2MSP2;tÞ (6)


where, MSP1 is the first order, two state Markov switching process thatdrives the return and has the transition probability of:

Pm pm 1� pm

1� qm qm

" #(7)

Depending on the state governed by MSP1 the mean return could beeither m1ðState ¼ 1Þ or m2ðState ¼ 2Þ, where m1om2. The variance of theerror term, et, is driven by another first order, two state independentMarkov switching process, MSP2 whose transition probability is:

Ps ps 1� ps

1� qs qs

" #(8)

Thus, the variance could be either s21ðState ¼ 1Þ or s22ðState ¼ 2Þ,depending on the state. We have s21os22. It is clear from (6) that the modelfor the return generating process is conditionally normal and the parametersof the distribution depend on the state under consideration. But thenature of the two independent Markov switching processes suggeststhat we have four different state combinations to consider. These arefMSP1;MSP2g ¼ fðm1;s

21Þ; ðm2;s

21Þ; ðm1;s

22Þ; ðm2; s

22Þg. That is, there are four

separate regimes that need to be considered: Regime 1, low mean (negativereturn) state and low volatility state; Regime 2, high mean (positive return)and low volatility; Regime 3, low mean (negative return) and highvolatility; and Regime 4, high mean (positive return) and high volatility.Using Eqs. (7) and (8), the overall transition probability of the combinedprocess can be written as:

Pm:ps Pmð1� psÞ

Pm:ð1� qsÞ Pm:qs

" #(9)

Since the return generating process is conditionally normal, it isstraightforward to write the conditional density function of the jointprocess given a state pair (a regime). We multiply the conditional densitiesfor different states by the corresponding probabilities of the states and sumthem to obtain the likelihood function. Next, we maximize the weightedlikelihood function numerically with respect to the parameters of the model,which are Y ðm1; s

21; m2; s

22; pm; qm; ps; psÞ. This algorithm generates the

filtered probabilities of each state, i.e., the probability of a particular stateoccurring given the information up to that point in time. These are the timeseries of return/volatility regime probabilities that represent the market


participants’ view of the state of return/volatility in the individual country. Inthis chapter, the time series of the resulting regime probabilities are used toexplain the time-varying nature of conditional return autocorrelations. As theRegimes 1 and 2 probabilities will contain the same information (withopposite sign) as the high volatility regime probabilities (Regimes 3 and 4),our model only formally considers the latter as exogenous variables in Eq. (5).

2.2. Conditional Stock Market Integration

Bekaert and Harvey (1995) specify two regimes of market integration:(1) complete integration to world market where individual market returnsare a function of the covariance between the individual market return andthe world index returns, scaled by a world covariance risk factor and(2) complete segmentation where the individual market return is determinedin isolation and by own variance scaled by a representative investor’srelative risk aversion. We adopt their model and generate the time-varyingintegration probabilities. The completely integrated market return forcountry i is given by

rit ¼ a1 þ b1 r

it�1 þ lt COVðri

t; rwt Þ þ �

i1;t (10)

Where rit is a daily index return for country i, COV( ) is the conditional

covariance between the country i’s index return and the world index return,lt is time-varying world price of covariance risk, and �i

1;t is iid with ðmi1;s

i1Þ.

On the other hand, in completely segmented markets, the index returnsare determined as

rit ¼ a2 þ b2 r

it�1 þ li

t VARðritÞ þ �

i2;t (11)

Where VAR( ) is the conditional variance of country i’s index return, and lit

is country i’s time-varying price of risk, and �i2;t is iid with ðmi

2;si2Þ.

5

The standard Hamilton (1989, 1990) model of two state Markov regimeswitching with constant transition probabilities is adopted where the twotransition probabilities are shown as below:

P ¼ prob½St ¼ 1jSt�1 ¼ 1�; Q ¼ prob½St ¼ 2jSt�1 ¼ 2� (12)

Using Eqs. (10), (11) and (12), we generate the time series of the smoothedprobabilities, p�t , of individual countries being in the integration state(St=1),6 and this is used as the time-varying probability of integrationMarkovIntt in Eq. (5).


3. DATA AND RESULTS

The analysis of this chapter focuses on Datastream national stock marketindex prices for a range of Asia-Pacific emerging markets or markets whichup until recently were classified as emerging.7 Although data for a wide rangeof countries in this region is available, only those that provided a sufficientlylong sample period for analysis were included in this study. Data for a totalof nine Asia-Pacific stock markets were sampled at a daily frequency8 overthe longest period January 1988 to May 2004 giving a total of 4,266observations (China and Indonesia have shorter sampling periods, starting inJuly 1993 and April 1990, respectively). The continuously compoundingreturns data were computed from these index data and descriptive statisticsare provided in Panel A of Table 1. The US and Japanese stock markets arealso included to provide a control benchmark against which the estimationresults for the emerging markets may be compared. As such, returns data forthe stock market indices of these two countries are sampled and theirdescriptive statistics are included in Panel B of Table 1.9

The average annualized return across most of these markets is lowercompared to the US market. Hong Kong is the only exception with amarginally higher mean return. Japan’s long-suffering economy is mirroredin the poor performance of its stock market, which recorded an average of�1.25%. The only emerging market to generate a negative average returnwas Indonesia (�10.85) whose market continue to suffer from theaftereffects of the 1997 currency crisis. Consistent with the previousliterature, the volatility of these markets is substantially higher thandeveloped markets. All nine Asia-Pacific markets generated higher annual-ized standard deviation estimates compared to the US. The distribution ofthese returns is skewed and also feature excess kurtosis. The daily maximumrise in the value of the index exceeds 20% for Indonesia, Korea andMalaysia. Similarly, daily price falls in excess of 20% were observed forthese same countries plus Hong Kong. This suggests that the potential forsubstantial capital gains as well as losses are more common in these markets.

3.1. Regime Switching Estimates of Volatility

The literature suggests that one of the primary determinants of autocorrela-tion is volatility. In this chapter, volatility proxies are generated usingMarkov regime switching models as detailed in Section 2.1. Table 2 reportsthe estimated parameters of the four-regime Markov models driven by the


Table 1. Summary of International Stock Market Returns.

Annualized

Mean

Annualized

Standard

Deviation

Skewness Kurtosis Daily

Max

Daily

Min

Autocorrelation

ria Average ri,t Max ri,t Min ri,t

Panel A: Sample markets

China 5.05 31.24 0.04 4.69 10.71 �14.29 0.18 0.17 0.46 �0.16

Hong Kong 9.38 25.73 �1.08 22.32 15.56 �25.41 0.02 0.10 0.26 �0.08

Indonesia �10.85 48.25 �0.68 76.30 52.25 �52.95 �0.04 0.24 0.66 �0.77

Korea 2.42 37.36 0.32 12.07 26.87 �21.65 0.05 0.02 0.09 �0.06

Malayasia 6.04 28.53 �1.42 64.30 22.99 �36.77 0.09 0.13 0.30 �0.12

Philippines 5.32 27.19 0.77 10.28 19.55 �9.71 0.14 0.15 0.42 �0.04

Singapore 5.67 19.94 �0.10 7.28 10.62 �9.94 0.09 0.08 0.20 �0.02

Taiwan 4.56 34.25 0.00 2.52 13.73 �12.30 0.04 0.04 0.12 �0.07

Thailand 5.44 33.06 0.36 6.74 16.35 �15.89 0.12 0.11 0.35 �0.02

Panel B: Control sample markets

Japan �1.25 22.30 0.21 3.72 11.53 �8.22 0.08 0.08 0.21 �0.01

USA 9.20 16.05 �0.23 4.54 5.37 �7.03 0.02 0.08 0.24 �0.21

Note: This table presents a statistical summary and unconditional autocorrelation (r1) estimates for a range of daily stock market returns

sampled over the longest period January 1988 to May 2004. The mean, maximum and minimum conditional autocorrelation (rit) estimate

generated by a bivariate GARCH model as specified in Eqs. (3) and (4), are also provided.aAll are significance at least at 5%.

SUK-JO

ONG

KIM

AND

MIC

HAEL

D.McK

ENZIE

74

Table 2. Markov Regime Switching Volatility Model Estimates.

Pm Qm Ps Qs s1 s2 m1 m2 Log-L


China 0.5882�� 0.9452�� 0.9463�� 0.9102�� 0.0001�� 0.0008�� 0.0016�� 0.0163�� 7,490

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0003} {0.0000}

Hong Kong 0.9982�� 0.0462 0.9828�� 0.9554�� 0.9364�� 5.7929�� 12.0725�� 0.0909�� 7,256

{0.0000} {0.7508} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Indonesia 0.9567�� 0.4315�� 0.8218�� 0.9753�� 0.0002�� 0.0060�� 0.0275�� 0.0014�� 9,395

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Korea 0.9265�� 0.0809� 0.9864�� 0.9734�� 1.1393�� 8.1182�� 0.1579�� 2.2501�� 8,366

{0.0000} {0.0792} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Malaysia 0.9538�� 0.4300�� 0.9726�� 0.8826�� 0.4621�� 8.8443�� 0.0238 1.5425�� 6,428

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.2459} {0.0000}

Philippines 0.9083�� 0.6198�� 0.9619�� 0.8945�� 0.5171�� 5.2236�� 0.1714�� 1.0242�� 6,816

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Singapore 0.4115�� 0.9211�� 0.9713�� 0.9287�� 0.3609�� 3.2651�� 0.0846�� 1.0941�� 6,012

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Taiwan 0.9239�� 0.0000 0.9663�� 0.9399�� 1.2112�� 8.4068�� 0.1118�� 2.4657�� 8,656

{0.0000} {0.9505} {0.0000} {0.0000} {0.0000} {0.0000} {0.0003} {0.0000}

Thailand 0.9438�� 0.3338�� 0.9756�� 0.9381�� 0.9688�� 8.5143�� 0.1047�� 2.2245�� 8,027

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0284} {0.0000}


Japan 0.7837�� 0.7593�� 0.9748�� 0.9596�� 0.4747�� 2.7589�� 0.2787�� 0.3712�� 6,361

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0004} {0.0008}

USA 0.9673�� 0.1579�� 0.9889�� 0.9815�� 0.3522�� 1.9792�� 1.3776�� 0.1080�� 5,551

{0.0000} {0.0495} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Note: Regime 1, low mean and low volatility; Regime 2, high mean and low volatility; Regime 3, low mean and high volatility, and Regime 4,

high mean and high volatility.�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.

Co

nd

ition

al

Au

toco

rrelatio

na

nd

Sto

ckM

ark

etIn

tegra

tion

75

two independent Markov switching processes. The mean returns m1, m2indicate negative and positive stock returns, respectively, with respect to themarket indices analyzed. The transition probabilities Pm and Qm help us inferthe persistence of these two different regimes. A high value of Pm relative toQm, indicates that the probability of encountering a negative return period isvery high during the sample period. Similarly, the probability ofencountering positive return period is quite low. The two estimated varianceparameters suggest different levels of variances in the two regimes. Thehigher variance is bigger by a factor ranging from about seven to fortycompared to the variance in the low-variance regime. This is similar toresults reported in Bollen et al. (2000). The transition probabilities for thevariance regimes suggest that in all cases indices have high propensity to stayin a particular variance regime once they are in that regime. Bollen et al.(2000) explore this particular finding in the context of currency optionpricing.

To provide a feel for these regime probabilities, Fig. 1 presents arepresentative plot of these four regime states for the Korean stock. Notethat Regime 1, negative returns and low volatility; Regime 2, positive returnand low volatility; Regime 3, negative returns and high volatility; andRegime 4, positive returns and high volatility. These probability plots aretypical of the Markov model results for all of the countries included in thesample. These coefficients reveal that the probability of the market being inone of the two low volatility states is high a majority of the time. Quite sharpand sudden reversals of these probabilities can be seen however, suggestingthat these tranquil periods are interspersed with a number of high volatilityepisodes, which is consistent with the volatility clustering phenomena. Forthese Korean probabilities, the correlation between Regime 1 and Regime 3(4) is �0.7130 (�0.7320) while the correlation between Regime 2 andRegime 3 (4) is �0.3967 (�0.2626). The two high volatility regimes exhibit apositive association with a correlation between Regimes 3 and 4 of 0.5650.

3.2. Regime Switching Estimates of Integration

In this chapter, we investigate the impact of the presence of foreign investorson emerging stock market autocorrelation, where the Bekaert and Harvey(1995) time-varying measure of capital market integration is used to proxyfor the presence of foreign investors. As such, Eqs. (10)–(12) are estimatedfor the 9 national stock market indices which comprise our sample and theintegration probabilities, p�t , are presented in Fig. 2. Table 3 reports the


Markov Regime Switching Probabilities for Korea

Regime 1: Negative returns and low volatility

1988 1990 1992 1994 1996 1998 2000 2002 20040.00

0.25

0.50

0.75

1.00

Regime 3: Negative returns and high volatility

1988 1990 1992 1994 1996 1998 2000 2002 20040.0

0.2

0.4

0.6

0.8

1.0

Regime 2: Positive returns and low volatility

1988 1990 1992 1994 1996 1998 2000 2002 20040.00

0.25

0.50

0.75

1.00

Regime 4: Positive returns and high volatility

1988 1990 1992 1994 1996 1998 2000 2002 20040.00

0.25

0.50

0.75

1.00

Fig. 1. Markov Regime Switching Probabilities for Korea.

Co

nd

ition

al

Au

toco

rrelatio

na

nd

Sto

ckM

ark

etIn

tegra

tion

77

Fig. 2. Time-Varying Integration Measures. The following figures present estimates

of time-varying integration which is the time series of the smoothed probabilities of

individual counties being in the integration state (St=1) where the transition

probabilities are P=prob[St=1|St�1=1], and the integrated market returns are

given by rit ¼ a1 þ b1 r


t; rwt Þ þ �

i1;t.


Table 3. Markov Integration Model Parameter Estimates.

P Q a1 a2 b1 b2 s1 s2 Log-L Ave-P� P�-Pre-1997 P�-Post-1997


China 0.9353�� 0.9427�� 0.0125 �0.2623�� 0.1957�� 0.1069�� 2.6229�� 0.9658�� 5,461 0.4588 0.3444 0.5263

{0.0000} {0.0000} {0.8788} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000}

Hong Kong 0.9357�� 0.9821�� 0.3704�� 0.0808�� 0.0063 0.0412�� 2.8242�� 1.044�� 7,400 0.2097 0.1462 0.2993

{0.0000} {0.0000} {0.0058} {0.0001} {0.8956} {0.0277} {0.0000} {0.0000}

Indonesia 0.9385�� 0.9827�� 0.5186�� 0.3293�� 0.0686�� 0.0601�� 5.9372�� 1.3565�� 7,781 0.2149 0.0936 0.3458

{0.0000} {0.0000} {0.0163} {0.0000} {0.0000} {0.0003} {0.0000} {0.0000}

Korea 0.9711�� 0.9882�� 0.0921 �0.3015�� 0.0769�� 0.0216 3.7412�� 1.4319�� 8,936 0.2792 0.1222 0.5005

{0.0000} {0.0000} {0.3288} {0.0000} {0.0048} {0.2342} {0.0000} {0.0000}

Malaysia 0.9258�� 0.9856�� 0.3051�� 0.0894�� 0.0733�� 0.1256�� 3.9021�� 0.8945�� 6,869 0.1628 0.0928 0.2615

{0.0000} {0.0000} {0.0208} {0.0000} {0.0043} {0.0000} {0.0000} {0.0000}

Philippines 0.9601�� 0.891�� 0.0053 �0.5104�� 0.1051�� 0.1322 1.0087�� 2.8448�� 7,639 0.7379 0.7288 0.7508

{0.0000} {0.0000} {0.7805} {0.0000} {0.0003} {0.0000} {0.0000} {0.0000}

Singapore 0.9454�� 0.9833�� 0.1923�� 0.0537�� 0.1033�� 0.0518�� 2.0671�� 0.8361�� 6,400 0.2298 0.1126 0.395

{0.0000} {0.0000} {0.0038} {0.0004} {0.0000} {0.0064} {0.0000} {0.0000}

Taiwan 0.9467�� 0.9711�� 0.1346 �0.2747�� 0.0505�� 0.0071 3.1131�� 1.3759�� 8,887 0.3546 0.3765 0.3238

{0.0000} {0.0000} {0.1067} {0.0000} {0.0426} {0.7516} {0.0000} {0.0000}

Thailand 0.939�� 0.9729�� 0.1555� �0.207�� 0.1218�� 0.0928�� 3.2627�� 1.1843�� 8,399 0.3063 0.2109 0.4409

{0.0000} {0.0000} {0.0932} {0.0000} {0.0005} {0.0000} {0.0000} {0.0000}

Pane B: Control sample markets

Japan 0.9663�� 0.9765�� 0.1322�� 0.1262�� 0.0797�� 0.0713�� 1.8777�� 0.9237�� 7,130 0.4022 0.2922 0.5573

{0.0000} {0.0000} {0.0039} {0.0000} {0.0005} {0.0010} {0.0000} {0.0000}

USA 0.974�� 0.9865�� 0.194�� 0.0418 �0.0035 0.0506�� 1.4602�� 0.6421�� 5,534 0.3461 0.1514 0.6206

{0.0000} {0.0000} {0.0000} {0.2333} {0.9054} {0.0061} {0.0000} {0.0000}

Regime 1: rit ¼ a1 þ b1 r


t; rwt Þ þ �

i1;t.

Regime 2: rit ¼ a2 þ b2 r

it�1 þ li

t VARðritÞ þ �

i2;t.

Where rit is a daily index return for country i, COV ( ) is the conditional covariance between the country index i and the world index returns,

lt is time-varying world price of covariance risk, VAR ( ) is the conditional variance of country index i returns, and lit is country i’s time-

varying price of risk.�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.

Co

nd

ition

al

Au

toco

rrelatio

na

nd

Sto

ckM

ark

etIn

tegra

tion

79

Markov model estimations. It is noticeable that both P and Q are fairly highfor all countries, suggesting that once a market enters a state it tends to stayin that state. The coefficients for the lagged returns, b1 and b2, measure theextent of autocorrelation in return in each state, and the average of the twois essentially the same as the relevant ri for each country. They are, inessence, disaggregated ri in Table 1. In five out of nine countries theautocorrelation is higher in the integrated state, so there is no generalpattern of significant difference in autocorrelation coefficient between thetwo states. Another interesting result of note is that the standard deviationof returns in the integrated state is significantly higher than that of the non-integrated state in all cases except for the Philippines. This suggests thatonce a country moves to the integrated state, its exposure to the vagaries ofthe world market causes an increase in the level of market volatility.

Bekaert and Harvey (1995) found that shifts in the indicated level ofintegration could be traced back to political and economic events, whichimpacted on either the willingness or the ability of international investors toaccess the local stock market. A qualitative assessment of the probabilitiesestimated in this chapter produces similar evidence.10 At a general level, it isinteresting to note the impact of the 1997 currency crisis on the integrationparameter for the Asian markets. A clear increase in integration is evidentfor Hong Kong, Indonesia, Korea, Malaysia, Singapore and Thailand thatwere at the center of the speculative attacks. This change reflects thedominance of the global information set over the local one in local assetpricing. Except for the Philippines, there is a clear and interesting trend of asteady decline of the integration parameter starting mid- to late-1998. Onepossible explanation is that the increased dominance of foreign investors inthese markets shortly after the breakout of the crisis was to take advantageof the emerging profitable opportunities, which have started to dissipate asthese markets began the process of recovery from around 1998. As themarkets started to recover, the local information began to dominate thelocal asset pricing once again. This suggests that the bouts of heightenedintegration in the countries were only temporary.

3.3. Conditional and Unconditional Autocorrelation

Unconditional autocorrelation estimates (ri) may be estimated for each ofour indices, i, and the results are presented in Table 1. Except for Indonesia,all of the data series exhibit significant positive first-order autocorrelation.The highest observed level of autocorrelation is 0.18 for China and the


lowest level is �0.02 for Indonesia. To investigate this further time-varyingautocorrelation estimates are generated using the M-GARCH modelspecified in Eqs. (1)–(4). The estimated model coefficients and diagnosticproperties of the residuals are not presented to conserve space and areavailable on request.

The final three columns of Table 1 present a summary of the averageconditional autocorrelation estimates as well as the maximum and minimumobserved values. A comparison of the point estimates of autocorrelation tothe average conditional autocorrelation estimate reveals that these twotechniques provide a similar degree of information about the general level ofobserved autocorrelation, which is consistent with previous research. Theunconditional specification, however, omits important information aboutthe variability of autocorrelation as evidence by the range of conditionalestimates. Indonesia exhibits the greatest range of observations recording amaximum of 0.66 and a minimum value of �0.77 while Korea exhibits thesmallest range of observations (0.09 to �0.06).

The conditional autocorrelation estimates exhibit a good deal ofvariation. To gain a fuller appreciation of the variability of this data,consider Fig. 3 which presents a plot of the data for Korea. The plot clearlyhighlights the variability of autocorrelation and a number of otherinteresting features can also be identified from the data. For example, theestimate hovers around 0.1 and started a downward trend from aroundearly 1996 and stayed close to zero until mid-2002 before it starts to rise.

Indonesia provides another interesting example as the point autocorrela-tion estimate reported in Table 1 is negative and insignificant. Examinationof the conditional values reveals that the early part of the data is chara-cterized by high negative autocorrelation, while the latter part of the sample

1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004-0.20

-0.15

-0.10

-0.05

-0.00

0.05

0.10

0.15

0.20

0.25

Fig. 3. Conditional Correlation for Korea.


exhibits positive autocorrelation. Sub-period analysis using point estimatesverifies this pattern as the autocorrelation from 1998 to 1992 for Indonesia is�0.43 whereas over the latter part of the sample period it takes a value of0.12.

It is reassuring that these shifts in the conditional autocorrelationestimates are consistent with the unconditional values. The variability of theconditional autocorrelation estimates however, suggests that the use ofpoint estimates may be potentially misleading. Further, it is an interestingempirical issue to consider the extent to which the observed variability inautocorrelation can be explained using economic factors and the remainderof this chapter considers this issue.

3.4. Stock Market Autocorrelation and International Investors

To test the determinants of autocorrelation, the bivariate GARCH modelsummarized in Eqs. (1)–(3) and (5) is fitted to the data where MRP3 andMRP4 are the volatility proxies which correspond to Markov regimeprobability (MRP) 3 and 4, respectively, and MarkovIntt is the time-varyingprobability of integration. In addition, above average return and the day-of-the-week dummies are also considered which have been found in theprevious literature to be important. Tables 4A and 4B present the estimationresults of the Eqs. (1)–(3) and (5), respectively.

3.4.1. Conditional Mean and Volatility Estimation Results

We refrain from formally presenting the full model output to keep thepresentation of our results to a manageable level. We present the results forthe R1,t equation (the results for R2,t tend to mirror those of R1,t since theformer is just the one period lagged values of the latter) in (1) and (2). Fullresults are available on request. For the mean equation, a number ofsignificant day-of-the-week terms were estimated and they were almostexclusively negative suggesting the average market movement is typicallyhigher on a Friday, which is the assumed base case. This is especially so forthe first 2 days of the week as 5 (4) countries generated significant andnegative Monday (Tuesday) day-of-the-week coefficients, a1,Mon (a1,Tue).Significant evidence of a relationship between the world market return andthe Asia-Pacific market returns is in evidence as all nine markets showsignificant and positive influence of the world market return.

In terms of the ARCH and GARCH coefficients, all of the estimates aresignificant at the 5% level except for the ARCH (be11) term in the model


fitted to the Chinese return data which had a P-value of 0.15. The thresholdterms (be1,2) capture the presence of asymmetry in the volatility response ofshocks to the market. Ten of the countries generated threshold term that issignificant, although the sign on the term was mixed as half were positive. Incontrast to the mean equations, the day-of-the-week dummy variables in thevariance equation exhibit a mix of positive and negative signs. Overall, thereis certainly evidence of day-of-the-week effects in the volatility of these indexreturns series as all but two coefficients are significant. Notably, Mondayand Wednesday exhibit clear evidence of higher volatility compared to thebase case of Friday with 4 and 5 positive and negative significantcoefficients, respectively. The relationship between the variance of theworld market return and Asia-Pacific stock market volatility is also found tobe strong in six out of nine markets. In 5 out of 6 markets, higher globalmarket volatility is associated with a heightened volatility response in thelocal market the subsequent trading day.

The last two columns of Table 4A present the Ljung Box test of whitenoise for the estimated standardized residuals, zt ¼ et=

ffiffiffiffiht

p. There is

evidence of remaining first moment serial correlation but the secondmoment dependencies are reduced in most cases. Attempts to address thisimperfection led to the differing functional forms (especially with the lagstructures of the B-GARCH models and the number of lagged dependentvariables included in the mean equations) being relevant for the most of the11 (including Japan and the US) return series examined. This addressed theissue, however, the results of the conditional autocorrelation Eq. (4)estimations remain robust regardless of the functional form of the Bivariate-GARCH models selected. Thus, we report the results for the parsimoniousmodels and any conclusion we draw is not dependent on the modelselection.

3.4.2. Conditional Autocorrelation Results

The specification of the covariance equation in the MGARCH modelpresented in Eqs. (1)–(3) and (5), includes a time-varying autocorrelationcoefficient, which is specified as a function of volatility, large returns and theday-of-the-week which the past literature has found to be important. In thischapter, volatility is proxied by the MRP3 (MRP4) variables, which are thetime series of filtered Markov regime probabilities of return Regime 3 (4)which correspond to a period of high volatility and negative (positive)returns. The estimated coefficients for c1 and c12 capture the nature ofthe relationship between autocorrelation and volatility for MRP3 andMRP4,respectively, and are presented in Table 4B. The estimated results reveal that


Table 4A. B-GARCH(1,1) Estimations (January 1988–May 2004, Eqs. (1)and (2)).

a1,c a1,Lag a1,WRTN a1,Mon a1,Tue a1,Wed a1,Thu b1,c b1,h


China 0.1234�� 0.1004 0.3067�� 0.1119 �0.1538� �0.1531�� 0.1680�� 0.1845 0.9216��

{0.0004} {0.1011} {0.0000} {0.1051} {0.0616} {0.0291} {0.0253} {0.2021} {0.0000}

Hong Kong 0.1101�� 0.1838�� 0.5260�� 0.1155�� 0.0073 0.0064 �0.1090�� 0.1006�� 0.8723��

{0.0000} {0.0000} {0.0000} {0.0085} {0.8529} {0.8770} {0.0028} {0.0000} {0.0000}

Indonesia 0.1330 �0.0529 0.2867�� 0.0942 �0.1495 �0.0716 0.0136 �4.0061�� 0.8011��

{0.1343} {0.2161} {0.0000} {0.3393} {0.1129} {0.4608} {0.9008} {0.0005} {0.0000}

Korea 0.0122 �0.3664�� 0.3456�� 0.0108 0.0265 �0.0170 �0.0093 �0.1406�� 0.9040��

{0.5037} {0.0000} {0.0000} {0.8320} {0.7290} {0.7803} {0.8410} {0.0000} {0.0000}

Malaysia 0.1016�� 0.1262�� 0.2375�� 0.2162�� 0.0604 0.0012 �0.0147 0.0854�� 0.9277��

{0.0413} {0.0000} {0.0000} {0.0003} {0.3744} {0.9838} {0.7926} {0.0000} {0.0000}

Phillippines 0.1169�� 0.0301 0.3152�� 0.1309�� 0.1708�� 0.1159�� 0.0209 �0.0321�� 0.8189��

{0.0000} {0.1298} {0.0000} {0.0009} {0.0000} {0.0080} {0.5825} {0.0000} {0.0000}

Singapore 0.1194�� 0.0682�� 0.3143�� 0.1787�� 0.0853�� 0.0376 �0.0230 0.0648�� 0.7993��

{0.0000} {0.0000} {0.0000} {0.0000} {0.0147} {0.2242} {0.4309} {0.0000} {0.0000}

Taiwan 0.0325� �0.1576�� 0.3905�� 0.0107 �0.1200 0.0157 0.0544 0.0608�� 0.9161��

{0.0882} {0.0004} {0.0000} {0.8776} {0.2527} {0.7904} {0.3773} {0.0000} {0.0000}

Thailand 0.2145�� 0.2273�� 0.3407�� 0.2988�� 0.2045�� 0.0626 �0.1249�� 0.0936�� 0.9204��

{0.0000} {0.0000} {0.0000} {0.0000} {0.0006} {0.2379} {0.0149} {0.0000} {0.0000}


Japan �0.0419�� 0.1013�� 0.5557�� 0.0697 0.0704� 0.0187 0.0792�� 0.0429�� 0.8550��

{0.0289} {0.0000} {0.0000} {0.1199} {0.0582} {0.6345} {0.0435} {0.0000} {0.0000}

USA 0.0062 �0.3729 0.0188 0.0372 0.0274 0.0410 0.0141 0.2338�� 0.3397��

{0.6894} {0.2901} {0.3252} {0.2170} {0.4195} {0.1201} {0.6481} {0.0000} {0.0000}

Note: Q(20) and Q2(20) are the Ljung–Box test of white noise for the linear and non-linear (squared)

standardized residuals.�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.


b1e,11 b1e,12 b1,WVLTt

b1,Mon b1,Tue b1,Wed b1,Thu Log-L Q(20) Q2(20)

0.0751 �0.0064 �0.0177 0.6352�� 0.0454 0.6126�� 0.1167 �5667 32.0963�� 28.7913�

{0.1425} {0.8067} {0.3943} {0.0005} {0.8169} {0.0136} {0.4840} {0.0423} {0.0920}

0.1144�� 0.0241�� 0.0016 0.8479�� 0.4035�� 0.2980�� 0.0643�� 6640 26.9071 41.7179��

{0.0000} {0.0000} {0.6488} {0.0000} {0.0000} {0.0000} {0.0000} {0.1379} {0.0030}

0.1199�� 0.0959�� 0.2946� 4.2354�� 3.9340�� 4.0089�� 8.9229�� 8899 52.4673�� 0.72791

{0.0078} {0.0068} {0.0551} {0.0001} {0.0011} {0.0007} {0.0003} {0.0001} {1.0000}

0.0575�� 0.0272�� 0.1305�� 1.0817�� 0.9269�� 0.2692�� 0.4631�� 9607 29.5808� 15.8129

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0769} {0.7282}

0.0514�� 0.0316�� 0.0031 0.1956�� 0.5182�� 0.0980�� 0.1056�� 6047 36.238�� 7.80048

{0.0000} {0.0000} {0.2692} {0.0381} {0.0000} {0.0000} {0.0000} {0.0144} {0.9931}

0.1458�� 0.0042 0.0740�� 0.8863�� 0.5400�� 0.0379�� 0.1635�� 7437 48.4629�� 3.19074

{0.0000} {0.3857} {0.0000} {0.0000} {0.0000} {0.0032} 0.0291 {0.0004} {1.0000}

0.1389�� 0.0287�� 0.0668�� 0.4039�� 0.3250 �0.1057�� 0.0411�� 4836 21.5871 14.8716

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0056} {0.3633} {0.7837}

0.0728�� 0.0182�� 0.0271�� 1.5890�� 1.9068 0.1646�� 0.1596�� 9598 50.79�� 24.392

{0.0000} {0.0000} {0.0001} {0.0000} {0.0000} {0.0000} {0.0000} {0.0002} {0.2257}

0.0603�� 0.0210�� 0.0097�� 0.2083�� 0.1716�� 0.1398�� 0.4005�� 8783 52.0058�� 26.931

{0.0000} {0.0000} {0.0006} {0.0002} {0.0238} {0.0000} {0.0000} {0.0001} {0.1372}

0.0910�� 0.0074 0.0887�� 0.4647�� 0.1640�� 0.1622�� 0.0383�� 6082 27.6897 18.7694

{0.0000} {0.3769} {0.0000} {0.0000} {0.0000} {0.0000} {0.0067} {0.1169} {0.5369}

0.0348 0.0067 0.9641�� 0.1288� �0.1033�� 0.2019�� 0.1004�� 3322 33.0391�� 11.2671

{0.4754} {0.7734} {0.0000} {0.0713} {0.0000} {0.0000} {0.0013} {0.0334} {0.9390}


Table 4B. The Determinants of Autocorrelation (January 1988–May 2004, Eq. (5)).

d0 d1 d2 c1 c12 c2 c3 c4 H0: c1=c12 Q(20) Q2(20)


China 0.5772�� 0.3939�� 0.0073 �0.1449�� 0.0447 �0.0366 �0.0748� �0.1095�� 9.7316�� 32.0963�� 28.7913�

{0.0000} {0.0000} {0.3266} {0.0000} {0.2265} {0.3353} {0.0716} {0.0003} {0.0018} {0.0423} {0.0920}

Hong Kong 0.1495�� 0.0470 0.0013 �0.0458 0.0628�� 0.0567�� 0.0540�� 0.1026�� 2.6291 26.9071 41.7179��

{0.0000} {0.5164} {0.7613} {0.4556} {0.0038} {0.0150} {0.0042} {0.0000} {0.1049} {0.1379} {0.0030}

Indonesia 0.4009�� 0.4324�� 0.0126 �0.3685�� 0.3288�� 0.1953�� 0.1166� �0.0890�� 24.7838�� 52.4673�� 0.72791

{0.0000} {0.0008} {0.1447} {0.0000} {0.0005} {0.0002} {0.0621} {0.0001} {0.0000} {0.0001} {1.0000}

Korea 0.4303�� 0.1154�� 0.0068 �0.2750�� 0.2226�� 0.0616�� 0.0355� 0.0072 26.3239�� 29.5808� 15.8129

{0.0000} {0.0000} {0.2704} {0.0000} {0.0004} {0.0061} {0.0683} {0.6550} {0.0000} {0.0769} {0.7282}

Malaysia �0.0774 �0.9797�� 0.0045 0.0324 0.0887 �0.0067 0.0076 0.0098 0.9325 36.238�� 7.80048

{0.3368} {0.0000} {0.4967} {0.7167} {0.3006} {0.7877} {0.7514} {0.8005} {0.3342} {0.0144} {0.9931}

Philippines �0.0388�� 0.9808�� 0.0021 0.0438 �0.0110 �0.0264 �0.0460 �0.0066 0.0372 48.4629�� 3.19074

{0.0248} {0.0000} {0.8846} {0.6434} {0.9582} {0.6405} {0.2900} {0.8924} {0.8471} {0.0004} {1.0000}

Singapore 0.0785�� 0.2048 0.0025 �0.0649 0.1361�� 0.0906�� 0.0219 0.0133 7.1910�� 21.5871 14.8716

{0.0048} {0.2756} {0.7346} {0.1076} {0.0028} {0.0022} {0.2034} {0.4249} {0.0073} {0.3633} {0.7837}

Taiwan 0.1490�� 0.0376 0.0056 �0.2045�� 0.0301 0.0789�� 0.1227�� 0.0212 16.1408�� 50.79�� 24.392

{0.0000} {0.6588} {0.4198} {0.0000} {0.5471} {0.0029} {0.0000} {0.2546} {0.0001} {0.0002} {0.2257}

Thailand 0.3297�� 0.1972�� 0.0000 0.0331 0.0106 �0.0551�� 0.0186 �0.0987�� {0.2692} 52.0058�� 26.931

{0.0000} {0.0095} {0.9938} {0.1593} {0.7621} {0.0078} {0.4014} {0.0000} {0.6038} {0.0001} {0.1372}


Japan �0.0340�� 0.6558�� 0.0073 �0.0331 0.1616�� 0.0710�� 0.0394�� 0.0139 6.7796�� 27.6897 18.7694

{0.0007} {0.0000} {0.2904} {0.5027} {0.0051} {0.0000} {0.0123} {0.3701} {0.0092} {0.1169} {0.5369}

USA 0.1470� 0.6545�� 0.0188� 0.0312 0.0496 0.0593 �0.0700�� 0.0866 0.0263 33.0391�� 11.2671

{0.0728} {0.0000} {0.0852} {0.8470} {0.4051} {0.2130} {0.0481} {0.3698} {0.8713} {0.0334} {0.9390}

Note: Q(20) and Q2(20) are the Ljung–Box test of white noise for the linear and non-linear (squared) standardized residuals.�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.

SUK-JO

ONG

KIM

AND

MIC

HAEL

D.McK

ENZIE

86

the coefficient for c1 is not significantly different from zero for all countriesexcept China, Indonesia, Korea, and Taiwan where a negative coefficient isestimated. The estimate for c12 is positive and significant for four indices.A Wald test of coefficient equality (i.e., H0: c1=c12) is undertaken and theresults reject the null hypothesis of equality in five cases. This evidencesuggests that volatility is not as significant a determinant of autocorrelationin country index returns as has previously been found in the individual stocksetting. Further, the limited evidence of a relationship in our sample is moremixed compared to the past literature where higher levels of volatility aretypically associated with lower levels of autocorrelation.

A second determinant of autocorrelation which the past literature hasfound to be important is large changes in price which are proxied by aboveaverage positive or negative returns. Six of the coefficients capturing theimpact of above average positive returns (c2 on AAPt�1) are significant andfour are negative. Only five of the above average negative return coefficients(c3 on AANt�1) are significant and three of those are negative. In terms ofthe day-of-the-week effects the only discernible trend across the markets inour sample is for the autocorrelation to be lower on a Tuesday (eightcountries produced a significant and negative coefficient for cTue).

11

In general, it is interesting to note that the past literature has identifiedvolatility, large returns and day-of-the-week effects as significant determi-nants of individual stock autocorrelation. When the impact of thesevariables is considered in a market context, the evidence is generally weakeralthough not entirely inconsistent. These results motivate our search foradditional factors, which may be significant in determining autocorrelationat a market level and in this chapter we propose the presence ofinternational investors. It is to this hypothesis which we now turn ourattention.

The presence of international investors in a market is proxied by the levelof integration which is estimated using the Bekaert and Harvey’s (1995)conditional integration model. The impact of the presence of internationalinvestors on stock market autocorrelation is captured by the c4 coefficient inthe model and parameter estimates are reported in Table 4B. A significantrelationship is found in China, Hong Kong, Indonesia, and Thailand andthe coefficient is negative in all cases. These results suggest a fall in the levelof conditional autocorrelation in returns in response to the increasedpresence of international investors, which accompanies heightened integra-tion. This result is consistent with the view that the international investorsare positive feedback traders (see Dahlquist & Robertsson, 2004; Kim &Wei, 2002; Choe et al., 1999). As their presence in the market increases, their


positive feedback trading activities lessen the observed level of autocorrela-tion, and may even lead to negative autocorrelation in the extreme. The USand the Japanese stock markets are included in this study as a controlsample and the estimation results for this data are presented in Panel B ofTable 4B. Most relevant to the current discussion, the c4 coefficient is notsignificant for either market, which is consistent with expectations. Thevalue and volume of transactions in developed markets are substantial andthe trading strategies employed by investors spans the full spectrum. Assuch, the presence of foreigners is not expected to significantly impact on thedominant trading strategy in the market.

3.5. The Asian Currency Crisis and Emerging Market Autocorrelation

In the early 1990s, international investors began to seek out alternativeinvestment opportunities as bearish sentiment came to dominate traditionalfinancial markets. This resulted in a marked increase in the amount of fundsdirected into the emerging markets sector, which provided a valuable sourceof diversification and high expected returns. In 1997, however, the Asiancurrency crisis caused many international investors to revise theirexpectations of the emerging markets sector and a flight to quality resulted.These events suggest that it is appropriate to test the robustness of theresults presented in the previous section to this regime change. As such, thebivariate GARCH model summarized in Eqs. (1)–(3) and (5) is fitted to datafrom the pre- and post-crisis periods where the onset of the crash is setrelative to the floating of the Bhat on July 2, 1997.

The estimation results are summarized in Table 5 for the pre-crisis period.With respect to the central hypothesis, six of the estimated c4 coefficients aresignificant and five exhibit a negative sign. Thus, while the results are broadlyconsistent with the results estimated over the entire sample, some differencesare noteworthy. First, Hong Kong is insignificant in this pre-crash period.Second, the Philippines, Singapore, and Taiwan are all significant in thecurrent sample and the latter two exhibit a negative sign. These results areconsistent with Choe et al.(1999) who report strong evidence of positivefeedback trading by foreign investors prior to the crisis period.

A summary of the estimation results for the post-crisis period arepresented in Table 6 and the results suggest the speculative attack episode of1997 did cause a change in the market dynamics. Five of the estimated c4coefficients are significant and of those, only China and Malaysia generateda negative sign. Korea, Singapore and Thailand all exhibited a positive and


Table 5. The Determinants of Autocorrelation: Pre-crash Results (January 1988–June 1997, Eq. (5)).

d0 d1 d2 c1 c12 c2 c3 c4 H0: c1=c12 Q(20) Q2(20)


China 0.2387�� 0.3745 0.0257� 0.2175�� 0.1857�� 0.2674�� 0.0713�� 0.2487�� 42.9718�� 18.7938 12.7907

{0.0000} {0.2159} {0.0635} {0.0000} {0.0000} {0.0000} {0.0065} {0.0000} {0.0000} {0.5353} {0.8862}

Hong Kong 0.9969�� 0.0909�� 0.0012 �0.2303�� 0.0111 0.0018 �0.0254� �0.0898 24.5191�� 33.8816�� 44.6629��

{0.0000} {0.0179} {0.2722} {0.0004} {0.7467} {0.8931} {0.0751} {0.1446} {0.0000} {0.0269} {0.0012}

Indonesia 0.1212�� 0.0200 �0.0067�� 0.2801�� 0.0432 0.0477 0.1015�� 0.0636�� 17.8548�� 28.22 0.82492

{0.0003} {0.8551} {0.0266} {0.0006} {0.5299} {0.4761} {0.0000} {0.0062} {0.0000} {0.1043} {1.0000}

Korea 0.0625�� 0.7209�� 0.0077 �0.0718 0.0587 0.0134 �0.1280�� 0.0148 0.9738 32.2118�� 16.2987

{0.0194} {0.0000} {0.6017} {0.3730} {0.3499} {0.7505} {0.0017} {0.6287} {0.3237} {0.0411} {0.6979}

Malaysia �0.0865�� 0.6383�� 0.0093 �0.1959�� 0.4199�� 0.0116 �0.0029 �0.0352 31.9933�� 32.95�� 3.74423

{0.0116} {0.0000} {0.1668} {0.0002} {0.0000} {0.7760} {0.9342} {0.3161} {0.0000} {0.0342} {1.0000}

Philippines 0.3265�� 0.6401�� 0.0059 0.3321�� 0.2184�� 0.0042 �0.0781� 0.2558�� 1.5916 28.6826� 25.3091

{0.0000} {0.0000} {0.5814} {0.0001} {0.0000} {0.9150} {0.0645} {0.0000} {0.2071} {0.0942} {0.1899}

Singapore 0.0644 0.7289�� 0.0044 0.1638�� 0.2111�� 0.0762�� 0.0969�� 0.1207�� 0.3748 19.7066 11.3711

{0.4425} {0.0000} {0.5637} {0.0001} {0.0000} {0.0164} {0.0004} {0.0002} {0.5404} {0.4764} {0.9360}

Taiwan 0.1500�� 0.2728� 0.0032 �0.1632�� 0.0509�� 0.1292�� 0.1370�� 0.0257�� 122.4524�� 43.0634�� 19.194

{0.0000} {0.0505} {0.2448} {0.0000} {0.0027} {0.0000} {0.0000} {0.0149} {0.0000} {0.0020} {0.5092}

Thailand 0.7227�� 0.0346 �0.0087 0.0213 0.0787�� 0.0085 �0.0115 �0.2033�� 3.4113� 38.5413�� 27.2577

{0.0000} {0.1962} {0.1215} {0.4533} {0.0292} {0.7724} {0.7028} {0.0000} {0.0648} {0.0076} {0.1282}


Japan �0.2389�� 0.7269�� 0.0129�� 0.1315�� 0.1377�� 0.0732�� 0.0170 0.1063�� 0.0600 20.4261 65.051��

{0.0000} {0.0000} {0.0000} {0.0002} {0.0000} {0.0000} {0.3761} {0.0000} {0.8066} {0.4316} {0.0000}

USA 0.0811�� 0.6129 0.0058 �0.3276�� 0.0399 0.1531�� 0.0010 0.0269 8.1795�� 17.3647 55.977��

{0.0206} {0.1376} {0.6945} {0.0092} {0.4603} {0.0001} {0.9616} {0.3669} {0.0042} {0.6292} {0.0000}

�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.

Co

nd

ition

al

Au

toco

rrelatio

na

nd

Sto

ckM

ark

etIn

tegra

tion

89

Table 6. The Determinants of Autocorrelation: Post-crash Results (July 1997–May 2004, Eq. (5)).

d0 d1 d2 c1 c12 c2 c3 c4 H0: c1=c12 Q(20) Q2(20)


China 0.0573 0.8566�� 0.0049 0.0821�� 0.2398�� 0.1426�� 0.1114�� 0.0541�� 8.2487�� 30.8723� 38.2253��

{0.1071} {0.0000} {0.7344} {0.0005} {0.0000} {0.0000} {0.0001} {0.0010} {0.0041} {0.0569} {0.0083}

Hong Kong 0.0307�� 0.2965�� 0.0075�� 0.0103 0.0025 0.2012�� 0.1232�� 0.0226 0.3136 19.4091 23.6571

{0.0145} {0.0314} {0.0320} {0.6152} {0.8227} {0.0000} {0.0000} {0.4198} {0.5755} {0.4954} {0.2577}

Indonesia 0.3386�� 0.7441�� 0.0031 �0.2054 0.0800 �0.0775 0.0036 �0.0717 0.7771 39.7735�� 24.724

{0.0000} {0.0000} {0.8145} {0.1282} {0.7117} {0.3476} {0.9126} {0.1600} {0.3780} {0.0053} {0.2122}

Korea 0.3750�� 0.5851�� 0.0069 �0.5017�� 0.0389 �0.0106 0.0697�� 0.1060�� 91.9744�� 22.8767 11.268

{0.0000} {0.0000} {0.3314} {0.0000} {0.2169} {0.6623} {0.0083} {0.0000} {0.0000} {0.2949} {0.9390}

Malaysia 0.9299�� 0.0322 �0.0172�� 0.1376�� 0.0010 �0.0320 �0.0060 �0.4904�� 3.3377� 65.7169�� 5.82855

{0.0000} {0.7219} {0.0008} {0.0034} {0.9843} {0.3077} {0.8421} {0.0000} {0.0677} {0.0000} {0.9991}

Philippines 0.5339� �0.7142 0.0238 �0.3247�� 0.4722 0.1727 0.1056 �0.1887 0.0413 28.0167 2.32623

{0.0502} {0.6682} {0.8993} {0.0000} {0.5277} {0.8820} {0.9154} 0.9136 {0.8390} {0.1090} {1.0000}

Singapore 0.0666�� 0.1021 0.0033 �0.5092�� 0.0456� �0.0886�� 0.1928�� 0.2276�� 295.724�� 15.4847 25.441

{0.0000} {0.3408} {0.2729} {0.0000} {0.0877} {0.0000} {0.0000} {0.0000} {0.0000} {0.7480} {0.1851}

Taiwan 0.0655 �0.2067� 0.0094�� 0.4029�� 0.2708�� 0.0775 0.3345�� 0.0493 14.8593�� 21.8972 15.9624

{0.6296} {0.0985} {0.0022} {0.0000} {0.0039} {0.4045} {0.0210} {0.4540} {0.0001} {0.3461} {0.7190}

Thailand 0.2575�� 0.4632�� 0.0222�� 0.1196�� 0.1937�� 0.0482�� 0.1613�� 0.1090�� 6.5628�� 40.2385�� 20.6534

{0.0000} {0.0000} {0.0000} {0.0000} {0.0000} {0.0027} {0.0000} {0.0000} {0.0104} {0.0047} {0.4178}


Japan �0.0471 0.7399 �0.0140� �0.0813 0.0874 0.1495�� 0.0453 �0.0803�� 1.8362 29.828� 87.3749��

{0.3733} {0.1385} {0.0749] {0.4465} {0.3433} {0.0052} {0.4435} {0.0461} {0.1754} {0.0727} {0.0000}

USA 0.0233� 0.5070�� 0.0307�� 0.2548�� 0.1810�� 0.0356 �0.0924�� 0.1888 2.4636 20.7124 19.925

{0.0728} {0.0000} {0.0186} {0.0000} {0.0000} {0.3637} {0.0001} {0.0000} {0.1165} {0.4142} {0.4626}

�, Significance at 10%.��, Significance at 5%.��, Significance at 1%.

SUK-JO

ONG

KIM

AND

MIC

HAEL

D.McK

ENZIE

90

significant sign suggesting the increased presence of foreign traders led tohigher levels of autocorrelation. Two possible interpretations of our resultsexist. On the one hand, international investors may have withdrawn fromthe market and the dominant trading strategy among the local investorsmay have been contrarian in nature. As the exodus of foreign capitalcontinued immediately after the breakout of the Asian financial crisis, in theabsence of the positive feedback trading of foreign investors, the influence ofthe contrarians on the market would have increased, and this resulted inhigher levels of autocorrelation. This also explains the gradual decline in theintegration probabilities in all Asian countries (see Fig. 2) except for thePhilippines. Alternatively, international investors may have modified theirpreferred trading strategy to suit the new regime. The trading strategieswhich prove profitable during the bull run observed in the lead up to thecrisis are unlikely to prove successful in the post crisis market and so this is arational response of investors to such a significant change to the market.This interpretation of our results is consistent with Choe et al. (1999) whofound that the evidence of positive feedback trading by foreigners all butdisappeared after the crisis.

4. CONCLUSIONS

The capital flows of international investors have been subject of a great dealof interest in the academic literature. In this chapter, we investigate theimpact of the trading strategies employed by international investors on nineemerging Asia-Pacific stock market dynamics. Specifically, the stock marketwill exhibit a given level of autocorrelation, which reflects the amount andtype of feedback trading. The presence of international investors mayinfluence the observed level of autocorrelation if they pursue feedbacktrading strategies and the nature of the relationship will reflect the typeof feedback trading strategy employed. Drawing from a sample ofstock indices for a range of emerging or newly emerged markets in theAsia-Pacific, we test this hypothesis where the presence of foreigners isproxied by a time-varying measure of capital market integration.

The results of our analysis find important evidence of a significantrelationship between the presence of international investors and the level ofstock market autocorrelation. Specifically, lower levels of conditionalautocorrelation in returns are associated with the increased presence ofinternational investors. This result is consistent with the view that theinternational investors are positive feedback traders and is supported by


previous research. The nature of the relationship however, may change overtime. For example, analysis of our model for post-1997 Asian currency crisisdata suggests that the extent to which positive feedback trading is a feature ofthe market has diminished and foreign investors either withdrew from themarket or modified their trading strategies to suit the new regime. In addition,we find that volatility is not as significant a determinant of autocorrelation ofemerging market stock index returns as has previously been found in theindividual stock setting. The limited evidence of a relationship in our sampleis more mixed compared to the past literature where higher levels of volatilityare typically associated with lower levels of autocorrelation.

NOTES

1. The use of feedback trading strategies by international investors does not implyirrationality. Choe et al. (1999) argue that where informational asymmetries exist,the trades of local investors reveal their informational advantage to foreigners whowill then trade based on this information embodied in price changes. Thus, upwardprice movements suggest good news, which causes foreigners to trade in what may beincorrectly interpreted as irrational positive momentum trading.2. Albuquerque, Loayza, and Serven (2005) argue that, ‘‘the process of integration

starts with the removal of capital market restrictions, most notably the liberalizationof foreign investors’ participation in domestic stock markets, the listing of domesticfirms in foreign markets, and the privatization of state-owned companies’’3. EGARCH specifications were also tested and the results (available on request)

are qualitatively unchanged to those obtained using the GJR models reported in thischapter.4. McKenzie and Faff (2003) generated conditional autocorrelation estimates

using an M-GARCH model and subsequently tested the relationship betweenautocorrelation and its determinants in a SUR framework. The conditional variancefrom this GARCH model was used to proxy volatility and also appeared as thedenominator in the autocorrelation estimate.5. The time varying price of risk for each country is generated in a similar fashion

to Bekaert and Harvey (1995, p417, 419). It is a time varying coefficient, lit, attached

to the conditional variance (ARCH-M term) included in the conditional meanequation of the ARCH-M model of the index returns. It is conditioned on eachcountry market’s dividend yield and exchange rate volatility. The price of worldcovariance risk, lt, is similarly generated. It is a time varying coefficient on theconditional variance term in the mean equation of the ARCH-M model of the dailyreturn of the world index. It is conditioned by world market dividend yield in accessof the 30-day Eurodollar rate, the spread between the US 10 year bond and 3-monthrates, and the change in the 30-day Eurodollar rate.6. Bekaert and Harvey (1995) provide detailed discussions on the modeling issues

and interested readers are referred to their chapter for further discussions on theissue.


7. This distinction is necessary as Singapore, Hong Kong are no longer classifiedas emerging countries according to the International Finance Corporation.8. A relevant issue given our choice of daily data is whether, as assumed by the

theoretical model developed earlier, investors undertake shifts in risk bearingactivities on a daily basis. The following comments justify our stance. First, themajority of the technical trading literature focuses on daily decisions made byinvestors, which implicitly assumes that they do modify (or at least act as if theymodify) their risk bearing activities to reflect changing conditions in the market on adaily basis. Second, not all investors must update their portfolios every day. Whereonly a subset of investors update at any point in time, say weekly, and imperfectcorrelation exists between the trading activities of each subset (such that their tradingis spatially distinct), we will be able to observe shifts in risk bearing activities on acontinual basis. Third, the bulk of previous literature in this area has also used dailydata and for reasons of consistency, the same interval is chosen for analysis in thischapter.9. Eighteen developed markets were included as a control sample and the results

are qualitatively consistent across all markets. To limit the presentation of our resultsto a manageable level, we chose to focus on Japan and the US only which are the twolargest stock markets in the world (2003, World Federation of Exchanges data). Theregional index of Asia ex-Japan across our entire sample was also tested. Full detailsof the estimation results for all developed markets and indices are available onrequest.10. A detailed account of financial, political and economic events for a wide range

of emerging and developing markets can be found at www.duke.edu/~charvey/Country_risk/couindex.htm.11. The day of the week effects are not reported in the table to conserve space.

Interested readers can obtain these results upon request.

REFERENCES

Albuquerque, R., Loayza, N., & Serven, L. (2005). World market integration through the lens

of foreign direct investors. Journal of International Economics, 66, 267–295.

Bekaert, G., & Harvey, C. R. (1995). Time-varying world market integration. The Journal of

Finance, 50, 403–444.

Bekaert, G., & Harvey, C. R. (2002). Research in emerging markets finance: Looking to the

future. Emerging Markets Review, 3, 429–448.

Bekaert, G., & Harvey, C. R. (2003). Emerging markets finance. Journal of Empirical Finance,

10, 3–55.

Black, F. (1988). An equilibrium model of the crash. NBER Macroeconomics Annual 1988.

Black, F. (1989). Mean reversion, consumption smoothing. NBER Working chapter 2946.

Bollen, N. P. B., Gray, S. F., & Whaley, R. E. (2000). Regime switching in foreign exchange

rates: Evidence from currency option prices. Journal of Econometrics, 94, 239–276.

Bollerslev, T., Engle, R., & Nelson, D. (1994). ARCH models. In: R. F. Engle & D. McFadden

(Eds), Handbook of econometrics (Vol. 4). North-Holland: Amsterdam.

Booth, G. G., & Koutmos, G. (1998). Interaction of volatility, autocorrelation in foreign stock

returns. Applied Economics Letters, 5, 715–717.


Chan, K. (1993). Imperfect information and cross-autocorrelation among stock prices. Journal

of Finance, 48, 1211–1230.

Choe, H. C., Kho, B.-C., & Stulz, R. M. (1999). Do foreign investors destabilize stock markets:

The Korean experience in 1997. Journal of Financial Economics, 54, 227–264.

Dahlquist, M., & Robertsson, G. (2004). A note on foreigners’ trading and price effects across

firms. Journal of Banking and Finance, 28, 615–632.

Dornbusch, R., & Park, Y. (1995). Financial integration in a second-best world: Are we still

sure about our classical prejudices. In: R. Dornbusch & Y. Park (Eds), Financial

Opening: Policy Lessons for Korea. Seoul, Korea: Korea Institute of Finance.

Farmer, J. D. (2000). Market force, ecology, evolution. Working chapter available from http://

www.santafe.edu/Bjdf/

Froot, K. A., O’Connell, P. G. J., & Seasholes, M. S. (2001). The portfolio flows of

international investors. Journal of Financial Economics, 59, 151–193.

Froot, K. A., & Ramadorai, T. (2001). The information content of international portfolio flows.

Working chapter W8472, NBER.

Grinblatt, M., & Keloharju, M. (2000). The investment behavior, performance of various

investor types: A study of Finland’s unique data set. Journal of Financial Economics, 55,

43–67.

Hamilton, J. D. (1989). A new approach of the economic analysis of nonstationary time series,

the business cycle. Econometrica, 57, 357–384.

Hamilton, J. D. (1990). Analysis of time series subject to changes in regime. Journal of

Econometrics, 45, 39–70.

Harvey, C. R. (1995). Predictable risk, returns in emerging markets. Review of Financial Studies,

8, 773–816.

Kim, C.-J., Nelson, C. R., & Startz, R. (1998). Testing for mean reversion in heteroscedastic

data based on Gibbs sampling augmented randomization. Journal of Empirical Finance,

5, 131–154.

Kim, W., & Wei, S. (2002). Foreign portfolio investors before and during a crisis. Journal of

International Economics, 56, 77–96.

Knif, J., Pynnonen, S., & Luoma, M. (1996). Testing for common autocorrelation features of

two Scandinavian stock markets. International Review of Financial Analysis, 5, 55–64.

Koutmos, G. (1997). Feedback trading, the autocorrelation pattern of stock returns: Further

empirical evidence. Journal of International Money and Finance, 16, 625–636.

McKenzie, M. D., & Faff, R. W. (2003). The determinants of conditional autocorrelation in

stock returns. The Journal of Financial Research, 26, 259–274.

Radelet, S., & Sachs, J. (1998). The East Asian financial crisis: Diagnosis, remedies, prospects.

Brookings Paper, 1, 1–74.

Safvenblad, P. (2000). Trading volume, autocorrelation: Empirical evidence from the

Stockholm stock exchange. Journal of Banking, Finance, 24, 1275–1287.

Seasholes, M. (2004). Re-examining information asymmetries in emerging stock markets.

Working Paper, Berkeley, CA.

Sentana, E., & Wadhwani, S. (1992). Feedback traders, stock return autocorrelations: Evidence

from a century of daily data. Economic Journal, 102, 415–425.

Stiglitz, J. (2000). Capital market liberalization, economic growth, and instability. World

Development, 28, 1075–1086.

Turner, C. M., Starz, R., & Nelson, C. R. (1989). A Markov model of heteroscedasticity risk,

learning in the stock market. Journal of Financial Economics, 25, 3–22.


CHAPTER 5

THE IMPACT OF THE OPENING UP

OF THE B-SHARE MARKETS ON

THE INTEGRATION OF CHINESE

STOCK MARKETS

Langnan Chen, Steven Li and Weibin Lin

ABSTRACT

The opening up of B-share markets to domestic investors in 2001 is a

landmark event in the development of the Chinese stock markets. This

chapter aims to assess the possible changes in the market mechanism

associated with this important event. A VECM-DCC-MVGARCH model

is employed to investigate the market integration process in Chinese stock

markets around the opening up of the B-share market to domestic

investors. Our empirical results reveal that the Chinese stock markets

were segmented before the opening up whereas they were integrated to

some extent in the long-run after the opening up of B-share markets.

Moreover, it is also found that A-share markets played a dominant role on

the information flows between A-share and B-share markets; the short-

run information flows between A-share and B-share markets were more

rapid after the opening up of B-share markets.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00005-2

95

dx.doi.org/10.1016/S1569-3767(07)00005-2.3d

1. INTRODUCTION

Chinese stock markets have grown rapidly since their inception in the early1990s. There are two stock exchanges in China: The Shanghai StockExchange (SHSE) and the Shenzhen Stock Exchange (SZSE). Currently,there are two classes of shares listed and traded on both SHSE and SZSE:A- and B-shares.1 A-shares are restricted to domestic investors. B-shareswere only available to foreign investors until February 2001. Chinesecitizens are now allowed to trade B-shares if they have the foreign currencyrequired, i.e. US dollars for B-shares on SHSE and Hong Kong dollars forB-shares on SZSE. The opening up of the B-share markets to domesticinvestors in 2001 has been widely regarded as a landmark event to theintegration of Chinese stock markets.

Assessing the impact of the opening up of B-share market is veryimportant for a few reasons. First, it is important to understand the changein the association between A- and B-share markets after this landmarkevent. This may have a lot of implications to current investors. For example,how important is it for investors to diversify their investment in bothB-share markets and the A-share markets after the event? Second, it isimportant to know if this event is a positive move in the direction formerging the A-share and B-share markets after foreign exchange controlis lifted up in the future. Hence this may have important implications forthe policy makers in the Chinese stock markets. This chapter aims toprovide some empirical evidence on the impact of this important event onthe integration of Chinese stock markets.

The recent research literature on market integration can be separated intotwo groups (Kasa, 1992; Francis & Leachman, 1998). On the one hand,there are many chapters investigating the independence of returns or returnvolatility across national equity markets, see e.g. Hamao, Masulis, and Ng(1990), King, Sentana, and Wadhwani (1994) and Bekaert and Harvey(1995). The evidence from these studies indicates that national equitymarkets exhibit a variety of short-term linkages and interactions and arebecoming reasonably well integrated. On the other hand, there are manychapters investigating long-term co-movements and relationships amongnational equity markets. These studies including Kasa (1992), Francis andLeachman (1998), and Bessler and Yang (2003) etc., employ the procedureof Johansen (1991) for cointegration testing to assess the long-runrelationship(s) among international stock markets. The results of this groupof research indicate that national equity markets possess common long-runequilibrium path(s) or equivalently, common stochastic trends. These

LANGNAN CHEN ET AL.96

findings are therefore consistent with linkages between national equitymarkets, albeit over a longer time horizon.

The approach used to investigate the relationship(s) among internationalstock marketed can be applied to address the A-share and B-share marketsegmentation/integration and the information asymmetry patterns inChinese stock markets. In this chapter, we shall consider the four Chinesestock markets: A-share market in SHSE (SHA), B-share market in SHSE(SHB), A-share market in SZSE (SZA) and B-share market in SZSE (SZB).

There are a few studies in the literature investigating the relationshipsbetween A-share and B-share markets. Bailey (1994) first documented thebig price discounts of B-shares relative to comparable A-shares, which couldbe evidence of the segmentation between A-share and B-share markets.Fung, Lee and Leung (2000) provided supportive evidence for the argumentof segmented A-share and B-share markets. Yang (2003) employed the VARand ECM methods to analyse the relationship among A-, B- and H-sharemarkets. He found that the share markets are not linked in the long-run andforeign investors in the B-share market in SHSE is better informed thanthe investors in the two A-share markets and foreign investors in B-sharemarket in SZSE and Hong Kong market over time. More recently, Chanand Kwok (2005) provided evidence that the premium for A-shares isdetermined by the limited alternative investment opportunities available toretail investors.

It appears that there is a lack of studies in the literature focusingspecifically on the event of the opening up of the B-share market. Thischapter aims to fill the gap in the literature. We use the share market priceindexes to investigate the integration/segmentation of A- and B-sharemarkets. Moreover, we consider both the long-run and short-run aspects ofthe share market integration before and after the opening up of the B-sharemarkets.

The remainder of the chapter is organized as follows. Section 2 provides adescription of the data. The research methodology is discussed in Section 3.Section 4 presents the empirical results. Finally, we conclude in Section 5.

2. DATA AND DESCRIPTIVE STATISTICS

The B-share markets were opened up to domestic investors in February2001. This chapter is concerned with the stock market integration aroundthis important event. To this end, we need the data in two periods: thethree-year period before this event (January 1998 to January 2001), and the

The Impact of B-Share Markets 97

three-year period after the event (March 2001 to March 2004). We use fourChinese stock market price indexes: Shanghai A-Share (SHA), ShanghaiB-Share (SHB), Shenzheng A-Share (SZA) and Shenzheng B-Share (SZB).In addition, we also need Hang Seng Composite Index (HSI) of Hong Kongand Standard & Poor’s 500 Composite Index (S&P500) of the US forrobustness tests. The daily closing prices of these six indexes are used for thepurpose of this study.

Fig. 1 presents the paths of the four Chinese stock price indexes in thewhole sample period, i.e. from 1998 to 2004. It is clear that the Chinesestock markets have an upward trend from 1998 to 2001, but a downwardtrend from 2001 to 2004.

Table 1 reports the return correlations among the Chinese stock priceindexes. The results show that the two A-share markets as well B-sharemarkets are highly correlated before the opening up event with correlationcoefficients above 0.94, while the correlation between A-share and B-sharemarkets are relatively lower. However, after the opening up of B-share

Fig. 1. Time Series of Chinese Stock Price Indexes from 1998 to 2004.


market, the correlation coefficient between SHB and SZB is much smaller(0.567) though correlation between SHA and SZA remains similar.

This chapter focuses on the integration process of the four Chinese stockmarket indexes, and we are not concerned with the H-share market as wellas the N-share market. In addition, the HSI and S&P 500 indexes are usedfor a robustness tests.

Table 2 reports the results on the augmented Dickey–Fuller (Dickey &Fuller, 1981) and Phillips–Perron (Phillips & Perron, 1988) unit root tests

Table 1. The Correlations among the Chinese Stock Market Indexes.

SHA SHB SZA SZB

Panel I: Before the opening up of B-share markets

SHA 1.0000

SHB 0.7997 1.0000

SZA 0.9962 0.8085 1.0000

SZB 0.8602 0.9436 0.8633 1.0000

Panel II: After the opening up of B-share markets

SHA 1.0000

SHB 0.8087 1.0000

SZA 0.9669 0.8725 1.0000

SZB 0.7035 0.5674 0.5697 1.0000

Table 2. The Unit Root Test for All the Indexes.

Before After

Without Trend With Trend Without Trend With Trend

ADF PP ADF PP ADF PP ADF PP

SHA �0.75 �0.80 �2.01 �2.11 �1.50 �1.50 �2.10 �2.10

SHB �0.26 �0.29 �1.70 �1.69 �1.35 �1.58 �5.50 �5.50

SZA �0.70 �0.75 �1.96 �2.04 �1.23 �1.24 �2.22 �2.28

SZB �0.88 �1.26 �2.22 �2.50 �3.16 �3.19 �3.50 �3.51

HSI �1.14 �1.13 �2.42 �2.45 �2.06 �2.13 �2.66 �2.67

S&P500 �1.91 �1.86 �2.96 �2.70 �1.92 �1.88 �1.46 �1.37

This table reports the results on the augmented Dickey–Fuller (Dickey & Fuller, 1981) and

Philips–Perron (Philips & Perron, 1988) unit root tests. The numbers of lags are selected

according to the Akaike Information Criterion. The critical values of the augmented Dickey–

Fuller unit root tests without trend and with trend are �2.86 and �3.41 at the 5% level,

respectively. The critical values of the Philips–Perron unit root tests without trend and with

trend are �14.1 and �21.7 at the 5% level, respectively. Note that ADF stands for augmented

Dickey–Fuller test, and PP stands for Philips–Perron test.


for all the six indexes considered in this chapter. Similar to the findingsby Long, Payne and Feng (1999) and Yang (2003), we find that thereis one unit root in each of the stock price indexes under our study, butthere is no unit root in the first difference of each index at the 5%significance level (not reported here). Thus, all the six stock price indexes areI(1) process.

3. METHODOLOGY

For this study, we need to jointly test the hypotheses of A-share and B-sharemarket integration and the differential demand argument (Sun & Tong,2000) based on cointegration analysis. Whether segmented trading betweenA-share and B-share markets results in informationally segmented marketsin the long-run also needs to be tested. This is equivalent to testing thehypothesis of no cointegration between A-share and B-share markets inboth SHSE and SZSE. Moreover, if we find one or more long-run relationin the system, we should proceed to examine whether the evidence is simplyfor the differential demand argument or against the A-share and B-sharemarket segmentation argument.

Given the purpose of this chapter, it is natural to base our testing methodon the procedure developed by Johansen (1991). It is well known that, likethe standard VAR, the individual coefficients of the VECM are difficult tointerpret, which may cause difficulty in exploring the short-run dynamicstructure. Accordingly, innovation accounting may be good description ofthe dynamic relationship among time series (Sims, 1980; Lutkepohl &Reimens, 1992; Swanson & Granger, 1997). However, the imposition ofcointegration constraints in the non-stationary VAR recently has beendemonstrated to be crucial in yielding consistent impulse responses andforecast error decomposition (Phillips, 1998). Some researchers algebraicallyconvert the estimated VECM to its corresponding level VAR to summarizethe dynamic interactions among markets (Bessler & Yang, 2003). In thischapter, we estimate a generalized form of dynamic conditional correlation(DCC) model proposed by Engle (2002) to accommodate time-varyingvariance, covariance and correlations.

One of the advantages of the DCC-GARCH model is to allow the modelto be estimated more easily even when the covariance matrix is very large.This model is essentially a two-step approach to capture the dynamics. First,we identify and estimate univariate GARCH models. We also allow forasymmetric effect in first stage. Second, we capture the time-varying market


interdependencies through a multivariate GARCH structure for thecorrelation matrix of the standardized return. The standardization usesthe conditional variance retrieved from the univariate GARCH modelestimates. This approach appeals to us as it has the flexibility of univariateGARCH, but not the complexity of conventional multivariateGARCH, which is numerically demanding to estimate for a large set ofmarkets. Thus, the following VECM-DCC-GARCH model is proposed forthis study:

DX t ¼ PX t�1 þPk�1i¼1

GiDX t�i þ mþ rt ðt ¼ 1; . . . ;TÞ

rtjIt�1 � Nð0;DtRtDtÞ

D2t ¼ diagfotg þ diagfktg � rt�1r

0t�1 þ diagfltg �D2

t�1

�t ¼ D�1t rt

S ¼1

T

Pt

ðrtr0tÞ

Qt ¼ S � ðii0 � A� BÞ þ A � �t�1�0t�1 þ B �Qt�1

Rt ¼ diagfQtg�1QtdiagfQtg

�1

8>>>>>>>>>>>>>>>><>>>>>>>>>>>>>>>>:

(1)

Here Xt denotes the vector consisting of the price indexes (SHA, SHB, SZAand SZB) for the four markets under consideration, D is the differenceoperator, � denotes the Hadamard product, P ¼ ab0 is a coefficient matrix,m and Gi are matrixes of short-run dynamic coefficients, and rt is a vector ofinnovations. Given the existence of cointegration, the data generatingprocess of Xt can be appropriately modelled with vector error correctionmodel (VECM) with k�1 lags.

The parameter estimates on the VECM can provide information on thelong-run and short-run structure. The long-run structure can be identifiedthrough testing hypotheses on b, while the short-run structure can beidentified through testing hypotheses on a, Gi and the variance–covariancematrix based on the innovation vector rt (Johansen & Juselius, 1994;Juselius, 1995; Johansen, 1995).

We assume the error is conditionally normal with mean zero andcovariance matrix:

rtjIt�1 � Nð0;DtRtDtÞ (2)

where Dt is an n� n diagonal matrix with time-varying standard deviation,i.e.

ffiffiffiffiffiffihi;t

p, from univariate GARCH models on the diagonal and Rt is the


time-varying correlation matrix:

Dt ¼

ffiffiffiffiffiffiffih1;t

p0 0

0ffiffiffiffiffiffiffih2;t

p 0

..

. ... . .

. ...

0 0 ffiffiffiffiffiffiffihn;t

p

2666664

3777775; Rt ¼

r1;1;t r2;1;t r1;n;t

r2;1;t r2;2;t r2;n;t

..

. . .. ..

.

rn;1;t rn;2;t rn;n;t

2666664

3777775Without the assumption of normality, the estimator will still have Quasi-Maximum Likelihood (QML) interpretation (Engle, 2002).

The equation D2t ¼ diagfotg þ diagfktg � rt�1r0t�1 þ diagfltg �D2

t�1 assu-mes that the each of the residual series follows a univariate GARCHprocess. We can furthermore investigate the asymmetric GARCH process(i.e. GJR-GARCH, Glosten, Jagannathan, & Runkle, 1993),

D2t ¼ diagfotg þ diagfktg � rt�1r

0t�1 þ diagfltg �D2

t�1

þ diagfgtg � I½rt�1o0�rt�1r0t�1

(3)

where I ½rt�1o0� denotes indicator function. If gt is equal to zero, the modelis reduced to a standard GARCH(1,1). We test this by utilizing theappropriate model.

After estimating GARCH model, we standardize the residuals as:

�t ¼ D�1t rt (4)

where Qt is defined as the diagonal DCC model. As Qt does not generallyhave ones on the diagonal, we scale it to get a proper correlation matrix Rt.According to the definition, Qt is positive definite. Thus, via scaling, we canget a proper correlation matrix with ones on the diagonal and off-diagonalelements smaller or equal to one in absolute value.

Bollerslev (1990) introduces a multivariate GARCH model with constantcorrelations. He assumes that the conditional correlation matrix of theresiduals is constant and the conditional variance assumes the form:

Ht ¼ diagffiffiffiffiffiffiffiffiffih1;1;t

p; . . . ;

ffiffiffiffiffiffiffiffiffihn;n;t

p� �Rdiag

ffiffiffiffiffiffiffiffiffih1;1;t

p; . . . ;

ffiffiffiffiffiffiffiffiffihn;n;t

p� �(5)

where R is the constant correlation matrix.


In general

Rt ¼ R ¼

1 r12 r1n

r12 1 ...

..

. . ..

rn�1n

r1n rn�1n 1

26666664

37777775 (6)

When rij ¼ 0, DCC-GARCH becomes diagonal vech-GARCH.Testing if the correlation is constant has proven to be difficult, as testing

the dynamic correlation with the data that have time-varying volatilitiescan lead to wrong conclusions. Bera and Kim (2002) and Tse (2000) providetwo tests on the null of constant correlation against an alternative. Oneshortcoming of both tests is that they do not work well for higherdimensions. Engle and Sheppard (2001) construct a test under their DCCspecification. Given the equations of the DCC they propose the test:

H0 : Rt ¼ R

against

H1 : vechðRtÞ ¼ vechðRÞ þ b1vechðRt�1Þ

þ b2vechðRt�2Þ þ þ bsvechðRt�sÞ

The idea of this test is to use the standardized residuals from theestimation of the first stage (�t ¼ D�1t rt). These residuals have to bestandardized again by the symmetric square root decomposition of theconstant correlation R,

vt ¼ �0tR�1=2 (7)

Let Y t ¼ vechu½vtv0t � Ik�, where vechu is the vech operator which only

selects elements above the diagonal and Ik is the covariance matrix ofresiduals vt. Under the null of constant correlation, the residuals vt should bei.i.d., and the constant and the lagged parameters in the vector autoregres-sion Y t ¼ aþ b1Y t�1 þ þ bsY t�s þ Zt should be zero. The test statistic isthus given by bdX 0Xbd 0bd2 � w2ðsþ 1Þ (8)


4. EMPIRICAL RESULTS

In this section, we present the empirical results which consist of two parts:the long-run and short-run structure of the integration of the Chinese stockmarkets.

4.1. Long-Run Structure of the Integration

The unit root tests for the four Chinese stock price indexes reveal that thefour indexes are all I(1). We use the LR and AIC criteria to select the orderof lag for the level VAR model on each of the four stock indexes (SHA,SHB, SZA and SZB). The order of lag for the level VAR model should be 2before, and 5 after the opening up of the B-share markets, respectively.Thus, the order of lag for the first-order difference VAR and VECM shouldbe 1 before and 4 after the opening up of the B-share market, respectively.

Table 3 reports the Johansen (1991) trace test results. We fail to reject thatthere is a zero cointegrating vector either with a constant included in thecointegration space or with a linear trend at the 5% significance level for theperiod before the opening up of the B-share markets to domestic investors.However, for the period after the opening up of B-share markets, we findthat two cointegrating vectors exist either with a constant included in thecointegrating space or with a linear trend at the 5% significance level. Thisimplies that the Chinese stock markets have become partially integratedafter the opening up of the B-share market to domestic investors, which is incontrast to the completely segmented status before the opening up.

Table 3. Johansen Trace Test for Chinese Stock Price Indexes.

H0: Before After

Linear Trend Without Linear Trend Linear Trend Without Linear Trend

T C (5%) T C (5%) T C (5%) T C (5%)

r=0 40.83 47.86 42.93 54.07 101.90� 47.86 107.38� 54.08

rr1 19.26 29.80 20.47 35.19 41.91� 29.79 45.52� 35.19

rr2 9.01 15.49 10.16 20.26 9.49 15.49 12.99 20.26

rr3 0.70 3.84 1.33 9.16 0.55 3.84 4.01 9.16

This table reports the trace test statistics (T) and the critical values (C) from Osterwald-Lenum

(1992). We determine the lags in the underlying VARs by considering the minimization of the

Akaike Information Criterion and the absence of significant serial corrections in the residuals.�Denotes significant at 5%.


According to Table 4, the null of multinormality is rejected at 1% level forboth before and after the opening up of the B-share markets. Furthermore,the LM tests reveal that there is no first-order and second-orderautocorrelation before the opening up of B-share markets, whereas thereis significant first-order and second-order autocorrelation after the openingup of B-share markets. It should be noted that ARCH effects and deviationfrom normality assumption do not appear to seriously affect the inferenceon cointegration (Gonzalo, 1994; Lee & Tse, 1996).

We then explore the possibility that one of the four series is not in thecointegrating space (i.e. in the cointegrating vector). It is possible that thevector is a linear combination of a subset of the four series. The resultsare reported in Table 5 below.

The null hypothesis is that one series is not in the cointegrating space.The test statistic is distributed chi-squared with two degrees of freedom.

Table 4. Residual Tests.

Test H0: Before After

Stat. p-Value Stat. p-Value

Normality Multivariate normal 35.26 0.000 250.43 0.0000

LM test No autocorrelation (lag 1) 14.16 0.59 34.36 0.0049

No autocorrelation (lag 2) 14.74 0.54 35.44 0.0035

This table reports the results of the multivariate normality test and the Lagrangian Multiplier

(LM) tests (as described in Hansen & Juselius, 1995) for the residuals before and after the

opening up of the B-share market.

Table 5. Tests of Exclusion of Each Stock Market Series from theCointegration Space (Given Two Cointegrating Vectors).

Series Chi-Squared p-Value Decision

SHA 20.91 0.000 R

SHB 40.99 0.000 R

SZA 28.26 0.000 R

SZB 23.62 0.000 R

Tests are on the null hypothesis that the particular series listed in the first column is not in the

cointegrating space. The heading ‘‘Decision’’ relates to the decision to reject (R) or fail to reject

(F) the null hypothesis at the 1% level of significance. Under the null hypothesis, the test

statistic is distributed chi-squared with two degrees of freedom.


We clearly reject the null for all series.2 This implies that all the four indexescan be in the cointegrating space, hence supports the hypothesis that theChinese stock markets are partially integrated after the opening up of theB-share markets to domestic investors.

Table 6 reports the results on the possibility that some markets do notrespond to perturbations in the cointegrating vector. Here we are interestedin the weak exogeneity of each series, relative to the long-run equilibrium.The null hypothesis for each market is that it does not make adjustmenttoward the estimated long-run relation. The test statistic is distributed chi-squared with two degrees of freedom. Our results show that the nullhypothesis cannot be rejected for SHA and SZA, while it is rejected for SHBand SHA at 5% level of significance.3 This implies that A-share markets donot make adjustment toward the estimated long-run relation, while B-sharemarkets do. A-share markets demonstrate weak exogeneity while B-sharemarkets do not. Thus, A-share markets lead the B-share markets. Ourconclusion supports the view that domestic investors have more informationon Chinese stocks than foreign investors as in Chakravarty, Sarkar, and Wu(1998), Su and Fleisher (1999). The information advantage of domesticinvestors may be due to language barrier for foreign investors, difference inaccounting standards, the limited availability of information on listed firmsfor foreign investors. However, our conclusion is contrary to Yang (2003)who claims that foreign investors have information advantage, and Chen,Lee, and Rui (2001) who claim that there is no information asymmetry.

Table 7 reports the Johansen (1991) trace test results for the six indexesincluding the Hong Kong market (HSI) and the US market (S&P500). Wefail to reject that there is a zero cointegrating vector either with a constantincluded in the cointegrating space or with a linear trend at the 5%

Table 6. Tests of Weak Exogeneity (Given Two Cointegrating Vectors).

Series Chi-Squared p-Value Decision

SHA 1.44 0.487 F

SHB 33.49 0.000 R

SZA 0.61 0.736 F

SZB 33.68 0.000 R

Tests are on the null hypothesis that a particular series listed in the first column is weakly

exogenous with respect to perturbations in the cointegrating vector. The heading ‘‘Decision’’

relates to the decision to reject (R) or fail to reject (F) the null hypothesis at a 1% level of

significance. Under the null hypothesis, the test statistic is distributed chi-squared with one

degree of freedom.


significance level for the period before the opening up of the B-sharemarkets to domestic investors. However, for the period after the opening upof B-share markets, we find that two cointegrating vectors exist either with aconstant included in the cointegrating space or with a linear trend at the 5%significance level. These conclusions are identical to Johansen trace testwithout the two international indexes (Table 5). Thus, the results in Table 7further enhance the hypothesis that the Chinese stock markets have becomemore integrated after the opening up of B-share markets. It should benoticed that including international markets does not increase the numberof cointegrating vectors. This implies that the Chinese stock markets are stillsegmented from the international markets after the opening up of B-sharemarkets. This is consistent with our intuition and the literature (Chui &Kwok, 1998).

In sum, after the opening up of the B-share market, the Chinese stockmarkets are no longer segmented; however, they are still segmented from theinternational markets.

4.2. The Short-Run Structure of the Integration

Based on the long-run equilibrium tests, we further estimate the VECM-DCC-MVGARCH model. Before the opening up of B-share markets, there

Table 7. Johansen Trace Test for the Six Stock Price Indexes.

H0: Before After

Linear Trend Without Linear Trend Linear Trend Without Linear Trend

T C (5%) T C (5%) T C (5%) T C (5%)

r=0 82.91 95.75 86.01 103.85 161.89� 95.75 167.90� 103.85

rr1 52.55 69.82 54.77 76.97 77.65� 69.81 82.23� 76.97

rr2 31.04 47.86 32.87 54.08 37.62 47.85 41.55 54.08

rr3 19.78 29.80 21.07 35.19 21.36 29.79 25.29 35.19

rr4 10.00 15.49 11.28 20.26 6.44 15.80 9.53 20.26

rr5 2.51 3.84 3.78 9.16 1.12 3.84 3.77 9.16

This table reports the trace test statistics (T) and the critical values (C) from Osterwald-Lenum

(1992). We determine the lags in the underlying VARs by the consideration of the minimization

of the Akaike Information Criterion and by the absence of significant serial corrections in the

residuals.�Denotes significant at 5%.


is no cointegrating vector in the level VAR, thus we use the VAR modelfor the first difference of the level series instead of the VECM model. Afterthe opening up of B-share markets, there are cointegrating vectors and thuswe can use the VECM to filter the series and make observations from theresiduals.

According to Section 3, the estimates of the DCC-MVGARCH can besplit into two steps. The first step is to identify and estimate the GARCHeffect of the residuals for each series. The second step is estimate theconditional correlation coefficients based on the standardized residualseries.

Table 8 reports that GJR-GARCH effect for the four stock market priceindexes.4 We can draw some conclusions by observing the estimates ofthe coefficient l and g. Table 8 reveals that the Chinese stock marketsoverall have strong GARCH effect and leverage effect. Before the opening

Table 8. The Estimates for the GARCH Model.

Before After

Coefficient Z-Statistic p-Value Coefficient Z-Statistic p-Value

SHA C 18.32 4.06 0.0000 22.52 3.21 0.0013

RESID(�1)2 0.13 6.18 0.0000 0.08 5.86 0.0000

Leverage – – – 0.09 4.12 0.0000

GARCH(�1) 0.85 42.61 0.0000 0.84 32.43 0.0000

SHB C 2.11 4.20 0.0000 7.15 6.38 0.0000

RESID(�1)2 0.17 7.89 0.0000 0.10 8.42 0.0000

Leverage �0.12 �5.54 0.0000 0.05 2.34 0.0192

GARCH(�1) 0.87 61.99 0.0000 0.87 77.09 0.0000

SZA C 6.51 4.28 0.0000 1.63 3.33 0.0009

RESID(�1)2 0.24 6.79 0.0000 0.09 6.64 0.0000

Leverage – – – 0.10 4.45 0.0000

GARCH(�1) 0.66 13.21 0.0000 0.83 38.01 0.0000

SZB C 0.41 5.96 0.0000 0.75 6.39 0.0000

RESID(�1)2 0.25 7.74 0.0000 0.17 8.78 0.0000

Leverage �0.15 �3.84 0.0001 �0.07 �2.93 0.0034

GARCH(�1) 0.77 27.97 0.0000 0.85 60.47 0.0000

This table reports the estimates for the GJR-GARCH model (Glosten et al., 1993):

D2t ¼ diagfotg þ diagfktg � rt�1r0t�1 þ diagfltg �D2

t�1 þ diagfgtg � I ½rt�1o0�rt�1r0t�1. For each index

series, we report coefficient estimates under three headings: under C we report the con-

stant term; under RESID(�1)2, GARCH (�1) and leverage we report the estimate for

diagfktg;diagfltg; diagfgtg, respectively.


up of B-share markets, B-share markets in both SHSE and SZSE havesignificant leverage effect, and the estimates of g are all negative. Thisimplies that the two B-share markets are very sensitive to good news onprofits. The two A-share markets are more resistant to bad news. After theopening up of the B-share markets, all estimates of g are positive except theone for the SZB. This implies that most markets react more to bad newsthan good news.

Table 9 reports the estimation for the dynamic conditional correlation(DCC) model. We use the Likelihood Ratio (Engle & Sheppard, 2001) totest if the conditional correlation coefficients are constant. If the hypothesisof constant conditional correlation coefficients is rejected, we then furtherestimate the dynamic correlation coefficients (Engle, 2002) where A and B

are the coefficients of �t�1�0t�1 and Qt�1, respectively. The correlationcoefficients are calculated according to Rt ¼ diagfQtg

�1QtdiagfQtg�1. The

tests are carried out for each of the 6 pairs of markets. Our empirical resultsreveal that only two pairs (SHA and SHB, SHB and SZA) have time-varying correlation, while other pairs have constant correlation before theopening up of B-share markets. This implies that Chinese stock markets arelargely segmented as far as volatility is concerned.

After the opening up of B-share markets, four pairs of markets (SHA andSHB, SHA and SZB, SHB and SZA, and SZA and SZB) have demonstratedsignificant time-varying correlation. Furthermore, the averages of dynamicconditional correlation coefficients for the two pairs (SHA and SHB, SHBand SZA) have increased dramatically across the opening up of B-sharemarkets. For example, for SHA and SHB, the average of the dynamicconditional correlation coefficient estimates has increased from 0.472 to0.750, the coefficient B in the DCC model decreased from 0.9 to 0.7, andcoefficient A increased from 0.046 to 0.15 across the opening up of B-sharemarkets. Therefore, the conditional covariance between each pair ofmarkets has become lower and the time-varying coefficients are becomingmore sensitive to market changes.

Fig. 2 shows the dynamic conditional correlation coefficients are mostlybelow 0.5 before the opening up of B-share markets. After the opening up ofB-share markets, dynamic conditional correlation coefficients increaseddramatically, in particular in the year 2001. They reached a peak of 0.97.The estimation for the DCC model reveals that the speed of informationflows between markets has increased; the volatility spillover effect hasbecome more significant after the opening up of B-share markets. Thus, thediversification effect of investing in A- and B-shares has nearly diminished.


Table 9. The Identification and Estimate of the DCC Model.

Before After

Constant

Correlation Test

DCC Coefficients Correlation

Coefficient

Constant Correlation Test DCC Coefficients Correlation

Coefficient

Statistic p-Value A B Statistic p-Value A B

SHA_SHB 5.48� 0.065 0.046 0.936 0.472 11.91�� 0.003 0.178 0.764 0.750

(0.0022) (0.0053) (0.1790) (0.0027) (0.005) (0.1323)

SHA_SZA 3.98 0.137 – – – 1.51 0.469 – – –

SHA_SZB 3.78 0.151 – – – 5.08� 0.079 0.135 0.816 0.735

(0.0055) (0.0168) (0.1275)

SHB_SZA 5.31� 0.070 0.039 0.947 0.459 6.73�� 0.034 0.172 0.775 0.761

(0.0007) (0.0015) (0.1842) (0.0025) (0.0048) (0.1244)

SHB_SZB 0.01 0.996 – – – 1.91 0.385 – – –

SZA_SZB 2.62 0.269 – – – 22.49�� 0.00 0.174 0.755 0.753

(0.0025) (0.0084) (0.1320)

Based on the univariate GARCH test, we use the Likelihood Ratio (Engle & Sheppard, 2001) to test if the conditional correlation coefficients

are constant. If the constant conditional correlation coefficients are rejected, we then further estimate the dynamic correlation coefficients

(Engle, 2002) where A and B are the coefficients of �t�1�0t�1 and Qt�1, respectively. The correlation coefficients are calculated according to

Rt ¼ diagfQtg�1QtdiagfQtg

�1. This table reports the test results for each of the 6 pairs of markets. The numbers in parentheses are the

standard error of the estimates.�Denotes significant at 10%.��Denotes significant at 5%.

LANGNAN

CHEN

ET

AL.

110

5. CONCLUSION

This chapter is concerned with the integration of Chinese stock marketsaround the opening up of the B-share markets to domestic investors. Ourstudy addresses this problem from both long-run and short-run aspects.

First, we find that the Chinese stock markets are largely segmented beforethe opening up of the B-share markets. This is consistent with the findings inthe literature.

Second, we find that the Chinese stock markets have become moreintegrated after the opening up of the B-share markets. This is reflected in

Fig. 2. The Time-Varying Correlation Coefficients. (a) Before the Opening Up of

B-Share Markets. (b) After the Opening Up of B-Share Markets.


the facts that there are more cointegrating vectors, increased number ofmarkets in the cointegrating space, decreased number of markets with weakexogeneity and a stronger time-varying correlation. However, it appearsthat Chinese stock markets are still segmented from the international equitymarkets.

Third, we find that A-share markets play a dominant role in theinformation flows, and the speed of information flows has increasedsignificantly after the opening up of B-share markets to domestic investors.

In sum, our results reveal that the two stock markets have become moreintegrated and the speed of information flows across the markets hasincreased after the opening up of B-share markets to domestic investors.

Fig. 2. Continued


Thus, the opening up event has been a successful step for increasing theintegration of the Chinese stock markets.

Our findings have some important implications to the Chinese stockmarkets. For example, after the opening up of the B-share markets todomestic investors, the four markets (SHA, SHB, SZA and SZB) havebecome more highly correlated. Thus, investors are more likely to beindifferent to investing in either A-share markets or B-share markets thoughdomestic investors do have the access to B-share markets. Furthermore, ourresults indicate that the opening up of B-share market is an important stepin the right direction for the merging of A-share markets and B-share in thefuture when the foreign exchange control is lifted in China.

Fig. 2. Continued


NOTES

1. Chinese companies are also allowed to list shares overseas. Most Chineseoffshore stocks are traded on the Hong Kong Stock exchange (H-shares), althoughthere are now some Chinese stock traded on other stock exchanges includingNew York, London and Singapore stock exchanges. Clearly, B-shares are listedand traded on the home market whereas H-shares are not.2. Here we consider the case of given two cointegrating vectors. In the case of

given one cointegrating vector, SZB can be excluded from the cointegrating space,but this does not contradict the partial integration of the Chinese stock markets.3. The weak exogeneity test for the two A-share markets gives a statistic value of

9.14, the p-value is 0.058 (4-degrees of freedom). Hence, the weak exogeneity cannotbe rejected at 5% level of significance. This means that the A-share markets haveweak exogeneity.4. Since the leverage effect has not been found through our study and previous

studies for A-share indexes in the period before the opening up of B-share market, itis not necessary to estimate the leverage coefficients for SHA and SZA for thatperiod.

ACKNOWLEDGMENT

The authors are grateful to the anonymous referee, whose comments havehelped to improve the chapter significantly. Any remaining errors are, ofcourse, our own. Li is the corresponding author. This research is supportedby China Natural Science Foundation under Grant No. 70473106 and70673116, the 985 Project, the Research Base Project from MOE underGrant No. 05JJD790075 and the China SSF under Grant No. 07BJY167.

REFERENCES

Bailey, W. (1994). Risk and return on China’s new stock markets: Some preliminary evidence.

Pacific-Basin Finance Journal, 2, 243–260.

Bekaert, G., & Harvey, C. R. (1995). Time-varying world market integration. Journal of

Finance, 50, 403–444.

Bera, A. K., & Kim, S. (2002). Testing constancy of correlation and other specifications of the

BGARCH model with an application to international equity returns. Journal of

Empirical Finance, 9, 171–195.

Bessler, D. A., & Yang, J. (2003). The structure of interdependence in international stock

markets. Journal of International Money and Finance, 22, 261–287.

Bollerslev, T. (1990). Modeling the coherence in short-run nominal exchange rates:

A multivariate generalized ARCH approach. Review of Economics and Statistics, 72,

498–505.


Chakravarty, S., Sarkar, A., & Wu, L. (1998). Information asymmetry, market segmentation

and the pricing of cross-listed shares: Theory and evidence from Chinese A and B shares.

Journal of International Financial Markets, Institutions and Money, 8, 325–355.

Chan, K., & Kwok, J. K. H. (2005). Market segmentation and share price premium: Evidence

from Chinese stock markets. Journal of Emerging Market Finance, 4, 43–61.

Chen, G. M., Lee, B., & Rui, Q. (2001). Foreign ownership restrictions and market

segmentation in China’s stock markets. Journal of Financial Research, 24, 133–155.

Chui, A., & Kwok, C. (1998). Cross-autocorrelation between A shares and B shares in the

Chinese stock market. Journal of Financial Research, 21, 333–353.

Dickey, D. A., & Fuller, W. A. (1981). The likelihood ratio statistics for autoregressive time

series with a unit root. Econometrica, 49, 1057–1072.

Engle, R., & Sheppard, K. (2001). Theoretical and empirical properties of dynamic conditional

correlation multivariate GARCH. NBER Working Paper 8554, National Bureau of

Economic Research.

Engle, R. F. (2002). Dynamic conditional correlation: A simple class of multivariate generalized

autoregressive conditional heteroskedasticity models. Journal of Business and Economic


Francis, B., & Leachman, L. (1998). Superexogeneity and the dynamic linkages among

international equity markets. Journal of International Money and Finance, 17, 475–492.

Fung, H., Lee, W., & Leung, W. (2000). Segmentation of the A- and B-share Chinese equity

markets. Journal of Financial Research, 23, 179–195.

Glosten, L., Jagannathan, R., & Runkle, D. (1993). Relationship between the expected value

and the volatility of the nominal excess return on stocks. Journal of Finance, 48,

1779–1801.

Gonzalo, J. (1994). Five alternative methods of estimating long-run equilibrium relationships.

Journal of Econometrics, 60, 203–233.

Hamao, Y., Masulis, R., & Ng, V. (1990). Correlation price changes and volatility across

international stock markets. Review of Financial Studies, 3, 281–307.

Johansen, S. (1991). Estimation and hypothesis testing of cointegration vectors in Gaussian

vector autoregressive models. Econometrica, 59, 1551–1580.

Johansen, S. (1995). Identifying restrictions of linear equations with applications to

simultaneous equations and cointegration. Journal of Economics, 69, 111–132.

Johansen, S., & Juselius, K. (1994). Identification of the long-run and the short-run structure:

An application to the ISLM model. Journal of Econometrics, 63, 7–36.

Juselius, K. (1995). Do purchasing power parity and uncovered interest rate parity hold in the

long run? An example of likelihood inference in a multivariate time-series model. Journal

of Econometrics, 69, 211–240.

Kasa, K. (1992). Common stochastic trends in international stock markets. Journal of Monetary


King, M., Sentana, E., & Wadhwani, S. (1994). Volatility and links between national stock

markets. Econometrica, 62, 901–933.

Lee, T., & Tse, Y. (1996). Cointegration tests with conditional heteroscedasticity. Journal of


Long, D., Payne, J., & Feng, C. (1999). Information transmission in the Shanghai equity

market. Journal of Financial Research, 22, 29–45.

Lutkepohl, H., & Reimens, H. (1992). Impulse response analysis of cointegrated systems.

Journal of Economic Dynamics and Control, 16, 53–78.


Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regression.

Biometrika, 75, 335–346.

Phillips, P. (1998). Impulse response and forecast error variance asymptotics in nonstationary

VARs. Journal of Econometrics, 83, 21–56.

Sims, C. (1980). Macroeconomics and reality. Econometrica, 48, 1–48.

Su, D., & Fleisher, B. (1999). Why does return volatility differ in Chinese stock markets?


Sun, Q., & Tong, W. (2000). The effect of market segmentation on stock prices: The China

syndrome. Journal of Banking and Finance, 24, 1875–1902.

Swanson, N. R., & Granger, C. W. J. (1997). Impulse response functions based on a causal

approach to residual orthogonalization in vector autoregressions. Journal of the

American Statistical Association, 92, 357–367.

Tse, Y. K. (2000). A test for constant correlations in a multivariate GARCH model. Journal of


Yang, J. (2003). Market segmentation and information asymmetry in Chinese stock markets:

A VAR analysis. The Financial Review, 38, 591–609.


CHAPTER 6

A SINGLE CURRENCY FOR

ASEAN-5: AN EMPIRICAL STUDY

OF ECONOMIC CONVERGENCE

AND SYMMETRY

Zhi Lu Xu, Bert D. Ward and Christopher Gan

ABSTRACT

Ng (2002), and Lim and McAleer (2003) explained that if the national

economies are not converging, or if the responses of national economies to

random shocks are asymmetric, the cost of premature monetary

integration would be high. This chapter investigates the feasibility of

adopting a single currency for ASEAN-5 countries. The research uses the

Kalman Filter procedure to test the economic convergence among

ASEAN-5 countries, relative to Japan and the US. In addition, the

symmetry of underlying structural shocks is also examined by applying a

structural vector autoregression (SVAR) model. The research findings

showed that Singapore, Malaysia, and Thailand (ASEAN-3) appear to

be relatively suitable for forming an Optimum Currency Area. However,

the results did not show significance evidence whether the Japanese Yen or

the US dollar will be a suitable currency for the ASEAN-3 countries to

adopt commonly.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00006-4

117

dx.doi.org/10.1016/S1569-3767(07)00006-4.3d

1. INTRODUCTION

A single currency implies a single central bank replacing all the existingindividual central banks and adopting common monetary and exchange ratepolicies in the region. The debate on the feasibility of a currency union hascentred on the theory of Optimum Currency Areas (OCA), which wasinitially suggested by Mundell (1961). The theory addresses the question:‘What is the appropriate area for one common currency?’ The introductionof the Euro in January 1999, suggests that a similar currency union could beintroduced to other region of the world. One proposed region is theAssociation of Southeast Asian Nations (ASEAN). ASEAN was establishedin August 1967, with five original member countries: Singapore, Malaysia,Indonesia, Thailand, and the Philippines (hereafter referred to as ASEAN-5).Brunei, Vietnam, Laos, Myanmar, and Cambodia became members of theassociation between 1984 and 1999. This study focuses on ASEAN-5countries in the region, which have the most advanced economies in theregion. They have high level of GDP and high volume of the trade, and theyalso attract most of the foreign direct investment in the region (ASEANSecretariat, 2005).

Mundell (1961) argued that a currency union can facilitate internationaltrade, and a single medium of exchange reduces transaction costs inregional trade. The new regime would thereby stimulate capital flowsand investment, increase growth and employment, and improve thebalance of payments performance. Madhur (2002) emphasised that a singlecurrency promotes greater trade among countries; therefore, opennessand the volume of intra-regional trade will be greater under a commoncurrency than under a regime of national currencies with floating exchangerates. Rose (2004) carried out a meta-analysis by applying a ‘gravity model’to examine the effect of a common currency on trade. The author foundstrong evidence that members of currency unions traded over threetimes as much as otherwise similar pairs of countries ceteris paribus.

The benefits of forming an OCA are not achieved without costs. Previousresearch suggests that the costs of adopting a single currency come from theloss of monetary independence. Member countries may not have anoverabundance of monetary policy and exchange rate adjustment options asstabilisation policies in the face of unexpected macroeconomic shocks (SeeBayoumi & Mauro, 1999; Madhur, 2002; Bunyaratavej & Hahn, 2003).Mundell (1961) believed that the costs of adopting a single currency dependon how easily an economic shock in one country is transmitted to othercountries in the same region. Bayoumi and Mauro (1999) supported the idea

ZHI LU XU ET AL.118

that the larger and more dissimilar the underlying shocks that membercountries face, the higher the costs of adopting a common currency.However, if the correlation of the shocks is high, or the shocks that are facedby individual economies are symmetric, the costs can be lowered (Bayoumi &Eichengreen, 1993, 1994).

2. PRECONDITIONS FOR A CURRENCY UNION

At the very heart of the currency union is the issue of economic convergence,which depends on a high degree of convergence of national economies.Economic convergence occurs when poor countries, with low initial incomeand productivity levels, tend to have high rates of economic growth thatwould enable them to catch up, or converge upon, the living standards ofrich countries over the long run (Lim & McAleer, 2003). This effort can beachieved only when all member nations face both external and domesticshocks in a similar pattern (Jayaraman, Ward, & Xu, 2005). Furthermore, ifthe responses of national economies to the macroeconomic shocks wereasymmetric, the cost of premature monetary integration would be high.Therefore, economic convergence of national economies and symmetry ofunderlying macroeconomic shocks are the two preconditions for a currencyunion.

Based on previous research, the exchange rate, GDP growth rate andinflation seemed to be the main three factors affecting economic convergence.Ocampo (2004) asserts that exchange rate stability would promote trade andinvestment, and give rise to peer pressure for macroeconomic coordination,as witnessed in Europe. Kuroda (2004) also argued that an exchange rateregime that is flexible outside but relatively stable inside would beappropriate for the East Asian economies. Per capita GDP is also one ofthe main indicators used to measure real convergence. If a region displays ahigh degree of heterogeneity in the region, Bunyaratavej and Hahn (2003)believed, the region may not be an ideal candidate for forming a currencyunion. As for the inflation rate, Laabas and Limam (2002) establish that thecountries with similar inflation rates would like to coordinate their policies toachieve the requirements of a currency union, because similar inflation ratessignal similarity in structure and in the conduct of economic policies.Therefore, the convergence in the nominal exchange rates, GDP growth ratesand inflation rates will be examined in this study.

An economy is affected by two types of shocks, demand and supplyshocks. Demand shocks come from the changes in aggregate demand, while

A Single Currency for ASEAN-5 119

supply shocks result from the changes in potential GDP (Taylor & Dalziel,2002). Mundell (1961) argued that a high degree of economic convergencerelies heavily on the similarity in response to both domestic demand andsupply shocks. This is supported by a number of researchers (see Yuen,2001; Fidrmuc & Korhonen, 2003; Zhang, Sato, & McAleer, 2003), whoexamined size, correlations of the shocks and the responses to the bothshocks on different region. However, Ng (2002) emphasised that correlationof external shocks should also be considered in the study of ASEANeconomies. This is because ASEAN economies are open and could besusceptible to the external shocks. The author claims that the nominalexchange rate is used to adjust external shocks, so a positive correlation ofexternal shocks could strengthen the case for a currency union in ASEANcountries.

Even though a number of studies on economic convergence and symmetryof the shocks have been conducted on Asian countries, majority of thestudies did not test both conditions simultaneously in the ASEAN region.The purpose of this study is to examine the feasibility of adopting a singlecurrency in ASEAN-5 countries by testing economic convergence andsymmetry of underlying shocks simultaneously, in order to find out whethera single currency can be adopted among these five countries or anysubgroup of the countries for a start. The nominal exchange rate, real GDPgrowth rate and inflation rate are the three economic indicators used to testeconomic convergence in ASEAN-5 countries. In addition, three types ofshocks were included in this study, namely, domestic supply shocks,domestic demand shocks, and external shocks. The following sectionintroduces the methodologies used in this research.

3. MODELLING AND METHODOLOGY

This study seeks to test both economic convergence and symmetry ofunderlying shocks in ASEAN-5 countries; therefore, two models and twotypes of methodologies were used in the empirical analysis.

3.1. Economic Convergence

3.1.1. Methodology

To test for economic convergence in Europe, Hall, Roberson, and Wickens(1992) developed a model consisting of the differentials between any two

ZHI LU XU ET AL.120

countries and the differential between one of the countries and a thirdcountry (or a world index) based on the model provided by Haldane andHall (1991).1 Hall et al. (1992) model is given as follows.

½XDM � XUK�t ¼ at þ bt½XDM � XUS�t þ et (1)

where XDM, XUK, and XUS are the logs of the economic factors. The authorsshowed that if the DM and UK series have converged, then bt is expected toapproach zero and if the UK and US series have converged, then bt isexpected to approach one. Moreover, at and bt are time-varying parameters,which can be estimated by applying the Kalman Filter procedure asproposed initially by Kalman (1960). It allows for stochastic regressioncoefficients right from the start.

The Kalman Filter is an algorithm for sequentially updating a linearprojection for a state-space form. The Kalman Filter when applied to amodel in state-space form provides an algorithm for producing predictionerrors ~ut and prediction error variances ~Ft. The example of a linear statespace representation of the dynamics of the n� 1 vector Yt is given by thefollowing system of equations.

Y t ¼ dt þ atX t þ �t (2a)

at ¼ ct þ at�1Tt þ ot (2b)

where at=(K� 1) vector of state variables; ct=non-stochastic (K� 1)vector; Xt=(N�K) matrix; Tt=fixed (K�K) matrix and et, ot=M- andm-dimensional white noise vectors, respectively (et, ot are assumed to beserially independent).

Eqs. (2a) and (2b) comprise a Kalman Filter model. Eq. (2a) is an ordinaryregression with time-varying parameters, which is called a measurementor observation equation. It describes how well the actually observeddata are generated from the state variables. Eq. (2b) describes the evolutionof the (unobserved) ‘state’ variable, at over time, and is called stateor transition equation. Note that there can be more than one transitionequation.

The Kalman Filter is able to update an estimator as soon as newobservations become available. There are two steps in the process. Firstly,given all information available, the procedure can make an optimalprediction of correct observations. Secondly, when the new observation isincluded in the estimator of the state vector, the estimates can be updated inthe next round. The Kalman Filter has a recursive nature in determining


coefficient at, using updated information on Yt and Xt. Estimation takesplace by means of Maximum Likelihood techniques. Under the assumptionthat et, ot are Gaussian error terms, and the sample log Likelihood is shownas follows.

logLðyÞ ¼ �nT

2log 2p�

1

2

Xt

log j ~F ðyÞj �1

2

X~�1t ðyÞ

�1~�tðyÞ (3)

Therefore, the Kalman Filter itself does not estimate unknown para-meters of the model, but merely provides ~ut and their variances ~Ft, henceconventional maximisation routines can then be used to determine theunknown parameters.

3.1.2. Modelling

Following the Hall et al. (1992) model, this study selects Japan and theUS as reference countries in testing economic convergence amongASEAN-5 countries. The state-space model used in this study is presentedas follows:

Measurement equation : ½XA � XB�t ¼ at þ bt½XA � XC�t þ et

Transition equation :at ¼ at�1 þ v1t

b1 ¼ bt�1 þ v2t

(4)

where XA and XC represent the set of economic indicators for Japan andthe US, respectively, and XB represents the economic indicators of thefive respective ASEAN-5 countries. In Eq. (4) at and bt are time-varyingcoefficients, which can be estimated by applying the procedures discussedabove. The model is then used to test the convergence of each indicatorseparately. For example, to test the economic convergence of inflation, XA

represents the inflation rate in Japan, XC represents the inflation rate of theUS, and the XB series represents the inflation rate of the five respectiveASEAN-5 countries.

Based on Hall et al. (1992) findings, if Japan and the ASEAN-5countries’ series have converged, we would expect bt tend to be zero, while ifthe series of the ASEAN-5 countries and the US have converged, bt isexpected to equal one. Mathematically, this relationship can be expressed asfollows.

When bt=0, Eq. (4) can be written as: [XA�XB]t=at+et, indicating thatXA and XB series have converged. However, when bt=1, Eq. (4) can be

ZHI LU XU ET AL.122

written as follows:

½XA � XB�t ¼ at þ 1½XA � XC�t þ et

[XC�XB]t=at+et, indicating that XC and XB series have converged.In addition to the condition on bt, Hall et al. (1992) state that the time-

varying coefficient at needs to be constant. The reason is that the expression‘at+et’ cannot be stationary under any condition if b=1 or b=0, given thatthe coefficient at is non-stationary. Therefore, the dual requirements forconvergence are bt tends to zero and at tends to a constant.

3.2. Symmetry of Underlying Shocks

3.2.1. Modelling

Sims (1980) initially developed the Vector Autoregressive (VAR) model toidentify supply and demand shocks. In order to allow for theory basedidentification restrictions, Sims (1982, 1986), Bernanke (1986), and Shapiroand Waston (1988) subsequently developed the Structural Vector Auto-regressive (SVAR) model. Blanchard and Quah (1989) presented long-runrestrictions in SVAR, while Bayoumi and Eichengreen (1993) developed amodel with identification restrictions on the long-run parameters to assesswhether the European monetary union satisfies the criteria of an optimumcurrency area. The methodology used in this study is based on the SVARmodel provided by Bayoumi and Eichengreen (1994).

The class of SVAR models that EViews estimates can be written asfollows:

Aet ¼ But (5)

where et is the observed residuals of Vector Autoregression (VAR) model,while ut is the unobserved structural shocks. A and B are k� k matrices andcan be estimated by providing long-run identifying restrictions. Bayoumiand Eichengreen (1993, 1994) argued that a positive demand shock wouldresult in a temporary increase in output and a permanent increase in theprice level. However, a positive supply shock would cause a permanentincrease in the output and a permanent decrease in the price level. As for theexternal shocks, Ng (2002) reasoned that they stem from the movementsin world business cycles and are outside the control of a domestic economyand have a permanent impact on world output domestic output anddomestic price.


The structural VAR model that contains long-run restrictions, using lagoperators Li is specified as follows:

Dy�t

Dyt

Dpt

264375 ¼X1

t¼0

Li

c11 c12 c13

c21 c22 c23

c31 c32 c33

264375 �et

�st

�dt

264375 (6)

In Eq. (6), y�t , yt, pt are the logarithms of world output, domestic output,and domestic prices, respectively. C is a 3� 3 matrix representing theaccumulated long-run response of the variables to the structural shocks, eet,est, edt are independent external, supply and demand shocks, respectively.

3.2.2. Methodology

Following the process provided by Bayoumi and Eichengreen (1993, 1994),the first step for SVAR is estimation of a Standard VAR model, taking theform of:

X t ¼ C1X t�1 þC2X t�2 þ . . .þCpX t�p þ et ¼ CðLÞX t (7)

where Xt is a vector of differences of 3 endogenous variables, the Ci are(3� 3) matrices of reduced form parameters, and et are the residuals havingGaussian process and L is the polynomial lag operator. Before estimatingthe VAR model, unit root tests of the three endogenous variables will beconducted.

According to the class of SVAR models in Eq. (5), the unobservedstructural shocks ut take the following form:

ut ¼ B�1Aet (8)

Since EViews (2005) requires all restrictions to be linear in the elements ofA and B, to specify a long-run restriction, the A matrix must be an identitymatrix. The B matrix can be estimated from the VAR model by imposinglong-run restrictions C. The long-run identifying restrictions are specified interms of the elements of the C matrix in Eq. (6). We assume the domesticsupply shocks have long-run effects on domestic output and prices, while thedemand shocks only have effects on domestic prices. Only external shockshave long-run effects on the world output. Taking into account theserestrictions, the C matrix will be a 3� 3 lower triangular matrix. Since thevalue of the residuals (et) can be generated from the estimated VAR model

ZHI LU XU ET AL.124

(Eq. (7)), the structural shocks ut (i.e. eet, est, edt) can then be identifiedusing Eq. (8).

3.3. Data

The data used in this study is retrieved from International FinancialStatistics (IFS), 2000–2005. Annual data from 1970 to 2004 is used toestimate the models. The investigated countries include the ASEAN-5,namely, Singapore, Indonesia, Malaysia, Thailand, Philippines, and Japanand the US. In examining the external shocks to ASEAN-5 countries, worldoutput will be investigated by using total industrial countries’ production asthe indicator.

When we test convergence in nominal exchange rates, the logarithm ofSDR2 is used. The two variables that will be used in both the economicconvergence test and the symmetry of structural shocks test are real GDPgrowth rates and inflation rates. Real GDP growth rates can be obtained bytaking the natural log of the ratio of the GDP index of the current period tothat of the GDP index of the previous period, where GDP index equals thedollar amount of GDP divided by the corresponding GDP deflator. CPI is agood indicator for inflation rate, which can be obtained by taking thelogarithm of the ratio between the CPI in the current period to the CPI inthe previous period. Note that both real GDP growth rates and inflationrates will be multiplied by 100 in the process of examination.


4.1. Unit Root Tests

The variables for both models were first tested for stationarity usingalternative unit root tests. Three types of unit root tests were used to identifythe stationarity of the series, namely, the Augmented Dickey–Fuller (ADF),Phillips and Perron (1988) approach (PP), and Kwiatkowski, Phillips,Schmidt, and Shin (1992) (KPSS) tests. In the ADF test, the augmenting laglengths were chosen with Hannan–Quinn selection criterion, starting with amaximum length of 4 lags. For the ‘level’ tests the ADF test equationsincluded a constant and trend term, whereas for the ‘1st difference’ tests,only a constant term was included. All hypotheses were tested at 5%significance level unless specified differently. In the KPSS and PP tests, the


equations included a constant term in ‘level’ tests with Bartlett Kernelspectral estimation method and Newey–West bandwidth selection criterion.As for the ‘1st difference’ test, the PP test equation excluded a constantterm.

Three variables are tested as conditions for the Kalman Filter procedure.They are the logarithm of the SDR series, real GDP growth rates andinflation rates for ASEAN-5 countries and Japan and the US. As for theSVAR model, unit root tests for three variables are required. They are worldGDP growth rate as well as the GDP growth and inflation rates of theASEAN-5 countries.

However, the results computed by the three types of unit root tests do notall point to the same inference (detailed results available on request). Forexample, ADF and PP test suggest that logarithm of SDR (LSDR) in theASEAN-5 countries and Japan are all integrated of order one, I(1), andLSDR in US is stationary. However, the KPSS test gave some differentresults for LSDR in Singapore, Indonesia, and the Philippines – they are allI(0), which indicates stationarity of the series. The three unit root testsprovide different results for the logarithm of SDR, real GDP growth ratesand inflation rates, suggesting that these unit root tests are not able tounambiguously classify finite time series variables (see Johnston & Dinardo,1997). Since there is no unequivocal evidence that all three series areintegrated of the same order, the Kalman Filter procedure and the SVARmodel will be applied to test the economic convergence and to identify theshocks, which may limit the robustness of the tests.

4.2. Economic Convergence – Kalman Filter Procedure

As discussed previously, the application a Kalman Filter to the state-space model in Eq. (4) provides estimates of the time-varying parameters,at and bt. If the bt estimates tend to zero, the economies in ASEAN-5countries may have converged with Japan, whereas bt tending to 1 indicatesconvergence with the US. The relevant empirical results for all 5 ASEANcountries may be summarised as follows.3

4.2.1. Estimation of State Variable Coefficients bt

For nominal exchange rates (SDR), the estimates of bt fluctuated during theearly 1970s as a result of the oil shocks in 1973/1974. Except for Indonesia,the fluctuation eased off since the early 1980s, and the bt estimates seem tocentre around one between 1986 and 1997, indicating a relatively stronger

ZHI LU XU ET AL.126

relationship with the US dollar. However, after 1997 onwards, bt has atendency to move away from the numerical value of one and stayed at thevalue between 0 and 1 in 2004, suggesting that the relationship betweenASEAN-5 currencies and the US dollar has lessened. On the whole, thenominal exchange rates in ASEAN-5 countries have converged with eachother since 2000, but no strong evidence emerges showing they haveconverged with either the US or Japan.

Indonesia deviated from the pattern exhibited for the other four ASEANcountries from 1978 to 2000, especially during the years of 1978, 1983, 1986,and 1997. In the mid-1970s, the Bank of Indonesia announced gradualadjustments in the exchange rate in order to follow the industrial world’sabandonment of fixed exchange rate regimes. In 1978, the Bank ofIndonesia was compelled to devalue the Rupiah by 50% to address theeroding profits of exporters. A second major devaluation was undertaken in1983 accompanied by a major financial reform in the Bank of Indonesia.Further, a third devaluation was undertaken in September 1986, mainly inresponse to the decline in foreign exchange earnings through oil exports. OnAugust 14, 1997 the Bank of Indonesia announced that the Rupiah wouldbe allowed to float (Country Studies, 2003–2005). On balance, the Rupiahhad a tendency to move together with the currencies of the other fourASEAN countries from 1997 onwards.

As regards the other nominal variable, namely inflation, there was a muchhigher degree of convergence toward zero among Singapore, Thailand, andMalaysia after the oil shock in 1973. As for the Philippines, except for thefluctuations during the 1980s, the inflation series have been moving togetherwith the other three ASEAN countries. Indonesia, however, converged withother countries smoothly until the outbreak of the 1997 Asian financialcrisis. Results from the Kalman Filter suggest that the inflation series inSingapore, Malaysia, and Thailand may have converged with inflation inJapan. However, there was no evidence that nominal exchange rates in thosethree countries have converged with the Japanese series.

Moreover, it was found that Indonesia and the Philippines are the twocountries that deviated from the other three countries during certainperiods. Firstly, high inflation has been a major problem in Indonesia sincethe mid-1960s. During the period of the oil crisis, the Bank of Indonesiaprinted Rupiah currency in exchange for oil-generated revenue, and annualinflation surged to around 40%. In 1997, inflation increased sharply againafter the Bank of Indonesia allowed the Rupiah to float.

As for the Philippines, the economy experienced considerable difficultiesin the early 1980s mainly because of the declining world market for


Philippine exports, and the difficulties in borrowing from the internationalcapital markets (Country Studies, 2003–2005). In 1983, the country suffereda political and economic crisis that did not improve until 1986. In additionto the trade deficit and the government budget deficit in 1988, the economyonce again began to encounter difficulties, causing the state variablecoefficient bt to fluctuate in the 1980s. However, the fluctuations have easedoff since the 1990s and the series display a tendency to move together withSingapore, Malaysia, and Thailand.

As for real GDP growth rate, there has been considerable divergence ingrowth in ASEAN-5 economies over the period 1970–2004. However,regardless of these fluctuations, the value for the state coefficient bt seems tobe distributed relatively evenly around the value of zero, except for somelarge negative values around the time of the Asian financial crisis during thelate 1990s.

4.2.2. Estimation of State Variable Coefficients at

Turning now to the coefficient estimates for at, one of the dual requirementsfor convergence is that at tends to be constant (Hall et al., 1992). This wasindeed the case for the nominal exchange rate series for all ASEAN-5countries during the period of 1980 to 2004. For the real GDP growth rateseries, there were certain fluctuations in coefficient values at during theperiod examined, especially in the case of Indonesia from 1997 onwards.Since no convergence in real GDP growth rates has been found based on thecoefficient bt, not surprisingly, little evidence is shown that the coefficient at

tend to a constant. Recall the estimates bt for inflation, Philippines, andIndonesia deviated from the other three ASEAN countries during the periodof the 1980s, and 1997–2004, respectively. As for the estimates at,Philippines and Indonesia deviated during the same period, and the otherthree countries displayed a constant tendency.

4.2.3. Summary for Economic Convergence Test

According to the examination of nominal and real variables, havingconsidered the dual requirements for convergence, the following evidencehas been found. Nominal exchange rates in Singapore, Malaysia, Thailand,and the Philippines have converged among themselves, but no strongevidence has been found that nominal exchange rates in these four countrieshave converged with the series in Japan or the US. As for the inflation rates,the series in Singapore, Malaysia, and Thailand have converged not onlyamong themselves, but also have converged with the series in Japan. As forreal GDP growth rate, no convergence has been found in ASEAN-5

ZHI LU XU ET AL.128

countries. Therefore, a group of countries comprising Singapore, Malaysia,and Thailand (ASEAN-3), has been found to have converging nominalexchange rates and inflation rates among themselves. However, no strongevidence suggests whether ASEAN-3 countries should adopt the US dollaror the Japanese Yen.

The next step is to examine the feasibility of a currency union forASEAN-5 countries based on the other precondition, which is the symmetry(or otherwise) of underlying shocks. If different countries experience similarshocks and have similar responses to the shocks, then those countries may

be candidates for a currency union. The results derived from this test can becompared to the findings found by the Kalman Filter procedure, in order toclarify our final results.

4.3. Symmetry of Underlying Shocks4

As discussed previously, there are three shocks to be examined – externalshocks, domestic supply shocks, and domestic demand shocks, whichare represented by world output, domestic output, and domestic price,respectively. The measures for world output, domestic output, and domesticprice are world real GDP, domestic real GDP, and domestic inflation,respectively. Before estimating and examining the correlation of shocks, wewill look at the correlations for real GDP and inflation in ASEAN-5countries.

4.3.1. Correlation of Real GDP Growth and Inflation

The ASEAN-5 economies seem to have displayed a lower degree ofcorrelation in their growth relationships than in their inflation movementsFor real growth, the highest correlation was between Singapore andMalaysia (0.829), but for three pairs of countries (Indonesia–Philippines,Indonesia–Thailand, and Philippines–Thailand) the correlation coefficientswere numerically small and not statistically significant at the 5% level(Table A1). As for inflation, the cross-country correlations tended to begreater, with fully 9 out the 10 combinations being statistically significantlycorrelated5 with each other, except for the Philippines–Indonesia combina-tion (Table A2). This is consistent with the patterns exhibited by statevariable coefficient bt for inflation rates, which suggests that the ASEAN-5countries may have been pursuing fairly similar monetary policies. As forthe growth movements, they have exhibited high inter-country correlation.The correlation of growth seems to provide a more distinct geographic


pattern. Two sub-regions can be distinguished according to relatively highcorrelation, comprising Singapore, Malaysia, and Thailand, and the other,Philippines and Indonesia.

This analysis of the correlations suggests that ASEAN-5 countriesmay consist of a core of three countries, namely, Singapore, Malaysia,and Thailand, exhibiting high correlations for both output growth andinflation.

4.3.2. Correlation of the Shocks

We also analysed the cross-country correlations among the ASEAN-5countries for three types of shocks – external shocks, domestic supplyshocks, and domestic demand shocks.

The objective of examining the correlation of external shocks in theASEAN-5 countries is to find out how closely correlated the external shocksare across the economies. Considering firstly the effects of external shocks,all pairs of ASEAN-5 countries exhibited high positive correlations, rangingfrom 0.7806 for Malaysia–Thailand to 0.4525 for Singapore–Indonesia(Table A3). All 10 correlation coefficients were statistically significant. Thismay be because most of the countries are relatively open and possess export-oriented economies. Now considering supply and demand shocks, we foundthat Singapore, Malaysia, and Thailand had highly correlated (and mostlystatistically significant) supply shocks (Table A4). For the demand shocks,the correlations were even higher (e.g. r=0.7624 for Singapore–Malaysia)(Table A5). In this regard, Ng (2002) claims that when interpreting demandshocks, it is important to note that monetary policy can affect the degree ofcorrelations. If countries follow similar monetary policies, the demandshocks for those countries are expected to be correlated. The findings ofsignificant correlations in demand shocks together with a high degree ofcorrelation in inflation among ASEAN-5 countries are consistent with thefact that similar monetary policies have been implemented in those countriesduring the examination period. According to Phui and Yuen (2001), if thecorrelation is positive, the shocks are categorised as symmetric, and if thecorrelation is negative or not statistically significant, the shocks areasymmetric. Therefore, in terms of correlation of shocks, the effect of theshocks on ASEAN-5 countries seems to be symmetric, especially forSingapore, Malaysia, and Thailand.

By examining the correlation of growth and inflation, correlation of thevarious shocks, the above results indicate that there is a core group of threeASEAN countries – Singapore, Malaysia, and Thailand – that seemssuitable for a currency union. The correlation of the accumulated impulse

ZHI LU XU ET AL.130

response to the shocks is the next point we need to investigate in order toconfirm whether Singapore, Malaysia, and Thailand exhibit similarpatterns.

4.3.3. Correlation Coefficients of Accumulated Impulse Response Functions to

Three Types of Shocks

Based on economic theory, high correlations of responses to shocks suggestsymmetric shocks between the countries, hence low costs of monetary union.The correlation coefficients of accumulated impulse response functions todomestic supply shocks, domestic demand shocks, and external shocks arediscussed below.

First of all, we found that the correlations of accumulated impulseresponses to domestic supply and domestic demand shocks were all high(reaching rW0.9 in several country pairings) and statistically significant atthe 5% level (Tables A6 and A7). The correlations of price responses to bothsupply and demand shocks were all positive. Indonesia had a negativerelationship with the other four countries in terms of output responses tosupply shocks. Again, except for Indonesia, the other four countries hadsignificant correlations of output responses to external shocks. As forthe price responses to external shocks, those in most of the ASEAN-5countries were statistically significant except for Singapore–Philippines,and Malaysia–Philippines. Frenkel and Nickel (2002) argued that the higherthe correlations, the quicker the adjustment to the shocks. The overallpicture that emerges from the analysis of the response dynamics is thatSingapore, Malaysia, and Thailand seem to adjust more quickly to the sameshocks than the Philippines and Indonesia (Table A8).

4.3.4. Summary for Symmetry of Underlying Shocks

We have examined the symmetry and asymmetry of domestic demand,supply shocks, and external shocks among ASEAN-5 countries. Firstly, allASEAN-5 countries exhibited high degrees of correlation in terms ofinflation and growth. Secondly, in regard to the correlation of the shocks,we found that all ASEAN-5 countries were statistically significantlypositively correlated with each other in regard to the external shocks.Singapore, Malaysia, and Thailand were found to have highly correlatedsupply and demand shocks. Finally, high correlations of responses to theshocks were found in Singapore, Malaysia, and Thailand, which suggestedquicker adjustments to the same shocks in the core group.


4.4. Summary

In this section, two preconditions for an OCA, namely, economicconvergence and symmetry of underlying shocks have been examined. Byapplying the Kalman Filter procedure, we found evidence that nominalexchange rates and inflation rates tend to be converging in Singapore,Malaysia, and Thailand (ASEAN-3). However, no strong evidence wasfound to suggest whether the ASEAN-3 countries should adopt the USdollar or the Japanese Yen as their common currency. Moreover, we alsofound high correlations of inflation rates in those countries. This evidencesuggests that similar monetary policies have been implemented. The SVARanalysis showed some evidence that the effect of shocks on the core group,comprising Singapore, Malaysia, and Thailand, may be symmetric (hencepossible candidates for currency union). Based on the empirical resultsdiscussed above, the next section provides a detailed discussion andconclusion of this study.

5. DISCUSSION AND CONCLUSIONS

This research examined the feasibility of adopting a single currency inASEAN-5 countries by looking at two economic preconditions for acurrency union – economic convergence and symmetry of underlyingshocks. This study found that a subgroup, comprising Singapore, Malaysia,and Thailand (hereafter referred to as ASEAN-3), may be candidates foradopting a single currency for a start in the ASEAN region. However, thisstudy did not find strong empirical evidence for whether the Japanese Yenor US dollar would be a suitable currency for the ASEAN-3 countries toadopt commonly.

Singapore, Malaysia, and Thailand exhibit common economic character-istics. They have had relatively high rates of GDP growth compared to theother seven countries during the examination period (1970–2004), excludingthe period covering the 1997 Asian financial crisis. Singapore, Malaysia, andThailand were the top three countries in the region that had a high value ofmerchandise exports, which on average made up 73% of total merchandiseexports in the region during 1996 to 2004 (ASEAN Annual Report, ASEANStatistics, 2004–2005). Not only have the ASEAN-3 countries been tradingheavily among themselves, but also have been major trading partners for therest of seven countries in the ASEAN region. Furthermore, the ASEAN-3countries are more attractive to foreign investors compared to the other

ZHI LU XU ET AL.132

seven ASEAN countries because they are able to provide an environmentthat stimulates economic growth. These three countries received 86% offoreign direct investments in the entire ASEAN region in 2004 (ASEANSecretariat). Finally, it can be observed within the ASEAN-3 countries thatthere is a shift from the agriculture sector to the industry and services sectorsin the structures of their GDP. These common economic characteristics thatwere found for the ASEAN-3 countries are consistent with the empiricalfindings that the ASEAN-3 countries may be suitable candidates for acurrency union.

As for the common currency that the ASEAN-3 countries may adopt, thisstudy did not find evidence for them to adopt a single currency such as theJapanese Yen or the US dollar. The ASEAN-3 countries may be able to pegto a basket of currencies, comprising the US dollar, Japanese Yen, and theEuro. This idea is similar to China which replaces its fixed dollar peg with abasket peg and this could be a possible method the ASEAN-3 countriescould adopt in their pursuit of a currency union. However, in this basket,the Yen has been much more volatile than the US dollar and the Euro. If theYen continued to fluctuate as much as it has in the past, ASEAN-3 countrieswould have significant variations in their effective exchange rates.Furthermore, these three countries have different exchange rate regimes.Singapore and Thailand practise managed floating regimes, but Malaysiahas a fixed peg to the US dollar. It remains to be seen whether Malaysia willabandon its fixed peg to the US dollar in favour of a managed floatingregime requires political support and commitment by the Malaysiangovernment. Both a single-currency peg and basket-currency peg have prosand cons. Hence regardless of whether a single currency or a basket peg ischosen to be implemented in the ASEAN-3 countries, regional exchangerate stability and competitiveness are always important for economicdevelopment. As the ultimate objective is to adopt a single currency for thewhole ASEAN region, we should look at the whole picture of the region.

The economic progress towards a currency union is not likely to comeeasily for the ASEAN region. The main difficulties that the region is facingare high degree divergence in economic development and exchange rateregimes. Singapore, Malaysia, Thailand, and Brunei have relative high GDPper capita. Cambodia, Lao PDR, Myanmar, Vietnam (CLMV) have muchless developed economies. The economies in Indonesia and the Philippinesremain in the middle. It is questionable as to whether CLMV will growfaster and be able to catch up the other ASEAN countries and it is alsouncertain as to how long it will take to achieve economic convergence in thewhole region. The diversity of exchange rate regimes in ASEAN countries is


another underlying problem. The foreign exchange regimes range from ahard peg (Brunei) to an independent float (Thailand, Indonesia, and thePhilippines). It is very difficult for 10 member countries to agree with acommon exchange rate regime, therefore to adopt a common currency.Another stumbling block, coming from weak political commitments, wouldmake the road to a currency union even more difficult.

In addition to Japan and the US, ASEAN-5 countries also have strongtrade relationships with China. Since joining the World Trade Organisation in2001, China’s GDP growth has grown almost 10% per year and reached11.1% in the first quarter of 2007. China’s booming economy boosts its tradesurplus over the years and the trade relationship with the ASEAN countrieshas deepened. China has become ASEAN’s second largest import market in2005 (ASEANTrade Data). In 2002, China and the ASEAN countries formedplans to establish a bilateral free trade area (ASEAN Secretariat, 2005).

Huang (2006) believed that the revaluation of the Chinese RMB helpsASEAN monetary integration since the market-oriented regime for theRMB will be relatively more stable against the US dollar, which has recentlycome under increased pressure. However, Huang (2006) also cautioned thatthis system has functioned for only a short period of time and it is too earlyto ascertain whether the RMB could be a core currency for the ASEANcountries. Furthermore, the Chinese government only allows gradualappreciation of the RMB within a certain band, which may make the RMBundervalued. If the RMB is truly undervalued, this will certainly makeChinese goods cheaper in the world market, which leads to imbalancedbalance of payments. Thus, if the RMB were included in the study, it isdoubtful that this would affect our results and conclusions.

NOTES

1. Haldane and Hall (1991) investigated the following relationship between theUS dollar ($), the Deutsche Mark (DM) and Pound sterling (d)

DM

d

t

¼ at þ btDM

$

t

þ et

where DM/d is the logarithm of the nominal deutschmark–sterling exchange rateand DM/$ is the corresponding deutschmark–US dollar rate. If the coefficient a isconstant and b is zero, then the pound sterling converged on the deutschmark, whichwas the case in the mid-1980s.2. The SDR is special drawing rights created by the IMF. It is an international

reserve asset to supplement the existing official reserves of member countries. SDRs

ZHI LU XU ET AL.134

are allocated to member countries in proportion to their IMF quotas. The SDR alsoserves as the unit of account of the IMF and some other international organisations.Its value is based on a basket of key international currencies.3. The detailed charts showing the evolution of all estimates are not shown in

order to conserve space, but are available upon request to the authors.4. Detailed empirical results available on request to the authors.5. The statistic value is calculated in the form of t ¼ ðr

ffiffiffiffiffiffiffiffiffiffiffin� 2p

Þ=ðffiffiffiffiffiffiffiffiffiffiffiffiffi1� r2p

Þ

(Wackerly, Mendenhall, & Scheaffer, 1996, pp. 512–513), where r is the computedcorrelation coefficient and n the number of observations. The critical value is 2.042 at5% significance level. These t-statistic is also used to test the linear correlationcoefficients for the shocks and impulse response functions discussed below.

REFERENCES

ASEAN Secretariat (Association of Southeast Asian Nations). (2005), available at http://

www.aseansec.org/

ASEAN Annual Report, ASEAN Statistics. (2004–2005).

ASEAN Trade Data, ASEAN Secretariat (Association of Southeast Asian Nations). (1993–2003).

available at http://aseansec.org/trade/

Bayoumi, T., & Eichengreen, B. (1993). Shocking aspects of European monetary unification. In:

F. Torres & F. Giavazzi (Eds), Adjustment and growth in the European monetary union.

Cambridge: Cambridge University Press.

Bayoumi, T., & Eichengreen, B. (1994). One money or many? Analysing the prospects for monetary

unification in various parts of the world. Princeton Studies in International Economies, 76,

1–44, International Economics Section, Department of Economics, Princeton University.

Bayoumi, T., & Mauro, P. (1999). The suitability of ASEAN for a regional currency

arrangement. IMF working paper No. 162, Research Department.

Bernanke, B. S. (1986). Alternative explanations of the money-income correlation. NBER

Working Papers 1842, National Bureau of Economic Research, Inc.

Blanchard, O., & Quah, D. (1989). The dynamic effects of aggregate demand and supply

disturbances. American Economic Review, 79, 655–673.

Bunyaratavej, K., & Hahn, E. D. (2003). Convergence and its implications for a common

currency in ASEAN. ASEAN Economic Bulletin, 20, 49–59.

Country Studies. (2003–2005). Federal Research Division of the Library of Congress, available

at http://countrystudies.us/

Fidrmuc, J., & Korhonen, I. (2003). Similarity of supply and demand shocks between the Euro

Area and the CEECs. Economic System, 27(3), 313–334, Elsevier.

Frenkel, M., & Nickel, C. (2002). How symmetric are the shocks and the shock adjustment

dynamics between the Euro area and Central and Eastern European countries? EViews

User guide 5.0, Quantitative Micro Software, Irvine: California, 2005. International

Monetary Fund Working Paper, WP/02/222.

Haldane, A. G., & Hall, S. G. (1991). Sterling’s relationship with the dollar and the

Deutschmark: 1976–89. Economic Journal, 101, 436–443.

Hall, S. G., Roberson, D., & Wickens, M. R. (1992). Measuring convergence of the EC

economies. The Manchester School, LX(Suppl.), 99–111.


Huang, C.-J. (2006). An Asian currency band after the RMB revaluation. Unpublished Report,

Department of Finance, Tunghai University, China.

International Financial Statistics (IFS). (2000–2005). Browser, available at http://ifs.apdi.net/imf/

Jayaraman, T. K., Ward, B. D., & Xu Z. L. (2005). Are the Pacific Islands ready for a currency

union? An empirical study of degree of economic convergence. USPEC Working Paper,

No. 2005/2, February.

Johnston, J., & Dinardo, J. (1997). Econometric methods. New York: McGraw-Hill.

Kalman, R. E. (1960). A new approach to linear filtering and prediction problems. Journal of

Basic Engineering, 82, 306–321.

Kuroda, H. (2004). Transitional steps in the road to a single currency in East Asia. Asian

Development Bank, Seminar on May 14.

Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis

of stationary against the alternative of a unit root: How sure are we that economic time

series have a unit root? Journal of Econometrics, 54, 159–178.

Laabas, B., & Limam, I. (2002). Are GCC countries ready for currency union? Kuwait: Arab

Planning Institute.

Lim, L. K. & McAleer, M. (2003). Convergence and catching up in ASEAN: A comparative

analysis. CIRJE Discussion Papers, available at http://www.c.u-tokyo.ac.jp/cirje/

research/03research02dp.html

Madhur, S. (2002). Costs and benefits of a common currency for ASEAN. ERD Working Paper

No. 12, Economic and Research Department, Asian Development Bank.

Mundell, R. A. (1961). Theory of optimum currency areas. American Economic Review, 51, 663–686.

Ng, T. H. (2002). Should the Southeast Asian countries form a currency union? The Developing

Economies, XL(2), 133–134.

Ocampo, R. F. De. (2004). A single currency for Asia: Is it time? 37th ADB Annual Meeting,

May 14.

Phillips, P. C. B., & Perron, P. (1988). Testing for a unit root in time series regressions.

Biometrica, 75, 335–346.

Phui, L., & Yuen, H. (2001). Optimum currency areas in East Asia: A structural VAR

approach. ASEAN Economic Bulletin (August 1), 1–24.

Rose A. K. (2004). A Meta-analysis of the effect of common currencies on international trade.

NBER Working Paper No. 10373, JEL No. F34, 1–26.

Sims, C. (1980). Macroeconomics and reality. Econometrica, 48, 1–48.

Sims, C. (1982). Policy analysis with econometric models. Brooking Papers of Economic

Activity, 2, 107–152.

Sims, C. A. (1986). Are forecasting models useable for policy analysis? Quarterly Review of the

Federal Reserve Bank of Minneapolis, 10(1), 1–16.

Shapiro, M., & Waston, M. (1988). Sources of business cycle fluctuations. In: O. Blanchard &

S. Fischer (Eds), NBER Macroeconomics Annual (pp. 111–148). Cambridge: MIT Press.

Taylor, J. B., & Dalziel, P. (2002). Macroeconomics, New Zealand edition, Milton, Qld: John

Wiley & Sons, Boston, MA: Houghton Mifflin.

Wackerly, D., Mendenhall, W., & Scheaffer, R. (1996). Mathematical statistics with applications

(pp. 512–513). New York: Wadsworth Publishing.

Zhang, Z., Sato, K., & McAleer, M. (2003). Asian monetary integration: A structural VAR

approach. CIRJE Discussion Papers, No. F-212.

ZHI LU XU ET AL.136

APPENDIX

Table A1. Correlation of Growth in ASEAN-5 Countries.

Singapore Malaysia Indonesia Philippines Thailand

Singapore 1.000

Malaysia 0.829 1.000

Indonesia 0.438 0.491 1.000

Philippines 0.381 0.413 0.190� 1.000

Thailand 0.497 0.692 0.339� 0.278� 1.000

�Indicates that correlation coefficient is not statistically significant at 5%.

Table A2. Correlation of Inflation in ASEAN-5 Countries.


Singapore 1.000


Indonesia 0.445 0.608 1.000

Philippines 0.486 0.453 0.237� 1.000

Thailand 0.856 0.873 0.547 0.359 1.000


Table A3. Correlation Coefficients of External Shocksin ASEAN-5 Countries.


Singapore 1.0000

Malaysia 0.6392 1.0000

Indonesia 0.4525 0.4971 1.0000

Philippines 0.6286 0.7528 0.6971 1.0000

Thailand 0.7471 0.7806 0.5755 0.6166 1.0000

Correlation coefficients of external shocks are all statistically significant.


Table A4. Correlation Coefficients of Supply Shocksin ASEAN-5 Countries.


Singapore 1.0000

Malaysia 0.6745 1.0000

Indonesia 0.3722 0.4877 1.0000

Philippines 0.3253� 0.3753 0.1949� 1.0000

Thailand 0.4293 0.5935 �0.0371� 0.4065 1.0000


Table A5. Correlation Coefficients of Demand Shocksin ASEAN-5 Countries.


Singapore 1.0000

Malaysia 0.7624 1.0000

Indonesia 0.4293 0.5349 1.0000

Philippines 0.4479 0.3296� 0.0441 1.0000

Thailand 0.8353 0.7371 0.4448 0.3871 1.0000


Table A6. Correlation Coefficients of Accumulated Impulse ResponseFunctions to Domestic Supply Shocks.


Impulse Response of Output

Singapore 1.000


Indonesia �0.680 �0.440 1.000

Philippines 0.970 0.768 �0.642 1.000

Thailand 0.959 0.769 �0.635 0.994 1.000

Impulse Response of Prices

Singapore 1.000


Indonesia 0.749 0.716 1.000

Philippines 0.793 0.884 0.364 1.000

Thailand 0.936 0.976 0.570 0.956 1.000

Correlation coefficients are all statistically significant at 5%.

ZHI LU XU ET AL.138

Table A7. Correlation Coefficients of Accumulated Impulse ResponseFunctions to Domestic Demand Shocks.



Singapore 1.000


Indonesia �0.688 �0.339 1.000

Philippines �0.581 �0.649 0.512 1.000

Thailand 0.790 0.902 �0.471 �0.904 1.000

Impulse Response of Prices

Singapore 1.000


Indonesia 0.737 0.913 1.000

Philippines 0.344 0.791 0.805 1.000

Thailand 0.590 0.976 0.872 0.739 1.000

Correlation coefficients are all statistically significant at 5%.

Table A8. Correlation Coefficients of Accumulated Impulse ResponseFunctions to External Shocks.



Singapore 1.000


Indonesia 0.300� 0.290� 1.000

Philippines �0.851 �0.782 �0.227� 1.000

Thailand 0.990 0.989 0.181� �0.836 1.000

Impulse Response of Price

Singapore 1.000


Indonesia 0.465 0.604 1.000

Philippines �0.020� 0.311� 0.384 1.000

Thailand 0.682 0.888 0.847 0.519 1.000



PART III:

BUBBLES AND SPILLOVERS

CHAPTER 7

PERIODICALLY COLLAPSING

BUBBLES IN THE ASIAN

EMERGING STOCK MARKETS

Ako Doffou

ABSTRACT

This chapter investigates empirically the existence of periodically

collapsing bubbles in the Asian emerging stock markets using the

Enders–Siklos (2001) momentum threshold autoregressive model. As

explained in Bohl (2003), this non-linear time series technique can be

used to analyze bubble driven run-ups in stock prices followed by a crash

in a non-cointegration framework with asymmetric adjustment. This

technique offers a more potent insight in the stock prices behavior than

can possibly be obtained using conventional non-cointegration tests. The

empirical findings for 10 Asian emerging stock markets from 1993 to

2005 refute the bubble hypothesis.

1. INTRODUCTION

The standard present value rule of asset pricing may fail in financial marketswhen infinitely many assets can be traded. It can be shown that asset pricescan be meaningfully decomposed into a fundamental value and a pricing





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00007-6

143

dx.doi.org/10.1016/S1569-3767(07)00007-6.3d

bubble. The fundamental value obeys the present value rule. Most of thedeviations of stock prices from the present value model can be captured bythe bubble. Since the early 1980s, new developments in the stock marketsand renewed investors’ interest in those markets have motivated academicresearchers to show continuous interest in the phenomenon of speculativebubbles. The emergence of bubbles is explained in the finance literature as aself-organizing process of infection among traders leading to equilibriumprices which deviate from fundamental values. This economic explanationmakes bubbles transient phenomena and leads to repeated fluctuationsaround fundamentals.

Rational bubbles can follow either explosive AR(1) processes withdeterministic time trends or more complex stochastic processes. These classesof bubbles assume that stock prices and dividends are not cointegrated, thatis, there does not exist a stationary linear combination of the stock price anddividend. Standard tests for non-cointegration are often subject to substantialsize distortion in the presence of periodically collapsing bubbles. Advances ineconometrics allow a deeper study of bubbles and can lead to a betterunderstanding of the characteristics of stock markets.

Earlier studies of the consistency of dividend and stock price data with themarket fundamental hypothesis found it difficult to distinguish thecontribution of hypothetical rational bubbles to stock prices from that ofunobservable market fundamentals. Diba and Grossman (1988a) proposedan alternative testing strategy using the standard unit root test and a test fornon-cointegration between real stock prices and dividends as a test forbubbles. The intuition behind this approach is as follows: If stock prices arenot more explosive than dividends, then rational bubbles do not existbecause if they do, the stock price time series will have an explosiveconditional expectation. But the standard unit root and non-cointegrationtests assume a unit root as the null hypothesis and a linear autoregressiveprocess. A special class of rational bubbles called periodically collapsingbubbles follow a non-linear process and therefore cannot be detected usingthe Diba and Grossman test methodologies. Using simulated data in thepresence of periodically collapsing bubbles, Evans (1991) showed that thestandard unit root and non-cointegration tests led to the incorrectconclusion of the absence of bubbles most of the cases. But, Evans’ resultis based only on Monte Carlo simulations, not on empirical evidence. Usingthe annual and monthly US real stock price and dividend time series for theperiod 1871–1995, Bohl (2003) investigates empirically the existence ofperiodically collapsing bubbles in stock prices using the Enders and Siklos(2001) momentum threshold autoregressive (MTAR) model. This model can

AKO DOFFOU144

handle non-linear processes in a non-cointegration framework and take intoaccount asymmetries in departures from the long-term equilibrium relation-ship. Hence, the MTAR model, by design, can capture empirically thecharacteristics of periodically collapsing bubbles. Bohl’s findings refuteEvans’ hypothesis of periodically collapsing bubbles in the US stock market.

This chapter also uses the Enders and Siklos (2001) MTAR model toinvestigate the existence of periodically collapsing bubbles in the AsianEmerging stock markets. The empirical findings, using the annual andmonthly real stock and dividend time series for the period 1993–2005 for 10Asian emerging markets, refute the bubbles hypothesis.

The chapter proceeds as follows. Section 2 explains the theoreticalunderpinnings of periodically collapsing bubbles. Section 3 describes theeconometric concepts and methodologies underlying the MTAR techniqueand how this technique is appropriate to capture the behavior of this class ofrational bubbles in stock prices. Section 4 provides the application andestimation results for the Asian emerging stock markets as well as the datadescription. Finally, Section 5 concludes the chapter.

2. THEORY OF PERIODICALLY COLLAPSING

BUBBLES

A stock non-arbitrage or fundamental value is typically defined as thepresent value of its expected future dividends based on all currentlyavailable information. Mathematically,

Pt ¼ ZEtðPtþ1 þDtþ1Þ (1)

where Pt is a real stock price at time t (non-arbitrage or intrinsic value), Z aconstant discount rate (Z ¼ 1=ð1þ rÞ), r the constant real expected return,Dt+1 the real dividend to the holder of the stock between t and t+1, and Et

denotes the expectations conditional on information at time t.The market-fundamentals solution to Eq. (1) is

Pt ¼ F t ¼X1k¼1

ZkEtDtþk (2)

provided the transversality condition limn!1

ZnEtPtþn ¼ 0 holds. This occurs

when the conditional expectations are defined and the sum converges. Whenthe transversality condition fails to hold, Eq. (1) has not one unique solution

Periodically Collapsing Bubbles in the Asian Emerging Stock Markets 145

given by Eq. (2), but an entire class of solutions called homogeneoussolutions given by

Pt ¼ Ft þ Bt (3)

where Bt, the bubble term, is any random variable that satisfies

Bt ¼ ZEtBtþ1 (4)

or equivalently

Btþ1 ¼Bt

Zþ btþ1 ¼ Btð1þ rÞ þ btþ1 (5)

where

btþ1 ¼ Btþ1 � EtðBtþ1Þ (6)

The bubble in the equity price is Bt, and the innovation in the bubble at timet+1 is bt+1 which has zero mean (Etbt+1=0). A stochastic bubble is createdwhen the innovation in the bubble bt has a constant, nonzero variance. Hence,if bubbles exist, they must be expected to grow at the real rate of interest.Bt embodies the notion of a rational speculative bubble and, if present, it willcause Pt to deviate from the market fundamental path defined by Ft.

In the absence of bubbles (Bt=0, 8k), then Eqs. (2) and (3) lead to

Pt � r�1Dt ¼ ðrZÞ�1X1k¼1

ðZÞkEtDDtþk (7)

Clearly, Eq. (7) shows that if Pt and Dt are generated by I(1) processes,then Pt�r�1Dt is generated by a stationary process (there is a stationarylinear combination of Pt and Dt, Pt and Dt must be cointegrated withcointegrating parameter r�1).

In the presence of bubbles, the bubble term Bt must be added to the right-hand side of Eq. (7) above. Because the bubble term Bt given in Eq. (4)follows a non-stationary process, Pt and Dt cannot be cointegrated in thepresence of bubbles because Pt�r�1Dt will have an explosive conditionalexpectation. Therefore, Diba and Grossman (1988a) suggest testing for non-cointegration between real stock prices and dividends as a test for bubbles.But, Evans (1991) pointed out the limitation of this procedure which leadsto the incorrect conclusion of non-existence of rational bubbles whenperiodically collapsing bubbles are present.

Evans (1991) periodically collapsing bubbles are a class of bubbles whichare extremely attractive in that they collapse almost surely in finite time and

AKO DOFFOU146

are strictly positive (Diba & Grossman, 1988b):

Btþ1 ¼ Z�1Bt2tþ1 if Bt � a (8a)

Btþ1 ¼ ½dþ ðpZÞ�1ytþ1ðBt � ZdÞ�2tþ1 if Bt4a (8b)

where Z=(1+r)�1, a and d are positive parameters with 0odoaZ�1, 2tþ1 isan exogenous independently and identically distributed positive randomvariable with Et2tþ1 ¼ 1, and yt+1 an exogenous independently andidentically distributed Bernoulli process (independent of 2tþ1) which takesthe value 1 with probability p and the value 0 with probability 1�p, where0opo1. Hence, p is the probability of continuation of the bubble.

It is easy to verify that the process in Eq. (8) satisfies Eq. (4) and that Bt>0implies Bm>0, 8m4t. As long as Btra, the bubble grows at mean rate1+r=Z�1. When Bt>a, the bubble moves into a phase in which it grows atthe faster mean rate (pZ)�1 as long as the eruption continues, but in whichthe bubble collapses with probability 1�p per period. When the bubblecollapses, it falls to a mean value of d, and the process begins again. Varyingd, a, and p leads to an alteration of the frequency with which bubbles erupt,the average length of time before collapse, and the scale of the bubble.

Eqs. (8a) and (8b) show that Evans’ bubbles model satisfies two theoreticallywell-grounded properties of stochastic bubbles. First, this class of bubblescannot completely burst because after a complete collapse they cannot emergeagain. Second, a negative stock price bubble cannot exist because it wouldimply a negative expected stock price which is not economically sound.

Periodically collapsing bubbles clearly satisfy Eq. (4). Using Monte Carlosimulations, Evans (1991) shows that this class of bubbles may appear to bestationary on the basis of standard tests even though they are explosive byconstruction. This may be due to the sudden collapse of the bubble whichstandard tests may interpret as a mean reversion, biasing the test towardsrejection of non-cointegration. This chapter explores the consequences ofusing the Enders–Siklos MTAR model to investigate empirically theexistence of periodically collapsing bubbles in the Asian emerging marketsstock prices. A brief description of this model follows.

3. THE MOMENTUM THRESHOLD

AUTOREGRESSIVE MODEL

The momentum threshold autoregressive (MTAR) model in Enders andSiklos (2001) can capture the characteristics of periodically collapsing


bubbles. When periodically collapsing bubbles are present in stock prices,the estimated residuals o�t from the cointegration regression

Pt ¼ l�0 þ l�1Dt þ o�t (9)

shows patterns of increases in stock prices followed by a sudden drop. Thiskind of behavior of the stock price can be captured in the following regression

Do�t ¼ Ktf1o�t�1 þ ð1� KtÞf2o

�t�1 þ

Xtj¼1

xjDo�t�j þ mt (10)

where Kt, the indicator variable, is defined as follows: Kt=1 if Do�t�1 � O andKt=0 if Do�t�1oO, with O being the value of the threshold.

In the MTAR model, the null hypothesis of no cointegration isH0 : f1 ¼ 0;H0 : f2 ¼ 0, and H0 : f1 ¼ f2 ¼ 0. The critical values for thecorresponding t- and F-statistics are provided in Enders and Siklos (2001),Tables 1 and 2. The null hypothesis of symmetric adjustment H0 : f1 ¼ f2

can be tested using the F-statistic if the null hypothesis of no cointegration isrejected. When the null hypothesis of symmetric adjustment is not rejected,we can conclude that the stock price series Pt and dividend series Dt arecointegrated. That is, there is a stationary linear combination of Pt and Dt

with symmetric adjustment. A special case of the MTAR test is the Engleand Granger (1987) test. However, for a wide range of adjustmentparameters, the MTAR test is more powerful when asymmetric departuresfrom equilibrium occur.

Table 1. Monte Carlo Simulation Results Based on the MTARMethodology.

Significance Level 10% 5% 1%

Null Hypothesis f1=f2=0 f1=f2 f1=f2=0 f1=f2 f1=f2=0 f1=f2

Exact rejection

of the null

hypothesis for

different

values of the

probability p

0.99 0.991 0.718 0.982 0.601 0.968 0.513

0.95 0.991 0.715 0.982 0.598 0.967 0.511

0.85 0.991 0.708 0.983 0.589 0.967 0.499

0.75 0.991 0.694 0.984 0.579 0.969 0.482

0.65 0.992 0.648 0.986 0.553 0.978 0.464

0.50 0.993 0.561 0.990 0.541 0.982 0.447

0.25 0.994 0.476 0.994 0.463 0.986 0.396

0.10 0.996 0.402 0.998 0.417 0.989 0.365

Note: Each entry in this table represents the percentage of cases in which the null hypothesis is

correctly rejected. The details of the Monte Carlo simulation are provided in the text.

AKO DOFFOU148

As clearly stated in Bohl (2003), the MTAR model is designed toempirically detect periodically collapsing bubbles because theoretically,there is a potential for these bubbles to take positive but not negative values.Moreover, the run-ups or increases in stock prices before a crash occurs arean indication of an asymmetry in the evolution of the residuals of thecointegration regression, i.e., Eq. (9). The path of changes in o�t�1 above thethreshold followed by a sharp drop to the threshold captures periodicallycollapsing bubbles. But, the path changes in o�t�1 below the threshold doesnot show bubble eruptions followed by a collapse.

If the threshold is constrained to zero (O=0), a positive change in theestimated residuals (Do�t 40) indicates a rise in stock prices relative todividends followed by a crash, where the departure from present value rulescan be persistent and substantial according to Evans (1991). In contrast,when Do�t o0, decreases in stock prices relative to dividends followed by asharp rebound back to the equilibrium position is less likely. Theseasymmetric deviations from the equilibrium position are indicative of theexistence of periodically collapsing bubbles in stock prices. In this case, the

Table 2. Unit Root Tests.

Pt Dt DPt DDt

Panel A: Annual Data

DF �0.058 �0.093 �12.472� �11.033�

t 0 0 0 0

KPSS 1.975� 3.022� 0.384 0.269

Panel B: Monthly Data

DF 0.082 �1.323 �16.398� �12.104�

t 5 5 4 4

KPSS 14.109� 16.481� 0.43 0.13

Note: Pt is the real stock price at time t, Dt the real dividend at time t, DPt the change in the

stock price at time t, DDt the change in dividend at time t, DF the augmented Dickey and Fuller

(1981) statistic and KPSS is the Kwiatkowski et al. (1992) statistic. Hall (1994) procedure is

used to determine the time lag t of the DF tests. The Schwert (1989) approximation,

t ¼ int½4ðT=100Þ�1=4, is used to compute the time lag of the KPSS tests. For the KPSS tests, the

time lag is t ¼ 4 for annual data and t ¼ 7 for monthly data. Annual and monthly stock and

dividend time series for 10 Asian emerging stock markets are used. These markets include

Hong Kong, Singapore, Taiwan, Thailand, Malaysia, India, Pakistan, Indonesia, Philippines,

and South Korea. These data are obtained from the International Finance Corporation (IFC)

Emerging Markets Data Base (EMDB). Tests are performed on the IFC Emerging Market

Investable Indexes.�Statistical significance at the 1% level.


estimated coefficient f�1 is statistically significant and negative and greaterthan f�2 in absolute value, and the null hypothesis of symmetric adjustmentH0 : f1 ¼ f2 is rejected.

As opposed to a test of the null hypothesis of no cointegration, a test ofcointegration with MTAR adjustment, even though an indirect test ofthe presence of periodically collapsing bubbles, overcomes the problemsinherent in standard unit root and cointegration tests identified inEvans (1991).

The key objective and contribution of this chapter is the investigation ofthe null hypothesis of symmetry, not the rejection of the null hypothesis ofno cointegration. Therefore, using Eqs. (8a) and (8b), Evan’s (1991) MonteCarlo simulations are replicated by setting the parameter values as follows:r ¼ 0:05; Z ¼ 1=ð1þ rÞ ¼ 0:9524; a=1; d=0.50; Bt value at time zero =d;and T=100. In this chapter, 10,000 runs of the simulations are conductedand the corresponding regressions are assessed. Because the true value of thethreshold parameter O is not known ex ante, Chan’s (1993) approach is usedto estimate this parameter. The estimated residuals are sorted in ascendingorder, with the 15% largest and smallest values deleted. From the remaining70% residuals, the threshold parameter which yields the lowest residual sumof squares is selected (e.g., Enders & Siklos, 2001). The degree of rejection ofthe null H0 : F1 ¼ F2 ¼ 0 and H0 : F1 ¼ F2 is compiled in Table 1 at the10, 5, and 1% significance level and for different probabilities p varyingfrom 0.99 to 0.10. The null hypothesis F1 ¼ F2 ¼ 0 is highly rejected foralmost all significance levels and for almost all levels of the probability ofcontinuation of the bubble per period p. The degree of rejection increasesslightly as the probability p decreases. The degree of rejection of the nullhypothesis F1 ¼ F2 is more than acceptable and increases with thesignificance level. Overall, the explanatory power of both tests is very high.Hence, the F-test for the symmetry hypothesis is robust enough to identifyany asymmetry when the actual data generating process is dictated byEvans’ bubble model.

4. DATA AND EMPIRICAL RESULTS

Data were collected from 10 emerging Asian stock markets: Hong Kong,Singapore, Taiwan, Thailand, Malaysia, India, Pakistan, Indonesia,Philippines, and South Korea. The data were obtained from the Interna-tional Finance Corporation (IFC) Emerging Markets Data Base (EMDB).Tests are performed on the IFC Emerging Market Investable Indexes. The

AKO DOFFOU150

IFC investable indexes were introduced in March 1993. The IFC investableindexes are adjusted to reflect the accessibility of markets and individualstocks to foreign investors. These indexes offer a performance benchmarkfor international investors who might view the illiquid or restricted securitiesin a market to be irrelevant. Unit root tests and cointegration approachesare applied to the real annual and monthly stock price and dividend data forAsian investable emerging markets for the period 1993–2005. The indexprice series are the market capitalization weighted series of individual stockprice series in the index. The index dividend series are also the marketcapitalization weighted series of the individual stock dividend series in theindex. The index price series are regressed over the index dividend series.The empirical results are summarized in Tables 2 and 3.

The stochastic properties of real Asian emerging markets stock priceseries and real dividend series are examined separately by applying theDickey and Fuller (1981) or DF method and the Kwiatkowski, Phillips,Schmidt, and Shin (1992) or KPSS approach. For these tests, theapproximate critical values are taken from MacKinnon (1991) and Sephton(1995), respectively. Table 2 shows the results of the real Asian emergingmarkets stock price series Pt and real dividend series Dt as well as the seriesassociated with the changes in these variables, namely DPt and DDt. Hall(1994) procedure is used to determine the time lag t of the DF tests while theSchwert (1989) approximation, t ¼ int½4ðT=100Þ�1=4, is used for the KPSStests. The KPSS tests investigate the null hypothesis of level stationarity andthe DF tests are undertaken with a constant term. All test statistics arereported at the 10, 5, and 1% significance level.

In Table 2, the DF tests cannot reject the null hypothesis of a unit root inthe real stock price and dividend time series but they reject the nullhypothesis of a unit root in both time series of the changes in value DPt andDDt. The KPSS tests reject the null hypothesis of level stationarity butcannot reject the same null hypothesis for the DPt and DDt time series.A careful observation of the statistics in Table 2 leads to the conclusion ofthe existence of one unit root in the level of both types of time series.Another set of tests such as DF tests with a constant term and a linear timetrend in the alternative hypothesis and KPSS tests that investigate the nullhypothesis of trend stationarity are also examined. The findings of thesealternative tests, not reported here, support the results presented in Table 2.The data frequency does not affect the results in Table 2, consistent withBohl (2003) and other recent research in the literature of bubbles studies.The results of the unit root tests in Table 2 refute the existence of speculativebubbles in the Asian Emerging Stock Markets.


The test for cointegration between the real stock prices and dividends isthen conducted using the Engle and Granger (1987) methodology based onEq. (9) and the support regression Do�t ¼ Fo�t�1 þ

Ptj¼1 xjDo�t�j þ mt. The

lag lengths t are picked based on the statistically significant coefficientsof the lagged values Do�t�j. The results of the cointegrating regressionDurbin–Watson (DW) tests and the cointegrating regression augmented

Table 3. Cointegration Tests.

Monthly

Data

Annual Data

Panel 1: Engle–Granger Results

Estimated cointegrating parameter (l�1) 37.781 33.146

Cointegrating regression Durbin–Watson statistic

(DW)

0.085 0.611��

Cointegrating regression augmented Dickey–Fuller

statistic (DF)

�6.174� �4.295��

Coefficient of determination (R2) 0.848 0.912

Lag length (t) 1, 5 0

Panel 2: Johansen Procedure (Trace Test)

Estimated cointegrating parameter (l�1) 39.011 35.951

Number of cointegrating vectors W ¼ 0 33.264� 14.625��

Number of cointegrating vectors W � 1 0.214 0.087

LM1 – Type test of first-order autocorrelated

residuals

3.726 3.382

LM4 – Type test of fourth-order autocorrelated

residuals

6.083 4.513

Lag length (t) 1, 2, 3 1

Panel 3: MTAR Methodology

Estimated threshold parameter (O�) using Chan

(1993)

0.782 11.228

Estimated parameter of the MTAR model (f�1) �0.053 �0.625

(5.221)� (4.241)�

Estimated parameter of the MTAR model (f�2) �0.027 �0.313

(2.13)�� (2.371)��

F-statistic for the null hypothesis of no cointegration

(F�NC)

11.491� 8.053��

F-statistic for the null hypothesis of symmetric

adjustment (F�SA)

3.978 2.492

Lag length (t) 1, 5 0

� Statistically significant at the 1% level.�� Statistically significant at the 5% level.�� Statistically significant at the 10% level, respectively. t-statistics are in parentheses.

AKO DOFFOU152

Dickey–Fuller (DF) tests are reported in Table 3, Panel 1. Both tests rejectthe null hypothesis of no cointegration at the 5% significance level. Inaddition, the Johansen’s (1991) maximum likelihood approach is appliedwith the lag lengths picked based on the criteria of serially uncorrelatedresiduals. To this end, the LM-type tests for first- and fourth-orderautocorrelation (LM1 and LM4) are carried out.

The finding based on the trace test statistics is that the real stock price seriesand real dividend series are cointegrated. Moreover, the estimated values ofthe cointegrating coefficients l�1 are stable for all the cointegration techniquesimplemented. Based on the conventional Engle–Granger and Johansencointegration tests (Table 3), which both assume linear and symmetric

adjustment, the real stock price and dividend time series are cointegrated.Hence, these two conventional cointegration analyses refute the existenceof speculative bubbles in the Asian emerging stock markets. The resultsachieved here are not affected by the alternative specifications and testmethodologies.

But the conventional tests indicated above cannot rule out the existence ofperiodically collapsing bubbles. To be able to test for asymmetric adjustmentpatterns in favor of the existence of periodically collapsing bubbles, theMTAR univariate model in Enders and Granger (1998) is applied separatelyto the time series DPt and DDt. The results, not displayed here, are asfollows: (1) the annual time series do not show asymmetries; (2) the monthlytime series show statistically significant adjustment patterns at the 10% levelsupporting the existence of periodically collapsing bubbles.

The test results for the MTAR model appear in Table 3, Panel 3.These results include the estimated parameters f�1 and f�2 in Eq. (10) and therelated t-statistics for the null hypotheses H0 : f1 ¼ 0 and H0 : f2 ¼ 0; theF-statistics, F�NC, which tests the null hypothesis of no cointegrationH0 : f1 ¼ f2 ¼ 0; the F-statistics, F�SA, which tests the null hypothesis ofsymmetric adjustment H0 : f1 ¼ f2; and the consistently estimated attractorparameter O� using Chan’s (1993) approach. The estimated parametersrelated to the deviations below and above the threshold are negative andstatistically significant at the 5 and 1% level. The F�NC statistics are statisticallysignificant at the 5 and 1% levels for the annual and monthly time series,respectively, and therefore reject the null hypothesis of no cointegration. Inabsolute terms, the estimated values for f�1 are higher than those for f�2. TheF�SA statistics cannot reject the null hypothesis of symmetric adjustment. Thisis most likely due to a synchronized asymmetric behavior across the two timeseries. The results of the MTAR cointegration tests in Panel 3 of Table 3provide the evidence that refutes the existence of periodically collapsing


bubbles in the Asian emerging stock markets: the null hypothesis of nocointegration is rejected and the residuals generated by the run-ups in thestock prices followed by a crash do not exhibit an asymmetric development.

5. CONCLUSIONS

This chapter investigates empirically the existence of periodically collapsingbubbles in monthly and annual Asian emerging markets stock prices, usingthe Enders and Siklos (2001) MTAR cointegration model. Although thesebubbles clearly satisfy Eq. (4), Evans (1991) shows, using Monte Carlosimulations, that they may often appear to be stationary on the basis ofstandard tests, even though they are by construction explosive. Intuitively,this may be due to the sudden collapse of the bubble, which standard testsmay in some sense ‘‘mistake’’ for mean reversion, biasing the test towardsrejection of non-cointegration. The proposed model is a generalization ofEngle and Granger (1987) two-step procedure and can be used to formallytest for rational speculative bubbles which may burst after they have reachedcertain levels. The bubbles component can be seen as a non-linear process inthe alternative hypothesis. Even in the case the actual data generating processis given by Evans (1991) bubble model, the MTAR technique remains a veryrobust test to detect periodically collapsing bubbles. The results of the MonteCarlo simulations conducted here support this assertion.

Based on the MTAR approach, the empirical results in this chapter refutethe existence of periodically collapsing bubbles in the Asian emerging stockmarkets for the period 1993–2005. Moreover, deviations from the long-termequilibrium relationship do not appear to show an asymmetric adjustmentof the residuals from the long-run relationship. These results do not supportEvans’ (1991) claim of periodically collapsing bubbles, but are consistentwith Bohl (2003). These results are also consistent with Taylor and Peel(1998) who propose a test based on a modification to the least squaresestimator designed to be robust in the presence of error terms which mayexhibit strong skewness and kurtosis.

ACKNOWLEDGMENT

This chapter won the Best Paper Prize Award at the Asian FinanceAssociation conference held in Hong Kong, China, on July 4–7, 2007. ThisAward was sponsored by the University Utara Malaysia.

AKO DOFFOU154

REFERENCES

Bohl, M. T. (2003). Periodically collapsing bubbles in the US stock market? International

Review of Economics and Finance, 12, 385–397.

Chan, K. S. (1993). Consistency and limiting distribution of the least squares estimator of a

threshold autoregressive model. The Annals of Statistics, 21, 520–533.

Diba, B. T., & Grossman, H. I. (1988a). Explosive rational bubbles in stock prices? American

Economic Review, 78, 520–530.

Diba, B. T., & Grossman, H. I. (1988b). The theory of rational bubbles in stock prices.

Economic Journal, 98, 746–754.

Dickey, D. A., & Fuller, W. A. (1981). The likelihood ratio statistics for autoregressive time

series with a unit root. Econometrica, 49, 1057–1072.

Enders, W., & Granger, C.W.J. (1998). Unit-root Tests and asymmetric adjustment with an

example using the term structure of interest rates. Journal of Business and Economic


Enders, W., & Siklos, P. L. (2001). Cointegration and threshold adjustment. Journal of Business

and Economic Statistics, 19, 166–176.

Engle, R. F., & Granger, C. (1987). Cointegration and error correction: Representation,

estimation and testing. Econometrica, 55, 251–276.

Evans, G. W. (1991). Pitfalls in testing for explosive bubbles in asset prices. American Economic

Review, 81, 922–930.

Hall, A. (1994). Testing for a unit root in time series with pretest data-based model selection.

Journal of Business and Economic Statistics, 12, 461–470.



Kwiatkowski, D., Phillips, P., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis of

stationarity against the alternative of a unit root. Journal of Econometrics, 54, 159–178.

MacKinnon, J. G. (1991). Critical values for cointegration tests. In: R. F. Engle &

C. W. Granger (Eds), Long-run economic relationships: Readings in cointegration

(pp. 267–276). Oxford: Oxford University Press.

Schwert, G. W. (1989). Tests for unit roots: A Monte Carlo investigation. Journal of Business

and Economic Statistics, 7, 147–159.

Sephton, P. S. (1995). Response surface estimates of the KPSS stationarity test. Economics

Letters, 47, 255–261.

Taylor, M. P., & Peel, A. P. (1998). Periodically collapsing stock price bubbles: A robust test.

Economics Letters, 61, 221–228.


CHAPTER 8

CURRENCY CRISES IN ASIA:

A MULTIVARIATE LOGIT

APPROACH

Jan P. A. M. Jacobs, Gerard H. Kuper and Lestano

ABSTRACT

Indicators of financial crises generally do not have a good track record.

This chapter presents an early warning system (EWS) for six countries in

Asia in which indicators do work. We extract a full list of currency crisis

indicators from the literature, apply factor analysis to combine the

indicators, and use these factors as explanatory variables in logit models

which are estimated for the period 1970:01–2001:12. The quality of the

EWS is assessed both in-sample and out-of-sample. We find that money

growth (M1 and M2), national savings, and import growth correlate with

currency crises.

1. INTRODUCTION

In view of the large costs associated with financial crises being able topredict a crisis is crucial. Market indicators of default and currency risks,such as interest rate spreads and changes in credit ratings, hardly provide awarning of financial crises either because lenders do not have access to





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00008-8

157

dx.doi.org/10.1016/S1569-3767(07)00008-8.3d

timely and comprehensive information on the creditworthiness of theborrower or because lenders expect an official bailout of a troubledsovereign borrower. This resulted in the construction of a monitoring tool,the so-called early warning system (EWS).1 A EWS consists of a precisedefinition of a crisis and a mechanism for generating a prediction of thecrisis. Typically, a EWS has an empirical structure that forecasts thelikelihood of a financial crisis with indicators that show a country’svulnerability to a crisis. EWS models differ widely in terms of the definitionof a financial crisis, the time span on which the EWS is estimated and theforecast horizon, the selection of indicators, and the statistical oreconometric methods used. A common feature of all existing EWSstudies is the use of fundamental determinants of both the domesticsectors commonly reflected by financial and real sector activities, andexternal sectors approximated by current account and capital accountindicators.

The list of studies on EWS of financial crises is long. A full list isbeyond the scope of this chapter. The literature distinguishes the threevarieties of financial crises: currency crises, banking crises, and debt crises.Interested readers are referred to Kaminsky, Lizondo, and Reinhart (1998)for papers on currency crises prior to the Asian crisis; Bustelo (2000) andBurkart and Coudert (2002) on the Asian crisis; and Abiad (2003) for recentstudies on currency crises in emerging markets as well as industrialcountries. Gonzalez-Hermosillo (1996) and Demirguc--Kunt and Detragiache(1997, 2005) focus on banking crises, while Cline (1995) and Marchesi (2003)survey debt crises. We restrict our attention in this chapter to currency crisessince this type of crises is considered to have triggered the Asian financialcrisis.2

Several methods have been suggested to construct EWS models. The mostpopular one is used in this chapter, namely qualitative response (logit orprobit) models. Examples are Frankel and Rose (1996) and Frankeland Wei (2005), who study currency crises, and Demirguc- -Kunt andDetragiache (1997, 2000) and Eichengreen and Arteta (2002) on bankingcrises. Alternatives are cross-country regression models with dummyvariables as put forward by Sachs, Tornell, and Velasco (1996), graphicalevent studies as suggested by Eichengreen, Rose, and Wyplosz (1995) andAziz, Caramazza, and Salgado (2000), and the signal extraction approach,a probabilistic model proposed by Kaminsky et al. (1998), Goldstein,Kaminsky, and Reinhart (2000), and Edison (2003). In the last methodvalues of individual indicators are compared between crisis periods andtranquil periods. If the value of an indicator exceeds a threshold, it signals

JAN P. A. M. JACOBS ET AL.158

an impending crisis. For a review and comparison of different indicators,we refer to Perez (2005). Recently, Martinez-Peria (2002), Coe (2002),and Abiad (2003) proposed a Markov-switching EWS; Tudela (2004) useda duration model approach, while Falcetti and Tudela (2006) focus onintertemporal linkages using a dynamic probit model.

This chapter develops an econometric EWS of currency crises of six Asiancountries, Malaysia, Indonesia, the Philippines, Singapore, South Korea,and Thailand. These countries have been selected because the Asian fluhit Thailand and spread to other countries in the region almostinstantaneously. We set up logit models for currency crises with indicatorsextracted from a broad set of potentially relevant financial crisis indicators.The models are estimated using panel data for the January 1970–December2001 period. A panel data set can be useful because it allows us to sort outeconomic effects that cannot be distinguished with the use of time seriesdata alone.

The set-up of our EWS is similar to Kamin, Schindler, and Samuel (2007)and Bussiere and Fratzscher (2006), who also adopt a binomial multivariatequalitative response approach. However, while the final result of their(unreported) specification search is a set of combinations of indicators asexplanatory variables, we apply factor analysis to reduce this informationset. For investigations involving a large number of observed variables, it isuseful to simplify the analysis by considering a smaller set of linearcombinations of the original variables. The development of the factors overtime seems to have important consequences for the probability of a currencycrisis to occur. The factor analysis outcomes in combination with theestimation results and the ex post and ex ante track record allow the generalconclusion that (some) indicators of financial crises do work, at least in ourEWS of Asia. This finding is in contrast with IMF (2002) and Edison (2003),who observe that the performance of a EWS is generally poor and at bestmixed. Our method – the combination of factor analysis and logit modelling– enables us in principle to answer the question posed by Bustelo (2000)whether additional indicators have explanatory power to predict financialcrises. It also allows the dismissal of uninformative indicators.

The organisation of the chapter is as follows. Section 2 describes howwe date currency crises. The results – dummy variables indicating dates ofvarious crises – are used in binary choice models that explain the probabilityof a crisis. Section 3 describes our set of indicators, while Section 4 presentsfactor analysis and factors. Section 5 presents the binomial multivariatelogit models for currency crises. We analyse the performance of the modelsin-sample and out-of-sample in Section 6. Section 7 concludes.

Currency Crises in Asia 159

2. DATING CURRENCY CRISES

Generally, a currency crisis is defined to occur if an index of currencypressure exceeds a threshold.3 Eichengreen et al. (1995) made an importantearly effort to develop a method to measure currency pressure and to datecurrency crises. Their definition of exchange rate pressure is inspired by themonetary model of Girton and Roper (1977). The exchange rate is underpressure if the value of a constructed index exceeds a certain threshold. Theindex consists of weighted relative changes of the nominal exchange rate,international reserves, and interest rates to capture successful as well asunsuccessful speculative attacks. All variables in their index are relative to areference country and their threshold is time-independent. For the dating ofcurrency crises they set the exchange market pressure index threshold to twostandard deviations from the mean. The method of Eichengreen et al. washeavily criticised, which led to alternatives based on the same methodology.Kaminsky et al. (1998) and Kaminsky and Reinhart (1999) followed theconcept of Eichengreen et al. fairly closely, but they excluded interest ratedifferentials in their index and comparisons to a reference country.

In this chapter, we identify episodes of currency crisis in East Asia withour own version of Kaminsky et al. in which we include interest rates in theindex. This choice is based on a more extensive evaluation of currency crisesdating methods (Lestano & Jacobs, 2007). In addition, experimentation withdifferent currency crisis concepts revealed that the concept used hereperformed best in an in-sample signal extraction experiment (Jacobs, Kuper,& Lestano, 2004).

Table 1 summarises the distribution of the currency crises over thesix Asian countries in our sample, 1970:01–2001:12. The total number of

Table 1. Currency Crises: Distribution over Countries.

Currency Crises

Indonesia 9 (2.34%)

Malaysia 10 (2.60%)

Philippines 12 (3.13%)

Singapore 11 (2.86%)

South Korea 7 (1.82%)

Thailand 9 (2.34%)

All countries 58 (2.52%)

Note: The number between parentheses shows the frequency of crisis occurrence, which is

calculated by dividing the total number of crisis months by the total number of observations.


currency crises identified with our method is 58 (2.52 percent of the sampleobservations), distributed more or less evenly over the six countries.

3. INDICATORS

We focus on indicators of macroeconomic development and externalshocks.4 Worsening of these indicators affects the stability of financialsystem and may result in a financial crisis. The indicators are selected onthe basis of economic theory and recent findings of empirical studies onfinancial crises. See Jacobs, Kuper, and Lestano (2005) for details andreferences. Another major consideration was the data availability on amonthly basis for our country coverage and sample. The indicators areclustered into four major groups:

� External: Real exchange rate (REX), export growth (EXG), importgrowth (IMP), terms of trade (TOT), ratio of the current account to GDP(CAY), the ratio of M2 to foreign exchange reserves (MFR), and growthof foreign exchange reserves (GFR).� Financial: M1 and M2 growth (GM1 and GM2), M2 money multiplier(MMM), the ratio of domestic credit to GDP (DCY), excess real M1balances (ERM), domestic real interest rate (RIR), lending and depositrate spread (LDS), commercial bank deposits (CBD), and the ratio ofbank reserves to bank assets (RRA).� Domestic (real and public): The ratio of fiscal balance to GDP (FBY), theratio of public debt to GDP (PBY), growth of industrial production(GIP), changes in stock prices (CSP), inflation rate (INR), GDP percapita (YPC), and growth of national saving (NSR).� Global: the growth rate of world oil prices (WOP), the US interest rate(USI), and OECD GDP growth (ICY).

The main source of the data is the International Financial Statistics ofthe IMF for the macroeconomic and financial indicators (IMF, 2003) andthe World Bank Development Indicators for the debt variables (WorldBank, 2002). Missing data are supplemented from Thomson Datastreamand various reports of the countries’ central banks. All data in localcurrency units are converted into US dollars. Some annual indicators areinterpolated to obtain a complete monthly database.


The appendix lists definitions, sources and transformations of our crisesindicators. Two types of transformations are applied to make sure that theindicators are free from seasonal effects and stationary: 12 monthspercentage changes, and deviation from linear trends. In case the indicatorhas no visible seasonal pattern and is non-trending, its level form is main-tained. Some unavailable indicators are approximated by closely relatedindicators, for example OECD GDP is substituted by industrial productionof industrial countries.

4. FACTOR ANALYSIS

The aim of this chapter is to calculate the probability of a currency crisis.However, the set of economic indicators that is informative to determinewhether or not crises will occur is huge. It is not feasible to include allindicators in the logit model because of a lack of observations andmulticollinearity among the indicators. So, we reduce the information set foreach country into a limited number of factors. These factors are then used asexplanatory variables in the logit model.

Technically speaking, factor analysis transforms a set of random variableslinearly and orthogonally into new random variables. For a detailedexposition of factor analysis including references see for example, Venablesand Ripley (2002). The first factor is the normalised linear combination ofthe original set of random variables with maximum variance; the secondfactor is the normalised linear combination with maximum variance of alllinear combinations uncorrelated with the first factor; and so on. Byconstruction factors are uncorrelated.

Unfortunately, there is no ‘best’ criterion for dropping the least importantfactors. The so-called Kaiser criterion drops all factors with eigenvaluesbelow one. The Cattell scree test is a graphical method in which theeigenvalues are plotted on the vertical axis and the factors on the horizontalaxis. The test suggests selecting the number of factors that corresponds tothe place of the curve where the smooth decrease of eigenvalues appears tolevel off. In general, the scree test provides a lower bound on the number ofrelevant factors. In this chapter, we use the Kaiser criterion since thiscriterion is widely used in the literature.

Table 2 lists eigenvalues and the total variance explained by the factorsfor each country. For most countries, eight factors emerge with aneigenvalue above unity.5


5. LOGIT MODEL

Since our dependent variable is a binary variable (where 1=crisis and 0=nocrisis) we use a binary choice model. Two popular versions are the probitand the logit model. The major difference is that the probit model is basedon the standard normal probability density function, whereas the logitmodel uses an S-shaped logistic function to constrain the probabilities tothe [0,1] interval. Predicted probabilities calculated by these models differonly slightly in practise. We opt for the logit model since it is easier to usecomputationally than the probit model. The logit model is specified as

PðZ ¼ 1Þ ¼ F ðZÞ ¼1

1þ e�Z¼

1

1þ e�ðaþbX Þ

where P is the probability that Z takes the value 1 and F is the cumulativelogistic probability function, X is the set of regressors and a and b areparameters. It can be shown that the regression equation is equal to

lnP

1� P

� �¼ Z ¼ aþ bX

In our model, the vector of explanatory variables X consists of the eightfactors rather than the full list of economic indicators themselves. Theestimation results are presented in Table 3. Intercepts and country-specificintercepts (fixed effects) are not reported. From the likelihood ratiostatistics, which tests the joint null hypothesis that all slope coefficient

Table 2. Eigenvalues and the Cumulative Sum of the Varianceof Eight Factors.

Eigenvalues Indonesia Malaysia Philippines Singapore South Korea Thailand

Factor 1 5.93 7.79 5.58 7.88 7.55 6.69

Factor 2 3.40 3.19 3.71 3.28 3.37 3.96

Factor 3 2.84 2.38 2.60 2.78 2.60 3.37

Factor 4 2.01 2.15 2.41 1.91 1.85 2.22

Factor 5 1.91 1.93 1.72 1.66 1.63 1.72

Factor 6 1.46 1.34 1.52 1.37 1.39 1.42

Factor 7 1.20 1.12 1.11 1.01 1.25 1.34

Factor 8 1.06 1.05 1.05 0.92 1.10 0.78

h2 0.76 0.81 0.76 0.83 0.80 0.83

Note: h2 represents the cumulative sum of the variance proportion explained by each factor.


except the constant are equal to zero, we conclude that the explanatoryvariables contribute significantly to the explanation of the variation in thecrises dummies. In addition, we observe that factor 1 has the largest impacton the predicted probability of a currency crises; it is significantly differentfrom zero at the 1% level.

Factor 1 has by far the largest contribution to predicting crises probabilities.Although interpretation of the estimated coefficients in terms of the underlyingindicators is not trivial, the eigenvector of factor 1 is informative, since factor 1is a linear combination of the indicators with weights given by the firsteigenvector. Table 4 reports indicators that have dominant weights in factor 1.The largest weights in factor 1 are related to the growth of money (M1 andM2), supporting Kamin et al. (2007), the growth of national saving, the rate ofgrowth of GDP per capita, and import growth. These variables are dominantfor all countries in our sample. Other variables that have an impact in somecountries are commercial bank deposits, growth of foreign exchange reserves,export growth, and to a lesser extent, the domestic real interest rate, terms oftrade, and the growth rate of world oil prices.

6. PERFORMANCE

The logit models discussed above produce estimated probabilities ofcrises. High probabilities signal crises, low probabilities tranquil periods.

Table 3. Estimation Results of the Binomial Logit Model.

Factor Coefficient z-statistic

Factor 1 �0.37 �5.80

Factor 2 0.01 0.06

Factor 3 0.18 1.77

Factor 4 0.33 3.75

Factor 5 0.24 2.12

Factor 6 0.13 1.10

Factor 7 0.02 0.28

Factor 8 0.15 1.08

McFadden R2 0.16

Observations with Z=1 58

Likelihood ratio statistic, w2 (8 degrees of freedom) 85.16

Note: The model is estimated with Huber–White robust standard errors. Fixed effects are not

reported. Critical values of the z-statistic at the 1% and 5% level are 2.57 and 1.96, respectively.

The critical value of the likelihood ratio test at 1% (8 degrees of freedom) is 20.09.


The model might give false signals, that is, a crisis does not take placedespite the logit model producing a high probability. There are fourpossibilities. A model may indicate a crisis (high estimated probability)when a crisis indeed occurs (P(1,1)) or it may indicate a crisis when no crisisactually takes place (P(1,0)). It is also possible that the model does notsignal a crisis (low estimated probability) where in fact a crisis does occur(P(0,1)). The final possibility (P(0,0)) is a situation in which the model doesnot predict a crisis and no crisis occurs. Table 5 lists the four possibilities.

Once we generate time series of crisis probabilities, we can evaluate theforecasting ability of the model. Instead of carrying out a standardsignalling experiment along the lines of for example, Frankel and Rose(1996) and Berg and Pattillo (1999), which both require an ad hocassumption on the translation of estimated crisis probabilities into crisisdummies, we use the quadratic probability score (QPS) and the logprobability score (LPS) proposed by Diebold and Rudebusch (1989). Bothscores give an indication of the average closeness of the predictedprobabilities and the observed realisations, as measured by a dummyvariable that takes on a value of one when there is a crisis and zerootherwise. Suppose we have a time series of T probability forecasts fPgTt¼1,where Pt is the prediction probability of the occurrence of crisis or no crisis

Table 4. Loadings for the First Factor that has the LargestContribution to Predicting Crises Probabilities.

Indicator Indonesia Malaysia Philippines Singapore South Korea Thailand

CBD 0.02 0.08 0.06 0.07 0.01 0.05

EXG 0.07 0.06 0.09 0.03 0.08 0.06

GFR 0.06 0.03 0.03 0.00 0.02 0.07

GM1 0.11 0.09 0.09 0.12 0.06 0.09

GM2 0.10 0.10 0.09 0.12 0.06 0.10

IMP 0.05 0.07 0.08 0.09 0.08 0.09

NSR 0.09 0.11 0.10 0.13 0.10 0.12

RIR 0.03 0.01 0.05 0.01 0.05 0.02

TOT 0.07 0.01 0.05 0.05 0.02 0.01

WOP 0.06 0.01 0.04 0.06 0.03 0.03

YPC 0.10 0.10 0.10 0.13 0.10 0.12

Abbreviations: CBD, commercial bank deposits; EXG, export growth; GFR, growth of foreign

exchange reserves; GM1, growth of M1; GM2, growth of M2; IMP, import growth; NSR,

growth of national saving; RIR, domestic real interest rate; TOT, terms of trade; WOP, growth

rate of world oil prices; and YPC, rate of growth of GDP per capita.


event by the model in period t. Similarly, let fZtgTt¼1 be the corresponding

time series of realisations; Zt equals one if the event occurs in period t andequals zero otherwise. The QPS and LPS are then given by

QPS ¼1

T

XT

t¼1

2ðPt � ZtÞ2

LPS ¼ �1

T

XT

t¼1

ðð1� ZtÞ lnð1� PtÞ þ Zt lnðPtÞÞ

The QPS has the desirable property of being strictly proper, meaningthat it achieves a strict minimum under the truthful revelation probabilitiesby the forecaster. In addition, it is the unique proper scoring rule thatis a function only of the discrepancy between realisations and assessedprobabilities. The QPS ranges from 0 to 2, with a score of 0 correspondingto perfect accuracy if the estimated probability is 1(0) and a crisis does(not)occur for all t. A score of 2 shows that the model indicates a perfect falsesignal in which the estimated probability is 0(1) and a crisis does(not) occurfor all t.

The LPS depends exclusively on the probability forecast of the event thatactually occurred, assigning as a score the log of the assessed probabilities.In two events of crises, that is, crisis (Z=1) and no crisis (Z=0), the LPS isa fully general scoring rule, because the probability forecast of a crisis (Pt)implicitly determines the probability forecast of a tranquillity (1�Pt).Clearly, LPS 2 ð0;1Þ with LPS=0 being perfect accuracy and LPS ¼ 1being a perfect false signal. Interpretation of this boundary value is similarto the QPS. LPS and QPS imply different loss functions with large mistakesmore heavily penalised under LPS.

We evaluate crisis probabilities in-sample (1970:01–2001:12) and out-of-sample (2002:01–2002:12). Table 6 reports the goodness of fit of the model.The second and the third column report in-sample performance, while thelast two columns examine out-of-sample forecast performance. Recall thatthe closer the score statistics in Table 6 are to zero, the more accurate the

Table 5. The Probabilities of Right and Wrong Crisis Predictions.

Estimated Probability Crisis (Z=1) No Crisis (Z=0)

High P(1,1) P(1,0)

Low P(0,1)=1�P(1,1) P(0,0)=1�P(1,0)


model predictions. For all countries, the model performs quite well in-sample. The out-of-sample forecasts perform better than the in-sampleprojections. This should not come as a complete surprise considering thefact that hardly any currency crisis occurred in the forecast period.

7. CONCLUSION

This chapter builds an econometric EWS of six Asian countries, Malaysia,Indonesia, Philippines, Singapore, South Korea, and Thailand. We set up aqualitative choice – in our case logit – model. From the literature we extracta broad set of potentially relevant financial crisis indicators, which arecombined into factors using factor analysis. These factors are used asexplanatory variables in a panel covering the period January 1970–December 2001.

The factor analysis outcomes in combination with the estimation resultsof the logit model and the in-sample and out-of-sample performance allowthe general conclusion that (some) indicators of financial crises do work,at least in our EWS of six Asia countries. We find that the growth rates ofmoney (M1 and M2), GDP per capita, national savings, and importscorrelate with currency crises. Other variables that have an impact insome countries are growth rates of commercial bank deposits, foreignexchange reserves, exports, and to a lesser extent domestic real interest rates,terms of trade, and world oil prices changes. So, our method – thecombination of factor analysis and logit modelling – offers a solution to thebad performance (mixed and weak in timing of crisis) of EWS as noted byIMF (2002) and Edison (2003).

Table 6. Performance of the Logit Model.

Within Sample Out-of-Sample

QPS LPS QPS LPS

Indonesia 0.042 0.088 0.003 0.038

Malaysia 0.044 0.098 0.001 0.023

Philippines 0.056 0.120 0.001 0.023

Singapore 0.053 0.125 0.000 0.012

South Korea 0.033 0.086 0.004 0.043

Thailand 0.034 0.075 0.000 0.009

Abbreviations: QPS, quadratic probability score; and LPS, log probability score, respectively.


Our findings have the following implications for macroeconomic andexchange rate policies, for others there are no obvious links. Financialindicators of currency crisis, like money growth (M1 and M2) and foreignreserves growth, suggest the desirability to avoid large excesses of liquidliabilities and large shortages of liquid assets. Emerging market countriesshould maintain low discrepancies between the two liquidities by prearrang-ing lines of credit and keeping a sufficient stock of international reserves.Other indicators, like the terms of trade and growth rates of imports andexports are associated with external competitiveness; the real interest ratewith financial liberalisations; and commercial bank deposits with domesticbank runs and capital flight. Indeed, the reliability of these indicators inanticipating currency crisis may alert against current account problems,rapid liberalisations, and financial panic.

An EWS provides insights into which variables signal the likelihood ofcountries to experience a financial crisis. The models should be used withcare though. Applying our EWS to developed economies could easilyproduce a result similar to what The Economist (2003) reported, the USbeing at risk according to Damocles, Lehman Brothers’ EWS (Subbaraman,Jones, & Shiraishi, 2003). To avoid pitfalls like these, EWS analyses shouldbe accompanied by country risk assessments.

NOTES

1. For example, the IMF has put a lot of effort in EWS models; see Evans, Leone,Gill, and Hilbers (2000), IMF (2002), and Berg, Borenstein, and Pattillo (2004).2. Lestano, Jacobs, and Kuper (2003) present early warning systems for bank and

debt crises as well.3. Alternatives to dating schemes with thresholds are event-based methods or

Markov switching models. Event-based methods are commonly used in thecontagion literature to date crises from high volatility exchange rate events or newsrecorded by newspapers and journals, academic reviews, and reports of internationalorganisations. Examples of the former are Granger, Huang, and Yang (2000) and Itoand Hashimoto (2002); Kaminsky and Schmukler (1999), Glick and Rose (1999),and Dungey and Martin (2004) use news based currency crises. Martinez-Peria(2002) and Abiad (2003) adopt a Markov switching framework in their EWS model,which yields currency crisis dates.4. This study does not consider data on structural problems such as banking,

corporate governance, or corruption. Data on these issues are hard to obtain,especially for a long time span and higher frequency.5. For Singapore and Thailand, we maintain eight factors although only seven

factors have an eigenvalue above unity.


ACKNOWLEDGMENTS

The support of the Atma Jaya Catholic University (Jakarta, Indonesia)is gratefully acknowledged. The present version of the chapter benefitedfrom discussions with Mardi Dungey, helpful suggestions from MarcelFratzscher, Niels Hermes, and Elmer Sterken, and comments received atvarious seminars and workshops.

REFERENCES

Abiad, A. (2003). Early warning systems: A survey and a regime-switching approach. IMF

Working Paper 32. International Monetary Fund, Washington, DC.

Aziz, J., Caramazza, F., & Salgado, R. (2000). Currency crises: In search of common elements.

IMF Working Paper 67. International Monetary Fund, Washington, DC.

Berg, A., Borenstein, E., & Pattillo, C. (2004). Assessing early warning systems: How have

they worked in practice? IMF Working Paper 52. International Monetary Fund,

Washington, DC.

Berg, A., & Pattillo, C. (1999). Predicting currency crises: The indicators approach and an

alternative. Journal of International Money and Finance, 18(4), 561–586.

Burkart, O., & Coudert, V. (2002). Leading indicators of currency crises for emerging countries.

Emerging Markets Review, 3(2), 107–133.

Bussiere, M., & Fratzscher, M. (2006). Towards a new early warning system of financial crises.

Journal of International Money and Finance, 25(6), 953–973.

Bustelo, P. (2000). Novelties of financial crises in the 1990s and the search for new indicators.

Emerging Markets Review, 1(3), 229–251.

Cline, W. R. (1995). International debt re-examined. Washington, DC: Institute for

International Economics.

Coe, P. J. (2002). Financial crisis and the great depression: A regime switching approach.

Journal of Money, Credit, and Banking, 34(1), 76–93.

Demirguc- -Kunt, A., & Detragiache, E. (1997). The determinants of banking crises: Evidence

from developing and developed countries. Policy Research Working Paper Series 1828.

The World Bank.

Demirguc- -Kunt, A., & Detragiache, E. (2000). Monitoring banking sector fragility:

A multivariate logit approach. World Bank Economic Review, 14(2), 287–307.

Demirguc- -Kunt, A., & Detragiache, E. (2005). Cross-country empirical studies of systemic

bank distress: A survey. National Institute Economic Review, 192(1), 68–83.

Diebold, F., & Rudebusch, G. (1989). Scoring the leading indicators. Journal of Business, 62(3),

369–391.

Dungey, M., & Martin, V. L. (2004). A multifactor model of exchange rates with unanticipated

shocks: Measuring contagion in the East Asian currency crisis. Journal of Emerging

Market Finance, 3(3), 305–330.

Edison, H. J. (2003). Do indicators of financial crises work? An evaluation of an early warning

system. International Journal of Finance and Economics, 8(1), 11–53.


Eichengreen, B., & Arteta, C. (2002). Banking crises in emerging markets: Presumptions

and evidence. In: M. I. Blejer & M. Skreb (Eds), Financial policies in emerging markets

(pp. 47–94). Cambridge, MA: MIT Press.

Eichengreen, B., Rose, A. K., & Wyplosz, C. (1995). Exchange rate mayhem: The antecedents

and aftermath of speculative attacks. Economic Policy, 21, 251–312.

Evans, O., Leone, A. M., Gill, M., & Hilbers, P. (2000). Macroprudential indicators of financial

system soundness. IMF Occasional Paper 192. International Monetary Fund,

Washington, DC.

Falcetti, E., & Tudela, M. (2006). Modelling currency crises in emerging markets: A dynamic

probit model with unobserved heterogeneity and autocorrelated errors. Oxford Bulletin

of Economics and Statistics, 68(4), 445–471.

Frankel, J., & Wei, S. J. (2005). Managing macroeconomic crises: Policy lessons. In:

J. Aizenman & B. Pinto (Eds), Managing volatility and crises. A practitioner’s guide.

Cambridge: Cambridge University Press, Chapter 7.

Frankel, J. A., & Rose, A. K. (1996). Currency crashes in emerging markets: An empirical

treatment. Journal of International Economics, 41(3–4), 351–366.

Girton, L., & Roper, D. (1977). A monetary model of exchange market pressure applied to the

postwar Canadian experience. American Economic Review, 67(4), 537–548.

Glick, R., & Rose, A. K. (1999). Contagion and trade: Why are currency crises regional?

Journal of International Money and Finance, 18(4), 603–617.

Goldstein, M., Kaminsky, G., & Reinhart, C. (2000). Assessing financial vulnerability: An

early warning system for emerging markets. Washington, DC: Institute for International

Economics.

Gonzalez-Hermosillo, B. (1996). Banking sector fragility and systemic sources of fragility.

IMF Working Paper 12. International Monetary Fund, Washington, DC.

Granger, C. W. J., Huang, B. W., & Yang, C. W. (2000). A bivariate causality between stock

prices and exchange rates: Evidence from recent Asian flu. Quarterly Review of

Economics and Finance, 40(3), 337–354.

International Monetary Fund. (2002). Global Financial Stability Report, A Quarterly Report

on Market Developments and Issues. International Monetary Fund, Washington, DC.

International Monetary Fund. (2003). International Financial Statistics CD-ROM. Washington,

DC: International Monetary Fund.

Ito, T., & Hashimoto, Y. (2002). High frequency contagion of currency crises in Asia. NBER

Working Paper 9376. National Bureau of Economic Research, Cambridge, MA.

Jacobs, J. P. A. M., Kuper, G. H., & Lestano (2004). Currency crises in Asia: A multi-

variate logit approach. Mimeo April 2004. University of Groningen, Groningen, the

Netherlands.

Jacobs, J. P. A. M., Kuper, G. H., & Lestano (2005). Identifying financial crises. In:

M. Dungey & D. N. Tambakis (Eds), Identifying international financial contagion:

Progress and challenges (pp. 86–110). New York: Oxford University Press.

Kamin, S. B., Schindler, J. W., & Samuel, S. L. (2007). The contribution of domestic and

external sector factors to emerging market devaluations crises: An early warning systems

approach. International Journal of Finance and Economics, 12(3), 317–336.

Kaminsky, G. L., Lizondo, S., & Reinhart, C. M. (1998). Leading indicators of currency crisis.

IMF Staff Papers 45/1. International Monetary Fund, Washington, DC.

Kaminsky, G. L., & Reinhart, C. M. (1999). The twin crises: The causes of banking and

balance-of-payments problems. American Economic Review, 89(3), 473–500.


Kaminsky, G. L., & Schmukler, S. L. (1999). What triggers market jitters? A chronicle of the

Asian crisis. Journal of International Money and Finance, 18(4), 537–560.

Lestano, & Jacobs, J. P. A. M. (2007). Dating currency crises with ad hoc and extreme value-

based thresholds: East Asia 1970–2002. International Journal of Finance and Economics,

12(4), 371–388.

Lestano, Jacobs, J. P. A. M., & Kuper, G. H. (2003). Indicators of financial crises do work! An

early-warning system for six Asian countries. CCSO Working Paper 13. Department of

Economics, University of Groningen, the Netherlands.

Marchesi, S. (2003). Adoption of an IMF programme and debt rescheduling. Journal of

Development Economics, 70(2), 403–423.

Martinez-Peria, S. M. (2002). A regime-switching approach to the study of speculative attacks:

A focus on EMS crises. Empirical Economics, 27(2), 299–334.

Perez, J. (2005). Empirical identification of currency crises: Differences and similarities between

indicators. Applied Financial Economics Letters, 1(1), 41–46.

Sachs, J. D., Tornell, A., & Velasco, A. (1996). Financial crises in emerging markets: The

lessons from 1995 (with comments and discussion). Brookings Papers on Economic

Activity, 1, 147–198.

Subbaraman, R., Jones, R., & Shiraishi, H. (2003). Financial crises: An early warning system.

Research Report. Lehman Brothers.

The Economist. (2003). America the risky? The Economist (October 2).

Tudela, M. (2004). Explaining currency crises: A duration model approach. Journal of

International Money and Finance, 23(5), 799–816.

Venables, W. N., & Ripley, B. D. (2002). Modern applied statistics with S (4th ed.). New York:

Springer.

World Bank. (2002). World Development Indicators CD-ROM. Washington, DC: World Bank.

APPENDIX. EXPLANATORY VARIABLES –

DEFINITIONS AND SOURCES

External Sector (Current Account)

REX Real exchange rate (deviation from trend). The nominal exchange rate is

local currency unit (LCU) per USD, IFS line AE. The CPI is IFS line

64. The real exchange rate is the ratio of foreign (US CPI, IFS line

64ZF) to domestic prices (measured in the same currency). Thus,

REX=ePf/P, where e=nominal exchange rate, P=domestic price

(CPI), and Pf=foreign price (US CPI). A decline in the real exchange

rate denotes a real appreciation of the LCU.

EXG Export growth (12 months percentage changes). IFS line 70.D.

IMP Import growth (12 months percentage changes). IFS line 71.D.

TOT Terms of trade (12 months percentage changes). Unit value of exports

divided by the unit value of imports. Unit value of exports is IFS line

74.D. Import unit value for country (IFS line 75.D) is not available,


instead exports prices of industrialised countries are used, IFS line

110.74.D.

CAY Ratio of the current account to GDP. Current account (IFS line 78AL)

divided by nominal GDP (interpolation of IFS line 99B)

External Sector (Capital Account)

MFR Ratio of M2 to foreign exchange reserves (12 months percentage

changes). Ratio of M2 (IFS lines 34 plus 35) converted into USD and

international reserves (IFS line 1L.D).

GFR Growth of foreign exchange reserves (12 months percentage changes).

IFS line 1L.D.

Financial Sector

GM1 M1 growth (12 months percentage changes). IFS line 34.

GN2 M2 growth (12 months percentage changes). IFS line 35.

MMM M2 money multiplier (12 months percentage changes). Ratio of M2

(IFS lines 34 plus 35) to base (reserve) money (IFS line 14).

DCY Ratio of domestic credit to GDP (12 months percentage changes). Total

domestic credit (IFS line 32) divided by nominal GDP (interpolation

of IFS line 99B).

ERM Excess real M1 balance (based on estimated money demand equation).

Percentage difference between M1 (IFS line 34) deflated by CPI (IFS

line 64) and demand for real M1 estimated as function of real GDP,

nominal interest rates (IFS line 60L), and a time trend. If monthly real

GDP data is not available for a country, then its annual counterpart

(IFS line 99BP) is interpolated to monthly data.

RIR Domestic real interest rate. 6 months time deposit (IFS line 60L) deflated

by CPI (IFS line 64).

LDS Lending and deposit rate spread. Lending interest rate (IFS line 60P)

divided by 6 months time deposit rate (IFS line 60L).

CBD Commercial bank deposits (12 months percentage changes). Demand

deposits (IFS line 24) plus time, savings and

foreign currency deposits (IFS line 25) deflated by CPI (IFS line 64).

RRA Ratio bank reserves to bank assets. Bank reserves (IFS line 20) divided

by bank assets (IFS line 21 plus IFS line 22a to IFS line 22f).

Domestic Real and Public Sector

FBY Ratio of fiscal balance to GDP. Government budget balance (IFS line

80) divided by nominal GDP (interpolation of IFS line 99B).

PBY Ratio of public debt to GDP. Public and publicly guaranteed debt

(World Bank) divided by nominal GDP (interpolation of IFS line

99B).

APPENDIX. (Continued )


GIP Growth of industrial production (12 months percentage changes). If

industrial production index for a country is not available, then index

of primary production (crude petroleum, IFS line 66AA) is used.

CSP Changes in stock prices (12 months percentage changes). IFS line 62.

INR Inflation rate (12 months percentage changes). IFS line 64.

YPC GDP per capita (12 months percentage changes). GDP (interpolation

of IFS line 99B) divided by total population (interpolation of IFS

line 99Z).

NSR National savings (12 months percentage changes). Public (IFS line 91F)

and private consumption (IFS line 96F) subtracted from GDP

(interpolation of IFS line 99B).

Global Economy

WOP Growth of world oil prices (12 months percentage changes). IFS line

176.AA.

USI US interest rate (12 months percentage changes). The US Treasury bill

rate (IFS line 60C).

ICY OECD GDP growth (12 months percentage changes). Approximated by

industrial production (IFS line 66).

APPENDIX. (Continued )


CHAPTER 9

EVIDENCE OF BUBBLES IN THE

MALAYSIAN STOCK MARKET

Gary J. Rangel and Subramaniam S. Pillay

ABSTRACT

We tested for evidence of stock price bubbles in the Malaysian stock market

from 1978 to 2004. Four different tests were used namely excess volatility

tests, unit root/co-integration tests, duration dependence tests, and the

intrinsic bubbles model. All four tests indicate that during the sample

period, there was evidence of stock price bubbles. All tests results conform

to the theoretical literature on asset price bubbles except for the results on

the intrinsic bubbles model, which concludes that Malaysian investors under

react to information on dividends. We find this result hardly surprising as

anecdotal evidence does indicate that Malaysian investors place more

importance on capital gains as compared to dividends. Although we do not

go into a debate on whether authorities should be prick the bubble to stem

its negative effects, we argue that transparent information dissemination

will ensure that the stock market becomes more efficient in pricing stocks.

1. INTRODUCTION

Financial researchers have long been interested in the propagation of stockmarket bubbles. A bubble is essentially a significant deviation between





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00009-X

175

dx.doi.org/10.1016/S1569-3767(07)00009-X.3d

actual market value and its fundamental value. Most empirical research onstock market bubbles has focused on developed countries (Craine, 1993;Taylor & Peel, 1998; Kelleher, Kim, & Kim, 2001; Bohl, 2003; Wu & Xiao,2004). Empirical research on stock market bubbles in emerging countrieshas been sparse (Chan, McQueen, & Thorley, 1998; Chung & Lee, 1998;Sarno & Taylor, 1999; Herrera & Perry, 2003). As such, this chapter seeks toextend research on emerging market bubbles by focusing on evidence ofstock market bubbles in the Malaysian context.

The Kuala Lumpur Composite Index (KLCI), which is the mostcommonly used broad-based measure of the Kuala Lumpur Stock Exchange(KLSE) has experienced at least seven boom-bust cycles since its inception.The first crash, which occurred in 1973, lasted nearly 2 years. Neoh (1989)argued that an increase in speculation by first-time investors caused the run-up in stock prices and the subsequent crash. These first-time investors weremainly civil servants who did not have much experience in stock marketinvestment. The lack of alternative investments (the 3 months fixed depositrate at that time was 5%) caused the money in circulation to be channelledinto the local bourse and the bull-run took off in January 1971 and wentinto free fall from 13 February 1973 (Neoh, 1989).

As can be seen from Fig. 1, the second crash occurred in 1981, whichended a 16-month bull-run. The fall from peak to trough was about 58%, avery rapid fall by any standard. A less severe downturn occurred in thebeginning of 1984 and reached its trough only in May 1986. The next crashwas in 1987, which coincided with the 1987 crash of the Dow JonesIndustrial Average (DJIA). Another huge mania occurred in 1993 and manyordinary folk from hawkers to clerks were attracted to the market. Therewas nothing but talk of stock market speculation among many Malaysiansat that time. However, this came to an abrupt end when the market headedsouthwards in 1994 after foreign investors pulled out. The Malaysian stockmarket moved upward for the next 3 years culminating in another run up instock prices just before the 1997 Asian financial crisis. When the crisis hit,the KLCI registered its sharpest fall from its peak of 1271.57 points on25 February 1997 to 262.7 points on 1 September 1998, which is a fall of79.34%. Almost US$225 billion of market capitalisation was wiped offwhich made it the biggest stock market loss among the five crisis hitcountries (Athukorala, 1998).

The Malaysian stock market mirrored the Malaysian economic recoverywith the KLCI reaching a high of 1013.27 points on 18 February 2000.Nevertheless, this euphoria did not last long and the index tumbled to553.34 points on 9 April 2001.

GARY J. RANGEL AND SUBRAMANIAM S. PILLAY176

The fact that there are two extremes of valuation on the KLSE seems to bean indication of evidence of a stock market bubble. In fact, this cycle of boomsand busts seems to keep repeating itself as if to denote a lack of learning frompast mistakes by stock market investors. This chapter’s goal is simple.Analysis will be conducted to determine whether over a long period of time,there is evidence of bubbles occurring in the KLSE as measured by the KLCI.

The motivation of this chapter stems from the fact that how an asset isvalued at the present moment or in the future influences major economic andsocial policy decisions that affect not only investors but also society, andeven the world (Shiller, 2005). This contributes to a misallocation of scarceresources, for example business start-ups during the Internet boom of the late1990s and the proliferation investment trusts or investment companies whosesole duty was merely to arrange so that people could own stock in maturecompanies through a medium of new ones (Galbraith, 1971).

Several tests are used to detect evidence of stock prices bubbles. They arethe unit root/co-integration tests, excess volatility tests (Shiller, 1981), Chanet al.’s (1998) duration dependence test, and Froot and Obstfeld’s (1991)intrinsic bubbles model. The rest of the chapter is as follows. Section 2examines the literature from a historical perspective with a focus onevidence of stock price bubbles in emerging markets. We also look at themethodology to derive the fundamental values of stock prices. Section 3

0

200

400

600

800

1000

1200

1400

1980 1985 1990 1995 2000 2005

Inde

x (B

ase

1977

=10

0)

Month End Closing

Fig. 1. Kuala Lumpur Stock Exchange Composite Index (KLCI).

Evidence of Bubbles in the Malaysian Stock Market 177

presents the various tests. Section 4 describes the data used in this chapter.Section 5 focuses on the results and in Section 6, the conclusions arepresented.


Several episodes of run-ups in stock prices have been historically termed asstock price bubbles. One of the most famous is the US October 1929 crash.The stock market bubble built up until October 1929 when the frenzy onWall Street ended culminating in a 19.6% fall of the DJIA index for thatmonth alone. The index fell by a further 22% in the following month. As aconsequence of the 1929 Great Crash, the US economy slipped into arecession, which spread across the globe and lasted for several years.

Closer to our times is the 1987 Crash of the DJIA. Mishkin and White(2003) document significant aggregate price increases just before the collapsein October 1987. In a single day (19 October 1987), the DJIA suffered a22.6% decline, and for that month, the DJIA was down by 23.2%. The mostremarkable thing about these two episodes is the apparent similarities intheir patterns. Both bull markets started in the second quarter of the year(1924 and 1982, respectively) and lasted 63 months. Stock market pricesreached their zenith in the third quarter of the year (3 September 1929 and25 August 1987) and 54 days elapsed between the peak and market collapse(Brooks & Katsaris, 2003).

Japan’s bubble economy of the late 1980s is yet another example of theexcesses during boom times. The Nikkei 225 index began its upward ascentin 1986 and the index hit a peak of Yen 38,916 on the last trading day of1989. By the following year end, prices had fallen by 40%. The declinecontinued for another 2 years until stock prices bottomed out at the endof 1992. This represented a loss of 60% of market capitalisation whencompared to the peak. For the last 15 years, stock prices on the Tokyo StockExchange have shown little movement (Alexander, 1997).

If developed countries have experienced stock market bubbles, whatabout developing countries? Herrera and Perry (2003) document theexistence of stock market bubbles in Latin American countries in theirsample period from 1980 to 2001 based on unit root/co-integration tests andthe intrinsic bubbles framework. In quantifying stock price bubbles acrossthe region, they found 22 bubble episodes during 1980–2001. Bubbles andcrashes seem to have similar duration with bubbles persisting for an averageof 8 months while crashes last on average about 10 months.


Ahmed, Rosser, and Uppal (1999) test for evidence of non-linearspeculative bubbles in Pacific-Rim stock markets (including Malaysia)using a regime-switching methodology for a sample period from the late1980s to the middle 1990s. A vector autoregression (VAR) of fundamentalstock price values was estimated and the residuals from the VAR modelwere examined for absence or presence of trends using the regime switchingtechnique. The null hypothesis of no trends was rejected for all ten countriesthereby concluding that the presence of speculative bubbles could not beruled out.

Chan et al. (1998), however, reach the opposite conclusion. Although theyfound positive autocorrelation, negative skewness, and leptokurtosis invarying degrees of significance across six Asian stock markets (includingMalaysia) implying the presence of rational speculative bubbles, they foundlittle or no evidence when tests were conducted for duration dependence andconditional skewness. They attribute this to the nature of crashes inemerging market which exhibit a gradual progression from their peaks ascompared with the instantaneous crash as predicted by theory, which isexhibited by the US stock market.

We examine stock market bubbles within the framework of the theoryof rational bubbles. The theory of rational bubbles states that pricesmay deviate from fundamental values even though investors act rationally.Therefore, our assumption is that a stock market bubble can occureven within the confines of the Efficient Market Hypothesis (EMH).Blanchard and Watson (1982) illustrated this fact by incorporating aconcept of risk compensation. According to the rational bubble theory,as prices overshoot their fundamental values, there is an increase in theprobability of the bubble bursting. In turn, the possibility of financialloss increases the risk associated with the ownership of bubbling stock,thereby justifying the acceleration of its price (Pratten, 1993). Therefore,there are no arbitrage opportunities since all relevant information is stillcontained in stock prices. We operate base on this premise as Lai, Low, andLai (2001) indicate that Malaysian investors are generally rational and donot unduly react to extraneous events when it comes to making investmentdecisions.

Although the rational bubble theory provides the avenue for a multipleseries of tests to examine its evidence in a stock market, it does not explainhow bubbles are formed in the first place. Kindleberger (2000) offers aninsight on this. He theorised that speculation develops in two stages. In thefirst stage deemed as the sober stage of investments, households, firms, andinvestors respond to macroeconomic shock in a limited and rational way.


However, in the second stage, over optimism that positive events willcontinue into the future leads all actors to anticipate higher stock pricevalues in the future, which drives up demand for stocks.

3. TESTS FOR STOCK PRICE BUBBLES

In most finance texts, the fundamental value of a stock is equal to sum of allfuture discounted dividends. However, the actual stock price may deviatefrom the fundamental value whereby

pat ¼ pf

t þ bt þ ut (1)

where pat is the actual stock price in period t, bt a bubble component, and ut

the random error term. Therefore, the actual stock price has twocomponents, one being the ‘‘fundamental component’’ described as thediscounted value of future dividends and a bubble component. This bubblecomponent is expected to grow by at least the required rate of return as inEq. (2) below

Etðbt þ 1Þ ¼ ð1þ iÞbt (2)

Eq. (2) therefore rules out the possibility of profit-making arbitrageopportunities even though a bubble component is present in the stock price.Under the assumption that dividends grow slower than i, the marketfundamental part of the stock price converges. The bubble componenthowever is explosive or non-stationary in nature. Even though investors areaware of the existence of the bubble, they hold on to their investments inanticipation that there will be further price increases. Notice that theexpectation of making high capital gains from the sales of the stock in thefuture is consistent with no-arbitrage pricing as the value of the right to sellthe stock has already been factored in the current price (Gurkaynak, 2005).By substituting, it can be shown that

pat ¼

X1g¼1

1

ð1þ iÞgEtðdtþgÞ þ b0ð1þ iÞt þ ut (3)

A rational speculative bubble exists when both the fundamental and thebubble component grow by at least the required rate of return. In other


words, the bubble component grows at the same rate as the discount rate(Brooks & Katsaris, 2003). When this is the case, an investor cannot earnarbitrage profits by shorting the stock and buying it back at a future date(Gilles & LeRoy, 1992). However, the bubble component need notnecessarily grow at the rate of return. It can be more explosive as describedby Blanchard (1979) when investors are aware of the existence of the stockprice bubble and factor the probability of a crash back to fundamentalvalues occurring in the investment horizon. This notion of increasingrequired return from participating in a market where observed prices havedeviated significantly from fundamental value fits in well with the fact thatrisk averse investors required higher returns in order to compensate forhigher risks undertaken.

One assumption that is important in the development of the fundamentalvalue is the transversality condition. The fundamental value of the stockprice is broken up into two terms. One is the sum of the net present value ofexpected dividends and the other is the expected terminal value. This isrepresented by Eq. (4).

pat ¼

X1g¼1

1

ð1þ iÞgEtðdtþgÞ þ lim

g!1

1

1þ ig

� �Ptþg (4)

The transversality condition asserts that the second term on the right-hand side of Eq. (4) is zero. The reason for this is simple. If there is a positivebubble, and this term is not zero, the infinitely lived agent could sell thestock and the lost utility, which is the discounted value of the dividendstream, will be lower than the terminal value. This cannot be an equilibriumprice as then all agents will want to sell the stock and the price will fall to thefundamental level (Gurkaynak, 2005).

3.1. Variance Bounds Test

One of the first ways to test for the presence of stock price bubbles used theconcept of volatility. This is made possible by comparing the variance of theobserved stock prices and the variance of the fundamental stock pricesconstructed from the Dividend Discount Model. A case for the existence ofstock prices bubbles can be made if the variance of the actual stock prices issignificantly greater than variance of the constructed fundamental stockprices. Shiller (1981) performed the first tests of excess volatility by


comparing the volatility of actual prices and fundamental prices constructedusing ex post analysis. The fundamental prices were constructed fromactual dividend data with the assumption of perfect foresight, and aconstant discount rate. In order to obtain the infinite sum of dividends, aterminal value using the mean de-trended price of the sample period wasused. Variance bound tests are built on the assumption that the variance ofthe observed prices should be no bigger than the variance of fundamentalprices as derived from the Dividend Discount Model. This is represented byEq. (5).

Varð pft Þ ¼ Varð pa

t Þ þ VarðutÞ � Varð pat Þ (5)

Shiller (1992) presents two other inequalities, which refers to stock returnsand price changes. In all cases, the discount rate is constant where

gt ¼ g_¼

1

1þ r(6)

and the inequalities are

sðdtPtÞ �sðDtÞffiffiffiffi

r2p (7)

sðDPtÞ �sðDtÞffiffiffiffiffiffiffiffið2rÞ

p (8)

where dtPt is the innovation in price defined as Pt � Et�1Pt ¼ Ptþ Dt�1�

ð1þ rÞPt�1, r2 the two-period interest rate (1+r)2�1, and DPt ¼ Pt � Pt�1.The results of his analysis showed that the variance bounds are violatedimplying evidence of stock price bubbles. LeRoy and Porter (1981) concurwith these findings. Several criticisms have emerged.1 Mankiw, Romer, andShapiro (1985) proposed volatility tests that circumvent the issue ofdividend stationarity. They develop a naive forecast of the stock price

P0t ¼

X1k¼0

gkþ1F tDtþk (9)

where F tDtþk denotes a naive forecast of Dtþk made at time t. The naiveforecast does not need to take rational properties. However, it is assumedthat rational investors are privy at time t of this naive forecast. The


relationship between the naive forecast price, perfect foresight price, and theobserved prices is given by the identity in Eq. (10)

P�t � P0t ¼ ðP

�t � PtÞ þ ðPt � P0

t Þ (10)

Here ðP�t � PtÞ equals to ut as in Eq. (5) and thus is uncorrelated withinformation at time t. Therefore

Et ¼ ½ðP�t � PtÞðPt � P0

t Þ� ¼ 0 (11)

since Pt and P0t are known at time t. If both sides of Eq. (11) are squared and

taking the expectations, it thus implies

EtðP�t � P0

t Þ2¼ EtðP

�t � PtÞ

2þ EtðPt � P0

t Þ2 (12)

This equality further implies

EtðP�t � P0

t Þ2� EtðP

�t � PtÞ

2 (13)

and

EtðP�t � P0

t Þ2� EtðPt � P0

t Þ2 (14)

With the law of iterated projections, this allows the replacement ofexpectations conditional of information at time t with expectationsconditional on information available prior to the beginning of the sampleperiod (Mankiw et al., 1985). Therefore, if we let E denote the expectationsconditional on the initial conditions, Eqs. (12)–(14) can be re-written as

EðP�t � P0t Þ

2¼ EðP�t � PtÞ

2þ EðPt � P0

t Þ2 (120)

EðP�t � P0t Þ

2� EðP�t � PtÞ

2 (130)

EðP�t � P0t Þ

2� EðPt � P0

t Þ2 (140)

As long as the expectations are taken conditional on the informationavailable a finite amount of time before date t, non-stationarity poses nodifficulties for the existence of these conditional expectations (Mankiw et al.,1985). Eqs. (13u) and (14u) are the volatility relationships that are examined.


Eq. (13u) implies that the market price is a better forecast of the expost rational price, in terms of the mean squared error, than the naiveforecast stock price. If the naive forecast is better than the marketforecast, then the inequality (13u) is violated. Eq. (14u) states that the ex postrational price is more volatile around P0 than is the market price. Eq. (14u) isthus the same as Shiller’s volatility test. Mankiw et al. (1985) find that bothvariance inequalities are violated using Shiller’s dataset which indicates thatobserved prices are much more volatile as compared to present value priceseven after accounting for non-stationarity in dividends. They also weight theerrors in order to rule out heteroskedasticity but the results remainunaltered.

3.2. Non-Stationarity and Co-Integration Tests

Based on the Dividend Discount Model, prices are exclusively determinedby future discounted dividends. Diba and Grossman (1988a, 1988b)theorized that there would be no possibility of rational speculative bubblesif dividends and prices are stationary in the mean. However, even ifdividends and prices are non-stationary, the fact that both variables are co-integrated means that there is no possibility of stock price bubbles. Theirspecification of the fundamental price is:

Pft ¼

X1g¼1

1

1þ i

� �g

Etðdtþg þ otÞ (15)

where ot denotes the fundamentals that cannot be observed by theresearcher. If ot is assumed to be no more stationary than dt, then themarket fundamental price will be as stationary as the dividends. In theabsence of bubbles, if dividends are stationary in levels, stock prices will beequal to market fundamentals and should be stationary in levels as well.Under the null hypothesis of no bubbles in stock prices, and assuming thatot is stationary, dividend and stock prices should exhibit a long-termrelationship i.e. be co-integrated. Diba and Grossman found both dividendand price series for the S&P 500 were difference stationary implying a lackof evidence for stock price bubbles using Dickey–Fuller stationarity tests.A follow-up test using co-integration found both variables to be co-integrated which supports the earlier findings. Campbell and Shiller (1987)arrive at the same conclusion but add a caveat that the results are dependentupon the discount factor that is used.


3.3. Intrinsic Bubbles Model

The tests outlined above showed stock price bubbles can take on explosivecharacteristics. If bubbles are uncorrelated with fundamentals, they mustgrow at an expected rate of at least (1+i) per period in order to be arbitragefree or grow more than (1+i) should the risk of a collapse be factored intothe bubble return. However, bubble behaviour can also be tied together withfundamentals as suggested by Froot and Obstfeld (1991). In their model,stock price bubbles are driven non-linearly by exogenous fundamentaldeterminants of stock prices. The fundamental determinant is dividends. Intheir chapter, the dividend generating process is assumed to be following ageometric martingale as depicted in Eq. (16).

dtþ1 ¼ mþ dt þ �tþ1 (16)

where m is the trend growth in dividends, dt is the log of dividends at time t,and �tþ1 is a normal random variable with conditional mean zero andvariance s2. From Eq. (16), the present value stock price is proportionalto dividends with the assumption that t period dividends are known whenPf

t is set.

Pft ¼ kDt (17)

where k ¼ ðer � emþs2=2Þ�1. The non-linear function of the bubble is given by

Eq. (18).

BðDtÞ ¼ cDlt (18)

where c and l are constants. l is the positive root of the quadratic equationl2s2=2þ lm� r ¼ 0. Therefore, Froot and Obstfeld (1991) intrinsic bubblesmodel comprises of the summation of Eqs. (17) and (18)

Pat ¼ Pf

t þ cDlt (19)

or

Pat ¼ kDt þ cDl

t þ �t (20)

Intrinsic bubbles impart non-linearity into the relationship between stockprices and dividends. In this case, the price/dividend ratio is

Pt

Dt

¼ kþ cDl�1t þ �t (21)


where �t is a well-behaved error term. In order to test for bubbles, the price/dividend ratio is regressed on a constant and dividends. Not finding anysignificant coefficients except for the constant in these regressions willindicate a lack of bubbles, while finding a non-linear relationship betweenprices and dividends will be interpreted as signalling the presence of anintrinsic bubble (Gurkaynak, 2005). In Froot and Obstfeld’s (1991) case,they find strong positive significance for the coefficient c with data on USstock prices and dividends. Ma and Kanas (2004) reaffirm Froot andObstfeld’s findings by performing an out of sample forecast throughbootstrapping of their model. The intrinsic bubbles model outperforms twoalternative measures of stock prices, namely the random walk model and therational bubbles model.

3.4. Duration Dependence Test

The underlying methodology of this test stems from the implication that if arational speculative bubble were to occur, it would have typified a sequenceof observations of the same sign to denote positive reinforcement.Therefore, the probability of a run (sequence of same sign observations)of positive abnormal returns ending should decline with the length ofthe run (positive duration dependence or a negative hazard function)(McQueen & Thorley, 1994). The duration dependence test addresses theissue of non-linearity in bubble formation. This implies that the testparameters are allowed to vary (the probability of the run ending)depending on the run length and is also dependent on whether the runhas positive or negative abnormal returns.

The data set, ST comprises of T observations of random length, I. A run isdefined as abnormal returns of the same sign. Thus, I is a positive valueddiscrete random function generated by some discrete density function, f i

PrðI ¼ iÞ, and corresponding cumulative density function, F i PrðIoiÞ.Ni would denote the count of completed run length i. Thus, the densityfunction of the log likelihood is

LðyjST Þ ¼X1i¼1

Ni ln f i (22)

y is a vector of parameters to be estimated. The hazard function,hi ¼ PrðI ¼ ijI � iÞ, represents the probability that a run ends at i, giventhat it lasts at least until i. A hazard function describes the data in terms of


conditional probabilities in contrast to the density function specificationwhich focuses on unconditional probabilities. The hazard function relation-ship with the density function is based on

hi ¼f i

ð1� FiÞand f i ¼ hi

Yi�1j¼1

ð1� hjÞ (23)

Using the relationship in Eqs. (22) and (23) can be modified as

LðyjST Þ ¼X1i¼1

Ni ln hi þMi lnð1� hiÞ (24)

where Mi is the count of runs with length greater than i. In order to carryout the test for duration dependence, the hazard function must take afunctional form. McDonald, McQueen, and Thorley (1995) use the log-logistic functional form as

hi ¼1

1þ e�ðaþb ln iÞ(25)

The log-logistic function transforms the unbounded range of aþ b lnðiÞinto the (0, 1) space of hi, the conditional probability of ending run.Therefore, the null hypothesis of no bubbles implies that the probability of arun ending is independent of the prior returns. Therefore, b=0. Thealternative hypothesis on the other hand suggests that the probability of arun ending should decrease with run length. In this case the parameter of theslope, b, would be less than 0 or negative (decreasing hazard rate). Eq. (25)is substituted into Eq. (24) and the log-likelihood function is maximised withrespect to a and b. The likelihood ratio test (LRT) of b=0 is asymptoticallydistributed w2 with one degree of freedom (Chan et al., 1998).

The notion of applying the four tests listed for this chapter stems from thefact that no one true test can unequivocally detect the evidence of stock pricebubbles. Therefore, applying a multiple test approach exploits the strengthsof each test while ameliorating the weaknesses (Flood & Hodrick, 1990).

4. DATA

Several data sources were used in this chapter. The month end KLCI indexwas obtained from Bursa Malaysia (formerly known as the KLSE).Dividends were obtained from our own calculations of total net dividends


by month recorded by ex date paid out by companies who make up thecomponents of the KLCI. Most dividend information was obtained fromthe Kuala Lumpur Stock Exchange Gazette, Investors Digest, and variousyears of annual companies’ reports. Once the monthly total net dividendswere obtained, a 12-month moving average of annual dividends was dividedby the month end market capitalisation to obtain the dividend yield. Impliedmonthly dividends are obtained as a product of monthly dividend yield andprevious period monthly closing values of the KLCI.

5. RESULTS

In Section 3, several techniques to test for bubble presence were described.We discuss the results of the variance bounds test first before moving on tothe unit root and co-integration tests, duration dependence tests, and finallydiscuss the results of the intrinsic bubbles model.

5.1. Results of the Variance Bounds Test

Table 1 depicts the definitions of the symbols used for this test.2

The growth trend factor (l) is 1.002221 based on the regression of the logof price on an intercept and time trend. r is calculated to be 0.018426. Thismeans if one invests on a weighted portfolio based on the KLCI componentstocks over the sample period, an investor would be expected to receive aaverage monthly return of 1.84%. The results of the volatility analysis isshown in Table 2.

In essence, the variance bound is violated. The simple comparisonbetween the standard deviation of actual prices and the standard deviationof ex post prices shows that volatility of actual prices exceeds the volatilityof ex post prices by a factor of 4.01 times. Even when the comparisons aredone using price innovations as in Eqs. (7) and (8), the inequalities are stillviolated by a factor from 3.72 to 3.73 times.

There is, however, one differentiating finding between our results vis-a-visShiller (1981) for US data and Heaney (2004) for Australian data. Thecorrelation coefficient between ex post prices and the actual prices arepositive for both these studies. However, our results indicate a negativecorrelation albeit a very small one. To resolve this puzzle, we also analysedthe correlation coefficient between observed prices and ex post pricescalculated from Datastream obtained data. The sample period for


Datastream data is from 1986 to September 2004. Datastream does not havestock price index and dividend yield data for Malaysia prior to 1986. Thecorrelation coefficient turned out to be positive 0.3947.

To reconcile this, we split our sample into two whereby the first sampleperiod ranged from 1978 to 1985 and the second sample period ranged from1986 to end September 2004. The correlation co-efficient for the sampleperiod 1978 to 1985 turned out to be negative 0.28151. As for the periodfrom 1986 to September 2004, the correlation co-efficient between the actualprices and ex post prices was positive 0.4393, which is almost similar to thecorrelation co-efficient obtained using Datastream data. This indicates thatthe negative correlation co-efficient obtained for the entire sample periodfrom 1978 to September 2004 is heavily influenced by the results obtainedfor the first sample period. It also reflects the possible fact that therelationship between dividends and prices from 1978 to 1985 was differentfrom the second sample period of 1986 to September 2004 which seems morein-line with conventional theory.

We did consider whether our method of calculating the dividendyields could have influenced the findings on the negative correlation

Table 1. Definitions of Principal Symbols.

g = real discount factor for series before de-trending; g ¼ 1=ð1þ rÞ

g = real discount factor for de-trended series; g lgDt = real dividend accruing to stock index (before de-trending)

dt = real de-trended dividend; dt Dt=lt�T

D = first difference operator Dxt xt � xt�1

dt = innovation operator; dtxtþk Etxtþk � Et�1xtþk; dx dtxt

Et = mathematical expectations operator conditional on information at time t; Etxt

EðxtjItÞ where It is the vector of information variables known at time t

l = trend factor for price and dividend series; l ¼ 1þ g, where g is the long-run

growth rate of price and dividends

Pt = real stock price index (before de-trending)

pt = real de-trended stock price index; pt ¼ Pt=lt�T

p�t = ex post rational stock price index

r = one-period real discount rate for series before de-trending

r = real discount rate for de-trended series; r ¼ ð1� gÞ=gr2 = two-period real discount rate for de-trended series; r2 ¼ ð1þ rÞ2 � 1

t = time (month)

T = base month for de-trending and for consumer price index; pT=Pt= nominal

stock price index at time T

E = unconditional mathematical expectations operator. EðxÞ is the true (population)

mean of x


co-efficient. The major difference between our calculated dividend yieldsas compared with Datastream yields is that our dividend yields are based onnet dividends (after tax) whereas Datastream’s dividend yields are basedon gross dividends. The correlation co-efficient between the two dividendyields is very high at 0.9186 and is strongly significant for the sample periodfrom 1986M1 to 2004M10 where there was available overlapping of data.Fig. 2 depicts the ex post price as compared to actual observed prices overthe sample period.

The comparison of the ex post rational price and the actual prices showsthat several positive bubbles show up notably the early 1980s, 1984–1985,1987, and early 1990s. From this period on, there was a prolonged positivebubble right up to the 1997 Asian financial crisis, which caused a collapse inthe Malaysian stock market. A recovery took place thereafter and amillennium boom occurred for a brief period. As of the last month in thesample, actual prices are below fundamental values.

Table 2. Sample Statistics for Price and Dividend Series.

Sample Period 1978M1–2004M9

E(p) 1051.417

E(d) 19.37358

r ¼ EðdÞ=EðpÞ 0.018426

r2 ¼ ð1þ rÞ2 � 1 0.037192

b ¼ lnðlÞ 0.002219

sðbÞ (0.000196)

corðp; p�Þ �0.0871

s(d) 4.638813

Elements of inequality

Inequality (5)

sðp�Þ4sðpÞ violated 4.01 times

s(p) 350.8486

sðp�Þ 87.41795

Inequality (7)

sðDpþ dt�1 � rpt�1ÞosðdÞ=ffiffiffiffir2p

violated 3.72 times

sðDpþ dt�1 � rpt�1Þ 89.51905

sðdÞ=ffiffiffiffir2p

24.05376

Inequality (8)

sðDpÞosðdÞ=ffiffiffiffiffi2rp

violated 3.73 times

sðDpÞ 90.16067

sðdÞ=ffiffiffiffiffi2rp

24.16431

Note: In this table, E denotes sample mean, s the standard deviation, and s the standard error.

The rest of the symbols are as defined in Table 1.


In order to take into consideration of the criticisms by Marsh and Merton(1986) on the assumption of stationarity of dividends, we re-analyse the datausing the variance bounds tests developed by Mankiw et al. (1985). The expost rational price is defined as

P�t ¼XðT�tÞ�1

k¼0

gkþ1Dtþk þ gT�tPT (26)

Notice that instead of using the mean de-trended price as the terminalvalue, Mankiw et al. (1985) use the last observed de-trended real price.P�t corresponds to the ‘‘perfect foresight price’’ for the policy of holding theportfolio until time T and the selling it at the prevailing price. The naive

0

400

800

1200

1600

2000

2400

1978 1981 1984 1987 1990 1993 1996 1999 2002

Ex-Post Price Actual Price

Sto

ck in

dex

valu

es

Fig. 2. Actual KLCI Month End Values and Constructed Fundamental Values.

Note: Kuala Lumpur Composite Index (KLCI, dashed line) and ex post rational

price (solid line), from 1978M1 to 2004M9, both de-trended by dividing a long-run

exponential growth factor. The ex post rational is the present value of actual

subsequent real de-trended dividends, subject to an assumption about the present

value in 2004M9 of dividends thereafter.


forecast price obtained from

P0t ¼

gð1� gÞ

Dt�1 (27)

Table 3 presents the results for the various mean squared error terms atthe actual required rate of return of 1.84% per month. Fig. 3 shows the threeprice series.

Inequality (13u) is tested by comparing the second and the third columnbelow. The third column is higher than the second. This implies that thenaive forecast price, P0 is a better forecast of the perfect foresight price, P�

than is the market price P. Inequality (14u) as earlier mentioned is analogousto Shiller’s variance bounds test and is also violated. The comparison is

0

400

800

1200

1600

2000

2400

1978 1981 1984 1987 1990 1993 1996 1999 2002

P

P*

Po

Sto

ck in

dex

valu

es

Fig. 3. The Perfect Foresight Price (P�), the Naive Forecast Price (Po), and the

Market Price (P) for Case r=1.84%.

Table 3. Unbiased Volatility Tests (Not Weighted).

r (%) EðP� � P0Þ2 EðP� � PÞ2 EðP� P0Þ

2

1.84 114325.49 133719.54 255664.09


between the second column and the fourth column whereby the mean squareerror of column four is 2.2 times greater than that of column two. Inconclusion, when accounting for non-stationarity in dividends, the volatilityof actual values of stock prices still exceeds fundamental values.

Apart from volatility tests, we also compute how big the standarddeviation of the discount rate would have to be to account for the resultsobtained from inequality (8) using the Eq. (28).

sðrÞ �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2EðrÞsðDpÞ � sðdÞ

p� � EðrÞ

EðdÞ(28)

If we make the assumption that r in Table 2 equals EðrÞ and that thesample variances equals population variances, we find that the standarddeviation of rt would have to be at least 8.152 percentage points. Using theMalaysia’s Base Lending Rate (BLR) as published by the InternationalMonetary Fund’s International Financial Statistics, we find that thestandard deviation of the BLR was only 1.5851 percentage points (sampleperiod 1977–2002) which is dramatically lower than what the discount ratehas to be for fundamental prices to equal to actual prices. If we take, as anormal range for rt implied by these figures, a 2 standard deviationaround the real interest rate, r given by Table 2, then the real interest ratewould have to range from �14.4614 to 18.1466% (real interest rates havenever seen these kinds of levels).

5.2. Results of the Unit Root/Co-Integration Test

We next examined the time series properties of real prices and real dividendsby looking at unit root and co-integration analysis. The results are outlinedin Table 4.

The below results indicate that taken individually, both prices anddividends have unit roots suggesting that both series are not mean revertingwhen no constant or time trend is introduced. When a constant or aconstant and time trend is introduced, prices continue to be non-stationarybut dividends take on mean reverting properties. The estimated statistic alsoindicates that the price-dividend ratio is stationary when a constant or aconstant and time trend is introduced. Therefore, it is questionable whetherthese tests can be decisive, given the ambiguity of the results.

We thus conducted a co-integration test to substantiate the findings of theunit root tests. If dividends and prices exhibit a long-term relationship, then


there is no possibility of bubbles in stock prices. We use the Johansen test toidentify the presence or otherwise of co-integrating vectors in the levels ofprices and dividends. The results are outlined in Table 5.

The Johansen test statistic is less than its 5% critical value, implying thatthere are no linear combinations of the series that are stationary, andtherefore there is no co-integration over the sample period. Such a resultimplies that there are no long run relationship between dividends and prices.The theory thus concludes that the presence of speculative bubbles cannotbe rejected. It is quite possible that speculative bubbles may be transient, i.e.appearing in some periods but not in other periods.

5.3. Results of Duration Dependence Tests

The third test we conducted is the duration dependence test. As mentionedearlier, should stock prices exhibit bubble-like tendencies, the value of the

Table 4. Augmented Dickey–Fuller Unit Root Tests for MonthlyPrices, Monthly Dividends, and Monthly Price Dividend Ratio.

No Constant and No

Time Trend

Constant Only Constant and Time

Trend

Real price �0.424065 �2.381319 �2.441431

(lag length=3) (lag length=3) (lag length=3)

Real dividend �0.349815 �4.191269�� 4.837745��


Price:dividend ratio,

Pt/Dt

�1.279987 �3.757006�� 3.764814��


Note: �� denotes significance at the 1% level. The lag length is determined automatically by

using the Schwartz information criterion.

Table 5. Johansen Test Results for Co-integration betweenPrices and Dividends.

Trace Statistic Critical Values

(1%)

Eigenvalue Number of Co-integrating

Vectors Under the Null

Hypothesis

18.03541 30.45 0.037401 None

5.989912 16.26 0.018777 At most 1


slope the parameter, b, is negative (decreasing hazard rates). Otherwise,there is no duration dependence or b ¼ 0 (constant hazard rate or geometricdensity function). The results are shown in Table 6.

The monthly excess returns are obtained as a comparison between actualreal returns and the in-sample mean real return. If the actual real monthlyreturn is greater than the in-sample mean real return, a positive run isrecorded and likewise a negative run is recorded should the actual realmonthly return be less than the in-sample mean return. The results inTable 6 indicate that for the Malaysian stock market, there is no possibilityof a rational speculative bubble as b co-efficient is not significantly differentfrom zero implying constant hazard rates. The results of the negative runcounts are as per theory. However, a question can be put forth on the usage

Table 6. Run Counts, Hazard Rates, and Tests of DurationDependence for Runs of Monthly Excess Value-Weighted Portfolio

Returns (January 1978–August 2004).

Run Length Positive Negative

Actual Run

Counts Total=81

Sample Hazard

Rates

Actual Run

Counts Total=80

Sample Hazard

Rates

1 43 0.530864198 44 0.55

2 17 0.447368421 14 0.388888889

3 10 0.476190476 9 0.409090099

4 6 0.545454545 9 0.692307692

5 2 0.4 2 0.5

6 1 0.33333333 1 0.5

7 1 0.5 1 1

8 1 1

Log-logistic test

a 0.474349 0.428961

b 0.015548 0.102971

LRT of H0: b=0 0.037632 0.950485

p value 0.846183 0.329596

Note: A run length i is a sequence of i abnormal returns of the same sign. The sample hazard

rate hi ¼ Ni=ðMi þNiÞ, represents the conditional probability that run ends at i, given that it

lasts until i, where Ni is the count of runs of length i and Mi is the count of runs with length

greater than i. The log-logistic function is hi ¼ 1=1þ e�ðaþb ln iÞ. The likelihood ratio test (LRT)

of the null hypothesis, H0: b=0, of no duration dependence (constant hazard rate) is

asymptotically distributed w2 with one degree of freedom. p value is the marginal significance

level, which is the probability of obtaining the value of the LRT or higher under the null

hypothesis.


of the in-sample mean of monthly real returns, which is compared againstthe actual monthly real returns.

The weakness lies in the upward bias of the in-sample mean real monthlyreturns if the sample data does indeed have bubble-like tendencies.We therefore take an alternative approach. Instead of comparing returns,we compare the perfect foresight real price as obtained earlier with theactual observed real prices. A positive run is recorded should the actualprices exceed the perfect foresight price and a negative run is recordedshould the opposite be found. The advantage in using the perfect foresightprice is that there is no upward bias in its formation. We take the cue fromShiller (1992) who commented that the usual analysis in finance literature

Table 7. Run Counts, Hazard Rates, and Tests of DurationDependence for Runs of Comparison of Ex Post Monthly Real Prices

and Actual Monthly Real Prices (January 1978–August 2004).

Run Length Positive Run Length Negative

Actual Run

Counts

Total=10

Sample

Hazard Rates

Actual Run

Counts

Total=11

Sample

Hazard Rates

1 2 0.2 1 3 0.27272

2 1 0.10204 2 2 0.25

3 2 0.28571 5 1 0.16667

8 1 0.2 7 1 0.2

17 1 0.25 11 1 0.25

23 1 0.33333 23 1 0.33333

26 1 0.5 28 1 0.5

109 1 1 47 1 1

Log-logistic test

a 0.260033 0.216113

b �0.317293 �0.256438

LRT of H0: b=0 13.82929 5.607642

p value 0.0000 0.01788

Note: A run length i is a sequence of i abnormal returns of the same sign. The sample hazard

rate hi ¼ Ni=ðMi þNiÞ, represents the conditional probability that run ends at i, given that it

lasts until i, where Ni is the count of runs of length i and Mi is the count of runs with length

greater than i. The log-logistic function is hi ¼ 1=1þ e�ðaþb ln iÞ. The LRT of the null hypothesis,

H0: b=0, of no duration dependence (constant hazard rate) is asymptotically distributed w2

with one degree of freedom. p value is the marginal significance level, which is the probability of

obtaining the value of the LRT or higher under the null hypothesis. Those run length not listed

in the above table have zero run counts. For example run lengths 4 through 7 are not listed

because the run counts is zero.


tends to analyse data on returns, rather than price or value. The durationdependence analysis re-done using prices instead of returns as shown inTable 7.

The results in Table 7 using prices rather than excess real returns showsignificant duration dependence for both positive runs as well as negativeruns. The LRT test is also significant implying that the co-efficient theindependent variable, log(run) is different from zero. Therefore, rationalspeculative bubbles are present during the sample period. The case of thesignificance of the log(run) co-efficient in negative runs requires somefurther explanation. Chan et al. (1998) also found significant durationdependence in negative excess returns. They claim this duration dependenceis either driven by chance or fads but not by rational bubbles. However, ifwe look at Fig. 3, there are cases in which the actual price is consistentlybelow the ex post price. So, although the stock price index does not reachzero, but prolonged downturns do take place owing to investors’ receipt ofcontinuous bad information regarding the stock market, the economy, andthe political situation.

5.4. Results of the Intrinsic Bubbles Model

We conduct a final test based on the Froot and Obstfeld (1991) intrinsicbubbles model. In this model, stock prices depend exclusively ondividends. The first step in developing this model is the determination offundamental prices. Therefore, we calculate the implied fundamentalvalues based on stochastic version of Gordon’s (1962) model of stockprices which is essentially Eq. (17). Recall that k ¼ ðer � emþs

2=2Þ�1. From

the analysis of the data, r, the geometric average return of the KLCI isestimated to be 2.39% per month. The estimate of the parameters inEq. (16) are m ¼ 0:002525 and s ¼ 0:128144. Therefore, k ¼ ð1:023993�e0:002525þ0:128144

2=2Þ�1¼ 75:7591. This means that the fundamental monthly

values of the KLCI is 75.7591 times dividends. The results are differentfrom Froot and Obstfeld (1991) and also Ma and Kanas (2004) and thuscannot be used as a benchmark as not only is the country of analysisdifferent but also the frequency in their case is yearly whereas ours usesmonthly data.

Our results also differ from Ibrahim and Abdul Rahman (2003) whose kvalue was 1.54 for quarterly Malaysian stock data. The intrinsic bubblemodel of Eq. (21) defines that the price-dividend ratio is non-linearly relatedto dividends. We test this relationship by regressing the price-dividend ratio


onto dividends. As in Eq. (21), we constrain the value of l based onl2s2=2þ lm� r ¼ 0. The calculated value of l is 1.5626. The results areshown below in Table 8.

The regression results indicate that intrinsic bubbles were present duringthe whole sample period as the coefficient of the regressor is significant.However, the sign of the co-efficient is negative. This means that price-dividend ratio in general under reacts to information on dividends.Therefore, a preliminary conclusion would be that the behaviour ofaggregate prices on the Malaysian stock market as represented by theKLCI is different from that modelled by Froot and Obstfeld’s intrinsicbubbles model. Although our sample period is longer than Ibrahim andAbdul Rahman’s (2003) sample period, their results seems to imply that theintrinsic bubbles model mimics the behaviour of aggregate price on theMalaysian stockmarket.

The fact that the co-efficient is negatively significant and thus refers tounder reaction towards dividends is hardly surprising in an emerging marketsuch as Malaysia. Since the early period of the sample, Fig. 4 shows that thedividend yield has been on the downtrend except for spikes in 1985 and justafter the 1997 Asian financial crisis.

Neoh (1989) has presented anecdotal evidence that Malaysian investorsseem to care less for dividends as compared to capital gains. There areseveral reasons for this. Stock market capital gains are not considered astaxable income in Malaysia. However, dividends are taxed at the corporatetax rate. This is unlike in the US where stock market capital gains aretaxable. We conducted the same test for the Singaporean stock market,which also does not tax stock market capital gains and obtain the sameresults for the intrinsic bubbles model (Rangel & Pillay, 2007). We can thusconclude that in an environment where capital gains are not considered

Table 8. Estimates of Eq. (18), Pt=Dt ¼ kþ cDl�1t þ �t (Full Sample

Period 1978–2004).

Regression Method k c F Test R2 DW

Ordinary least squares (OLS) 73.33199�� 0.276112�� 28.13846� 0.08 0.107472

(5.927063) (0.075729)

Note: Standard errors are reported in parentheses; OLS regressions report Newey–West

standard errors allowing for serial correlation and conditional heteroscedasticity.�Statistically significant at the 5% level.��Statistically significant at the 1% level.


taxable income, investors in general tend to place greater importance incapital gains and thus under react to dividends.

6. CONCLUSION

All four tests show evidence of stock market bubbles to a certain extent duringthe sample period analysed. This indicates that as an emerging market economy,Malaysia is not free from bubble-like tendencies that have plagued the stockmarket. Nevertheless, from a graphical relationship between fundamental priceswhichever way you define it and actual prices, the KLCI did in fact convergeto its fundamental values after each stock market crash be it the crash of 1981,1987, 1997, and the year 2000. A comparison of fundamental prices and actualvalues also reveals the prolonged stock market bubble of the early 1990s rightup to the 1997 Asian financial crisis. All other observed deviations betweenactual prices and fundamental prices pale in comparison with this bubble.

Were the excesses of this stock market bubble the cause of meltdownduring the 1997 Asian financial crisis? The question is left for future

0

1

2

3

4

5

1978 1981 1984 1987 1990 1993 1996 1999 2002

Div

iden

d yi

eld

Fig. 4. Monthly Dividend Yield for KLCI.


researchers to answer. As pointed out earlier, stock market bubbles canpropagate should investors be optimistic of the future ahead of them. Onlyby analysing whether the effects of sustained economic growth inmacroeconomic fundamentals did invoke excessive optimism amonginvestors in the Malaysian stock market can we conclude what caused thepropagation of stock price bubbles. After all, before the 1997 Asian financialcrisis, Malaysia was touted by the IMF as one of the Asian ‘‘tigers’’.

Lastly, what is the role of the monetary authorities in controlling the illeffects of bubbles? Although we will not get into the ongoing debate onwhether central banks should prick a bubble in order to deflate it, much canbe done to ensure adequate information is provided for the investor to makea well informed decision on his/her investment. Therefore, informationdissemination should be transparent and although there is alwaystemptation to project the economy and the stock market in a good light,we are of the opinion that sometimes, bad news is also a necessary evil to joltthe markets back into a sober mood.

NOTES

1. Several papers offer a comprehensive review of the criticisms of all tests(Camerer, 1989; Flood and Hodrick, 1990). For more recent reviews, see Campbell(2000) and Gurkaynak (2005).2. See Heaney (2004) the technical details of deriving the trend factor, de-trended

dividends and de-trended prices, and present value prices.

ACKNOWLEDGEMENTS

We would like to thank Richard Heaney of RMIT for helping us with theformat used in estimations of the volatility tests. We also thank Kenneth A.Froot of Harvard Business School and Ma Yue of Lingnan University,Hong Kong for insights into the co-efficient of the intrinsic bubbles modeland its general methodology.

REFERENCES

Ahmed, E., Rosser, J. B., Jr., & Uppal, J. J. (1999). Evidence of nonlinear speculative bubbles in

Pacific-Rim stock markets. Quarterly Review of Economics and Finance, 39, 21–36.


Alexander, A. J. (1997). Asset prices in Japan: The bubble and its breaking. Japan Economic

Institute Report. Japan Economic Institute, Washington, DC.

Athukorala, P. (1998). Swimming against the tide: Crisis management in Malaysia. ASEAN

Economic Bulletin, 15, 28–40.

Blanchard, O. J. (1979). Speculative bubbles, crashes and rational expectations. Economics

Letters, 3, 387–389.

Blanchard, O. J., &Watson, M.W. (1982). Bubbles, rational expectations and financial markets.

In: P. Wachtel (Ed.), Crises in the economic and financial structure (pp. 295–316).

Lexington, MA: Lexington Books.

Bohl, M. T. (2003). Periodically collapsing bubbles in the US stock market? International

Review of Economics and Finance, 12, 385–397.

Brooks, C., & Katsaris, A. (2003). Rational speculative bubbles: An empirical investigation of

the London Stock Exchange. Bulletin of Economic Research, 55, 319–346.

Camerer, C. (1989). Bubbles and fads in asset prices. Journal of Economic Surveys, 3, 3–41.

Campbell, J. Y. (2000). Asset pricing at the millennium. Journal of Finance, 55, 1515–1567.

Campbell, J. Y., & Shiller, R. J. (1987). Cointegration and tests of present value models. Journal

of Political Economy, 95, 1062–1088.

Chan, K., McQueen, G., & Thorley, S. (1998). Are there rational speculative bubbles in Asian

stock markets? Pacific-Basin Finance Journal, 6, 125–151.

Chung, H. T., & Lee, B. S. (1998). Fundamental and nonfundamental components in stock

prices of Pacific-Rim countries. Pacific-Basin Finance Journal, 6, 321–346.

Craine, R. (1993). Rational bubbles: A test. Journal of Economic Dynamics and Control, 17, 829–846.

Diba, B. T., & Grossman, H. I. (1988a). Explosive rational bubbles in stock prices? American


Diba, B. T., & Grossman, H. I. (1988b). The theory of rational bubbles in stock prices.

Economic Journal, 98, 746–754.

Flood, R. P., & Hodrick, R. J. (1990). On testing for speculative bubbles. Journal of Economic

Perspectives, 4, 85–101.

Froot, K. A., & Obstfeld, M. (1991). Intrinsic bubbles: The case of stock prices. American


Galbraith, J. K. (1971). The great crash 1929. London, UK: Andre Deutsch Limited.

Gilles, C., & LeRoy, S. F. (1992). Bubbles and charges. International Economic Review, 33,

323–339.

Gordon, M. (1962). The invesment, financing, and valuation of the corporation. Homewood, IL:

Irwin.

Gurkaynak, R. S. (2005). Econometric tests of asset price bubbles: Taking stock. Finance and

Economics Discussion Series. Washington, DC: The Federal Reserve Board.

Heaney, R. (2004). Excess volatility? The Australian stock market from 1883 to 1999.

Managerial Finance, 30, 76–94.

Herrera, S., & Perry, G. E. (2003). Tropical bubbles: Asset prices in Latin America, 1980–2001.

In: W. C. Hunter, G. G. Kaufman & M. Pomerleano (Eds), Asset price bubbles:

The implications for monetary, regulatory, and international policies (pp. 127–162).

Cambridge, MA: The MIT Press.

Ibrahim, F. W., & Abdul Rahman, R. (2003). Speculative influences in the stock market: A case

study of Kuala Lumpur stock exchange. Capital Markets Review, 11, 41–63.

Kelleher, D., Kim, G. S., & Kim, S. (2001). Evidence of bubbles in the Korean stock markets.

European Financial Management Association, Lugano, Switzerland.


Kindleberger, C. P. (2000). Manias, panics, and crashes: A history of financial crises. New York,

NY: Wiley.

Lai, M. M., Low, K. L. T., & Lai, M. L. (2001). Are Malaysian investors rational? Journal of

Psychology and Financial Markets, 2, 210–215.

LeRoy, S. F., & Porter, R. D. (1981). The present-value relation: Tests based on implied

variance bounds. Econometrica, 49, 555–574.

Ma, Y., & Kanas, A. (2004). Intrinsic bubbles revisited: Evidence from nonlinear cointegration

and forecasting. Journal of Forecasting, 23, 237–250.

Mankiw, N. G., Romer, D., & Shapiro, M. D. (1985). An unbiased reexamination of stock

market volatility. Journal of Finance, 40, 677–687.

Marsh, T. A., & Merton, R. C. (1986). Dividend variability and variance bounds tests for the

rationality of stock market prices. American Economic Review, 76, 483–498.

McDonald, J. B., McQueen, G., & Thorley, S. (1995). Testing for duration dependence with

discrete data. Brigham Young University, Provo, UT.

McQueen, G., & Thorley, S. (1994). Bubbles, stock returns, and duration dependence. Journal

of Financial and Quantitative Analysis, 29, 379–401.

Mishkin, F. S., & White, E. N. (2003). US stock market crashes and their aftermath:

Implications for monetary policy. In: W. C. Hunter, G. G. Kaufman & M. Pomerleano

(Eds), Asset price bubbles: The implications for monetary, regulatory, and international

policies (pp. 53–79). Cambridge, MA: The MIT Press.

Neoh, S. K. (1989). Stock market investment in Malaysia and Singapore. Kuala Lumpur,

Malaysia: Berita Publishing.

Pratten, C. F. (1993). The stock market. Cambridge, UK: Cambridge University Press.

Rangel, G. J., & Pillay, S. S. (2007). Evidence of bubbles in the Singaporean stock market.

Singapore Economic Review Conference, Singapore.

Sarno, L., & Taylor, M. P. (1999). Moral hazard, asset price bubbles, capital flows, and the East

Asian crisis: The first tests. Journal of International Money and Finance, 18, 637–657.

Shiller, R. J. (1981). Do stock prices move too much to be justified by subsequent changes in

dividends? American Economic Review, 71, 421–436.

Shiller, R. J. (1992). Market volatility. Cambridge, MA: The MIT Press.

Shiller, R. J. (2005). Irrational exuberance. Princeton, NJ: Princetress.

Taylor, M. P., & Peel, D. A. (1998). Periodically collapsing stock price bubbles: A robust test.

Economics Letters, 61, 221–228.

Wu, G., & Xiao, Z. (2004). Are there speculative bubbles in stock markets? Evidence from an

alternative approach. Ross School of Business. University of Michigan.


PART IV:

STOCK MARKETS

CHAPTER 10

ABNORMAL RETURNS AFTER

LARGE STOCK PRICE CHANGES:

EVIDENCE FROM ASIA-PACIFIC

MARKETS

Vu Thang Long Pham, Do Quoc Tho Nguyen and

Thuy-Duong To

ABSTRACT

This chapter aims to expand the overreaction literature by examining

whether the price reversals occur in the short-term period (i.e., 3 days) and

long-term period (i.e., up to 20 days), following large 1-day price changes

in Asia-Pacific markets over the period 2001–2005. Our results based on

firm data in three Asia-Pacific markets, namely, Australia, Japan, and

Vietnam, and static and dynamic measures of large price changes indicate

the followings. First, stock prices tend to reverse over the short-term period

after large price changes. Second, in the case of large price declines defined

by arbitrary trigger values, investors may earn profit from exploiting the

phenomena of price reversals; however, the profit is not large enough to

exploit since it is less than the profit from passive funds. This result is

supportive of the weak form of efficient market hypothesis. Third, we find

mixed evidence of long run price reversal across markets. Forth, market

conditions (i.e., bear or bull) may not explain the magnitude of price





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00010-6

205

dx.doi.org/10.1016/S1569-3767(07)00010-6.3d

reversals. Finally, the dynamic measures of large price changes based on

individual firms provide more consistent evidence across markets, which is

supportive of short-term price reversals and overreaction hypothesis. This

evidence exists in the emerging market of Vietnam as well as developed

Australian and Japanese markets.

1. INTRODUCTION

Fama (1970) introduced well-known efficient market hypothesis (EMH) withthree forms: strong form, semi-strong form, and weak form. Stock prices atany time fully reflect all available information in the strong form, all publicavailable information in the semi-strong form and historical information inthe weak form of EMH. Two important implications of EMH are that futurestock prices are unpredictable and expected stock returns can only bedetermined by rational asset pricing models. Evidence from empirical studieshas suggested that stock prices do not always accurately reflect availableinformation. In particularly, research in experimental psychology hassuggested that ‘‘most people overreact to unexpected and dramatic news’’(De Bondt & Thaler, 1985). Motivated by this, De Bondt and Thaler (1985)develop the overreaction hypothesis that suggests: ‘‘Extreme movements in

stock prices will be followed by subsequent price movements in the opposite

direction’’, and ‘‘The more extreme the initial price movement, the greater will

be the subsequent adjustment’’ (De Bondt & Thaler, 1985). The overreactionhypothesis implies a violation of weak form of EMH, i.e., future stock pricescannot be predicted from past stock prices. A great number of studies haveassessed whether short-term price movements in the opposite direction orprice reversals occur following 1-day extreme price movement. Some notablechapters in this area include Brown, Harlow, and Tinic (1988, 1993), Atkinsand Dyl (1990), Bremer and Sweeney (1991, 1996), Cox and Peterson (1994),Park (1995), and Bremer, Hiraki, and Sweeney (1997). Stock prices aregenerally found to be reversed following 1-day sharp declines. Exceptionalcase is Cox and Peterson (1994), whose study finds that abnormal returnsof longer term (4–20 days) are negative after a large 1-day decline. Besidesoverreaction, two other factors may explain short-term price reversalsinclude bid-ask bounce, i.e., a shift from bid to ask or ask to bid maypartially account for the reversal patterns (see e.g., Atkins & Dyl, 1990;Park, 1995) and the level of market liquidity, i.e., the more the liquidmarkets the weaker the reversals (see e.g., Bremer & Sweeney, 1991; Cox &Peterson, 1994).

VU THANG LONG PHAM ET AL.206

This chapter aims to expand the overreaction literature by examiningwhether the price reversals occur in the short-term period, i.e., 3 days, andlong-term period, i.e., up to 20 days following the large 1-day price changesin Asia-Pacific markets over the period 2001–2005. The sample consists ofall firms included in three major indices, i.e., 100 firms traded on theAustralian Securities Exchange (ASX) and comprising the ASX 100 index,300 firms traded on the Tokyo Stock Exchange (TSE) and included in theNikkei 300 and finally, 33 firms trading on the Ho Chi Minh City SecuritiesTrading Center (HCMC STC) and comprising the VN-INDEX as of 2005.As far as Asia-Pacific markets are concerned, Wong (1997) examinedmarket indexes such as All Ordinaries Index and Nikkei 225 Index. He findsthat stock prices tend to rise (fall) after a large 1-day increase (decrease),which is inconsistent with Debondt and Thaler’s overreaction hypothesis.

Our study fits into the literature in at least two ways. First, we use dailydata at firm level of three Asia-Pacific markets with different levels ofdevelopment, where Australia and Japan are two major and advancedmarkets and Vietnam represents a new emerging market in the region. In thisregards, we seek to provide new evidence on the price reversal hypothesis forthe same markets as well as across countries, and to answer the question ofwhether the price patterns following large price changes differ from countryto country. If the reversal patterns on the HCMC STC exist and are similarto those on ASX and the TSE, as suggested by Bremer et al. (1997), thepatterns of price reversals may result from the fundamental behavior ofinvestors, regardless of the institutional features, which are different acrossthe markets. Second, our approach of estimating expected returnsdistinguishes itself from exiting ones in a major way. This study appliesthe method developed by MacKinlay and Richardson (1991) using genera-lized method of moments (GMM) to estimate the expected stock returnsdescribed by the CAPM. GMM is chosen for its many advantages. It is ageneral estimator that encompasses many standard econometric estimatorsincluding ordinary least square (OLS), instrumental variables (IVs), andmaximum likelihood (ML). Not only that GMM is valid under weakerassumptions about the normality of data distribution, but it also has thepotential to improve the estimation since it allows serially correlatedresiduals and conditional dependency of residuals on the factors. Our resultsindicate short-term reversal patterns across Asia-Pacific markets.

The remainder of this chapter is organized as follows. Section 2 describesthe dataset and reviews the methodology for analyzing abnormal returns.In Section 3, the empirical results are presented. Section 4 provides economicimplications of the empirical findings, and Section 5 concludes the chapter.

Abnormal Returns After Large Stock Price Changes 207

2. DATA AND METHODOLOGY

2.1. Data

In this chapter, we use daily data at firm level as described in Table 1. Thestudy investigates the behavior of stock returns on days after a large pricechange occurred for firms of three different markets in the Asia-Pacific region.The daily returns are computed based on the closing price of each trading

Table 1. Description of Dataset.

Market Data Data Source

Australia 100 stocks included in the

ASX100, which accounts for

approximately 87% of the total

market capitalization of the All

Ordinaries Index.

Datastream

A proxy for market index, i.e., All

Ordinaries Index, which is

Australia’s premier market

indicator representing the

weighed value of 500 largest

firms listed on the Australian

Stock Exchange (ASX).

Datastream

A proxy for risk free rate, i.e.,

equivalent rate of return on

Australia Interbank 3-month.

Datastream

Japan 300 firms included in the Nikkei

300.

The Nikkei Economic Electronic

Databank System (NEEDS)

A market proxy, i.e., Tokyo

Stock Exchange Price Index

(TOPIX).

NEEDS

A proxy for risk free rate, i.e., the

overnight Tokyo call rate.

NEEDS

Vietnam 33 stocks listed on the HCMC

STC, as at the end of

December 2005.

The Bank for Investment and

Development of Vietnam

(BIDV) Securities Co., Ltd.

A proxy for market index, i.e.,

VN-INDEX, which is

calculated base on weighed

value of all stocks traded on

the HCMC STC.

BIDV Securities Co., Ltd.

A proxy for risk free rate, i.e.,

equivalent daily rate of return

on 1-year Treasury Bill.

International Financial Statistics

(IFS) database provided by the

International Monetary Fund

(IMF)


day. If two successive closing prices are not available, the daily returns are notrecorded. The sample period extends from January 2001 to December 2005.

Table 2 shows the summary statistics of the daily returns. Among the threemarkets, Australian and Vietnamese stock returns have smaller standarddeviations compared to returns on Japanese stocks.

2.2. Methodology

Our approach to investigate whether there are predictable patterns of stockreturns after large price changes is similar to that of Brown et al. (1993),Atkins and Dyl (1990), Bremer and Sweeney (1991), Cox and Peterson(1994), Park (1995), Bremer et al. (1997), and Wong (1997). Two measuresfor large price changes are examined.

The first measure considers all daily rates of returns that were greater/lessthan or equal to pre-specify trigger values over the 5-year period fromJanuary 2001 to December 2005 as ‘‘large’’ price increase/decrease events.Consistent with Bremer and Sweeney (1991), Cox and Peterson (1994),Park (1995), and Bremer et al. (1997), the trigger values are (+/�) 10% forAustralian and Japanese markets. For Vietnamese market, the triggervalues are lower at (+/�) 4% since HCMC STC applies narrow daily pricelimits of (+/�) 5%.

As of Wong (1997), the second measure utilizes dynamic trigger valuesbased upon firm’s expected return and volatility. Define all daily rates ofreturns that were significantly greater/less than their sample mean returnsat around 2.5% level as large price increases (Rit4mi þ 2si)/decreases(Ritomi � 2si). Sample mean return mi and sample standard deviation si ofstock i are estimated over the period from 2001 to 2005.

As explained by Atkins and Dyl (1990) and Bremer et al. (1997), large1-day price changes are primarily caused by unexpected, new firm specificor market wide information pertinent to the value of the stock such as

Table 2. Summary Statistics of Daily Returns in the Sample(2001–2005).

Country Sample Size Mean (%) Standard

Deviation (%)

Minimum (%) Maximum (%)

Australia 114,770 0.0718 1.6918 �39.0244 26.4009

Japan 355,632 0.0626 2.3082 �33.3333 35.7143

Vietnam 23,155 0.0346 1.7030 �7.0000 7.0000


unexpected operating results, or unanticipated government decisions likea change in the level of interest rate, etc. These events provide goodopportunity to assess whether stock prices fairly reflect unexpected, newinformation or overreact to such information.

Following Bremer and Sweeney (1991) and Cox and Peterson (1994), onlyone event per day is allowed in order to minimize correlation across samples.The event observations are ordered first by date and then alphabetically bystock name. If a date has more than one event, only the observation appearingfirst in the ordering sort for that date is retained. Table 3 shows the number oflarge price changes across three markets measured using pre-specified triggervalues and dynamic trigger values. Event samples ‘‘3’’ in the table indicate thefinal event samples appropriate for analyzing abnormal returns.

In order to investigate whether abnormal returns are present, the returnson days following a large price change are compared to expected return,estimated using unrestricted CAPM model and market model, via popularGMM method. We apply a standard event study approach, which is similarto that of MacKinlay (1997) to calculate abnormal returns as follows.

The abnormal return ARi;t is estimated for a 41-day event windowcomprised of 20 pre-event days and 20 post-event days, by deducting realizedrate of return from the estimated expected return,

ARi;t ¼ ~Ri;t � Eð ~Ri;tÞ (1)

Table 3. Number of Large Price Changes over the Period 2001–2005.

Event Sample 1 2 3 1 2 3

Large price

declines

AU, JP: R0r�10%; VN: R0r�4% Ritomi � 2si

Australia 77 72 68 2451 837 789

Japan 373 192 185 8099 873 838

Vietnam 584 249 202 779 334 269

Large price

increases

AU, JP: R0Z10%; VN: R0Z4% Rit4mi þ 2si

Australia 93 84 80 2848 947 869

Japan 787 380 356 10547 1041 986

Vietnam 818 282 234 991 341 286

N.B.

Event Sample 1: Initial event sample including all large price change events for all stocks.

Event Sample 2: Event sample with only one event per day.

Event Sample 3: Final event sample with only one event per day and post-event estimation

periods of at least 10 trading days.


where ~Ri;t, Eð ~Ri;tÞ are the realized rate of return and expected return forstock i on day t, respectively.

Under the unrestricted CAPM,

ARi;t ¼ ~Ri;t � Rf � ai � bið ~RM;t � Rf Þ (2)

where ai, bi are the simple averages of the ai intercepts and bi coefficients,respectively, from following GMM regression equations estimated over twoperiods: 120–21 days before the event day (pre-event), and 21–120 day afterthe event day t=0 (post-event):

~Ri;t � Rf ¼ ai þ bið ~RM;t � Rf Þ þ ~�i;t t 2 pre-event; post-event (3)

The GMM approach applied here is relatively similar to that of MacKinlayand Richardson (1991). There are two sample moments 1=T

PTt¼1 ~�i;t;

1=TPT

t¼1 ~�i;tð ~RM;t � Rf Þ, and two parameters ai, bi to be estimated for eachstock. Therefore, the moment condition in Eq. (3) is exactly identified, andthe associated Hansen’s (1982) J-statistic is zero.

Under the market model,

ARi;t ¼ ~Ri;t � ai � bið ~RM;tÞ (4)

where ai, bi are the simple averages of the ai intercepts and bi coefficients,respectively, from following GMM regression equation:

~Ri;t ¼ ai þ bið ~RM;tÞ þ ~�i;t t 2 pre-event; post-event (5)

Similarly, two sample moments 1=TPT

t¼1 ~�i;t; 1=TPT

t¼1 ~�i;tð ~RM;tÞ exactlyidentify two parameters ai, bi for each stock. Therefore, the associatedHansen’s (1982) J-statistic is zero.

If less than 100 days of returns are available during the post-eventestimation periods, bi are estimated using however many days of returns areavailable, provided there are at least 10.

The mean abnormal return across event observations on day t denoted asARt is the sum of individual abnormal returns on day t divided by thenumber of events,

ARt ¼1

N

XN

i¼1

ARi;t (6)

where N is the number of events.The cumulative abnormal return for stock i from day t1 to day t2

denoted as CARiðt1; t2Þ is simply summed daily abnormal returns over day


t1 to day t2,

CARiðt1; t2Þ ¼Xt2t¼t1

ARi;t (7)

The sample average cumulative abnormal return across event observationsfrom day t1 to day t2 denoted as CARðt1; t2Þ is the sum of mean abnormalreturn over day t1 to day t2,

CARðt1; t2Þ ¼Xt2t¼t1

ARt ¼1

N

XN

i¼1

CARiðt1; t2Þ (8)

Let y1 denote the cross-sectional test statistic (t-statistic) based on thehypothesis, which asserts that:

H0. Expected abnormal return is zero for each stock for each day t.

H1. Expected abnormal return is different from zero for each stock foreach day t.

Let y2 denote the test statistic (t-statistic) on the basis of the hypothesis,which asserts that:

H0. Expected cumulative abnormal return is zero for each stock for dayt1 to day t2.

H1. Expected cumulative abnormal return is different from zero for eachstock for day t1 to day t2.

Then the value of y1 and y2 are calculated based on cross-sectionalvariances as,

y1 ¼ARt

varðARtÞ1=2� Nð0; 1Þ (9)

y2 ¼CARðt1; t2Þ

varðCARðt1; t2ÞÞ1=2� Nð0; 1Þ (10)

where varðARtÞ is the cross-sectional variance of abnormal returns onday t, whereas varCARðt1; t2Þ is the cross-sectional variance of cumulativeabnormal returns from day t1 to day t2. These values of test statistics for theevent day (t=0) and subsequent days provide evidence on whether statisticallysignificant price reversals exist. This is discussed in the next section.


3. EMPIRICAL FINDINGS

3.1. Abnormal Returns Following Large Price Changes that are

Greater/Less than Pre-Specified Trigger Values

Table 4 presents the frequency of price continuations and price reversalsoccurring on days 1, 2, and 3 and days 1–3, 1–5, 1–10, and 1–20 followingthe day of large price changes. The triggers values are (+/�) 10% for theAustralian and Japanese markets and 4% for the Vietnamese market. Ingeneral, the results across the three markets indicate higher frequency ofprice reversals than of price continuations over the short-term period, i.e.,3 days following the large price changes.

Fig. 1 plots the average cumulative abnormal returns around 20 days of10% or greater price decrease, calculated using unrestricted CAPM model.The figure shows that cumulative abnormal returns keep falling on day 1after the large price decrease in the Australian and Japanese markets. Inthese two markets, price reversals occur on days 2 and 3. Price reversalsoccur immediately on day 1 and continue to day 3 after large price decline inthe Vietnamese market. The figure also shows that price reversals persist inlonger term up to 20 days in Australia, whereas price tends to decline overthe 20 days period in Vietnam and there is no clear return pattern during 20days following large price decrease in Japan.

Table 5 presents mean abnormal returns of large stock price decreases andincreases over the 5-year period from January 2001 to December 2005,computed using the two methods described in the previous section.

As shown in Panel A of Table 5, the mean abnormal returns are �14.6807,�12.2346, and �4.1637% on the day of large price decrease (i.e., day 0) inthe Australian, Japanese, and Vietnamese markets, respectively, measuredusing the unrestricted CAPM model, and of similar magnitude using themarket model. The mean abnormal returns are positive for day 2 and day 3of the 3 trading days following the day of the large price decrease in Australiaand Japan, and positive for these 3 days in Vietnam. Significantly positivemean abnormal returns are observed on day 2 in Australia and day 1 inVietnam, as indicated by two-tailed test. The clearest evidence of short-termprice reversal is seen in the Vietnamese market where average cumulativeabnormal return for 3 days following large price decline (CAR1–3) is0.5727% and statistically significant. CAR1–3 in the Australian andJapanese markets are 0.9020 and �0.0281%, respectively, and both arestatistically insignificant. With respect to longer term, the average cumulativeabnormal returns over days 1 through 20 (CAR1–20) are 2.6399, �0.2279,


and �1.0717%, respectively in the three markets. We interpret these cross-country results as mixed evidence on the price reversal and overreactionhypothesis over the long-term period.

Fig. 2 plots the average cumulative abnormal returns around 20 days of10% or greater price increase. The abnormal returns are calculated usingunrestricted CAPM model. Following the rise on event day, cumulative

Table 4. Distribution of Price Continuations and Reversals(2001–2005).

Panel A: Frequency of price continuations and price reversals after a large 1-day decline:

continuation: ARo0, reversal: AR>0; continuation: CARo0, reversal: CAR>0

Australia (N=68) Japan (N=185) Vietnam (N=202)

(ARo0) (AR>0) (ARo0) (AR>0) (ARo0) (AR>0)

Day 1 0.500 0.500 0.486 0.514 0.446 0.554

Day 2 0.382 0.618 0.454 0.546 0.490 0.510

Day 3 0.471 0.529 0.497 0.503 0.421 0.579

(CARo0) (CAR>0) (CARo0) (CAR>0) (CARo0) (CAR>0)

Days 1–3 0.529 0.471 0.497 0.503 0.386 0.614

Days 1–5 0.441 0.559 0.541 0.459 0.396 0.604

Days 1–10 0.544 0.456 0.530 0.470 0.505 0.495

Days 1–20 0.529 0.471 0.546 0.454 0.475 0.525

Panel B: Frequency of price continuations and price reversals after a large 1-day advance:

continuation: AR>0, reversal: ARo0; continuation: CAR>0, reversal: CARo0


(AR>0) (ARo0) (AR>0) (ARo0) (AR>0) (ARo0)

Day 1 0.400 0.600 0.368 0.632 0.517 0.483

Day 2 0.475 0.525 0.430 0.570 0.470 0.530

Day 3 0.562 0.438 0.449 0.551 0.521 0.479

(CAR>0) (CARo0) (CAR>0) (CARo0) (CAR>0) (CARo0)

Days 1–3 0.462 0.538 0.388 0.612 0.402 0.598

Days 1–5 0.488 0.512 0.390 0.610 0.440 0.560

Days 1–10 0.550 0.450 0.483 0.517 0.547 0.453

Days 1–20 0.462 0.538 0.444 0.556 0.551 0.449

Large price changes are measured using pre-specify trigger values.

Large 1-day declines (AU, JP: R0r�10%; VN: R0r�4%).

Large 1-day advances (AU, JP: R0Z10%; VN: R0Z4%).


abnormal returns decrease on days 1 and 2; days 1, 2, and 3; and days 2 and 3in the Australian, Japanese, and Vietnamese markets, respectively. Over thelonger term, stock price increases significantly in Australia, whereas declineslightly in Japan and Vietnam over the 20 days period following the largeprice advance.

Panel B of Table 5 shows that the mean abnormal returns are 13.8499,11.9320, and 3.9259% on the day of the large price advance in theAustralian, Japanese, and Vietnamese markets, respectively, measured usingthe unrestricted CAPM model. The mean abnormal returns are negative forday 1 and day 2 in the Australian market; day 2 and day 3 in the Vietnamesemarket; and 3 trading days following the initial large price increase in theJapanese market. Significantly negative mean abnormal returns are evidenton day 1 and day 2 in Japan market. Among the three markets, Japan showsstrongest evidence of short-term price reversal where total abnormal returnover days 1–3 (CAR1–3) is �1.1187% and statistically significant. Over thelonger term, price reversals are evident in the Japanese market withsignificant and negative CAR1–5 and in the Vietnamese market withsignificant and negative CAR1–5 and CAR1–10. In contrast, there is no

-18

-16

-14

-12

-10

-8

-6

-4

-2

0

CA

R (

%)

1 2 3 4-20 -15 -10 -5 0 5 10 15 20Event Time

AU JPVN

Fig. 1. Cumulative Abnormal Returns for Stocks that Exhibited a Large Decline in

Price at Day 0 (AU, JP: R0r�10%; VN: R0r�4%).


Table 5. Abnormal Returns and Cumulative Abnormal Returns After a Large 1-Day Decline or Advance(2001–2005).

Country Un-restricted CAPM Market Model

Australia Japan Vietnam Australia Japan Vietnam

Panel A: Large price declines (AU, JP: R0r�10%; VN: R0r�4%)

Sample Size N=68 N=185 N=202 N=68 N=185 N=202

AR0 �14.6807% �12.2346% �4.1637% �14.6807% �12.2346% �4.1633%

(�18.10)�� (�38.00)�� (�49.71)�� (�18.10)�� (�38.00)�� (�49.75)��

AR1 �0.6520% �0.2776% 0.3354% �0.6520% �0.2776% 0.3357%

(�1.03) �(0.64) (1.68)� (�1.03) (�0.64) (1.69)�

AR2 1.0255% 0.0810% 0.1016% 1.0254% 0.0810% 0.1012%

(2.06)�� (0.22) (0.61) (2.06)�� (0.22) (0.60)

AR3 0.5286% 0.1685% 0.1357% 0.5286% 0.1685% 0.1355%

(1.22) (0.58) (0.81) (1.22) (0.58) (0.81)

CAR1–3 0.9020% �0.0281% 0.5727% 0.9020% �0.0281% 0.5725%

(1.36) (�0.06) (2.29)�� (1.36) (�0.06) (2.29)

CAR1–5 0.7107% �0.6668% 0.3971% 0.7106% �0.6668% 0.3968%

(1.04) (�1.24) (1.41) (1.04) (�1.24) (1.41)

CAR1–10 1.8618% �1.1912% 0.0232% 1.8617% �1.1912% 0.0220%

(2.12)�� (�1.70)� (0.05) (2.12)�� (�1.70)� 0.05)

CAR1–20 2.6399% �0.2279% �1.0717% 2.6397% �0.2279% �1.0729%

(2.09)�� (�0.22) (�1.53) (2.09)�� (�0.22) (�1.53)

VU

THANG

LONG

PHAM

ET

AL.

216

Panel B: Large price increases (AU, JP: R0Z10%; VN: R0Z4%)

Sample Size N=80 N=356 N=234 N=80 N=356 N=234

AR0 13.8499% 11.9320% 3.9259% 13.8498% 11.9320% 3.9257%

(30.61)�� (58.05)�� (45.58)�� (30.61)�� (58.05)�� (45.58)��

AR1 �0.2400% �0.6567% 0.0298% �0.2401% �0.6567% 0.0296%

(�0.40) (�2.25)�� (0.18) (�0.40) (�2.25)�� (0.18)

AR2 �0.0434% �0.3896% �0.2018% �0.0436% �0.3896% �0.2019%

(�0.12) (�1.94)� (�1.32) (�0.12) (�1.94)� (�1.32)

AR3 0.5957% �0.0725% �0.1556% 0.5955% �0.0725% �0.1558%

(1.72)� (�0.39) (�1.01) (1.72)� (�0.39) (�1.01)

CAR1–3 0.3122 �1.1187 �0.3277 0.3118 �1.1187 �0.3280

(0.59) (�4.14)�� (�1.38) (0.59) (�4.14)�� (�1.38)

CAR1–5 1.4803% �0.6987% �0.4752% 1.4796% �0.6987% �0.4758%

(2.29)�� (�2.13)�� (�1.76)� (2.29)�� (�2.13)�� (�1.76)�

CAR1–10 0.8630% �0.5129% �0.6441% 0.8614% �0.5129% �0.6452%

(1.05) (�1.18) (�1.85)� (1.05) (�1.18) (�1.85)�

CAR1–20 3.5471% �0.6438% �0.6216% 3.5437% �0.6439% �0.6236%

(2.90)�� (�0.99) (�1.06) (2.89)�� (�0.99) (�1.07)

Large price changes are measured using pre-specify trigger values.

R0: Return on a large 1-day decline or advance; AR0: abnormal return on a large 1-day decline or advance; AR1, AR2, AR3, AR4, AR5:

abnormal returns on days 1, 2, 3, 4, 5 after a large 1-day decline or advance; CAR1–3, CAR1–5, CAR1–10, CAR1–20: 3-, 5-, 10-, and 20-day

cumulative abnormal returns after a large 1-day decline or advance; cross-sectional t-values in parentheses.�Significantly different from 0 at the 0.10 level (two-tailed test).��Significantly different from 0 at the 0.05 level (two-tailed test).��Significantly different from 0 at the 0.01 level (two-tailed test).

Ab

no

rma

lR

eturn

sA

fterL

arg

eS

tock

Price

Ch

an

ges

217

evidence in support of long-term price reversal in the Australian market. Thismarket shows significant price continuation, i.e., cumulative abnormal returnof 3.5471% over 20 days period after large price increase.

3.2. Abnormal Returns Following Large Price Changes that are

Greater/Less than Dynamic Trigger Values

Table 6 presents the frequency of price continuations and price reversalsoccurring on days 1, 2, and 3 and days 1–3, 1–5, 1–10, and 1–20 followingthe day of large price increases (Rit4mi þ 2si)/decreases (Ritomi � 2si).Similar to results from pre-specified trigger values, the results based ondynamic trigger values across the three markets show higher frequency ofprice reversals than of price continuations over the short-term period.

Fig. 3 shows the average cumulative abnormal returns for the period from20 days before to 20 days after the day of initial price decline, calculatedusing unrestricted CAPM model. The figure shows clear short-term patternsof price reversals across the three markets. Cumulative abnormal returns

-4

-2

0

2

4

6

8

10

12

14

16

CA

R (

%)

-20 -15 -10 -5 0 5 10 15 20

Event Time

1 2 3 4

AU JPVN

Fig. 2. Cumulative Abnormal Returns for Stocks that Exhibited a Large Advance

in Price at Day 0 (AU, JP: R0Z10%; VN: R0Z4%).


rise immediately from day 1 to day 3 in Australia and Vietnam and on day 1and day 2 in Japan. The price patterns from day 4 through day 20 quitediffer from market to market.

As shown in Panel A of Table 7, the mean abnormal returns are �4.4453,�5.5064, and �3.7591% on the day of large price decrease (i.e., day 0) in theAustralian, Japanese, and Vietnamese markets respectively, measured using

Table 6. Distribution of Price Continuations and Reversals(2001–2005).

Panel A: Frequency of price continuations and price reversals after a large 1-day decline:

continuation: ARo0, reversal: AR>0; continuation: CARo0, reversal: CAR>0


(ARo0) (AR>0) (ARo0) (AR>0) (ARo0) (AR>0)

Day 1 0.447 0.553 0.453 0.547 0.398 0.602

Day 2 0.465 0.535 0.507 0.493 0.502 0.498

Day 3 0.525 0.475 0.506 0.494 0.465 0.535

(CARo0) (CAR>0) (CARo0) (CAR>0) (CARo0) (CAR>0)

Days 1–3 0.435 0.565 0.469 0.531 0.375 0.625

Days 1–5 0.450 0.550 0.498 0.502 0.383 0.617

Days 1–10 0.511 0.489 0.533 0.467 0.535 0.465

Days 1–20 0.540 0.460 0.519 0.481 0.480 0.520

Panel B: Frequency of price continuations and price reversals after a large 1-day advance:

continuation: AR>0, reversal: ARo0; continuation: CAR>0, reversal: CARo0


(AR>0) (ARo0) (AR>0) (ARo0) (AR>0) (ARo0)

Day 1 0.446 0.554 0.399 0.601 0.493 0.507

Day 2 0.449 0.551 0.453 0.547 0.451 0.549

Day 3 0.481 0.519 0.468 0.532 0.493 0.507

(CAR>0) (CARo0) (CAR>0) (CARo0) (CAR>0) (CARo0)

Days 1–3 0.426 0.574 0.386 0.614 0.388 0.612

Days 1–5 0.455 0.545 0.418 0.582 0.399 0.601

Days 1–10 0.486 0.514 0.476 0.524 0.528 0.472

Days 1–20 0.510 0.490 0.466 0.534 0.542 0.458

Large price changes are measured using dynamic trigger values.

Large returns are significantly different from sample mean value at 2.5% level.


the unrestricted CAPM model. Average cumulative abnormal return for 3days following large price decline (CAR1–3) are positive and significant inthree markets. This evidence on short-term price reversals therefore supportsthe overreaction hypothesis.

Fig. 4 plots the average cumulative abnormal returns around 20 days of10% or greater price increase. Cumulative abnormal returns fall immediatelyfrom day 1 to day 3 in the three markets, indicating clear short-term patternsof price reversals.

Panel B of Table 7 shows that the mean abnormal returns are 4.5862,5.8811, and 3.6710% on the day of the large price advance in the Australian,Japanese, and Vietnamese markets, respectively, measured using the unrest-ricted CAPM model. The mean abnormal returns are negative for days 1, 2,and 3 of the 3 trading days following the initial large price increase in threemarkets. Panel B of Table 7 also shows that total abnormal returns overdays 1–3 (CAR1–3) are significant and negative. These results indicatesignificant short-term price reversals that are supportive of the overreactionhypothesis.

-4

-3

-2

-1

0

1

2

3

CA

R (

%)

1 2 3 4-20 -15 -10 -5 0 5 10 15 20Event Time

AU JPVN

Fig. 3. Cumulative Abnormal Returns for Stocks that Exhibited a Large Decline in

Price at Day 0 (Large Returns are Significantly Different from Sample Mean Value

at 2.5% Level).


Table 7. Abnormal Returns and Cumulative Abnormal Returns after aLarge 1-Day Decline or Advance (2001–2005).

Australia Japan Vietnam

Panel A: Large price declines

Sample Size N=789 N=838 N=269

AR0 �4.4453% �5.5064% �3.7591%

(�40.23)�� (�51.14)�� (�47.85)��

AR1 0.1985% 0.2618% 0.4927%

(2.34)�� (2.64)�� (3.10)��

AR2 0.1750% 0.0439% 0.0162%

(2.47)�� (0.50) (0.12)

AR3 0.0366% �0.0437% 0.0971%

(0.50) (�0.56) (0.74)

CAR1–3 0.4101% 0.2621% 0.6061%

(4.08)�� (2.25)�� (3.09)��

CAR1–5 0.4278% 0.1921% 0.4441%

(3.65)�� (1.42) (1.94)�

CAR1–10 0.7446% 0.1389% 0.1137%

(4.28)�� (0.74) (0.32)

CAR1–20 0.9467% 0.2431% �0.3821%

(3.68)�� (0.83) (�0.68)

Panel B: Large price increases

Sample size N=869 N=986 N=286

AR0 4.5862% 5.8811% 3.6710%

(51.00)�� (64.10)�� (47.03)��

AR1 �0.0353% �0.4282% �0.1523%

(�0.42) (�4.25)�� (�1.10)

AR2 �0.2333% �0.2084% �0.2829%

(�3.46)�� (�2.77)�� (�2.09)��

AR3 �0.0090% �0.1793% �0.1804%

(�0.14) (�2.25)�� (�1.33)

CAR1–3 �0.2776% �0.8158% �0.6157%

(�3.00)�� (�7.84)�� (�3.03)��

CAR1–5 �0.2675% �0.6258% �0.7605%

(�2.45)�� (�4.97)�� (�3.26)��

CAR1–10 �0.2407% �0.8067% �0.7859%

(�1.49) (�4.66)�� (�2.58)��

CAR1–20 �0.0336% �1.1259% �0.7345%

(�0.14) (�4.18)�� (�1.47)��

Large price changes are measured using dynamic trigger values.

R0: Return on a large 1-day decline or advance; AR0: abnormal return on a large 1-day decline

or advance; AR1, AR2, AR3, AR4, AR5: abnormal returns on days 1, 2, 3, 4, 5 after a large

1-day decline or advance; CAR1–3, CAR1–5, CAR1–10, CAR1–20: 3-, 5-, 10-, and 20-day

cumulative abnormal returns after a large 1-day decline or advance; cross-sectional t-values in

parentheses; large returns are significantly different from sample mean value at 2.5% level.�Significantly different from 0 at the 0.10 level (two-tailed test).��Significantly different from 0 at the 0.05 level (two-tailed test).��Significantly different from 0 at the 0.01 level (two-tailed test).


We discussed abnormal return patterns using two measurements of largeprice changes. In general, both measurements provide evidence in supportiveof the price reversals and overreaction hypothesis in the short-term period,i.e., 3 days period. The measure of large price changes based on dynamictrigger values provide stronger evidence of short-term price reversalsthan such measure based on pre-specified trigger values. This suggests thatfirm specific characteristics play a significant role in explaining the pricereversals.

The existence of price reversals, measured by static dynamic trigger values,has some other important implications. First, the reversal patterns areconsistent regardless of market development level, indicating that they may

result from the fundamental behavior of investors rather than institutionalfeatures. In Vietnamese market where there are more unsophisticatedinvestors, the magnitude of price reversals are evidently larger than inAustralian and Japanese market, i.e., CAR1–3s in Vietnamese, Australian,and Japanese markets are equivalent to 16.1% (0.6061/3.7591), 9.2%(0.4101/4.4453), and 4.7% (0.2621/5.5064) of initial price drops, and 16.8%

-1

0

1

2

3

4

5

6

CA

R (

%)

1 2 3 4-20 -15 -10 -5 0 5 10 15 20

Event Time

AU JPVN

Fig. 4. Cumulative Abnormal Returns for Stocks that Exhibited a Large Advance

in Price at Day 0 (Large Returns are Significantly Different from Sample Mean

Value at 2.5% Level).


(0.6157/3.6710), 6.1% (0.2776/4.5862), and 13.9% (0.8158/5.8811) of initialprice increases, respectively.

Secondly, other important factors including bid-ask spread and the level ofmarket liquidity that have been discussed extensively in previous literature(see e.g., Atkins & Dyl, 1990; Bremer & Sweeney, 1991; Cox & Peterson, 1994;Park, 1995) may also explain the price reversals in the three markets. Investorstend to have substantial selling pressure, i.e., more investors want to sellthe stock than to buy it, in response to bad news arrival on the day of initialprice decrease, increasing the probability of transaction at bid price, i.e., priceat which someone is willing to buy. On the day that overreaction to bad newsis completed, i.e., on day 1 in the three markets, investors may realize thatinitial price decrease has actually been excessive. As a consequence, there willbe more buyers than sellers, enhancing the probability of transaction at askprice, i.e., price at which someone is willing to sell. In this case, systematicshifts from trading at bid prices to ask prices may partially account for short-term price reversal. Similar market forces apply when there is excessivereaction to good new arrivals, resulting in systematic shifts from trading atask prices to bid prices, and short-term price reversals will be observed in themarket.

Market liquidity, i.e., the possibility of changing stocks into cash quicklywithout loss, may also explain price reversals in the three markets. Assuggested by Cox and Peterson (1994), if market liquidity plays a significantrole in explaining price reversal, we should observe stronger reversals in lessliquid market, and vice versa. Our finding of larger price reversals in HCMCSTC, a low liquidity market (stocks only have two official transaction pricesper trading day due to the market convention of matching orders only twicea day) are consistent with Cox and Peterson (1994) in that market liquidityis an important factor in price reversal process.

Compared with the extant literature, for the cases of price pattern followinglarger price decreases, our study is consistent with Atkins and Dyl (1990),Bremer and Sweeney (1991), Cox and Peterson (1994), and Bremer et al.(1997), i.e., stock price is short-term reversed after large 1-day price decreases.

Our results based on individual firms are in contrast to Wong’s (1997)findings based on market indices, though utilizing similar dynamic measureof large price changes. For the cases of large price increases, our resultsindicate CAR1–10 and CAR1–20 are negative and significant in Australiaand Japan, whereas in Wong’s chapter, these figures are found to be positivein the same markets. Similarly, for the cases of large price decreases, we findpositive CAR1–10 in Australia and Japan which are associated withnegative values according to Wong’s results.


The differences between results obtained from dynamic measures of largeprice changes and statically pre-specified measures of large price changes,and between results obtained from dynamic measures based on individualfirms and such measures based on overall market indices as discussed abovesuggest that the dynamic measures based on individual firms provide moreconsistent evidence across markets, which is supportive of short-term pricereversals and overreaction hypothesis.

In summary, we have argued that market overreaction, bid-ask spreads,market liquidity, firm specific characteristics, and the behavior of invertors/traders may attribute to the sources of price movements in oppositedirections following large 1-day price changes.

4. ECONOMIC IMPLICATIONS

This section discusses the economic significance, i.e., whether opportunitiesfor investors to earn excess profit from the observed patterns exist. It shouldbe noted first that short selling is prohibited in the ASX and HCMC STC,whereas it is allowed in the TSE. Therefore, investors cannot exploit thereversal patterns following price increases in Australian and Vietnamesemarkets. Second, because the dynamic trigger value is ex post, which isestimated using the whole sample statistics, therefore we cannot really form aportfolio to earn excess profit from the reversal patterns following large pricechanges measured by dynamic trigger values.

As reported in Panel A of Table 5, the average cumulative abnormalreturn for 10 trading days following large price decrease (CAR1–10) is1.8618% in Australia, and CAR1–3 is 0.5727% in Vietnam. These values arestatistically significant which may offer following opportunities for contra-rian investors. Consider two investment strategies for investors in Australianstock market over the 6-year period from January 2000 to December 2005 asfollows.

The first strategy is passive management, i.e., buy and hold the AllOrdinaries Index over 5 years from 2001 to 2005. By following this strategy,the profit before transaction cost is 46.9%, i.e., buying All-Ords at 3205.4 onfirst trading day of 2001 and selling at 4708.8 on last trading day of 2005.

The second strategy is active management. We periodically invest the sameamount that otherwise is invested in passive funds, in any stock whose returnequals to, or falls below the trigger, i.e., 10%, and selling all the same stocks10 trading days later. Over the 5 years period, this strategy can be repeated


45 times out of 68 trading days that the stock price falls bellow the triggervalue. The return on every 1 Australian Dollar of initial investment beforetransaction costs is: ROI45 ¼

Q45i¼1ð1þ riÞ � 1

h i� 100% ¼ 25:3%, with the

cumulative abnormal return for 10 trading days following the initial pricedecrease. Clearly, the profit from active strategy is less than that of passivemanagement.

Similar results are obtained for the Vietnamese market. The evidenceacross countries indicates that although there are evident of price reversals,they are not large enough to exploit. This is consistent with the EMH.

To investigate whether market conditions affect the magnitude of pricereversals, we examine the cumulative abnormal returns following large pricechanges in bear period, i.e., the period that market experiences stockdeclining, and bull period, the period that market experiences stockadvancing. Fig. 5 shows the graphs of market indices of three markets overthe period 2001–2005. The bear periods are from January 2001 toMarch 2003in Australia and Japan and from July 2001 to October 2003 in Vietnam. Theremaining periods are bull periods in those markets. In the Australian market,the magnitude of price reversals following large price decreases in bear periodis larger than that in bull period, i.e., CAR1–10 are 2.7313 and 0.0436% in

100

200

300

400

500

600

VN

-IN

DE

X

25jan2001 09jun2002 22oct2003 05mar2005 18jul2006

Time

25jan2001 09jun2002 22oct2003 05mar2005 18jul2006

Time

25jan2001 09jun2002 22oct2003 05mar2005 18jul2006Time

2500

3000

1600

1400

1200

1000

800

3500

4000

4500

5000

ALL

-OR

DS

TOP

IX

Fig. 5. Market Indices (2001–2005).


bear and bull period, respectively. Whereas, in Vietnamese market, themagnitude of price reversals following large price decreases in bear period isless than that in bull period, i.e., CAR1–3 are 0.4404 and 0.6493% in bear andbull period, respectively. The different results between two markets indicatethat market conditions may not explain the magnitude of price reversals.

5. CONCLUSIONS

This chapter investigates whether there are predictable short-term patternsof stock abnormal returns (i.e., 3 days) and long-term pattern (i.e., up to20 days) following the large 1-day price changes in Asia-Pacific markets overthe period 2001–2005.

Our results based on daily firm data in three Asia-Pacific markets, namely,Australia, Japan, and Vietnam indicate the following. First, stock prices tendto be reversed after large price changes. Second, in the case of large pricedeclines defined by arbitrary trigger values, investors may earn profit fromexploiting the phenomena of price reversals, however, the profit is not largeenough to exploit since it is less than the profit from passive funds. Thisresult is supportive of the weak form of EMH. Third, we find mixed evidenceof whether the price reverses or not over the long-term period. Forth, marketconditions (i.e., bear or bull) may not explain the magnitude of pricereversals. Finally and most importantly, the dynamic measures of large pricechanges based on individual firms provide most consistent evidence acrossmarkets, which are supportive of short-term price reversals and overreactionhypothesis. This evidence exists in emerging market such as Vietnam as wellas developed markets such as Australia and Japan.

ACKNOWLEDGMENTS

The authors are greatly appreciative to Bruce Grundy (the AsianFA/FMAdiscussant), Nabil Maghrebi, Kazuhiko Nishina, and participants at theAsianFA/FMA Meeting in Auckland, New Zealand, 2006 for their valuablecomments.

REFERENCES

Atkins, A., & Dyl, E. (1990). Price reversals, bid-ask spreads, market efficiency. Journal of

Financial and Quantitative Analysis, 25, 535–547.


Bremer, M., Hiraki, T., & Sweeney, R. J. (1997). Predictable patterns after large stock price

changes on the Tokyo Stock Exchange. Journal of Financial and Quantitative Analysis,

33, 345–365.

Bremer, M., & Sweeney, R. J. (1991). The reversals of large stock-price decreases. Journal of

Finance, 46, 747–754.

Bremer, M., & Sweeney, R. J. (1996). Short-run rebounds after large stock-price decreases: The

virtue of resisting panic selling. Nanzan Management Review, 10, 1–23.

Brown, K., Harlow, W. V., & Tinic, S. M. (1988). Risk aversion, uncertain information, market

efficiency. Journal of Financial Economics, 22, 355–385.

Brown, K., Harlow, W. V., & Tinic, S. M. (1993). The risk, required return of common stock

following major price innovations. Journal of Financial and Quantitative Analysis, 28,

101–106.

Cox, D. R., & Peterson, D. R. (1994). Stock returns following large one-day declines: Evidence

on short-term reversals, long-term performance. Journal of Finance, 49, 255–267.

De Bondt, W. F. M., & Thaler, R. (1985). Does the stock market overreact? Journal of Finance,

40, 793–805.

Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. The Journal

of Finance, 25(2), 383–417.

Hansen, L. P. (1982). Large sample properties of the generalized method of moments

estimators. Econometrica, 50, 1029–1054.

MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic

Literature, 35, 13–39.

MacKinlay, A. C., & Richardson, M. (1991). Using generalized method of moments to test

mean-variance efficiency. Journal of Finance, 46, 511–527.

Park, J. (1995). A market microstructure explanation for predictable variations in stock returns

following large price changes. Journal of Financial and Quantitative Analysis, 30,

241–256.

Wong, M. C. S. (1997). Abnormal stock returns following large one-day advances, declines:

Evidence from Asian-Pacific markets. Financial Engineering and the Japanese Markets, 4,

71–177.


CHAPTER 11

PRICE LIMITS IN ASIA-PACIFIC

FINANCIAL MARKETS: THE CASE

OF THE SHANGHAI STOCK

EXCHANGE

Bert Scholtens and Liu Yao

ABSTRACT

Several Asia-Pacific financial markets impose price limits to reduce

excessive fluctuations. We examine stock price behavior following daily

limit moves on the Shanghai Stock Exchange for 200 firms in the period

1997–2004. We find weak evidence for the occurrence of overreaction on

the Shanghai stock market on the basis of price limits. We conclude that

investors do not exhibit overreaction to the event of limit activation except

in the case of 1-day up limit moves. We also conclude that the Shanghai

Stock Exchange can be regarded as a (semistrong) efficient market.

1. INTRODUCTION

At the Dojima exchange in Japan during the early eighteenth century, pricelimits were imposed on rice futures to reduce excess volatility (Moser, 1990).Price limits did not attract much attention until the 1987 US market crash.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00011-8

229

dx.doi.org/10.1016/S1569-3767(07)00011-8.3d

Following this crash, ‘‘circuit breakers’’ were suggested to prevent themarket from suffering excessive volatility caused by investor overreaction.Circuit breakers soon became popular in many countries and served asthe main measure to reduce excessive fluctuations. Generally, they consist oftrading halts and price limits. Under trading halts, market-wide circuitbreakers halt trading on the whole market for a specified duration when theindicator (e.g., market index) reaches a specified level. Under price limits,trading is permitted when the price is within a specified range.

Especially in the emerging markets in the Asia-Pacific region, price limitshave been widely used. Examples are China, Hong Kong, Japan, Korea,and Taiwan. Lee and Kim (1995) investigate the effect of price limits inKorea and suggest that price limits significantly reduce price volatility.Similar results are established by Fung (1999) for Hong Kong. However,Chung (1991) suggests that there is no strong evidence that price limitsreduce volatility. Choi and Lee (2001) provide evidence that delayed pricediscovery in Korea is due to price limits. Chen (1993) examines the effect ofprice limits on stock price volatility using data from Taiwan. He finds littleevidence that price limits help reduce price volatility. Huang (1998) findssignificant price reversals following both up and down limit moves inTaiwan. Also for Taiwan, Kim and Sweeney (2001) test how price limitsinduce an informed investor to shift part or all of his/her profit-motivatedtrades to the next trading day, thus holding back the spread of information.Cho, Russell, Tiao, and Tsay (2003) examine intraday data from the TaiwanStock Exchange. They find that stock prices accelerate toward the upperbound but there is weak evidence of acceleration toward the lower bound.Chen, Rui, and Wang (2005) examine the effectiveness of price limits inChina. In bullish periods, price limits effectively reduce stock volatility fordownward price movements, but not for upward price movements. Inbearish periods, price limits are effective in reducing stock volatility forupward but not for downward price movements. Kim and Rhee (1997)examine the daily stock price data of the Tokyo Stock Exchange from 1989to 1992 and find delayed price discovery.

The most popular rationale for imposing price limits is to reduce overre-action (see DeBondt & Thaler, 1985, 1987 on overreaction). Proponents ofprice limits advocate that, during periods of extreme price fluctuations, pricelimits provide extra time that allows investors to re-evaluate marketinformation and modify their strategies. Hence, limits can reduce traders’overreaction and reduce price volatility. However, opponents argue thatprice limits serve no purpose but slowing down or delaying the pricediscovery process (e.g., Fama, 1989). Even though limits can temporally

BERT SCHOLTENS AND LIU YAO230

stop the price of a share from falling or rising to an extreme extent on agiven trading day, they argue that the price will continue to move in thesame direction toward a new equilibrium price in subsequent trading day(s).Moreover, rather than stabilizing the stock market, price limits may have amagnet effect that pushes prices toward the limits: When prices movetoward the limits, traders rush to trade as a result of fearing that orders willnot be executed once the limits are hit. That is, price limits result ininvestors’ overreaction when prices are approaching the limits.

Two hypotheses can be formulated. First, according to the (behavioral)overreaction hypothesis, stock market participants overreact to events.In this case, the event is defined as the increase or decrease of stockprices which activate the price limit for 1, 2, and 3 days. The overreactionhypothesis as applied to price limits states that stock prices, upon hittingtheir daily up (down) limits, exhibit a subsequent decrease (increase) on thefollowing days. Therefore, pre- and post-event abnormal returns will havedifferent signs. Second, the (semistrong) efficient market hypothesis predictsthat stock market investors will not overreact to price limit moves becausethe activation of price limits, which is available to the public immediatelyafter it occurs, is fully reflected in current stock prices. Consequently, stockprice reversal will not occur when stock prices hit their limits. Therefore,pre- and post-event abnormal returns will have the same sign. We will testthese different views in the remainder of this chapter.

To sum up, the purpose of our study is to test if investors overreact to theactivation of daily limits on the Shanghai Stock Exchange (SSE). Weapply an event study to investigate the information content of stock prices,which activate the price limits, and the relationship with the overreactionhypothesis from December 1996 to December 2004. We find that over-reaction hardly occurs with respect to price limits. This result is robust to theestimation procedure, sampling period, and event window. The remainderof this chapter is organized as follows. Section 2 describes the data andmethodology. Section 3 presents and discusses our findings. Section 4 putsour findings about China in a regional perspective. The conclusion is inSection 5.


We analyze the SSE. All data are extracted from the China Stock Market &Accounting Research Database (CSMAR).1 Our sample covers the periodbetween 1997 and 2004. The number of stocks ranges from 383 in 1997 to

Price Limits in Asia-Pacific Financial Markets 231

837 in 2004. On the SSE, the stocks of firms that experience losses from theircontinuous operating activities during the last 2 financial years are requiredto be marked as special treated (ST) or particular transfer (PT). The tradingof these stocks is restricted and treated differently from others by the stockexchange agency. One such restriction is that a tighter price limit of 75% isimposed on these stocks. Since the purpose of this study is to investigateinvestors’ reaction to ordinary price limit events, all ST and PT stocks areexcluded from the stock selection. As a result, there are 767 stocks left. Twohundred stocks are selected on the basis of stratified random samplingalthough none of similar studies uses this approach. This sampling methodis adopted because of the frequent IPOs in the period 1997–2004. About 100new firms are listed each year in the investigation period. To eliminate theinfluence of different listing time, the listing time distribution of firm in thesample is made similar to that of the overall population. Thus, the 767 listedstocks are divided into eight strata according to the year of listing. Eachstock in a single stratum is listed in the same time period. After the strata areformed, a simple random sample without replacement is taken from eachstratum. For each trading day, the highest and lowest trading prices allowedfor the particular trading day were calculated from the closing price of theprevious trading day. Then, we compare the daily high/low price with theresults calculated by Eqs. (1a) or (1b), and the activation of the price limitcan be identified.

Phigh;t ¼ ð1þ 10%Þ � Pt�1 (1a)

Plow;t ¼ ð1� 10%Þ � Pt�1 (1b)

where Pt�1 denotes the closing price of previous trading day t�1; Phigh;t

denotes the highest price that can be reached at day t; Plow;t denotes thelowest price that can be reached at day t.

After checking whether the price changes of 710% are caused by reasonslike stock splits, stock dividends, etc., the activation of the limit can beconfirmed as a ‘real’ limit move. The price changes of 710% (or more)caused by special reasons above do not activate the limits, and will not beinvestigated.

Here, the event is the stock price move that activates the price limits(up or down) for 1, 2, or 3 days. A 1-day limit move means that pricechanges within 1 day have reached the highest or the lowest prices calculatedby Eqs. (1a) or (1b). Therefore, an event may contain one, two, or threelimit moves. Table 1 presents the numbers of limit moves for the period


1997–2004. During this period, the total number of limit moves for all listedstocks accumulates to 3,942, with 2,600 up limit moves and 1,342 down limitmoves. One-day limit moves account for the majority of all moves. Onlyevents with no more than three limit moves were included in the research. Asa result, 1,820 up limit moves and 760 down limit moves are selected. Thisamounts to 79% of all up limit moves and to 63% of all down limit moves inthe period 1997–2004.

The distribution of sampled returns series has a relatively long right tailand is peaked relative to normal. One of the causes to this asymmetricdistribution is the existence of outliers. Therefore, we replace the simplereturns with the logarithmic returns. Furthermore, with respect to testingfor significance, we will test on the basis of the Student’s t-test and aCorrado test (Corrado, 1989). This test does not require symmetry in cross-sectional excess return distributions. Therefore, it is viewed as being morepowerful than the parametric Student’s t-test.

We use the event study methodology to test the hypotheses (see Brown &Warner, 1980, 1985, and MacKinlay, 1997, for an overview and discussion.Daily abnormal returns are estimated by the market model:

Rit ¼ ai þ biRmt þ �it (2)

assuming E½�it� ¼ 0 and variance �it ¼ s2� . Rit is the daily return of stock i onday t and is defined as the logarithmic difference of consecutive closing

Table 1. Limit Moves in the Sample.

Total Sample

Number of Limit

Moves

Number of Limit

Moves per Event

Number of

Events

Number of Limit

Moves

Up limit moves 2,600 1-day 1,612 1,612

(100%) 2-day 183 366

3-day 25 75

Subtotal 1,820 2,053

(78.96%)

Down limit

moves

1,342 1-day 695 695

(100%) 2-day 51 102

3-day 14 42

Subtotal 760 839

(62.52%)

Total limit moves 3,942 2,580 2,892

(100%) (73.36%)


prices. The market return on day t, Rmt, is the return of the capitalization-weighted market index compiled by the SSE. The market model parameters,ai and bi, are estimated over a 125-day estimation period of t ¼ ½�140; �16�in which t ¼ 0 is the event day of limit moves. The number of event days isin accordance with the number of days for which limit moves last. In thecase of missing returns, parameter estimation excludes both the day of themissing return and the return of the subsequent day (Brown & Warner,1980, 1985). To ensure that the parameter estimation is not subject to aninfrequent trading bias, at least 100 trading days of the return data must beavailable when the market model is applied. Abnormal daily returns, Ait, forthe event period t ¼ ½�15; 15� around the event day are computed asfollows:

Ait ¼ Rit � ða_it þ b_

it þ RmtÞ (3)

Average abnormal returns (AARs) and cumulative average abnormalreturns (CAARs) are examined for portfolios that are constituted of thestocks with up limit moves and down limit moves. Furthermore, AARs andCAARs are also investigated in the 1-, 2-, and 3-day limits cases.

Brown and Hartzell (2001) suggest an alternative to estimate the a’sand b’s in the market model. They regress the stock daily returns on themarket index returns using the entire sample period. In other words, onlyone pair of a and b is estimated by Eq. (2): Rit ¼ ai þ biRmt þ �it. Comparedto Huang (1998)’s 125-day market model, which asks for separateestimations of 2,580 pairs of a and b according to our sample size, Brownand Hartzell’s model needs only 200 estimations since each stock requiresonly one pair of a and b. Then these two parameters are used to calculatethe abnormal returns around the event days. We will use this alternativeapproach as a robustness test. To avoid a sampling bias, we adopt astratified random sampling method, which, as far as we are aware of,has never been adopted by any prior researches in investigating stockmarket overreaction. The test of sampling bias involves two steps. First,a second sample (sample 2) is selected by repeating the stratified randomsampling procedure with different random numbers. Then, we test ifoverreaction exists in sample 2 and check if the results of sample 2 areconsistent with those of the first sample. Furthermore, we design asensitivity test in order to determine whether the changes in the patternsof the cumulative abnormal returns are related to the changes in the lengthof event windows. To this extent, we reduce the event windows from day[�15, +15] to day [�5, +15].


3. RESULTS

Tables 2–4 present AARs, CAARs, t-statistics, and Corrado-statistics forthe cases of 1-, 2-, and 3-day up limit moves and down limit moves,respectively. The AARs and CAARs are computed for the entire test periodfrom 15 days before limit activation days to 15 days after the limit activationdays. For brevity sake, we only report the abnormal returns in the [�5; 10]interval. In line with Huang (1998), the at-event period in this study is thelimit activation days plus one subsequent trading day (post limit activationday): the at-event period consist of days 0 and 1 for 1-day limit moves;days 0, 1, and 2 for 2-day limit moves and days 0, 1, 2 and 3 for 3-daylimit moves.

Table 2. One-Day Up and Down Limit Moves: Average AbnormalReturns (AAR), Student’s t-Statistics of AAR (t(AR)), Corrado’st-Statistics (Corrado t(AR)), and Cumulative Average AbnormalReturns (CAAR) of 200 Firms Listed on the Shanghai Stock

Exchange in 1997–2004.

Day One-day Up Limit Moves (N=1,612) One-day Down Limit Moves (N=695)

AAR t(AR) Corrado

t(AR)

CAAR AAR t(AR) Corrado

t(AR)

CAAR

�5 0.00100 0.73 �0.05 0.01059 0.00034 0.30 �0.12 0.00055

�4 0.00209 1.52 0.38 0.01268 0.00055 0.49 0.59 0.00111

�3 0.00244 1.78� 0.44 0.01512 0.00231 2.03�� 0.23 0.00342

�2 0.00205 1.49 0.33 0.01717 0.00170 1.49 �0.18 0.00511

�1 0.00217 1.58 0.11 0.01933 �0.00612 �5.38�� 2.47�� 0.00101

0 0.06551 47.73�� 11.08�� 0.08485 �0.05173 �45.49�� 11.02�� 0.052741 �0.00427 �3.11�� 1.58 0.08058 �0.00633 �5.57�� 2.06�� 0.05907

Reversal

2 �0.00484 �3.52�� 1.73� 0.07575 0.00024 0.21 �0.09 �0.05883

3 0.00002 0.01 �0.82 0.07576 0.00108 0.95 0.17 �0.05775

4 �0.00137 �1.00 �1.10 0.0744 0.00077 0.68 �0.11 �0.05697

5 �0.00542 �3.95�� 1.74� 0.06898 0.00289 2.54�� 0.75 �0.05408

6 �0.00449 �3.27�� 1.29 0.06449 0.00136 1.20 0.58 �0.05272

7 0.00444 3.24�� 0.62 0.06893 0.00141 1.24 0.39 �0.05131

8 �0.00144 �1.05 �0.81 0.0675 �0.00024 �0.21 �0.23 �0.05155

9 0.00082 0.14 �0.44 0.06769 0.00145 1.28 0.40 �0.05009

10 �0.00218 �1.59 �0.62 0.06551 �0.00014 �0.12 �0.61 �0.05023

�Significant at the 0.10 level.��Significant at the 0.05 level.��Significant at the 0.01 level.


Table 2 shows that in the pre-event period, the AARs are statisticallyinsignificant. On day 0 (limit activation day), there is an AAR of 0.06551,which is statistically significant at the 1%-level. On day 1 (post limitactivation day) there is a significantly negative AAR only by means ofStudent’s t-statistics. On day 2, a reversal of the significant AAR is observedwhich lasts until day 6. From the 1-day down limit moves, the AAR is�0.0517 on the limit activation day is highly statistically significant. TheAAR on the post limit activation day is also negative and significant.A possible explanation for the negative AAR on day 1 is that the activationof the �10% limit did not allow the market to reach equilibriumautomatically on day 0 and lead to a transfer of transactions to day 1.Then, the downward trend continued on day 1, when pending orders wereexecuted. From day 2 to 4, the AARs are all (insignificantly) positive. Wefind some reversal, but it is insignificant in term of Corrado’s t-statistics.Following the above analysis, we can draw the conclusion that, for 1-day up

Table 3. Two-Day Up and Down Limit Moves: AAR, t(AR), Corradot(AR), and CAAR of 200 Firms Listed on the Shanghai Stock Exchange

in 1997–2004.

Day Two-day Up Limit Moves (N=183) Two-day Down Limit Moves (N=51)

AAR Student

t(AR)

Corrado

t(AR)


t(AR)

CAAR

�5 0.00017 0.09 �0.49 0.00286 0.00831 2.02� 0.82 0.03459

�4 �0.00077 �0.41 �0.89 0.00208 0.01578 3.83�� 2.43�� 0.05038

�3 �0.00144 �0.77 �0.80 0.00064 0.00988 2.40�� 0.90 0.06026

�2 0.00124 0.66 0.26 0.00188 0.00374 0.91 0.33 0.06400

�1 �0.00222 �1.19 �1.01 �0.00034 �0.00719 �1.75� �1.55 0.05681

0 0.05164 27.73�� 6.17�� 0.05131 �0.07565 �18.38�� 7.00�� 0.01883

1 0.06029 32.37�� 8.69�� 0.11159 �0.05492 �13.34�� 5.68�� 0.07376

2 �0.00305 �1.64� �0.89 0.10854 �0.00665 �1.61 �0.67 �0.08040

Reversal

3 �0.00590 �3.17�� 1.58 0.10265 �0.01107 �2.69�� 0.99 �0.09148

4 �0.00310 �1.67� �1.00 0.09954 0.00649 1.58 1.02 �0.08499

5 �0.00512 �2.75�� 1.35 0.09442 0.00606 1.47 1.33 �0.07892

6 0.00284 1.52 0.32 0.09726 0.01146 2.78�� 1.12 �0.06746

7 �0.00068 �0.37 �0.90 0.09657 �0.00198 �0.48 �0.12 �0.06943

8 0.00113 0.61 0.75 0.09770 �0.00754 �1.83� �1.14 �0.07698

9 0.00354 1.90 0.15 0.10124 0.00113 0.28 0.04 �0.07584

10 0.00074 0.40 �0.36 0.10198 �0.00269 �0.65 �0.20 �0.07853



limit moves the reversal with an AAR occurred on day 2, which is 1 dayafter the at-event days. In addition, there is weak evidence that supports theexistence of reversal for 1-day down limit moves. The 3-day lagged reversalis insignificant when Corrado’s t-statistics are examined.

Table 3 presents the 2-day up and down limit moves. On day 0 (2-daylimit activation) the AARs of 0.0516 is highly statistically significant. TheAAR on day 1 is also positive and statistically significant. On day 2, which isthe post limit activation day, there is an AAR of �0.00301. Furthermore,there is a reversal. Both only are significant when Student’s t-statistics areexamined. In the case of the 2-day down limit moves, the AARs on the limitactivation days are statistically significant. We find evidence for pricereversal for 2-day up and down limit moves only on the basis of Student’st-statistics. In the 2-day up limit case, a significant reversal is observed onday 3, which lasts until day 5. For the 2-day down limit case, a 3-day laggedreversal occurs on day 6. However, this evidence is weak compared to that

Table 4. Three-Day Up and Down Limit Moves: AAR, t(AR), Corradot(AR), and CAAR of 200 Firms Listed on the Shanghai Stock Exchange

in 1997–2004.

Day Three-day Up Limit Moves (N=25) Three-day Down Limit Moves (N=14)

AAR Student

t(AR)

Corrado

t(AR)


t(AR)

CAAR

�5 0.00765 1.79� 0.86 0.03892 0.01314 1.77� 0.75 0.10727

�4 0.00739 1.73� 0.24 0.04631 �0.00342 �0.46 �0.54 0.10385

�3 �0.00705 �1.65 �1.63 0.03926 0.01794 2.41�� 1.67 0.12179

�2 0.00281 0.66 0.25 0.04207 0.00945 1.27 0.57 0.13124

�1 �0.01168 �2.73�� 1.89� 0.03039 0.00626 0.84 0.89 0.13750

0 0.06984 16.34�� 5.52�� 0.10023 �0.06966 �9.36�� 4.04�� 0.06784

1 0.07036 16.46�� 5.22�� 0.17059 �0.06993 �9.40�� 3.96�� 0.00209

2 0.07153 16.73�� 4.71�� 0.24212 �0.05893 �7.92�� 3.12�� 0.06102

3 0.00025 0.06 �0.39 0.24237 �0.01113 �1.50 �1.33 �0.07215

Reversal

4 0.00053 0.12 �0.73 0.24290 �0.00873 �1.17 �1.50 �0.08088

5 0.00299 0.70 0.22 0.24589 �0.00120 �0.16 0.37 �0.08208

6 �0.00451 �1.05 �0.61 0.24138 �0.00506 �0.68 0.28 �0.08713

7 �0.00336 �0.78 �0.88 0.23803 0.01235 1.66 �0.13 �0.07479

8 �0.00215 �0.50 �0.26 0.23588 �0.00338 �0.45 0.40 �0.07817

9 0.00139 0.32 �0.24 0.23727 �0.00146 �0.20 �0.42 �0.07963

10 0.01624 3.80�� 1.84� 0.25351 �0.01103 �1.48 �1.27 �0.09066



found in the case of 1-day up and down limit moves due to the fact that it isinsignificant when we take the Corrado’s t-statistics into account.

Table 4 gives the results for the 3-day up and down limit moves. From the3-day up limit moves, we observe that all AARs on the limit activation days(day 0, 1, and 2) are positive and significant. Though the AAR on day 3(post limit activation day) is positive, it is statistically insignificant. There isno clear evidence of price reversal for 3-day up limit moves. The 3-day downlimit moves shows a different pattern. On day 0 the AAR is positive andsignificant, which continues until day 2 (3-day limit activation). Though theAAR on day 3 (post limit activation day) is negative, it is statisticallyinsignificant. We did not observe any significantly positive abnormal returnsduring the post-limit days. Thus, there is no price reversal for the 3-daydown limit moves either. In short, no reversal for the 3-day up limit movesor down limit moves was detected.

Figs. 1–3 plot CAARs (from Tables 2–4) for 1-day (Fig. 1), 2-day (Fig. 2),and 3-day (Fig. 3) limit moves. For the up limit moves, price reversalsexisted in two of the three cases (1- and 2-day). The AARs on the ‘‘reversalday’’ are �0.0048 and �0.0059 for the 1- and 2-day case, respectively. Thelatter AAR is only statistically significant in term of Student’s t-statistics.The CAARs are both negative over the post-limit days for 1- and 2-day up

-0.30

-0.20

-0.10

0.00

0.10

0.20

0.30

-15 -13 -11 -9 -7 -5 -3 -1 1 3 5 7 9 11 13 15

Event Days

CA

Rs

1D LU1D LD

Fig. 1. Cumulative Abnormal Returns (CARs) for 1-Day Up and Down Limit

Moves (1D LU and 1D LD).


-0.30

-0.20

-0.10

0.00

0.10

0.20

0.30

-15 -13 -11 -9 -7 -5 -3 -1 1 3 5 7 9 11 13 15

Event Days

CA

Rs

2D LU2D LD

Fig. 2. CARs for 2-Day Up and Down Limit Moves (2D LU and 2D LD).

-0.30

-0.20

-0.10

0.00

0.10

0.20

0.30

-15 -13 -11 -9 -7 -5 -3 -1 1 3 5 7 9 11 13 15 17

Event Days

CA

Rs

3D LU

3D LD

Fig. 3. CARs for 3-Day Up and Down Limit Moves (3D LU and 3D LD).


limit cases. For the down limit moves, price reversals appear to be lagged3 days for 1- and 2-day down cases. There is no price reversal at all for thecase of 3-day down limit moves. Moreover, only the CAARs over the post-limit days for the 1-day down limit case are all positive.

With respect to the robustness of our results, the method applied byBrown and Hartzell (2001) is used to test whether the different length ofestimation period can bias the results. In general, similar results were found,although the magnitude of the AARs and t-statistics differed slightly. Forexample, for 1-day up limit moves, it is observed that on day 2 the AARis statistically significant. From this robustness check, we conclude thatsignificant reversals exist for the 1- and 2-day up limit moves if Student’st-statistics are examined independently. When the Corrado’s t-statistics isconsidered, the reversal is only significant for the case of 1-day limit moves.For 1-day down limit case, we observed a 3-day lagged reversal, which isonly statistically significant by Student’s t-statistics. On day 5, a significantAAR is observed. However, we find a different pattern for the case of 2-daydown limit case. Instead of the 3-day lagged reversal observed by usingthe method of Huang (1998), we found a 1-day lagged reversal by usingthe methodology of Brown and Hartzell (2001). That is, a (significantly)negative AAR on day 3 is followed by a significant reversal. This reversallasted until day 6. Again, the significance is only supported by Student’st-statistics. Besides, the AAR on day 1 for both cases of 1- and 2-daydown limit moves is significantly (Student’s t-value) negative, and can beinterpreted as the indicator of upcoming down limit moves. The results for3-day up and down limit cases were consistent with those of Huang (1998).That is, no evidence of overreaction or reversal is found.

Since the sampling method of this chapter differs from that used inprevious studies, it seems necessary to test whether our results are biased bythe sampling method. Using the same procedure of stratified sampling and adifferent group of random numbers, we obtained another sample. Theresults of this sample two differed in magnitude, but not in sign or level ofsignificance.

Third, we reduce the event window in line with the suggestion by Huang(1998). We start cumulating abnormal returns from day 5 onwards, insteadof from day 15 (for brevity sake, these results are not reported but areavailable upon request with the authors). Again, we find nearly identicalresults, which suggest that most of the CAARs are cumulated on or aroundthe event day(s). Thus, we may conclude that our results are not sensitivewith respect to the length of the event window.


In general, the evidence of overreaction found in this study is not asstrong as that found elsewhere. Only in case we would rely on the Student’st-statistic, we establish that there is reversal in four out of six cases.However, on the basis of the Corrado test statistic, which can be regardedas a robustness check on the Student test, we only have a price reversal inone case. This rank test is better specified under the null hypothesis andmore powerful under the alternative hypothesis than the parametric t-test(Corrado, 1989). Second, we also find fewer information leakages on pre-event days than previous studies did (Huang, 1998). Third, the Chinesestock market tends to overreact slower than others market. For instance,Huang (1998) finds that reversals occurred immediately after the post limitactivation day. However, our study suggests that investors on the SSE tendto overreact to the 1- and 2-day down limit moves 3 days later than theircounterparties in Taiwan. This seemingly typical feature of the SSE isconsistent with the finding of Chen et al. (2005), who found for downwardprice movements that price limits effectively reduce panic as a result of badnews. In our study, the evidence is also more favorable for the up limit casesthan the down limit cases. That is, when Student’s t-test is taken as adecision criterion, as is the case with our reference studies, price reversalsoccur immediately after the event day(s) for the 1- and 2-day up limit cases,but only 3 days after event day(s) for the 1- and 2-day down limit cases.When the Corrado’s t-test is taken into account, only the reversal for the1-day up limit case remain significant. The last difference with other studiesis that, when the result of the Corrado test is taken into account, pricereversals only exist for the 1-day up limit case, which is one out of the sixcases. Investors in the SSE may overreact to other information, butapparently not to the limit activation except for the 1-day up limit moves.The results of our study are robust since the tests of potential estimationbias, sampling bias, and sensitivity do not change them significantly.Referring to our hypotheses, the overreaction hypothesis can only beaccepted for the 1-day up limit case. Therefore, in general, we conclude thatthe Shanghai stock market appears to be (semistrong) efficient on the basisof the market response to hitting price limits.

4. REGIONAL PERSPECTIVE

This section puts the findings for the SSE in a regional perspective. It seemsworthwhile to point out that price limits in Asia date back to the early


eighteenth century. At the Dojima exchange in Japan, price limitswere imposed on rice futures in order to limit price volatility (Moser,1990). Price limits in Asia became quite accepted with the development ofstock exchanges (Rhee & Chang, 1993). Examples of markets that haveprice limits to reduce excess volatility are Korea, Taiwan, Japan, Malaysia,Thailand, and China. Hong Kong and Singapore can be regarded as more‘liberal’ markets in the sense that they do not use price limits.

The evidence regarding the impact of price limits in Asia is quite mixed.Lee and Kim (1995) find that in Korea price limits significantly reduce pricevolatility. Furthermore, Choi and Lee (2001) establish for Korea thatprice limits result in delayed price discovery. Fung (1999) also found thatprice limits reduced volatility in Hong Kong during the periods in which thismarket applied the limits. However, this might be due to the fact that HongKong had these limits only for relatively short periods of time. On the otherhand, Chung (1991) suggests that there is no strong evidence that pricelimits reduce volatility. Chen (1993) examines the effect of price limits onstock price volatility using data from Taiwan. He finds little evidence thatprice limits help reduce price volatility. Furthermore, Huang (1998) findssignificant price reversals following both up and down limit moves inTaiwan. Cho et al. (2003) examine intraday data from the Taiwan StockExchange and find that stock prices accelerate toward the upper bound butthere is weak evidence of acceleration toward the lower bound. Chen et al.(2005) examine the effectiveness of price limits in China. In bullish periods,price limits effectively reduce stock volatility for downward price move-ments, but not for upward price movements. In bearish periods, price limitsare effective in reducing stock volatility for upward but not for downwardprice movements. Kim and Rhee (1997) examine the daily stock price dataof the Tokyo Stock Exchange from 1989 to 1992 and find delayed pricediscovery.

Our results have several implications from the perspective of Asia-Pacificfinancial markets. First, there is little scope for profiting when the limit isreached as most price reversal cannot be predicted. Second, the SSE appearsto have become a mature market. Third, the way to assess the significance ofabnormal returns is very important as we sometimes do find a significantreaction on the basis of a parametric test, whereas the more correct non-parametric test shows that the price reversals are insignificant. Fourth, thisobservation can have an effect on the assessment of price limits in Asia-Pacific markets as most research so far tends to rely on parametric testingwhereas we have the impression that the underlying data do not allowfor using this approach but, instead, warrant non-parametric tests for the


significance of any abnormal returns. This also is one of the challenges forfuture research.

5. CONCLUSION

Price limits do frequently occur in Asia-Pacific financial markets. Thischapter tests for overreaction by examining the price movements following1-, 2-, and 3-day cases of price limit moves in China. The sample contains200 listed firms on the SSE from 1997 to 2004. A short-overreaction pattern,which is a predictable price reversal, is observed in the case of 1-day up limitmoves. On day 2, a negative and significant abnormal return is observed.According to our tests, the price reversal only exist in one out of the sixcases, therefore, we conclude that the investors on the SSE do not exhibitoverreaction to the event of limit activation except in the 1-day up limit case.Strictly speaking, the overreaction hypothesis is only accepted in the case of1-day up limit move. In general, our results suggest that the Shanghai stockmarket can be regarded as (semistrong) market efficient. Our findings arerobust for an alternative estimation method, a different sampling method,and a shorter event window.

NOTE

1. This database is compiled by the China Accounting and Finance ResearchCentre of the Hong Kong Polytechnic University and Shenzhen GTA InformationTechnology Ltd. We are very grateful that we were allowed to use this database.

REFERENCES

Brown, G. W., & Hartzell, J. C. (2001). Market reaction to public information: The atypical

case of the Boston Celtics. Journal of Financial Economics, 60, 333–370.

Brown, S. J., & Warner, J. B. (1980). Measuring security price performance. Journal of Financial


Brown, S. J., & Warner, J. B. (1985). Using daily stock returns: The case of event studies.

Journal of Financial Economics, 14, 3–31.

Chen, G. M., Rui, O. M., & Wang, S. S. (2005). The effectiveness of price limits and stock

characteristics: Evidence from the Shanghai and Shenzhen Stock Exchanges. Review of

Quantitative Finance and Accounting, 25, 159–182.

Chen, Y. M. (1993). Price limits and stock market volatility in Taiwan. Pacific-Basin Finance

Journal, 1, 139–153.


Cho, D. D., Russell, J., Tiao, G. C., & Tsay, R. (2003). The magnet effect of price limits:

Evidence from high-frequency data on Taiwan Stock Exchange. Journal of Empirical

Finance, 10, 133–168.

Choi, W. S., & Lee, S. B. (2001). A transitory and asymmetric properties in price limits: Korean

evidence with intra-day data. Working Paper.

Chung, J. R. (1991). Price limit system and volatility of Korean stock market. In: S. G. Rhee &

R. P. Chang (Eds), Pacific-basin capital markets research (Vol. 2, pp. 283–294).

Amsterdam: Elsevier Science Publishers.

Corrado, C. (1989). A nonparametric test for abnormal security-price performance in event

studies. Journal of Financial Economics, 23, 385–395.

DeBondt, W. F. M., & Thaler, R. H. (1985). Does the stock market overreact? Journal of

Finance, 40, 793–805.

DeBondt, W. F. M., & Thaler, R. H. (1987). Further evidence on investor overreaction and

stock market seasonality. Journal of Finance, 42, 557–581.

Fama, E. (1989). Perspectives on October 1987, or what did we learn from the crash?

In: R. W. Kamphuis, Jr., R. C. Kormendi & J. W. H. Watson (Eds), Black Monday

and the future of the financial markets (pp. 71–82). Homewood, IL: Irwin.

Fung, A. K.-W. (1999). Overreaction in the Hong Kong stock market. Global Finance Journal,

10, 223–230.

Huang, S. H. (1998). Stock price reaction to daily limit moves: Evidence from the Taiwan Stock

Exchange. Journal of Business Finance and Accounting, 25, 469–483.

Kim, K. A., & Rhee, S. G. (1997). Price limit performance: Evidence from the Tokyo Stock

Exchange. Journal of Finance, 52, 885–901.

Kim, K. A., & Sweeney, R. J. (2001). Effects of price limits on information revelation: Theory and

evidence. Working Paper.

Lee, S. B., & Kim, K. J. (1995). The effect of price limits on stock price volatility: Empirical

evidence in Korea. Journal of Business Finance and Accounting, 22, 257–267.

MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic

Literature, 35, 13–39.

Moser, J. T. (1990). Circuit breakers. Economic Perspectives, 14, 2–13.

Rhee, S. G., & Chang, R. P. (1993). The microstructure of Asian equity markets. Journal of

Financial Services Research, 6, 437–454.


CHAPTER 12

CHINA’S SECURITIES MARKETS:

CHALLENGES, INNOVATIONS,

AND THE LATEST DEVELOPMENTS

Xinyi Yuan, Wei Fan and Qiang Liu

ABSTRACT

Important developments of China’s securities markets within the last two

years, namely, the Share Reform, the warrant market, the innovative

listed open-end funds (and exchange-traded funds), corporate bonds

with detachable warrants, exchange-traded asset-backed securities, are

described. The discussion focuses on unique, innovative features of these

products, as compared to their counterparts available in more mature

markets (when applicable), and points to possible future research themes.

The proposed rules with regard to stock index futures and credit trading

are also discussed.

1. INTRODUCTION

It took China’s leading Shanghai Stock Exchange (ShSE) Composite Indexonly 562 days to go from 1,011 to 2,975 points, an increase of almost 200%.At the same time, the index’s trading volume shot up an enormous 835%(Fig. 1).1 For comparison, during the height of the Internet bubble the





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00012-X

245

dx.doi.org/10.1016/S1569-3767(07)00012-X.3d

NASDAQ Composite Index took 889 days to rise the same percentageamount, closing at its historical high of 5,048 points on March 10, 2000,while its corresponding trading volume increased a merely 160%.2 At theend of January, the daily trading value of China’s stock markets reachedover 100 Billion Chinese Yuan (CNY).3 At this rate, the expected annualtrading value would surpass China’s GDP of 20.94 trillion for 20064 by awide margin. New customers were overflowing branch offices of brokerages,with the A-share5 market adding over 90 thousand new accounts daily.6

Continuing its bull run, the Composite Index closed at 3,841 points on April30, 2007. Clearly, all those are signs of a possible stock market bubble.7

With a domestic market capitalization more than 50% of its GDP and theworld’s fourth-largest economy, no wonder China attracts global attentionsthose days. A keynote speaker discussed China’s financial role in the AsiaPacific region,8 addressing the 19th Australasian Finance and BankingConference held in Sydney in December 2006. Recently, the New YorkTimes ran front page stories about China’s hot stock market (Yardley, 2007)

500

1000

1500

2000

2500

3000

05-05

-10

05-0

7 -19

05-0

9-27

05-1

2-06

06-0

2-14

06-0

4-25

06-0

7-04

06-0

9-12

06-1

1-21

07-0

1-30

0

50

100

150

200

250

300

Clo

sing

Pric

e

Tra

ding

Vol

ume

Trading Date

Fig. 1. Daily Closing Prices (the Upper Curve) and Trading Volumes (the Lower

Black Area) of the ShSE Composite Index from May 10, 2005 to January 31, 2007.

XINYI YUAN ET AL.246

and its increasing influence on markets around the world (Norris & Peters,2007), while a BusinessWeek article featured government officials warningChinese investors of bubble and a possible correction (Bremner, 2007).According to a China Securities Journal article (Lu, Niu, & Huang, 2007),Goldman Sachs however did not believe the market was overvalued, whichis of course a typical stance for investment banks during bull markets andshould be taken with a grain of salt.

Established in December 1990, China’s young stock markets ‘‘havehistorically been stagnant financial backwaters, marred by scandal, weakoversight, and fundamental contradictions’’ (Yardley, 2007). Giant state-owned enterprises went public while holding their majority of sharesnontradable, T+0 trading (securities bought and sold on the same day) isnot allowed, credit trading (short-selling and margin-buying) is forbidden,and most importantly government policies have had a major influence onthe market. As a result, the market nearly collapsed in June 2005. In the pastyear and a half, a series of reforms, such as the so-called Share Reform thatconverts the (state-owned) nontradable shares to tradables and theinnovative IPOs while companies get listed in the A-share markets and inHong Kong’s market simultaneously, were implemented to boost investors’confidence in the market.

China now has organized markets for commodity futures, commonstocks (A- and B-share), closed-end funds, exchange-traded funds (ETFs),listed open-end funds (LOFs), warrants, government debts, corporate andfinancial bonds, convertible bonds, and asset-backed securities (ABSs).A judiciously selected group of latest developments, current innovations, andfuture trends of China’s financial markets will be discussed in this article.

First, Share Reform and its by-product, the warrant market, will bedescribed. The innovative LOFs as well as the traditional ETFs will then bepresented. Corporate bonds with detachable warrants (BDWs), followed byABSs, will be discussed next. Finally, an overview of the much anticipatedindex futures and credit trading is going to be given.

2. THE SHARE REFORM9

In the 1990s, most of the companies that sold shares to the public were bigstate-owned enterprises, whose majorities of shares, originally owned by thegovernment, were nontradable according to government policies (Jin &Yuan, 2006). As of January 31, 2005, tradable A-shares accounted for only30% of the total outstanding market value of the A-share market (Liu, 2007).

China’s Securities Markets 247

This two-tier structure of equity ownerships caused two major problems forthe healthy development of the securities market. First, holders ofnontradable shares, who control the board, tend to favor policies thatincrease the net assets of the firm but ignore completely or hurt the long-termperformance of the stock in the secondary market. As a result, equityfinancing became a tool to enrich the owners of nontradable shares, whilepunishing the investing public, which as a whole lost a significant amount ofmoney. Second, the pricing ability of the market was quite weak, sincenobody knew how to value those nontradable shares and thus the whole firm.Coupled with a small (tradable) market capitalization, the market had beenprone to manipulations and quite volatile (Wu & Ruan, 2004).

To solve those problems, the China Securities Regulatory Commis-sion (CSRC) launched the Share Reform in May 2005, when the A-sharemarket was almost at a point of eight-year low. Reform plans of listedcompanies were proposed and approved by their shareholders, with littlegovernment intervention. In order to carry out the reform smoothly, holdersof nontradable shares had to pay for the right to trade. As a result, inves-tors of tradable shares were awarded additional shares, free warrants, orcash payments (or any combinations of these); some companies did areverse-split of their nontradable shares (or reduced their total nontradableshares).

As of December 31, 2006,10 1,452 companies, or 97% of all the listedcompanies in China, had finished their Share Reform (Table 1).11 Evenwithout the running up of the stock market, the Share Reform alone wouldhave increased the number of tradable shares or the market capitalizationroughly by two-fold. This is certainly a benign and welcome development,since the problems caused by the two-tier structure of equity ownerships areno longer a thorny issue for investors.

Table 1. Progress of the Share Market Reform.

No. of Firms

Listed

No. of Firms

Reformed

Percent of

Firms

Reformed

Market Value

(Billion)

Market Value

(Reformed)

Shanghai Stock

Exchange

813 795 97.9 7,161 6,961

Shenzen Stock

Exchange

679 657 96.8 1,779 1,692

Total 1492 1452 97.3 8,940 8,653


3. THE WARRANT MARKET

Even though China does not yet have an options market, it has a warrantmarket now, which is a by-product of the Share Reform discussed in theprevious section. The issuing of warrants in the Share Reform was a veryimportant event in the development of China’s financial markets, becausewarrant was the first financial derivative in China that is so closely related tooptions.12

On August 22, 2005, the first warrant, issued by Bao Steel Company,began to trade on ShSE. Eighteen months later, China’s warrant market wasthe biggest in the world in terms of annual total traded value, surpassingthat of Hong Kong (Mitchell, 2007). So far 28 companies had issued 34warrants and long-dated options13 to shareholders (Table 2).

In addition to listed companies, securities firms are also allowed to issuecovered options. To write calls, a securities firm has to deposit with the exchan-ge the full number of shares of the underlying stock for exercising; a marginequal to the exercise price needs to be maintained in order to issue puts.

Margin accounts are not required to trade warrants. Trades are matchedby computers, not market makers. Unlike stocks, warrants trade on a T+0basis. On average, a warrant may trade or change hands 150 times beforeexercising, while in more mature markets, this rate is on the order of tens(Mitchell, 2007).

As of January 31, 2007, 27 warrants were listed and actively traded. Mostof the warrants are nominally Bermudan, but nearly European in terms ofvaluation, since they can only be exercised within five days of maturity. Thewarrants are long-dated with a maturity of one or two years. Among thesewarrants, 16 are calls. Due to the running up of the stock market, calls arecurrently deep in-the-money, while puts deep out-of-the-money.

Table 2. Warrant Issuances as of January 2007.

Panel A: Derivative Type

Warrant Long-Dated Call Long-Dated Put

6 10 18

Panel B: Exercising Style

European Bermudan

4 30


This new market had shown several interesting characteristics. Many ofthe warrants were much too overpriced initially, but following the recent runup in the stock market, they became under-priced. As of January 2007,the average ratio of time value to stock price of the active warrants was amere 4.5%, the lowest among global warrant markets (Yang & Ma, 2007).Particularly puzzling are four calls, namely, MagangCWB1,14 Qiaocheng-HQC1, YiliCWB1, and WuliangYGC1, that traded below their intrinsicvalues, max(S–K,0) (Hull, 2003). On its last trading (or exercising) day,WugangJTB1 still had market prices that were below its intrinsic values fora total of 117 minutes. On the other hand, the prices of many puts in 2006were too high compared with their theoretical (Black–Scholes) fair values.Of the 13 puts, eight had a fair value below 0.002 CNY, four below 0.04,while their closing prices ranged from 0.383 to 1.306 on January 31, 2007.Some of the matured puts did show big price drops several days beforematurity, however.

Several factors may have contributed to those abnormal behaviors.Institutional investors acted only as writers of warrants, and accountedroughly for a tiny 1% of the trading volumes (Liu, 2006). Among theinvestors in warrants, the majority were small, individual investors, whomay not know or understand how to price options and were still learningthis new derivative product. It is puzzling that the securities firms, whichshould have better knowledge, did not take advantages of those possiblearbitrage opportunities.

The huge trading volumes and high turnover rates could be a result ofT+0 trading, since warrants provide the only T+0 trading available inChina. The lack of short-selling mechanism in the stock market may havealso contributed to those problems.

All the warrants issued so far are dividend protected; that is, the strikeprice of a warrant is adjusted according to pre-specified rules when dividendis paid over the life of the warrant. This adjustment of strike price makes thepricing of warrant much more complicated, and poses a significant challengeto theorists of derivatives theory.

4. LISTED OPEN-END FUNDS

The investment fund industry in China has had a staggering growth sincethe first fund, a closed-end company, came onto the market in 1998. Asof December 31, 2006, 53 companies managed 321 funds (of which 53 areclosed-end funds) with 856 billion in assets.15 During this rapid development,


two other types of funds, namely, LOF and ETF, were also introduced toinvestors.

The idea of LOF, or exchange-traded mutual funds (open-end compa-nies), was a Chinese financial innovation. On December 20, 2004,Nanfangjipei16 (or Active Allocation Fund of China Southern FundManagement), the first LOF, began to trade on the Shenzhen StockExchange (SzSE). As of March 31, 2007, 17 LOFs with 69.7 billion totalassets were listed on SzSE (Table 3).

LOFs have the following innovative features that are markedly differentfrom the well-established practices of mutual funds in the United States.

1. A LOF comes into existence through a period of public offering, wheninvestors can purchase shares through either SzSE or the fund manager.In general, after it is legally established, but before trades on SzSE, thefund is closed to investors (for purchasing or redeeming) for a period ofno more than three months.

2. After a fund is listed on SzSE, its shares can be purchased or redeemedthrough either SzSE or the fund manager at its daily net asset value(NAV); further, shares can be traded continuously at market-determined(by supply and demand) prices when SzSE is open. Shares bought(through trading) on SzSE can be either sold or redeemed the nexttrading day (or T+1), while shares purchased (through issuing) will beonly available the day after the next trading day (or T+2).

3. Shares purchased through fund managers are recorded in the MutualFund Depository and Clearing System of China Securities Depositoryand Clearing Corporation (CSDCC), while shares obtained on SzSE arerecorded in the Shenzhen Securities Depository and Clearing System ofCSDCC. Shares in one system cannot be redeemed (or sold) directlyin the other, and it takes two days to transfer shares across systems(or T+2).

LOFs offer at least three advantages over mutual funds. First, lowertransaction costs. It costs less than 0.3% of the total proceeds to trade, whilethe cost of purchase (via issuing) is around 1.5% and that of redemptionaround 0.5%. Second, price certainty. Investors can get a certain price throughtrading, while the price could not be known for sure when issuing/redeem-ing at NAV. Third, reduced transaction time (available in T+1, compared toT+2 for mutual funds). As a result, the liquidity of the funds is expected toincrease due to the possibility of trading of LOFs throughout the day.Further, LOFs offer more flexibility to investors and broaden its investorbase. Choosing between trading and purchasing/redeeming, different types of


Table 3. Listed Open-End Funds as of March 31, 2007.

LOF Name Listing Date NAV Asset (Million) Fund Manager

Nanfangjipei 04/12/20 1.047 5,786 China Southern Fund

Management Co.

Boshizhuti 05/02/22 1.442 4,680 Bosera Fund

Management Co., Ltd.

Zhongyinzhongguo 05/02/23 1.194 2,023 BOC International

Investment Managers

Guangfaxiaopan 05/04/29 1.780 5,065 GF Fund Management

Co., Ltd.

Jingshundingyi 05/05/25 1.610 2,489 Invesco Great Wall

Fund Management

Rongtongjuchao 05/06/16 1.240 1,864 Rongtong Fund


Wanjiagongyong 05/08/15 1.152 293 Wanjia Asset


Nanfanggaozheng 05/09/21 1.464 8,245 China Southern Fund

Management Co.

Jiashi300 05/10/17 1.113 5,932 Harvest Fund


Zhaoshangchengzhang 05/12/09 1.014 6,106 China Merchants Fund

Management Co.

Xingyequshi 06/01/19 3.182 2,470 Industrial Fund


Fuguotianhui 06/02/16 1.075 6,798 Fullgoal Fund


Jingshunziyuan 06/04/07 1.307 2,951 Invesco Great Wall

Fund Management

Heyinxiaolu 06/07/21 1.545 2,481 ABN Amro Teda Fund

Management Co.

Penghuajiazhi 06/09/18 1.816 1,318 Penghua Fund


Changshengtongzhi 07/02/16 1.032 793 Changsheng Fund


Penghuadongli 07/03/09 1.099 10,433 Penghua Fund


Note: Data from Financial Street Holding Co., Ltd. (www.jrj.com). The initial minimum asset

size for a fund to be established and listed is 200 million CNY. When the asset of a fund has

been below 50 million for a period of consecutive 60 trading days, the fund manager may

choose to close the fund. The date format, yy/mm/dd, is used in this article.


traders, such as hedgers, speculators, and arbitrageurs, as well as long-terminvestors, would be attracted to LOFs.

Of course, nothing useful comes for free. SzSE charges LOFs fees forinitial listing as well as monthly maintenance, both of which are passed on toinvestors eventually as costs and thus reduce the overall performance of thefunds. Furthermore, the existence of two parallel systems in LOFs couldresult in market inefficiency and thus arbitrage opportunities. The ease oftrading with LOFs may also attract speculative behaviors in the LOFmarket.

5. EXCHANGE-TRADED FUNDS

The first ETF, the Standard & Poor’s Depository Receipts (SPDRs)17 thattracks the S&P 500 Index, began to trade on the American Stock Exchangeon January 22, 1993. As of December 2005, 201 ETFs with $296 billionassets were traded in the United States.18

The first ETF in China, the Exchange-traded, Open-end ShSE50Index Fund (Shangzheng50ETF),19 started trading on ShSE on February23, 2005. As of March 31, 2007, five ETFs, namely, Shangzheng50ETF,Shangzheng180ETF, HongliETF, Shenzheng100ETF, and Zhongxiaoban-ETF, were traded on ShSE and SzSE (Table 4).

Even though ETF is not a Chinese invention, the trading of ETFs inChina does have some unique features when compared with the trading ofETFs in the United States. For example, ETF shares issued can be sold, but

Table 4. Exchange-Traded Funds as of March 31, 2007.

ETF Name Listing Date NAV Asset (Billion) Fund Manager

Shangzheng50ETF 05/02/23 2.258 6.7 China Asset

Management Co.

Shenzheng100ETF 06/04/24 2.653 4.1 E Fund Management

Co., Ltd.

Shangzheng180ETF 06/05/18 6.340 0.5 Hua’an Fund

Management Co.

ZhongxiaobanETF 06/09/05 1.740 3.0 China Asset

Management Co.

HongliETF 07/01/18 2.481 2.4 AIG Huatai Fund

Management

Note: Data from Financial Street Holding Co., Ltd. (www.jrj.com).


not redeemed on the same day, shares bought (through trading) can beredeemed, but not sold on the same day, stocks obtained via redemption canbe sold, but not used to obtain issued ETF shares, and stocks bought onthe market can be used to obtain issued ETF shares, but not sold on thesame day.20

These rules may appear to be strange and confusing, but they make sensein light of the current mechanism of trading stocks, which is still on a T+1basis. As a matter of fact, the trading of ETFs can be viewed as a first step inmoving toward T+0, since several types of same-day trading against thecorresponding indices are made possible. For example, one could buy on thestock market a bucket of stocks that matches an index proportionally, usethose securities to obtain issued ETF shares, and then sell the ETF shares onthe ETF market, all done in one day. Furthermore, certain (risky) arbitrageopportunities could be exploited using these rules.

6. LOFS VS. ETFS: SOME COMPARISONS

Both LOFs and ETFs are new products in China. As of March 31, 2007,17 LOFs with 69.7 billion total assets were listed, while five ETFs with16.7 billion total assets were listed.21 On average, the sizes of ETFs andLOFs are roughly the same.

The traded values of some of these funds are shown in Table 5.22

Shangzheng50ETF was the most actively traded funds, with 198 million

Table 5. Daily Traded Values of ETFs and LOFs as of March 30, 2007.

Fund Name Traded Value

(Million)

Fund Name Traded Value

(Million)

Shangzheng50ETF 198.0 Xingyequshi 6.8

HongliETF 43.0 Guangfaxiaofan 6.4

ZhongxiaobanETF 33.5 Shenzheng100ETF 4.6

Nanfanggaozheng 17.6 Zhongyinzhongguo 3.9

Changshengtongzhi 15.9 Boshizhuti 2.7

Jiashi300 15.1 Rongtongjuchao 2.6

Zhaoshangchengzhang 13.8 Jingshundingyi 1.5

Wanjiagongyong 10.2 Shangzheng180ETF 1.5

Penghuadongli 8.9 Penghuajiazhi 1.1

Fuguotianhui 7.9 Jingshunziyuan 0.6

Nanfangjipei 7.7 Heyinxiaolu 0.3


CNY changing hands. Shangzheng180ETF had the lowest traded valueamong ETFs, 1.5 million. The most actively traded LOF was Nanfang-gaozheng, 17.6 million, while the lowest value was a tiny 300 thousand forHeyinxiaolu. On average, ETFs were much more active than LOFs.

Several reasons may explain why LOFs are not as active as ETFs. First, itis cheaper to own and trade ETFs. For example, Shangzheng50ETF charges0.5% for management, and pays 0.1% for custodian, while the managementfee and custodian fee for Nanfangjipei are 1.5% and 0.25%, respectively.Second, ETFs attract institutional investors, such as fund managers, whileLOFs are more suitable to individual investors. The unit for purchasing orredeeming is one million shares (or units) for Shangzheng50ETF, forexample, which is too high in most cases for individual investors.

Given that traditional mutual funds are red-hot23 in China right now, it ispuzzling that LOFs are in some sense ignored by investors. Thisphenomenon is definitely worth further investigation.

7. CORPORATE BONDS WITH DETACHABLE

WARRANTS

According to the latest CSRC regulations,24 listed companies can issuecorporate BDWs, whose bond and warrant may be listed and tradedindependently. Even though it is called ‘‘detachable convertible bond’’ in theofficial directive, BDW is in essence a bundle (or package) of corporatedebts and stock warrants, which is frequently seen in international marketsbut really a new product in China’s financial market.

On November 29, 2006, China’s first BDW Magang Detachable began totrade on ShSE. As of January 31, 2007, three BDWs with a total issuing sizeof 9.9 billion CNY were on the markets (Table 6).

All three BDWs are AAA-rated, but only 06Zhonghua Debt is unsecured.The maturities of the warrants, which count from the first day of their

Table 6. Bonds with Detachable Warrants.

Bond Maturity

(Year)

No. of Warrant

Per Bond

Warrant Maturity

(Month)

Exercising

Style

06Magang Debt 5 23 MagangCWB1 24 Bermudan

06Zhonghua Debt 6 15 ZhonghuaCWB1 12 Bermudan

07Gangfan Debt 6 25 GangfanGFC1 24 Bermudan


trading, are much shorter than those of the corresponding bonds. Investorsshowed great interests in these BDWs when they were first offered to thepublic; only 9.15% of registered buyers of the new issue of 06Magang Debtwas allocated the security, while the numbers for 06Zhonghua Debt and07Gangfan Debt were even much lower, at 2.63% and 3.72%, respectively.25

The trading data for the debts and warrants from the BDWs for theclosing of January 31, 2007 are shown in Table 7.26 It can be seen clearlythat the trading of warrants was much more active than that of theircorresponding bonds. Two reasons may account for the difference. First, theunit of trading for bonds in terms of value is much higher than that ofwarrants, so bonds are less liquid, and thus less attractive to individualinvestors. Second, the trading of warrants is not subject to daily limits ofprice movements, and can be done on T+0.

8. ASSET-BACKED SECURITIES

The first exchange listed asset-backed security (ABS) with a size of 9.4billion CNY, the so-called China Unicom CDMA Network Leasing IncomePlan, was issued by China International Capital Company for ChinaUnicom in September 2005. In December 2005, two OTC ABSs were issuedby China Credit Trust Company; they were the 4.1 billion Kaiyuan 2005-#1Loan-Backed Security supported by commercial loans from ChinaDevelopment Bank, and the 3 billion Jianyuan 2005-#1 Mortgage-BackedSecurity based on household mortgages from China Construction Bank.

In 2005, four ABSs with 17.1 billion CNY assets were issued. Thefollowing year, 10 ABSs with 28 billion assets were issued (an increment of64% over the previous year), seven of which with 16.4 billion were traded on

Table 7. Trading of Bonds with Detachable Warrants as ofJanuary 31, 2007.

Bond Value Traded

(Million)

Closing Price Warrant Value Traded

(Million)

Closing Price

06Magang

Debt

19.01 84.85 MagangCWB1 1,449 2.38

06Zhonghua

Debt

7.28 82.07 ZhonghuaCWB1 290 4.62

07Gangfan

Debt

8.65 83.95 GangfanGFC1 431 2.34


ShSE and SzSE. Among those issuances in 2006, seven ABSs were issued byseven securities companies, and three by one trust company.27

As of January 2007, 14 ABSs (roughly classified into five kinds) are on themarket (Table 8).28 The largest single issuance is over 10 billion CNY, whileseveral products are issued on a rolling basis. Among these, only Jianyuan,Kaiyuan, Dongyuan, and Xinyuan are traded among banks (or on the OTCmarket).

ABSs on the international markets are usually OTC products, but most ofthe ABSs in China are traded on either ShSE or SzSE. The trading of ABSon exchanges in China is not very active, however. For example, ShSE had atotal of 10 trades of ABS in January 2007, with seven trades done onJanuary 31st alone; the number for SzSE for the same month was 19.

Table 8. Asset-Backed Securities as of January 2007.

Product Name Issuer Asset (Billion)

China Netcom Account

Receivable-Backed Certificate

China International Capital 10.3

China Unicom CDMA Network

Leasing Income Plan

China International Capital 9.4

Kaiyuan 2006-#1 Loan-Backed

Security

China Credit Trust Co., Ltd. 5.7

Xinyuan 2006-#1 Re-Structuring

Asset Trust


Kaiyuan 2005-#1 Loan-Backed

Security


Jianyuan 2005-#1 Mortgage-

Backed Trust


China Huaneng Lancangjiang

Electricity Income Plan

China Merchants Securities Co. 2.0

Jiangsu Wuzhong BT Project

Repurchase-Fund Certificate

CITIC Securities Co., Ltd. 1.7

Dongyuan 2006-#1 Re-structuring

Asset-Backed Security


Nantong Tiandian Electricity

Sales-Backed Certificate

Huatai Securities Co., Ltd. 0.8

Nanjing Public Water Treatment

Fee Plan

East Securities Co., Ltd. 0.7

Guan-Shen Freeway Toll Income

Plan

Guangdong Securities Co., Ltd. 0.6

Yuandong First Phase Leasing

Income Plan

Orient Securities Co., Ltd. 0.5

Pudong Road-Bridge BT Project

Income Plan

Guotai Junan Securities Co., Ltd. 0.4


Those trades, which had quite large sizes, were done by institutional investors.Apparently, small investors lack either the necessary capital or knowledge totrade those products, so new ways of trading may need to be invented in orderto attract more investors and increase the liquidity of this new market.

9. COMING SOON: MARGIN TRADING AND

FINANCIAL FUTURES

As was mentioned earlier, T+0 trading is not allowed in China. In addition,short-selling is currently forbidden on China’s markets. It is well known to stu-dents of finance that without short-selling, no arbitrage could not be enforced.Government officials are aware of those problems, and credit trading andfinancial futures are being considered and may be introduced very soon.

According to the draft rules regarding short-selling and margin-buying,29

the standard of trading on margin is quite high. For example, margin tradecan be only done with a list of stocks chosen by an exchange. The initialmargins for margin-buying and short-selling are both 50%, while themaintenance margins are 30%. For any security, margin trading will behalted as soon as over 25% of its tradable shares are either bought onmargin or sold short, and then not resumed until this percentage is below20%. For short-sell orders, the offer price should not be lower than the mostrecent trading price of the stock.28

One important innovation will be that, in addition to common stocks,bonds (government and corporate), mutual funds, and other securities listedon the exchange can also be used in margin trading.

Financial futures will be another important development. With the rapidadvance of China’s financial markets, international investors are increasinglyturning to China for investment opportunities, and several stock indexfutures are already introduced overseas. On October 1, 2004, Chicago BoardOptions Exchange launched the first China related futures with the CBOEChina Index as the underlying. Hong Kong Exchanges and Clearingintroduced futures and options on the FTSE/Xinhua China 25 Index onMay23, 2006.30 Singapore Exchange on September 5, 2006 started trading theso-called A50 Index futures, the first futures based on a stock index of theA-share market, the FTSE Xinhua China A50 Index (Shen & Zhang, 2006).

ChinaFinancialFuturesExchange (CFFE)was establishedonSeptember8,2006 to prepare for the launching of stock index futures. Currently, investorscan simulate the trading of futures on terminals set up by brokerage firms.The underlying is the ShSE–SzSE 300 Index by China Securities Index


Company. The minimum price change for one contract is 30 CNY(corresponding to 0.1 point of the index), and contracts with four deliverymonths (the current month, next month, and two near months from March,June, September, and December) are traded. One contract costs 30 CNY totrade. At the end of January 2007, CFFE had 155 members and over 100thousand registered accounts (Li, 2007b).

The required margin is 8%. Unlike the well-known S&P 500 Stock Indexfutures, this futures contract does not offer a lower maintenance margin.Accounts are marked-to-market and settled on every trading day without anyloans outstanding.

Five minutes before the opening of the market, orders are accepted andcollectively matched to determine an open price; while the market is open,orders are matched continuously by computer. To determine the settlementprice, prices from the final hour of trading are averaged using tradingvolumes as weights; for comparison, the S&P 500 Stock Index futures uses thefinal 30 seconds of trading to set the settle price. In addition to the daily limitsof 10% price change (up or down), trading will be halted at 6% for 10min.31

The simulated trading of ShSE–SzSE 300 Index futures showed someunique characteristics. For example, the prices of futures sometimes did notobey the well-known no arbitrage conditions, the basis (or the pricedifference between spot and futures) did not decrease with the passing oftime, and the trading of far-month contracts was abnormally active (Meng,2006). Two reasons may explain those phenomena. First, investors used upof all their allocated capitals during simulation, which did not incur reallosses. Second, investors had been quite bullish about the stock market sinceearly 2006. To curb that speculative behavior, the exchange recently raisedthe margin for the IF0704 contract32 from 12 to 15%, for example.

It is widely expected that the trading of stock index futures will begin inJune (Du, 2007), after the revised rules for trading futures become effectiveon April 15th. The rules will be applicable not only to commodity and stockindex futures, but also to futures and options on securities, interest rates,foreign exchange rates, and their related indices (if applicable),33 and set thestage for the development of China’ derivatives markets.

10. CONCLUSIONS

As is evident from the above, the speed of developments of China’s financialmarkets is astounding. Nevertheless, to become fully functioning, China’ssecurities markets still face many obstacles.


One such obstacle is the insignificant size of the debt markets. The allimportant (risk-free) interest rates are still not allowed to float according tosupply and demand. A welcome development in this regard is the launch ofShIBOR, or Shanghai Inter-Bank Offer Rate, on January 4, 2007 (Guo &Yu, 2007). On the other hand, the corporate bond market is not welldeveloped either, and non state-owned companies have had trouble obtainingfinancing for years.

History shows that financial innovations are primarily driven by marketdemands. To overcome the obstacles faced by the Chinese financial markets,the government should probably consider allowing market forces to worktheir wonders and placing fewer restrictions on financing, securitization, andmarket innovations.

NOTES

1. Numbers computed and figure plotted using data from Guotai Junan Securities(www.gtja.com).2. Numbers computed using data from Yahoo! Finance (finance.yahoo.com).3. CNY is the unit of Renminbi (RMB). If not indicated explicitly, CNY is

assumed to be the unit of currency throughout this article.4. Bloomberg News. China: Economy Grew 10.7% in 2006. The New York Times,

January 25, 2007.5. Refer to Liu (2007) as well as Neftci and Menager-Xu (2007) for the definition

of A-share and other details about China’s financial market.6. Source: China Securities Depository and Clearing Corporation (CSDCC), Ltd.7. At the end of April 2007, the average P/E ratio was 47.6 for Shenzhen Stock

Exchange and 53.2 for ShSE.8. The talk, titled China’s Rise and the Changing Pattern of APEC Capital Flows,

was given by John Edwards of HSBC Australia.9. Officially called the Split Share Structure Reform.10. Share Reform is considered to be finished by the end of 2006.11. SzSE stands for Shenzhen Stock Exchange. Data from Dragoninfo Financial

Information References System (www.dfirs.com).12. Convertible bonds have been traded on the stock exchanges for many years

(Liu, 2007).13. Hereafter the term warrant will be used to mean both when it is not necessary

to distinguish them.14. The exchange-assigned abbreviation for a warrant is used here, which has the

following pattern: Pinyin of the first two Chinese characters, three (Roman) letters,and one digit.15. Center for Fund Research, China Galaxy Securities. The Chinese Securities

Investment Funds Annual Report 2006. China Securities Journal (in Chinese),January 8, 2007.


16. The Pinyin of the exchange-assigned abbreviation for a fund’s name is usedhere.17. SPDRs is not an open-end company, but a Unit Investment Trust.18. Data from Investment Company Institute (www.ici.org/funds/abt/faqs_

etfs.html).19. The exchange-assigned abbreviation for an ETF is used here, which has the

following pattern: Pinyin of the initial Chinese characters, which might be followedby a number, and then ‘‘ETF.’’20. Rules enacted by Shanghai Stock Exchange.21. Numbers computed using data from Financial Street Holding Co., Ltd.

(www.jrj.com).22. Data from Guotai Junan Securities (www.gtja.com).23. 334 thousands new mutual fund accounts were opened on March 6 alone

(Li, 2007a).24. CSRC Directive No. 30, Rules for Issuing Securities by Listed Companies

(in Chinese), May 8, 2006.25. Data from Shenzhen Securities Information Co., Ltd. (www.cninfo.com.cn).26. Numbers computed using data from Guotai Junan Securities (www.gtja.com).27. Numbers computed using data from China Government Securities Depository

Trust and Clearing Co., Ltd. (www.chinabond.com.cn), ShSE, and SzSE.28. ShSE, Rules for Trading on Margin (Pilot version), August 21, 2006.29. CSRC No. [2006] 69, Rules for Providing Short-Selling and Margin-Buying

Services by Securities Companies (Pilot version), 30 June 2006.30. FTSE/Xinhua Index Limited (www.ftse.com/xinhua).31. CFFE, Rules for Simulated Trading of Stock Index Futures, October 30, 2006.32. Futures contract on ShSE–SzSE 300 index with a delivery month of April

2007.33. The State Council, Rules for Futures Trading, March 16, 2007.

REFERENCES

Bremner, B. (2007). Talking investors down from China high. BusinessWeek (February 1).

Du, Z. (2007). Stock index futures will trade in June. Securities Times (in Chinese), March 20.

Guo, F., & Yu, L. (2007). ShIBOR based pricing in financial markets. China Securities Journal

(in Chinese), January 4.

Hull, J. C. (2003). Options, futures, and other derivatives (5th edn.). Upper Saddle River, New

Jersey: Prentice Hall.

Jin, Q., & Yuan, H. (2006). An empirical analysis of the factors affecting the price in Share

Reform. Working Paper (in Chinese).

Li, J. (2007a). New investors lined up to purchase newly established mutual funds, and

daily numbers of newly registered accounts reached a new high. No. One Financial Daily

(in Chinese), March 8.

Li, Z. (2007b). Over 100 thousands accounts registered for simulated trading of index futures.

China Securities Journal (in Chinese), March 8.

Liu, Q. (2007). China’s convertible bond market. In: S. N. Neftci & M. Y. Menager-Xu (Eds),

China’s financial markets: An insider’s guide to how the markets work (pp. 171–185).

Burlington, MA: Elsevier Academic Press.


Liu, Y. (2006). The highly risky warrant. Changzhou Daily (in Chinese), December 26.

Lu, Z., Niu, J., & Huang, J. (2007). Goldman Sachs: No bubbles in China’s stock markets.

China Securities Journal (in Chinese), February 6.

Meng, Y. (2006). Chinese characteristics of stock index futures. No. One Financial Daily

(in Chinese), December 6.

Mitchell, T. (2007). China’s love affair with warrants. The Financial Times, available at

www.ft.com, article dated January 4.

Neftci, S. N., & Menager-Xu, M. Y. (2007). China’s financial markets: An insider’s guide to how

the markets work. Burlington, MA: Elsevier Academic Press.

Norris, F., & Peters, J. W. (2007). Wall St. Tumble adds to worries about economies. The New

York Times (February 28).

Shen, S., & Zhang, S. (2006). Singapore starts trading A-share futures before China. Bloomberg

(September 5).

Wu, J., & Ruan, T. (2004). The structure of split shares and financing behavior of China’s listed

companies. Journal of Financial Research (in Chinese), 6, 56–67.

Yang, G., & Ma, J. (2007). Lowest ratio of time value to stock price among global warrant

markets. Shanghai Securities News (in Chinese), January 11.

Yardley, J. (2007). Chinese united by common goal: A hot stock tip. The New York Times

(January 30).


CHAPTER 13

TEMPORAL CAUSALITY OF

RETURNS OF INDEX FUTURES

AND STOCK MARKETS: EVIDENCE

FROM MALAYSIA

Wee Ching Pok

ABSTRACT

This chapter investigates the impact change of the composition of market

agents on the timing of the arrival of information in Bursa Malaysia. The

price discovery role of futures trading on the spot market is examined

through three distinct sub-periods: pre-crisis, crisis and after capital

controls. For this purpose, the Johansen Cointegration (1988, 1991) and

VECM and Granger causality are used. The analysis shows that there is

no significant long-run relationship. As for short-run, the results show

futures lead spot. However, futures’ lead is shorter in pre-crisis and crisis

periods where foreign institutional investors dominate. This study deduces

that the significant change in the composition of market agents could

contribute to the variation of lead–lag relationship.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00013-1

263

dx.doi.org/10.1016/S1569-3767(07)00013-1.3d

1. INTRODUCTION

The Asia-Pacific region has been one of the main beneficiaries of theincrease in international investments by large institutional investors fromdeveloped countries. Before the Asian financial crisis in 1997, the Asianregion had captured more than 45% of the net private capital inflows to theemerging markets. The main pull-factor has been the growing deregulationand financial liberalisations undertaken by less developed emerging econom-ies in the late 1980s which has given foreign investors the opportunity toinvest in domestic shares. This, however, has led to excessive fluctuations instock prices in the equity market.

TheMalaysian derivative market developed in response to the economic riskassociated with the equity market. High volatility led to a demand for hedginginstruments/risk management tools to protect investments. One such is thestock index futures contract. In new emerging derivative markets, investorsneed time to become acquainted with derivatives and the benefits which theycan offer. There are also bottlenecks and restrictions which the regulatorybody needs to overcome before the market can grow. The public at large andthe companies in particular are also expected to be receptive to the newopportunities for investment. In Malaysia, stock index futures trading waslaunched on 15 December 1995. To the extent that index futures provideshedging and profit earning opportunities, the introduction of futures tradingwas expected to play the role of price discovery; price discovery means thatthe futures market reflects new information before the spot market does.To investigate this, the lead–lag or causality relationship between the pricemovements of stock index futures returns and the underlying spot marketreturns is examined. The lead–lag relationship between the price movementsof stock index futures returns and the underlying spot market returns willillustrate how fast one market absorbs new information relative to the other.Understanding the process by which new information is incorporated into thespot and futures prices not only allows market agents to use the leading marketas a source of price discovery but also offers an investment strategy to takeequity positions and allow hedging. Studying the process is interesting because,while the local investors take time to get to know the market, the market’sperformance is greatly influenced by the participation of foreign institutionalinvestors. Hence, this study explores the possibility that the participation offoreign investors influences the lead–lag or causality relationship between theprice movements of the stock index futures and the underlying spot.

There is voluminous literature on the lead–lag relationship between the spotmarket and the futures market. Previous research has shown that derivative

WEE CHING POK264

markets in developed countries are more efficient at incorporating newinformation because the market agents in these markets are better acquaintedwith derivative securities. Most studies have shown that returns of futuresmarkets significantly lead those of spot markets. Generally, most of theempirical results show a bi-directional asymmetric lead–lag relationship, thatis, a strong lead of the futures over the spot and a weak feedback effect of thespot lead the futures.1 There is also evidence to suggest uni-directional lead–lag relationship where the evidence shows futures leads spot or spot leadsfutures.2 So far, only Wahab and Lashgari (1993) report stronger evidence ofspot market leads futures. Thus, the research has provided evidence of theprice discovery role of the futures market with different timing of leads. Thisimplies that information flows to futures markets are faster than to the stockmarket and the two markets are not contemporaneously correlated.

While many of the studies focus on timing issues, many also investigatethe hypotheses that explain the possible causes of variation of lead–lagrelationship between the futures and the spot markets. Chan (1992) andStoll and Whaley (1990) argue that many of the component stocks in anygiven index are not traded frequently enough to allow prices to updateinformation quickly. The futures price, on the other hand, adjustsinstantaneously to new information. Frino and West (1999) discuss marketmaturation as another factor whereby the futures lead over the spot marketdeclines as the futures market matures through time because stock andfutures markets become more integrated (in a more correlated sense) as thefutures market matures. According to Abhyankar (1995), the relativedifferences in liquidity between the index and futures markets could alsoinduce a lead–lag relationship. It is said that the average time between tradesfor component stocks in the market index is longer than the average timebetween trades for the futures contract. Fleming, Ostdiek, and Whaley(1996) and Kim, Szakmary, and Schwarz (1999) argue that informed tradersare more attracted to derivatives markets because of the leverage andtransaction cost benefits. They demonstrate that the cost of taking aposition in the stock index futures is considerably lower than the cost oftaking an equivalent position in stocks. Hence, on average, informed tradersare more likely to trade in stock index futures market and price movementsin stock index futures are likely to precede price movements on stocks.Different ‘‘market architectures’’ can also give rise to the futures marketsleading the spot markets. Grunbichler et al. (1994) argue that screen-tradedmarkets enhance price discovery by reducing trading costs, reducing time toexecute orders, reducing time to disseminate trade information andincreasing the quality of information reported to the market. Moreover,

Temporal Causality of Returns of Index Futures and Stock Markets 265

the anonymity provided by screen trading has induced traders in futuresmarket to trade more frequently than in the spot. In some situations,informed traders may choose to trade in the spot market rather than in thefutures markets. Chan (1992) and Frino, Walter, and West (2000) show thatif an informed trader has firm-specific information, it may be optimal totrade the shares of the firm directly rather than trade on the futures index.Thus, for some types of information, the transmission of information mayrun from the spot to the futures market. As such, this raises the possibility ofa bi-directional lead–lag relationship between the futures and spot returns.

In the case of Malaysia, Tan (2002) investigates the causality relation-ship between the Malaysian stock market and the futures market andfinds bi-directional short-run causality for the periods before and after theimplementation of selective capital controls. For the long-run relationship,he finds evidence of the futures market leading the spot (being weaklyexogenous) in the period before the implementation of selective capitalcontrols but not in the period after their implementation.3

The present study offers evidence of the kind provided in previous studiesby Ghosh (1993), Wahab and Lashgari (1993), Tse (1995), Pizzi et al. (1998),Brooks, Rew, and Ritson (2001) and Tan (2002), by looking at the pricediscovery role of the Malaysian stock index futures market through threedistinct sub-periods, instead of two sub-periods: pre-crisis (15 December1995 to 31 July 1997), during the crisis (1 August 1997 to 14 September1998) and after the enforcement of capital controls (15 September 1998 to 31July 2001). These subdivisions allow comparison of the price discoveryrole of futures returns in different situations, in particular where there issignificant presence of foreign institutional investors. To examine the lead–lag or the causality relationship, the Johansen (1988, 1991) cointegrationtest and VECM are used. The Granger causality test is used to examine thedynamics of the returns. It is believed that this analysis differs from that ofTan (2002) in terms of the robustness of its econometric procedures incarrying out the cointegration tests. The roots of the companion matrix andthe recursive least squares diagnostics tests on the residuals of the VARmodel and the Chow tests of parameter stability are examined before testingfor cointegration. A comprehensive econometric procedure has been carriedout to ensure that the cointegration results and the causality tests are valid.The possibility that foreign institutional investors influenced the timing ofthe arrival of information is explored. In the initial phase of futures trading,the local investors including the institutional investors had not been tradingactively and hence the market was dominated by foreign institutional inve-stors. Although the Securities Commission has provided support through

WEE CHING POK266

the policy changes, the situation did not change substantially even duringthe Asian financial crisis. The situation changed, however, when theMalaysian government imposed selective capital controls, which led to largewithdrawals of foreign funds from the futures and spot markets. The authorbelieves that the participation and/or non-participation of certain groups ofmarket agents, in particular groups who are more experienced, influence thetiming of the arrival of information in the markets. So far, no research hasexplored the possible impact of change in the composition of market traderson the speed at which information is transmitted. The author conjecturesthat this attempt has widespread appeal because this phenomenon isevidential in most emerging markets where foreign institutional investorsare actively participating in the markets.

The findings suggest that there is no cointegration between the spot andfutures markets. Hence, there is no significant long-term relationship or pricelinkage between the spot and the futures markets. However, in the short-term, the Granger causality test reveals that these markets are significantlylinked to each other in a uni-directional manner. Further, there is evidence tosuggest that the futures markets lead or cause changes in the spot markets inall the sub-periods examined. This study also finds that the futures lead isshorter in the pre-crisis and crisis periods (as seen with t=�2 during the pre-crisis and t=�1 during the crisis) when the foreign institutional investorswere participating actively in the Malaysian market compared to the periodafter the crisis (as seen with t=�11 after the imposition of capital controls)when foreign investors withdrew their funds. This also implies that the speedof transmitting information is hastened when foreign institutional investorsdominate the market.

The rest of this chapter is organised as follows. The following sectionprovides background details about the performance of stock index futurestrading and the profile of the market traders. Data used in this study arediscussed in Section 3. Empirical findings are discussed in Section 4; andSection 5 provides a summary of the chapter.

2. BACKGROUND ON THE STOCK AND

FUTURES MARKETS

At its launch, on 15 December 1995, the KLSE was the largest bourse in theASEAN-5 region in terms of market capitalisation and the third largest inthe Asia-Pacific region after Hong Kong and Australia. The KLOFFE is thethird futures exchange to open in Asia after Singapore and Hong Kong.


Figs. 1 and 2, respectively, present the graph of the monthly average dailytrading volume and open interests after the launch of futures trading and theprofile of traders of KLOFFE. As shown in Fig. 1, for the first six monthsafter the launch of the futures trading, the average daily trading volumestands between 200 and 300 contracts per day. As shown in Fig. 2, the bulkof the volume of trade came from foreign institutional investors. Theycontributed 54% of the trade. Local independent traders who trade on theirown account constituted 20% of trade, local retail investors made up 15%and foreign retail investors and proprietary trading by members eachcomprised 5% of total trading. Disappointingly, local institutional investorsonly accounted for 1% of the trade.4 Thus, on September 1996, toencourage greater participation of local investors, the Securities Commis-sion proposed the following: (1) a dual-licensing scheme for security dealerrepresentatives so that they can trade in both spot and futures markets,(2) introduce market makers to improve futures market liquidity, (3) moreeducational packages for local investors and intermediaries and (4)collateral lodgement by the spot market players as a cover for initialmargins. However, by December 1996, as shown in Fig. 2, the profile oftraders in the KLOFFE had not changed significantly though the averagedaily trading volume continued to rise.5

From May 1997, the average daily trading volume began to hit the liquidlevel of 2,000 contracts per day.6 Despite this, the foreign institutionalinvestors continued to dominate, while the local fund managers did notmuch use futures contracts to hedge to reduce the risk exposure in the spotmarket.7 In August 1998, amidst the Asian financial crisis, the marketperformance began to hit new record levels with the average daily tradingvolume and the open interests at 4,650 and an outstanding 23,000 contractsper day, respectively.8 Nonetheless, the performance of the futures marketcould not be sustained and it took a drastic turn when Bank NegaraMalaysia announced its measures of selective capital controls purportedly(effective 1st September 1998 but which actually took effect only after 14thSeptember 1998) to prevent the Ringgit being attacked further by currencyspeculators. The implementation of selective capital controls caused largeforeign funds to be withdrawn from local markets. In the first three weeksafter the implementation of currency controls, foreign institutions’contribution to futures trading had drastically reduced to only 4%. Thecontribution of local retail investors, however, increased from 31% to 63%,but this significant increase was partly due to the decline in total tradingvolume. The average daily trading volume and the average open interestsdropped significantly to a dismal 800 and 1,600 contracts per day in October

WEE CHING POK268

0

5,000

10,000

15,000

20,000

25,000

Nu

mb

er o

f C

on

trac

ts

VolumeOpen Interest

Time

Dec

-95

Mar

-96

Jun-

96

Sep-

96

Dec

-96

Mar

-97

Jun-

97

Sep-

97

Dec

-97

Mar

-98

Jun-

98

Sep-

98

Dec

-98

Mar

-99

Jun-

99

Sep-

99

Dec

-99

Mar

-00

Jun-

00

Sep-

00

Dec

-00

Mar

-01

Jun-

01

Fig. 1. Monthly Average Daily Trading Volume and Open Interests. Source:

www.mdex.com.my

0%

10%

20%

30%

40%

50%

60%

Jun-

98

Mar-

99

Jun-

00

Sep-0

0

Foreign Institutions

Local Institutions

Overseas Retail

Local Retail

Locals

Proprietary

Time

Per

cen

tag

e

Mar

& June

96

Sep-9

6

Dec-9

6

Mar-

97

Jun-

97

Sep-9

7

Dec-9

7

Mar-

98

Sep-9

8

Dec-9

8

Jun-

99

Sep-9

9

Dec-9

9

Mar-

00

Dec-0

0

Mar-

01

Jun-

01

Sep-0

1

Dec-0

1

Fig. 2. Profile of Traders of KLOFFE. Source: www.mdex.com.my


1998 (see Fig. 1).9 After the imposition of the selective capital controls, therewas a significant drop in participation of foreign institutional investors andlocal retail investors began to play a more significant role in the market.

3. DATA

The data used in this study are daily closing spot and futures prices10 for theKLSE CI for the period between 15 December 1995 and 31 July 2001adjusted for public holidays. This time period is particularly interestingbecause within it, three distinct periods of differing performances areobserved in terms of (1) the average daily trading volume and open interestfor each of the months (see Fig. 1), (2) the monthly standard deviations ofthe spot and futures returns (see Fig. 3). The period between December 1995and July 1997 was when the local traders began to get acquainted with themarket and the mechanics of trading futures. Hence, the performance fromlocal retail investors during this period was uninspiring. Also during thisperiod, local institutions did not participate actively in the trading offutures. Hence, the futures market was dominated by foreign institutionalplayers. Then, between August 1997 and September 1998, Malaysia had tocope with the financial crisis. To counter the effect of the high volatility in

0.000

2.000

4.000

6.000

8.000

10.000

12.000

14.000Spot ReturnsFutures Returns

Per

cen

tag

e

Time

Dec

-95

Mar

-96

Jun-

96

Sep-

96

Dec

-96

Mar

-97

Jun-

97

Sep-

97

Dec

-97

Mar

-98

Jun-

98

Sep-

98

Dec

-98

Mar

-99

Jun-

99

Sep-

99

Dec

-99

Mar

-00

Jun-

00

Sep-

00

Dec

-00

Mar

-01

Jun-

01

Fig. 3. Monthly Standard Deviations for Spot and Futures Returns. Source:

www.mdex.com.my

WEE CHING POK270

the spot market, the market traders, particularly the local retail investors,were observed to begin using futures for hedging and speculation. Theperformance of the futures in this period improved by three times more overthat of the pre-crisis period. However, the futures market was stilldominated by foreign players (see Fig. 2). On 1st September 1998, BankNegara Malaysia announced the imposition of selective capital controls inMalaysia.11 After the imposition, the average daily futures trading volumeand average daily futures open interests plunged sharply. However, theysoon stabilised with higher trade levels compared to those of the pre-crisisperiod (see Fig. 1). Similarly, the monthly standard deviations for spot andfutures returns escalated after July 1997 and stabilised after September1998 – also at much higher standard deviation levels compared to those ofthe pre-crisis period (see Fig. 3). Taking these factors into account, it isuseful to divide the sample into three distinct periods for analysis: the pre-crisis period, from 15 December 1995 to 31 July 1997; the crisis period, from1 August 1997 to 14 September 1998; and after selective capital controlsperiod, from 15 September 199812 to 31 July 2001.

The daily returns of the KLSE spot and futures are computed by taking thelogarithm of the ratio between the current price and that of the previous periodand multiplying the result by 100. There were 1,386 total daily observations inthe period. The KLSE CI spot was collected from DataStream Internationaland the futures data were obtained from the Malaysia Derivatives website.Stock index futures prices are those of the nearby contract.

4. DATA ANALYSIS AND EMPIRICAL RESULTS

4.1. Descriptive Statistics

Table 1 presents descriptive statistics for the returns series of the spot andthe futures of the entire sample and Table 2 presents the descriptive statisticsfor the return series of the spot and futures for the sub-periods. The resultshown in Table 1 reveals that the mean (�0.029) returns of the spot ismarginally higher than the mean (�0.030) returns of the futures. Thestandard deviation (2.189) of the returns of the spot, however, is lower thanthe standard deviation (2.721) of the returns of the futures. This is expectedbecause futures prices are more sensitive to information than spot prices. Interms of skewness, the spot records a positive value (0.537) as opposed tofutures (�0.747) which is negative and the kurtosis of spot (29.910) is muchlower than that of the futures (46.214). On closer examination, via the


histogram, it is found that the extreme minimum value of the daily futuresreturns (�38.839) had impacted the skewness and the kurtosis. The highkurtosis in both of the daily spot and daily futures returns, consequently ahigh Jacque–Bera, rejects the null hypothesis of normal distribution. Itwould seem that the crisis and the imposition of selective capital controls

Table 1. Descriptive Statistics for the Return Series of Spot (st) andFutures (ft) for the Period 15 December 1995 to 31 July 2001.

st ft

Mean �0.029 �0.030

Median �0.071 �0.093

Maximum 20.817 28.781

Minimum �24.153 �38.839

Standard Deviation 2.189 2.721

Skewness 0.537 �0.747

Kurtosis 29.910 46.214

Jarque–Bera 41857.280 107894.200

Probability 0.000 0.000

Observations 1385 1385

Table 2. Descriptive Statistics for the Return Series ofSpot (st) and Futures (ft).

st ft

Pre-Crisis Crisis After Capital

Controls

Pre-Crisis Crisis After Capital

Controls

Mean 0.006 �0.342 0.073 0.005 �0.335 0.070

Median 0.003 �0.651 �0.033 �0.018 �0.751 �0.075

Maximum 2.675 20.817 6.523 2.916 28.781 7.991

Minimum �3.329 �24.153 �6.342 �4.244 �38.839 �7.337

Standard

Deviation

0.889 4.058 1.565 0.963 5.112 1.927

Skewness �0.321 0.622 0.045 �0.453 �0.450 0.338

Kurtosis 4.578 12.575 5.096 5.075 18.243 4.857

Jarque–Bera 48.234 1075.917 130.005 85.212 2691.176 115.403

Probability 0.000 0.000 0.000 0.000 0.000 0.000

Observations 399 277 709 399 277 709

Pre-crisis sample period – from 15 December 1995 to 31 July 1997.

Crisis sample period – from 1 August 1997 to 14 September 1998.

After capital controls sample period – from 15 September 1998 to 31 July 2001.

WEE CHING POK272

significantly influenced the distribution of returns of the daily spot and thedaily futures, particularly, the extreme values.

The results in Table 2 show that the mean returns of the spot during thepre-crisis and after selective capital controls (respectively, 0.006 and 0.073)are relatively higher than the mean returns of the futures (respectively, 0.005and 0.070). Nonetheless, during the crisis, spot suffers higher losses (�0.342)than the futures (�0.335). With respect to standard deviation, the spot’sstandard deviation was relatively lower than the futures for all the three sub-periods analysed (respectively, 0.889, 4.058, 1.565 vs. 0.963, 5.112, 1.927).These results also revealed that both spot and futures had higher volatilityduring the crisis and significantly reduced volatility after the selective capitalcontrols. As for skewness of returns, during the pre-crisis, it was negative forboth spot and futures (respectively, �0.321 and �0.453). The spot returnswere positively skewed (0.622) contrary to negatively skewed (�0.450)futures returns during the crisis. Again this negative skewness is due to theinfluence of extreme negative values although a greater part of the futuresreturns was in the negative. However, after the selective capital controls, thedistribution of returns of futures reverts back to positive skewness as a resultof the large withdrawal of foreign institutional investors from the futuresmarket. As expected, the kurtosis is comparatively higher during the crisisthan during the pre-crisis and after selective capital controls which leads tohigh Jacque–Bera statistics. The performance of mean, standard deviation,skewness and kurtosis of both the futures and spot markets improved aftercapital controls.

4.2. Unit Root and Structural Break Tests

The results are highly conclusive; the Augmented Dickey and Fuller (ADF,1981) and Phillips and Perron (PP, 1988) tests reveal that the null hypothesisof non-stationary or I(1) or unit root is accepted for the logarithm of dailyspot, logSt, and logarithm of daily futures, logFt at 5% significance levels.13

The results are also highly conclusive with the null hypothesis of non-stationary or I(1) or unit root is rejected for the first difference of thelogarithm of daily spot, st and logarithm of daily futures, ft at 5%significance levels.14 Thus, overall, these variables are determined to be non-stationary or integrated of order 1 (or I(1)). In estimating the regression withtrend and constant, it is found that the series examined has a unit root, i.e.,non-stationary or I(1). It can thus be concluded that the logarithm of dailyspot and daily futures exhibits stochastic trends.


The results of unit root tests on log price series and first difference seriesof the spot and futures for the sub-periods: pre-crisis, crisis and after capitalcontrols are also examined. The results are highly conclusive. Respectively,the ADF and PP tests reveal that the null hypothesis of unit root or I(1)is accepted for the logarithm of daily spot, logSt, and logarithm ofdaily futures, logFt

15 and that null hypothesis of unit roots or I(1) is rejectedfor the first difference of the logarithm of daily spot, st and logarithm ofdaily futures, ft at 5% significance levels for all the three sub-periodsanalysed.16

Since the logarithm of daily spot, logSt, and the logarithm of dailyfutures, logFt are non-stationary or I(1) for the entire sample period andalso for the sub-periods, the Perron (1997) test can now be conducted tocheck the possibility of a series becoming stationary or I(0) once thestructural break was accounted for. Tables 3 and 4, respectively, present theresults of the Perron (1997) tests for structural break for the whole periodand for the sub-periods. As shown in Table 3, all the models (IO1, IO2 andAO), with their different methods of determining the break (Tb),

17 confirmthat the alternative hypothesis of stationary or I(0) with structural break isrejected for the entire sample period. The Perron (1997) test also confirmsthat the log price series for all three sub-periods by and large are non-stationary or I(1) processes even after structural break is accounted for. Asseen in Table 4, in the period between observations 1 to 400 (15 December1995 to 31 July 1997), evidence of stationary or I(0) for logarithm of dailyspot, logSt, and logarithm of daily futures, logFt is found after taking intoaccount the structural break for Model IO2 and Method UR at the 1%significance level. However, the other counterparts did not show suchevidence and moreover, Tb determined by this method is very different fromTb determined derived by other methods. Hence, there is insufficientevidence to substantiate a stationary or I(0) process for both the series. Asfor the period between observations 401 and 677 (1 July 1997 to 14September 1998), evidence of stationary or I(0) for logarithm of daily spot,logSt, is found significant at 5% level after taking into account thestructural break for Models IO1 and method UR, IO1 and method Studabs,IO1 and method Stud and IO2 and method UR but not at the 1%significance level. Again, this evidence is insufficient to substantiate astationary or I(0) process for both the series. Hence, this test confirms thatthe logarithm of daily spot, logSt, and the logarithm of daily futures, logFt

are non-stationary or I(1) for the entire period as well as for the sub-periodsanalysed.

WEE CHING POK274

4.3. Statistical and Mathematical Stability Tests of VAR

In the examination of the stability of the VAR, statistically andmathematically, the following is found for the entire sample period.Although the roots of the companion matrix lie inside the unit circle,18

which indicate the VAR satisfies the mathematical stability condition, it isfound the required properties of residuals are not met19 and that it also failsthe Chow tests of parameter stability where the time pattern of theserecursive graphs exceeds the 5% critical value.20

Table 3. Test of Structural Break on Log Price Series of Spot (logSt)and Futures (logFt) for the Entire Sample Perioda.

logSt logFt

Tb Perron Test Statistics Tb Perron Test Statistics

Model IO1 and method UR 90 �4.3102 10 �4.1609

Model IO1 and method Studabs 1179 �2.7970 1179 �3.0536

Model IO1 and method Stud 1179 �2.7970 1179 �3.0536




Model AO and method UR 79 �4.3205 75 �4.0173

Model AO and method Studabs 1179 �2.9187 1179 �3.0330

Model AO and method Stud 1179 �2.9187 1179 �3.0330

�Critical values at 5% significant level, 100 observations.

Model IO1 and Method UR, �5.10, Model IO1 and Method Studabs, �5.05, Model IO1 and

Method Stud, �5.05.


Method Stud, �5.19�Critical values at 5% significant level, 200 observations.

Model AO and Method UR, �4.65, Model AO and Method Studabs, �4.41, Model AO and

Method Stud, �4.41.��Critical values at 1% significant level, 100 observations.






Method Stud, �5.02.aFrom 15 December 1995 to 31 July 2001.


Table 4. Test of Structural Break on Log Price Series of Spot (logSt)and Futures (logFt).

logSt logFt

Tb Perron Test Statistics Tb Perron Test Statistics

Pre-crisis (From 1 to 400a)




Model IO2 and method UR 59 �6.0842�� 59 �6.5728��



Model AO and method UR 79 �4.7615� 75 �4.0002



Crisis (From 401 to 677b)

Model IO1 and method UR 522 �5.4952� 522 �4.9559

Model IO1 and method Studabs 522 �5.4952� 515 �4.9271

Model IO1 and method Stud 522 �5.4952� 515 �4.9271

Model IO2 and method UR 522 �5.6672� 522 �5.3097






After capital controls (From 678 to 1386c)










�Critical values at 5% significant level, 100 observations.




Method Stud, �5.19.�Critical values at 5% significant level, 200 observations.








Method Stud, �5.02.aFrom 15 December 1995 to 31 July 1997 and n=400.bFrom 1 August 1997 to 14 September 1998 and n=277.cFrom 15 September 1998 to 31 July 2001 and n=709.

WEE CHING POK276

This study proceeds by examining the statistical and mathematicalstability of VAR over the sub-periods. In addition to the three sub-periodsanalysis, an additional analysis in the crisis period (called crisis-adjustedperiod) where 10 significant observations (i.e., observations 668 to 677 orfrom 1 September 1998 to 14 September 1998) which represent the result ofmarket overreaction due to imposition to capital controls are not includedas observations. Except for the crisis-adjusted period, the diagnostic testsresults of the residuals fail to indicate that these lag lengths are theappropriate ones. Significant combination effects in ARCH, heteroscedas-ticity and non-normality for sub-periods: pre-crisis, crisis and after capitalcontrols are observed.21 It is found that except for crisis-adjusted and aftercapital controls periods, the pre-crisis sub-period fails the 1-step Chow test,forecast Chow test and break point Chow test.22 This shows that theparameters of the VAR model are statistically unstable. However, the rootsof the companion matrix for all periods23 examined lie inside the unit circleimplies the VARs satisfy the mathematical stability condition.

4.4. Cointegration and Causality Results

With these constraints, Johansen’s (1988, 1991) cointegration test wasconducted to examine the long-run equilibrium relationship between thespot and futures market. First, the result of the regression of the seriesagainst time is reported.24 The results show a linear deterministic trend in theseries logarithm of daily spot, logSt and logarithm of daily futures, logFt.On account of this, Model 3 (that series has linear deterministic trend withthe constant in the cointegrating vector and also outside the cointegratingvector) of Johansen (1995) was chosen to test for cointegration.

Table 5 presents the results of the cointegrating test for the all sub-periodsexamined. As shown in Table 5, there seems to be evidence of onecointegrating vector (r=1 or reduced rank) in the pre-crisis, crisis and crisis-adjusted periods where the calculated trace statistic, ltrace, of no cointegrat-ing vector exceeds the 95% critical value of the trace statistic, given at 15.41.Hence, the null hypothesis of no cointegrating vectors (r=0) is rejected andthe alternative hypothesis of one or more cointegrating vectors (r>0) isaccepted. For the after capital controls period, the cointegration testprovides evidence of the two cointegrating vectors (r=2 or full rank) becausethe calculated trace statistic, ltrace, rejects the null hypothesis of onecointegrating vectors (r=1) and accepts the alternative hypothesis of one ormore cointegrating vectors (r>1).


Before conclusions are drawn, there is a need to query whether the abovecointegration results are valid. For a valid cointegration test, there is a needto confirm that the VARs are statistically and mathematically stable. If anyof these criteria are not met, the trace test statistic will be affected, whichmay lead to an invalid cointegration result. In the pre-crisis period, there isone cointegrating vector but the graphical outputs of the Chow tests ofparameter stability and the diagnostic tests of the residuals of VAR do notlend support to this result. No statistical stability is obtained although thereis mathematical stability (as observed in roots of companion matrix). Thesecriteria have not been met, hence, the calculated trace statistic, ltrace, isunsuitable in these instances.25 This implies that there is no long-runequilibrium relationship between spot and futures prices. There is no causaleffect of one market causing the other market in the long run. The same lineof argument applies to crisis, crisis-adjusted and after capital controlsperiods. In the crisis period, it shows one cointegrating vector but the resultsalso have no meaning. This VAR might appear to be mathematically stablebut it is, in fact statistically unstable (the diagnostic tests of residuals ofVAR and also the stability of coefficients conditions are not met). In the

Table 5. Tests for the Number of Cointegrating Vectors – ltrace Test.

Null Hypothesis Alternative

Hypothesis

ltrace ValueStatistic

95% Critical

Value

99% Critical

Value


r=0 r>0 24.6424�� 15.41 20.04

rr1 r>1 1.9920 3.76 6.65

Crisis (From 401 to 677b)

r=0 r>0 18.8852� 15.41 20.04

rr1 r>1 0.8831 3.76 6.65

Crisis-adjusted (From 401 to 677b)

r=0 r>0 19.9530� 15.41 20.04

rr1 r>1 0.0473 3.76 6.65


r=0 r>0 38.2156�� 15.41 20.04

rr1 r>1 7.8789�� 3.76 6.65

�Rejects null hypothesis at 5% significant level.��Rejects null hypothesis at 1% significant level.aFrom 15 December 1995 to 31 July 1997 and n=400.bFrom 1 August 1997 to 14 September 1998 and n=277.cFrom 15 September 1998 to 31 July 2001 and n=709.

WEE CHING POK278

case of the crisis-adjusted period, although these results are better than thatof the crisis period (where stability of coefficients in the Chow’s tests isfound), there is an inability to conclude that there appears to becointegration because the diagnostic test of the residuals of VAR has notfulfilled the criterion of normality (high Jacque–Bera statistics).26 Similarbut not identical remarks could be made about the results reported for theafter capital controls period. The cointegration tests results are not reliablefor similar reasons (see Table 5).

Hence, it is concluded that there seems no cointegration for any ofthe sub-periods examined. This is because the required conditions on theVAR – statistical and mathematical stability – are not met for any laglength. If there is no cointegration, the next best means to examine thecausality of both spot and futures returns is through a VAR in firstdifferences.

Tables 6–8, respectively, present the causal effect results (using VAR withfirst difference) for returns of spot and futures for pre-crisis, crisis-adjusted

Table 6. Causality Testing Results for the Pre-Crisis Perioda.

DlogSt DlogFt

Coefficients Standard

Error

t-Statistics Coefficients Standard

Error

t-Statistics

Constant 0.0081 0.0435 0.1854 0.0060 0.0482 0.1237

DlogSt�1 �0.1391 0.1162 �1.1969 0.1919 0.1288 1.4904

DlogSt�2 �0.2889 0.1207 �2.3932� �0.0927 0.1337 �0.6931

DlogSt�3 �0.1785 0.1192 �1.4980 0.0938 0.1320 0.7103

DlogSt�4 �0.1774 0.1198 �1.4808 �0.0640 0.1327 �0.4819

DlogSt�5 �0.2386 0.1134 �2.1053� �0.1414 0.1256 �1.1262

DlogFt�1 0.2901 0.1054 2.7533� �0.0603 0.1167 �0.5165

DlogFt�2 0.2641 0.1119 2.3595� 0.0492 0.1240 0.3966

DlogFt�3 0.1413 0.1112 1.2705 �0.0829 0.1232 �0.6731

DlogFt�4 0.0725 0.1114 0.6509 �0.0291 0.1235 �0.2355

DlogFt�5 0.1629 0.1056 1.5429 0.0723 0.1170 0.6180

Q(5) 0.2532 (0.9980) 0.2171 (0.9990)

Jacque–Bera 51.3471 (0.0000)�� 73.2431 (0.0000)��

B-G Serial Corr. LM

Test-F-Statistic

1.1427 (0.3200) 1.1492 (0.3180)

ARCH Test-F-Statistic 17.3491 (0.0000)�� 11.3572 (0.0008)��

White Test-F-Statistic 2.6439 (0.0000)�� 2.1624 (0.0029)��

�Rejects null hypothesis at 5% significant level.��Rejects null hypothesis at 1% significant level.aFrom 15 December 1995 to 31 July 1997 and n=399.


and after capital controls periods. The diagnostics tests of residuals (thebottom section of each of the tables), although not ideal, provide evidence ofthe best available lag length for causality analysis. The stability tests forcausality of returns of spot27 and causality of returns of futures28 provideevidence that the parameters are more or less stable for all the periodsexamined. In the case of the pre-crisis period, there is evidence of uni-directional causality. The results in Table 6 show that returns of futures att=�1 and t=�2 and returns of spot at t=�2 and t=�5 are significant inexplaining the current returns of spot; none of the lagged returns of spot andfutures are significant in explaining the current returns of futures.

With regards to the crisis-adjusted period, there is again evidence ofuni-directional causality. The results in Table 7 show that the returns offutures at t=�1 and returns of spot at t=�1 are significant in describingthe current returns of spot and none of the lagged returns of spot andfutures are significant in explaining the current returns of futures.

Lastly, with regards to the after capital controls period, there is also anobserved uni-directional causality up to t=�11 (see Table 8). The returnsof futures at t=�1, t=�2, t=�3, t=�5, t=�7 and t=�11 and returns ofspot at t=�5 are significant in describing the current spot returns and noneof the lagged returns of spot and futures are significant in describing thecurrent futures returns.

Table 7. Causality Testing Results for the Crisis-adjusted Perioda.

DlogSt DlogFt


Error


Error

t-Statistics

Constant �0.3861 0.1937 �1.9938� �0.4512 0.2402 �1.8786

DlogSt�1 �0.4949 0.1546 �3.2006� �0.1585 0.1917 �0.8265

DlogSt�2 �0.1691 0.1317 �1.2843 0.0282 0.1633 0.1725

DlogFt�1 0.5676 0.1256 4.5177� 0.1097 0.1558 0.7044

DlogFt�2 0.2305 0.1215 1.8972 �0.0133 0.1507 �0.0880

Q(5) 3.2463 (0.6620) 4.0268 (0.5460)

Jacque–Bera 881.2110 (0.0000)�� 97.2643 (0.0000)��

B-G Serial Corr. LM

Test-F-Statistic

1.0805 (0.3410) 0.5987 (0.5503)

ARCH Test-F-Statistic 0.0001 (0.9934) 0.0249 (0.8747)

White Test-F-Statistic 1.1834 (0.3093) 1.5842 (0.1297)

�Rejects null hypothesis at 5% significant level.��Rejects null hypothesis at 1% significant level.aFrom 1 August 1997 to 14 September 1998 and n=277.

WEE CHING POK280

In addition to above, a VAR Pairwise Granger causality test (a strictercausality sense) is applied to determine whether the returns of the futuresmarket cause the spot market in the Granger sense. The results (see Table 9)confirm that futures market returns do indeed Granger-cause the spotmarket returns for all the periods examined.

Table 8. Causality Testing Results for the After CapitalControls Perioda.

DlogSt DlogFt


Error


Error

t-Statistics

Constant 0.0603 0.0578 1.0441 0.0661 0.0722 0.9153

DlogSt�1 �0.0888 0.0804 �1.1048 0.0262 0.1004 0.2613

DlogSt�2 �0.1391 0.0807 �1.7235 �0.0023 0.1008 �0.0226

DlogSt�3 �0.1145 0.0810 �1.4124 0.0622 0.1012 0.6147

DlogSt�4 �0.1030 0.0810 �1.2713 0.0822 0.1012 0.8119

DlogSt�5 �0.1946 0.0813 �2.3935� �0.0547 0.1016 �0.5383

DlogSt�6 �0.0752 0.0818 �0.9192 0.1040 0.1021 1.0180

DlogSt�7 �0.1254 0.0808 �1.5526 �0.0241 0.1009 �0.2393

DlogSt�8 �0.1132 0.0807 �1.4022 �0.0058 0.1008 �0.0573

DlogSt�9 �0.0019 0.0800 �0.0240 0.1210 0.0999 1.2111

DlogSt�10 �0.0422 0.0795 �0.5313 �0.0028 0.0992 �0.0282

DlogSt�11 �0.1244 0.0718 �1.7324 �0.1155 0.0897 �1.2880

DlogFt�1 0.2045 0.0645 3.1735� �0.0702 0.0805 �0.8722

DlogFt�2 0.1932 0.0675 2.8610� 0.0811 0.0843 0.9622

DlogFt�3 0.1589 0.0679 2.3412� 0.0097 0.0847 0.1149

DlogFt�4 0.0580 0.0682 0.8494 �0.1171 0.0852 �1.3741

DlogFt�5 0.2448 0.0686 3.5671� 0.0951 0.0857 1.1091

DlogFt�6 0.0710 0.0695 1.0222 �0.0588 0.0868 �0.6774

DlogFt�7 0.1471 0.0688 2.1382� 0.0302 0.0859 0.3519

DlogFt�8 0.1235 0.0685 1.8032 �0.0024 0.0855 �0.0279

DlogFt�9 0.0200 0.0682 0.2937 �0.1018 0.0851 �1.1959

DlogFt�10 0.0615 0.0678 0.9064 0.0133 0.0847 0.1566

DlogFt�11 0.1533 0.0614 2.4971� 0.1452 0.0767 1.8941

Q(5) 0.0386 (1.0000) 0.0882 (1.0000)

Jacque–Bera 116.6356 (0.0000)�� 129.8553 (0.0000)��

B-G Serial Corr. LM

Test-F-Statistic

1.1785 (0.3084) 0.0281 (0.9722)

ARCH Test-F-Statistic 8.2154 (0.0043)�� 12.7525 (0.0004)��

White Test-F-Statistic 1.3505 (0.0685) 1.5556 (0.0138)�

�Rejects null hypothesis at 5% significant level.��Rejects null hypothesis at 1% significant level.aFrom 15 September 1998 to 31 July 2001 and n=709.


5. SUMMARY AND CONCLUSIONS

The purpose of this analysis is to examine whether the spot and the futuresmarkets influence each other over the short-term and the long-term.Establishing a long-run equilibrium with the spot market is important ifthe futures market is to be of service in providing risk managementopportunities to industry and price discovery information to spot markettraders.

No evidence of a long-term relationship between the stock and futuresmarkets is found. This is in contrast to the evidence provided by Tan (2002),who found a long-run relationship before the imposition of selectivecapital controls. Before the cointegration test, there is a prerequisite thatthe VAR should be statistically and mathematically stable. No evidence

Table 9. VAR Pairwise Granger Causality Results.

Exclude Chi-Sq df Prob.


Dependent variable: DlogSt

DlogFt 11.5476 5 0.0415

All 11.5476 5 0.0415

Dependent variable: DlogFt

DlogSt 5.8686 5 0.3192

All 5.8686 5 0.3192

Crisis-adjusted (From 401 to 677b)


DlogFt 20.4157 2 0.0000

All 20.4157 2 0.0000


DlogSt 0.8300 2 0.6603

All 0.8300 2 0.6603



DlogFt 31.9350 11 0.0008

All 31.9350 11 0.0008


DlogSt 5.9154 11 0.8789

All 5.9154 11 0.8789

aFrom 15 December 1995 to 31 July 1997 and n=399.bFrom 1 August 1997 to 14 September 1998 and n=277.cFrom 15 September 1998 to 31 July 2001 and n=709.

WEE CHING POK282

of such is found in any of the sub-periods examined. Because thesecriteria have not been met, the trace statistic cannot be accuratelydefined; hence the cointegration results are dubious. This suggests that thechange in spot prices is not responding to the previous equilibrium error.Whether this is because of pricing inefficiency in the spot and/or futuresmarkets or due to the different institutional design of the two markets isless clear.

The causality of both spot and futures returns through VAR with firstdifference is then examined. The results show uni-directional causality fromfutures to the spot market in all the sub-periods examined. In the pre-crisisperiod, the returns of futures up to t=�2 is significant in explaining thecurrent returns of spot. During the crisis period, the returns of futuresat t=�1 is significant in explaining the current returns of spot. However,after the introduction of capital controls, the returns of futuresup to t=�11 is significant in the current returns of spot. When thecausality is tested in the Granger sense, it is found that the futures marketsdo indeed Granger-cause the spot market. The significant change in thecomposition of foreign institutional investors before and after theimplementation of selective capital controls lead to examination ofthe influence of foreign institutional investors on the speed of transmittinginformation or the price discovery role of futures trading. From theresults, it is found that the active participation of foreign institutionalplayers improved or shortened the speed of information transmission;hence the price discovery role of futures trading is seen to be influencedby the change in the composition of market agents as seen in theMalaysia case.

Overall, the short-run causality from the futures to the spot market isfound. It is also found that the lead is shorter in the pre-crisis and crisisperiods when foreign institutional investors were participating actively in theMalaysian market. Whether this phenomenon could be a causal factorrequires further research on the role and influence of foreign institutionalinvestors in emerging capital markets. If a price relationship between thespot and futures markets in the long run does not emerge, it would beappropriate to question whether the institutional structure of the spot andfutures markets restricts efficient pricing behaviour. It may be that themarkets have not gained sufficient trading experience to establish a long-runequilibrium relationship. Another possible explanation could be the result ofthe profile of market traders which has changed drastically over the periodof study. Two possible future directions have emerged from this study. Thecomposition of market traders, in particular, the role of foreign participants


(who are more experienced traders) in the emerging markets may provide auseful insight into the causal relationship between the stock index futuresand the stock market in the long run and the short-run. The other would beto examine the test for cointegration using intra-day data such as 5-minuteinterval data, which may provide further insights into the interaction of thespot and futures markets in Malaysia.

NOTES

1. Stoll and Whaley (1990) find evidence that S&P500 and MMI futures leads thespot by about 5min and occasionally as much as 10min or more, and weak evidenceof spot leads futures; Chan (1992) reports MMI futures leads spot up to 15min;Grunbichler, Longstaff, and Schwartz (1994) find DAX futures leads spot by about15–20min and spot leads futures by 5min; Kawaller, Koch, and Koch (1987) findS&P500 futures prices lead spot prices by 20–45min while stock only affects futuresprices rarely beyond 1min; Pizzi, Economopoulos, and O’Neill (1998) find S&P500futures market leads the spot market by at least 20min and spot lead futures by 3 to4min; and, Chiang and Fong (2001) find HSI futures lead the spot by 10min andspot leads the futures by 5min.2. Frino and West (1999) find ASX futures lead index by 20–25min; Abhyankar

(1995) and Brooks et al. (2001) also find evidence that FTSE 100 futures leads spotreturns; Tse (1995) finds NSA futures lead the spot, Ghosh (1993) finds evidence ofinformation flows from futures to spot for S&P500; and, Tan (2002) finds the KLSEfutures leads spot only in the period before the implementation of capital controls.3. He attributes these findings to several factors: the diversion during the financial

crisis of market-wide information to company-specific information; the tradingrestrictions imposed by the authorities to halt plummeting prices and spillover fromthe regional contagion effect of the financial crisis at its height in October 1997 andthe fall of the Dow in the same period.4. This is attributed to the fact that they were not allowed to trade stock index

futures at the time of the launch. In addition, local unit-trust funds were restricted byboth the conditions of the trust deeds and the guidelines imposed by the industrywatchdog, Securities Commission. Other local institutions were faced withrestrictions such as statutory provisions and unfamiliarity. Insurance companieswere also not allowed to use futures.5. In a survey carried out by KLOFFE, it was found that the local institutional

investors were yet to participate actively in the futures market. Although more localfund managers and insurance companies had been given the green light to tradefutures, the expected increased in participation did not occur because of the Baringsdebacle (resulting in collapse of Britain’s Barings bank in February 1995) was stillfresh in their minds and many were still uneasy about derivatives. Lack ofunderstanding about futures trading among local retail and institutional investorsfurther contributed to lower participation of the local and institutional investors.Additionally, the brokers did not have much incentive to push for futures trade

WEE CHING POK284

because the commissions were low compared to those available in spot market. Thesurvey also found that the dissemination of KLOFFE market data to the investingpublic was rather restricted.6. This change was due to a host of significant events that occurred in the regional

markets. The Thai Baht, which came under the attack from currency speculators,had affected other Asian stock markets. While the KLSE declined under aweakening Ringgit and rising interest rates, the Securities Commission allowed thelocal unit-trust companies to trade in futures market.7. It was also discovered that many company charters still do not permit their

managers to invest in futures markets. As for state-owned unit-trust companies,many did not participate because they felt that they were sufficiently hedged – theymanaged to buy new shares in the market at low issuance prices. Nevertheless, theimplementation of a dual-licensing scheme and the wider dissemination of KLOFFEprices to the investing public had increased the participation of local retail investorsin the futures market.8. The high open interests showed that more and more market participants were

holding their positions for longer-term gains instead of indulging in intra-daytrading. Many began to look for signals from the futures market to help them todecide whether to buy, sell or hold their investments in the spot market.9. To make up for the loss of trade in the futures markets, in November 1998, the

Securities Commission allowed local fund-managers to apply for futures fund-manager licences. However, this move failed to restore the previous high tradingvolume. The Exchange continued to woo foreign fund managers to return to themarket and some did in early 1999. By June 1999, foreigners’ share of the tradingvolume on KLOFFE rose to only 14% (see Fig. 2). During the period covering thesecond half of year 1999, the whole of year 2000 and the first half of year 2001,the market did not regain the trading volumes observed in KLOFFE before theimplementation of selective capital controls.10. A significant number of studies have used 5-minute interval data (see for

example, Chiang and Fong (2001), Frino and West (1999), Chan (1992), Kim et al.(1999), Grunbichler et al. (1994), Stoll and Whaley (1990) and Abhyankar (1995,1998)). There are also studies that use high frequency minute-to-minute data (see forexample, Shyy, Vijayraghavan, & Scott-Quinn (1996), Frino et al. (2000) and Pizziet al. (1998)). It is also noted that the span of data ranges from as short as 17 days(Shyy et al., 1996) to as long as 6 years (Kim et al., 1999), and, for the majority of thecases the span of data less than a year. The short span of data especially those whichare less than a year is compensated with high frequency minute-to-minute or 5-minuteinterval data. Short span cum high-frequency data are currently being employed bymost researches for the study of the lead–lag relationship. The question of whethersuch data are valid for investigation of long-run equilibrium relationship is contro-versial. Moreover, such data are only available for well-developed exchanges.11. The move had widespread implications for the financial industry in Malaysia.12. The imposition of capital controls takes place on 1 September 1998, but the

effect was only observed after 15 September 1998. Hence, the after capital controlsperiod is analysed from this day onwards.13. This information can be provided upon request.14. This information can be provided upon request.


15. This information can be provided upon request.16. This information can be provided upon request.17. Method UR chooses the break date, which minimises the t-statistic for testing.

Method Studabs chooses the break date without any priori assumption on the sign ofchange. Method Stud allows the date of change to be unknown but restricts theanalysis to the cases of a ‘‘crash’’ or a slow down in growth.18. For the entire sample period (observations 1 to 1,386 or 15 December 1995 to

31 July 2001), the roots are: 0.9969, 0.9407, 0.8136, 0.8136, 0.7890, 0.7890, 0.7542,0.7542, 0.7289, 0.7289, 0.7120, 0.7120, 0.6830, 0.6830, 0.6678, 0.6678, 0.6379, 0.6379,0.6191 and 0.1775 for lag length of 10.19. This information can be provided upon request.20. This information can be provided upon request.21. This information can be provided upon request.22. This information can be provided upon request.23. For pre-crisis period (observations 1 to 400 or 15 December 1995 to 31 July

1997), the roots are: 0.9855, 0.9145, 0.8780, 0.8780, 0.8629, 0.8629, 0.8462, 0.8462,0.8458, 0.8458, 0.8391, 0.8391, 0.8313, 0.8313, 0.8304, 0.8304, 0.8134, 0.8134, 0.7847,0.7847, 0.7713, 0.7713, 0.7606, 0.7606, 0.7003 and 0.3101 for lag length of 13. For thecrisis period (observations 401 to 677 or 1 August 1997 to 14 September 1998), theroots are: 0.9926, 0.8367, 0.8367, 0.8162, 0.6877, 0.6877, 0.6515, 0.6515, 0.6294,0.6294, 0.6263, 0.6263, 0.2275 and 0.2275 for lag length of 7. For the crisis-adjustedperiod (observations 401 to 667 or 1 August to 1997 to 31 August 1998), the rootsare: 0.9982, 0.8369, 0.4779, 0.4423, 0.4423, 0.2912, 0.2912 and 0.2718 for lag lengthof 4. For the after capital controls period (observations 678 to 1,386 or 15 September1998 to 31 July 2001), the roots are: 0.9892, 0.9102, 0.7916, 0.7916, 0.7450, 0.7450,0.7055, 0.7043, 0.7043, 0.6927, 0.6927, 0.6763, 0.6763, 0.6391, 0.6391, 0.4304, 0.1569and 0.1569 for lag length of 9.24. This information can be provided upon request.25. Furthermore, when the VECM is examined, it is found that the ECM term or the

adjustment coefficients (the alphas) were both statistically insignificant different fromzero. The VECM results are intentionally omitted because the cointegration test resultsare not valid. For a valid cointegration test, it has to satisfy the criteria where the VARhas to statistically and mathematically stable. These criteria have not been met.26. Further, when the VECM is examined, it is found that the ECM term or

the adjustment coefficients (the alphas) were both statistically insignificant differentfrom zero.27. This information can be provided upon request.28. This information can be provided upon request.

ACKNOWLEDGMENT

The author gratefully acknowledges the comments of Prof. J. L. Ford ofUniversity of Birmingham, UK.

WEE CHING POK286

REFERENCES

Abhyankar, A. H. (1995). Return and volatility dynamics in the FTSE 100 stock index and

stock index futures markets. Journal of Futures Markets, 15(4), 457–488.

Abhyankar, A. H. (1998). Linear and non-linear Granger causality: Evidence from the UK

stock index futures market. Journal of Futures Market, 18, 519–540.

Brooks, C., Rew, A. G., & Ritson, S. (2001). A trading strategy based on the lead-lag

relationship between the spot index and futures contract for the FTSE 100. International

Journal of Forecasting, 17(January–March), 31–44.

Chan, K. (1992). A further analysis of the lead-lag relationship between the cash market and

stock index futures market. Review of Financial Studies, 5, 123–152.

Chiang, R., & Fong, W. M. (2001). Relative informational efficiency of cash, futures and

options markets: The case of an emerging market. Journal of Banking and Finance, 25,

355–375.

Dickey, D. A., & Fuller, W. A. (1981). Likelihood ratio statistics for autoregressive time series

with a unit root. Econometrica, 49, 1057–1072.

Fleming, J., Ostdiek, B., & Whaley, R. (1996). Trading costs and the relative rates of price

discovery in stock, futures and options markets. Journal of Futures Markets, 16, 353–387.

Frino, A., & West, A. (1999). The lead-lag relationship between stock indices and stock index

futures contracts: Further Australian evidence. ABACUS, 35, 333–341.

Frino, A., Walter, T., & West, A. (2000). The lead-lag relationship between equities and stock

index futures markets around information releases. Journal of Futures Markets, 20(5),

467–487.

Ghosh, A. (1993). Cointegration and error correction models: Intertemporal causality between

index and futures prices. Journal of Futures Markets, 13, 193–198.

Grunbichler, A., Longstaff, F. A., & Schwartz, E. S. (1994). Electronic screen trading and the

transmission of information: An empirical examination. Journal of Financial Inter-

mediation, 3, 167–187.

Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics

and Control, 12, 231–254.



Johansen, S. (1995). Likelihood-based inference in cointegrated vector autoregressive models.

Oxford: Oxford University Press.

Kawaller, I. G., Koch, P. D., & Koch, T. W. (1987). The temporal price relationship between

S&P 500 futures and the S&P 500 index. Journal of Finance, 42, 1309–1329.

Kim, M., Szakmary, A. C., & Schwarz, T. V. (1999). Trading costs and price discovery across

stock index futures and cash markets. Journal of Futures Markets, 19, 475–498.

Perron, P. (1997). Further evidence on breaking trend functions in macroeconomic variables.

Journal of Econometrics, 80, 355–385.

Phillips, P., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75,

335–346.

Pizzi, M. A., Economopoulos, A. J., & O’Neill, H. M. (1998). An examination of the

relationship between stock index cash and futures markets: A cointegration approach.

Journal of Futures Markets, 18(3), 297–305.


Shyy, G., Vijayraghavan, V., & Scott-Quinn, B. (1996). A further investigation of the lead-lag

relationship between the cash market and stock index futures market with the use of bid/

ask quotes: The case of France. Journal of Futures Markets, 16, 405–420.

Stoll, H. R., & Whaley, R. E. (1990). The dynamics of stock index and stock index futures

returns. Journal of Financial and Quantitative Analysis, 25, 441–468.

Tan, J. H. (2002). Temporal causality between the Malaysian stock price and stock-indexed

futures market amid the selective capital controls regime. ASEAN Economic Bulletin, 19,

191–203.

Tse, Y. K. (1995). Lead-lag relationship between spot index and futures price of the Nikkei

stock average. Journal of Forecasting, 14, 553–563.

Wahab, M., & Lashgari, M. (1993). Price dynamics and error correction in stock index and

stock index futures markets: A cointegration approach. Journal of Futures Markets,

13(7), 711–742.

WEE CHING POK288

CHAPTER 14

PRICE BEHAVIOUR

SURROUNDING BLOCK

TRANSACTIONS IN STOCK INDEX

FUTURES MARKETS:

INTERNATIONAL EVIDENCE

Alex Frino, Jennifer Kruk and Andrew Lepone

ABSTRACT

This chapter examines the price impact of large trades in futures markets

across 14 stock index futures contracts in 11 different international

markets. On the balance, we find that part of the initial price effect of

futures trades is temporary. These initial price effects are partially

reversed, implying that they incur a liquidity premium; though there is

some variation in this finding across markets. We also find strong

evidence that large buyer- and seller-initiated trades have positive and

negative permanent effects on prices, implying they convey information.

We conclude, similar to research based on equities markets, that traders

in futures markets are informed.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00014-3

289

dx.doi.org/10.1016/S1569-3767(07)00014-3.3d

1. INTRODUCTION

Numerous equity market studies analyse the impact of institutional tradeson stock prices.1 These studies examine two important issues. First, theyexamine whether temporary and/or permanent price effects are associatedwith the execution of large trades. A temporary price effect (price reversal)occurs when the stock price moves momentarily due to short-run liquiditycosts. A permanent price effect (price continuation) occurs when there is achange in the fundamental value of a stock and its price permanently movesto a new level.2 Equity market studies provide evidence that large tradeshave statistically significant permanent price effects, suggesting they areexecuted by informed traders and contain information. Second, equitymarket studies examine asymmetries in price behaviour between buys andsells, finding evidence of price continuations following buys and partial pricereversals following sells. This suggests that in equities markets, sellers pay aliquidity premium while buyers do not.

There is a dearth of empirical research examining the price behavioursurrounding institutional trades in futures markets. In contrast to thenumerous empirical studies in equities markets, only Berkman, Brailsford,and Frino (2005) explicitly examine the price impact of large trades infutures markets. The price behaviour surrounding trading in futuresmarkets is expected to differ from equity markets. Subrahmanyam (1991)proposes that index products reduce information asymmetries andencourage liquidity trading as they diversify away any stock-specificinformation. Compared with underlying equity products, the probabilitythat trades in stock index futures contain information is lower. Chan andLakonishok (1993) suggest restrictions on short-selling in equities marketsgenerate asymmetrical price behaviour in buys and sells. This implies buysand sells should behave symmetrically in futures markets, as there are noshort-selling restrictions. Berkman et al. (2005) test both of these issuesusing a sample FTSE 100 stock index futures traded on the LondonInternational Financial Futures and Options Exchange (LIFFE). Theyprovide evidence of significant permanent price effects and partial reversals(i.e., information and liquidity effects), but no evidence of asymmetricalprice effects between buys and sells.

The analysis in Berkman et al. (2005) is limited to a single stock indexfutures contract traded on LIFFE; an electronic order-driven market withan off-market facility to trade blocks greater than 750 contracts.3 Under-lying index stocks trade on the London Stock Exchange (LSE) in anelectronic order-driven market that interacts with a network of dealers.

ALEX FRINO ET AL.290

Further, the sample used in Berkman et al. (2005) covers two relatively shorttime periods and includes the beginning of the dot com crash, with averagedaily returns of �0.13% in the first data period and �0.26% in the seconddata period (Berkman et al., 2005, p. 567).

This study contributes to the literature by extending Berkman et al. (2005)in two important ways. First, this study examines 14 stock index futurescontracts traded on 11 markets with differing market structures.4 Second,the data period in this study incorporates five years, in contrast to thesample of three months examined by Berkman et al. (2005). Specifically, we(i) measure total, temporary and permanent price effects associated withblock trades, (ii) test for potential asymmetries in the permanent price effectand (iii) discuss price impact differences across markets. The permanentprice effect is the primary focus of this study.

The remainder of this chapter is structured as follows. Section 2 describesthe data and methodology. Sections 3 and 4 present results and severaladditional tests, respectively, while Section 5 provides a summary andseveral future research avenues.


The data used in this study are sourced from Reuters and describetransactions executed in 14 stock index futures from 1 January, 2001 to 31December, 2005. The sample includes trades from the DAX, FTSE100,CAC40, OMXS30, S&P500 GLOBEX, Hang Seng Index, KOSPI 200, MSCISingapore, MSCI Taiwan, SPI 200, TOPIX, Nikkei 225 (OSE), Nikkei 225(SGX) and TAIEX stock index futures contracts. Each trade record containsfields which document the date, time, price, volume, best bid and best askassociated with each trade. Bid and ask quotes are the prevailing best quotesimmediately prior to the trade.

Block trades are defined as the largest 2% of trades, by volume, for eachcontract. Trades are classified as buyer- or seller-initiated using theclassification algorithm from Ellis, Michaely, and O’Hara (2000).5 In thisalgorithm, trades are initially classified using a quote-based rule. Tradesexecuted at the best ask quote are classified as buyer-initiated and tradesoccurring at the best bid quote are classified as seller-initiated. Anytrades not captured by this classification rule are classified using a tick rule,where trades occurring on an up-tick are classified as buyer-initiated andtrades occurring on a down-tick are classified as seller-initiated. Anyremaining unclassified trades are excluded from the sample.6

International Evidence 291

The sample is restricted to electronic trading in the near contract duringdaytime trading hours.7 Trades occurring on the expiration day of the nearcontract are excluded.8 Some exchanges have a facility for trading largeblocks off-market. Transactions that meet off-market size requirements areexcluded from the sample as off-market trades arrive to the market crossedand in some instances reporting is delayed.9

The pre- and post-trade benchmarks employed in this study to measureprice impact are the transaction prices five trades before and five trades afterthe block trade, respectively; analogous to the benchmarks used in Berkmanet al. (2005). The calculation of total, temporary and permanent price effectsis consistent with Chan and Lakonishok (1993). Total measures the totalprice impact of a trade, and can be decomposed into Temporary (liquidity)and Permanent (information) effects, as follows:

Totali;t ¼Pricet � Pricet�5

Pricet�5

� 100 (1)

Temporaryi;t ¼Pricetþ5 � Pricet

Pricet

� 100 (2)

Permanenti;t ¼Pricetþ5 � Pricet�5

Pricet�5

� 100 (3)

For each trade, Pricet is the transaction price, Pricet�5 the price five tradespreceding the trade and Pricet+5 the price five trades after the trade.10

Table 1 presents descriptive statistics for block trades in the 14 stock indexfutures contracts examined in this study. Panel A reports statistics for buysand Panel B reports statistics for sells. There are significant differences insample sizes across contracts. The DAX, FTSE100, CAC40 and KOSPI 200have sample sizes greater than 100 thousand for block buys and sells; theHang Seng, SPI 200, TOPIX, Nikkei 225 (OSE) and TAIEX have samplesizes for buys and sells between 20 thousand and 100 thousand; and theOMXS30, S&P500 GLOBEX, MSCI Singapore, MSCI Taiwan and Nikkei225 (SGX) have sample sizes less than 20 thousand for both block buysand sells.

Panels A and B of Table 1 describe the size of transactions in terms ofcontract volume and notional trade value in US dollars.11 The sample has alarge range in mean trade volume. OMXS30 futures have the greatestaverage volume of 1,164.01 contracts and MSCI Singapore futures have the


Table 1. Descriptive Statistics for Block Trades: Volume Traded andNotional Trade Value.

Contract Volume Traded (Contracts) Notional Trade Value (USD ’000s) N

Mean Median Std Dev. Mean Median Std Dev.

Panel A: Buys

DAX 74.34 47.00 58.60 7,119.90 4,494.31 6,539.64 235,064

FTSE100 71.46 40.00 103.41 3,828.30 2,223.31 6,605.42 250,916

CAC40 64.65 49.00 46.55 1,685.65 1,387.08 1,734.09 204,489

OMXS30 1,085.08 387.00 2,688.06 8,760.04 2,737.53 20,541.07 2,411

S&P500 Globexs 30.41 19.00 197.75 6,445.94 4,076.06 46,062.40 17,540

Hang Seng 34.88 30.00 15.99 1,890.79 1,910.28 1,479.17 91,294

KOSPI 200 271.80 230.00 125.96 8,936.76 8,533.71 6,407.01 208,755

MSCI Singapore 9.67 8.00 6.01 194.74 174.91 174.15 15,304

MSCI Taiwan 26.39 21.00 12.87 467.09 486.18 365.82 8,283

SPI200 33.96 28.00 18.36 1,164.81 1,017.84 1,202.62 25,038

TOPIX 59.77 52.00 13.72 5,660.69 5,513.28 2,082.38 23,997

Nikkei 225 (OSE) 79.56 79.00 10.74 7,009.30 7,064.30 2,043.25 25,428

Nikkei 225 (SGX) 73.10 60.00 33.54 2,374.35 2,587.73 2,129.04 5,021

TAIEX 34.88 30.00 15.99 1,890.79 1,910.21 1,479.16 91,294

Panel B: Sells

DAX 69.34 40.00 59.63 7,237.10 4,530.18 6,725.30 142,070

FTSE100 69.04 40.00 100.49 4,088.29 2,478.74 6,739.70 156,892

CAC40 62.29 48.00 43.63 1,851.93 1,581.27 1,661.57 141,443

OMXS30 1,164.01 400.00 2,698.81 9,256.18 2,966.26 21,567.18 2,506

S&P500 GLOBEX 40.07 19.00 704.21 8,929.75 4,045.65 198,098.59 18,101

Hang Seng 34.78 30.00 16.07 1,863.99 1,889.14 1,473.52 92,634

KOSPI 200 271.19 230.00 156.85 8,880.50 8,484.03 7,044.86 216,453

MSCI Singapore 9.68 8.00 6.59 46.51 168.34 179.18 14,969

MSCI Taiwan 26.63 21.00 13.61 466.02 479.38 383.21 7,976

SPI200 34.06 28.00 18.89 1,179.46 1,017.35 1,266.28 24,548

TOPIX 59.75 51.00 13.84 5,659.67 5,511.35 2,110.19 23,024

Nikkei 225 (OSE) 79.46 79.00 10.69 6,952.28 7,004.48 2,025.27 25,765

Nikkei 225 (SGX) 72.28 60.00 32.96 2,325.92 2,584.08 2,038.73 4,865

TAIEX 34.78 30.00 16.07 1,863.99 1,889.14 1,473.52 92,634

This table reports descriptive statistics for block trades in the 14 stock index futures contracts

examined in this study. Block trades represent the largest 2% of trades in each contract after

removing trades that meet the minimum volume threshold for off-market block transactions.

Panel A reports statistics for buys and Panel B reports statistics for sells. The mean, median and

standard deviation are reported for volume traded and notional trade value. Volume traded is

the number of contracts per trade. Notional trade value is calculated as [price� volume� index

multiplier� fx rate] where price is the trade price, volume is the number of contracts, index

multiplier is the dollar value per index point as reported in Table A1 and fx rate is the daily

exchange rate to USD as provided by the US Federal Reserve.


lowest average volume of 9.67 contracts. The mean notional trade value alsohas a large range, with values ranging from USD 0.46 million for MSCISingapore futures up to USD 9.256 million for OMXS30 futures.12 Overall,buyer- and seller-initiated transactions across all markets are relativelysimilar.

3. RESULTS

Table 2 reports estimates of total, temporary and permanent price effects forblock buys and block sells. The total price effect for all contracts is positiveand statistically significant for block buys and negative and statisticallysignificant for block sells. The direction and statistical significance of thetotal price effect is consistent with prior equities and futures marketsresearch; however, the magnitude is smaller than previously reported, withall estimates less than (or equal to) 2.5 basis points. The total price effectreported for the largest trade size category in Berkman et al. (2005) is lessthan six basis points, compared with the maximum 2.5 basis points reportedfor S&P500 GLOBEX futures in this study.13

Results for the temporary effect are reported in Table 2. For block buys,the temporary price effect is negative in 11 of the 14 contracts examined, andstatistically significant in nine. This implies buyers incur a liquidity premiumto transact large blocks in the majority of the contracts examined in thisstudy. Results for block sells are analogous to block buys, with the majorityof contracts incurring a statistically significant positive temporary priceeffect. On the balance, this study provides strong evidence that traders pay aliquidity premium to transact large blocks in futures markets, consistentwith Berkman et al. (2005).

Results for the permanent effect are reported in Table 2. All contractshave a positive permanent price effect for block buys, and the permanentprice effect for buys is statistically significant in 10 of the 14 contractsexamined. This suggests that block buys in these 10 contracts are executedby informed traders. For block sells, the permanent price effect is negative in13 of the 14 contracts examined, and statistically significant in 12 of thecontracts. This study provides overwhelming evidence that large trades infutures markets contain information. Berkman et al. (2005) report acomplete price reversal for both buys and sells in their largest trade sizecategory, finding no evidence of a significant permanent price effect for largetrades. In this study, only FTSE 100 and Nikkei 225 (OSE) futures havea complete price reversal for buys and sells, suggesting the findings in


Table 2. Total, Temporary and Permanent Price Effects for Block Trades.

Contract Total Temporary Permanent

Buys Sells Buys Sells Buys Sells Abs(buys) �Abs(sells)

DAX 0.0045�� 0.0063�� 0.0043�� 0.0047�� 0.0002 �0.0016�� 0.0014��

FTSE100 0.0084�� 0.0028�� 0.0067�� 0.0040�� 0.0017 0.0012 0.0005

CAC40 0.0075�� 0.0068�� 0.0037�� 0.0024�� 0.0038�� 0.0044�� 0.0006

OMXS30 0.0177�� 0.0176�� 0.0073�� 0.0087�� 0.0104�� 0.0089�� 0.0015

S&P500 GLOBEX 0.0227�� 0.0250�� 0.0008 0.0007 0.0219�� 0.0245�� 0.0026

Hang Seng 0.0117�� 0.0123�� 0.0060�� 0.0058�� 0.0057�� 0.0065�� 0.0008

KOSPI 200 0.0107�� 0.0112�� 0.0106�� 0.0101�� 0.0001 �0.0011�� 0.0010��

MSCI Singapore 0.0206�� 0.0197�� 0.0045�� 0.0030�� 0.0251�� 0.0227�� 0.0024

MSCI Taiwan 0.0155�� 0.0153�� 0.0007 �0.0007 0.0162�� 0.0160�� 0.0002

SPI200 0.0158�� 0.0155�� 0.0032�� 0.0033�� 0.0190�� 0.0188�� 0.0002

TOPIX 0.0103�� 0.0091�� 0.0060�� 0.0056�� 0.0043�� 0.0035�� 0.0008

Nikkei 225 (OSE) 0.0209�� 0.0200�� 0.0186�� 0.0177�� 0.0023 �0.0023 0.0000

Nikkei 225 (SGX) 0.0122�� 0.0111�� 0.0005 0.0009 0.0117�� 0.0102�� 0.0015

TAIEX 0.0117�� 0.0123�� 0.0060�� 0.0058�� 0.0057�� 0.0065�� 0.0008

�Significantly different from zero at the 5% level.��Significantly different from zero at the 1% level.

This table reports returns surrounding block trades for each of the 14 contracts examined in this study. Block trades represent the largest 2%

of trades in each contract after removing trades that meet the minimum threshold for off-market block transactions. Total is the percentage

return from the price five trades prior to the trade to the trade price. Temporary is the percentage return from the trade price to price five

trades after the trade. Permanent is the percentage return from the price five trades prior to the trade to the price five trades after the trade.

Abs(buys)�Abs(sells) is the mean difference in the permanent price effect for buys and sells. A t-test is used to test the deviation of mean

values from zero and critical t-values are adjusted for sample size.

Intern

atio

na

lE

viden

ce295

Berkman et al. (2005) cannot be generalised to incorporate all futurescontracts.

Berkman et al. (2005) also hypothesise that buy and sell trades in futuresmarkets will behave symmetrically post-execution. Their analysis providesevidence to support this hypothesis, as they find no significant differencebetween permanent price effects for buy and sell trades. For each contractexamined in this study, Table 2 reports the mean difference between buysand sells for the permanent price effect. Consistent with Berkman et al.(2005), the majority of contracts in Table 2 have symmetrical permanentprice effects for buys and sells. There is no significant difference in the meanpermanent price effect for 12 of the 14 futures contracts examined.14

The results presented in Table 2 of this study provide evidence thatprice behaviour surrounding block trades differs across contracts andmarkets. There are numerous market and contract design issues potentiallycontributing to this inconsistency. It is beyond the scope of this chapterto discuss them all; this is an issue for future research. Some examples ofpotential differences include contract size (Karagozoglu & Martell, 1999),the availability of off-market trading facilities (Madhavan & Cheng, 1997)and the transfer of information between futures and cash markets (Fleming,Ostdiek, & Whaley, 1996). The appendix contains Table A1 describingcontract specifications, contract size and some aspects of market design foreach contract examined in this study.15

4. ADDITIONAL TESTS

This section discusses various robustness tests employed to confirm resultspresented in Section 3. We repeat the study using midpoint quotes, analternative definition of execution costs and alternative pre- and post-tradebenchmarks.16

Koski and Michaely (2000) recognise a potential bid-ask bias whenmeasuring price impact using transaction prices.17 To overcome thispotential problem, they calculate price impact using quoted returns. Totest whether price movements reported in this study capture bid-ask bounce,we recalculate Eqs. (1)–(3) using contemporaneous quote midpoints insteadof transaction prices. The results reported in this chapter are consistent withresults based on quote midpoints. This suggests price effects reported inSection 3 are not driven by bid-ask bounce.

The calculations of total, temporary and permanent price effects inBerkman et al. (2005) are different to this study. The second test ensures


results are consistent with Berkman et al. (2005) and provides an additionaltest of the technique employed to measure total, temporary and permanentprice effects in block trades. In this analysis, effective half spreads, realisedspreads and permanent price effects are synonymous with total, temporaryand permanent price effects, respectively. These alternative measures ofprice impact and its components are taken directly from Berkman et al.(2005) and are calculated as

Effective half spread ¼ 100Di lnPricei

MQBeforei

� �(4)

Realised half spread ¼ 100Di lnPricei

MQAfteri

� �(5)

Permanent price impact ¼ 100Di lnMQAfteri

MQBeforei

� �(6)

where Di is a binary variable that equals 1 for buys and �1 for sells, Pricei

the value-weighted average price of the trade, MQBeforei the mid-quotefive trades before the block trade and MQAfteri the mid-quote five tradesafter the block trade. Analysis based on these alternative measures producesresults consistent with those reported in this chapter. The majority ofcontracts incur price impact which is permanent, indicating significantinformation content.

Chan and Lakonishok (1993, 1995) recognise the importance of bench-mark selection in price impact studies. The final test examines the choice ofbenchmark by replacing the five-trade benchmarks used in Eqs. (1)–(6) with10-trade benchmarks. Changing the pre- and post-trade benchmark doesnot significantly affect results. Results are thus robust to the choice of pre-and post-trade benchmarks.

5. CONCLUSIONS

This chapter extends Berkman et al. (2005) and produces broad interna-tional evidence of the price impact incurred by block trades in futuresmarkets. The chapter examines 14 stock index futures contracts from 11different exchanges, and provides evidence of statistically significant priceimpact associated with block trades in all contracts. Consistent with theanalysis of large trades in Berkman et al. (2005), block trades in the majority


of contracts examined in this study incur a statistically significant temporaryprice effect. This suggests traders executing large transactions in futurespay a liquidity premium. In contrast with Berkman et al. (2005), this studyprovides evidence that block trades in futures markets contain information.This suggests the findings in Berkman et al. (2005) are not applicable to allfutures contracts in all markets. Consistent with futures market literature,block buys and sells in the majority of contracts have symmetricalpermanent price effects.

This chapter provides a preliminary analysis of price impact and itscomponents in a selection of stock index futures. Many potential avenuesfor future research arise from results presented here. For example, analysisof the determinants of price impact could formally highlight why blocktrades in some contracts contain information and others do not. Thediffering information content of buys and sells for several futures contractsalso warrants future research, as thus far no explanation for this differenceexists.

NOTES

1. Equity market studies examining price impact include Kraus and Stoll (1972),Dann, Mayers, and Raab (1977), Holthausen, Leftwich, and Mayers (1987, 1990),Ball and Finn (1989), Reinganum (1990), Choe, McInish, and Wood (1991), Blumeand Goldstein (1992), Kumar, Sarin, and Shastri (1992), Chan and Lakonishok(1993, 1995, 1997), Keim and Madhavan (1995, 1996, 1997), Korthare and Laux(1995), Aitken and Frino (1996a, 1996b), Bessembinder and Kaufman (1996), Huangand Stoll (1996), Gemmill (1996), Madhavan and Cheng (1997), Bonser-Neal,Linnan, and Neal (1999), Domowitz, Glen, and Madhavan (2001), Jones and Lipson(2001), Saar (2001), Conrad, Johnson, and Wahal (2001), Nimalendran andPetrella (2003), Bortoli, Frino, and Jarnecic (2004), Chiyachantana, Jain, Jiang,and Wood (2004), Frino, Jarnecic, Johnstone, and Lepone (2005), and Frino,Jarnecic, and Lepone (2007).2. Kraus and Stoll (1972) and Scholes (1972) discuss temporary and permanent

price effects in this context.3. Madhavan and Cheng (1997) find that upstairs markets are primarily used by

traders able to credibly signal their trades are not information motivated. This couldresult in unique dynamics in the downstairs market, as large liquidity traders areattracted to the upstairs market where upon negotiation they can receive a betterprice for their block transaction.4. Details of the market structure are provided in the appendix.5. This classification algorithm is similar to the algorithm of Lee and Ready

(1991).6. Over 99% of trades in the sample are classified using this algorithm.


7. The exception for this is S&P500 stock index futures. This contract is traded onthe floor during daytime hours, and traded electronically through Globexs overnight.8. Frino and McKenzie (2002) report abnormal price behaviour in the period

prior to contract expiration. We remove trades on the contract expiration day toremove potential bias from the sample as traders roll their positions from the nearto deferred contract. Consistent with Frino and McKenzie (2002), we also excludetrades within 10 days of expiration of the near contract and results are consistent.These results are available on request.9. Berkman et al. (2005) also exclude off-market trades from their analysis.10. The large samples examined in this study necessitate adjustment of the t-value

critical level to alleviate Lindley’s paradox. The new critical value t� is calculatedusing the following formula:

t� ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi½c2=T T1=T � 1�ðT � kÞ

qwhere c is the ratio between the Bayesian probabilities of the null and alternativehypotheses; T the sample size and k the number of regressors in the model.Derivation and further explanation of this adjustment is found in Johnstone (2005).11. Notional trade value is calculated as [price� volume� index multiplier� fx

rate] where price is the trade price, volume the number of contracts, index multiplierthe dollar value per index point as reported in the appendix and fx rate is the dailyUSD exchange rate provided by the US Federal Reserve.12. OMXS30 futures do not have an off-market block trading facility. This

explains the large mean and variance for trades in this contract.13. Sample differences are one potential cause of this variation in the magnitude

of price impact. This study examines data covering five years from 2001 to 2005,while Berkman et al. (2005) examine three months in 2000. Average daily turnover infutures markets has increased dramatically since the Berkman et al. (2005) sampleperiod, and this enhanced liquidity could have contributed to a fall in the magnitudeof the total price effect over time.14. Block sells in DAX and KOSPI 200 futures have a permanent price effect

significantly larger in magnitude than block buys.15. This table is by no means exhaustive; there are many more market and

contract design characteristics relevant to these contracts.16. For space considerations, results from robustness tests are not reported but

are available on request.17. Numerous studies recognise a potential bid-ask bias when using returns

calculated with transaction prices, including Vijh (1988), Foerster, Keim, and Porter(1990), Lease, Masulis, and Page (1991), Bhardwaj and Brooks (1992), Gosnell,Keown, and Pinkerton (1996), Rhee and Wang (1997) and Frino et al. (2005).

ACKNOWLEDGMENTS

This research was funded by the Sydney Futures Exchange underCorporations Regulation 7.5.88(2). The authors wish to thank Angelo Aspris,


George Li, Reuben Segara and participants of the Sydney Futures ExchangeSeminar Series for useful comments, and the Securities Industry ResearchCentre of Asia-Pacific (SIRCA) for providing the Reuters data.

REFERENCES

Aitken, M., & Frino, A. (1996a). Asymmetry in stock returns following block trades on the

Australian Stock Exchange: A note. Abacus, 32, 54–61.

Aitken, M., & Frino, A. (1996b). Execution costs associated with institutional trades on the

Australian Stock Exchange. Pacific-Basin Finance Journal, 4, 45–58.

Ball, R., & Finn, F. (1989). The effect of block transactions on share prices: Australian

evidence. Journal of Banking and Finance, 13, 397–419.

Berkman, H., Brailsford, T., & Frino, A. (2005). A note on execution costs for stock

index futures: Information versus liquidity effects. Journal of Banking and Finance, 29,

565–577.

Bessembinder, H., & Kaufman, H. (1996). Quotations and trading costs on the domestic equity

exchanges. Unpublished working paper. Arizona State University.

Bhardwaj, R., & Brooks, L. (1992). The January anomaly: Effects of low share price,

transaction costs, and bid-ask bias. Journal of Finance, 47, 553–575.

Blume, M., & Goldstein, M. (1992). Displayed and effective spreads by market makers.

Unpublished working paper. University of Pennsylvania.

Bonser-Neal, C., Linnan, D., & Neal, R. (1999). Emerging market transaction costs: Evidence

from Indonesia. Pacific-Basin Finance Journal, 7, 103–127.

Bortoli, L., Frino, A., & Jarnecic, E. (2004). Differences in the cost of trade execution services

on floor-based and electronic futures markets. Journal of Financial Services Research, 26,

73–88.

Chan, L., & Lakonishok, J. (1993). Institutional trades and intraday stock price behaviour.


Chan, L., & Lakonishok, J. (1995). The behavior of stock prices around institutional trades.

Journal of Finance, 50, 1147–1174.

Chan, L., & Lakonishok, J. (1997). Institutional equity trading costs: NYSE versus NASDAQ.


Chiyachantana, C., Jain, P., Jiang, C., & Wood, R. (2004). International evidence on

institutional trading behavior and price impact. Journal of Finance, 54, 869–898.

Choe, H., McInish, T., & Wood, R. (1991). Market microstructure effects on the measurement of

the impact of block trades. Unpublished working paper. University of Memphis.

Conrad, J., Johnson, K., & Wahal, S. (2001). Institutional trading and soft dollars. Journal

of Finance, 56, 1–20.

Dann, L., Mayers, D., & Raab, R., Jr. (1977). Trading rules, large blocks and the speed of price

adjustment. Journal of Financial Economics, 4, 3–22.

Domowitz, I., Glen, J., & Madhavan, A. (2001). Liquidity, volatility and equity trading costs

across countries and over time. International Finance, 4, 221–255.

Ellis, K., Michaely, R., & O’Hara, M. (2000). The accuracy of trade classification: Evidence

from NASDAQ. Journal of Financial and Quantitative Analysis, 35, 529–551.


Fleming, J., Ostdiek, B., & Whaley, R. (1996). Trading costs and the relative rates of price

discovery in the stock, futures, and options markets. Journal of Futures Markets, 16,

353–387.

Foerster, S., Keim, D., & Porter, D. (1990). Intraday spreads, returns and variances: Tests of the

informed trader hypothesis. Unpublished working paper. University of Western Ontario.

Frino, A., Jarnecic, E., Johnstone, D., & Lepone, A. (2005). Bid-ask bounce and the

measurement of price behavior around block trades on the Australian Stock Exchange.


Frino, A., Jarnecic, E., & Lepone, A. (2007). The determinants of the price impact of block

trades: Further evidence. Abacus, 43, 94–106.

Frino, A., & McKenzie, M. (2002). The pricing of stock index futures spreads at contract

expiration. Journal of Futures Markets, 22, 451–469.

Gemmill, G. (1996). Transparency and liquidity: A study of block trades in the London Stock

Exchange under different publication rules. Journal of Finance, 51, 1765–1790.

Gosnell, T., Keown, A., & Pinkerton, J. (1996). The intraday speed of stock price adjustment to

major dividend changes: Bid-ask bounce and order flow imbalances. Journal of Banking

and Finance, 20, 247–266.

Holthausen, R., Leftwich, R., & Mayers, D. (1987). The effect of large block transactions on

security prices: A cross-sectional analysis. Journal of Financial Economics, 19, 237–267.

Holthausen, R., Leftwich, R., & Mayers, D. (1990). Large-block transactions, the speed of

adjustment, and temporary and permanent stock-price effects. Journal of Financial


Huang, R., & Stoll, H. (1996). Dealer versus auction markets: A paired comparison of

execution costs on NASDAQ and the NYSE. Journal of Financial Economics, 41,

313–357.

Johnstone, D. (2005). Significance levels and large samples in finance: Correction of Connolly

(1989). Unpublished working paper. University of Sydney.

Jones, C., & Lipson, M. (2001). Sixteenths: Direct evidence on institutional execution costs.


Karagozoglu, A., & Martell, T. (1999). Changing the size of a futures contract: Liquidity and

microstructure effects. The Financial Review, 34, 75–94.

Keim, D., & Madhavan, A. (1995). The anatomy of the trading process: Empirical evidence on

the behavior of institutional traders. Journal of Financial Economics, 25, 75–98.

Keim, D., & Madhavan, A. (1996). The upstairs market for large-block transactions: Analysis

and measurement of price effects. Review of Financial Studies, 9, 1–36.

Keim, D., & Madhavan, A. (1997). Transaction costs and investment style: An inter-exchange

analysis of institutional equity trades. Journal of Financial Economics, 46, 265–292.

Korthare, M., & Laux, P. (1995). Trading costs and the trading systems for NASDAQ stocks.

Financial Analysts Journal, 51, 42–53.

Koski, J., & Michaely, R. (2000). Prices, liquidity and the information content of trades. Review

of Financial Studies, 13, 659–696.

Kraus, A., & Stoll, H. (1972). Price impacts of block trading on the New York Stock Exchange.


Kumar, R., Sarin, A., & Shastri, K. (1992). The behavior of option prices around large block

transactions in the underlying security. Journal of Finance, 47, 879–889.

Lease, R., Masulis, R., & Page, J. (1991). An investigation of market microstructure impacts on

event study returns. Journal of Finance, 46, 1523–1536.


Lee, C., & Ready, M. (1991). Inferring trade direction from intraday data. Journal of Finance,

46, 733–746.

Madhavan, A., & Cheng, M. (1997). In search of liquidity: Block trades in the upstairs and

downstairs markets. Review of Financial Studies, 10, 175–204.

Nimalendran, M., & Petrella, G. (2003). Do ‘thinly-traded’ stocks benefit from specialist

intervention? Journal of Banking and Finance, 27, 1823–1854.

Reinganum, M. (1990). Market microstructure and asset pricing: An empirical investigation of

NYSE and NASDAQ securities. Journal of Financial Economics, 28, 127–147.

Rhee, S., & Wang, C. (1997). The bid-ask bounce effect and the spread size effect: Evidence

from the Taiwan Stock Market. Pacific-Basin Finance Journal, 5, 231–258.

Saar, G. (2001). Price impact asymmetry of block trades: An institutional trading explanation.

Review of Financial Studies, 14, 1153–1181.

Scholes, M. S. (1972). The market for securities: Substitution versus price pressure and the

effects of information on share prices. Journal of Business, 45, 179–211.

Subrahmanyam, A. (1991). A theory of trading in stock index futures. Review of Financial

Studies, 4, 17–51.

Vijh, A. (1988). Potential biases from using only trade prices of related securities on different

exchanges: A comment. Journal of Finance, 43, 1049–1055.


APPENDIX

Table A1. Contract Specifications, Contract Size and Market Design.

Contract Contract Specifications Contract Size Market Design

Exchange Minimum

Tick

Notional Value

Per Index Point

Relative

Minimum

Tick

Notional Value of

One Contract

(USD)

Off-Market

Threshold

Overnight

Trading

Cash/

Futures

Open First

Cash/

Futures

Close First

DAX Eurex 0.5 points EUR 25 0.000091 161,942.52 250 No Futures Cash

FTSE100 Euronext.liffe 0.5 points GBP 10 0.000089 96,580.42 750 No Same Cash

CAC40 Euronext.liffe 0.5 points EUR 10 0.000106 55,856.61 N/A No Futures Cash

Hang Seng HKE 1 point HKD 50 0.000067 95,936.12 100 No Futures Cash

KOSPI 200 KSE 0.05 points KRW 500,000 0.000036 683,533.20 N/A No Same Cash

MSCI Singapore SGX 0.1 points SGD 200 0.000359 33,498.92 200 No Futures Cash

MSCI Taiwan SGX 0.1 points USD 100 0.000362 27,581.00 200 Yes Futures Cash

Nikkei 225 OSE 10 points JPY 1,000 0.000621 136,676.53 100 No Same Cash

Nikkei 225 SGX 5 points JPY 500 0.000312 68,338.27 300 Yes Futures Cash

OMXS30 OMX 0.25 points SEK 100 0.000260 12,095.38 N/A No Same Same

S&P500 GLOBEX CME 0.1 points USD 250 0.000080 312,072.50 N/A Yes N/Aa N/Aa

SPI200 SFE 1 point AUD 25 0.000209 87, 432.20 300 Yes Futures Cash

TOPIX TSE 0.5 points JPY 10,000 0.000303 139,952.49 100 No Same Cash

TAIEX TFE 1 point TWD 200 0.000153 39,928.90 N/A No Futures Cash

This table reports contract specifications and market design details for each of the 14 stock index futures contracts examined in this study. Exchange is

the main exchange on which the contract is traded, minimum tick is the minimum price increment, notional value per index point is the dollar value (in

local denominations) of each index point, relative minimum tick is the minimum price increment divided by the average index level at 30/12/2005,

notional value of one contract is the dollar value of each index point multiplied by the index level at 30/12/2005 and converted to US dollars using the

exchange rate provided by the US federal reserve on that day, off-market threshold is the minimum number of contracts per trade required for off-

market trading, overnight trading indicates an overnight trading session, cash/futures open first indicates if the cash or futures market opens first (based

on regular trading hours), and cash/futures close first indicates if the cash or futures market closes first (based on regular trading hours).aThe data used in this study is for the overnight trading session. The cash market is not open during this session.

Intern

atio

na

lE

viden

ce303

PART V:

CORPORATE FINANCE

CHAPTER 15

THE DETERMINANTS OF CAPITAL

STRUCTURE: EVIDENCE

FROM VIETNAM

Nahum Biger, Nam V. Nguyen and Quyen X. Hoang

ABSTRACT

This study examines financing decisions by Vietnamese firms and

compares the results with the findings observed in economies character-

ized by market mechanisms and property rights. It uses data from

Vietnamese enterprises census 2002–2003. Similar to findings in other

countries, financial leverage of Vietnamese firms increases with firm size

and managerial ownership and decreases with profitability, and with non-

debt tax shield. It is also correlated with industry characteristics.

Financial leverage was negatively correlated with fixed assets and

positively correlated with growth opportunities, contrary to the findings

in other countries. Corporate income tax has a negative, albeit small

effect on financial leverage.

1. INTRODUCTION

The financing decision is a central issue in financial economics. The moderntheory of capital structure was developed by Modigliani and Miller (M&M)



r 2008 Published by Elsevier Ltd.

ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00015-5

307

dx.doi.org/10.1016/S1569-3767(07)00015-5.3d

(1958) and since the publication of their chapter many researchers havefollowed and extended the path they mapped. A vast literature continues todevelop theories that explain capital structure choice. Researchers studydepartures from the M&M assumptions and examine the implications. Theempirical evidence of alternative theories is still inconclusive (Rajan &Zingales, 1995). Much of the theoretical setting and the empirical evidencereported in the literature is based on firms in the United States where themarket mechanism and property rights have been functioning for manyyears. Are the tentative conclusions and empirical evidence based on highlydeveloped capital markets applicable in less developed economies? In thischapter, we examine the practice of capital structure decisions in theVietnamese context where the institutional environment for Vietnamesefirms is very different from that of highly developed economies. First,Vietnam is in a transitional period from a centrally controlled to a marketeconomy. Second, the state still is the controlling power in a large number ofmajor firms. Third, the Vietnamese stock market was established just sixyears ago with limited number of listed firms. Private firms (PF) and foreignowned firms are still relatively rare. With such salient features one wonderswhether the factors that affect corporate financing decisions in developedcapital markets have similar effects in the Vietnam corporate context.

This chapter examines the determinants of Vietnamese firms’ capitalstructure. We refer to factors that have been identified by both financialtheories and empirical studies. We believe that it is probably the first chapterin this arena of research regarding Vietnamese firms.

The M&M (1958) model in a perfect capital market setting implies thatcapital structure has no effect on a firm’s value. In less than perfect marketswhere firms operate in environments with taxes and differential tax rates;where there is an information asymmetry between insiders and outsiders,principals and agents this result does not hold. Furthermore firm’s attributesvary by industry and business environment (markets, products, businesslines, industry, etc.) and such attributes may imply that the classical M&Mirrelevance proposition does not hold. Following the seminal work of M&Ma number of theories have been developed (Jensen & Meckling, 1976;Fama & Friend, 2002; Myers & Majluf, 1984; Harris & Raviv, 1991; Chang,1992) to explain variation in debt ratios across firms. Two models seem tohave gained popularity, the static tradeoff model and the pecking theory.

The static tradeoff model suggests that there is an optimal capitalstructure for the firm. A firm’s specific optimal leverage is created as resultof a process that balances the effects of corporate and personal taxes (taxshield), agency costs (bondholders and equity holders and managers and

NAHUM BIGER ET AL.308

shareholder conflicts), bankruptcy costs, etc. (Jensen and Meckling (1976),Bradley, Jarrel, & Kim (1984), Jensen (1986), Stulz (1990), Harris and Raviv(1991) and Chang (1992)). On the other hand, the pecking order theory firstproposed by Myers and Majluf (1984) predicts that there is an order ofpreference over sources of financing. Firms prefer to finance investmentsusing internal equity (retained earning) then by less risky debts and onlyafter these two sources seem insufficient do firms approach external equityfinancing. The argument here is based on information asymmetry betweenthe outsiders (investors) and insiders (managers/shareholders). Faced with agood investment project, managers do not wish to communicate thatinformation to new shareholders. Conversely, if the firm faces a poorprospect project the owners-managers might wish to issue new shares as thiswould benefit existing shareholders. Outside investors view the issue of newshares as a signal of poor prospect. Thus, issuing new equity would bedifficult and costly. This description assumes that managers act on behalf ofexisting shareholders’ interests, an assumption that may be questionablewhen managers are not part of firm’s owners (Watson & Wilson, 2002).

Which hypothesis, tradeoff static or pecking order, is more relevant inexplaining a firm’s leverage behavior is still an empirical question. Thepecking order theory was supported in the study by Chaplinsky andNiehaus (1993) and Shyam-Sunder and Myers (1999) but rejected in thestudy by Korajczyk, Lucas, and McDonald (1990). Fama and French (2002)show that both models explain some company’s financing behavior andnone can be rejected. Titman and Wessels (1988) comment that empiricalfindings have lagged behind theoretical research because firms attributesinfluencing financing decisions are expressed in term of abstract conceptsand are not directly observable nor easy to measure.

The present study employs data from enterprises’ census (2002–2003) toexamine the determinants of capital structure of Vietnamese firms.Theoretical research suggests a number of factors that correlate with firm’sfinancial leverage.1 In this study, we examine a variety of firm’s attributessuch as collateralized assets (CA), profitability, tax rates, non-debt taxshield (NDTS), size, growth opportunities, industry classification and firminstitutional and managerial ownership that have been claimed to affectcapital structure decisions in western economies. Empirical evidencereported as Bradley et al. (1984), Titman and Wessels (1988), Rajan andZingales (1995), Wald (1999) supports these contentions. We also documenttypical features associated with Vietnamese firms such as current assets,taxes and ownership that affect firm financing and are not reported in otherstudies. The empirical findings lend support to the contention that similar to

The Determinants of Capital Structure: Evidence from Vietnam 309

other countries, Vietnamese firms financial leverage increases with firm sizeand with managerial ownership and decreases with profitability, NDTS andsome industry characteristics. Several salient features are found inVietnamese firms. First, in contrast to prior empirical studies where firm’sleverage increases with fixed asset and decreases with growth opportunities,the correlations have the opposite sign for Vietnamese firms. Second,corporate income tax considerations have negative, albeit very small effecton firm’s leverage. Growth opportunities induce firms to have higherfinancial leverage – in contrast to the theory and prior empirical evidence inother countries. We also find evidence that there is a significant institutionalimpact on capital structure for all three alternative measures of financialleverage as explained below.

Our study has several important implications. First, it suggests that thebasic factors that affect financing decisions are generally applicable inVietnamese firms. Exceptions are due to the fact that Vietnam is still intransition towards a market economy and several elements of a marketeconomy have not been reached. Second, the evidence indicates thatVietnamese firms find it difficult to obtain debt and especially externalequity financing. The stock market is still in a state of infancy and firms relyheavily on debt financing. Debt funding is based on the value of firm’scurrent assets and rarely on their fixed assets. This can be attributed to thefact that firms generally lack strong infrastructure that might ensure lowbankruptcy costs. Finally, this study is probably the first chapter to examinethe association of leverage and firm’s attributes in Vietnamese firms. AsVietnam joins the WTO, this study may serve as a basis for researchers andpractitioners to further investigate the firm’s characteristics and their effectson financing decisions in a market economy environment.

The chapter is organized as follows. Section 2 discusses the proxies fordeterminants of capital structure. Section 3 presents the descriptive statist-ics, empirical analysis and discusses some salient features of Vietnamesefirms. Section 4 presents our conclusions.

2. DETERMINANTS OF CAPITAL STRUCTURE

In this section, we review the factors that different theories of capitalstructure suggest to influence firm’s choice of financial leverage. Thesefactors include ownership of CA by firms, income tax, NDTS, profitability,size, growth opportunities, industry classification and firm ownership.


2.1. Collateralized Assets (CA)

Capital structure theories suggest that firm’s asset structure affects itsleverage in both positive and negative ways. Agency theory (Jensen &Meckling, 1976) and works by Galai and Masulis (1976), Myers (1977) pointout that the existence of agency costs of debt may cause firms to take onriskier investment after the issuance of debt to expropriate wealth from thefirm’s bondholders because the firm equity is effectively a stock option.Firms with a high level of assets that can be used as collateral tend to usemore debt rather than issue new equity because costs associated with issuingequity rise due to the asymmetry of information possessed by insiders andoutsiders. This line of argument suggests a positive relation between debtratios and the firm capacity of CA. In contrast, Grossman and Hart (1982)proposed that as managers have incentives to consume excessive perquisites,higher debts level would mitigate this tendency because of a higher threat ofbankruptcy. The costs associated with this agency problem would be higherfor firms with a lower level of CA. Firms that have less CA may choose ahigher level of debts to reduce a manager’s consumption of perquisites.

We measure CA as the ratio of fixed assets to total assets, denoted as CA.Different theories suggest that the presence CA on capital structure mighteither be positive or negative.

2.2. Profitability (ROA)

The predicted relationship between a firm’s profitability and its capitalstructures has been mixed. In a taxable environment many models concludethat the presence of taxes would induce profitable firms to use more debt inorder to take advantage of tax shield from corporate tax. In contrast, Myersand Majluf (1984) refer to the pecking order and conclude that profitablefirms tend to use less debt because then have internally generated funds(equity). In an agency setting, financial theories predict a mixed direction.Firms with a free cash flow or high profitability may tend to use earnings topay up debt in order to overcome possible restraints on managementdiscretion. In another approach, Chang (1992) considers a combination ofdebt and equity that can be interpreted as an optimal contract betweencorporate insiders and outsiders. It follows that profitable firms tend to useless debt.

Empirical studies provide mixed results. The pecking order theory wassupported in a study by Chaplinsky and Niehaus (1993) but rejected in the


study by Korajcryk et al. (1990). In general, most empirical studies report anegative correlation between firm profitability and leverage (Kester, 1986;Friend & Lang, 1988; Titman & Wessels, 1988; Rajan & Zingales, 1995;Wald, 1999). At variance with these studies Long and Maltyz (1985) reportleverage to be positively correlated with profitability but the relationshipis weak.

In the present study we use earnings before interests, tax and extra-ordinary income scaled by total assets, denoted as ROA, to be a proxy fora firm’s profitability.

2.3. Corporate Income Tax

Corporate income tax has important impact on debt–equity choices. TheModigliani–Miller proposition for a world with corporate taxes suggeststhat firms that face higher marginal tax rates would use more debts to takeadvantage of tax shields. Tax shields however do not apply if firms canobtain or issue interest-free liabilities. An empirical study by Mackie &Jeffrey (1990) found debt to be positively related to marginal effect tax rates.Huang and Song (2006) using average effective tax rate (ETR) to examinethe Chinese listed firms report similar results.

In this chapter, we use average effective income tax rate as a proxy for taxrates to examine the effect of tax on leverage.

2.4. Non-Debts Tax Shield (NDTS)

The tax deduction for depreciation and investment tax credits are NDTS.DeAngelo and Masulis (1990) suggest that tax deductions for depreciationand investment tax credits might substitute the tax deduction of debtfinancing. Empirical studies use different indicators as a proxy for NDTS,including annual depreciation expenses plus investment credit tax deflatedby earnings before interests, taxes and depreciation (EBIDA) (Bradley et al.,1984); ratio of depreciation to total assets (Wald, 1999); ratio of depreci-ation and amortization expenses scaled by total assets (Huang & Song,2006)). These studies find that leverage is negatively correlated withNDTS.

In the present study, we use the sum of depreciation and amortizationexpenses scaled by total assets as a proxy for NDTS.


2.5. Firm Size (LnS)

Many studies suggest that leverage ratio may be affected by firm size. Twocontradicting arguments are provided. One line of reasoning is based on afirm’s costs and earnings. Large firms enjoy economies of scale and credit-worthiness in issuing long-term debt (LTD) and have bargaining powerover creditors (Marsh, 1982); they are more diversified and less prone tobankruptcy (Warners, 1997; Ang, Chua, & McConnel, 1982), and theyenjoy a more stable cash flow. Hence, larger firms would have a tendency touse higher financial leverage.

Conversely, Smith (1977) suggests that small firms bear high costs ofissuing new equity and LTD securities and may therefore prefer to relyon short-term debt (STD) and be more highly leveraged compared to largersize firms. Furthermore, Rajan and Zingales (1995) claim that larger firmsare willing to disclose more information to outsiders, operate under lessasymmetric information and may therefore tend to use more equity thandebt. Overall, these arguments suggest a negative relationship betweenleverage and firm size.

Empirical studies generally support the positive relationship hypothesis(Marsh (1982); Rajan & Zingales (1995); Wald (1999); Booth et al. (2001),Huang & Song (2006)). In contrast, Kester (1986), Kim and Sorensen(1986), Titman and Wessels (1988) find a negative, albeit weak and nothighly significant relationship.

Many empirical studies use the natural logarithm of sales or total assets tobe proxy for firm size. We follow the same approach and use log-sale (LnS)to measure the firm size.

2.6. Growth (GTA)

Growth opportunities may be considered assets that add value to firms butthey cannot be used by firms as collaterals. They are not subject to incometax (Titman & Wessels, 1988). Furthermore, under agency theory, debtserves as a tool to mitigate managerial discretion when firms do not havemany investment opportunities (Jensen, 1986; Stulz, 1990). These argumentssuggest a negative relationship between growth opportunities and financialleverage.

Researchers used several indicators to measure a firm’s growthopportunities. These include Tobin’s Q-market-to-book ratio of total assets(Rajan & Zingales, 1995; Booth et al., 2001); capital investment scaled by


total assets (Titman & Wessels, 1988); five year average sale growth (Wald,1999); and percentage change in total assets (Titman & Wessels, 1988).

In the present study, we measure growth opportunities by the percentagechange in total assets (GTA) as suggested by Titman and Wessels study(1988). They found a significant negative correlation between leverage andgrowth.

2.7. Industry

Titman (1984) suggests that a leverage increases with the extent to which afirm’s product is not unique and does not requires specialized services. Firmsthat manufacture machines and equipment, for example, should be financedwith less debt.

We used a dummy variable to distinguish between manufacturing firmsand others. Firms with 4-digit code from 2911 to 4021 in the industry codingsystem are classified as manufacturing firms.

2.8. Ownership Structure

As stated above a significant part of the literature is devoted to models inwhich capital structure is determined by agency costs. Jensen and Meckling(1976) discuss two types of conflicts: conflicts between shareholders andmanagers and conflicts between debt holders and equity holders. Theysuggest an optimal structure of leverage and ownership that may be used inorder to minimize total agency costs. Leland and Pyle (1977) argue thatleverage is positively correlated with the extent of managerial equityownership. Empirical findings by Cornett and Travlos (1989) and Berger,Ofek, and Yermack (1977) support this prediction. However, Friend andLang (1988) found opposite results.

In this study, we divide the sample of firms into six groups according toownership type: (i) State owned enterprises (SOE) which include firms thatare fully owned by the State (used as the benchmark group); (ii) Stateshareholding firms (SF) include those who have already sold shares to thepublic but the State still has a controlling voting power with over 50% ofshares; (iii) Private firms (PF); (iv) Shareholding firms (SHF) that are eithernot owned by the State or where the State owns less than 50% of the shares;(v) Foreign joint-venture firms (JF) in which foreign investors own shares


but do not have full control and (vi) Fully foreign firms where foreigninvestors own 100% of shares.

Table 1 below summarizes the determinants of capital structure, defini-tions and theoretical predicted signs.

2.9. Measures of Capital Structure

Five indicators are often used to represent capital structure: Total liabilitiesdivided by total assets (TL); total debt divided by total debt and equity(DE); the ratio of debt to total assets; STD and LTD divided by book valueof equity. International studies usually use both book and market values tomeasure leverage indicators. Because at the time of the census Vietnam’sstock market has been in primary stages, no market values are available for

Table 1. Variable Definitions.

Proxy Variables Definitions Predicted

Sign

1 Collateralized Assets (CA) Fixed assets divided by total assets +

2 Profitability (ROA) Earnings before interest, tax and

depreciation divided by total

assets, lagged one year period

+/�

3 Effective Tax Rate (ETR) Income tax divided by earnings

before tax

+

4 Non-Debt Tax Shield

(NDTS)

Depreciation and amortization

expenses divided by total assets

�

5 Firm Size (LnS) Natural logarithm of firm sales,

lagged one year period

+/�

6 Growth Opportunity

(GTA)

Change in total assets between two

consecutive years (2002–2003)

scaled by previous year fixed

assets (2002)

�

7 Industry Classification (IC) Dummy variable for

manufacturing firms. Firm is

assigned value one if firm code

falls in range of 2911 to 4021 and

zero otherwise

+

8 Ownership Structure (OS) Five dummy variables for six types

of firm ownership types, using

state-owned firm as benchmark

+/�


most companies and therefore balance sheet values represent the value offirms in the present study.

We report the findings on the first three measures of leverage.

2.10. The Data

Data for the empirical analysis was obtained from an enterprise’s censusconducted by General Statistical Office for 2002–2003. This census was thefirst in Vietnam. The number of enterprises has been growing quickly sincethe year 2000 as result of the passage of the new enterprises law.

From the total sample, we exclude (i) small firms (under 10 workers)because these firms are mainly household businesses with very small capitaland revenue; (ii) co-operatives since they have very special characteristics(operating under the co-operative law, small capital and revenue, out-standing debts carried from many previous decades); (iii) firms that operateas state administration management or as public utilities.

We use the firm data for the years 2002–2003 to measure the variables. Allvariables except profitability and firm size are averaged over two years. Theprofitability attribute is taken from previous year (2002) to allow us todetermine whether profitability has more than a short-term effect onleverage. It is also relevant since lenders usually consider previous yearearnings as a basis to consider a firm’s requests for loans. In addition,Titman and Wessels (1988) suggest that measuring firm size by usingprevious LnS overcomes the problem of possible spurious relations betweensize and debt ratios. Past profitability affects both size and debt ratio in theshort term (profitable firms become larger and profitable firm increase theirnet worth).

3. EMPIRICAL FINDINGS


Table 2 reports descriptive statistics of three leverage measures – totalliabilities deflated by total assets (TL), ratio of debt and debt plus equity(DE) and total debt divided by total assets (DEBT) and of explanatoryvariables affecting leverage choice.2

Given the data constraint we use the average figure for the years2002–2003 in order to minimize possible effects of the adjusting process on


leverage. We believe that total liabilities divided by total assets is the bestmeasure of financial leverage. This is because creditors consider not onlyfirm’s debt but also other liabilities that affect the possibility to liquidate thefirm in case of financial distress. The two additional indicators of leverageare important and serve to check robustness of the results. In comparison toleverage level of developed countries, Vietnamese firm’s leverage level issomewhat similar to listed Chinese firms (0.51) but lower than that indeveloped countries where the figures are 0.58, 0.69, 0.73, 0.71, 0.73,respectively, in the US, Japan, Germany, France, South Korea (Sources:extracted from Huang & Song, 2006). Alternative leverage indicators suchas the ratio of debt to debt and equity or the ratio of debt to total assetsexhibit higher level in Vietnamese firms. The Vietnamese figures are 0.41 and0.42, respectively, while these indicators are in range of 0.16–0.27 for thecountries listed above.

In general, the size of Vietnamese enterprises is small in comparison withthose in developed countries. Median average annual sale is around VND3 billions (less than 200 thousand USD). The average rate of return on assetsis low (3%) and the effective income tax rate of 19%.

Table 3 reports a correlation matrix among leverage indicators and theexplanatory variables. Some points are worth noting here. First, a very high

Table 2. Descriptive Statistics of Leverage and Independent Variablesfor Vietnamese Firms (2002–2003).

Descriptive Statistics N=3,778

Variables Mean Min. Max. Std.

MTL 0.52 – 1.00 0.28

MDE 0.42 – 1.10 0.26

MDEBT 0.41 – 1.00 0.25

MCA 0.39 – 0.99 0.26

ROA 0.03 (0.69) 0.70 0.08

METR 0.19 (6.11) 0.32 0.17

MNDTS 0.16 – 1.00 0.18

LnS 8.98 1.39 16.19 1.91

GTA 0.17 (0.99) 1.00 0.31

All variables are calculated using book values. MTL is the average total liabilities divided by

total assets; MDE – average total debt divided by total debt and equity; MDEBT – average

total debt divided by total assets; MCA – average total fixed assets divided by total assets;

ROA – Operating income divided by total assets in the year 2002; METR – average income tax

rates; MNDTS – average non-debt tax shield divided by total assets; LnS – logarithm of sales in

the year 2002; GTA – changes in total assets through 2002–2003 deflated by total asset 2002.


correlation exists among three leverage indicators (r=0.99, 0.88, 0.99).It follows that the choice of the ‘‘appropriate’’ proxy for financial leverageis of little relevance. Second, the lower half portion of the matrix revealscorrelations between independent variables that are all statistically signifi-cantly different from zero. In light of the very low correlation between theseindependent variables the possible multi-colinearity problem is a non-issueas well.

3.2. Empirical Analysis

In this section, we present the empirical findings on the determinants ofcapital structure (Table 4).

We report both estimates of non-standardized and standardizedcoefficients. A quick glance shows that firm size, profitability and valuesof CA are strongly correlated with leverage. The standardized coefficientsimply that among the explanatory variables firm size and CA have thestrongest effects on leverage.

3.2.1. Collateralized Assets

Theoretical research predicts collateralized asset to be positively correlatedwith leverage. Prior empirical studies use fixed assets as its proxy and thefindings were consistent with theoretical predictions. Our empirical findingshowever yielded opposite results: leverage decreases as fixed asset valueincreases. We interpret this finding is as follows. Fixed assets and current

Table 3. Correlation Matrix among Variables.

MDE MDEBT ROA MCA METR MNDITS LnS GTA

MTL .99 .88 �.09 �.29 .07 �.04 .33 .13

MDE .99 �.15 �.28 .03 �.02 .36 �.03

MDEBT �.16 �.28 .02 �.03 .37 �.02

ROA �.03 .11 .09 .16 .05

MCA �.25 .26 �.27 .02

METR �.06 .06 .05

MNDITS .19 �.14

LnS �.11

Note: All the correlation coefficients that are reported in this table are significantly different

from zero. Notice the very high correlation between the three measures of leverage, MTL, MDE

and MDEBT in the sample.


Table 4. OLS Regression Estimates of Factors Affecting Financial Leverage.

Independent Variables

Model 1: MTL Model 2: MDE Model 3: MDEBT

Estimated

Coefficients

Standardized

Coefficients

Estimated

Coefficients

Standardized

Coefficients

Estimated

Coefficients

Standardized

Coefficients

(Constant) 0.15 0.09 0.08

(5.53)�� (3.91)�� (3.42)��

MCA �0.22 �0.20 �0.19 �0.19 �0.17 �0.18

(�11.81)�� (�11.28)�� (�10.73)��

ROA �0.54 �0.15 �0.68 �0.21 �0.68 �0.21

(�9.94)�� (�14.08)�� (�14.30)��

METR �0.09 �0.04 �0.15 �0.07 �0.15 �0.07

(�2.35)�� (�4.17)�� (�4.23)��

MNDTS �0.03 �0.02 �0.03 �0.02 �0.05 �0.03

(�1.04) (�1.47) (�2.02)��

LnS 0.05 0.35 0.05 0.38 0.05 0.39

(21.14)�� (23.12)�� (23.46)��

GTA 0.16 0.17 0.02 0.03 0.03 0.03

(11.62)�� (1.91)� (2.15)��

IC 0.00 0.00 0.00 0.01 0.01 0.01

(0.18) (0.31) (0.34)

SSF 0.00 0.00 �0.01 �0.03 �0.01 �0.02

(�0.10) (�1.63) (1.19)

PF 0.06 0.03 0.06 0.03 0.06 0.03

(1.83)� (2.02)�� (2.00)��

Th

eD

etermin

an

tso

fC

ap

ital

Stru

cture:

Evid

ence

from

Vietn

am

319

Table 4. (Continued )

Independent Variables

Model 1: MTL Model 2: MDE Model 3: MDEBT

Estimated

Coefficients

Standardized

Coefficients

Estimated

Coefficients

Standardized

Coefficients

Estimated

Coefficients

Standardized

Coefficients

SF 0.04 0.04 0.03 0.03 0.03 0.03

(2.53)�� (1.91)�� (1.98)��

JF �0.01 �0.01 0.00 0.00 0.00 0.00

(�.54) (0.144) (0.11)

FF �0.10 �0.10 �0.10 �0.12 �0.09 �0.11

(�5.02)�� (�5.74)�� (�5.52)��

N=3,755 Adjusted R-squared=0.21 Adjusted R-squared=0.23 Adjusted R-squared=0.23

Note:

The numbers in bracket are t-value.

DW=1.84, autocorrelation is rejected at 1% significance level; F-test=F (12, 3742)=85.9 are strongly significant.

Test for multi-colinearity: All VIF coefficients are less than 2.

All variables are calculated using book value. MTL is defined as average total liabilities divided by total assets; MDE – average total debt

divided by total debt and equity; MDEBT – average total debt divided by total assets; MCA – average total fixed assets divided by total assets;

ROA – operating income divided by total assets in year 2002; METR – average income tax rates; MNDTS – average non-debt tax shield

divided by total assets; LnS – logarithm of sales in the year 2002; GTA – changes in total assets through 2002–2003 deflated by total asset

2002. IC – dummy variable for Industry: IC=1 if firms type code falls in 2911–4921 (manufacturing machinery and equipment).

SSF, PF, SHF, JF and FF are dummy variables for firm ownership type using fully state owned firm as benchmark. SSF=1 denotes for

shareholding firms with over 50% state ownership; PF=1 for private firms; SHF=1 for other shareholding firms including ones with state

ownership under 50%; JF=1 for firms involve with foreign owners but less than 100%; FF =1 for wholly owned foreign firms.� mean statistically different from zero at the 1% level.�� mean statistically different from zero at the 5% level.�� mean statistically different from zero at the 10% level.

NAHUM

BIG

ER

ET

AL.

320

assets are two components of total assets, so they are perfectly negativelycorrelated. Current assets are more liquid than fixed assets. Lendinginstitutions in Vietnam generally examine the liquidation capacity of thefirm when they evaluate loan applications and therefore current assets areused as proxy for CA in Vietnamese firms. This explains why firms that ownless fixed assets relative to their current assets may have higher leverage.

3.2.2. Profitability

Our finding is consistent with prior empirical studies that found leverage tobe negatively correlated with profitability. Higher profitable firms use lessdebt. This evidence is in line with the pecking order hypothesis. The impactof profitability on leverage is somewhat stronger for debt–equity ratio(b=�.21) than to total liabilities (b=�.17).

3.2.3. Effective Income Tax Rate

Theoretical research predicts a positive impact of tax rate on leverage due tothe tax shield of interest on debt. In the present study, we found that theimpact of income tax is negative, although the extent of effect relativelysmall (a 1% increase in income tax rate could bring 0.04% to 0.07% drop intotal liabilities and debt–equity ratio). We interpret this result as follows.Income tax affects the choice of leverage in two directions. It has a positiveimpact through tax shields of interest on debt and negative impact becausehigher tax rates reduce the firm’s total profitability and hence firms find itmore difficult to obtain loans. It may be that the second effect dominates thefirst in the Vietnamese context.

NDTS – the effect is negative as predicted by theory but the magnitude ofimpact is small (0.02–0.03%).

3.2.4. Firm Size

The positive correlation of size with leverage is empirically valid. Thisresult is consistent with the findings of various prior studies (Marsh, 1982;Rajan & Zingales, 1995; Wald, 1999; Booth et al., 2001; Huang & Song,2006). The findings support the hypothesis that firm size may be interpretedas a reversed proxy of bankruptcy costs and a firm’s cost in issuance ofdebts. The size effect is similar for all three measures of leverage. A 1%increase in sales could result in an increase of leverage 0.05%. Observing thestandardized regression coefficients firm size shows the largest effects onleverage.


3.2.5. Growth Opportunities

We use the change in total assets between two consecutive years (2002–2003)scaled by average total assets 2002 to be a proxy for a firm’s growthopportunities. We found all three measures of leverage are positivelycorrelated with growth opportunities at the 1, 5 and 10% level ofsignificance. The impact of growth opportunities on total liabilities indicatoris stronger (b=.16) than on both debt–equity ratio and debt indicators(b=.03).

3.2.6. Industry Classification

Empirical finding is consistent with theoretical predictions. Firms thatoperate in the manufacturing machinery and equipment industry have alower financial leverage as compared to others. The coefficients for theindustry classification dummy variable are all negative (b=�.18, �.31, �.34for the three measures of leverage, respectively) because these firms’products are more difficult to quickly liquidate. This finding is alsoconsistent with our prior argument that fixed asset level is negativelycorrelated with leverage.

3.2.7. Ownership

Conflicting theories exist regarding the relationship of managerial andinstitutional ownerships and a firm’s financial leverage. In the present study,we used fully state-owned firms as a benchmark group, and several resultswere evident. First, controlling for other factors, both PF and SF (where theState owns less then 50% of the shares) have on average 3–4% higher levelof financial leverage. Comparing fully state-owned firms and Statecontrolled shareholding firms no statistically significant differences betweentheir financial leverage was found. Second, foreign JFs and fully foreignowned firms have lower financial leverage than state-owned firms, but thedifference is only statistically significant for fully foreign firms. Foreignfirms in Vietnam are primarily financed by parent companies and this sourceof financing is equity financing.

In addition to the empirical findings in this study, several featuresregarding Vietnamese firms are worth noting.

First, regarding capital structure, Vietnamese firms have low level of LTDin their total liabilities – the median LTD was only 20% of the total debt(median LTD of VND 78 millions or 5,000 USD) out of total liabilities(median 383 millions VND or 24,000 USD). A low level of long-term debtalso implies a low long-term investment and therefore low return in thelong run.


Second, the findings suggest that common factors that play a role in amarket economy are generally applicable to Vietnamese firms, with somenotable exceptions. Vietnam is still in transition towards a market economy,and the basic foundations of market mechanisms have not yet beenestablished. The stock market is still in its infancy stage. Firm financingmust therefore rely on external debt or on internally retained earnings. Bothsources are difficult to obtain because firms are required to have highliquidation capacity (current assets) and profitability.

Finally, the evidence also shows that taxes not have been a ‘‘push’’ factorsas firms are unable to take advantage of tax shields. Furthermore, personalincome tax has little or no impact on corporate financing decisions. Itfollows that firms trying to raise their value would tend to distributeearnings through dividends rather than undertake major investments inorder to get investment tax credits.

4. CONCLUSION

This study documents the factors affecting Vietnamese firm capitalstructure. The study shows that basic market principles of financingdecisions are generally applicable to Vietnamese firms. Empirical evidenceshows that for Vietnamese firms leverage increases with firm size andmanagerial ownership and decreases with profitability and NDTS. Leveragewas found to be correlated with industry characteristics. However, theextent of the market forces work is limited. First, in contrast to priorempirical studies that showed that a firm’s leverage increases with fixedassets and decreases with growth opportunities, we found the correlationsbetween leverage and these variables have opposite signs. Second, corporateincome tax has a negative effect on a firm’s financial leverage levels incontrast to the positive effect that is predicted by theoretical researchalthough the effect magnitude is weak. The correlations between firm’sattributes and leverage are robust across three different measures ofleverage.

The present study is believed to be probably the first to examine theassociation of leverage and firm’s attributes in Vietnamese firms. AsVietnam will soon joint WTO, this study should serves as a basis forresearchers and practitioners to further investigate the firm’s characteristicsand their effects on financing decisions in a market economy environment.Further research should focus on how firms distribute profit and why. Whatis influence of tax policy on leverage? The effect of firm size on leverage is


also an important issue. Because most Vietnamese firms are small sized, theywould face difficulty in finance by external debt. We feel that more work isneeded to examine the institutional ownership effects and in particular theeffects of managerial ownership on financing decisions.

NOTES

1. For a survey of findings see Harris and Raviv (1991).2. Total liabilities=debt (short- & long-term debt)+current liabilities; Total

assets= total liabilities+equity. The first measure (TL) reflects the propo-rtion of assets owed to external creditors, TL affects liquidation capacity,(TL ¼ ðdebtþ current_liabilitiesÞ=ðDebtþ current_liabilitiesþ equityÞ). As firms thatseek to grow and finance their operations using external financing they can borrow orsell equity. Thus, the second measure is defined as the ratio of debt-to-debt plus equity(DE ¼ ðdebtÞ=ðdebtþ equityÞ). DE thus reflects the financing choice. The thirdindicator (Debt ¼ ðdebtÞ=ðdebtþ current_liabilitiesþ equityÞ) reflects the fraction ofdebt financing out of total assets. The three measures of leverage are different for firmswith high level of current liabilities.

ACKNOWLEDGMENT

We thank Vietnam’s General Department of Statistics for providing us withthe data used in this study.

REFERENCES

Ang, J., Chua, J., & McConnel, J. (1982). The administrative costs of corporate bankruptcy:

A note. Journal of Finance, 37, 219–226.

Berger, P. G., Ofek, E., & Yermack, D. (1977). Managerial entrenchment and capital structure

decisions. Journal of Finance, 52, 1411–1438.

Booth, L., Aivazian, V., Kunt, A. D., & Maksimovic, V. (2001). Capital structure in developing

countries. Journal of Finance, 56, 87–130.

Bradley, M., Jarrel, G., & Kim, E. H. (1984). On the existence of an optimal capital structure:

Theory and evidence. Journal of Finance, 39, 857–880.

Chang, C. (1992). Capital structure as an optimal contract between employees and investors.


Chaplinsky, S., & Niehaus, G. (1993). Do inside ownership and leverage share common

determinants? Quarterly Journal of Business and Economics, 32(4), 51–65.

Cornett, M. M., & Travlos, N. G. (1989). Information effects associated with debt-for-equity

and equity-for-debt exchange offers. Journal of Finance, 44, 451–468.


DeAngelo, H., & Masulis, M. W. (1990). Optimal capital structure under corporate and

personal taxation. Journal of Financial Economics, 8, 3–29.

Fama, E. F., & French, K. R. (2002). Testing trade-off and pecking order prediction about

dividends and debts. Review of Financial Studies, 15, 1–23.

Friend, I., & Lang, L. H. P. (1988). An empirical test of the impact of managerial self-interest

on corporate capital structure. Journal of Finance, 43, 271–281.

Galai, D., & Masulis, R. W. (1976). The option pricing model and the risk factor of stock.


Grossman, S. J., & Hart, O. D. (1982). Corporate financial structure and managerial incentives.

NBER Working Paper No. R0398. Available at SSRN: http://ssrn.com/abstract=

578641

Harris, M., & Raviv, A. (1991). The theory of capital structure. Journal of Finance, 46, 297–355.

Huang, G., & Song, F. M. (2006). The determinant of capital structure: Evidence from China.

China Economic Review, 17, 14–36.

Jensen, M., & Meckling, W. (1976). Theory of the firm: Managerial behavior, agency costs and

ownership structure. Journal of Financial Economics, 3, 305–360.

Jensen, M. C. (1986). Agency cost of free cash flow, corporate finance and takeovers. American


Kester, W. C. (1986). Capital and ownership structure: A comparison of United States and

Japanese Corporations. Financial Management, 15, 5–16.

Korajczyk, R. A., Lucas, D., & McDonald, R. L. (1990). Understanding stock price behavior

around the time of equity issues. In: G. Hubbard (Eds), Asymmetric information,

corporate finance, and investment. University of Chicago Press, Chicago.

Leland, H., & Pyle, D. (1977). Information asymmetric, financial structure and financial

intermediation. Journal of Finance, 32, 371–388.

Long, M., & Maltyz, E. (1985). The investment financing nexus: Some empirical evidence.

Midland Corporate Finance Journal, 3, 53–59.

Mackie, M., & Jeffrey, K. (1990). Do taxes affect corporate financing nexus: Some empirical

evidence. Midland Corporate Finance Journal, 3, 53–59.

Marsh, P. (1982). The choice between equity and debt: An empirical study. Journal of Finance,

37, 121–144.

Modigliani, F., & Miller, M. H. (1958). The cost of capital, corporation finance and theory of

investment. American Economic Review, 48, 261–275.

Myers, S. (1977). Determinants of corporate borrowings. Journal of Financial Economics, 9,

147–176.

Myers, S. C., & Majluf, N. (1984). Corporate financing and investment decisions when firms

have information investors do not have. Journal of Financial Economics, 13, 187–221.

Rajan, R., & Zingales, L. (1995). What do we know about the capital structure? Some evidence

from international data. Journal of Finance, 50, 1421–1460.

Shyam-Sunder, L., & Myers, S. C. (1999). Testing static tradeoff against pecking order models

of capital structure. Journal of Financial Economics, 51, 219–244.

Smith, C. (1977). Alternative methods for raising capital right versus underwritten offerings.


Stulz, R. M. (1990). Managerial discretion and optimal financing policies. Journal of Financial


Titman, S. (1984). The effects of capital structure on a firm’s liquidation decision. Journal of

Financial Economics, 13, 137–151.


Titman, S., & Wessels, R. (1988). The determinant of capital structure choice. Journal of

Finance, 43, 1–19.

Wald, J. (1999). How firm characteristics affect capital structure: An international comparison.

Journal of Financial Research, 22, 161–187.

Warners, J. (1997). Bankruptcy costs: Some evidence. Journal of Finance, 32, 337–347.

Watson, R., & Wilson, N. (2002). Small and medium size enterprise financing: A note on some

of the empirical implication of pecking order. Journal of Business Finance and

Accounting, 29, 557–578.

Wisaeng, K., & Sorensen, E. H. (1986). Evidence on the impact of agency costs of debt in

corporate debt policy. Journal of Financial and Quantitative Analysis, 21, 131–144.


CHAPTER 16

SHAREHOLDERS’ VALUE

CREATION AND DESTRUCTION:

THE STOCK PRICES’ EFFECTS

OF MERGER ANNOUNCEMENT

IN JAPAN

Ognjenka Zrilic and Yasuo Hoshino

ABSTRACT

Based on the event study methodology this chapter tests value creation,

buying growth, and hubris hypotheses on the sample of 62 Japanese

mergers with announcement in period 1993–2005. We find an average

1.19% cumulative abnormal return in 3 days surrounding the merger

announcement. The findings suggest that differences in financial resources

allocation pattern may provide a source of value gain. Further, mergers

with fast-growing target are value enhancing when acquirer has prior

ownership in target. Announcement returns are adversely related to

acquirer’s past performance, implying that well-performing acquirers

possibly overestimate the true value of deal and overpay target.



r 2008 Published by Elsevier Ltd.

ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00016-7

327

dx.doi.org/10.1016/S1569-3767(07)00016-7.3d

1. INTRODUCTION

Abundant research literature has found that mergers represent a mixedblessing for shareholders of acquiring firms. Research in U.S. suggests thattargets experience significant wealth gains, while gains to acquirers are oftennull or insignificantly positive (Asquith, 1983; Jensen & Ruback, 1983;Mueller & Sirower, 2003).

Historically, mergers in Japan have been notably fewer than in U.S. andconsequently, less explored. Due to deregulation of Japanese financialmarket in recent years, merger activity has increased, in terms of both, thenumber and the value of deals. Increase is especially prominent following1998 due to changes in Government policy and amendments to Anti-monopoly Law. During 1999, total number of mergers first time reachedmore than 1000, and in consequent 6 years more than doubled to 2725in 2005 (MAAR Magazine, August 2006). After 1998, domestic mergersaccount for more than 60% of the number of total mergers and thistrend continues in years following. The dramatic increase in merger activitymakes it important to understand motives behind and consequences ofmergers.

Previous event studies on Japanese mergers reach contrasting resultsin different periods of analysis; acquirers gain (Pettway & Yamada, 1986;Kang, Shivdasani, & Yamada, 2000; Inoue, 2002), acquirers experiencewealth losses (Yeh & Hoshino, 2001). Moreover, skepticism regarding theconsequences of merger activity has grown with studies based on accountingdata finding that mergers tend to distort long-term profitability of mergingfirms (Odagiri & Hase, 1989; Yeh & Hoshino, 2002). A number of eventstudies on Japanese mergers have been focused on examining theimplications of keiretsu groupings, main bank system (Kang et al., 2000;Yeh & Hoshino, 2001) and cross-corporate shareholdings (Van Schaik &Steenbeek, 2004). However, deregulation of Japanese financial marketshas brought substantial changes in corporate structures and weakeningthe role of main bank in recent years. Thus, it is likely that currentlyincreased merger activity can be attributed to different factors frompreviously stated.

The objective of this chapter is to investigate the conditions under whichJapanese acquirers in domestic market earn abnormal returns by exami-ning alternative managerial behavioral assumptions regarding themerger activity decision making. Focus on post-bubble period allows usto examine recent merger activity, not extensively explored in previousliterature.

OGNJENKA ZRILIC AND YASUO HOSHINO328

2. THEORY AND HYPOTHESIS

Mergers have been topic broadly explored in different scientific disciplines.Abundant literature describes motives behind merger activity. One of themost comprehensive classification of merger motives is provided byTrautwein (1990). The classification is based on rationality of mergeractivity decision making. The rationality approach suggests that mergeroccurs as a result of sound decision making, aimed to achieve benefits forshareholders or pursue managerial personal goals. On the other hand,occurrence of a merger as a non-rational decision is suggested by processtheory. Process theory argues that merger occurs as a result of individuals’bounded rationality, organizational routines, or political power (Trautwein,1990). According to such non-rational approach, individuals have limitedinformation processing capabilities resulting in incomplete and biasedevaluations.

In this chapter, we integrate both perspectives and explore three theoriesof mergers often described as having high degree of plausibility; valuationtheory, empire-building theory, and process theory (Trautwein, 1990).

2.1. Rationality Perspective

2.1.1. Valuation Theory

Valuation theory argues that firm’s excess resources are not readily availableto other firms since existing market impediments (such as governmentregulations, limited information transfer, etc.) prevent smooth distributionof excess resources among firms. In light of this view, merger occurs whenacquirer has private information about target that would increase valueof combined entity through purchase of an undervalued target. Thus,acquirer’s management is motivated by valuable information aboutpotential advantages to be achieved from combining with the target’sbusiness (Trautwein, 1990). Barney (1988) suggests that value for acquirer isinduced when private and unique (inimitable) cash flow exists betweenmerging parties. Unique or inimitable cash flow means that a particulartarget has higher value for one acquirer than for the others. Harrison, Hitt,and Hoskisson (1991) provide evidence that this type of synergy isprominent when specific differences rather than similarities in resourcesallocation pattern exist between merging parties, since such differences arenot easily observable, neither easy to replicate by other market participants.Consequently, acquirer with source of synergy based on dissimilarity in

Shareholders’ Value Creation and Destruction 329

allocation of resources is likely to have an advantage, since reducedcompetition allows purchasing target at a lower price.

Models incorporating imbalance in financial resources and growthprospects are often used to explore acquisition likelihood, as well assynergy potential between merging firms. Palepu (1986) provides empiricalevidence that imbalance in financial resources and growth opportunities of afirm increases its probability of becoming an acquisition target.

Myers and Majluf (1984) developed a model in which high-liquid acquirerspurchase low-liquid targets with growth potential. In their model, merger haspotential for value enhancement when one firm’s excess liquid resourcescompletely cover the other firm’s investment needs. Liquid resources arevaluable since allowing firm to avoid undesirable external financing, forexample issuing of stocks in periods when firm is undervalued. The modelassumes asymmetric information (management has information that investorsdo not have; for example management knows more about firm’s value). Underasymmetric information, low-liquid firm not willing to issue stocks may notundertake all beneficial investment opportunities. Therefore, such firm hasa potential to increase its value by merging with high-liquid partner. Thus,the resource availability of one company can be combined with investmentneeds of the other in order to advance shareholders’ value. We omit Myersand Majluf ’s assumption regarding the specific direction of complementarywhere high-liquid acquirers purchase low-liquid, high-growing targets. Thus,the complementary in liquid resources and growth opportunities is proposedin both ways; high-liquid acquirers purchase high-growing targets and viceversa, high-growing acquirers combine with high-liquid targets.

H1a. Merger between one party with higher liquidity and the other withhigher growth will have a positive effect on acquirer’s shareholders returnat merger announcement.

The other source of synergy can arise from the difference in financialleverage of merging parties. If one of merging parties is leveraged and theother has unused debt capacity, the value of tax savings on incremental debtcould provide that both parties gain from exploiting unused debt capacity(Sudarsanam, Holl, & Salami, 1996). Bruner (1988) found that targets priorto merger have significantly more leverage than their acquirers and thecontrol sample, though his result does not support hypothesis that themarket value of merger is affected by this type of financial dissimilarity.

H1b. Purchasing target with higher financial leverage will have a positiveeffect on acquirer’s shareholders return at merger announcement.


2.1.2. Empire Building Theory

Empire building theory has its origins based on separation of ownership andcontrol in corporation. Management in public company act as agent forshareholders, performing with delegated authority on shareholders’ behalf.Agency problems can arise when managers serve their own interest that isnot aligned with shareholders’ interest. Evidence of interest conflicts can beperceived through large compensations, excessive perquisites or offensivegrowth, often referred to as ‘‘empire building.’’ Model proposed by Marris(1964) suggests that management pursue growth maximization at theexpense of shareholders’ wealth. Morck, Shleifer, and Vishny (1990) forU.S. acquisitions provide evidence that acquirer’s returns decrease whencompany acquires fast growing target, result consistent with hypothesis thatmanagement expropriate shareholders’ value by pursuing growth max-imization.

Previous studies on Japanese management commonly indicate internalgrowth as their preference in comparison to American management (Odagiri &Hase, 1989). This difference is attributed to the practice of long-termemployment, management attitudes to retain employees and employees’loyalty to company. Nevertheless, the recent surge of mergers in Japanrequires the reexamination of underlying factors that shape merger activity.

H2a. Purchasing fast growing target will have a detrimental effect onacquirer’s shareholders’ return at merger announcement.

However, the detrimental effect could be mitigated if acquirer hasownership in target prior to merger.

H2b. The detrimental effect of purchasing fast growing target will beweaker if acquirer has greater ownership in target prior to mergerannouncement.

2.2. Non-Rationality Perspective

2.2.1. Process Theory

Historically, scholars have commonly adopted rationality perspective thatportrays managerial decision to merge companies as a result of rationaldecision. According to non-rationality notion of process theory, individualshave limited ability of processing information, leading to incompleteevaluations, and tendency to make irrational decisions. The scarcity ofempirical evidence in respect to process theory can be seen as caused by


managerial common attempt to rationalize their actions (Trautwein, 1990).In this study, we examine the implications of managerial overconfidence,denoted as hubris (Roll, 1986).

Roll (1986) argues that management on basis of previous successmistakenly overestimates synergy potential and overpays target, resultingin the decrease of acquirer’s shareholders wealth. Even in mergers withsynergy potential, management can still commit valuation errors. Thus,hubris can be viewed as a factor affecting the size of bid rather than a motivefor merger (Mueller & Sirower, 2003).

Contrasting empirical evidence suggests that performance has a negativerelationship with risk taking decisions; poorly performing organizationstend to engage in riskier projects than well-performing organizations (Singh,1986). Morck et al. (1990) provide evidence that better-performing U.S.acquirers also make better mergers.

However, according to Roll’s (1986) hypothesis, successful managers aremore prone to overconfidence on the basis of previous success.

H3. Acquirer’s past performance will be negatively related to share-holders’ return at merger announcement date.

3. METHODOLOGY

3.1. Sample and Data

We identify merger events from M&A Data Book for period 1993–2002 andMAAR Magazines from 2003–2005. The sample consists of 62 domesticmerger events between stock listed companies that had merger announce-ment date in the period 1993–2005. Following previous studies (Yeh &Hoshino, 2001), financial industry mergers are excluded due to differentaccounting practices. Mergers in which acquirer purchases two or moretargets during 1 year are screened out since this would representconfounding event with difficulty to measure the effect of stock prices’changes. Mergers in which parent acquires already owned subsidiary arealso eliminated since they represent cases of legal status changes rather thanmerger in its pure form. The final sample comprises 62 acquirers for whichwe could obtain daily stock prices from Toyo Keizai Kabuka CD-ROM2002, 2005 and Yahoo.jp finance. Accounting and ownership data aresourced from Nikkei NEEDs CD-ROM 2006, various issues of NikkeiKaisha Nenkan and Kigyo Keiretsu Soran.


3.2. Event Study Methodology

In order to access the market reaction at merger announcement, we usetraditional event study methodology as proposed by Brown and Warner(1985). The price change at merger announcement is referred to asabnormal return, calculated as the difference between the observed returnof the security and the predicted (normal) return that would occur if themerger was not announced. This can be mathematically expressed asfollows:

uit ¼ Rit � ðai þ biRmtÞ (1)

where uit is abnormal return of acquirer security on day t, Rit daily return ofacquirer security on day t, Rmt the daily return of Topix on day t, a and bare estimated parameters from the market model

Rit ¼ ai þ biRmt þ eit (2)

The estimation window used in this chapter is 180 days prior to the eventwindow, that is, from 211 to 31 days before merger announcement. Theabnormal return is calculated on the basis of estimated parameters frommarket model for the test period from 30 days before merger announcementdate to 60 days following merger announcement.

The average abnormal return for N securities on a common day t iscalculated as follows:

AARt ¼

PNi¼1

uit

N(3)

Cumulative abnormal returns (CARs) up to date T are calculated as:

CART ¼XTt¼1

AARt (4)

To access statistical significance, average abnormal return on day t isstandardized by its standard deviation; the standard deviation is estimatedfrom initial 180 days time-series of average abnormal returns. Thestandardization procedure insures that abnormal returns are identicallydistributed, while time-series of average abnormal returns provide cross-sectional independence in the security-specific abnormal returns across time.


The test statistic for any event day t is

AARt

SðAARtÞ(5)

where

SðAARtÞ ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPt¼�31

t¼�210

ðAARt � AÞ2

179

vuuut(6)

A ¼1

180

Xt¼�31

t¼�210AARt (7)

For tests over the multiday intervals (t1, t2), the test statistics is the ratio ofcumulative mean abnormal return to its estimated standard deviation, asgiven by

Pt2t¼t1

AARt

Pt2t¼t1

S2ðAARtÞ

!1=2(8)

3.3. Dependent Variable

As dependent variable, we use CAR from 3 days before mergerannouncement to 1 day after, as employed in previous study (Yeh &Hoshino, 2001). In order to check for robustness, we use CAR from 5 daysbefore merger announcement to 2 days following the announcement.

3.4. Independent Variables

In order to evaluate the effect of merging partners’ liquidity-growthcomplementary (H1a) we use the product of (Acquirer’s Liquidity – Target’sLiquidity) and (Target’s Growth – Acquirer’s Growth) as proposed bySudarsanam et al. (1996). The interaction term is positive in cases when onepartner is more liquid and another grows faster. Liquidity is measured asthe ratio of working capital to total assets in the year before merger


announcement. Growth is estimated as 3 years average sales growth prior tomerger announcement year.

The second independent variable is the difference between target’s andacquirer’s financial leverage (H1b). Leverage is measured as the ratio oftotal liabilities to shareholders’ equity in year before merger announcement.

In order to test H2a and H2b we use target’s growth and the product oftarget’s growth and dummy variable that equals 1 when acquirer owns morethat 5% of target’s outstanding shares in year before merger announcement.We use 5% ownership level since for listed companies, disclosure ofownership above 5% is required according to shareholders’ rules.

We measure acquirer’s past performance (H3) as acquirer’s 3 yearsaverage industry adjusted ordinary income growth.

3.5. Control Variables

It has been suggested that mergers in Japan sometimes have ‘rescuemotives’; as in occasion when companies purchase business partners infinancial distress. These mergers could increase value if benefits ofcontinuing business relationship are larger than costs associated withrescuing target. We use target’s 3 years average industry adjusted ordinaryincome growth as a proxy for rescue merger.

Commonly used variable in mergers and acquisitions research is relativesize of merging companies. The relative size of target to acquirer is usuallyused as proxy for gains from economies of scale and scope. Asquith, Bruner,and Mullins (1983) found for U.S. mergers positive relationship betweenacquirers’ returns and relative size of target to acquirer; a bid for a targethalf the acquirer’s size produced 1.8% larger return comparing to bid for atarget one tenth of acquirer’s size. Villalonga and McGahan (2005) on basisof Hennart’s digestibility theory argue that merger is more complex topursue when size of partners is more balanced, since it is difficult for them tobecome digested by merging party. Relative size is estimated as log (totalassets of target/total assets of acquirer).

Based on previous literature, we also include acquirers’ large block-holdersand management ownership as control variables (Kang et al., 2000; Yeh &Hoshino, 2001). In large public companies, ownership components such aslarge block-holders and management ownership can be employed in order tomitigate agency problems. According to Prowse (1992), large block-holdersare usual in Japan with top five shareholders owning on average 33% offirm’s outstanding shares. Such concentrated ownership can provide strong


incentives for institutional investors to monitor management. We measureinstitutional ownership as a percentage of outstanding shares owned by top tenshareholders at the end of the year before merger announcement. Similarly, wedefine management ownership as the percentage of outstanding shares ownedby top management at the end of year before merger announcement.

Historically, M&A have not been attractive tools for corporaterestructuring in Japan. The significant increase of merger activity is evidentfollowing ‘‘Tokyo Big Bang’’ reforms in April 1998, focused on deregulationof financial markets, requirements for transparency in accounting practicesand corporate governance. The increase in the number of deals wasespecially prominent in 1999 due to revision of corporate laws; exchange(transfer) of shares for creating a 100% owned subsidiary was allowed in themajor amendment. We use dummy variable equal to one for mergersannounced after 1998 in order to examine the effects of policy changes.

4. ANALYSIS AND RESULTS

4.1. Univariate Analysis

Table 1 shows abnormal returns cumulated during specific windowintervals. The result suggests significantly positive market reaction before

Table 1. Cumulative Abnormal Returns (CARs) for Various Windows.

Window Interval Mean CAR (%) t-Statistic (Two-Tailed)

AD–5 to AD–1 1.22 1.705�

AD–3 to AD–1 0.90 1.622

AD–1 to AD 1.64 3.605��

AD–1 to AD+1 1.19 2.148��

AD–3 to AD+1 1.36 1.892�

AD–5 to AD+2 1.51 1.661

AD–5 to AD+5 0.84 0.719

AD–10 to AD+10 0.51 0.347

AD–20 to AD+20 1.39 0.679

AD–30 to AD+30 2.17 0.865

AD–30 to AD+60 4.14 1.350

AD+1 to AD+3 �0.31 �0.569

AD+1 to AD+5 �1.28 �1.783�

�Statistical significance at 10% level.��Statistical significance at 5% level.��Statistical significance at 1% level.


merger announcement (t=�5, �1). Also, market reacts favorably in variousintervals surrounding the initial announcement. If we examine patternfollowing deal announcement, CAR is significantly negative (t=1, 5). Kanget al. (2000) investigate mergers in period 1977–1993. They find gains in theshort window surrounding the deal announcement. Yeh and Hoshino (2001)for mergers in period 1981–1998 find statistically significant losses atannouncement date.

In order to preliminary access the effects of merging partners’ pre-mergercharacteristics on acquirers’ announcement return, we divide the sampleas shown in the Table 2. We compare the 5-days (t=�3 to t=1)

Table 2. Sample Divided According to Firms’ Pre-MergerCharacteristics for CAR (�3,1).

Pre-Merger

Characteristics of

Matched Firms

Number of

Observations

(% of CAR Positive)

Mean CAR (%)

and ( p-Value)

t-Test of

Difference

in Means

w2-Test ofDifference in

%>0

Rsize>0.5 24 1.17 t=�0.18 w2=0.01

(38.7) (0.31) (0.85) (0.906)

Rsizeo0.5 38 1.48

(61.3) (0.19)

Liqdif_Grdif>0 36 3.05�� t=2.57�� w2=6.23��

(66.6) (0.010) (0.0125) (0.0126)

Liqdif_Grdifo0 26 �0.98

(34.6) (0.312)

Levdif>0 38 1.74� t=0.58 w2= 2.84

(44.7) (0.101) (0.557) (0.091)

Levdifo0 24 0.75

(66.6) (0.569)

Tgrowth faster than

sample median

31 1.59 t=0.29 w2= 5.25��

(67.7) (0.108) 0.77 (0.022)

Tgrowth slower than

sample median

31 1.12

(38.7) (0.39)

Bperf higher than

industry average

32 0.76 t=�0.75 w2= 1.07

(46.8) (0.45) 0.454 (0.300)

Bperf lower than

industry average

30 1.27

(60) (0.12)

Tperf higher than

industry average

26 0.48 t=�0.92 w2= 0.19

(50) (0.643) 0.36 (0.665)

Tperf lower than

industry average

36 1.99�

(55.5) (0.099)

�Statistical significance at 10% level.��Statistical significance at 5% level.��Statistical significance at 1% level.


announcement return according to pre-merger characteristics of mergingpartners. The mean value of acquirers’ CAR (t=�3 to t=1) is 1.36%,statistically significant at 10% level. Then, we try to answer the questionwhich characteristics of matched firms affect the return more positively.

Table 2 shows acquirers’ 5-days mean CAR for various sub-samples offirms, t-test of the difference in means of sub-samples and w2 test of thedifference in CAR percent positive.

We check whether the relative size of merging parties affects acquirer’sreturn. After comparing return of acquirers with relatively small target(ratio less than 50% of acquirer’s size) to return of acquirers with relativelylarge target (ratio larger or equal to 50% of acquirer’s size), we found nodifference across sub-samples. The CARs are 1.17 and 1.48%, respectively.These differences are not statistically significant.

As a preliminary observation on the effect of liquidity-growth comple-mentary, we divide the sample into sub-categories according to positive andnegative Liqdif_Grdif variable. For positive value sub-group (36 companies),the mean acquirers’ return is 3.05% statistically significant at 1% level. Themean acquirers’ return for negative value sub-category (26 companies) is�0.98%, statistically insignificant. The t-test of difference in means issignificant at 5% level, suggesting that mergers between one partner beingmore liquid and another growing faster are favorably evaluated by marketat merger announcement. Statistically significant w2 test at 5% level confirmsthat mergers by companies with opposite growth-liquidity match areattractive options for business restructuring.

Second, we divide the sample according to difference in financial leverageof target and acquirer. We examine if the possibility to infuse capital intoleveraged target can be regarded as a source of gain for acquirer. The meanacquirers’ return for a sub-group with positive leverage difference(38 companies) is 1.74%, weakly significant at 10% level. The sub-groupwith negative leverage difference (24 companies) has insignificant meanacquirers’ return of 0.75%. Acquiring target with higher financial leverage isrelatively attractive in comparison to acquiring target with lower leverage.However, the t-test of difference in means and w2 test in difference of CARpercent positive are not significant.

Third, in order to access the effect of target’s growth on acquirers’ return,we divide sample into mergers with target growing faster than median andtarget growing slower than median. The mean return of sub-group withtarget growing faster than median is 1.59%, weakly significant ( p=0.108).The mean return of sub-group with target growing slower than median is1.12%, statistically insignificant. Both sub-groups have positive acquirers’


return above 1%, suggesting that in Japanese domestic mergers target’sgrowth is not factor significantly affecting acquirers’ return. t-test showsstatistically insignificant difference, confirming the previous finding.However, w2 test significant at 5% level indicates that mergers with fast-growing targets are relatively more attractive, since large percent ofacquirers in fast-growing target sub-sample have positive CAR (67.7%).

Fourth, we divide acquirers on sub-sample with ordinary income growthhigher than industry average and ordinary income growth lower thanindustry average to examine H3. We use ordinary income growth as a proxyfor quality of past performance. Acquirers with above-industry performancehave insignificant mean return of 0.75%, while acquirers with below-industry performance have mean return of 1.27%, also not significant atconventional level ( p=0.12). The pattern of acquirers’ returns of sub-groups is as predicted in H3. However, insignificant t and w2 tests do notsupport the predicted difference in returns across two sub-groups. Moreexplicit regression analysis examining the effect of acquirer’s pastperformance on stock return indicates that we cannot reject H3.

Finally, we examine the effect of target’s past performance as a proxy forrescue mergers in Japan. Acquirers purchasing targets with above-industryperformance have mean return of 0.48%, statistically indistinguishable fromzero. On the other hand, acquiring below-industry performing targetresulted in significant mean return of 1.99%. According to t and w2 tests, thedifference in sub-samples is not significant. However, above finding impliesthat acquiring poor-performing target does not affect negatively acquirer’sreturn at merger announcement. At contrary, Kang’s et al. (2000) study inperiod 1977–1993 indicates significantly negative effect of rescue mergers onacquirers’ return.

4.2. Multivariate Analysis

In this section, we show results of regression analysis using as the dependentvariable 5-days CAR (t=�3 to t=1).

Table 3 shows the results of Model 1 containing only control variables,Model 2 with valuation theory variables (H1a and H1b) added, Model 3with empire building theory variables (H2a and H2b) added and Model 4with hubris theory variable (H3) included. The F test of overall model lacksof significance in Model 1, and becomes significant as the main variables areadded. The first two hypotheses (H1a and H1b) test the effect of liquidity-growth and leverage differences on acquirers’ return and predict the positive


effect of respective differences on acquirers’ firm value at mergerannouncement. The coefficients on Liqdif_Grdif and Levdif are positiveand significant in all models; the results support H1a and H1b. In H2a, wepredict the negative effect of ‘buying growth’ on acquirer’s return. We argue

Table 3. Dependent Variable Cumulative Abnormal ReturnCAR(�3,1).

Variable Hypotheses (1) (2) (3) (4)

Liqdif �0.039 �0.037 �0.045

(0.87) (0.81) (0.92)

Grdif �0.007 0.002 0.053

(0.24) (0.04) (0.59)

Liqdif_Grdif H1a 0.365 0.335 0.340

(3.24)*** (3.08)*** (2.60)**

Levdif H1b 0.0003 0.0004 0.0005

(2.18)** (3.61)*** (3.79)***

Tgrowth H2a �0.085 �0.111

(0.53) (0.62)

Tgrowth_Own5 H2b 0.974 0.988

(3.14)*** (3.03)***

Own5 0.003 0.008

(0.22) (0.49)

Bperf H3 �0.008

(2.75)***

Tperf �0.002

(1.48)

Rsize �0.013 0.012 0.011 0.014

(0.51) (0.43) (0.38) (0.49)

Block 0.034 �0.013 �0.027 �0.028

(0.58) (0.28) (0.57) (0.56)

Mown �0.045 0.0002 �0.030 �0.039

(0.53) (0.03) (0.37) (0.45)

Year (after 1998) 0.044 0.043 0.044 0.047

(2.73)*** (2.77)*** (2.81)*** (3.10)***

Constant �0.023 �0.012 �0.001 �0.004

(0.80) (0.49) (0.04) (0.15)

Observations 62 62 62 62

R2 0.122 0.275 0.333 0.361

F-statistics 1.98 2.52a 2.27a 2.08a

(1) Absolute value of t statistics in parentheses.

(2) apo0.05.�Statistical significance at 10% level.��Statistical significance at 5% level.��Statistical significance at 1% level.


that prior ownership of acquirer in target will act as a factor bufferingthis detrimental effect (H2b). The coefficient on Tgrowth is negative, butinsignificant in all models. Moreover, positive and significant coefficient onTgrowth_Own5 in all models indicates that buying growing target is beneficial

Table 4. Dependent Variable Cumulative Abnormal ReturnCAR(�5,2).

Variable Hypotheses (1) (2) (3) (4)

Liqdif �0.020 �0.018 �0.029

(0.49) (0.42) (0.64)

Grdif 0.049 0.044 0.064

(1.37) (0.86) (0.81)

Liqdif_Grdif H1a 0.260 0.240 0.214

(2.65)** (2.42)** (1.76)*

Levdif H1b 0.0003 0.0004 0.0005

(2.56)** (2.99)*** (3.09)***

Tgrowth H2a �0.043 �0.043

(0.26) (0.24)

Tgrowth_Own5 H2b 0.964 0.975

(2.10)** (2.01)**

Own5 0.003 0.011

(0.18) (0.56)

Bperf H3 �0.010

(3.53)***

Tperf �0.001

(0.65)

Rsize �0.021 0.001 0.002 0.011

(0.85) (0.05) (0.07) (0.35)

Block 0.017 �0.012 �0.026 �0.022

(0.32) (0.24) (0.51) (0.42)

Mown �0.062 �0.006 �0.043 �0.050

(0.79) (0.08) (0.49) (0.56)

Year (after 1998) 0.052 0.050 0.050 0.057

(3.39)*** (3.30)*** (3.32)*** (4.00)***

Constant �0.014 �0.009 0.001 �0.011

(0.47) (0.31) (0.03) (0.37)


R2 0.168 0.256 0.315 0.365

F-statistics 2.88a 2.28a 2.09a 2.13a

(1) Absolute value of t statistics in parentheses.

(2) apo0.05.�Statistical significance at 10% level.��Statistical significance at 5% level.��Statistical significance at 1% level.


for acquirers with prior ownership in target above 5%. In H3, we examinethe effect of acquirer’s industry adjusted past performance on returns atannouncement. The logically expected effect would be that market favorsmergers by above-industry performing firms, since such firms have alreadyproven their ability to make good business decisions. However, thecoefficient on Bperf is negative and significant, implying that market issuspicious at announcement of mergers by successful firms.

In order to check for robustness, we estimate the models using asdependent variable 8-days CAR (t=�5 to t=2). Table 4 shows that resultsare unaltered, essentially supporting the same hypotheses.

5. CONCLUSION

Domestic mergers in Japan have expanded rapidly in recent years, in termsof both, the number and value of deals. Considering the growing importanceof mergers in Japan, this study investigates the valuation effects of Japanesedomestic mergers and the underlying sources of value gains and losses.

Our findings indicate that value can be advanced in mergers when onepartner is more liquid while the other grows faster, due to ability of liquidparty to finance partner with growth opportunity. Also, acquiring targetwith higher financial leverage can become a source of gain, possibly due totax savings on incremental debt for newly merged entity. Thus, differences infinancial resources allocation pattern may provide a source of valueenhancement. This view supports Harrison et al. (1991) argument thatuniquely valuable synergy might be generated when differences exist betweenresources of merging companies. Such differences are difficult to perceiveand emulate by potential competitive bidders due to asymmetric informa-tion, therefore decreasing the probability of competitive bids and an auction.

Further, our study examines managerial objectives in terms of ‘buyinggrowth’. Extensive research in U.S. and U.K. suggests that managementcompensation depends on the size of firm (Firth, 1980). Morck et al. (1990)provide evidence that U.S. market perceives unfavorably mergers with rapidlygrowing targets. They also find negative effect of diversifying mergers onacquirers’ return. Nevertheless, in our sample there are only eight cases ofmergers between firms belonging to different industries. This suggests thatmergers in Japan are largely concentrated in adjacent business field, wherenewly acquired business is complementary to acquirer’s main business in orderto strengthen utilization of existing resources, market power and economies ofscale and scope. Thus, growth through diversifying mergers in Japan is not as


popular as in case of U.S. merger market. Mergers concentrated in relatedfield of business are compatible to Japanese management preference forinternal growth, possibly occurring when internal growth efforts are hinderedby internal resource constraints (Odagiri & Hase, 1989). The tendency ofJapanese management to merge related business versus motivation ofAmerican management to grow through conglomerates might be the reasonfor different result we obtain in terms of buying growth.

Also, we explore the impact of the quality of acquirer’s past performance onthe stock return. Morck et al. (1990) provide evidence that U.S. market favorsmergers by well-performing firms in contrast to our findings for Japanesemergers. In Japan, stockholders’ return is lower following the announcementof mergers by well-performing firms. Our finding is consistent with hubristheory by Roll (1986), suggesting that management of well performing firms isprone to overconfidence, resulting in overvaluation of a merger and a detri-mental effect for acquirer’s shareholders. Since large scale mergers in Japan areoften described as friendly deals between firms with established relationships(Kester, 1991; Kruse, Park, Park, & Suzuki, 2006), it can be argued thatproblem of winner’s curse combined with hubris is less likely to occur inJapanese market. However, due to lack of market for corporate control,management in Japan does not face monitoring by external market forces.Thus, increase in merger activity by overoptimistic management is possible.

In conclusion, our findings support the view that focus on specificresources combination between acquirer and target could become essentialfor shareholders’ value upturn. In particular, financial resources spilloverbetween merging companies has a potential to generate value enhancingsynergy. Future research could access the impact of differences in resourceson long-term performance of a merger. We find that well-performingacquirers reach lower stock return at merger announcement, indicating thatsuccessful managers might be overconfident in their valuation of target firms.This evidence requires further consideration especially through examininglong run performance of mergers by above industry performing firms. Ourresults are a subject to constraints due to, most importantly, short-run eventstudy methodology. In order to access the full impact of complex events suchas mergers, it should be accounted for long-term measurements as well.

ACKNOWLEDGMENT

We appreciate comments and suggestions from anonymous reviewers. Thisstudy was supported by the second author’s research grant from the Japan


Society for the Promotion of Sciences’ Grant-in-Aid for Scientific Research (C).Earlier version of this paper was presented at the 2007 Eastern Japan RegionalConference of Japanese Association of Administrative Science in Japan.

REFERENCES

Asquith, P. (1983). Merger bids, uncertainty, and stockholder returns. Journal of Financial


Asquith, P., Bruner, R. F., & Mullins, D. W. (1983). The gains to bidding firms from merger.


Barney, J. B. (1988). Returns to bidding firms in mergers and acquisitions: Reconsidering the

relatedness hypothesis. Strategic Management Journal, 9, 71–78.

Brown, S. J., & Warner, J. B. (1985). Using daily stock returns, the case of event studies. Journal

of Financial Economics, 14, 3–31.

Bruner, R. F. (1988). The use of excess cash and debt capacity as a motive for merger. Journal of

Financial and Quantitative Analysis, 23, 199–217.

Firth, M. (1980). Takeovers, shareholder returns and the theory of the firm. Quarterly Journal

of Economics, 94, 235–260.

Harrison, J. S., Hitt, M. A., & Hoskisson, R. E. (1991). Synergies and post-acquisition

performance: Differences versus similarities in resource allocations. Journal of Manage-

ment, 17, 173–190.

Inoue, K. (2002). Nihon no M&A ni okeru Torihiki Keitai to Kabuka Kouka (Transaction type

and stock price effect in Japanese). Keiei Zaimu Kenkyu (Japan Journal of Finance),

22(2), 107–120(in Japanese).

Jensen, M. C., & Ruback, R. S. (1983). The market for corporate control. Journal of Financial


Kang, J. K., Shivdasani, A., & Yamada, T. (2000). The effect of bank relations on investment

decisions: An investigation of Japanese takeover bids. Journal of Finance, 55, 2197–2218.

Kester, W. C. (1991). Japanese takeovers: The global contest for corporate control. Boston, MA:

Harvard Business School Press.

Kruse, T. A., Park, H. Y., Park, K., & Suzuki, K. (2007). Long-term performance following

mergers of Japanese companies: The effect of diversification and affiliation. Pacific-Basin

Finance Journal, 15(2), 154–172.

Marris, R. (1964). Managerial capitalism in retrospect. New York: Palgrave MacMillan.

Morck, R., Shleifer, A., & Vishny, R. W. (1990). Do managerial objectives drive bad

acquisitions? Journal of Finance, 45, 31–48.

Mueller, D. C., & Sirower, M. L. (2003). The causes of mergers: Tests based on the gains to

acquiring firm’s shareholders and the size of premia. Managerial and Decision

Economics, 24, 373–391.

Myers, S., & Majluf, N. (1984). Corporate financing and investment decisions when firms have

information that investors do not have. Journal of Financial Economics, 13, 187–221.

Odagiri, H., & Hase, T. (1989). Are mergers and aquisitions in Japan going to be popular too?

International Journal of Industrial Organization, 7, 49–72.

Palepu, K. (1986). Predicting takeover targets: A methodological and empirical analysis.

Journal of Accounting and Economics, 8, 3–35.


Pettway, R. H., & Yamada, T. (1986). Mergers in Japan and their impacts upon stockholders’

wealth. Financial Management, 15, 43–52.

Prowse, S. D. (1992). The structure of corporate ownership in Japan. Journal of Finance, 47,

1121–1140.

Roll, R. (1986). The hybris hypothesis of corporate takeovers. Journal of Business, 59, 197–216.

Singh, J. V. (1986). Performance, slack, and risk taking in organizational decision making.

Academy of Management Journal, 29, 562–585.

Sudarsanam, S., Holl, P., & Salami, A. (1996). Shareholder wealth gains in mergers: Effect of

synergy and ownership structure. Journal of Business Finance and Accounting, 23,

673–698.

Trautwein, F. (1990). Merger motives and merger prescriptions. Strategic Management Journal,

11, 283–295.

Van Schaik, D., & Steenbeek, O. W. (2004). Price and volume effects of merger bids in Japan.

Social Science Research Network Working Paper, available at http://papers.ssrn.com/

sol3/papers.cfm?abstract_id=592081xPaperDownload

Villalonga, B., & McGahan, A. M. (2005). The choice among acquisitions, alliances and

divestitures. Strategic Management Journal, 26, 1183–1208.

Yeh, T., & Hoshino, Y. (2001). Shareholders’ wealth, bank control, and large shareholders: An

analysis of Japanese mergers. Japan Journal of Finance, 21, 150–166.

Yeh, T., & Hoshino, Y. (2002). Productivity and operating performance of Japanese merging

firms: Keiretsu-related and independent mergers. Japan and the World Economy, 14,

347–366.


CHAPTER 17

TAKEOVERS AND SHAREHOLDER

VALUE CREATION ON THE STOCK

EXCHANGE OF THAILAND

David E. Allen and Amporn Soongswang

ABSTRACT

There are few studies of take over effects in emerging stock markets and

of whether such events result in value-increasing or value-decreasing

effects for the successful targets and bidders. This study analyses the

impact of successful takeovers on the Stock Exchange of Thailand

(SET). Both target and bidding firms’ performances during a period of 12

months before and after the takeover are investigated. Abnormal returns

are measured using an event study approach; applying two models and

three parametric test statistics. The results suggest that Thai takeover

effects are wealth-creating for both offeree and offeror shareholders.

1. INTRODUCTION

Given that mergers and acquisitions take place under conditions ofuncertainty, it is not surprising that not all business combinations aresuccessful. Past studies show that successful firms that combine businessescan benefit from economies of scale, but diversification for other reasons





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00017-9

347

dx.doi.org/10.1016/S1569-3767(07)00017-9.3d

tends to be less successful. Forms of the event study methodology has beenthe predominant method used to measure share price responses to mergeror takeover announcements, and most studies suggest that takeovers createshareholder wealth. Jensen (2006) suggested that the market for corporatecontrol has generated large benefits of around US535 billion to event firms’shareholders in approximately the 50 largest US takeovers in the prior4 years. Some other prior studies suggest that takeovers have negativeeffects, but more recent studies during the 1990s have predominantlypositive outcomes. Therefore, the results are mixed, though they suggestthat anticipated wealth creation can be viewed as the likely rationale behindmerger and acquisition decisions.

2. REVIEW OF PRIOR STUDIES

2.1. The Evidence of the Existence of Positive Abnormal

Returns to Target Firms

Datta, Pinches, and Narayanan (1992) compare their results with theconclusions provided in two previous reviews by Jarrell, Brickley, andNetter (1988) and Jensen and Ruback (1983). The combined US evidence isfairly consistent in showing more value creation for target firms becausetheir shareholders earn significant gains and achieve larger gains in tenderoffers rather than in mergers.

Studies of other stock markets provide similar evidence. For example, DaSilva Rosa, Izan, Steinbeck, and Walter (2000), in an Australian study, andDumontier and Petitt (2002), in a French study, both report that the targetfirm’s shareholders benefit significantly from takeover announcements.Goergen and Renneboog (2004) examine large European takeovers,suggesting that short-term wealth effects are remarkably similar to thosereported in the US and UK studies. They find positive takeoverannouncement effects of 9% for target firm’s shareholders, but thecumulative abnormal returns (CARs) also include the price run-ups overthe 2-month period prior to the announcement date of around 23%.

Bruner (2002) summarises the findings of 21 studies and reveals that thetarget firm’s shareholders receive significantly and considerable positiveabnormal returns, despite variations in the time period, type of acquisition(mergers vs. tender offers) and observation period. The results are consistentwith two previous surveys1 conclusions that the target firm’s shareholdersrealise average abnormal returns (AARs) in the range of 20–30%.

DAVID E. ALLEN AND AMPORN SOONGSWANG348

Campa and Hernando (2004) summarise the findings of 13 studies andreport that the target firm’s shareholders obtain significantly positive returnsin almost all cases. Recent studies provide additional evidence. For example,Santos, Errunza, and Miller (2003) find significant wealth gains accrue tothe foreign target firm’s shareholders regardless of the type of acquisition;Campa and Hernando (2004) also find that the target firm’s shareholdersreceive on average a statistically significant CAR of 9% for mergers.

Conversely, Agrawal and Jaffe (2002) summarise prior studies of the pre-acquisition performance of target firms, in which eight from twelve studiesshow negative abnormal returns, but only two studies record significantlynegative abnormal returns, and the remaining four studies report insigni-ficantly positive abnormal returns. Their results are consistent with twoother studies; Danbolt (2002) and Karceski, Ongena, and Smith (2000),cited in Campa and Hernando’s (2004) survey. These report negativeabnormal returns (for windows smaller than 10 days prior to the event date).

In conclusion, the merger and acquisition transaction appears to delivera premium return for the target firm’s shareholders, which are on averagesignificantly positive in the range of 20–30%. The results also suggest thatthe larger the event window the greater the increase in the amount andsignificance of abnormal returns. In addition, positive abnormal returns inthe days prior to the announcement date also reported, suggesting marketanticipation of information subsequently disclosed about the takeovers.

2.2. The Evidence of Negative Abnormal Returns to Bidding Firms

Datta et al. (1992) cite some contrary evidence to that reported in Jensenand Ruback (1983) and Jarrell et al. (1988). In particular, they find that thebidding firm’s shareholders do not gain at all; whether successful or not.Jensen and Ruback (1983) find that the bidding firm’s shareholders gain insuccessful mergers and lose in unsuccessful transactions. Jarrell et al. (1988)report declining returns for the bidding firm’s shareholders in acquisitionsundertaken in the 1970s and 1980s, compared to the 1960s; while Datta et al.(1992) find the decline over time is insignificant. A recent study of successfulcross-border mergers by Black, Carnes, and Jandik (2001) suggested thatbidding firm’s shareholders suffer significantly negative abnormal returnsof �13.20 and �22.90% over 3- and 5-year windows, respectively.

Bruner (2002) summarises the findings of 44 studies and 20 of thesestudies report negative returns: with 13 of these 20 studies suggestingsignificantly negative returns. The negative abnormal returns vary between

Takeovers and Shareholder Value Creation 349

�1 and �3%. He concludes that one-third (13) show value destruction, one-third (14) show value conservation and one-third (17) show value creation.

Campa and Hernando (2004) summarise the findings of 17 studies, 10 ofthese studies report negative abnormal returns which vary between less than1 and �5%, and in most cases are significantly different from zero. Sevenmore studies report zero or positive abnormal returns ranging from 0 to 7%.It is noteworthy that most results are very small when compared with thereported abnormal returns for the target firm’s shareholders in previousstudies. The findings are distributed rather evenly amongst studies that showboth value-decreasing and value-increasing effects. Thus, the outcomes forthe bidding firm’s shareholders are inconclusive.

2.3. The Evidence of Zero or Positive Abnormal Returns to Bidding Firms

The survey by Jensen and Ruback (1983) suggests that the bidding firm’sshareholders in successful tender offers realise statistically significantpositive gains ranging from 2.40 to 6.70%,2 and the weighted averagereturns are 3.80%. The evidence about bidding firm shareholders’ returns inmergers is mixed,3 and it might be concluded that on the whole, the returnsfor bidding firm’s shareholders in mergers are approximately zero.

By contrast, surveys by Bradley, Desai, and Kim (1988) and Weston andCopeland (1992) suggest that the acquiring firm’s shareholders gainsignificantly positive returns. In Australia, Brown and da Silva Rosa(1998) report that acquisitions increase bidding firm shareholders’ equityvalue. Some other studies find zero or small positive abnormal returns suchas Eckbo and Thorburn (2000), Loderer and Martin (1990) and Maquieria,Megginson, and Nail (1998). A recent study by Goergen and Renneboog(2004) analyses the wealth effects of large (intra) European takeovers andfinds that share prices of the bidding firms positively respond with astatistically significant announcement effect of 0.70%. Herman andLowenstein (1988) report that ROC (pre-tax returns on total capital) to thebidding firms (using tender offers) increase from 14.70 to 19.60%. Parrinoand Harris (1999) find that the bidding firm’s shareholders experience asignificant and positive 2.10% operating cash flow return after mergers.Similarly, Ghosh (2001) shows that the acquiring firm’s shareholders obtaincash flows that increase significantly for cash acquisitions. Ghosh (2002)finds that on average, the post-acquisition market share of acquiring firmsincreases about 20% from the pre-acquisition level.


In conclusion, the evidence about returns achieved by bidding firm’sshareholders is inconclusive with reports of negative, zero and positiveabnormal returns. The negative returns vary between less than 1 to �7%. Incontrast, many studies find zero or positive returns ranging from 0 to 7%(except for Loughran & Vijh, 1997), but these effects are very small whencompared with the returns obtained by target firm’s shareholders.

Obviously, most studies have concentrated on short-term performanceinvestigation, but a number of more recent studies have been devotedto long-term return examination. However, Bruner (2002) notes thatmany studies show a slight tendency for declining returns over time. Thereturns seem to be more positive in the 1960s and 1970s than in the 1980sand 1990s, except for mergers and acquisitions in technology and bankingwhere returns for the bidding firm’s shareholders increased in the 1990s.Consistent with Bradley et al. (1988), Jarrell et al. (1988), and Loderer andMartin (1992).

2.4. Thai Literature Review

Most studies of mergers and acquisitions have focused on developedmarkets such as the US stock market, the UK or European stock marketsand the Australian stock exchange. Only a few merger studies concentrateon developing or emerging stock markets; see, for example, Estrada,Kritzman, and Page (2004) and Fernandes (2005). Lins and Servaes (2001)examine the value of corporate diversification in seven emerging markets,including the Thai stock market, and find that diversified firms experiencea discount of approximately 7% compared with single-segment firms. Veryfew studies have focused on merger and acquisition activities on the SET.

2.5. The Evidence of Positive Abnormal Returns to Target Firms and

Negative Abnormal Returns to Bidding Firms

Varaboontweesuk (2003) suggests that on the announcement date, theAARs and cumulative average abnormal returns (CAARs) for the targetfirm’s shareholders are insignificantly positive at 0.12% and significantlypositive at 12.72%, respectively. Meanwhile, those for the bidding firm’sshareholders are insignificant and negative at �0.35% and significant andnegative at �5.59% consecutively. Similarly, Leemakdej (1998) also findsthat the positive CAARs over the period (�2, 0) for the target firm’s


shareholders are about 20%. In addition, both studies suggest that there isevidence that takeover information leaks into the market about 40 days(approximately 2 months) prior to the announcement date, generatingpositive abnormal returns for the target firm’s shareholders. OnlyVaraboontweesuk (2003) reports that the news is available to the marketapproximately 10 days before the event date, resulting in negative abnormalreturns for the bidding firm’s shareholders.

2.6. The Evidence of Positive Abnormal Returns to Successful

Target and Bidding Firms

Anuchitworawong (2001) finds that abnormal returns occur before thetakeover announcement and are rising until the announcement date,resulting in positive CAARs of approximately 15% over the period(�2, 0) for the successful target firm’s shareholders. In addition, theabnormal returns continue to increase for another 25 days before graduallydecreasing. Anuchitworawong (2001) also suggests that on average, thesuccessful bidding firm’s shareholders realise positive CAARs of about2–12% during the offer period,4 but these returns decline graduallyafterwards. The market displays significantly positive abnormal returns15 days prior to the event date but negative abnormal returns 5 days priorto the event date.

Clearly, these prior results are consistent and suggest that on average,takeover effects are positive, resulting in substantially positive abnormalreturns for the target firm’s shareholders, but small negative abnormalreturns for the bidding firm’s shareholders. Meanwhile, both successfultarget and bidding firms’ shareholders gain positive abnormal returns. Also,the market positively reacts to the successful takeover news before theannouncement date. This is consistent with recent studies of developed stockmarkets by Jensen (2006) and Morellec and Zhdanov (2004).

However, most studies focus on stock returns over short time periods(a few days or a few months) around the takeover announcements.Furthermore, there has not been a great deal of attention paid to contrastingsuccessful target firms with successful bidding firms. Prior Thai studies havebeen predominantly one-sided, focussing on either a target or bidding firm’seffects rather than contrasting the two. Furthermore, these studies useddaily stock price data, examined short-window abnormal returns andapplied only the market model plus a limited range of statistical tests. Weknow that event study results are sensitive to the metrics used. Thus, a more


comprehensive examination of merger and acquisition performance on theSET is justified. In this study, in addition to including more sample data bycovering a longer period from year 1992 to 2002, both successful target andbidding firms’ are examined using a variety of metrics, For example; twomodels and three parametric significance tests were applied: these includethe market model and market-adjusted (zero-one) model; a standardised-residual test, a standardised cross-sectional test and conventional t-tests.

3. METHODOLOGY

3.1. Data

There are three major sources for the stock price data used in this studywhich were governed by the following considerations:

(1) To alleviate issues relating to survivor bias, we are careful to includedelisted companies and companies in ‘‘REHABCO’’5. Thus, wescrutinised the list of delisted companies, the list of companies tradedunder the rehabilitation sector or ‘‘REHABCO,’’ plus the list of totalcompanies listed on the SET, the list of listed companies that have theirnames changed, supplementary information about listed companies(in Form 56-1) and the SET’s rules and guidelines regarding takeoversand delisting were collected from the SET.

(2) All tender offer statistics between August 1992 and October 2002, andother specific information such as the rules, conditions and proceduresto be followed in tender offers, other additional information associatedwith tender offers, offerors (bidding firms) and offerees (target firms)was gathered from the Securities and Exchange Commission, Thailand(SEC).

(3) The Datastream database was used to provide the information aboutthe stock prices of the sample firms.

3.2. Research Method

The analysis in this study is based on the tender offer statistics obtainedfrom the SEC between 1992 and 2002. The sample firms were classifiedaccording to whether they were involved as a target or bidder. Moreover,when there were any tender offers that involved repeated targets or bidders,


any of the same target and the same bidder or the same target and adifferent bidder or a different target and the same bidder, the latest tenderoffer was first selected, then, the second, and then the third latest one, in thissequence. This was to optimise the data utilised from the limited availablesources of data. However, we imposed a requirement of an interval of noless than 1 year’s length between each tender offer.

During the time selected, the number of takeovers on the SET amountedto 151 tender offers. However, the initial sample finally was reduced to 52tender offers or 52 target firms, and 50 successful target firms. Likewise,for the bidding firms, there are about 70% of the total 151 tender offersthat were individuals, or made up of a group of investors, or non-listedcompanies, or a mixed combination with these characteristics. To obtaina larger sample size, any single bidder or consortium or a sub-set of theconsortium that tendered an offer to the targets on the SET at that timewere included in the sample. Thus, approximately 30% or 44 out of the total151 tender offers (74 bidders) were selected to be the bidder sample for thisstudy. Nevertheless, the initial sample reduced to 28 tender offers by 42bidding firms, and included 39 successful bidding firms, which are sub-setsof the total 42 bidding firms. These were selected to be the bidder samples.The event firms were selected according to the selection criteria set outbelow:

(1) A tender offer was classified as being ‘‘successful’’ if the bidder increasedits holding of the target shares or purchased at least some6 of theoutstanding target shares that were tendered for. Thai securitylegislation also defines a proportion from 25% of the target shares’holdings as a strategic shareholder and the bidder is required to tenderan offer for the total remaining outstanding shares of the target.

(2) Any tender offer was excluded from the sample when it occurred withthe purpose of a de-listing.7 Those cases were also deleted when thetender offer was cancelled later or the target firm was in the process oflisting.

(3) The ‘‘survivorship’’ period of time required in the study is a period ofmonths from over (�48, +16) around the event, due to the limitation ofavailable stock price data. Either selected targets or bidders had to becompany listed on the SET at that time and the stock price data had tobe available for the required period of time.

The research is largely based on a sample of successful tender offers,investigating successful target and biding firms. The analysis emphasisesabnormal performance measurement using monthly stock price data.


The firm’s stock price reaction to the takeover announcement was estimatedas the rate of abnormal returns for the target and bidding firm’s share-holders, respectively. The abnormal returns were derived from a particularstock that responded to the event study as previously mentioned. The rawreturns for 1 month were simply the change in stock price and any dividendspaid, divided by the closing stock price the month before. The abnormalreturn was defined as the difference between the realised return observedfrom the market and the benchmark return over the period around thetakeover announcements. Also, it was defined ‘‘at the announcement oftakeovers’’ or ‘‘around the takeover announcements’’ as the event-windowof the examination. The event period is the bid period or (�12, 0, +12)months, month ‘‘0’’ was defined as the event month, and the eventdate (month) was defined as the submission date (month) of the tender offerby the bidder to the SEC or the date (month) that the proposal was filedat the SEC.

3.3. Market Model

To examine the effect of the event on each stock, i, control is made for thenormal relation between the return on stock i during month t, and the returnon the market index Rm.

Rit ¼ ai þ biRmt þ �it

where Rit is the return of stocks, Rmt the return of market index, ai theintercept term, bi the systematic risk of stocks and eit the error term.

The market model was selected as an expected return model and the OLS(ordinary least squares) regression was used in regression of the stock returnover 3 years of the estimation period against the return on the valuedweighted SET index for the corresponding calendar months. The SET indexis calculated from all stocks listed on the SET and is a market capitalisationweighted index that was used as the market index. The regression yielded theintercept term and a measure of systematic risk that is used to calculatean abnormal return, or a residual. In each event related month for eachsample stock. Month 13 (or 0) was determined as the event month and wecalculated 25 abnormal returns on each stock over the period around thetakeover announcements, from month 1 (�12) through to month 25 (+12).This interval is the event window for the bid period investigation of thisstudy. The impact of the event on stock returns was examined through anumber of stocks that were affected by the takeover announcements at the


event time. The abnormal returns (ARs) were averaged as:

AARt ¼1

n

Xn

i¼1

�it

where n is the number of stocks (firms).

3.4. Cumulative Abnormal Return Method

The accumulated effect of the event was examined using the CAARmeasure. The values of the AARs were continuously cumulated for everymonth from T1 (month 1 or �12) to T2 (month 25 or +12) as

CAAR ¼XT2

t¼T1

AARt

The CAARs plot for the sample stocks will tend to show patterns that canbe summarised (as shown in Fig. 1). Also, this pattern reasonably suggestswhether a hypothesis test for abnormal performances in month ‘‘0,’’ ratherthan for the entire period, is appropriate.

3.5. Buy-and-Hold Abnormal Return (BHAR) Method

In addition to the CAR approach, which could be regarded as beingdescriptive in nature,8 the BHAR approach, which is accompanied by amore feasible investment strategy, was also used. A stock’s BHAR wasdefined9 as the product of one plus each month’s abnormal return,minus one. To obtain a holding-period BHAR (BHARiT), the abnormalreturns were calculated as

ARit ¼ Rit � ai � biRmt

BHARiT ¼YT�1t¼0

½1þARit� � 1

where t=0 is the event month or the beginning period and T�1 is the periodof investment (in months).

Abnormal performance (BHARpT) was defined as the cross-sectionalaverage of the BHAR of the number of stocks (n). That is the abnormal


return (BHARiT) was averaged as

BHARpT ¼1

n

Xn

i¼1

BHARiT

Obviously, under either the CAR or the BHAR method, the abnormalreturns are calculated the same as the returns to a trading rule. While theCAR method uses the sum of each month’s AARs as the abnormal

CAARs to Successful Target Firms

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

1 3 5 7 9 11 13 15 17 19 21 23 25

Event Time

Val

ues

of C

AA

Rs

market model

zero-one model

Cumulative Average Abnormal Returns Estimated from the Market and Market-Adjusted (Zero-one)Models to Successful Target Firms

CAARs to Successful Bidding Firms

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1 3 5 7 9 11 13 15 17 19 21 23 25

Event Time

Val

ues

of C

AA

Rs

market modelzero-one model

Cumulative Average Abnormal Returns Estimated from the Market and Market-Adjusted (Zero-one)Models to Successful Bidding Firms

Fig. 1. CAARs Estimated from the Market and Market-Adjusted (Zero-One)

Models Applied to Successful Target and Bidding Firms.


performance measure, the BHAR method first compounds each stock’sabnormal returns and then uses the average or mean compounded abnormalreturns as the performance measure. This implies that the CAR and the BHARmethods can be employed with the period around the takeover announcementsand post-event or after the announcement of the takeover period.

3.6. Market-Adjusted Model (Zero-One Model)

The market-adjusted model was another10 expected return model used forthis study.

Rit ¼ biRmt þ �it

where Rit is the return of stocks, Rmt the return of market index, bI thesystematic risk of stocks and eit the error term.

All the calculation procedures are the same as those applied with themarket model as previously described. Also, the CAR and BHAR methodswere used with the market-adjusted model, respectively.

3.7. Statistical Tests of Abnormal Returns

To test the null hypothesis that the mean cumulative or BHAR is equal tozero for a sample of n firms, we employed three parametric test statistics.

3.7.1. Standardised-Residual Test

The standardised residual=the event-period residual scaled by the standarddeviation of the estimation-period residuals.

The test statistic is the sum of the standardised residuals divided by

(approximately) the square root of the number of sample firms (The actual

denominator is

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPNi¼1ðTi � 2Þ=ðTi � 4Þ

q, where Ti is the number of days

(months) in security i’s estimation period and N is the number of firms in thesample. If for most firms there are a large number of days (months) in theestimation period,

PNi¼1ðTi � 2Þ=ðTi � 4Þ � N.

t ¼PNi¼1

SRiEffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPNi¼1

ðTi � 2Þ=ðTi � 4Þ

s


or

t ¼PNi¼1

SRiEffiffiffiffiffiNp

where SRiE is the standardised residual, Ti the number of days (months) insecurity i’s estimation period and N the number of firms in the sample.

3.7.2. Standardised Cross-Sectional Test

The test statistic is the average event-period standardised residual divided byits contemporaneous cross-sectional standard error.

t ¼ 1=NPNi¼1

SRiEffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1=NðN � 1Þ

PNi¼1

ðSRiE �PNi¼1

SRiE=NÞ2

s

3.7.3. Conventional t-Tests

The test statistic is the AAR divided by its cross-sectional standard error.

tCAR ¼ CARiT

ðsðCARiT Þ=ffiffiffinpÞ

tBHAR ¼ BHARiT

ðsðBHARiT Þ=ffiffiffinpÞ

where CARiT and BHARiT are the sample averages and s(CARiT) ands(BHARiT) are the cross-sectional sample standard deviations of abnormalreturns for the sample of n firms.

4. RESULTS

The following sections present the results of the market and market-adjusted(zero-one) model analyses for the bid period or (�12, +12) for successfultarget and bidding firms. The results are shown and explained in terms ofthe performances of the monthly AARs, due to their correlations with thesignificance tests for the total standardised abnormal returns (TSRs) and the


average event-period standardised abnormal returns (ASRs), CAARs andaverage buy-and-hold abnormal returns (ABHARs). The main issues are thesize and signs of these abnormal returns and whether or not they aresignificantly different from zero.

4.1. Successful Target Firms

Tables 1 and 2 show that the findings estimated from the market and zero-one models are in line with each other. In month 0, the monthly AARs andCAARs are about 13.40% when estimated from the zero-one model and14.40% when estimated from the market model; and approximately 31.10%as estimated from the market model and 32% as estimated from the zero-one model, respectively. In addition, the successful target firm’s share-holders earn significantly positive monthly AARs and CAARs immediatelyaround the takeover announcement month, showing the positive CAARsover the period (�2, +2) and (�1, +1) at about 32 and 30.40%; and 33.10and 27.40% when estimated from the market and zero-one models,respectively.

Before month 0, there is evidence indicating that the market anticipatesthe takeover news as potentially being good news at least 2 months prior tothe takeover announcement. The CAARs over the period (�2, �1) arepositive at 13 and 14.50% as estimated from the market and zero-onemodels consecutively. The largest monthly AARs occur in month �1 whichare significantly positive at 10.10 and 10.60%, resulting in the positiveCAARs over the period (�12, �1) of 18.70 and 16.70%, as estimated fromthe zero-one and market models, respectively.

After month 0, the CAARs over the period (+1, +12) and (�12, +12) arepositive at 7.80 and 14.40%; and 38.80 and 46.40% when estimated fromthe market and zero-one models, respectively. It can be argued that theshareholders show downward-biased estimates of the likely value increasesattached to takeover announcements, as suggested in Akbulut andMatsusaka (2003). For the purpose of further comparison and analysis,and to strengthen the results, Table 3 presents that post the announcementmonth, the ABHARs over the same time period (�12, +12) are positive at148.20%11 and 36.70% consecutively, the ATSRs and AASRs are alsopositive, but are insignificant. However, the results are supportive eventhough they are not equal in magnitude. Finally, the results are consistentwith past studies and it can be concluded that the successful target firm’sshareholders realise substantially positive abnormal returns in each time


Table 1. Abnormal Returns Estimated from the Market and Market-Adjusted Models to Successful Target and Bidding Firms.

Event

Month

Successful Target Firms (50 Firms) Successful Bidding Firms (39 Firms)

Market Model Market-Adjusted

Model

Market Model Market-Adjusted

Model

AARs CAARS AARs CAARS AARs CAARs AARs CAARs

�12 0.024 0.024 0.030 0.030 0.029 0.029 0.052 0.052

�11 0.011 0.035 0.021 0.051 0.046 0.075 0.060 0.112

�10 0.003 0.038 0.007 0.058 �0.051 0.024 �0.022 0.090

�9 �0.026 0.012 �0.016 0.042 �0.025 �0.001 0.002 0.092

�8 �0.033 �0.021 �0.022 0.026 �0.020 �0.021 0.002 0.094

�7 0.010 �0.010 �0.007 0.019 �0.006 �0.027 0.028 0.122

�6 0.008 �0.002 0.000 0.012 0.004 �0.023 0.005 0.127

�5 0.028 0.026 0.009 0.021 �0.028 �0.051 �0.028 0.099

�4 0.010 0.036 0.020 0.041 �0.001 �0.052 0.027 0.126

�3 0.001 0.037 0.001 0.042 0.061 0.009 0.065 0.191

�2 0.024 0.061 0.044 0.086 0.022 0.031 0.037 0.228

�1 0.106 0.167 0.101 0.187 0.009 0.040 0.041 0.269

0 0.144 0.311 0.134 0.321 �0.018 0.022 �0.008 0.262

+1 0.054 0.365 0.039 0.360 �0.025 �0.004 �0.015 0.247

+2 �0.008 0.357 0.013 0.373 �0.035 �0.039 �0.017 0.230

+3 0.027 0.384 0.023 0.396 �0.025 �0.064 �0.020 0.210

+4 �0.01 0.373 0.011 0.407 0.016 �0.048 0.022 0.232

+5 �0.005 0.368 0.001 0.408 �0.036 �0.084 �0.034 0.198

+6 0.026 0.394 0.051 0.459 �0.059 �0.143 �0.053 0.146

+7 �0.036 0.358 �0.050 0.409 �0.022 �0.165 �0.005 0.140

+8 0.008 0.366 0.021 0.430 �0.041 �0.206 �0.023 0.117

+9 0.049 0.415 0.047 0.477 �0.012 �0.218 0.013 0.130

+10 0.011 0.426 0.016 0.493 0.004 �0.214 0.025 0.155

+11 �0.030 0.396 �0.020 0.473 0.005 �0.209 0.026 0.181

+12 �0.008 0.388 �0.009 0.464 �0.031 �0.240 �0.005 0.176

Note: This table presents the monthly average abnormal returns (AARs) and the cumulative

average abnormal returns (CAARs) to successful target and bidding firms for tender offers

occurring from 1992 to 2002. The measurement of the takeover announcement effects on the

firms or the realised returns for the successful target and bidding firms’ shareholders for the bid

period (�12, +12) were measured by the market and market-adjusted models. The AARs are

monthly abnormal returns for the successful target and bidding firms’ shareholders from

12 months before the event month until 12 months after the event month were estimated then,

cross-sectional averages in each month were calculated over the number of the firms. The

CAARs are the monthly AARs which are accumulated from the first month of the investigation

period until the last month of the period. The sample sizes (N) for the successful target and

bidding firms are presented in the parentheses.


Table 2. The Sum of Standardised Residuals (TSRs) and Average Event-Period Standardised Residuals(ASRs) Estimated from the Market and the Market-Adjusted Models to Successful Target and Bidding Firms.

Event Months Successful Target Firms (50 Firms) Successful Bidding Firms (39 Firms)

Market Model Market-Adjusted Model Market Model Market-Adjusted Model

TSRs ASRs TSRs ASRs TSRs ASRs TSRs ASRs

�12 17.307 0.346 16.638 0.333 16.711 0.428 22.855 0.586

(2.37)� (0.081) (2.28)� (0.93) (2.60)� (1.42) (3.55)� (2.03)�

�11 4.866 0.097 2.485 0.050 131.321 3.367 92.435 2.37

(0.067) (0.42) (0.34) (0.25) (20.40)�� ((1.43) (14.36)�� (1.44)

�10 7.333 0.147 �5.581 �0.112 �15.302 �0.392 �3.908 �0.100

(1.1) (0.58) (0.34) (�0.90) (�2.38)�� (�1.95) (�0.61) (�0.48)

�9 0.816 0.016 �1.487 �0.030 6.120 0.157 12.260 0.314

(0.11) (0.08) (�0.20) (�0.14) (0.95) (0.28) (1.90) (0.58)

�8 �9.673 �0.193 �8.418 �0.168 �13.203 �0.339 �5.237 �0.134

(�1.33) (�1.25) (�1.15) (�1.16) (�2.05)� (�0.78) (�0.81) (�0.29)

�7 4.411 0.088 �12.144 �0.243 �7.011 �0.180 6.669 0.171

(0.61) (0.36) (�1.67) (�1.09) (�1.09) (�0.59) (1.04) (0.57)

�6 2.212 0.044 �5.545 �0.111 2.870 0.074 0.227 0.006

(0.30) (0.22) (�0.76) (�0.48) (0.45) (0.29) (0.04) (0.02)

�5 11.675 0.234 2.680 0.054 �9.640 �0.247 �11.51 �0.295

(1.60) (0.85) (0.37) (0.20) (�1.50) (�1.44) (�1.79) (�1.67)

�4 3.236 0.065 3.802 0.076 4.751 0.122 14.710 0.377

(0.44) (0.41) (0.52) (0.47) (0.74) (0.51) (2.29)� (1.35)

�3 3.750 0.075 3.133 0.063 73.932 1.896 59.556 1.527

(0.51) (0.32) (0.43) (0.28) ((11.49)�� (1.72) (9.25)�� (1.54)

�2 19.694 0.394 26.142 0.523 27.981 0.717 17.035 0.437

(2.70)�� (1.36) (3.59)�� (1.90) (4.35)�� (1.22) (2.65)� (0.84)

�1 46.681 0.974 39.016 0.780 11.952 0.306 21.745 0.558

(6.68)�� (2.92)�� (5.35)�� (2.57)� (1.86) (0.74) (3.38)�� (1.23)

0 80.316 1.606 68.464 1.369 �10.101 �0.259 �7.868 �0.202

(11.02)�� (3.07)�� (9.39)�� (2.71)�� (�1.57) (�0.99) (�1.22) (�0.86)

+1 39.014 0.780 26.494 0.530 �3.724 �0.095 �0.451 �0.012

(5.35)�� (2.39)� (3.63)�� (1.83) (�0.58) (�0.54) (�0.07) (�0.06)

+2 �38.861 �0.737 �7.864 �0.157 �8.094 �0.208 �3.352 �0.086

(�5.06)�� (�0.98) (�1.08) (�0.38) (�1.26) (�0.94) (�0.52) (�0.52)

DAVID

E.ALLEN

AND

AMPORN

SOONGSWANG

362

+3 12.737 0.255 �1.083 �0.022 �7.871 �0.202 �10.141 �0.260

(1.75) (0.81) (�0.15) (�0.06) (�1.22) (�1.04) (�1.58) (�1.57)

+4 8.962 0.179 16.367 0.327 55.432 1.421 39.222 1.007

(1.23) (0.46) (2.25)� (0.71) (8.61)�� (1.08) (6.09)�� (0.72)

+5 �14.797 �0.296 �9.358 �0.187 �22.449 �0.576 �23.179 �0.595

(�2.03)� (�1.08) (�1.28) (�0.73) (�3.49)�� (�2.45)� (�3.60)�� (�2.14)�

+6 18.537 0.371 21.402 0.428 �28.403 �0.728 �28.634 �0.734

(2.54)� (1.58) (2.94)�� (2.01)� (�4.41)�� (�3.90)�� (�4.45)�� (�3.14)��

+7 �14.435 �0.289 �24.587 �0.492 �3.712 �0.095 �0.045 �0.001

(�1.98) (�1.03) (�3.37)�� (�1.52) (�0.58) (�0.52) (�0.01) (�0.01)

+8 8.678 0.174 �1.198 �0.024 �17.244 �0.442 �13.251 �0.340

(1.19) (0.86) (�0.16) (�0.05) (�2.68)� (�1.90) (�2.06)� (�1.27)

+9 43.101 0.682 30.888 0.618 2.291 0.050 10.738 0.275

(4.68)�� (1.60) (4.24)�� (1.36) (0.36) (0.29) (1.67) (1.23)

+10 �0.479 �0.010 �0.468 �0.009 8.840 0.227 12.455 0.319

(�0.07) (�0.02) (�0.06) (�0.02) (1.37) (0.72) (1.93) (1.09)

+11 �1.782 �0.036 �3.197 �0.078 29.197 0.749 28.263 0.725

(�0.24) (�0.15) (�0.54) (�0.31) (4.54)�� (0.79) (4.39)�� (0.87)

+12 2.468 0.049 2.221 0.044 �17.386 �0.446 �6.310 �0.162

(0.34) (0.29) (0.30) (0.21) (�2.70)� (�1.63) (�0.98) (�0.69)

Note: This table presents the cross-sectional total and average monthly standardised abnormal returns (residuals) for the bid period

(�12, +12) for tender offers occurring from 1992 to 2002. Specifically, to strengthen the results of the successful target and bidding firms’

performances, the realised returns for the firms’ shareholders for the bid period (�12, +12) were estimated from the market and market-

adjusted models. The monthly abnormal returns for the successful target and bidding firm’s shareholders from 12 months before the event

month until 12 months after the event month were calculated. Then, the monthly abnormal returns were standardised and cross-sectionally

summed and averaged to form the monthly total or the sum of the standardised residuals (TSRs) and the average event-period standardised

residuals (ASRs), respectively. The results show the monthly TSRs and ASRs for the successful target and bidding firms’ shareholders. To test

the significance of the monthly abnormal returns, the standardised-residual and standardised cross-sectional tests were applied. The t-statistics

were calculated by means of the standardised-residual test and the standardised cross-sectional test, respectively. The t-statistics are the sum of

the standardised residuals divided by (approximately) the square root of the number of sample firms, and the average event-period

standardised residual divided by its contemporaneous cross-sectional standard error, respectively. The standardised residual equals the event-

period residual divided by the standard deviation of the estimation-period residuals, adjusted to reflect the forecast error. The formulas are as

follows: t ¼PN

i¼1SRiE=ffiffiffiffiffiNp

; t ¼ 1=NPN

i¼1SRiE=ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1=NðN � 1Þ

p PNi¼1ðSRiE �

PNi¼1SRiE=NÞ2. The sample sizes (N) for the successful target and

bidding firms are presented in the parentheses, 36 and 25 months were selected for the estimation-period and event-window consecutively. The

test statistics are shown in the parentheses below the values of the TSRs and ASRs.�Significant at 5% level.��Significant at 1% level.

Ta

keo

versa

nd

Sh

areh

old

erV

alu

eC

reatio

n363

Table 3. Summary of the Results Estimated from the Market and Market-Adjusted Models for SuccessfulTarget and Bidding Firms (Bid Period) Investigations.

Sample Market Model (12, +12) Market-Adjusted Model (�12, +12)

CAARs

(�12, 0)

CAARs ABHARs ATSRs AASRs CAARs

(�12, 0)

CAARs ABHARs ATSRs AASRs

Successful target firms

(50 firms)

0.311 0.388 1.482a 10.031 0.201 0.320 0.464 0.367 7.123 0.142

(NA) (1.96) (1.39) (1.38) (0.59) (NA) (3.52)�� (1.80) (0.98) (0.34)

Successful bidding

firms (39 firms)

0.021 �0.240 0.012 8.290 0.213 0.262 0.176 0.145 8.971 0.230

(NA) (1.31) (0.08) (1.29) (�0.33) (NA) (1.50) (1.29) (1.39)

Note: ABHARs, average buy-and-hold abnormal returns; ATSRs, the means of total or the sum of standardised residuals; AASRs, the means

of the average event-period standardised residuals. The test statistics are provided in the parentheses below the values of the abnormal returns.

According to the conventional t tests, the results of the significance tests are the tests for the CAARs and ABHARs over the period (�12,

+12) for the bid period investigation.aWhen excluded Q: UOXT which has the remarkably substantial stock price returns in the sample, the ABHARs are positive at 44.70%

(t=1.94).�Significant at 5% level.��Significant at 1% level.

DAVID

E.ALLEN

AND

AMPORN

SOONGSWANG

364

period, before the announcement, with most of the returns realised in month�1, during the announcement and post the announcement month.

4.2. Successful Bidding Firms

Tables 1 and 2 present that the results estimated from the market and zero-one models are mostly consistent, especially in terms of direction. In month 0,the monthly AARs for the successful bidding firm’s shareholders areinsignificantly negative at �0.80 and �1.80% when estimated from the zero-one and market models, respectively. The CAARs over the period (�12, 0),starting 12 months before and including the event month, are positive at2.10 and 26.20% as estimated from the market and zero-one modelsconsecutively. Therefore, the effect of takeover announcement on the wealthof the successful bidding firm’s shareholders is positive.

Prior to month 0, the takeover news apparently leaks into the market4 months early, when estimated from the zero-one model, and 3 months,when estimated from the market model, at least, before the takeoverannouncement month. This leads to positive CAARs of 17.0 and 9.10% forthe successful bidding firm’s shareholders, respectively. Meanwhile, theCAARs over the period (�12, �1) are positive at 3.90% as estimated fromthe market model, compared with the 26.90% as estimated from the zero-one model. It is suggested that the successful bidding firms perform betterthan expected prospects for making takeovers, as the explanations by thestudy of Firth (1980).

After month 0, the CAARs over the period (+1,+12) and (�12, +12)are negative at �8.60 and �26.10%; and positive at 17.60% and negative at�24% when estimated from the zero-one and market models, respectively.This is partly supported by the ABHARs over the period (�12,+12) whichare positive at 14.50 and 1.20% consecutively. Also, the ATSRs and AASRsare positive (See ABHARs in Fig. 2).

The results are consistent with past studies including a more recent studyby Black et al. (2001) suggesting that the successful bidding firm’sshareholders gain significant positive abnormal returns of 1.50%, andsimilar to Anuchitworawong’s (2001) Thai study. We conclude that priorto the announcement and during the announcement month, a successfultakeover results in positive abnormal returns meanwhile, post theannouncement month, this leads to both positive and negative abnormalreturns for the bidding firm’s shareholders.


5. CONCLUSION

Our main findings suggest that in the takeover announcement month, asuccessful takeover enhances the wealth of the target firm’s shareholders, onaverage to an extent of 31 and 32% when estimated from the market andzero-one models consecutively. This is also displayed in positive CAARs ofabout 30–32% and 27–33% instantly around the announcement month forthe target firm’s shareholders. The market appears to anticipate takeover

ABHARs (market model) toSuccessful Targets

0

10

20

30

40

50

60

1 5 9 13 17 21 25 29 33 37 41 45 49

Sample Firms

ABHARs (zero-one)toSuccessful Targets

0

1

2

3

4

5

6

7

8

9

10

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49

Sample Firms

Average Buy-and-Hold Abnormal Returns Estimated from the Market and Market-Adjusted (Zero-one) Models to Successful Target Firms

ABHARs (market model) toSuccessful Bidders

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

1 4 7 10 13 16 19 22 25 28 31 34 37

Sample Firms

ABHARs (zero-one model) toSuccessful Bidders

0

0.5

1

1.5

2

2.5

3

3.5

1 4 7 10 13 16 19 22 25 28 31 34 37

Sample Firms

Average Buy-and-Hold Abnormal Returns Estimated from the Market and Market-Adjusted (Zero-one) Models to Successful Bidding Firms

Val

ues

of A

BH

AR

s

Val

ues

of A

BH

AR

s

Val

ues

of A

BH

AR

s

Val

ues

of A

BH

AR

s

Fig. 2. ABHARs Estimated from the Market and Market-Adjusted (Zero-One)

Models Applied to Successful Target and Bidding Firms.


news starting in month �2, before the event at least. Most of the positivemonthly AARs occur in month �1 which are significant and positive at 10and 11%, as estimated from the zero-one and market models, respectively.

At the same time, the results suggest that in the announcement month,a successful takeover leads to small and substantial positive CAARs forthe bidding firm’s shareholders. These are around 2%, as estimated from themarket model and 26%, when estimated from the zero-one model. Thebidder firms stock returns appear to positively anticipate potential takeovernews approximately some 4 and 3 months, depending on the metric, prior tothe announcement month.

For the purposes of comparisons and to strengthen the results for postannouncement month, or over the period (�12, +12), the BHAR approachwas also used. The results show that for the successful target firms’investigation, the ABHARs are positive at 37 and 148%, when estimatedfrom the zero-one and market models respectively, and the ATSRsand AASRs are also positive. In the cases of successful bidding firms,the findings suggest that their ABHARs are positive at 15 and 1%consecutively, and that the ATSRs and AASRs are positive. The resultsare therefore fairly consistent with one another. These findings are alsoconsistent with past studies and extend the evidence about the impact ofthese activities in the Thai market. We conclude that on average, the Thaitakeovers considered in this study were value creating events for ‘successful’target and bidding firms.

NOTES

1. Jensen and Ruback (1983) conclude a 29.10% gain for tender offers and Dattaet al. (1992) conclude a 21.81% gain for target firm’s shareholders.2. See for more details Bradley (1980), Bradley, Desai, and Kim (1982), Dodd and

Ruback (1977), and Jarrell and Bradley (1980).3. See in Asquith, Bruner, and Mullins (1983) and Dodd (1980).4. Once the SEC has approved the bid, the tender offer has to take place during

the tender offer period of at least of 25–45 trading days.5. ‘‘Companies under Rehabilitation Sector’’ or ‘‘REHABCO’’ is a sector

established by the SET in March 1998 to clearly separate listed companies requiringmajor restructuring due to substantial losses over time. Initially, 33 companies wereclassified under REHABCO.6. The control of a firm can increase continuously from none for those who own

no shares to complete for those who own 100% of the target’s shares or voting rightsoperations (see more in Bradley et al., 1988; also see Dodd & Ruback, 1977). Theevidence shows that the tender offer for target shares varies from 43.79 to 52.11%,


the purchased target shares vary from 25.60 to 28.99%, after shares are purchased,the holding of target shares is 62.83%, on average, while the biggest target shareholding is 99.91%.7. There are about 22.52% of the total tender offers are engaged with delisted

purposes and approximately 60.78% of the total delisted companies are caused bymandatory delisting.8. See for details Ikenberry, Lakonishok, and Vermaelen (1995).9. This study follows Kothari and Warner (1997).10. See for details Gondhalekar and Bhagwat (2000, p. 8) and Leemakdej

(1998, p. 3).11. When excluded Q: UOXT which has the remarkably substantial stock price

returns in the sample, the ABHARs are positive at 44.70% (t=1.94).

REFERENCES

Agrawal, A., & Jaffe, J. F. (2002). Do takeover targets under-perform? Evidence from operating

and stock returns. Social Science Research Network, http://www.ssrn.com

Akbulut, M. E., & Matsusaka, J. G. (2003). Fifty years of diversification announcements. Social

Science Research Network, http://www.ssrn.com

Anuchitworawong, C. (2001). Tender offer effects in the Thai market. Working Paper, Osaka

University Graduate School of Economics, Osaka.

Asquith, P., Bruner, R. F., & Mullins, J. D. W. (1983). The gains to bidding firms from merger.


Black, E. L., Carnes, T. A., & Jandik T. (2001). The long-term success of cross-border mergers

and acquisitions. Social Science Research Network, http://www.ssrn.com

Bradley, M. (1980). Inter-firm tender offers and the market for corporate control. Journal of

Business, 53, 345–376.

Bradley, M., Desai, A., & Kim, E. H. (1982). Specialized resources and competition in the market

for corporate control. Working Paper, University of Michigan.

Bradley, M., Desai, A., & Kim, E. H. (1988). Synergistic gains from corporate acquisitions and

their division between the stockholders of target and acquiring firms. Journal of Financial


Brown, P., & da Silva Rosa, R. (1998). Research method and the long-run performance of

acquiring firms. Australian Journal of Management, 23, 23–38.

Bruner, R. F. (2002). Does M&A pay? A survey of evidence for the decision-maker. Journal of

Applied Finance, 12, 48–68.

Campa, J. M., & Hernando, I. (2004). Shareholder value creation in European M&As.

European Financial Management, 10, 47–81.

Danbolt, J. (2002). Target company cross-border effects in acquisitions into the UK. University of

Glasgow: Department of Accounting and Finance, Working Paper.

Da Silva Rosa, R., Izan, H. Y., Steinbeck, A., & Walter, T. (2000). The method of payment

decision in Australian takeovers: An investigation of causes and effects. Australian

Journal of Management, 25, 67–94.

Datta, D. K., Pinches, G. E., & Narayanan, V. K. (1992). Factors influencing wealth creation

from mergers and acquisitions: A meta-analysis. Strategic Management Journal, 13, 67–84.


Dodd, P. (1980). Merger proposals, management discretion and stockholder wealth. Journal of

Financial Economics, 8, 105–138.

Dodd, P., & Ruback, R. (1977). Tender offers and stockholder returns: An empirical analysis.


Dumontier, P., & Petitt, B. S. P. (2002). Determinants of returns of acquiring firms around tender

offer announcements: Evidence from French control-oriented and parent-subsidiary offers.

Social Science Research Network, http://www.ssrn.com

Eckbo, B. E., & Thorburn, K. (2000). Gains to bidder firms revisited: Domestic and

foreign acquisitions in Canada. Journal of Financial and Quantitative Analysis, 35,

1–25.

Estrada, J., Kritzman, M., & Page, S. (2004). Countries versus industries in emerging markets:

A normative portfolio approach. Social Science Research Network, http://www.ssrn.com

Fernandes, N. (2005). What level of portfolio disaggregation in emerging market investments?

Journal of Portfolio Management, 31, 41–49.

Firth, M. (1980). Takeovers, shareholder returns, and the theory of the firm. The Quarterly

Journal of Economics, 94, 235–260.

Ghosh, A. (2001). Does operating performance really improve following corporate acquisitions.

Journal of Corporate Finance, 7, 151–178.

Ghosh, A. (2002). Increasing market share as a rational for corporate acquisitions. Social Science

Research Network, http://www.ssrn.com

Goergen, M., & Renneboog, L. (2004). Shareholder wealth effects in large European takeover

bids. European Financial Management Journal, 10, 9–45.

Gondhalekar, V., & Bhagwat, Y. (2000). Evidence on takeover characteristics and motives in the

acquisitions of NASDAQ targets following the stock market crash of 1987. Social Science


Herman, E., & Lowenstein, L. (1988). The efficiency effects of hostile takeovers. In:

J. Coffee, Jr., L. Lowenstein & S. Rose-Ackerman (Eds), Knight, raiders and targets.

Oxford: Oxford University Press.

Ikenberry, D., Lakonishok, J., & Vermaelen, T. (1995). Market under-reaction to open market

share repurchases. Journal of Financial Economics, 39, 181–208.

Jarrell, G., & Bradley, M. (1980). The economics effects of federal and state regulations of cash

tender offers. Journal of Law and Economics, 23, 371–407.

Jarrell, G. A., Brickley, J. A., & Netter, J. M. (1988). The market for corporate control:

The empirical evidence since 1980. Journal of Economic Perspectives, 2, 49–68.

Jensen, M. C. (2006). Takeovers: The controversy and the evidence. Social Science Research

Network, http://www.ssrn.com

Jensen, M. C., & Ruback, R. S. (1983). The market for corporate control. Journal of Financial

Economics, 11, 593–638.

Karceski, J., Ongena, S., & Smith, D. C. (2000). The impact of bank consolidation on commercial

borrower welfare. International Finance Discussion Paper no. 679.

Kothari, S. P., & Warner, J. B. (1997). Measuring long-horizon security price performance.


Leemakdej, A. (1998). Measuring abnormal return of infrequently traded stock in event study.

Case study of takeover targets in Stock Exchange of Thailand. Working Paper,

Thammasat University, Bangkok.

Lins, K., & Servaes, H. (2001). Is corporate diversification beneficial in emerging markets? Social

Science Research Network, http://www.ssrn.com


Loderer, C., & Martin, K. (1990). Corporate acquisitions by listed firms: The experience of

a comprehensive sample. Financial Management, 19, 17–33.

Loderer, C., & Martin, K. (1992). Post-acquisition performance of acquiring firms. Financial

Management, 21, 69–79.

Loughran, T., & Vijh, A. M. (1997). Do long-term shareholders benefit from corporate

acquisitions? Journal of Finance, 52, 1765–1790.

Maquieria, C., Megginson, W., & Nail, L. (1998). Wealth creation versus wealth redistributions

in pure stock-for-stock mergers. Journal of Financial Economics, 48, 3–33.

Morellec, E., & Zhdanov, A. (2004). The dynamics of mergers and acquisitions. Social Science


Parrino, J. D., & Harris, R. S. (1999). Takeovers, management replacement, and post-

acquisition operating performance: Some evidence from the 1980s. Journal of Applied

Corporate Finance, 11, 88–97.

Santos, M. B. D., Errunza, V., & Miller, D. (2003). Does corporate international diversification

destroy value? Evidence from cross-border mergers and acquisitions. Social Science


Varaboontweesuk, V. (2003). Tender offer announcement and stockholders’ wealth. An

independent study for the Degree of Master of Science Program in Finance, Thammasat

University, Bangkok.

Weston, J. F., & Copeland, T. E. (1992). Managerial Finance (9th ed.). New York: The Dryden

Press.


PART VI:

FUNDS MANAGEMENT

CHAPTER 18

HERD BEHAVIOUR OF CHINESE

MUTUAL FUNDS

Jean Jinghan Chen, Xinrong Xiao and Peng Cheng

ABSTRACT

We develop our theoretical framework from the viewpoint of the

information asymmetry and the agency theory that the Chinese mutual

funds exhibit herd behaviour, and provide empirical evidence by using

cross-sectional data of all the Chinese mutual funds between 1999 and

2003. We find that the Chinese mutual funds show overall herding, buy

herding and sell herding, and the degree of sell herding is higher than that

of buy herding. The degree of Chinese herding is higher than their US

counterpart from all the three perspectives. This may be largely due to the

institutional factors rather than those firm-specific factors that influence

the US mutual funds investment decision.

1. INTRODUCTION

Classic finance theories, whose central paradigm is the efficient marketshypothesis (EMH), assume that investors behave rationally and reactinstantaneously to all available information. However, recent behaviouralfinance research has shown that investors do not always act rationally andmay not consider all available information in their decision-making process.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00018-0

373

dx.doi.org/10.1016/S1569-3767(07)00018-0.3d

Investors often make systematic mistakes and traditional models of choiceunder uncertainty cannot explain some of their investment decisions.

Herd behaviour has become one of the major issues of behaviouralfinance. The study of herding started in the early 1970s on the institutions inthe US market (Friend, Marshall, & Crockett, 1970; Kraus & Stoll, 1972).Most of the studies so far have focused on institutional trading behaviourin developed capital markets, in particular, the US market. However, therelacks research on herding in emerging markets. Emerging markets oftenhave a different institutional infrastructure from that in developed markets,together with undeveloped capital market and less sophisticated investors.Further, the quality of corporate information disclosure is questionable.These factors may result in investors behaving differently from theexpectations by the traditional finance theories. This chapter attempts toexamine institutional investors’ herd behaviour in emerging markets withparticular reference to China. The Chinese mutual funds are the dominateforce of institutional investment in China, and, therefore, are the focus ofthis study.


2.1. Theoretical Arguments

2.1.1. Intentional Herding versus Unintentional Herding

Lakonishok, Shleifer, and Vishny (1992) refer herding as ‘managers’ buying(selling) simultaneously the same stocks as other managers buy (sell).’Bikhchandani and Sharma (2000) point out there are two kinds of herding:‘intentional herding’ and ‘unintentional herding.’ Intentional herding is‘real herding’, which is the result of the intent by investors to copy thebehaviour of other investors with no regard to whether the others makesmart investment decisions. Intentional herding is not necessary to beefficient, because one investor’s investment decision depends on that of theothers, rather than his reaction to the information available.

Unintentional herding, also called ‘spurious herding,’ occurs wheninvestors make decisions independently from each other and their tradingactions do not correlate. This type of herding is the result of identicalinformation sources available to investors, and investors take sameinvestment strategy and share same attitude towards risk. It is an efficientoutcome of market portfolio choices; therefore, we should not exaggerateunintentional herding.

JEAN JINGHAN CHEN ET AL.374

Amore precise and generally recognised definition of herding includes onlyintentional herding. However, empirically, separate intentional herding fromunintentional herding is difficult and even not possible (Bikhchandani &Sharma, 2000). This chapter follows prior empirical studies and focuses onintentional herding, however, cannot fully rule out unintentional herding.

2.1.2. Rational Herding versus Non-Rational Herding

Intentional herding may be either rational or irrational (DeLong, Shleifer,Summers, & Waldman, 1990; Froot, Scharfstein, & Stein, 1993). Thischapter regards herding to be more rational than irrational by arguing thatrational herding may be more realistic, particularly when fund managersherd. Compared with individual investors, fund managers are generallymore experienced and well educated. Most fund management companieshave a dedicated research team and invest significantly in informationcollection and analysis. They also have an internal control system to overseefund managers’ portfolio choices. Thus, fund managers’ herd behaviour ismore likely to be rational than irrational.

2.1.3. Models of Rational Herding

The literature has revealed that the presence of rational herding could beeither information-based or compensation-based behaviour.

Information-based herding is consistent with the prediction by theinformation asymmetry theory. Shiller and Pound (1989), Banerjee (1992)and Bikhchandani, Hirshleifer, and Welch (1992) regard that institutions tryto infer information about the quality of investments from each other’strades, thus herd as a result.

Suppose that an individual investor is to make his investment decisionunder the same environment as the others. If he processes privateinformation about the correct course of action, which would be the resultsof his effort and/or his special access to the ‘insiders’ of the company, hisinvestment decision would infer the private information. Informational-based herding could arise because herding would optimise individual fundmanager’s benefit rather than publicly available information.

Compensation-based (or reputation-based) herding is in line with theprediction of the agency theory. The difficulty in evaluating fund managers’performance and separating ‘luck’ from ‘skill’ creates the agency problemsbetween fund managers and fund sponsors (Scharfstein & Stein, 1990).Typically, fund sponsors evaluate the performance of fund managers in apeer group. To avoid falling behind the peer group, fund managers haveincentive to hold the same stocks. Maug and Naik (1996) conclude that

Herd Behaviour of Chinese Mutual Funds 375

incentive provision for portfolio managers is an important factor in theirasset allocation. They regard that herd behaviour of fund managers arisesfrom a fully rational response to their compensation contracts.

The compensation package of a risk-averse fund manager is dependenton the bench-marking portfolio return, or his ranking of performance inthe peer group. It is optimal for fund sponsors to design a performancecontract based on the benchmark portfolio return or the ranking systemwhen the moral hazard prevails. Fund managers would skew theirinvestment selections towards the top performing managers’ portfolios orthe benchmark portfolio. This performance-linked contract would inducefund managers work hard and thus reduce the moral hazard.

2.2. Empirical Evidence

The empirical study of herding so far has generally focused on statisticaltests to gauge whether there is a cluster of decisions made by institutionalinvestors.

Lakonishok et al. (henceforth LSV, 1992) study 769 tax-exempted equityfunds between 1985 and 1989 in the US market, and find weak evidence ofherding in small stock holdings only. They conclude that their results do notpreclude market-wide herding in the US market.

Grinblatt, Titman, and Wermers (1995) use the US data on portfoliochanges of 274 mutual funds between 1974 and 1984, to examine herdbehaviour among fund managers and the relationship of such behaviour tomomentum investment strategies and performance. They find that herdingamong the US mutual funds exists although it does not have strongstatistical significance. Wermers (1999) uses the US data on quarterly equityholdings of virtually all mutual funds between 1975 and 1994, and also findsevidence of herding.

In the European markets, Wylie (2005) studies herding of the UKmutual funds. This chapter examines the portfolio holdings of 268 UKequity mutual funds between 1986 and 1993 and finds evidence of herding.The degree of herding increases with the number of managers tradinga particular stock over a period, particular for small and large stockholdings. The degree of herding of the UK mutual funds is similar to thatof the US mutual funds. However, in contrast to their US counterpart,the UK mutual fund managers undertake contrarian trading in large stockholdings.


3. DEVELOPMENT OF THE CHINESE MUTUAL

FUND INDUSTRY

The Chinese fund industry emerged in 1998 and has been growing rapidlysince the 2000s. The Chinese mutual funds, managed by governmentauthorised fund management companies, have become the driving force ofthe Chinese fund industry and the institutional investment in the Chinesestock markets, representing over 70% of the total institutional investment inChina (HKEx, 2004). Table 1 illustrates the development of the Chinesemutual fund industry. The mutual funds have been the largest institutionalinvestment, measured by both book value and market capitalisation, in theChinese stock markets. The first 6 closed-end funds appeared on thedomestic stock exchanges in 1998, marking a milestone for the Chinese fundindustry. Open-end funds emerged in 2001, and soon have become a moreattractive choice over closed-end funds. By the end of 2003, there were atotal of 54 closed-end funds and 56 open-end funds, with a total net assetvalue of US$20.94 billion. The percentage of the funds’ total stockholdingsin the shares’ total market capitalisation was only 1.81 in 1998, but soonreached 12.90 in 2003.

However, the Chinese mutual fund industry is still in its developing stage.The mutual funds’ stockholdings are about only 13% of the total tradable

Table 1. Development of the Chinese Mutual Funds (1998–2003).

Year

1998 1999 2000 2001 2002 2003

Number of fund management companies 6 10 10 15 21 34

Number of open-end funds 0 0 0 3 17 56

Number of closed-end funds 6 22 34 48 54 54

The aggregate NAVa at year-end

(US$billion)

1.28 7.09 10.44 9.97 14.62 20.94

Total dividend distributed to fund holders

(US$billion)

0 0.03 0.61 0.78 0.34 0.09

% of the total tradable shares held by all

fundsb (%)

1.81% 7.00% 5.27% 5.59% 9.5% 12.90%

Source: CSRC (2004) and Sinofin and Tianxiang Database 1998–2004.aNAV, net asset value.bThe percentage of the total stockholdings all mutual funds in the shares’ total market

capitalisation.


shares and their trading volumes account for about 30% of the total tradingvolume of the tradable shares (www.csrc.gov.cn/CSRCSite/deptlistcom.htm). Moreover, the government still firmly controls the domestic listedcompanies and holds about 50% of the total tradable shares on average(HKEx, 2004; Chen, 2005). There is a lack of effective implementationof regulations, rules and laws governing the trading activities in theChinese stocks markets (Chen, 2005). Under the current property rightslegislation, minority shareholders have little legal protection for theirinvestment and their interests in the listed companies (Chen, 2005). Thegovernment as the controlling shareholder can easily expropriate themutual funds. The influence of the Chinese mutual funds in corporategovernance and market stabilisation is much weaker than that of their UScounterpart. They typically stay in the companies for a short period.Therefore, the Chinese mutual funds may behave differently from the USmutual funds. The institutional difference is an important motivationfor this study.

4. CHINESE MUTUAL FUNDS AND HERDING

HYPOTHESIS

The Chinese mutual funds are the minority investors and they do nothave a representative sitting on the board of directors of their investedcompany. They can only obtain public information, such as a company’sannual reports and IPO prospectus, for making their investment decision.Therefore, any private (insider’s) information on their proposed invest-ment becomes much valuable. The fund managers in a peer group morelikely understand those managers who take their investment action firstlyas they may have access to valuable private information on the companiesand follow those initial investments. Information-based herding wouldthus arise.

China operates a ranking mechanism to reward fund managers. Fundsponsors evaluate fund managers in a peer group. Fund managers’compensation packages are dependent on their ranking among the peercompetitors. Morning Star (China) (http://cn.morningstar.com) publishesa league table for all the Chinese funds and rates the funds according to theMorning Star Risk-Adjusted Return, which is based on a fund’s historicalperformance and a risk assessment of the fund’s portfolio stocks at priorperiods. China Securities News (http://www.cs.com.cn/tzjj) also publishes a


ranking for all the Chinese funds on a yearly basis according to the growthrate of a fund’s net assets. These two ranking systems largely determine afund manager’s compensation package and his job security. This com-pensation design results in the fund managers imitating the other’s portfolioin the peer group. Compensation-based herding would thus arise. Therefore,our first hypothesis to be tested is:

H1. There exists herd behaviour in the Chinese mutual funds.

Compared with the US market, the information asymmetry is much moresevere in the Chinese stock markets due to earnings management and weakcorporate governance in the Chinese listed firms (Cheng & Chen, 2007).Being the minority investors, the Chinese mutual funds find difficult to judgethe quality of financial reporting of the listed companies.

Unlike their US counterpart, most of the Chinese fund managers com-panies employ fund managers domestically. Due to the short history of theChinese fund industry, the Chinese fund managers are generally lessexperienced and have only basic professional portfolio managementskills. Senior fund managers train their juniors in house; therefore, thejuniors inevitably inherit their seniors’ investment strategy. Fund sponsorsare more likely to accept the investment strategies designed by senior fundmanagers.

The Chinese government does not permit the Chinese mutual fundsto invest in overseas markets, including Hong Kong. This restrictionlimits fund managers’ portfolio choices only to the domestic markets. Themajority of the Chinese listed companies in the domestic stock markets areformer SOEs.

The undeveloped Chinese bond market is another factor, which affectsthe Chinese fund managers’ portfolio choices. By 2003, there were only 10corporate bond issued with a total market value of US$10 billion, which wasonly 0.7% of China’s GDP, while the market value of the US bonds was30% of the US GDP in 2003 (http://www.bondmarkets.com).

Under such a circumstance, the Chinese mutual funds may exhibitstronger herd behaviour than the US mutual funds. Moreover, uninten-tional herding may also be much noticeable in China. Therefore, the secondhypothesis to be tested is:

H2. The Chinese mutual funds exhibit stronger herd behaviour than theUS mutual funds.


5. EMPIRICAL STUDY

5.1. Methodology

Measuring funds’ herding is to test for the degree of joint dependencebetween the trades of fund managers in a peer group. Lakonishok et al.(LSV, 1992) present the first and then the most popular measurement ofherding by focusing on trading activities conducted by a subset of fundmanagers over a period of time, whose behaviour is of interest. This methodrests on the following proposition: in the absence of herding, fundmanagers, who buy a proportion of stocks, have the same expectationacross all stocks in any particular period. If we observe significant cross-sectional variation in this proportion, we reject the null hypothesis of noherding. The majority of empirical studies on herding have employed theLSV measurement.

The LSV (1992) present their model as the follows. Let H(i) equal to themeasure of herding by fund managers buying or selling stock i during agiven quarter t. The measure is as:

HðiÞ ¼BðiÞ

BðiÞ þ SðiÞ� pðtÞ

�AFðiÞ (1)

where B(i) is the number of fund managers who increase their holdings inthe stock in the quarter (net buyers), S(i) the number of fund managers whodecrease their holdings (net sellers), p(t) the expected proportion of fundmanagers buying in that quarter t, relative to the number of trading thatstock in the same quarter. Essentially, the Eq. (1) is a simple ‘count’ of thenumber of funds buying a stock during a given quarter, as a proportion ofthe total number of funds trading that stock in the same quarter, minus theexpected proportion of buyers. AF(i) is an adjustment factor to allow forrandom variation around the expected proportion of ‘buys’ under the nullhypothesis of independent trading decisions (no herding) by fund managers.The adjustment factor assumes that BðiÞ=ðBðiÞ þ SðiÞÞ follows a binomialdistribution with a probability p(t) of success. The average of ‘buys’ during t

period is a proxy for p(t) (Lakonishok et al., 1992).Implicitly, the Eq. (1) defines and measures herding as the tendency of a

subgroup of fund managers who trade a given stock together in the samedirection more frequently than we would expect them trading randomly andindependently (Wermers, 1999). The average of H(i) gives the extent towhich funds herd. A positive and significant H(i) shows herding.


Wermers (1999) modifies the LSV herding measure by including a buyherding measure, BH(i), and a sell herding measure, SH(i), to segregatestocks by whether they have a higher (or lower) proportion of buyers (orsellers) than the average stock trading during the same period. The relationbetween the unconditional herding measure, H(i), and the two conditionalherding measures, BH(i) and SH(i), is described as follows:

BHðiÞ ¼ HðiÞ ifBðiÞ

BðiÞ þ SðiÞ4pðtÞ (2)

SHðiÞ ¼ HðiÞ ifBðiÞ

BðiÞ þ SðiÞopðtÞ (3)

The average of BH(i) and SH(i) is useful in analysing herding by fundmanagers into and out of stocks separately. If funds tend to sell stocks inherds more than they buy in herds, then the average of SH(i) will be largerthan BH(i) (Wermers, 1999).

We adopt the original LSV (1992) model in our study to identify theoverall herding of the Chinese fund managers, and Wermers’s model (1999)to identify the funds buy herding and sell herding.

The null hypothesis is no herding exists, therefore herding measuresH(i)r0, BH(i)r0, and SH(i)r0. If H(i)>0, BH(i)>0 or SH(i)>0, wereject the null hypothesis.

Let’s define a random variable:Y ¼ jðX=nÞ � pj, and n is the number ofsample in every quarter, then HðiÞ ¼ Y � EðY Þ.

Since, EðHðiÞÞ ¼ 0

VarðHðiÞÞ ¼ VarðY Þ ¼ EðY 2Þ � ðEðY ÞÞ2

¼ EX

n� p

� �2" #

�pð1� pÞ

n2

p

¼npð1� pÞ

n2E

X � npffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinpð1� pÞ

p !224 35� pð1� pÞ

n2

p

¼npð1� pÞ

n2�

pð1� pÞ

n2

p¼

pð1� pÞ

n1�

2

p

� �ð4Þ

Theoretically,

HðiÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðpð1� pÞ=nÞð1� ð2=pÞÞ

p � Nð0; 1Þ


When using H(i) and its standard deviations as estimates of the statistics,then


p � tðn� 1Þ

As long as H(i) is in the range of mean72 standard deviations, we canaccept the null hypothesis of no herding. Therefore, we calculate the statistic


pand also run a t-test on it. We calculate and test BH(i) and SH(i) in thesame way.

5.2. Data Collection

Since the start of trading in 1998, there were 110 Chinese mutual funds listedon the Chinese stock markets by 2003. Among the 110 funds, 81 were stockfunds, and the rest were index funds or bond funds. We focus on the tradingactivities of the 81 stock funds. According to the Chinese regulation ofinformation disclosure, funds need to disclose their top ten stockholdings ona quarterly basis. Therefore, we are able to study the mutual funds’ top-tenportfolio stocks they disclosed quarterly. We remove those stocks whosetrading involved less than four funds in order to conclude our study moreconvincingly. Therefore, we start our investigation from the first quarter of1999 to the fourth quarter of 2003 (20 quarters in total) and include 531portfolio stocks. We collect the information from the databases of Sinofin(http://www.sinofin.net/) and Tianxiang (http://www.txsec.com/).


Table 2 presents the sample descriptive statistics. The number of Chinesemutual funds increased dramatically from 16 in 1999 to 82 in 2003. The totalsize of the funds measured by the funds’ total assets grew up from US$5.48billion in 1999 to US$20 billion in 2003.

The total number of top-ten funds’ portfolio stocks increased steadilyfrom 2 in the first quarter of 1999 to 45 in the fourth quarter of 2003.However, the total and the mean market values of the top-ten stockholdings


Table 2. Summary Statistics of the Chinese Mutual Funds and theFunds’ Top-Ten Stockholdings (1999–2003).

Quarter/

Year

Total

Number

of Funds

Total Size of

Funds (by

Total

Assets,

US$billion)

Stock

Counts of

the Top-

Ten

Holdings

Total Market

Value of the

Top-Ten

Holdings

(US$billion)

Mean Value

of the

Top-Ten

Holdings

(t-Test)

(US$billion)

% of the

Top-Ten

Holdings in

the Total

Tradable

Shares

Mean

(t-Test)

Standard

Deviation

Q1/1999 16 5.48 2 0.34 0.012�� 0.130

(0.046)

Q2/1999 2 1.47 0.024�� 0.088

(0.028)

Q3/1999 4 1.55 0.102�� 0.496�

(0.451)

Q4/1999 5 2.66 0.038� 0.233��

(0.122)

Q1/2000 28 6.90 15 3.07 0.039�� 0.278��

(0.065)

Q2/2000 17 3.83 0.016�� 0.313��

(0.119)

Q3/2000 17 2.49 0.028� 0.266��

(0.137)

Q4/2000 16 2.71 0.029� 0.241��

(0.115)

Q1/2001 46 10.00 19 2.28 0.023�� 0.265��

(0.189)

Q2/2001 24 3.00 0.009� 0.197��

(0.115)

Q3/2001 25 1.26 0.012� 0.108��

(0.075)

Q4/2001 39 2.22 0.004�� 0.130��

(0.084)

Q1/2002 64 15.79 41 1.19 0.011�� 0.108��

(0.068)

Q2/2002 37 3.91 0.004�� 0.164��

(0.147)

Q3/2002 41 1.61 0.009� 0.131��

(0.082)

Q4/2002 43 4.04 0.006� 0.209��

(0.105)

Q1/2003 82 20 53 2.57 0.018� 0.133��

(0.088)


did not show a significant increase and they fluctuated during 1999 and2003. There seems to be a pattern since 2000. Both the total and the meanmarket values of the top-ten stockholdings were higher in the second and thefourth quarters than those in the first and the third quarters. It may indicatethat the fund managers regularly increase their top-ten stockholdings in thesecond and the fourth quarters of a year, and decrease their stockholdings inthe first and third quarters. The reasons for this pattern are not clear but itmay associate with the instability of the Chinese stock markets and theinstitutional environment.

The top-ten stockholdings accounted for an average between 10 and 30%of a stock’s total market capitalisation during 1999 and 2003. Thispercentage was between 20 and 30 before 2001 but dropped to less than20 thereafter, in terms of the mean value, with 1% statistical significance.It may indicate that the fund managers have modified their investmentstrategy towards holding a diversified portfolio since 2001. Several largecorporate scandals in China in 2001 may accelerate this change of fundmanagers’ portfolio choices when the fund managers realised the high risk ofputting eggs into one basket. For example, YinGuangXia group committedaccounting fraud of boosting their earnings by US$91.8 million in 2001,


Quarter/

Year

Total

Number

of Funds

Total Size of

Funds (by

Total

Assets,

US$billion)

Stock

Counts of

the Top-

Ten

Holdings

Total Market

Value of the

Top-Ten

Holdings

(US$billion)

Mean Value

of the

Top-Ten

Holdings

(t-Test)

(US$billion)

% of the

Top-Ten

Holdings in

the Total

Tradable

Shares

Mean

(t-Test)

Standard

Deviation

Q2/2003 44 5.86 0.014� 0.176��

(0.128)

Q3/2003 42 3.55 0.032�� 0.125��

(0.130)

Q4/2003 45 4.55 0.025� 0.188��

(0.070)

Source: Sinofin and Tianxiang Database 1998–2004.� Denote significance (two-tailed) at 0.10 level.�� Denote significance (two-tailed) at 0.05 level.�� Denote significance (two-tailed) at 0.01 level.


which accounted for more than 90% of its reported earnings. The company’sshare price collapsed immediately after the news of the accounting scandal.

5.4. Results

5.4.1. Herding of the Chinese Mutual Funds

We firstly calculate the sample funds’ three herding indicators between1999 and 2003. Table 3 presents the results. The mean overall herding is0.153 and it is statistically significant at 1%. We also test the median overallherding and the result supports the mean statistics although at a 5%significance level.

The mean sell herding is 0.184 (at a 1% significance level), which is higherthan the 0.130 buy herding (at a 1% significance level). The results of themedian sell and buy herding statistics are consistent with those of meanstatistics. It indicates the sell herding of the Chinese mutual funds isrelatively stronger than its buying herding. The fund managers are moredependent on each other when they sell their stocks.

We further calculate the quarterly herding indicators. Table 4 presentsthe mean quarterly herding indicators. The results are consistent with thefindings of the overall herding. We also calculate the median herdingindicators (not presented in the chapter), which show the same results.Therefore, we can confirm the Hypothesis 1.

5.4.2. Comparison of Herding between the Chinese and the US Mutual Funds

Table 5 presents the comparison of herd behaviour between the Chinese andthe US mutual funds. There is evidence of herding in both the Chinese andthe US mutual funds, but the degree of herding of the Chinese mutual fundsis significantly higher than that of the US funds in terms of all the three

Table 3. Measures for Overall Herding, Buy Herding and Sell Herdingof Chinese Mutual Funds (1999–2003).

Minimum Maximum Mean (t-Statistic) Median (z-Statistic)

H(i) �0.134 0.731 0.153�� 0.151��

BH(i) �0.115 0.369 0.130�� 0.144�

SH(i) �0.134 0.731 0.184�� 0.160�

� Denote significance (two-tailed) at 0.10 level.�� Denote significance (two-tailed) at 0.05 level.�� Denote significance (two-tailed) at 0.01 level.


Table 4. Mean Quarterly Herding Indicators of the Chinese MutualFunds (1999–2003).

Quarter/Year H(i) BH(i) SH(i)

Q1/1999 0.0207� 0.0795�� 0.0968

Q2/1999 �0.0715 N/A �0.0715

Q3/1999 0.0616�� 0.0105 0.2662��

Q4/1999 0.0792�� 0.0807�� 0.0776�

Q1/2000 0.0746�� 0.0942�� 0.0494

Q2/2000 0.1635�� 0.1328�� 0.2299��

Q3/2000 0.2071�� 0.1238�� 0.3618��

Q4/2000 0.2459�� 0.1235�� 0.3988��

Q1/2001 0.2745�� 0.2805� 0.2703�

Q2/2001 0.2393�� 0.1826�� 0.3115��

Q3/2001 0.2061�� 0.2550�� 0.1669

Q4/2001 0.1263�� 0.0949�� 0.1753��

Q1/2002 0.1311�� 0.1066�� 0.1772��

Q2/2002 0.1254�� 0.0947�� 0.1576��

Q3/2002 0.1506�� 0.1292�� 0.1948��

Q4/2002 0.1011�� 0.1188�� 0.0808

Q1/2003 0.1856�� 0.1508�� 0.2361��

Q2/2003 0.1126�� 0.0962�� 0.1322��

Q3/2003 0.1406�� 0.1460 0.1330��

Q4/2003 0.1555�� 0.1335�� 0.1676��

Source: Sinofin and Tianxiang Database 1998–2003.� Denote significance (two-tailed) at 0.10 level.�� Denote significance (two-tailed) at 0.05 level.�� Denote significance (two-tailed) at 0.01 level.

Table 5. Comparison of Herding between the Chinese and the USMutual Funds.

Nations Investigated

Period

Studies Number of

Funds

Herding

Measures

Conclusions

USA 1985–1989 Lakonishok

et al. (1992)

769 tax-free

stock funds

H(i)=0.027 Herd behaviour is

significant, but

not very high1974–1984 Grinblatt

et al. (1995)

274 mutual funds H(i)=0.025

BH(i)=0.019

SH(i)=0.031

1975–1994 Wermers

(1999)

2,424 funds H(i)=0.034

BH(i)=0.030

SH(i)=0.037

China 1999–2003 Our study 81 stock funds H(i)=0.153 Herd behaviour is

significant, and

relatively high

BH(i)=0.13

SH(i)=0.184


herding indicators. The overall level of herding of the US mutual funds isbetween 0.02 and 0.03, while it is around 0.15 of the Chinese mutual funds.The finding confirms the Hypothesis 2. The institutional environment andweak corporate governance discussed in Section 4 shape the behaviour ofthe Chinese mutual funds.

Furthermore, we find sell herding to be higher than buy herding in boththe Chinese and the US mutual funds. The finding reflects that the fundsshare an aversion to the stocks whose prices have recently droppedsignificantly (Falkenstein, 1996). However, as discussed in Section 4, theChinese mutual funds face higher market systematic risks, and thus wouldexaggerate their sell herding. Moreover, there are few financial derivativesavailable to investors in the Chinese financial markets. The fund managersfind difficult to protect their investment by hedging. When market is downand stock price drops, the fund managers have few choices but to sell theirholdings to avoid further loss. It would be easier and safer for the fundmanagers selling losers than adopting an alternative investment strategy.

It is worth to mention that although we make the comparison betweendifferent time horizons, but both the US and the Chinese market envi-ronments remain stable during those investigation periods. Therefore, we couldlimit the impact of the possible bias resulted from the different time horizon.

5.4.3. Characteristics of Portfolio Stocks of the Chinese Mutual Funds

We further examine the characteristics of Chinese mutual funds’ top-tenportfolio stocks and attempt to find out whether the stock portfolios held bythe Chinese mutual funds exhibit similar features as those held by their UScounterpart. We segregate the sample portfolio stocks into four quartilesby the degree of herding. Table 6 sorts the sample stocks by ascending meanvalue of H(i), BH(i) and SH(i) in Panel A, B and C, respectively. Thecharacteristics of these portfolio stocks examined by the literature on the USmutual funds include: (1) prior price (Prc), the stock price at the end of theprior quarter (lagged one quarter) before the herding quarter; (2) liquidity(Liq), the monthly trading volume divided by the total trading shares;(3) size (Size), the market value of a stock at quarter-end (US$ billion);(4) age (Age), the number of months between the trading month of the stockand the herding month; (5) prior return (Ret(–1)), the prior-quarter stockreturn (lagged one quarter) before the herding quarter; (6) price to earningratio (P/E), the ration of stock price to earning per share; and (7) price tobook value (P/B), the ratio of stock price to its book value. We conductt-test on the mean value of each characteristic during 1999 and 2003


(not presented in this chapter). All the test results are significance at 1%level. Therefore, we use mean value as a measure for each characteristic.

The results in Table 6 show that, for all the three panels, none of the stockfeatures investigated exhibit a pattern with the level of herding. We alsoperform statistical tests but none of them shows a statistical significance(therefore no significance shown in Table 6). The literature on the study ofthe US mutual funds shows that only the size and the prior-quarter returnrelate to the level of herding (Wermers, 1999; Grinblatt et al., 1995;Lakonishok, Shleifer, Thaler, & Vishny, 1991). Therefore, the followingdiscussion focuses on the size and the prior-quarter return only.

Size. Wermers (1999) finds evidence of a higher level of herding only in small-cap stocks of the US mutual funds. He further argues that fund managers may

Table 6. Relationship between the Degree of Herding and theCharacteristics of the Top-Ten Portfolio Stocks.

H(i) Prc Liq Size Age Ret(�1) P/E P/B

Panel A: Herding degree and the stock characteristics

Mean 0.15 17.18 0.18 3.33 64.09 0.04 39.32 2.12

1 (smaller) �0.02 16.48 0.20 3.30 62.35 0.05 32.02 2.28

2 0.07 17.38 0.18 3.91 73.75 0.08 44.55 1.29

3 0.18 18.57 0.21 3.29 58.93 0.02 45.31 3.33

4 (larger) 0.34 16.27 0.13 2.83 61.26 0.00 36.97 1.99

BH(i) Prc Liq Size Age Ret(�1) P/E P/B

Panel B: Buy herding degree and the stock characteristics

Mean 0.13 16.69 0.19 3.30 66.16 0.06 45.59 2.23

1 (smaller) �0.02 15.32 0.19 3.33 58.93 0.02 40.50 3.73

2 0.08 18.13 0.19 4.06 79.06 0.13 51.81 1.41

3 0.17 17.85 0.22 3.21 63.34 0.04 44.92 3.33

4 (larger) 0.25 15.46 0.16 2.62 63.31 0.04 45.25 0.49

SH(i) Prc Liq Size Age Ret(�1) P/E P/B

Panel C: Sell Herding degree and the stock characteristics

Mean 0.18 17.79 0.17 3.37 61.51 0.01 31.20 1.96

1 (smaller) �0.02 18.08 0.22 3.28 66.57 0.09 20.72 0.35

2 0.07 16.34 0.16 3.76 68.26 0.00 33.98 0.81

3 0.19 17.43 0.16 3.39 62.62 �0.03 44.40 3.59

4 (larger) 0.44 19.37 0.14 3.02 48.48 �0.01 28.22 3.81

Source: Sinofin and Tianxiang Database 1998–2003.


not fully believe the earnings information disclosed by their invested companiesand are more likely to disregard this information if the consensus opinionturns out to be different from the information disclosure. Moreover, smallercompanies are more likely to disclose low quality information and receive lesspublic confidence. This leads to information-based herding.

Our results do not show any significant relationship between the sizeof Chinese funds’ portfolio stocks and the degree of the Chinese fundmanagers’ herding; therefore, do not support Wermer’s finding. The Chinesemutual funds herd with no regard to the size of the portfolio stocks inmarket value. The results reflect the severe information asymmetry problemin the Chinese stock markets discussed in Section 4. The problem of lowquality of financial reporting prevails in both small and large companies inChina. A good reflection of this problem is the wide-spread phenomenon ofupward earnings management in all Chinese listed companies (Cheng &Chen, 2007). Therefore, information-based herding is more likely to appearwith no regard to the size of stocks.

Prior Returns. The literature on the US market shows that the US mutualfunds use positive-feedback strategy widely, which is to sell past losers andbuy past winners (Grinblatt et al., 1995). This strategy can either stabilise ordestabilise stock prices. Alternatively, funds may herd due to the ‘window-dressing’ strategy, which amounts to sell past losers (Lakonishok et al.,1991). Therefore, past return may be a possible reason for mutual funds’herding. Wermers (1999) provides empirical evidence on funds investment asthey execute positive-feedback strategy. He shows that the US mutual funds’buy herding is the strongest in high prior-quarter return stocks while sellherding is the strongest in low prior-quarter return stocks.

The results in Table 6 reveal that the Chinese mutual funds do not herd oneither side of past-return stocks. We do not find any particular relationshipbetween the magnitude of funds’ herding and the prior-quarter return of theportfolio stocks. This finding is not consistent with that in the US. It mayimply that the Chinese mutual funds find difficult to employ the positive-feedback strategy due to the high volatility of the Chinese stock markets andtherefore the firms’ earnings.

6. CONCLUSIONS

This study finds that the Chinese mutual funds exhibit overall herding, buyherding and sell herding. The degree of sell herding is higher than that of


buy herding. Compared with their US counterpart, the Chinese mutualfunds show a higher level of herding from all the three perspectives. Thismay be largely due to the institutional factors rather than the firm-specificfactors.

The study does not find those firm-specific factors examined by theliterature and have affected the US mutual funds’ herd behaviour to beparticularly influential to the Chinese mutual funds’ investment decision.Our findings do not show any relationship between the Chinese mutualfunds’ herd behaviour and particular momentum investment strategies.

We believe that this study is the pioneer study of the Chinese mutualfunds from the perspective of behaviour finance. Although the study isof preliminary and exploring nature, it is of much importance becausenot only has it found the herd behaviour of the Chinese mutual funds, butalso it has raised two important and interesting future research questions:‘what are the determinants of mutual funds investment decision in theChinese markets?,’ and ‘how would the Chinese stock market react tothe mutual funds herd behaviour?.’ These future research directions reflectthe limitation of this study.

From the perspective of company management, firstly, Chinesemutual fund sponsors should be aware of the herd behaviour of the fundmanagers, then work out an appropriate incentive package to restrain theiropportunism. Secondly, when foreign mutual funds enter the Chinesemarket, they should not use similar investment strategies, which they haveadopted in developed markets such as in the US. They need to be aware ofthe constraints of the Chinese institutional environment and alter theirinvestment strategies accordingly. From the perspective of finance theory,further study on how the herd behaviour of mutual funds influences thestock price would provide empirical evidence for the EMH.

REFERENCES

Banerjee, A. V. (1992). A simple model of herd behaviour. Quarterly Journal of Economics, 107,

797–817.

Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A theory of fads, fashion, custom and

cultural change as informational cascades. Journal of Political Economy, 100, 992–1026.

Bikhchandani, S., & Sharma, S. (2000). Herd behaviour in financial markets: A review.

International Monetary Fund Working Paper no. WP/00/48, 14–27.

Chen, J. J. (2005). Institutional environment and corporate governance. Advances in Financial



Cheng, P., & Chen, J. J. (2007). Expropriation, weak corporate governance and post-IPO

performance: Chinese evidence. Advances in Financial Economics, 12.

CSRC (Chinese Securities Regulation Commission). (2004). China’s Securities and Futures

Markets, http://www.csrc.gov.cn/en/homepage/index_en.jsp

DeLong, B., Shleifer, A., Summers, L., & Waldman, R. (1990). Noise trader risk in financial

markets. Journal of Political Economy, 98, 703–738.

Falkenstein, E. G. (1996). Preferences for stock characteristics as revealed by mutual fund

portfolio holdings. Journal of Finance, 51, 111–135.

Friend, I., Marshall, B., & Crockett, J. (1970). Mutual funds and other institutional investors.

New York: McGraw-Hill.

Froot, K. A., Scharfstein, D. S., & Stein, J. C. (1993). Risk management: Coordinating

corporate investment and financial policies. Journal of Finance, 48, 1629–1658.

Grinblatt, M., Titman, S., & Wermers, R. (1995). Momentum investment strategies, portfolio

performance, and herding: A study of mutual fund behaviour. American Economic

Review, 85, 1088–1105.

HKEx (Hong Kong Exchanges and Clearing Limited). (2004). Institutional Investors in

Mainland China. http://www.hkex.com.hk/publication/newsltr/2004-01-19-e.pdf

Kraus, A., & Stoll, R. H. (1972). Parallel trading by institutional investors. Journal of Financial

and Quantitative Analysis, 7, 2107–2138.

Lakonishok, J., Shleifer, A., Thaler, R., & Vishny, R. (1991). Window dressing by pension fund

managers. American Economic Review, 81, 227–231.

Lakonishok, J., Shleifer, A., & Vishny, R. (1992). The impact of institutional trading on stock

prices. Journal of Financial Economics, 32, 23–44.

Maug E. G., & Naik, N. Y. (1996). Herding and delegated portfolio management: The impact of

relative performance evaluation on asset allocation. London Business School Working

Paper no. 223.

Scharfstein, D. S., & Stein, J. (1990). Herd behaviour and investment. American Economic

Review, 80, 465–479.

Shiller, R. J., & Pound, J. (1989). Survey evidence on diffusion of interest and information

among investors. Journal of Economic Behaviour and Organization, 2, 47–66.

Wermers, R. (1999). Mutual fund herding and the impact on stock prices. Journal of Finance,

54, 581–622.

Wylie, S. (2005). Fund manager herding: A test of the accuracy of empirical results using UK

data. Journal of Business, 78, 381–403.


CHAPTER 19

PERFORMANCE PERSISTENCE

OF PENSION FUND MANAGERS:

EVIDENCE FROM HONG KONG

MANDATORY PROVIDENT FUNDS

Patrick Kuok-Kun Chu

ABSTRACT

This chapter examines the performance persistence evidences of pension

fund managers who managed the constituent equity funds included in

Hong Kong Mandatory Provident Fund (MPF) schemes over the period

2001–2004. Nonparametric two-way contingency table and parametric

OLS regression analysis are employed to evaluate performance per-

sistence. The evidence suggests that the raw returns, traditional Jensen

alphas, and conditional Jensen alphas in the previous year possess

predictive abilities. When the funds are classified into high-volatile and

low-volatile samples, the high-volatile funds are found to possess stronger

performance persistence. Neither hot-hand nor cold-hand phenomena are

found in the equity funds managed by same investment manager.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00019-2

393

dx.doi.org/10.1016/S1569-3767(07)00019-2.3d

1. INTRODUCTION

Many international organizations have considered the problem of an ageingpopulation and most have also proposed some form of policy. Accordingto the 1994 World Bank study ‘‘Averting the Old-Age Crisis: Policies toProtect the Old and Promote Growth,’’ governments should protect theold and they should promote economic growth. The Mandatory ProvidentFund (MPF) system was implemented in Hong Kong on December 1, 2000.The main purpose of the system is as an employment-based protectionsystem. The problem of an ageing population has existed in Hong Kong andsuch problem has been highlighted since 1980s. Statistics showed that peopleaged 65 and above accounted for 6.6% of the population in 1981. Theproportion has grown to 11.5% in 2003, and is expected to increase to 14%by 2016, and to 24% by 2031.1 This shows that the need for retirementprotection is increasing. Before the implementation of the MPF system,only one-third of the 3.4 million Hong Kong workforces had some form ofretirement protection. With the implementation of MPF system, 86% of theworkforce had retirement protection by the end of 2001, either throughMPF or other retirement schemes.

With the launch of the MPF in Hong Kong on December 1, 2000, allHong Kong employers and employees have developed an interest in mutualfunds. Meantime, the need for research on the measurement of performanceof MPFs becomes higher, as investors may be better informed of theinvestment choices. So far, there has been a substantial amount of studiesdone on Hong Kong security markets and futures markets, however, theacademic research on the mutual funds industry in Hong Kong, and HongKong MPF, in particular, is just beginning to emerge.

The predictability of the performance of securities including recognizedfunds has long been of interest to academics. Although most of the MPFfunds have negative returns in the first 2 years of operations, activemanagers still try to outperform each other and the market. Historic alphasindicate the existence of past average abnormal performance. Of greatinterest is whether there is persistence in performance. For theorists andparticipants, understanding performance persistence is important. For thetheorists, the existence of performance persistence indicates the market isnot efficient. For the participants, especially most of whom have noinvestment experience and knowledge, the strategy of whether buying last-years winner is good or not will be the interest to them. As the MPFparticipants may change their fund managers, the usefulness of trackrecords is taken for granted by most participants.

PATRICK KUOK-KUN CHU394

Mutual fund performance persistence is substantially documented infinance literature but these previous studies found mixed evidences on short-term persistence in mutual fund performance in US. Early studies of mutualfund performance persistence have generally suggested that there is littleinformation in the performance track record. However, more recent studiesfind that when a shorter evaluation period is used, past performance doesprovide information about future performance of funds. They documentthat there exist funds exhibit short-term persistence in returns and suggestthat investors can earn abnormal returns by pursuing certain investmentstrategies to exploit this information. Compared with literature on mutualfunds in US and Europe, no research has been conducted on theperformance persistence of Hong Kong MPF so far. More precisely, thischapter aims at answering the following questions:

1. Do some Hong Kong MPF equity funds systematically outperform theirpeers? The usual methods employed to detect evidence of returnpersistence consist of two streams: parametric method that involvesusing regression model of current-period performance measures on last-period performance measures and nonparametric method that refers toconstructing contingency tables.

2. If the MPF constituent funds display risk-adjusted performancepersistence is examined. Indeed, one would like to know if somemanagers can consistently generate superior performance after account-ing for their systematic risk exposures. In order to address this issue, theJensen alpha measure is used as performance measure and theirpersistence will also be determined using the same parametric andnonparametric methodologies described above.

3. The relation between the fund volatility and performance persistence isexamined by separating the equity funds into two batches – high-volatileand low-volatile funds, the evidences of performance persistence are thenexamined separately in these two clusters of funds.

4. The performance persistence of the funds provided by each investmentmanager is examined. The rationale of performing this analysis is that thefunds under the same investment manager may share same slot ofsupports and resources, and are under the same supervision.

The remainder of the chapter is organized as follow. Section 2 summarizesthe literatures. Section 3 outlines the nonparametric and parametric researchmethodologies used to determine the return persistence; the methodologyto investigate the evidences of risk-adjusted return persistence; that todetermine the relation between the volatility and performance persistence.

Performance Persistence of Pension Fund Managers 395

Section 4 discusses the data set used in the study. Section 5 presents theempirical results of evidences of raw- and risk-adjusted return persistence bynonparametric approach and parametric approach; the result of evidencesof persistence in the rankings of the funds performances is also presented.Section 6 provides some further analysis of performance persistence such asthe relation between fund volatility and performance persistence and theevidence of performance persistence of constituent funds provided by sameinvestment manager. Section 7 provides a conclusion of this study.


Mutual fund performance persistence is well documented in financeliterature but these previous studies found mixed evidences on short-termpersistence in mutual fund performance in US. More evidences ofpersistence in mutual fund returns are found in the decade of 1990scompared with the prior two decades.

Goetzmann and Ibbotson (1994) use a two-way cross tabulation that isfresh in academics at that time, instead of cross-sectional regression, toinvestigate the persistence in monthly returns of 258 funds over the period1976–1988. Persistence analysis using two-way tables over successive 2-yearintervals shows evidence of persistence in both raw returns and Jensenmeasures in most of the years except that the raw returns showing reversalbetween 1980–1981 and 1982–1983. The authors query if the tests using rawreturns that are not adjusted for risk may document merely the differentialexpected returns between high-risk versus low-risk funds. The analyzes ofthe persistence of Jensen measures further document the evidence ofpersistence. Besides the two-way tables, some regression models are also setup to detect the magnitude of the 2-year alphas on the subsequent 2-yearalphas. The results are significant in four out of the five periods, and areextremely significant for the combined regression results. Besides finding theevidence of persistence in all sample funds, the authors separate the fundsinto high-variance funds and low-variance funds to detect if the survivorshipbias is exacerbated by different fund volatilities. The funds are categorizedas high-variance if the variances of the funds are above the median, whilemedian and below funds are categorized as low-variance funds. The resultshows that the phenomenon of persistence is stronger in high-variancefunds, indicating the survivorship is a possible source of bias in theperformance study.


Malkiel (1995) reports temporal differences in return persistence. Thedataset contains quarterly returns over the period 1971–1991. Using two-way cross tabulation, the author finds significant performance persistenceduring the 1970s. The evidence of persistence becomes weaker during the1980s. The percentage of winning funds tending to repeat their winningperformance reduces from 65.1 to 51.7%. The author also finds returnreversals for the years 1980 and 1987, which is consistent with Brown andGoetzmann (1995). Besides these 2 years, the author finds two additionalreversals in 1988 and 1990.

Brown and Goetzmann (1995) explore the persistence in annual fundperformance over the period 1976–1988 and also find return reversals for theyears 1980 and 1987. Following Brown, Goetzmann, Ibbotson, and Ross(1992) and Goetzmann and Ibbotson (1994) approaches, two-way tables areset up to test the performance persistence. Evidence of significantpersistences in seven or eight periods out of 12 years is found. Negativepersistence is found in 2 years, 1980 and 1987. They hypothesize the seculartrend in performance persistence is due to the difference between thesystematic risks across managers. Single-factor and three-factor alphameasures are employed to measure the excess returns. The results show thatthe R2 values of both models are higher than 0.9, which indicates usingsingle-factor or three-factor models to adjust for risk does not affect thepersistence patterns.

Cheng, Pi, and Wort (1999) document no significant evidence ofperformance persistence in mutual funds managed by Hong Kong fundhouses during the period 1992–1996. The authors take a differentprospective to explore the persistence in mutual fund performance. Theyexamine the performance of fund houses as a whole instead of individualfunds’ returns. This study contributes to the current literature on therelationship of common management strategies and supervision to fundhouse performance persistence. They find only 2 fund houses out of 32exhibit significant persistence, which contradict most pervious studies onAmerican mutual funds that found significant short-term persistence. Theauthors also explore the relationship between the persistence of fund houses’performance and economic significance by correlation analysis. They findthat there is no significant association between these two aspects and mayconclude that the investors may not earn significant excess returns frominvesting in hot hand houses. On the other hand, there is significant positiveassociation between the persistence of fund houses’ performance andperformance of individual funds provided by these fund houses. This showsthat the hot hand fund houses typically have more well performing funds,


which supports the view that some common management strategies andsupervision may be the underlying causes of short-term persistence.

Allen and Tan (1999) find similar result that the UK unit trusts exhibitperformance persistence. They document evidence of return persistence of131 UK managed fund over the period 1989 and 1995. They investigate thepersistence in performance measured by raw return and risk-adjustedreturns of individual unit trusts by employing two common parametric andnonparametric methods. The results show that long term (1- and 2-yearintervals) raw returns and alphas exhibit significant evidence of persistence.On the other hand, this evidence appears to reverse in the short-term (semi-annually and monthly). In addition, the relation between the volatility andthe persistence is studied by classifying the funds as high-variance and low-variance. The performance measured in alphas and raw returns still exhibitrepeat-winner patterns in two different classes of funds.

Christopherson, Ferson, and Glassman (1998) provide the first study onthe evidence of performance persistence of 273 US pension funds over theperiod 1979–1990 using conditional performance evaluation techniques,which were developed by Ferson and Schadt (1996). Their study documentsevidence of persistence in the performance measured by both unconditionalalphas and conditional alphas. Similar to the previous findings in mutualfund performance persistence, they find poor-performing funds tend to befollowed by low future returns, and that persistence is concentrated in thepoorly performing funds. Their study finds that persistence is concentratedin the poorly performing funds, which raises some puzzles left to beanswered. Why do the poorly performing managers survive? Is thereinefficiency in the market for pension manager services? Are the costs offiring poorly performing managers high enough to justify this persistence oflow returns? Do the poorly performing managers deliver valuable services tothe plan sponsors that offset their poor investment returns? What strategiesfor trading and trade execution that the persistently poor-performingmanagers use? The authors point out future research is needed, usingconditional models, to address these puzzles.

3. RESEARCH METHODOLOGY

3.1. Nonparametric Approach to Identify Performance Persistence

The first investigation of persistence uses the contingency table which isnamed ‘‘winner–winner, winner–loser’’ methodology applied by Goetzmann


and Ibbotson (1994), the persistence of performance measured by absolutereturns will be analyzed by constructing two-way tables showing perfor-mance over successive periods. The use of contingency table is referred to asnonparametric approach. As opposed to parametric approach, nonpara-metric approach is used to estimate percentiles of any continuousdistribution without the shape of the distribution being specifically definedby a formula and thus robust when the normality assumption does not hold.Following Goetzmann and Ibbotson, the fund is defined as a winner in thecurrent period if the raw return is above or equal to the median returns of allMPF equity funds over the stated holding period. In other words, thewinners (denoted by W) are distinguished from losers by ranking fundperformance and defining the top half of the list as winners and the bottomhalf as losers (denoted by L). Funds with returns equal to the median arealso called ‘‘winners’’. If a fund is in W for consecutive periods, it is definedas a winner–winner (WW). Thus, WW for 2002–2003 is the count of thewinners in 2002 that were also winners in 2003 if annual returns are beingused to evaluate the performance. If a fund remains in the bottom half of thereturns for two consecutive years, it is a loser–loser (LL). A fund that shiftsfrom W to L is a winner–loser (WL) and a fund that shifts from L to W isa loser–winner (LW). The frequencies inside are the numbers of funds thatare belong to one of four categories: (1) WtWtþ1, (2) LtLtþ1, (3) WtLtþ1, and(4) LtWtþ1. Funds in the first two categories are defined as persistent winner(loser) funds. The last two categories are defined as winner then loser, andloser then winner. The two-way contingency tables will be constructed asfollows:

Period (t+1)

Winner (W) Loser (L)

Period t Winner (W) WW WLLoser (L) LW LL

To the interest of the scheme participants, analyzing the annualperformance persistence is important to them as most of the schemeparticipants change and reallocate their fund portfolios inside their plansevery year, especially at the beginning of calendar year. The participantswho change their portfolios every year mostly check the previous annually


performance of the constituent funds in their portfolios and reallocate theirportfolios based on the previous performance.

The significance of evidence of performance persistence may beinvestigated by some statistical tests. The first one is the binomial test, ornamed as Malkiel z-test. This test detects if statistically there is evidenceshowing that winners (losers) in the period t have a significantly greater than50% chance of remaining winners (losers) in period t+1 exists. The teststatistic is computed as follows:

Z ¼X � npffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffinpð1� pÞ

p (1)

where X is the number of persistently winning (or losing) funds, n thenumber of funds in the sample, and p=1/2 which is the probability that awinning fund remaining in the winning category.

The other test that may investigate the statistical significance of theperformance persistence is the cross-product ratio (CPR) test. CPR isdefined as ðWW� LLÞ=ðWL� LWÞ, which captures the ratio of the fundsthat show performance persistence equals to one or not. The null hypothesisof no evidences of performance persistence is tested by hypothesizing CPRequals to one, in other words, each of the four categories denoted by WW,WL, LW, and LL is expected to have 25% of the total number of funds. Inlarge samples with independent observations, the standard error of thenatural log of the odds ratio is well approximated as slnðCPRÞ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1=WWÞ þ ð1=WLÞ þ ð1=LWÞ þ ð1=LLÞ

pin Christensen (1990). The test

statistic is the natural logarithm of odds ratio divided by its standard error,and is asymptotically normally distributed under the assumption ofindependence of the observations.

Z ¼lnðCPRÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ð1=WWÞ þ ð1=WLÞ þ ð1=LWÞ þ ð1=LLÞp (2)

The last test that may also investigate the significance of performancepersistence is the w2 independence test. Carpenter and Lynch (1999) find thatthe w2 independent test based on the number of winners and losers is wellspecified when they study the specification and power of various persistencetests. The rationale of the test is that because half of the funds are defined aswinners and losers respectively, if the evidence of persistence does not exist,the numbers in each bin should be equal or the actual distribution in eachbin should be 25% of the total number of funds. On the other hand, the


frequencies in the diagonal bins will be statistically significantly higher thanthe other two bins if performance persistence exists. The null hypothesis ofno evidence of performance persistence against the alternative of existenceof persistence is diagnosed by w2, which follows a w2 distribution with(R�1)� (C�1) degree of freedom in an R by C contingency table, so thedegree of freedom in the tests of this chapter is one. The test statistics isdefined as:

w2 ¼ðWW� E1Þ

2

E1þðWL� E2Þ

2

E2þðLW� E3Þ

2

E3þðLL� E4Þ

2

E4(3)

where E are known as expected values in the w2-test, and

E1 ¼ðWWþWLÞ � ðWWþ LWÞ

n ; E2 ¼ðWWþWLÞ � ðWLþ LLÞ

n

E3 ¼ðLWþ LLÞ � ðWWþ LWÞ

n ; E4 ¼ðLWþ LLÞ � ðWLþ LLÞ

n

The CPR- and w2-test usually lead to the same conclusions aboutperformance persistence. However, the latter has the disadvantage of notbeing able to find evidence of performance reversal since it is alwayspositive; while the former may detect evidence of performance reversal witha negative z-statistic.

3.2. Risk-Adjusted Return Persistence

Prior research showed that the evidence of return persistence is not affectedby the risk adjustment (Goetzmann & Ibbotson, 1994; Brown & Goetzmann,1995; Gruber, 1996). To test the hypothesis that the performance persistencepattern in our dataset is not influenced by the risk adjustment, this study usesnot only the raw returns but also the single-factor Jensen alpha and Fama-French three-factor Jensen alpha as measures of performance of equityfunds. The other rationale of adjusting risk is to document the differentialexpected returns between high-risk versus low-risk funds.

Likewise, the empirical tests outlined in the previous section – two-waycontingency table with Malkiel z-test on repeat winners, CPR-test, and w2

independence test; and OLS regression analysis of risk-adjusted perfor-mance in holding period on risk-adjusted performance in evaluation periodare used to test the persistence of MPF risk-adjusted returns (Jensen alphameasure and Fama-French three-factor alpha measure). The procedures to


find Jensen measure is outlined in Eq. (4). Since the Jensen measure has beenadjusted for risks and is a standard measurement of fund performance,persistence in Jensen measures may be considered as due to the existence ofconsistent stock selection skills.

Suppose Ri,t is the monthly return of the funds in the tth month, and Rm,t

is the monthly return on the mean-variance efficient market portfolio; theJensen measure refers to the intercept a in the regression model of return ofthe fund, i, in excess of the 1-month risk-free rate on the excess return on themarket portfolio as follow:

Ri;t � Rf ;t ¼ ai þ bðRm;t � Rf ;tÞ þ et (4)

If the CAPM is a correct model of equilibrium returns, the portfolio of afund should lie on the security market line and the value of Jensen alpha, ai

in Eq. (4), should be zero. Therefore, a significant positive Jensen alphaindicates superior performance if a fund manager possesses stock selectionability to outperform the market but no timing ability. The Jensen alphamay be estimated by the least squares regression of Eq. (4) and it representsthe constant periodic return that the fund manager is able to earn above anunmanaged portfolio, which is having identical market risk.

Sawicki and Ong (2000) suggest that further study may be done on theperformance persistence using the conditional models as no studies havebeen done on whether the extreme performers may be more easily detectedusing conditional methods. This study will try to examine if there areevidences of persistence of performance measured by conditional Jensenmeasure and whether there is difference in the persistence pattern betweentraditional Jensen measure and conditional Jensen measure.

Following Shanken (1990), Ferson and Schadt (1996) approximate thebeta in the conditional model that is assumed to be a linear function ofpublic information vector Zt that captures changing economic conditions,and is given by

biðZtÞ ¼ b1;i þ b02;iZt (5)

where b1,i is the unconditional mean of the conditional beta E½biðZtÞ�. Thecoefficient b2,i tracks how bi varies with the innovation of the conditioningvariable vector zt ¼ Zt � EðZtÞ. By multiplying the excess market returnRm;t � Rf ;t to biðZtÞ given by Eq. (5), the following regression equation isobtained:

Ri;t � Rf ;t ¼ aþ b1;iðRm;t � Rf ;tÞ þ b02;i½ZtðRm;t � Rf ;tÞ� þ et (6)


The additional factor may be interpreted as the returns on self-financingdynamic strategy that purchases zt units of market portfolio by borrowingon the risk-free market.

The conditional Jensen model uses public information variables that aresimilar to those have been identified as useful for predicting risks andsecurity returns over time in previous studies, but will be adjusted to complywith Hong Kong investment market. The following public informationvariables are used: (1) the lagged level of 1-month MPFA prescribed savingrate that is closest to 1 month to maturity at the end of the previous monthðSAVt�1Þ; (2) the dummy variable for the month of January ðJANtÞ; (3) thelagged dividend yield in the Hang Seng Index at the end of the previousmonth ðDIVt�1Þ; (4) the lagged measure of the slope of the term structurethat is the change in the term spread and is the difference between thematurity 10-year HKMA Exchange Fund Note and the 91-day HKMAExchange Fund Bill, both are annualized monthly averages ðTERMt�1Þ; and(5) the lagged quality spread in the corporate bond market that is the changein the corporate bond default-related yield spread and is the differencebetween the Moody’s BAA-rated corporate bond yield and the AAA-ratedcorporate bond yield, using the monthly average yields for the previousmonth ðDEFt�1Þ.

Given these five economic variables, the public information vector Zt

may be a vector of the five economic variables mentioned above and theproduct b02;i Zt will be a linear combination of these five variables asfollows:

b02j Zt ¼ bSAV;t SAVt�1 þ bJAN;t JANt�1 þ bDIV;t DIVt�1

þ bTERM;t TERMt�1 þ bDEF;t DEFt�1(7)

where bSAV;t, bJAN;t, bDIV;t, bTERM;t, and bDEF;t measure the extent to whichthe conditional beta diverges when market indicators are taken intoaccount.

Eq. (6) may be modified by combing Eq. (7) to derive the followingconditional Jensen measure (8):

Ri;t � Rf ;t ¼ aþ ðb1;i þ bSAV;t SAVt�1 þ bJAN;t JANt�1

þ bDIV;t DIVt�1 þ bTERM;t TERMt�1

þ bDEF;t DEFt�1Þ

� ðRm;t � Rf ;tÞ þ et

(8)


4. DATA

The data set consists of monthly prices of MPF constituent equity funds,from the date of the launch of MPF scheme on December 1, 2000 toDecember 31, 2004. All of these data were provided by Lipper AsiaLimited.2 The sample contains a total of 48 monthly observations. Most ofprevious studies suggested that using monthly or quarterly data for mutualfund performance studies is appropriate as the distribution of monthly orquarterly returns are closer to normal distribution of daily returns.According to the categories specified by Hong Kong Investment FundAssociation (HKIFA), the sample equity funds are separated into HongKong Equity, US Equity, Asia Excluding Japan Equity, Japanese Equity,Pacific Basin Excluding Japan Equity, European Equity, and Global Equity.This study excludes the category ‘‘other equity’’, which includes only oneKorean equity fund and there is no benchmark designed for this category.Separating the funds is important when using risk-adjusted alphas tomeasure the performance, because the risk-adjusted measures includedifferent benchmarks for different fund types.

It should be mentioned here that the NAV of equity fund is reducedby the exact amount of dividends or capital gain distributions paid to theshareholders. The monthly prices in the database have added thedistributions back to the NAV of equity fund.

Continuously compounded monthly returns are computed for each fundby taking the natural logarithm of the change in monthly NAV for eachmonth in the sample, i.e.,

Ri;t ¼ lnNAVi;t

NAVi;t�1(9)

where Ri,t is the return on fund i during the month t, NAVi;t is the net assetvalue of fund i at month t, and NAVi;t�1 is the net asset value of fund i atmonth t�1.

The MPFA prescribed saving rates quoted by the Mandatory ProvidentFund Scheme Authority was used as a proxy for the risk-free rate ðRf ;tÞ. Thesource of the quotes is from the official webpage of MPFA. As monthlyreturns are required, it is appropriate to convert the stated percent perannum to continuous monthly rates as follows:

Rf ;t ¼ln½1þ Rannum; f ;t�

12(10)

where Rannum;f ;t is the annual MPFA prescribed saving rates at month t.


The compounded monthly logarithmic returns on these benchmarks willthen be used as the market returns ðRm;tÞ in equations described in Section3.2. The benchmarks include the monthly returns on the following indiceswhich are currency hedged back into Hong Kong dollars, as required tomeet the regulations set by MPFA: (1) 90% FTSE MPF Hong Kong plus10% HSBC Index for HKEQ portfolio; (2) FTSE MPF USA (35% HK$Hedged) for USEQ portfolio; (3) FTSE MPF Asia Pacific ex Japan, AU andNZ for ASEQ portfolio; (4) FTSE MPF Japan (35% HK$ Hedged) forJPEQ portfolio; (5) FTSE MPF Asia Pacific ex Japan for PBEQ portfolio;(6) FTSE MPF Europe (35% HK$ Hedged) for EUEQ portfolio; and(7) FTSE MPF All-World (35% HK$ Hedged) for GBEQ portfolio. Theperformance of the benchmark should represent the performance that theinvestors would earn in the same class of securities. The data of the quotesof the series of these benchmark indices are obtained from the DataStream.

The conditional models described in Section 3.2 include five additionalvariables that are used to proxy the public information. The thirdadditional variable, ðDIVt�1Þ, represents the lagged dividend yield in theHang Seng Index; the series of HSI dividend yield are provided by the HSIServices Ltd. and obtained from its official webpage.3 The fourth variable,ðTERMt�1Þ, which involves the series of both interest rates of HKMAExchange Fund Note and HKMA Exchange Fund Bill, are provided by theHong Kong Monetary Authority and obtained from the DataStream. Thelast additional variable, ðDEFt�1Þ, which uses the series of Moody’s BAA-rated and AAA-rated corporate bond yields, are provided by the Moody’sInvestor Service.

One of the key issues to be considered for each analysis of mutual fundperformance is the potential survivor bias. If all funds of the populationbeing studied do not survive the entire study period, the data will includemeasures of the surviving funds only. Test results will thus be biased to somedegree, depending upon the attrition rate of the population, toward thesurvivors. The survivorship bias is minimal in this study because the numberof funds that did not survive constitutes a very small portion of all equityfunds. The only bias is that, if any funds closed and did not merge with anexisting fund, that fund would not have returns to be included for the year inwhich operations ceased. In fact, only one equity fund ceased operationswas operated by the trustee, which have ceased providing MPF services,Chamber CMG Choice. The data from this MPF trustee cannot becollected, so the funds provided by them are dropped from the database.Complete data were then assembled for all funds for which the data hadbeen published during the 4-year period of 2001–2004.



Tables 1–3 present the results of persistence tests by nonparametricapproaches. Three statistical tests are employed to test the significance ofthe persistence. They are Malkiel z-, CPR-, and w2-test. The Malkiel z-testseparately compares if the percentage of WW category is significantly higherthan 0.5 compared with WL category and if the percentage of LL categoryis significantly higher than 0.5 compared with LW category. The CPR- andw2-test take account of all four categories in the test statistic at the sametime. If the persistence effect is strong and significant, the results of theCPR- and w2-test should be consistent.

Table 1 presents the contingency table test for persistence in successiveannual raw returns of MPF equity funds. Panel B indicates 72.37% of allwinners in the current year are winners in the subsequent year. The Malkielz-test indicates the percentages repeat of winner equity funds aresignificantly higher than 50% at 1% significant level. Regarding the ‘‘coldhand’’ phenomenon – repeating losers – casually suggested by Malkiel(1995), Panel B indicates 68.92% of all losers repeat to be losers in thesubsequent year and the Malkiel z-test indicates the hypothesis of repeating-losers is not rejected. The CPR- and w2-test consistently indicate significantpersistences in annual raw returns at 1% significance level, regardless of‘‘hot-hand’’ or ‘‘cold-hand’’ phenomena. Considering consecutive annualperiods individually, Panel A of Table 1 shows more significant evidences ofperformance persistence exist in the more recent two periods 2002–2003, and2003–2004. The equity funds also have performance persistence in the firstperiod 2001–2002 but not significant. Confirmed by CPR- and w2-test withinsignificant statistics of 1.5342 and 2.4027, respectively. No reversalpattern, which is indicated by percentage of repeat winning funds less than50% and repeating losers less than 50%, can be observed in all periods.

Two-way contingency table is also constructed based on single-factorJensen measures which take account of risk to adjust the returns. Table 2shows the result of analogous persistence test on successive annual Jensenmeasures. Panel B of Table 2 indicates that the combined results still exhibitsignificant evidence of performance persistence suggested by consistentresults of both CPR- and w2-test. The Malkiel z-test suggests both thepercentages of repeat winning funds and repeat losing funds are significantlygreater than 50%. However, both percentages of repeat winners and repeatlosers are lower than those in previous table. Considering consecutive yearsindividually, interestingly, the funds show significant performance persis-tence at 5% level if the performance is measured by risk-adjusted returns in


Table 1. Two-Way Contingency Table: Ranked Fund Raw Returnsover Successive 1-Year Intervals.

Panel A: Individual annual periods

Subsequent Year 2002

Winners Losers

Initial year Winners 12 9

2001 (57.14%) (42.86%)

Total funds: 42

New funds: 10 Losers 7 14

(33.33%) (66.67%)

Malkiel z-test on repeat winners: Z=0.6547

Malkiel z-test on repeat losers: Z=1.5275��

Cross product ratio (CPR) test: Z=1.5342 CPR=2.6667

w2-test: w2=2.4027


Winners Losers


2002 (76.92%) (23.08%)

Total funds: 52


(26.92%) (73.08%)

Malkiel z-test on repeat winners: Z=2.7456�

Malkiel z-test on repeat losers: Z=2.3534�

CPR-test: Z=3.4307� CPR=9.0476

w2-test: w2=13.0193�


Winners Losers


2003 (79.31%) (20.69%)

Total funds: 56


(33.33%) (66.67%)



CPR-test: Z=3.3182� CPR=7.6667

w2-test: w2=12.069�


2001–2002 period. Consistent with the test on persistence of raw returns,significant performance persistence phenomenon occurs in the periods2002–2003 and 2003–2004. However, the evidences of persistence in thesetwo periods are weaker as the returns are adjusted by risks. The pattern ofpersistence appears to be not affected by the risk adjustment. The differentsystematic risk across the fund managers estimated by the single-factorJensen model is not great.

Table 3 presents the analogous two-way table for evaluating persistence inconditional Jensen measure and indicates that the persistence effect becomesweaker as the performance measure is conditional on public informationvariables, shown by fewer repeat-winners and repeat-losers percentages andsmaller value of z-statistics on repeat winners and repeat losers (2.3627 and2.2549 compared with 2.5236 and 2.3250, respectively) and w2 statistics


Panel B: Combined results of successive annual periods

Combined Results

Winners in

Holding Period

Losers in

Holding Period

Combined results Winners in

evaluation

period

55 21

Total funds: 150 (72.37%) (27.63%)

New funds: 24 Losers in

evaluation

period

23 51

(31.08%) (68.92%)



CPR-test: Z=4.9000� CPR=5.8075

w2-test: w2=25.6061�

Note: Winners and losers are ranked relative to the median raw return and determined over

1-year period, and then ranked over the subsequent 1-year periods. This provides three separate

periods. Winners are defined as funds with returns above or equal median and losers are funds

with returns below the median. WW and LL denote winners and losers in two consecutive

periods. LW denotes losers in the first period and winners in the subsequent period. WL denotes

winners in the first period and losers in the subsequent period.�Indicate significant persistence at 1% level.��Indicate significant persistence at 5% level.��Indicate significant persistence at 10% level.


Table 2. Two-Way Contingency Table: Ranked Fund Risk-AdjustedReturns (Jensen Alpha) over Successive 1-Year Intervals.



Winners Losers


2001 (59.09%) (40.91%)

Total funds: 42


(30.00%) (70.00%)



CPR-test: Z=1.8613�� CPR=3.3704

w2-test: w2=3.5788��


Winners Losers


2002 (65.38%) (34.62%)

Total funds: 52


(34.62%) (65.38%)

Malkiel z-test on repeat winners: Z=1.5689��


CPR-test: Z=2.1818�� CPR=3.5679

w2-test: w2=4.9231��


Winners Losers


2003 (67.86%) (32.14%)

Total funds: 56


(42.86%) (57.14%)


Malkiel z-test on repeat losers: Z=0.7559

CPR-test: Z=1.8600�� CPR=2.8148

w2-test: w2=3.5406��


(10.6420 compared with 11.7492). One of the individual 1-year periods,2002–2003, even becomes not exhibiting significant evidences of persistence.

6. FURTHER ANALYSIS OF PERFORMANCE

PERSISTENCE

6.1. Relation between Fund Volatility and Performance Persistence

Previous studies found that higher volatile funds have lower probability tosurvive and higher volatile surviving funds tend to have better performance(Brown et al., 1992). The finding reveals that the high-volatile funds maydominate the category ‘‘WW’’ and few high-volatile funds are in the



Combined Results

Winners in

Holding Period

Losers in Holding

Period

Combined results Winners in

evaluation

period

49 27

Total funds: 150 (64.47%) (35.53%)


evaluation

period

27 47

(36.49%) (63.51%)



CPR-test: Z=3.3809� CPR=3.1591

w2-test: w2=11.7492�

Note: Like the raw returns, winners and losers are ranked relative to the median Jensen alpha

and determined over 1-year period, and then ranked over the subsequent 1-year periods. This

provides three separate periods. The definitions of winners and losers, the interpretations of

WW, WL, LW, and LL, the formulae to compute the test statistics are same to the

nonparametric persistence analysis on raw returns.�Indicate significant persistence at 1% level.��Indicate significant persistence at 5% level.��Indicate significant persistence at 10% level.


Table 3. Two-Way Contingency Table: Ranked Fund ConditionalRisk-Adjusted Returns (Conditional Jensen Alpha) over Successive

1-Year Intervals.



Winners Losers


2001 (66.67%) (33.33%)

Total funds: 42


(33.33%) (66.67%)



CPR-test: Z=2.1176�� CPR=4.0000

w2-test: w2=4.6667��


Winners Losers


2002 (53.33%) (46.67%)

Total funds: 52


(45.45%) (54.55%)


Malkiel z-test on repeat losers: Z=0.4264

CPR-test: Z=0.5608 CPR=1.3714

w2-test: w2=0.3152


Winners Losers


2003 (71.43%) (28.57%)

Total funds: 56


(32.14%) (67.86%)



CPR-test: Z=2.8582� CPR=5.2778

w2-test: w2=8.6539�


category ‘‘WL’’ as the funds in this category may not survive. This bias isnamed as selection bias. Higher volatile funds are expected to have higherselection bias.

Although there are no funds in the current trustees ceased their operationsduring the sample period 2001–2004, it is interesting to investigate therelation between the fund volatility and performance persistence. Thesample funds are separated into two batches, high- and low-volatile fundsusing the median variance over the entire period 2001–2004 as the criticalvalue to split the funds. The funds that have variances equal or larger thanthe median variance are classified as high-volatile, and low-volatile funds arethen defined as the funds with variances lower than the median. Analogouscontingency tables are constructed for two different subsets of funds.

The two-way contingency tables of raw returns of funds separated ashigh-volatile and low-volatile funds over successive annual periods are



Combined Results

Winners in

Holding Period

Losers in Holding

Period

Combined

results

Winners in

evaluation

period

50 29

(63.29%) (36.71%)

Total funds: 150


evaluation

period

26 45

(36.62%) (63.38%)



CPR-test: Z=3.2216� CPR=2.9841

w2-test: w2=10.6420�

Note: Like the raw returns, winners and losers are ranked relative to the median conditional

Jensen alpha and determined over one-year period, and then ranked over the subsequent 1-year

periods. This provides three separate periods. The definitions of winners and losers, the

interpretations of WW, WL, LW, and LL, the formulae to compute the test statistics are same

to the nonparametric persistence analysis on raw returns.�Indicate significant persistence at 1% level.��Indicate significant persistence at 5% level.��Indicate significant persistence at 10% level.


presented in Panel A of Table 4. The combined result of all successiveannual periods is then summarized in Panel B of Table 4, which shows thenumbers and percentages in each batch. The combined results of allsuccessive annual periods show that the high-volatile funds have higherpercentage of funds in the category ‘‘WW’’ (82.35%) than the other threecategories (17.65% in WL, 43.18% in LW, and 56.82% in LL, respectively),and the evidence of repeat WW phenomenon in the sub-sample of high-volatile funds is stronger than that in the low-volatile funds (64.29%). Theseresults indicate the hypothesis stated by Brown et al. (1992) that the highvolatile funds should have better performance in order to survive is notrejected in the case of Hong Kong MPF. Goetzmann and Ibbotson (1994)also found that high-volatile funds have stronger persistence in theirperformance and cited that this phenomenon indicates survivorship may bea possible source of bias in the performance study. However, the differencebetween the respective percentages in the ‘‘WW’’ category for high-volatileand low-volatile funds is not quite significant, which implies the selectionbias does not mitigate the performance persistence study.

6.2. Performance Persistence of Constituent Funds Provided by Same

Investment Manager

The performance persistence of the constituent funds provided by eachinvestment manager is also studied. Conducting performance analysis foreach investment manager consists of three major reasons. The first reason isthat the funds under the same investment manager (i.e., same fund house)may be under the same evaluation and supervision of the same management.The investment teams of different funds under the same investment managermay share the same research, marketing, and administrative support. Thereis a high possibility that the constituent funds provided by the sameinvestment manager employ similar investment strategies although theyhave different investment objectives. The study of performance persistenceof the equity funds within the same investment manager may prove thehypothesis suggested by Brown and Goetzmann (1995) that short-termperformance persistence may be caused by the correlation across themanagers. This may contribute to the current literature on the associationacross the managers that are due to same strategy and supervision.

The second reason is due to the MPF system; the participants may onlychange the trustee or investment managers and have to choose the fundsprovided by the selected investment manager. They may not choose the


Table 4. Two-Way Contingency Table: Ranked Fund Raw Return over Successive 1-Year Intervals,Grouped by High-Volatility Funds, Low-Volatility Funds.


High-Volatility Low-Volatility Total Sample

Subsequent Year Subsequent Year Subsequent Year

2002 2002 2002

Winners Losers Winners Losers Winners Losers

Initial year 2001 Winners 3 3 9 6 12 9

Losers 7 9 0 5 7 14

2003 2003 2003



Losers 7 9 0 10 7 19

2004 2004 2004



Losers 5 7 4 11 9 18


Combined Results in Holding Period Combined Results in Holding

period

Combined Results in Holding

Period


Combined results in evaluation period Winners 28 (82.35%) 6 (17.65%) 27 (64.29%) 15 (35.71%) 55 (72.37%) 21 (27.63%)

Losers 19 (43.18%) 25 (56.82%) 4 (13.33%) 26 (86.67%) 23 (31.08%) 51 (68.92%)

Note: Winners and losers are ranked relative to the median raw return and determined over 1-year period, and then ranked over the

subsequent

1-year periods. This provides three separate periods. The funds are split into the high- and low-volatile funds by using median variance of all

equity funds over the entire period 2001–2004 as the benchmark. A fund is classified as high-volatile fund if its variance of annual returns is

higher than or equal to the median variance of all equity funds. A fund is classified as low-volatile fund if its variance of annual returns is

lower than the median variance of all equity funds.

PATRIC

KKUOK-K

UN

CHU

414

funds provided by different investment managers at the same time. Due tothis limitation in the system, the participants have more interest in whetherthe past performance records of investment managers are useful whenselecting the investment managers.

The last reason is that the current individual MPF scheme participantshave no right to select the trustees and investment managers, and such rightis actually transferred to employers. The employers have more interest onthe performance and performance persistence of the funds managed by thesame investment manager rather than performance of individual funds.

Following the nonparametric approach used in Section 5, two-waycontingency tables are constructed to examine the evidences of performancepersistence of the funds provided by the same MPF investment manager.The performance of the funds provided by the same investment managersare measured on monthly basis. The average monthly returns of allequity funds offered by the same investment managers are used as a proxyof monthly performance of each investment manager. The investmentmanagers are then separated into two groups: winners – which have averagereturns equal or above the median return of all investment managers forthat month; and losers – which have average returns below the median. Theprocedure is repeated every month. Similar to the nonparametric approachby contingency table in the previous section, an investment manager isdefined as WW if it is in the category W for consecutive months; LL ifa investment manager is in the category L for consecutive months; WL if ainvestment manager shifts from W to L and a investment manager shiftsfrom L to W is then defined as LW. An MPF investment manager isconsidered as having significant evidence of performance persistence if theprobability of repeating previous month’s above median returns (repeatwinning) is significantly more than 50%, which is diagnosed by the teststatistic given in Eq. (1).

Table 5 summarizes the two-way contingency table using average monthlyreturns of all equity funds of each investment manager and the number inthe cells of the two-way table is the number of repeat-winning, repeat-losing,winning–losing, and losing–winning monthly periods. The two-way table issupplemented with repeat-winning and repeat-losing z-statistics to investi-gate the significance of the persistence.

Contrary to the findings summarized in Table 1 that indicates there isevidence of performance persistence in raw returns over successive 1-yearintervals, the results in Table 5 indicate most of the investment managers donot have percentages of repeat-winning months significantly more than50%, 14 out of 21 investment managers which offer equity funds show the


Table 5. Two-Way Contingency Table: Ranked MPF Trustee and Investment Manager Returns overSuccessive Months from January 2001 to December 2004.

Trustees Investment Managers Initial

Month

Subsequent

Month

Repeat

Winners

(%)

Repeat

Winning

z-Statistic

Repeat

Losers

(%)

Repeating

Losing

z-StatisticWin Lose

AIA-JF AIG Win 6 10 37.50 �1.0000 42.11 �0.6882

Lose 11 8

AXA AXA Win 8 12 40.00 �0.8944 55.56 0.5774

Lose 12 15

BCT BCT Win 5 7 41.67 �0.5774 46.15 �0.2774

Lose 7 6

BOCI-Prudential BOCI-Prudential Win 1 6 14.29 �1.8898ww 50.00 0.0000

Lose 6 6

CMG First State Win 20 11 64.52 1.6164�� 25.00 �2.0000ww

Lose 12 4

Dexia (Standard Chartered MPF) Nexus Win 19 9 67.86 1.8898�� 47.37 �0.2294

Lose 10 9

HSBC MPF HSBC MPF Win 10 17 37.04 �1.3472www 20.00 �2.6833w

Lose 16 4

HSBC MPF Hang Seng MPF Win 10 17 37.04 �1.3472www 20.00 �2.6833w

Lose 16 4

HSBC MPF Schroder Win 14 13 51.85 0.1925 40.00 �0.8944

Lose 12 8

HSBC Institutional Kingsway Win 25 7 78.13 3.1820� 53.85 0.2774

Lose 6 7

HSBC Institutional Fidelity Win 6 15 28.57 �1.9640ww 42.31 �0.7845

Lose 15 11

PATRIC

KKUOK-K

UN

CHU

416

HSBC Institutional INVESCO Win 6 6 50.00 0.0000 44.44 �0.3333

Lose 5 4

HSBC Institutional DRESDNER Win 0 1 0.00 �1.0000 50.00 0.0000

Lose 1 1

ING ING Win 12 12 50.00 0.0000 47.83 �0.2085

Lose 12 11

Manulife Manulife Win 4 11 26.67 �1.8074ww 62.50 1.4142��

Lose 12 20

Mass Mutual Franklin Templeton &

Salomon Brothers

Win 3 7 30.00 �1.2649 30.00 �1.2649

Lose 7 3

MLC MLC Win 10 14 41.67 �0.8165 34.78 �1.4596www

Lose 15 8

PCI PCI Win 9 14 39.13 �1.0426 43.48 �0.6255

Lose 13 10

Principal Principal 800 Win 10 15 40.00 �1.0000 31.82 �1.7056ww

Lose 15 7

Principal Principal B300 (previously

DBS-Kwong On)

Win 3 3 50.00 0.0000 60.00 0.6325

Lose 4 6

Principal Zurich-Chinese Bank Win 8 13 38.10 �1.0911 50.00 0.0000

Lose 13 13

Note: The table presents the number of repeat-winning, repeat-losing, and reversal times of each investment manager that is providing equity

funds over the period from January 2001 to December 2004. The average monthly raw return of the equity funds provided by the same

investment manager is used as a proxy of monthly performance of that manager. Winners and losers are ranked relative to the median raw

return of all investment managers and determined over 1-month period, and then ranked over the subsequent 1-month periods. This provides

47 separate periods. The definitions of winners and losers, the interpretations of WW, WL, LW, and LL, the formulae to compute the test

statistics are same to the other nonparametric persistence analysis.�Indicate significant persistence at 1% level.��Indicate significant persistence at 5% level.��Indicate significant persistence at 10% level.wIndicate significant reversal at 1% level.wwIndicate significant reversal at 5% level.wwwIndicate significant reversal at 10% level.

Perfo

rma

nce

Persisten

ceo

fP

ensio

nF

un

dM

an

ag

ers417

percentages of repeating winning months less than 50%. Among the seveninvestment managers which exhibit a percentage of repeating winningmonths larger than 50%, three show percentages significantly larger than50% at either 10 or 1% significant level. They are First State Investments(HK) Ltd. (trustee: CMG Asia Trustee Company Limited), Kingsway FundManagement Limited (trustee: HSBC Institutional Trust Services (Asia)Limited) and Nexus Trust Services (HK) Limited (trustee: Dexia TrustServices HK Limited appointed by Standard Chartered MPF). It impliesthat the equity funds offered by these three investment managers tend torepeat winning monthly periods. More investment managers exhibitpercentages of repeat-winning months less than 50%, in other words,reversal. Among them, five investment managers BOCI-Prudential AssetManagement Limited (trustee: BOCI-Prudential Trustee Limited), HSBCProvident Fund (HK) Limited (trustee: HSBC Provident Fund Trustee(HK) Limited), Hang Seng MPF Services (trustee: HSBC Provident FundTrustee (HK) Limited), Fidelity Investments Management (Hong Kong)Limited (trustee: HSBC Institutional Trust Services (Asia) Limited), andManulife Provident Funds Trust Company Limited exhibit significantreversal at either at 10 or 1% level of significance. It suggests that the equityfunds offered by these investment managers tend to have more losingmonths after winning months.

Regarding the percentage of repeating losing months, only oneinvestment manager show significant persistence with percentage ofrepeat-losing months significantly larger than 50% – Manulife ProvidentFunds Trust Company Limited. Among the 21 investment managers thatoffer equity funds, only seven of them show percentages of repeat-losingmonths equal to or larger than 50%. It suggests that the cold-handphenomenon hypothesis is not supported. On the other hand, moreinvestment managers which are losers in the initial month are more likelyto be followed by being winners in the subsequent months. Five of themexhibit repeat losing percentage significantly less than 50% at 1–10%significance level.

Table 6 provides the comparison of conditional and unconditionalprobabilities of repeat-winning and repeat-losing monthly periods, respec-tively. The investment managers are listed according to their rank orders ofrepeat-winning percentages and repeat-losing percentages in the table.Columns 3 (and 6) present the repeat-winning (losing) percentages which usethe number of initial winning (losing) monthly periods as the base. Columns4 (and 7) show the overall W–W(L–L) percentages which on the other handuse the total number of monthly periods as the base. Columns 5 (and 8)


Table 6. Comparison of Conditional and Unconditional Repeating Winning and RepeatLosing Percentages.

Trustees Investment

Managers

Repeat

Winners

(%)

Overall

W–W

(%)

Overall

W–W (%)

Rank

Repeat

Losers

(%)

Overall

L–L

(%)

Overall

L–L (%)

Rank

HSBC Institutional Kingsway 78.13 55.56 1 53.85 15.55 16

Dexia (Standard

Chartered MPF)

Nexus 67.86 40.43 3 47.37 19.15 12

CMG First State 64.52 42.55 2 25.00 8.51 19

HSBC MPF Schroder 51.85 29.79 4 40.00 17.02 14

HSBC Institutional INVESCO 50.00 28.57 5 50.00 33.33 3

ING ING 50.00 25.53 6 47.83 23.40 8

Principal Principal B300 (previously

DBS-Kwong On)

50.00 18.75 13 60.00 37.50 2

BCT BCT 41.67 20.00 11 46.15 24.00 7

MLC MLC 41.67 21.28 7 34.78 17.02 14

AXA AXA 40.00 17.02 15 55.56 31.91 4

Principal Principal 800 40.00 21.28 7 50.00 27.66 6

PCI PCI 39.13 19.57 12 43.48 21.74 11

Principal Zurich-Chinese Bank 38.10 17.02 15 31.82 14.89 18

AIA-JF AIG 37.50 17.14 14 42.11 22.86 10

HSBC MPF HSBC MPF 37.04 21.28 7 20.00 8.51 19

HSBC MPF Hang Seng MPF 37.04 21.28 7 20.00 8.51 19

Mass Mutual Franklin Templeton &

Salomon Brothers

30.00 15.00 17 30.00 15.00 17

Perfo

rma

nce

Persisten

ceo

fP

ensio

nF

un

dM

an

ag

ers419


Trustees Investment

Managers

Repeat

Winners

(%)

Overall

W–W

(%)

Overall

W–W (%)

Rank

Repeat

Losers

(%)

Overall

L–L

(%)

Overall

L–L (%)

Rank

HSBC Institutional Fidelity 28.57 12.77 18 44.44 19.05 13

Manulife Manulife 26.67 8.51 19 62.50 42.55 1

BOCI-Prudential BOCI-Prudential 14.29 5.26 20 50.00 31.58 5

HSBC Institutional DRESDNER 0.00 0.00 21 42.31 23.40 8

Note: The table presents the comparison of the percentages of repeat-winners and repeat-losers with those of win–win and lose–lose for each

investment manager. The investment managers are ranked in the order of repeat-winner percentage shown in the column 3, while columns 4

and 7 show the percentages of win–win and lose–lose, and columns 5 and 8 show the respective ranking of investment managers in terms of

their win–win and lose–lose percentage.

Repeat-winners % ¼ WWWWþWL

Repeat-losers % ¼ LLLWþLL

:W2W % ¼ WW

WWþWLþLWþLLL2L% ¼ LL

WWþWLþLWþLL:

PATRIC

KKUOK-K

UN

CHU

420

exhibit the rank orders based on their overall W–W(L–L) percentagesranked from the largest percentage. The comparison results of columns 3and 4 indicate a clear incidence that the investment managers with higherpercentages of repeat-winning monthly periods also have relatively higherpercentages of overall W–W percentages. It indicates the investmentmanagers which are always winners in successive monthly periods havelower possibilities of being losers in the period 2001–2004 and implies thatthe hot-hand investment managers (with percentages of repeat winningmonthly periods more than 50%) are more likely to have relatively superiorperformance than the cold-hand investment managers. Column 8 confirmsthat the hot-hand investment managers have relatively less overall L–Lpercentages and lower overall L–L percentage ranks. It implies that thesuperior investment managers not only performs well but also are less likelyto persist inferior performance.

7. CONCLUSIONS

The primary focus of this chapter is upon the issue of performancepersistence of MPF equity funds. This study provides the first comprehen-sive study on the performance persistence of MPF equity funds. Severalstatistical tests (repeat winners test, CPR-, and w2-test) that supplement thetwo-way contingency table have been employed and compared to evaluatethe performance persistence and the result indicates that the pastperformance of a fund has long been used as an indication of futureperformance. Overall, there is strong evidence of persistence with asignificant w2 statistic of 25.6061, a significant z-statistic of 4.9 for CPR-test, significant z-statistic of 3.9001 and 3.2549 for repeat winners and repeatlosers, respectively. Previous studies outside Hong Kong found littleevidences of performance persistence; while evidences of annual raw returnpersistence were proved by both nonparametric contingency tables andparametric regression analysis in this study. Annual horizon seems to beappropriate as the data may be affected by noise if the time horizon is tooshort. On the other hand, choosing so long of a period may allow the skilllevel of the fund manager to change.

The hypothesis that the performance persistence evidences are not affectedby the risk adjustment was also tested in this study. The persistence evidencesof risk-adjusted returns measured by traditional Jensen alpha measures,conditional Jensen alpha measures and Fama-French three-factor alphameasures were investigated. The phenomenon that the past risk-adjusted


returns are useful in predicting risk-adjusted returns was found although theevidence of persistence becomes weaker after adjusting for risks.

The data were then controlled for volatility by splitting the funds intohigh-volatile and low-volatile funds and the results continue to exhibitrepeat-winner and repeat-loser patterns. The repeat-winner pattern is moresignificant in the group consisting of samples of high-volatile funds, whichimplies the high-volatile funds put more effort to repeat their goodperformance in order to survive.

Finally, this chapter takes on a different perspective to explore thepersistence pattern of Hong Kong MPF. Besides studying the performancepersistence of individual equity funds and different fund groups, theperformance persistence of investment managers, which are assigned by theMPF fund trustees and are offering equity funds, were also examined.Different from the studies on the performance persistence of individualfunds or portfolios of funds classified by their investment regions, only 7 outof 21 investment managers are found to exhibit repeat winning patterns andamong them only three investment managers have significant repeat-winning percentages on monthly basis. Thus, there does not appear to be ahot hand phenomenon in investment managers of MPF equity funds. Thephenomenon of persistent inferior performance shown by repeat losingpercentages is also not evident, as only one investment manager exhibitsignificant repeat-losing percentage. The comparison of conditional andunconditional probabilities of repeat-winning and repeat-losing monthlyperiods shows a strict association between the investment manager’sperformance persistence and its overall performance. The investmentmanagers exhibit high repeat-winning percentages, which are conditionalon their prior performance tend to also have higher overall W–Wpercentages which are unconditional on their prior performance. Theseinvestment managers also tend to have lower overall L–L percentages andimply they are less likely to persist inferior performance.

In conclusion, this chapter may provide us a picture that the pastperformance of the MPF equity especially the performance in the previousyear may be a good indication of the performance in the coming year. TheMPF participants may use historical information to beat the pack and thepast performance may also be a good indicator to find out good investmentmanagers versus bad ones. The study implies MPF mandate should be set upon an annual basis although this might ignore the fund shifting cost at suchregular intervals. The extensions of the methodologies supplemented thenonparametric contingency table to evaluate persistence in performance forsmall samples may be applicable for other emerging regional fund industries.


The Hong Kong Mandatory Provident Fund Schemes Authority (MPFA)did not require fund trustees release their fund characteristics and theirequity portfolios to the public until November 2005 and the investmentmanagers consider such information are their confidential during theobservation period 2001–2004. As the operations become transparent andmore information especially the fund operating characteristics such as fundcash flows, fund size, fund expense level, and turnover rates may beavailable in the future, more researches may be done on the determinants ofthe equity fund returns. Substantial amount of researches have been done inUS on the determinants of mutual fund returns. Among these fundoperating characteristics, the fund cash flows will be the major focus becausesome studies in US show that large unexpected cash flows to the funds maycause the fund managers make irrational investment decisions and thusinfluence the manager’s stock selection skill. The study on determinants ofthe performance persistence is also the other major interest in furtherresearches as well as the availability of the determinant data.

NOTES

1. Source: Census and Statistics Department, Hong Kong SAR Government.2. http://www.lipperweb.com3. http://www.hsi.com.hk

REFERENCES

Allen, D. E., & Tan, M. L. (1999). A test of the persistence in the performance of UK managed

funds. Journal of Business and Accounting, 26, 559–593.

Brown, S. J., Goetzmann, W., Ibbotson, R. G., & Ross, S. A. (1992). Survivorship bias in

performance studies. Review of Financial Studies, 5, 553–580.

Brown, S. J., & Goetzmann, W. N. (1995). Performance persistence. Journal of Finance, 50,

679–698.

Carpenter, J. N., & Lynch, A. W. (1999). Survivorship bias and attrition effects in measures of

performance persistence. Journal of Financial Economics, 54, 337–374.

Cheng, L. T. W., Pi, L. K., & Wort, D. (1999). Are there hot hands among mutual fund houses

in Hong Kong? Journal of Business Finance and Accounting, 26, 103–135.

Christensen, R. (1990). Log-linear models. New York: Springer-Verlag.

Christopherson, J. A., Ferson, W. E., & Glassman, D. A. (1998). Conditioning manager alphas

on economic information: Another look at the persistence of performance. Review of

Financial Studies, 11(1), 111–142.

Ferson, W. E., & Schadt, R. W. (1996). Measuring fund strategy and performance in changing

economic conditions. Journal of Finance, 51, 425–461.


Goetzmann, W. N., & Ibbotson, R. G. (1994). Do winners repeat? Journal of Portfolio

Management, 1994(Winter), 9–18.

Gruber, M. J. (1996). Another puzzle: The growth in actively managed mutual funds. Journal of

Finance, 51, 783–810.

Malkiel, B. G. (1995). Returns from investing in equity mutual funds 1971 to 1991. Journal of

Finance, 50, 549–572.

Sawicki, J., & Ong, F. (2000). Evaluating managed fund performance using conditional

measures: Australian evidence. Pacific-Basin Finance Journal, 8, 505–528.

Shanken, J. (1990). International asset pricing: An empirical investigation. Journal of



CHAPTER 20

FINANCIAL MARKET

IMPLICATIONS OF INDIA’S

PENSION REFORM

Helene K. Poirson

ABSTRACT

India’s planned pension reform will open the sector to private fund

managers. Drawing on international experiences, the chapter highlights

pre-conditions for the reform to kick-start financial development,

including (i) the buildup of critical mass, (ii) sufficiently flexible invest-

ment guidelines and regulations, and (iii) concurrent reforms in capital

markets. Given the limited scale of the planned reform, the key challenge

for India is to achieve sufficient critical mass. Options include granting

permission for existing workers to switch to the new system or

outsourcing all or part of the reserves of private sector provident funds

to the new pension fund managers.

1. INTRODUCTION

Several factors have given impetus to pension reform in India. Centralgovernment and state government pension liabilities have increasedconsiderably over the past decade1 and only 13 percent of the workforce is





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00020-9

425

dx.doi.org/10.1016/S1569-3767(07)00020-9.3d

currently covered by pension schemes. These are government employees, andworkers in the organized private sector covered by the Employees’ ProvidentFund (EPF) (defined contribution scheme) and the Employees’ PensionScheme (EPS) (defined benefit scheme).

Faced with these challenges, the government launched on January 1,2004, a New Pension System (NPS). The move shifted all new centralgovernment employees to a defined contribution plan from the currentnoncontributory defined benefit scheme. Participants in the new scheme willhave access to a range of investment products from selected companies.Once approved, the NPS would be open on a voluntary basis tonongovernment workers. Key legislation, however, is still under discussionin Parliament. As an interim arrangement, contributions from new civilservants – matched by government contributions – are being deducted andcredited a rate of return of 8 percent.

This chapter draws lessons from international experience on the financialmarket implications of India’s pension reform, with a focus on the followingtwo questions:

� How do the parameters of the NPS compare with privately managedsystems in other countries? Given its parameters, is the NPS likely togenerate fast growth of pension assets and stimulate financial marketdevelopment?� To what extent have regulatory limits, overly conservative investmentpractices, and regulatory restrictions hindered the ability of pension fundmanagers (PFMs) to achieve optimal portfolio diversification and con-strained the growth of the pension sector in countries that implementedsimilar reforms? What are other pre-conditions for pension reform todrive demand for bonds and equities?

2. BENCHMARKING INDIA’S PENSION SYSTEM

This section reviews to what extent the pension reforms’ track blazed byChile, and later followed by other Latin American and Eastern Europeancountries, is now being followed by India. The main features of India’s NPSare compared with privately managed systems in other countries (Table 1).The parameters of the pension reform envisaged in India appear in line withbest practice. However, two features set India apart from internationalcommon reform practice – the absence of a guaranteed minimum pensionfor participants (the so-called first pillar) and the only partially mandatory

HELENE K. POIRSON426

Table 1. Selected Countries: Main Features of Privately Managed Pension Systems.

Argentina Chile Mexico Peru Hungary Poland India United

Kingdom

Year of implementation of individual account

reforms

1994 1981 1992 1993 1998 1999 2005 1988

Mandatory (M) or voluntary (V) for new entrants to

workforce?

M M M M M M Ma V

Number of participants

Million 4.9 3.8 36.2 3.9 2.4 13.1 0.3 –

Percent of the potential contributor base – 58.0 – – – 70.0 – –

Contribution rate (percent of gross wage)b 7.0 12.3 6.5 10.9 8.0 7.3 10 2.5–5.25

Of which: PFM’s fee 1.12 0.99 1.47 1.99 0.97 1.6 – –

Investment performance

Average real rate of return on assets (2001–2005)c �7.1 7.0 6.3 15.1 – 9.8 – –

Average annual real rate of return (net of fees)d – 5.0 – – 3.9 – – –

Projected replacement ratio by 2030–2040 (percent)

Men 60 50–60 45 45 – – 43–95 48

Women (where different) – – 30 30 – – – –

Source: Faulkner-MacDonagh (2005), Impavido and Rocha (2006), Federacion Internacional de Administradoras de Pensiones, CONSAR,

OECD, IMF, and GAO.aMandatory only for central government employees recruited after January 1, 2004, and new employees of 19 state governments that have

joined.bTo the privately managed mandatory pension funds. For Argentina, the legally set contribution rate is 9 percent, but was modified by the

government in 2002 using temporary emergency powers.cAdjusted for inflation in US dollar of each year. 2000–2004 average for Poland. 2005–2003 average for Peru.d2001–2005 average for Chile. 1999–2003 average for Poland. 1998–2005 average for Hungary.

Fin

an

cial

Ma

rket

Imp

licatio

ns

of

Ind

ia’s

Pen

sion

Refo

rm427

character of the NPS – and may prevent the early achievement of sufficientcritical mass to stimulate financial market development.

2.1. India’s Pension Plan in International Perspective

The draft Pension Fund Regulatory and Development Authority (PFRDA)Bill, 2005, sets a framework for the development and regulation of pensionfunds in India. Once passed by Parliament, the Bill will allow the launch ofpersonal pension accounts in India and make the NPS available to workersin the unorganized private sector. It will also be available on a voluntarybasis (in addition to his/her mandatory cover) to any person governed bythe organized private sector schemes.

While the reform bill sets only the broad contours of the NPS and manydetails are yet to be finalized, its preliminary provisions place the new systemwell within international norms (see the appendix for highlights of relevantinternational experience).

� The employee contribution rate of 10 percent (matched by an equalgovernment contribution) is broadly within the international range.� The targeted terminal replacement rate (50 percent of the final wage) is inline with international experience and with the standards recommendedby the World Bank, and matches benefits under the existing system forgovernment employees.� Expected management costs of 0.5 percent of assets (Ministry of Finance,2005) are comparable to those in other emerging markets, although highcompared to low-cost providers in advanced economies. For instance, theUS federal civil servant Thrift Savings Program costs about 0.07 percentof assets (Faulkner-MacDonagh, 2005) and US low-cost private providerssuch as Vanguard and Fidelity charge fees of 0.2–0.3 percent of assets, lessthan half the levels envisaged in India. Larger volumes and larger averageaccounts for these US providers enable economies of scale and eliminatelow balance fees.� Participants are offered a menu of investment options, in line withbest practice, and have the option of switching between funds andschemes.2

� Voluntary participation over and above the mandated contribution rate isavailable to those participants that want additional coverage, providing aso-called third pillar.


Two features set India’s pension plan apart from common internationalpractice:

� The NPS only provides the second and third pillars. In other countriesthat have undertaken such reforms, the public pension system continuesto provide a first pillar, or comprehensive reform legislation is beingconsidered to introduce one (in the case of Chile). India’s organizedprivate sector is also covered under a two-pillar system.� Participation is mandatory only for new employees of the centralgovernment and 20 state governments that have joined the NPS. Existinggovernment employees and organized sector pensions schemes and fundsare exempt. Other countries, in contrast, mandated participation forall new entrants to the workforce and, in some cases, also for youngerworkers.

The potential for such schemes to build up assets and drive demand forpublic and private securities is sizeable (see Section 4). However, the twofeatures of India’s reform discussed above limit that potential.

� First, the absence of a first pillar may induce a relatively high share ofparticipants to opt for a conservative asset allocation, as subscribers seekto minimize the risk of an unfavorable ex post return on their assets. This‘‘safety bias’’ could be magnified if the regulator, concerned aboutinvestment risk, imposes excessive investment restrictions (see Section 3).� Second, a reform largely limited to new government workers may notgenerate sufficient critical mass early on to kick start financial marketdevelopment. Over time, while the entire government sector will becovered, the organized private sector will remain exempt (except forvoluntary participation in the third pillar).

The scope for voluntary take-up will depend on the relative attractivenessof the NPS. Existing private savings instruments in India include smallsavings (which provide a tax exempt and above-market rate of return),real estate, or own business, for the self-employed. In other countries thatimplemented similar reforms, while participation of the self-employed hasremained low, broad coverage was achieved by providing an option toswitch to workers covered under the old system (either in the initial law orthrough subsequent amendments) and by keeping the old system’s benefitsless generous. Tax incentives also played a role in other countries, butIndia’s fiscal situation constrains that option. Currently, NPS contributionsare tax-exempt, while benefits are taxed. To promote a level playing field, all

Financial Market Implications of India’s Pension Reform 429

private savings instruments should be subject to the same tax treatment. Inaddition to portfolio diversification (see Section 3), keeping costs low iscrucial to ensure net returns that attract new subscribers and provideadequate replacement rates.3 High management fees can dramaticallyreduce returns: net real returns in Chile averaged only 3 percent in the late1980s, after fees equivalent to 6 percentage points (ppts) of gross returns(Table 1). In Poland, total fees have also lowered net real returns in thefirst four years of the reform to an annual average of only 3 percent(Szekely, 2005).

Economies of scale and industry competition can help achieve costsavings. For instance, operations of an administrative nature – such ascollecting contributions – can be centralized (as planned in India). Fees alsotend to decline as growth of assets under management (AUM) enablesindustry consolidation. However, consolidation has raised concern aboutmarket power in some countries (Roldos, 2006). International experiencesuggests that industry competition is best enhanced by avoiding regulatoryimperatives that weaken PFMs’ ability to compete on the basis of rates ofreturn and result in excessive marketing costs – such as minimum returnrequirements relative to the industry average and overly tight investmentguidelines.

The fee structure can also encourage strong performance. An upfront feestructure results in providers focusing on attracting new accounts ratherthan achieving higher returns on existing accounts. A fee structure with bothfixed and variable components ensures better incentives. For example,private pension funds in the Dominican Republic can charge a monthlycommission of up to 1/2 percent of the individual wage plus a percentage ofannual returns above the benchmark (Samuel, 2006).

3. INVESTMENT POLICIES AND RETURNS ON

PENSION CONTRIBUTIONS

This section focuses on the extent to which regulatory restrictions andoverly conservative investment practices may have constrained the achieve-ment of optimal risk-adjusted returns by PFMs in existing DC systems.Sub-optimal returns have implications for the privately managed system’sattractiveness, as argued in the previous section. Tight investment controlscan also lessen the impact of pension reform on private securities’ demand(see Section 4).


3.1. Investment Limits

In the United States and the United Kingdom, regulations are based on the‘‘prudent person rule’’ or a self-regulatory framework. When accompaniedby prudential standards of diligence and expertise, this is generally viewedas superior to rules-based guidelines, because it places fewer restrictionson investment decisions. Most emerging markets, in contrast, regulateprivate pension funds via quantitative investment limits. There are typicallysignificant restrictions on equity and foreign investments, although bothtypes of restrictions have gradually been loosened over time in manycountries (Table 2).

Table 2. Equity and Foreign Investment Restrictionsfor Pension Funds.

Maximum Limits (In Percent of the Fund Size)

Equity Foreign Securities

Mature market

United Kingdom PPR PPR

United States PPR PPR

Germanya 30 20

Japanb 30 30

Canada No limit 30

France n.a. n.a.

Italy PPR 20

Emerging market

Argentina 50 20

Brazil 50 0

Chile 39 30

Colombia 30 20

Mexico 15 20

Peru 35 10.5

Hungary 50 30

Poland 40 5

Hong Kong SARc No limit No limit

Singapore PPR PPR

Note: PPR stands for prudent person rule.

Source: Chan-Lau (2004), Soueid (2005), and Roldos (2006).aSix percent in foreign equities of non-EU countries, 5 percent in non-EU bonds.bNo investment limits for employee pension funds.cAt least 30 percent of assets must be invested in Hong Kong dollar denominated assets.


Most countries also restrict investment in corporate bonds and derivatives.For example, Mexican institutional investors are not allowed to invest in bondsthat are rated below A, which limits their options to 20–30 large firms; more-over, they may invest no more than 5 percent of their assets in securities ratedsingle A (Soueid, 2005). Most countries have also adopted tight restrictionson the percentage of a company’s capital or outstanding bonds or on thepercentage of assets in a single issue that can be held by pension funds. Forinstance, Mexican pension funds cannot invest more than 20 percent of AUMin a single issue. Argentinean funds can only hold up to 5 percent of a com-pany’s capital and 5 percent of its bonds. Finally, investment in derivativeproducts is not allowed in most emerging countries, with the exception of Chile.

3.2. Asset Allocation

Investment practices in emerging markets tend to be conservative, withpension portfolios concentrated in fixed-income. In part, this could reflectthe rules-based guidelines. It could also be due to factors such as minimumreturn requirements, lack of financial sophistication of PFMs, weakperformance accountability, and dearth of private sector securities.

Indian pension funds have not participated in the corporate debtmarket, despite being allowed to do so.4 In part, this could be due tounderdeveloped and illiquid conditions of the corporate bond market (seeLuengnaruemitchai & Ong, 2005). In other emerging countries, the role ofriskier instruments also remains controversial (Table 3). In the US and theUK, in contrast, pension funds have a relatively low allocation to fixedincome and hold about 60 percent equities (in the form of shares or equity-linked mutual funds).

3.3. Diversification Abroad

Emerging market pension fund portfolios are also biased toward domesticassets, with the notable exception of Chile (Fig. 1).5 Polish pension fundsinvest only about 2 percent of their assets in foreign securities, less thanhalf the limit, perhaps because the foreign investment ceiling is too small tomake it worthwhile for pension funds to develop the related capacityand expertise. In El Salvador, pension portfolios are also home-biased asinvestments in foreign securities, until recently, were limited to those thatare traded on the local stock exchange (Samuel, 2006).6


Table 3. Pension Fund Portfolio Allocation, 2004 (As a Percent of Totala).

Selected

Countries

Cash and

Deposits

Bills and Bonds

Issued by Public

Administrations

Corporate

BondsbLoans Shares Land and

Buildings

Mutual

Funds

(CIS)

Unallocated

Insurance

Contracts

Other

Investmentsc

Czech

Republic

9.6 51.9 31.1 0.0 5.5 0.3 0.3 n.a. 1.3

Bulgaria 19.6 55.2 18.6 n.a. 3.3 1.7 n.a. n.a. 1.4

Estonia 4.4 33.9 23.3 0.0 35.1 1.0 6.2 0.0 0.8

Slovenia 13.3 46.3 32.4 n.a. 7.7 n.a. 0.3 n.a. n.a.

Hungary 1.3 74.9 2.0 n.a. 5.2 0.2 7.5 n.a. 8.9

Poland 5.8 58.9 1.4 0.0 33.4 n.a. 0.0 n.a. 0.5

Indonesiad 70.9 0.1 11.9 0.7 4.1 6.0 1.3 0.0 6.9

Korea 7.4 24.3 56.4 9.9 0.2 0.0 0.5 n.a. 1.4

Thailand 41.4 23.9 18.2 n.a. 13.7 n.a. 1.8 n.a. 1.0

Singapored 2.7 96.4 0.0 0.0 0.0 0.2 0.0 0.0 0.7

Colombia 0.8 48.5 30.1 0.0 6.2 0.0 2.2 0.0 12.2

Mexico 0.0 85.2 11.7 n.a. n.a. n.a. n.a. n.a. 3.1

Brazila 44.2 14.9 2.2. 3.9 15.9 6.7 11.6 0.0 0.6

Turkey 0.0 72.6 0.0 0.0 13.2 0.0 0.0 0.0 14.2

United

Kingdome

2.5 14.7 6.8 0.5 43.4 4.3 15.4 6.0 6.3

United States 8.3 6.4 5.0 0.1 35.5 0.6 30.7 9.4 4.0

Note: CIS stands for collective investment scheme.

Source: OECD and Global Pension Statistic.aTotal may not add up due to rounding or negligible value.b‘‘Corporate bonds’’ include corporate and financial sector debt instruments.cThe values include short-term payable accounts to the fund managers (commissions), payable loans, and the amount relative to the

liquidation of one pension fund (Pessoal da Caixa Geral de Depositos), transferred amount relative to the liquidation of one pension fund,

transferred to social security, worth about EUR 1 billion.d2002 data.e2003 data.

Fin

an

cial

Ma

rket

Imp

licatio

ns

of

Ind

ia’s

Pen

sion

Refo

rm433

Two factors appear to have contributed to Chile’s success in achievingglobal pension asset diversification: (i) allowing Chilean funds to hedgeforeign currency exposure using currency forwards (Walker & Lefort, 2002)and (ii) allowing them to invest in global mutual funds, thus providing awealth of investment options and bypassing the lack of experience of PFMs.After the 1998 crisis caused domestic returns to plummet, higher foreignallocations allowed Chilean funds to achieve higher returns and to meet theneeds of a sizeable retirement market without crowding-out the local capitalmarkets. Many countries are, however, reluctant to follow that route,in part because it complicates monitoring and involves additional fees,and also owing to the accompanying policy objective of developing localmarkets.

4. PENSION FUNDS AND CAPITAL MARKET

DEVELOPMENT

Pension reform is a logical catalyst for increased local institutionalinvestment and asset diversification, resulting in improved allocation offinancial savings and instruments. Sustainable fund inflows into local asset

10 15 20 25 30 35

Chile

Colombia

El Salvador

Mexico

Peru

Poland

Actual

Limit

Argentina

Hungary

0 5

Fig. 1. Selected Countries: Pension Fund Foreign Asset Allocation, as of June 2006

(In Percent of Total Assets). Note: The actual allocation may exceed the limit

in some cases due to investments in local foreign-currency denominated instru-

ments being counted as foreign assets but not counted toward the limit. The data

shown for Hungary is as of 2004. Sources: FIAP, Chan-Lau (2004), Soueid (2005),

Szekely (2005), and Samuel (2006).


markets reduce volatility and can induce a repricing of equities. Pensionreform can also have more qualitative effects, including better transparencyand governance, improvement of market microstructure, and innovation.

4.1. Financial Depth

India’s pension sector is small relative to more advanced Asian economiesand other emerging countries. While demographic trends and rising incomeshould contribute to rising demand for retirement services in the next twodecades, pension assets currently amount to only 53

4percent of GDP, much

below Singapore or Chile (Figs. 2 and 3).Emerging market pension fund assets are growing rapidly. Chile’s pension

AUM are nearing 65 percent of GDP after 22 years of operation of thefully funded system – a growth equivalent to nearly 3 ppts/year. While stillbelow the US level (95 percent of GDP), the size of Chile’s pension sectoris now similar to that of the UK. In the rest of Latin America, pensionassets have reached around 12 percent of GDP in the last decade, implyingannual growth of 1 to 11

2ppts, in line with G-7 experience since 1980

(Roldos, 2004). Later reformers, including Mexico and Hungary, have alsoexperienced rapid growth (Fig. 4).

0 10 20 30 40 50 60 70 80

Malaysia

Japan

Singapore

Hong Kong SAR

Korea

Thailand

India

ChinaPublic pension

Corporate pension

1.8

5.8

10.2

10.3

63.0

63.8

61.5

Taiwan POC

25.1

22.0

Fig. 2. Asian Countries: Pension Assets Under Management, 2005 (In Percent of

GDP). Source: HSBC.


The buildup of institutional assets has contributed to financial deepening(Fig. 5). In the G-7 countries, stock and bond market capitalization rose bymore than 40 and 20 ppts of GDP, respectively, between 1980 and 1998, ledby a 20 ppts of GDP increase in pension AUM (Roldos, 2004). Since 1981,

0 0.1 0.2 0.5 0.9 1.0 1.0 1.0 1.7 1.9 2.1 2.6 3.05.0 5.3

10.4 11.4 12.0 12.0 12.7 13.6

19.523.0 23.6

29.0

33.9

62.664.6

0

10

20

30

40

50

60

70

Per

u

(Weighted average 31.8 percent)

Ukr

aine

Indo

nesi

a

Lith

uani

a

Chi

na

Tai

wan

PO

CD

omin

ican

Rep

ublic

Latv

ia

Rus

sia

Slo

veni

a

Est

onia

Bul

garia

Cos

ta R

ica

Cro

atia

Tha

iland

Indi

a

Col

ombi

a

Bra

zil

Arg

entin

a

Uru

guay

El S

alva

dor

Bol

ivia

Ken

yaH

ong

Kon

gS

AR Is

rael

Sou

th A

fric

a

Sin

gapo

re

Chi

le

Fig. 3. Pension Fund Assets in Selected Non-OECD Countries, 2004 (In Percent of

GDP). Source: OECD.

0

2

4

6

8

10

12

14

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

Argentina Colombia

Peru Poland

Mexico

Fig. 4. Selected Emerging Countries: Pension Assets (In Percent of GDP). Sources:

FIAP, IFS, WEO, and staff calculations.


Chile’s market capitalization rose by nearly 30 ppts of GDP on the back ofsurging pension assets (Fig. 6).

Several studies have confirmed a positive impact of institutionalinvestment – including pension funds – on market capitalization usingpanel regressions, controlling for other determinants of stock and bondmarket capitalization, and encompassing both mature and emergingmarkets. Granger causality tests confirm that where causality exists, it runspredominantly from contractual savings to market capitalization, and notvice versa (see Roldos, 2004, for a comprehensive review).

The positive impact of pension reforms on market development, however,may take time to be reflected in the data. In Chile, since 1995, therelationship between growth of pension assets and market capitalizationhas been weak as returns on domestic equity investments turned flat ornegative from 1996 onward and pension funds diversified abroad. Thelater reformers are yet to experience a significant deepening of theirfinancial markets (relative to GDP) despite substantial growth in pensionassets, perhaps because AUM have not yet reached sufficient criticalmass. Factors such as the absence of supportive capital market regulationsand infrastructure may also have hindered financial deepening, in somecases causing the risk of significant distortions and asset price bubbles

0

0.1

0.2

0.3

0.4

0.5

0.6

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

United Kingdom

Netherlands

Zimbabwe U.S.A.

Canada

AustraliaDenmark

BelgiuNorway

Portugal

BrazilAustria Philippines

Pen

sion

ass

ets/

GD

P

Market capitalization/GDP

Fig. 5. Market Capitalization and Pension Assets, 1980–1985. Source: Beck,

Demirguc- -Kunt, and Levine (2000).


as growing imbalances emerge between the demand and supply of localsecurities (see Section 4.3).

4.2. Diversification of the Investor Base and Asset Allocation

A large part of the Indian financial sector is still mainly involved in deposit andloan services. Pension sector reforms could help increase institutional demandfor corporate bonds and help develop the local corporate bond market,enhancing the supply of long-term funds. The average maturity of bond issuein Chile increased from 10 to 15 years in the first half of the 1990s to 10–20years more recently (even 30 years for some issues). In Mexico, the bulk ofcorporate bonds was bought, held, and traded by institutional investors overthe past 5 years. Notably, pension funds held more than one third of alloutstanding bonds as of end-2004, despite relatively restrictive regulations(Soueid, 2005).

The proportion of Asian pension funds allocated to equity is significantlylower than in most other economies. However, young and growingpopulations in India and other Southeast Asian countries (Fig. 7) suggesta case for a more aggressive asset allocation. Institutional asset growthshould also – other things being equal – be an important factor in triggeringthe repricing of the stock market via reductions in liquidity and risk

0

10

20

30

40

50

60

70

80

90

1981

1982

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

Market capitalizations

Pension assets

Fig. 6. Chile: Pension Assets and Market Capitalization, 1981–2005 (In Percent of

GDP). Sources: FIAP, IFS, IFC, and staff calculations.


premiums and reduced cost of capital. Walker and Lefort (2002) confirmthis, finding a statistically significant effect of pension funds’ AUM onChile’s equity prices and the cost of capital.

Pension funds can also generate growing demand for new instruments,including high yield bonds, mortgage-backed securities (MBS), and currencyand interest rate derivatives. An increase in institutional investors’ demandfor such instruments however may require some relaxation of investmentrestrictions, accompanied by prudential standards of diligence and expertiseand the development of a ratings industry.

4.3. Increased Market Stability and Efficiency

In India, similar to the rest of Asia, asset markets remain characterizedby relatively high volatility; although volatility has declined recently. Thegrowth of pension and other institutional AUM could contribute to reducedmarket volatility, as a wider investor base and access to more informationand analysis facilitates price discovery.7 Walker and Lefort (2002) confirmthis link empirically in the case of Chile and for a broader sample of 33emerging economies.

The growth in private pensions’ AUM can have other qualitativeeffects on capital markets. Walker and Lefort (2002) show that in the cases

54

56

58

60

62

64

66

68

70

1950

1955

1960

1965

1970

1975

1980

1985

1990

1995

2000

2005

2010

2015

2020

2025

2030

2035

2040

2045

2050

India

Asia

Fig. 7. Asia: Population Aged 15–64 (In Percent of Total Population). Source:

United Nations.


of Chile, Argentina, and Peru, pension reform contributed significantlyto improvement of the regulatory and legal framework, increased trans-parency, and enhanced corporate governance. The reforms also increasedfinancial innovation, by fostering the growth of annuities, mortgage bonds,and other asset-backed securities; the creation of closed-end mutual fundsand local rating companies; and improved trading infrastructure.

However, the rapid growth of pension fund AUM may negatively affectlocal markets, when it outpaces the supply of private securities. This effectis magnified when tight controls limit the investment universe or whenregulations such as minimum required returns relative to an industry averageinduce herding behavior. The resulting concentration of investments ingovernment securities and securities from a limited number of localcompanies tends to magnify asset price swings and may make equity marketsmore prone to asset price bubbles. A large size of funds relative to localmarket supply may also result in liquidity constraints, since PFMs cannot sellassets without putting downward pressures on prices (Roldos, 2004).

5. CONCLUSIONS

While the broad parameters of the NPS appear in line with internationalbest practice, two features may limit the impact of the reform on financialmarkets. The absence of a first pillar and the only partially mandatoryparticipation set India’s plan apart and may result in concentration ofpension portfolios in government securities and higher-than-expectedmanagement fees as economies of scale are not realized early on.

Nonetheless, international experience points to several ways in which India’splanned pension reform could contribute to capital market development.

� Critical mass could be achieved faster by granting permission for exemptworkers to switch to the new system, and shifting all assets to privatePFMs.8 A less ambitious option could involve outsourcing of all or partof the management of accumulated reserves of partially funded schemessuch as EPF to the private sector under competitive bidding procedures(Holzmann, MacArthur, & Sin, 2000). Together with flexible investmentregulations, such reforms would ensure faster growth of pension assets.� Concurrent improvements in capital market regulations, laws, andinfrastructure are necessary. When such reforms are delayed, fast growthin pension AUM can generate imbalances between demand and supply oflocal securities and magnify volatility.


� Debt management agencies and regulators can support the provision of newinstruments for retirement savings by ensuring liquid government bonds(that serve an important benchmark function for the private sector) andissuing price-indexed bonds (to support the issue of price-indexed annuities).� A limited option for investments abroad can help PFMs diversify countryrisk, gain expertise and familiarity with new instruments, and relieve pressuresin local markets, when the supply of securities is restricted in the short term.

NOTES

1. India, similar to most other countries in Asia, has a system of statutory retirementpayments for government employees. Implicit pension debt is estimated at 25 percent ofGDP by the World Bank, with a significantly higher relative figure for some states.2. The current notification specifies four types of schemes of various risk–return com-

binations, reflecting differing combinations of government securities, corporate bonds,and equity shares, including an option with 100 percent investments in government bonds.3. For a subscriber contributing 10 percent of salary for 40 years, assuming annual

real wage growth of 2 percent, a net average real return on assets of 5 percent is necessaryto achieve the 50 percent replacement rate targeted by the Indian reform (Shah, 1997).4. Indian pension funds are allowed to invest up to 10 percent of new flows in

private corporate bonds.5. Even Chilean pension funds did not diversify meaningfully abroad until after

the 1997 Asian crisis, despite the gradual loosening of foreign investment limits, dueto high domestic returns (Roldos, 2004).6. A law passed in August 2006 allows 10 percent of pension portfolios to be

invested abroad.7. Moreover, if institutional investors’ risk tolerance is assumed to remain

relatively constant over time, volatility can be reduced as such investors takeadvantage of variations in risk premia (perhaps caused by variations in foreign orretail investors’ risk tolerance). This is done by purchasing securities when the riskpremia is high (at ‘‘low’’ prices) or vice versa.8. At the same time, it should be recognized that given the large share of the

informal sector in India, achieving full coverage would be difficult until theseworkers are brought in the more formal labor market.

ACKNOWLEDGMENTS

The author would like to thank J. Roldos, C. Kramer, M. De Broeck,J. Felman, W. Tseng, H. Shah, C. Klingen, M. Garcia-Escribano, A. Piris,R. Garcia-Saltos, J. Walsh, J. Canales, R. Lester, R. Palacios, W. Samuel, M.Kapoor, participants of a seminar held at the Ministry of Finance, India, andparticipants of a conference on India’s Financial System held at Wharton,University of Philadelphia, for their comments and useful discussions.


REFERENCES

Beck, T., Demirguc- -Kunt, A., & Levine, R. (2000). A new database on financial development

and structure. World Bank Economic Review, 14, 597–605.

Chan-Lau, J. A. (2004). Pension funds and emerging markets. IMF Working Paper No. 04/181.

Washington: International Monetary Fund.

Faulkner-MacDonagh, C. (2005). Addressing the long-run shortfalls of the Chilean pension

system. Chile-Selected Issues, IMF Country Report No. 05/316. International Monetary

Fund, Washington.

Holzmann, R., MacArthur, I., & Sin, Y. (2000). Pension systems in Asia and the Pacific:

Challenges and opportunities. Social Protection Discussion Paper Series No. 0014. World

Bank, Washington.

Impavido, G., & Rocha, R. (2006). Competition and performance in the Hungarian second pillar.

World Bank Policy Research Working Paper No. 3876. World Bank, Washington.

Luengnaruemitchai, P., & Ong, L. (2005). An anatomy of corporate bond markets: Growing pains

and knowledge gains. IMF Working Paper No. 05/152. International Monetary Fund,

Washington.

Ministry of Finance. (2005). The pension fund regulatory and development authority bill, twenty

first report. Lok Sabha Secretariat, New Delhi.

Roldos, J. E. (2004). Pension reform, investment restrictions, and capital markets. IMF Policy

Discussion Paper No. 04/4. International Monetary Fund, Washington.

Roldos, J. E. (2006). Pension reform and macroeconomic stability. Unpublished. International

Monetary Fund, Washington.

Samuel, W. (2006). Experience under pension reform. El Salvador-Selected Issues. Unpublished.

International Monetary Fund, Washington.

Shah, H. (1997). Toward better regulation of pension funds. Policy Research Working Paper

No. 1791. World Bank, Washington.

Soueid, M. (2005). Development of government securities and local capital markets in Mexico.

Mexico-Selected Issues, IMF Country Report No. 05/428. International Monetary

Fund, Washington.

Szekely, I. P. (2005). The Polish pension reform after six years. Republic of Poland-Selected

Issues, IMF Country Report No. 05/376. International Monetary Fund, Washington.

Walker, E., & Lefort, F. (2002). Pension reform and capital markets: Are there any (hard) links?

Social Protection Discussion Paper Series No. 0201. World Bank, Washington.

APPENDIX. INTERNATIONAL EXPERIENCE WITH

PRIVATELY MANAGED PENSION PLANS

Coverage

Chile’s system achieved fast coverage, with 40 percent of the labor force andnearly half of employees covered within five years of implementation(Faulkner-MacDonagh, 2005). Participation was initially mandatory onlyfor new entrants to the work force, but the government also offered


strong incentives to switch to other workers, including tax deductibility ofcontributions to the new system, not increasing the generosity of benefits ofthe old system, and lowering contribution rates dramatically.

In El Salvador, participation in the new system introduced in 1998 wasmandatory for workers under 36 years and new entrants to the labor force,while women 50 and over and men 55 and over had to remain in the oldsystem. Others had the option to switch. El Salvador provided strongincentives to switch, including income tax deductibility. The majority ofparticipants in the old system who had a choice did transfer to the newsystem, and coverage reached about a quarter of the economically activepopulation after one year (Samuel, 2006).

Fees

Chile’s management costs initially totaled up to 4 percent of wages (almostdouble their current levels). This in part reflected high fixed start up costs,e.g., computerization to manage millions of individual accounts andmarketing campaigns to entice contributors to switch to the new system.

El Salvador’s pension law initially limited pension fund fees to 312percent

of wages, but this was lowered to 3 percent in 2001, and further reduced to234percent in 2005 (Samuel, 2006). The law also sets out the services for

which private administrators (AFPs) can charge – including administrationof individual accounts, inactive accounts, and programmed withdrawals.At the outset, fees were set competitively to attract participants. However, inrecent years, the regulated maximum has been binding.

Replacement Rates

In Chile, according to recent estimates, the average worker’s pensionincome would replace 50–60 percent of the final salary over the mediumterm (2030–2040). However, over the longer term, the average replacementrate would substantially decline to just over 40 percent. These replacementrates are significantly lower than the 80 percent level promised at the time ofthe reform (Faulkner-MacDonagh, 2005) (Table 1).


PART VII:

BANKING AND DEBT MARKETS

CHAPTER 21

ON THE SAFETY AND SOUNDNESS

OF CHINESE BANKS IN THE

POST-WTO ERA

Lei Xu and Chien-Ting Lin

ABSTRACT

China’s accession to World Trade Organization (WTO) opened its

financial markets to foreign banks in December 2006. In addition to

foreign banks’ expertise and experience in modern banking activities, they

also appear to have the interest, competitiveness, and regulatory

advantages of competing with Chinese banks in the traditional Renminbi

(RMB) business. Such competition will lead to a loss of RMB deposits

and loans from local banks. Given that Chinese banks are currently ridden

with large non-performing loans and low capital adequacy, the foreign

bank entry will exert further pressure on the banks’ profitability and

solvency. Without larger regular bailouts from the central government

and fundamental changes on the roles of Chinese banks, China may

experience a banking crisis in the post-WTO era. We propose two types

of policy changes that may improve banks’ competitiveness and reduce the

likelihood of a banking crisis.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00021-0

447

dx.doi.org/10.1016/S1569-3767(07)00021-0.3d

1. INTRODUCTION

On December 11, 2006, as part of the conditions of accession to World TradeOrganization (WTO), China opened its financial markets to the world. Whenthe barriers to the largest emerging market are finally removed, will Chinesebanks face an unprecedented competition from their foreign counterparts?More importantly, given that the Chinese banks are ridden with large non-performing loans (NPLs) that require periodic bailouts from centralgovernment (see Ma, 2006), could a direct competition in the local RMBbusinesses with major international banks that are well capitalized and filledwith expertise cause a continuing deterioration in the quality of their assetsand further losses? Could the foreign competition ultimately lead to abanking crisis in China?

Our answers to the first two questions are yes and quite probable to thethird question. Our views are in contrast to Dobson and Kashyap (2006)who argue that foreign banks have little interests in the retail Reminbi(RMB) business as they tend to focus on the high margin activities and in theselected retail banking such as credit cards, mortgages, and investmentproducts. They reason that since domestic banks also lack the expertise inthese growth areas, direct competition over the same banking services seemsavoidable. Leung and Chan (2006) add to the argument that even if foreignbanks are interested in the local retail market, the improving competitivenessof Chinese banks, the additional regulatory requirements, and the localizedcultural and corporate practices will reduce their competitive advantage.Subsequently, the competition with foreign banks may not necessarily beone-sided.

In this chapter, we provide some anecdotal evidence based on recentactivities to suggest that foreign banks are not only interested in the RMBbusiness but have also demonstrated that they are capable of capturingsignificant market shares away from Chinese banks. In the core depositmarket, foreign banks are opening new branches in the strategic large moneycenters such as Shanghai, Beijing, and Guangzhou where the nation’sdeposits are concentrated. Foreign banks are also active in acquiring Chinesebanks that will in turn increase their size in deposit and loan portfolios.Furthermore, they have purchased NPLs from the big four banks thatprovide foreign banks with first-hand knowledge about the loan market anddetailed information about the customers.1 All of these moves taken byforeign banks suggest that they are prepared and ready to compete withdomestic banks in the lucrative RMB business when the regulatoryrestrictions were removed.

LEI XU AND CHIEN-TING LIN448

It is unlikely that Chinese banks can stand their grounds againsttheir foreign counterparts in the short run. The increase in foreigncompetition will, therefore, exert more pressure on the losses that Chinesebanks have experienced in recent years. Since domestic banks are alreadyin poor financial health due to large bad debts, low efficiency, andpoor management, the losses of low-cost deposits and quality loans to theforeign competitors will require larger regular capital injections fromthe government. It appears that the only way to avoid a banking crisis inthe short run is more government bailouts in the name of financial stabilityand social harmony. Without fundamental changes over the roles thatChinese banks play, it is doubtful that government intervention is sustainablein the long run.

The rest of the chapter is organized as follows. Section 2 provides somereasons of foreign banks’ interests in RMB business. Section 3 discussesregulations on deposit rates and foreign competitions on core deposits. InSection 4, we turn our attention to RMB loans. Section 5 highlights the skillsand expertise of foreign banks in modern banking. We then examineregulatory advantages that foreign banks enjoy over Chinese banks inbanking activities in Section 6. Section 7 examines the effect of foreigncompetition on the fragility of Chinese banks. In Section 8, we propose somepolicy changes based on the discussions in the earlier sections. Last sectionconcludes the chapter.

2. PURSUIT OF RETAIL RMB BUSINESS BY FOREIGN

BANKS

One reason that foreign banks may not forego RMB business in China is therelatively large domestic interest rate gap between domestic lending andborrowing rates. Fig. 1 shows the interest rate spreads between the bench-mark 1-year deposit rates and 1-year loan rates from 1996 to 2006. Thisofficial interest rate gap has increased steadily from 2.61 to 3.6% over the last10 years and appears to be driven by the continuing larger declines in thedeposit rates. An examination into the composition of loans in the Chinesebig four banks suggests that the average interest rate spread is larger than theofficial interest rate gap. As shown in Table 1, approximately 40% of loans inthese major banks are higher than the benchmark rates while 30% are belowthem. It indicates that the actual average interest rate gaps are higher thanthose reported by the People’s Bank of China (PBC).

On the Safety and Soundness of Chinese Banks 449

In contrast to the widening RMB interest rate spreads in Chinese banks,foreign banks have experienced declining interest rate gaps in their domesticmarkets due to bank deregulation and increased competition. Table 2 showsthat the average NIMs across seven different markets have declined over

0

2

4

6

8

10

12%

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

loan rate

deposit rate

interest rate gap

Fig. 1. RMB Deposit Rates, Loan Rates, and Interest Rate Gaps from 1996

to 2006.

Table 1. Loan Interest Rate Distribution of the Big Four Banks from2004 to 2006a.

Quarter Below Benchmark Benchmark Above Benchmark

(0.9, 1)b 1.0 Sub-total (1.0, 1.3) (1.3, 1.5) (1.5, 2.0) (2.0, N)

Q2 2006 29.56c 31.03 39.41 36.32 2.56 0.52 0.01

Q1 2006 28.26 31.83 39.92 36.80 2.62 0.47 0.02

Q4 2005 30.62 28.29 41.09 34.61 5.27 1.14 0.07

Q3 2005 26.81 32.11 41.08 36.45 4.07 0.45 0.11

Q2 2005 30.55 29.49 39.96 35.60 3.67 0.56 0.13

Q1 2005 27.90 32.40 39.80 35.10 3.80 0.60 0.20

Q4 2004 27.13 28.54 44.33 38.75 4.84 0.73 0.01

Source: PBC Monetary Policy Report, Q1 2001–Q2 2006.aLoan interest rates data only available since Q4 2004.b(0.9, 1) is the loan interest rates that lie in the range between 0.9 and 1.0 of the PBC benchmark

loan interest rates.c29.56 shows 29.56% of total loans at the big four banks are below PBC benchmark loan

interest rates.


comparable period.2 Most notably, NIMs in Italy, Sweden, the UK, and theUS have dropped between 13 and 39% over the 5-year period. These declinesare primarily due to smaller interest rate gaps rather than larger increases inearnings assets or bad debts. Given the presence of the attractive interest rategap, it is unlikely that foreign banks would not participate in the lucrativeretail RMB business. This view is further supported by a PriceWaterhouseCoopers (PWC) (2005) survey on 35 major foreign banks operating in Chinain which the respondents express the most attractive option to increasetheir market presence is through organic growth. Huang (2006a) also arguesthat foreign banks will apply this organic strategy when RMB businesses areopened to them.

The presence of the large interest rate gap in China is likely to persist in thenear future and thus allowing foreign banks sufficient time to establish astronghold in the retail RMB businesses. As part of the PBC’s interest ratepolicy for RMB commercial loans since 1999, commercial banks andmetropolitan credit cooperatives are free to establish their own lending rateswhile rural credit cooperatives are limited to 2.3 times of the benchmark rates.On the other hand, banks have long been prohibited to adjust RMB depositrates above the benchmark rates but are allowed to adjust deposit ratesdownwards since October, 2004.3,4 By allowing loan rates to increase butprohibiting similar increases in deposit rates, the PBC policy may contributeto the higher interest rate gaps than what the market would entail.

Another reason that the current interest rate gap will remain large is thatboth State Council and PBC are keen to see the retail sector profitable. It iswell known that Chinese banks are ridden with large bad debts that requireregular capital injections from the central government. The high interestmargins are often used as a cushion to absorb bank losses and reducegovernment bailouts.

Table 2. Average Net Interest Margins of Major Banks from2000 to 2005.

Country 2000 2001 2002 2003 2004 2005

Australia 2.27 2.19 2.16 2.20 2.07 2.06

France 0.94 0.84 0.80 0.96 0.88 0.85

Germany 0.83 0.90 0.81 0.80 0.72 0.66

Italy 2.06 2.13 2.49 2.33 2.08 1.63

Sweden 1.51 1.46 1.48 1.49 1.39 1.02

United Kingdom 2.36 2.07 2.04 1.97 1.67 1.44

United States 2.99 3.05 3.22 3.06 2.99 2.65

Source: BIS 72–76th Annual Report.


3. RMB DEPOSIT SERVICES

In addition to PBC’s regulation on deposit rates, PBC and China BankingRegulatory Commission (CBRC) limit the range of deposit services thatdomestic banks could provide.5 These services include demand deposits, termdeposits (3 months to 5 years), and notice deposits (1- and 7-day). Comparedwith a wider range of different accounts that foreign banks offer, the scope ofthese services that Chinese banks provide is small, and far from meeting theever changing needs of business and individuals. Therefore, foreign banksthat offer deposit services linking to market rates (such as cash managementaccounts) and catering to clients’ specific needs are likely to attract depositsaway from the domestic banks.

Dobson and Kashyap (2006), however, doubt that foreign entry wouldlead to large deposit outflows from the domestic banks. They reason thatmost depositors in countries such as Japan are slow to change their savingsbehavior and switch banks. Furthermore, there is little evidence ofinternational banks planning to build or acquire the branch infrastructure.The anecdotal evidence that we have gathered, however, paints a verydifferent picture.

First, foreign banks had been slow in building their branch infrastructurebut intend to speed up their expansions after restrictions had been removed.Huang (2006a) reports that major foreign banks plan to double the numberof its branches within 18 months. Among them, HSBC will increase itscurrent 26 branches by an additional 40–50 branches. Citigroup and Bank ofAmerica also indicate that they are ready to compete in RMB businessesafter probing the Chinese market in the past 5 years (see Shi, 2006). By June2006, there were 214 branches in China from 71 foreign banks. Based onthe projections, these major foreign banks could easily double or triple thenumber of branches in the next few years. Also, since 80% of total depositsin China are located in large money centers such as Beijing, Shanghai,Guangzhou, and Shenzhen, foreign banks do not need to compete withChinese banks across the country but on specific locations.

Besides setting up their own branches, foreign banks are rushing intobuying Chinese banks in major cities. The PWC (2006) survey reports acurrent flurry of M&A activities within China’s financial services industryand domestic banks will likely remain as major targets. Coupled with newbranches across China, the acquisition of domestic banks will allow foreignbanks to develop comprehensive retail networks.

Second, Liang and Yu (2006) find that depositors in China do concernabout the safety of their deposits. The issues on the safety and soundness of


domestic banking system, and the extensive presence of foreign banks mayencourage depositors to switch to foreign banks, which are well capitalizedand have better asset qualities than Chinese banks. Chen (2006) comparessome key indicators of listed major banks in China and the US. As shown inTable 3, US banks on average have higher capital adequacy ratios (CAR),lower leverage, and higher return on asset (ROA). In particular, Chinesebanks have an average equity to asset ratio of less than 4% compared to9.27% in the US banks. Furthermore, the average ROA of 0.59% amongChinese banks is more than twice lower than that of the US banks. The lowlevel of equity and poor financial performance of Chinese banks can largelybe traced to the large NPLs and the widespread corruption within thebanking industry (see Liu, 2005, 2006a). Therefore, flight to safety couldbecome a factor when foreign banks establish their retail networks.

Third, given the limits on deposit rates and types of deposit accounts thatdomestic banks could offer, the quality of bank services appears to be theonly competing basis for deposits. However, according to two surveys byNational Bureau of Statistics (2002a, 2002b), depositors seem critical aboutthe domestic bank services. It reports that only 10.1% of depositors aresatisfied with their overall services. Furthermore, more than 35% ofrespondents express little confidence about whether domestic banks couldimprove services to match with those of foreign banks within the next 5years. They have expressed their willingness to move to foreign banks. Tomake matters worse, domestic banks have started introducing bank fees since2003 without improving bank services. Various bank charges such as inter-city ATM transactions, debit cards annual fees, account administration fees,and inter-bank inquiry fees have gradually been added as part of the‘‘international standard practice.’’ These additional bank charges are likelyto force customers to pay more attention to the quality of services provided.

Fourth, foreign banks offer a wider array of financial services that providea one stop shopping banking service. Even with the recent expansions intowealth management services, domestic banks are still lagging behind theirforeign counterparts in innovation, sophistication, and the scope of financialservices. This view has been expressed by foreign banks as they see localcustomers are being underserved due to the lack of many standard bankingservices (see Dobson & Kashyap, 2006). The availability of a range offinancial services by foreign banks that offer market-based rates and returns,significantly higher than the ceiling rates imposed by PBC, will likely draindeposits away from domestic bank.

To insulate and protect domestic banks from direct competitions inindividual RMB deposits, the PRC has imposed a minimum of 1 million


Table 3. Some Key Indicators of Listed Banks in US and China.

Bank MV (US$

Million)

Total Asset

(US$ Million)

Net Asset (US$

Million)

CAR (%) Equity/Asset

(%)

ROA (%) ROE (%)

Comerica 8,600 53,000 5,100 11.75 9.62 1.64 16.90

M&T Bank 13,000 55,100 5,900 10.85 10.71 1.44 13.49

Regions 15,200 84,800 10,600 12.76 12.50 1.18 9.37

KeyCorp 1,4300 93,100 7,600 11.47 8.16 1.24 15.42

Fifth Third 20,500 1,05,200 9,300 10.51 8.84 1.50 16.60

BB&T 22,000 1,09,200 11,100 14.40 10.16 1.58 14.95

National City 22,200 1,42,400 12,600 10.56 8.85 1.40 15.54

SunTrust 27,800 1,79,700 16,900 10.57 9.40 1.18 12.02

US Bancorp 55,200 2,09,500 20,100 12.50 9.59 2.21 22.50

Wells Fargo 1,15,800 4,81,700 40,700 11.61 8.45 1.72 19.57

Wachovia 86,700 5,20,800 47,600 10.82 9.14 1.30 14.13

Bank of America 2,22,700 1,291,800 1,01,200 11.08 7.83 1.30 16.51

JPMorgan Chase 1,47,400 1,198,900 1,07,200 12.00 8.94 0.72 8.00

Citigroup 2,44,000 1,494,000 1,12,500 12.02 7.53 1.33 22.10

Average 72,529 4,29,943 36,314 11.64 9.27 1.41 15.51

China Merchants 11,400 11,000 3,100 9.06 3.37 0.60 15.93

MinSheng 5,300 69,000 1,900 8.26 2.78 0.54 17.48

Pudong Development 4,700 71,000 1,900 8.04 2.70 0.48 16.01

Hua Xia 2,300 44,100 1,300 8.23 2.95 0.39 12.33

Shenzhen Development 1,900 28,400 600 3.70 2.18 0.16 6.97

Bank of Communications 29,500 1,76,400 10,300 11.20 5.84 0.72 13.68

CCB 1,01,900 5,68,200 35,600 13.57 6.27 1.11 21.59

BOC 1,16,900 5,87,700 29,000 10.42 4.93 0.72 12.62

Average 34,238 1,94,475 10,463 9.06 3.88 0.59 14.58

Source: Chen (2006).

MV is based on the closing price on July 7, 2006 while other data on December 31, 2005.

PBC Exchange rate on December 31, 2005, US$1=RMB8.0702; July 7, 2006, US$1=RMB7.9936.

LEIXU

AND

CHIE

N-TIN

GLIN

454

RMB term deposits on foreign banks.6 Foreign banks, however, could easilyovercome the barrier by registering as ‘‘full foreign-funded banks’’ wherethe imposed requirements over the RMB businesses are the same as those ondomestic banks. They also can keep their tax preferential treatment with a15% income tax rate versus 33% for domestic banks. To take advantage ofthe favorable regulatory and tax rulings, Standard Chartered Bank hassubmitted its local registration application to CBRC less than 12 hours afterthe regulations on foreign funded banks were announced by the StateCouncil. Other major foreign banks have quickly followed suit to registerinto local banks. In the case of Hang Seng Bank, it has even prepared to belisted in the A share market.

4. RMB LOANS

Moving to the asset side of the balance sheet, foreign banks are active in thedomestic loan market in a number of ways. First, major international bankshave been buying NPLs well before the regulatory barriers were removed.The sale of these NPLs has been carried out by asset management companies(AMC) of the big four banks to improve their loan portfolio qualities. Theestablishment of the AMC parallels with the Resolution Trust Corporation(RTC) in the US during the Savings and Loans crisis in the 1980s. One mainexception, however, is that these AMCs are established by their parent banksrather than by government regulatory bodies. Between December 2001 andFebruary 2006, US$17.5 billion of NPLs were purchased by foreigninvestors, most of which were by foreign banks. The amount even thoughis a small fraction of total NPLs, it represents about 50% of NPLs recoveredby AMCs (see Li, 2006b). These loan purchases not only allow foreign banksto have a foothold in domestic loan market, but it also provides first-handknowledge about the market, detailed information of business customers,and problems associate with the exiting RMB loans.

In another front, the acquisition of domestic banks inadvertently allowsforeign banks to gain access to the loan market. Table 4 shows theacquisitions of major Chinese banks by foreign firms from 2001 to 2006.While the restriction of 20% share ownership of domestic banks limits theextent of the takeovers, the width and depth of foreign banks’ involvementremain significant.7 For example, Citigroup targets 20% of shares in PudongDevelopment Bank and is chosen as the lead underwriter of GuangdongDevelopment Bank. The investment by Citigroup in these two key regional


banks in the most wealthy and vibrant regions in China – the Pearl River andYangzi River delta, indicates a firm commitment in domestic RMB business.

In addition, foreign banks have been successful in originating RMB loansin recent years. Han (2006) reports that in Shanghai alone, the size of the

Table 4. Foreign Investments in Chinese Banks from 2001 to 2006.

PRC Bank Foreign Investor Date Shareholding (%)

Guangdong Development Bank Citibank Nov-06 20.00

Tianjin City Commercial Bank ANZ Jul-06 20.00

Bank of Beijing ING Bank Mar-05 19.90

Bank of Communications HSBC Aug-04 19.90

Bohai Bank Standard Chartered Bank Sep-05 19.90

Hangzhou City Commercial Bank Commonwealth Bank of Australia Apr-05 19.90

Shanghai Rural Commercial Bank ANZ Nov-06 19.90

Nanjing City Commercial Bank BNP Paribas Oct-05 19.20

Shenzhen Development Newbridge Capital Dec-04 17.89

Industrial Bank Hang Seng Apr-04 15.98

Jinan City Commercial Bank Commonwealth Bank of Australia Sep-04 11.00

Bank of China Temasek Aug-05 10.00

Huaxia Bank Deutsche Bank Oct-05 9.90

China Construction Bank Bank of America Jun-05 9.00

Bank of Shanghai HSBC Dec-01 8.00

Bank of Shanghai IFC Dec-01 7.00

Shenzhen Development GE Commercial Finance Sep-05 7.00

Huaxia Bank Pangaea Capital Management Sep-05 6.90

China Construction Bank Temasek Jun-05 6.40

ICBC Goldman Sachs Aug-05 5.80

Bank of Beijing IFC Mar-05 5.00

Bank of China Royal Bank of Scotland Aug-05 5.00

CITIC Bank BBVA Bank of Spain Dec-06 5.00

Nanjing City Commercial Bank IFC Nov-01 5.00

Shanghai Pudong Development Citibank Jan-03 4.62

Minsheng Temasek Jul-04 4.55

Huaxia Bank Sal. Oppenheim Oct-05 4.10

ICBC Allianz Aug-05 3.20

Bank of China Merrill Lynch Aug-05 2.50

Bank of China Li Ka-shing Foundation Aug-05 2.50

Xian City Commercial Bank Bank of Nova Scotia Sep-04 2.50

Xian City Commercial Bank IFC Sep-04 2.50

Bank of China UBS Sep-05 1.70

Minsheng IFC Jul-04 1.08

ICBC American Express Aug-05 0.80

Bank of China Asian Development Bank Oct-05 0.30

Source: PWC, 2006.

Shanghai Securities News, July 12, 2006.

China Business News, November 22, 2006.

China Economic Times, November 23, 2006.

Beijing Business Today, December 11, 2006.


loan portfolio has reached 96.5 billion Yuan in September 2006, a 236 timesof the total amount of loans originated by foreign banks in 1997. Huang(2006b) also documents that foreign banks issue more than 40% of newRMB loans in Shanghai during the first 11 months of 2006. Zhong (2006a)further argues that foreign banks are not only targeting Chinese firms, butthey are also pursuing SOEs in RMB loans, the most important clients todomestic banks.

It is also important to note that not only are domestic banks losingdomestic loans to foreign banks, but they are also losing loans of highquality. In the case of Nanjing Ericsson in 2002, it has decided to switch thelenders from two domestic banks, ICBC and Bank of Communications, toCitibank Shanghai in a 1.99 billion RMB loan before its maturity. In thefollowing year, Citigroup led a group of 18 foreign banks in providing asyndicated loan of 1 billion RMB without collateral to Shanghai Port Co. ata lending rate that was 10% below the PBC’s benchmark rate.8

Dobson and Kashyap (2006) argue that foreign banks’ full access to Chinamay have little impact on the RMB loan market since there are only 2 outof 35 foreign banks currently have more than 40% of their loans inRMB business. According to the PWC (2005) survey, foreign banks haveidentified that credit cards, mortgages, and investment products in retailbanking, and debt markets, credit derivatives/structured products, and riskmanagement in wholesale banking will become more important over the next3 years.

While these non-traditional banking are needed to meet new demands,they do not exclude traditional banking as part of the overall foreign banks’activities. At the time that the PWC survey was conducted, foreign bankswere largely not allowed to issue RMB loans and require special approvalsfrom CBRC. As a result, RMB loans formed a small part of the foreignbanks’ loan portfolios. Under the restrictions however, foreign banks werestill taking part in loan activities indirectly by issuing standby letters ofcredit that assist Chinese firms to obtain loans from local banks. Since 2006,these restrictions have been lifted and foreign banks are free to issue RMBloans directly to Chinese firms.

5. OTHER BANKING ACTIVITIES

Until China’s accession to WTO, Chinese banks found little needs to offerother financial services beyond accepting deposit and issuing loans. In the


highly regulated environment where both deposit and loan rates weredetermined by PBC, Chinese banks were guaranteed a healthy NIM. As aresult, they lack the necessary expertise in modern banking to compete withtheir large foreign counterparts. This view is also shared by the centralgovernment which encourages Chinese and foreign funded banks tocooperate in new financial products, business skills, information exchanges,personnel training, and management skills.9 It implicitly recognizes theshortcomings of domestic banks and directs them to acquire the knowledgeand skills from their international counterparts.

However, there is little incentive for foreign banks to share their expertisein other banking products. By the end of 2006, Citigroup had applied forpatents in China for more than 20 financial products. Even though CBRCofficials refused to grant these patents, Citigroup continues its applicationefforts. It suggests that foreign banks are keen to lock out new marketniches from domestic banks with exclusive rights. Zhu (2006) also reportsthat major foreign banks such as Citibank and HSBC have done theirhomework well in preparation for the Chinese market and are ready tochallenge domestic banks on retail networks, additional products, and newproduct development for local needs.

He (2001) further suggests that the fast expansions of foreign banks inChina will increase the demands for experienced staff from local banks. Theinefficiency of domestic banks with excessive number of staff is likely toexacerbate the brain drain at least in the short run when the banks areforced to downsize to improve productivity. Foreign banks may, therefore,further weaken the competitiveness of domestic banks by attracting talentedstaff from the domestic banks.

6. REGULATORY ADVANTAGES OF

FOREIGN BANKS

China’s current banking regulatory framework is similar to the bankingenvironment under the Glass-Steagall Act (1933) that separates commercialbanking from investment banking. The General Rules over Loans (1996)prohibit investment in equity from proceeds of loans.10 Similarly, theSecurities Law (1999) restricts fund flows from banks to the securitiesmarket.11 In addition, banks are also not allowed to invest in the securitiesmarket other than government bonds. Because banks, securities firms, andinsurance companies have been confined to their core activities, little


expertise and experiences do they possess over a wide range of financialservices compared to their foreign counterparts.

Foreign banks are also favored to compete in the domestic depositmarkets under the current regulations. They are only required to maintain amaximum loan to deposit ratio of 75% compared to the ratio of 60% forChinese banks. Furthermore, foreign banks, with a current average loan todeposit ratio of 200%, are granted a grace period of 5 years to meet therequirement. Therefore, foreign banks could compete in the loan marketwith the advantage of smaller amount of deposits over the next 5 years. Inthe mean time, with the intent of establishing retail network through newbranches and bank acquisitions, superior customer service, wider array offinancial services, and favorable regulatory environment, foreign banks arepositioned to compete and take away core deposits in the Chinese market inthe next several years.

Evidence that only foreign banks are allowed to participate in domestic Ashare market also highlights another regulatory advantage over the domesticbanks. After the stock market crash in 2001, China Securities RegulatoryCommission (CSRC) introduces the Qualified Financial InstitutionalInvestors (QFII) scheme in 2002 that encourages foreign funds to investand shore up the stock market. By the end of 2006, more than US$9 billionhave been allocated to foreign investors.12 Foreign banks have since takenthis opportunity and become the major investors in the domestic sharemarket. Table 5 shows that foreign banks are highly active in the market. Xu(2004) reports some foreign banks such as UBS and Morgan Stanley areamong the largest QFII shareholders in A share market. Although the quotafor QFII is relatively small compared to the US$64 billion (or 500 billionRMB) domestic investment funds in the stock market, QFII have been themarket movers. For example, Cao (2006) finds that a recent UBS report thatlowers the ratings of domestic bank shares listed in Hong Kong has broughtpanics to the domestic share market.

In another regulation, Qualified Domestic Institutional Investors (QDII,2006), domestic institutions and Chinese residents are allowed to investoffshore through domestic banks.13 Since domestic banks have littleknowledge or experience in overseas investments, they have no alternativesbut to work with foreign banks in the QDII investments. Furthermore, Li(2006a) reports that foreign banks in China are subsequently eligible tobecome QDII. Therefore, domestic investors could also directly turn toforeign banks. In November 2006, National Securities Fund with 230 billionRMB (or about US$30 billion) has chosen Northern Trust and Citigroup asits consignment banks for overseas investments.


The notion of financial innovations has been new to Chinese banks.Banks are restricted to their traditional activities and are often regarded asconduits of central government’s social policy and stability. Heilmann(2005) argues that the creation of Central Financial Work Commission

Table 5. QFII Investments in China A Share Marketon September 30, 2006.

Sequence Institution Name Bank Account with Companies Invested

1 Citibank Standard Chartered,

Shanghai

58

2 Credit Suisse Citibank, Shanghai 36

3 Morgan Stanley HSBC, HK 29

4 Merrill Lynch International HKBC, Shanghai 23

5 HSBC CCB 18

6 Nikko Asset Management

Co., Ltd.

Bank of Communications 13

7 Standard Chartered BOC 10

8 Lehman Brother ABC 9

9 Nomura Securities Citibank, Shanghai 8

10 Bill & Melinda Gates

Foundation

HSBC, HK 7

11 Deutsche Bank Citibank, Shanghai 6

12 SMBC ICBC 4

13 Barclays Bank Barclays Bank PLC 2

14 Hang Seng Bank CCB 2

15 Goldman Sachs HSBC, Shanghai 1

16 BNP Paribas ABC 1

17 UBSWarburg Citibank, Shanghai 1

18 Fortis Bank SA/NV BOC 0

19 ING Standard Chartered,

Shanghai

0

20 Dresdner Bank AG ICBC 0

21 ABN AMRO HSBC, Shanghai 0

22 SG HSBC, Shanghai 0

23 Banque Indosuez N/A 0

24 Templeton Asset

Management Ltd.

HSBC, Shanghai 0

25 JPMorgan Chase Bank,

JPM

HSBC, Shanghai 0

26 Power Corporation of

Canada

N/A 0

27 INVESCO Asset

Management Limited

BOC 0

Source: JRJ.com


(CFWC), a proxy for the central government, was to preserve thehierarchies in the finance industry and to restore central policy decisiveness.As such, senior executives of domestic banks are appointed by centralauthority. Since these executives are often former government officials andhave little experience in banking, they are less likely to meet customer needsin a fast changing business environment, to improve efficiency in fundallocation, and to conform to the emerging corporate governance. The lackof financial innovation and services of Chinese banks is clearly evidenced inthe composition of their incomes. Only 10% of the total banks’ incomes, onaverage, are non-interest incomes compared to 50% of the total incomes oftheir foreign counterparts (see Liu, 2006b).

CWFC was subsequently dissolved, replaced by CBRC whose creation isto lay the foundations for a nationwide market regulation and dismantle theold socialist institutions in China. However, since the establishment of CBRCin 2003, the only notable ‘‘innovation’’ by domestic banks has been to attractmore deposits and collect fees for various utilities companies and governmentagencies. It appears that while domestic banks are encouraged to beinnovative, restrictive bank regulations remain. In a recent case where thedemands on RMB NDF products are strong due to market’s expectation onRMB appreciation, domestic banks are banned from participating in themarket. However, foreign banks such as HSBC and Standard CharteredBank, the primary market makers of the products, are not subject to the sameregulations. Similarly, domestic banks attempt to issue securitized mortgageshave also been prohibited under the existing rules (see Qiao, 2005).

Leung and Chan (2006) argue that foreign banks in China cannot be surewinners after WTO because they lack network infrastructure, unfamiliarwith local business customers, and suffer from regulatory restrictions.However, all the anecdotal evidence discussed in Sections 3–6 suggestotherwise. Their argument relies much on the data until 2004. However, thepreparations of foreign banks in China for the post WTO have since beenmore complete and in depth than they have indicated.

7. THE IMPACT OF FOREIGN COMPETITION ON

THE SAFETY AND SOUNDNESS OF CHINESE BANKS

As discussed in earlier sections, foreign banks have shown the interests andcapabilities to compete with domestic banks in a wide array of financialservices including the retail RMB business. We have little doubt that foreign


banks are positioned to take a significant market share in the local business.He (2006) and Zhong (2006b) estimate that foreign banks will eventua-lly capture 30% of market shares from domestic banks in the traditionalbanking business. The bigger question that we would like to address is howthe loss of lucrative RMB business will affect the financial health of Chinesebanks? More specifically, could losses of deposits and high quality RMB loansthat provide the source of low cost funding and reliable returns lead to a furtherdeterioration in the quality of assets and earnings? Furthermore, since Chinesebanks have lower capital requirement and adequacy, could the deteriorationin earnings lead to a banking crisis? The following subsections examine theeffect of foreign competition on NPLs and capital adequacy of Chinese banks.

7.1. Non-Performing Loans

It has been well documented that Chinese banks are ridden with large NPLs.Zhang (2006) reports that the average NPL ratio among Chinese banks rangesfrom 20 to 25% of total loans or amounts to 4–5 trillion RMB by 2005.Li (2006b) estimates that the balance of NPLs from domestic banks should bemore than 3,590 billion RMB. The extent of the bad debt incurred bydomestic banks can also be measured by the transfer of 2,277 billion RMB inNPLs from the big four banks to their affiliated Asset ManagementCorporations (AMC), which are established to dispose the bad debts.

Even the official figures released by CBRC suggest that NPLs remain asignificant challenge to the commercial banks and the government. Table 6reports some summary statistics from CBRC on NPLs among Chinese banksfrom 2003 to 2006. In aggregate terms, NPLs have varied from 2.44 trillionRMB in 2003 to 1.31 trillion RMB in 2006. The sharp decline in NPLs overthe last 3 years, however, is a response to the government policy of loweringNPL ratios by 3–5% a year rather than an improvement in the loan quality.Chinese banks achieve their NPL targets by repackaging existing NPLs withnew loans that are likely to become future NPLs. Therefore, the currentdeclines in the NPL ratios do not reflect a true reduction on bad debts butrather a window dressing tactic that hides the current NPL problems. Whenstandardize the NPLs as a percentage of total loans, the average current NPLratio of Chinese banks is about 10 times as large as that of foreign banks.Table 6 shows that the average NPL ratios are 7.1 and 0.78% for Chinesebanks and foreign banks, respectively. In sum, it appears that the centralgovernment will have to continue to off-load trillions of NPLs and injectbillions of US dollar reserves into these domestic banks.


Table 6. NPLs of Chinese Banks from 2003 to 2006.

2003 2004 2005 2006

Balance

(RMB

Billion)

Share in

Total Loans

(%)

Balance

(RMB

Billion)

Share in

Total Loans

(%)

Balance

(RMB

Billion)

Share in

Total Loans

(%)

Balance

(RMB

Billion)

Share in

Total Loans

(%)

NPLs Classification 2440.6 17.8 n/a n/a 1313.4 8.61 1254.9 7.09

Sub-standard 334.2 2.5 307.5 2.36 333.6 2.19 267.5 1.51

Doubtful 1431.6 10.4 889.9 6.84 499.0 3.27 518.9 2.93

Loss 674.7 4.9 520.2 4.00 480.7 3.15 468.5 2.65

By Institutions

Major Commercial

Banks

n/a n/a 1717.6 13.21 1219.7 8.90 1170.3 7.51

SOCB n/a n/a 1575.1 15.57 1072.5 10.49 1053.5 9.22

Joint-Stock n/a n/a 142.5 4.94 147.2 4.22 116.8 2.81

City Commercial

Banks

n/a n/a n/a n/a 84.2 7.73 65.5 4.78

Rural Commercial

Banks

n/a n/a n/a n/a 5.7 6.03 15.4 5.90

Foreign Banks n/a n/a n/a n/a 3.8 1.05 3.8 0.78

Note: 2003 data cover SOCBs and policy banks. 2004, 2005, and 2006 data cover all commercial banks excluding policy banks.

Source: China Banking Regulatory Commission, 2003–2006.

The commercial banks include the state-owned commercial banks (SOCBs), joint-stock commercial banks, city commercial banks, rural

commercial banks, and foreign banks. The major commercial banks include the SOCBs and the joint stock commercial banks (JSCBs).

As the number of rural commercial banks and city commercial banks has increased in 2006, this year’s figures of these institutions are

incomparable with that of 2005.

On

the

Sa

fetya

nd

So

un

dn

esso

fC

hin

eseB

an

ks

463

Despite the prevailing large interest rate gap in China, Chinese banks havethe lowest NIMs among international banks largely due to large bad debts.Xie and Chen (2001) argue that Chinese loan risks have not been fully pricedespecially on those of state-owned enterprises (SOEs) because they play aspecial role in providing employment and social stability. The state-ownedcommercial banks are, therefore, encouraged to provide unmonitoredamount of loans to SOEs without sufficient credit worthiness. As a result,the big four banks are the SOEs’ biggest lenders but also incur heaviest SOElosses. Li (2006b) suggests that state ownership of commercial banks is theroot of NPLs and government regular bailout of NPLs is simply delaying thebanking crisis since there is no fundamental change in addressing the cause ofNPLs. As long as the governments at different levels play a central role indistributing financial resources and Chinese banks continue to behave asconduits of government’s social policy, we would likely see NPLs worsen.

With the current NPLs faced by Chinese banks and the governmentpractises in directing bank loans to SOEs, it is envisaged that the magnitudeof losses in domestic banks especially the big four banks will likely to growwhen foreign banks compete in the local markets. Furthermore, domesticbanks rely heavily on RMB deposits to provide liquidity and low cost sourceof funding. As shown in Table 7, 82.67–93.33% of the big four banks’ totalliabilities in 2005 were in the form of deposits compared to 0.71–0.93% inlong-term debt. The importance of deposits in generating income is alsoillustrated in the banks’ income composition, of which 79–92% are from netinterest income. If foreign banks manage to capture 30% of local RMBbusiness as predicted by He (2006) and Zhong (2006b), and Chinese banksare required to maintain the loan to deposit ratio below 75%, there will beless interest income for Chinese banks to offset the NPLs.

The quality of RMB loans among Chinese banks may also continue todeteriorate. Of the total assets in major Chinese banks, 47–59% are loans.As foreign banks increases their market shares in RMB loans throughacquisition of some NPLs, competing high quality loans with their Chinesecounterparts, and originating new loans, the outlook for Chinese banks andthe central government to tackle the NPL problems is not promising fromthe combining effects of losing market shares in both deposits and loans.

7.2. Capital Inadequacy

As NPLs worsen among Chinese banks due to increased foreign bankcompetition in RMB business, Chinese banks will face even greater challenge


to meet capital adequacy and requirement. In Section 3, we compare some ofthe key financial ratios reported in Table 3 on returns, profitability, CAR,and size between Chinese and US banks before foreign banks are allowed toconduct RMB business. Between them, US banks appear to be well capita-lized with an average CAR ratio of 11.6% compared to 9.1% for Chinesebanks. It is interesting to note that the current ratio of 9.1% among majorChinese banks is only achieved recently after the 8% CAR requirementimposed by the central government. The sharp increases in CAR ratios,

Table 7. Some Financial Ratios of Chinese Big Four Banks in 2005.

ICBC (RMB

Millions)

ABC (RMB

Millions)

BOC (RMB

Millions)

CCB (RMB

Millions)

Total assets 6,454,106 4,771,019 4,742,806 4,585,742

Interest earning assets 5,686,908 4,575,424 4,458,844 4,575,424

Loans 3,289,553 2,829,291 2,235,046 2,395,313

Loans/total asset 50.97% 59.30% 47.12% 52.23%

Securities 2,052,648 1,257,059a 1,562,320 1,413,871

Securities/total asset 31.80% 26.35% 32.94% 30.83%

Total liabilities 6,196,625 4,691,412 4,480,186 4,298,065

Deposits 5,660,462 4,036,854 3,703,777 4,006,046

Certificate of deposits n/a n/a n/a 5,429

Deposits/total Liabilities 91.35% 86.05% 82.67% 93.33%

Bonds 43,780 n/a 60,179 39,907

Bonds/total Liabilities 0.71% n/a 1.34% 0.93%

Operating incomes 150,551 55,495 116,028 128,714

Interest income 224,457 105,133 167,345 173,601

Interest expense 86,599 61,402 66,940 57,050

Net interest income 137,858 43,731 100,405 116,551

Net interest income/operating incomes 91.57% 78.80% 86.54% 90.55%

Net fee and comm 10,546 9,146 9,247 8,455

Net fee and comm/oper income 7.00% 16.48% 7.97% 6.57%

Investment income 4,016 23,282b –248 2,382

Other net operating income 340 2,618 2,412 2,086

Operating profit 89,258 10,943 64,744 128,714

Net profit after tax 33,704 7,878 27,492 47,096

Note: Interest Earning Assets include loans, securities investments, due from and placement

with other financial institutions, due from central bank.

Source: 2005 Annual Reports of Industrial and Commercial Bank of China (ICBC), Agriculture

Bank of China (ABC), Bank of China (BOC), and China Construction Bank (CCB).aABC’s investments include financial bonds, treasury bonds and other.bNo explanation on constituents or sources of investment incomes.


however, mask the true picture of Chinese banks’ capital adequacy. One waythat the big four banks boosts their CARs is by holding each other’ssubordinate bonds as ‘‘real’’ capital. Therefore, the actual CAR ratios forChinese banks may be lower than those indicated in Table 3. The equity toasset ratio provides some insight on the degree of leverage between Chineseand US banks. The average equity to asset ratio of the US banks is 9.27%,more than twice among Chinese banks.

The eight publicly listed Chinese banks in Table 3 that we use to drawcomparisons are also some of the better performing banks in China. Forexample, BOC and CCB are two of the better big four banks while ChinaMerchants Bank (CMB), MinSheng Bank, Pudong Development Bank, andBank of Communications are also some of the best joint-stock banks inChina. Therefore, a typical Chinese bank will have lower CAR and equity toasset ratio. These financial ratio indicators therefore suggest that Chinesebanks have relatively little capital to absorb losses from the NPLs. Withpoor asset management reflected in their low ROAs and their unique bankroles in providing employment, foreign competition may increase thelikelihood of bank failures even when bailouts from the central governmentare a common occurrence.

8. POLICY IMPLICATIONS AND

RECOMMENDATIONS

The lack of expertise in modern banking, the regulatory disadvantages andrestrictions, and the social roles of providing employment as discussed inthis chapter may lead to large bank failures after foreign banks are allowedto compete directly with domestic banks. To reduce the likelihood of abanking crisis in China, we suggest some changes in the current bankingpolicies that may address the problems associated with domestic banks.These proposed changes can be classified into two types – on thefundamental roles of Chinese banks and on leveling the playing field.

First and foremost, domestic banks should not be a conduit of government’spolitical and social policies. If domestic banks continue to be used for theSOE reforms, they are unlikely to be financially viable since large NPLs willpersist. To compete with foreign banks, Chinese banks need to be able toperform the same basic function as a bank. That is, to allocate financialresources to the most productive borrowers. By creating a ‘‘special’’ statefunded bank whose sole purpose is to provide credits to the SOEs while


the reforms continue, it frees the rest of the domestic banks into goodcommercial banking practices. Such separation of banks eliminates theZombie effect (Kane, 1989) that currently plagues the entire bankingindustry where banks are most willing to extend credits to SOEs in antici-pating for government bailouts.

In line with the creation of the new state funded bank, domestic banksshould also be encouraged to seek external capital outside of the government.This could be achieved by issuing stocks and long-term bonds as a publiclylisted firm. These measures will increase genuine equity capital and boostcapital adequacy ratio.

For the immediate term, several regulatory restrictions on domestic banksneed to be removed to provide an equal level of playing field. First, state-owned banks should be allowed to acquire foreign banks to gain experienceand expertise in modern banking activities. CBRC may also alert domesticbanks about new products that foreign banks are planning to introduce inChina. It would allow domestic banks more lead time to compete in the newproduct markets. Furthermore, similar schemes as with the QFII quotas forforeign banks could be introduced to encourage domestic banks into sharemarkets. We also suggest that unfavorable tax treatment and loan-to-deposit requirement on domestic banks be removed immediately. With thelarge presence of foreign banks in China, there is little need to attract moreforeign banks with the current incentives.

Finally, to improve management efficiency and bank productivity, bankexecutives should not be appointed based on their political connections orpast bureaucratic positions. The current management appears to have littleexperience in banking and tends to make its decisions based on politicalgrounds. At the same time, domestic banks need to stop the brain drain ontheir skilled managers by foreign competitors with better incentives andcareer path.

9. CONCLUSIONS

The liberalization of banking markets in a transition or emerging economybrings two benefits: Attracting new capital and restructuring inefficientbanking system. In the case of China, the former is not a primary motivationsince China has been receiving more foreign direct investment than any othercountries in recent years. The focus, therefore, is on improving China’sbanking system and efficiency. Lehner and Schnitzer (2006) develop a model


of spatial bank competition which suggests that an increasing number ofbanks from foreign bank entry tend to have some positive welfare effectsespecially in a competitive market environment. However, spillover effectsfrom foreign bank entry may counteract the positive effects from theincreased competition which in turn could result in an overall decline in thehost country’s social welfare.

We argue that in the current Chinese banking environment, the costs ofspillovers outweigh the benefits of increased competition. Our argument restswith foreign banks’ interests and demonstrated capabilities in competingwith Chinese banks in the lucrative RMB business. Since Chinese banks alsolack the expertise and experience in modern banking and financialinnovations, they are forced to compete narrowly among themselves in thetraditional business while foreign banks participate fully in both types ofbanking activities. Taking full advantage of new products and marketleadership by foreign banks, some domestic banks may be pushed out of themarket in the coming years.

Domestic banks’ inability to compete with foreign banks may continuelong after the foreign bank entry. Chinese banks, especially the big fourbanks, are burdened with large NPLs, low capital adequacy, lack of expertise,unequal treatments in taxation, and limited banking services. RMB busi-nesses have been the last resort for most Chinese banks to survive. Theimpact of foreign banks presence in China may, therefore, be significant inthe form of a banking crisis as cash flows in the form of RMB deposits andloans are transferred from Chinese banks to foreign banks.

We propose two types of policy changes that may address the inherentproblems and competitiveness of Chinese banks. First, it is important thatthe fundamental roles of Chinese banks be changed from a socially orientedprovider to a credit-based intermediary. The creation of a state fundedbank to meet the needs of the social reform may free the rest of domesticfor the latter roles. Second, Chinese banks need to compete with foreignbanks without unfavorable regulations and restricted scope of bankingactivities. The removal of these barriers will enhance the competitivenessof the banks and allow them to acquire skills and expertise in modernbanking.

NOTES

1. The big four banks include Industrial and Commercial Bank of China,Agricultural Bank of China, Bank of China, and China Construction Bank.


2. Since no reliable net interest margins (NIMs) are available from the Chinesebanks, we are forced to draw the comparisons using the official 1-year interestrate gap.3. PBC RMB Interest Rate Administration Rules, April 1, 1999 and October 28,

2004.4. The only exception is that commercial banks can provide higher rates to

domestic insurance companies with term deposits of more than 30 million RMB andof maturities of more than 5 years.5. CBRC was established on April 28, 2003 according to State Council’s (2003)

No. 8 Document (March 21, 2003). The People’s Congress Standing Committee laterapproved CBRC to take the banking supervision role over from PBC.6. Term deposits from businesses and all demand deposits are not restricted to a

minimum of 1 million RMB.7. CBRC (2003) No. 6 Decree, Administrative Rules Over Foreign Financial

Institutions’ Investment in Domestic Financial Institutions, issued on December 8,2003 and effective from December 31, 2003, Article 8.8. There are no penalties for borrowers to retire early from debts before maturity

in China.9. The 11th Five-Year Plan of Exploitation of Foreign Investments, November

9th, 2006, National Development & Reform Commission, National Development &Reform Commission.10. General Rules over Loans, Article 20, June 28, 1996, PBC.11. The Securities Law, Article 133, July 1, 1999, CBRC.12. China Insurance News, QFII Approved Quota Has Broken Through 9 Billion

US Dollars, December 27, 2006.13. CBRC (2006) No. 121 Document, Temporary Administration Rules over

Commercial Banks’ Offshore Wealth Management on Behalf of Clients, JointlyIssued by PBC, CBRC, and SAFE, April 17, 2006.

REFERENCES

Cao, Z. M. (2006). The lost pricing rights, http://www.jrj.com, December 4th.

Chen, Z. M. (2006). Comparison of four key indicators of commercial banks in US and China.

China Securities Journal (October 26).

Dobson, W., & Kashyap, A. (2006). The contradiction in China’s gradualist banking reforms.

Brookings Panel on Economic Activity.

Han, S. H. (2006). Foreign banks’ Shanghai speed: Assets increased 1.5 times during past

5 years. China Business News (December 11).

He, J. B. (2006). Domestic or foreign banks: Which is better for deposits? China Business Post

(November 18).

He, L. P. (2001). The challenge of the WTO accession on China’s banking industry.

International Economic Review, 2, 29–32.

He, L. P. (2006). Opening the banking sector and its impacts over China’s finance, bank decision

makers forum. Beijing, December 8th.

Heilmann, S. (2005). Regulatory innovation by Leninist means: Communist party supervision

in China’s financial industry. The China Quarterly, 181, 1–21.


Huang, L. (2006a). Foreign banks actively contacting domestic clients and focusing on network

expansion next year. Securities Times (November 20).

Huang, T. J. (2006b). Shanghai: Foreign banks takes forty percent of new RMB loans.

XinHua.net.com. December 8th.

Kane, E. J. (1989). The high cost of incomplete funding the FSLIC shortage of explicit capital.

Journal of Economic Perspective, 3, 31–47.

Lehner, M., & Schnitzer, M. (2006). Entry of foreign banks and their impact on host countries.

CEPR, Discussion Paper no. 5954.

Leung, M. K., & Chan, R. (2006). Are foreign banks sure winners in Post-WTO China?

Business Horizons, 49, 221–234.

Li, Y. (2006a). CBRC official explains foreign banks with China branches may apply for QDII.

China Securities Journal (May 31).

Li, Z. Y. (2006b). Non-performing loan of Chinese banks up to 3 Trillion Yuan. China Review

(8th ed.). pp. 28–29.

Liang, Y., & Yu, X. (2006). Factors affecting deposit market shares of commercial banks.

Financial Times (13th ed.).

Liu, M. K. (2005). Insiders and outsiders’ conspiracy to commit most serious crimes at China’s

banks. Beijing Youth Daily (April 9).

Liu, M. K. (2006a). Liu Mingkang gives orders in Northeast Provinces: Firmly prevent large

amount of crimes in banks. Shanghai Securities News (April 3).

Liu, M. K. (2006b). Speech on China financial derivatives forum. China Banking Regulation

Commission (October 24).

Ma, G. (2006). Sharing China’s bank restructuring bill. China and World Economy, 14,

19–37.

National Bureau of Statistics. (2002a). Survey shows services at domestic banks need to be

improved. National Bureau of Statistics Website, September 5th.

National Bureau of Statistics. (2002b). Majority of interviewed disagree with fees at domestic

banks. National Bureau of Statistics Website, September 25th.

PriceWaterhouseCoopers. (2005). Foreign banks in China. Beijing: PriceWaterhouseCoopers.

PriceWaterhouseCoopers. (2006). Going for growth – The outlook for M&A in the financial

services sector in Asia. PriceWaterhouseCoopers, Beijing.

Qiao, A. (2005). Securitization in China: Secured or not. International Financial Law Review, 24,

51–53.

Shi, C. G. (2006). Foreign banks rushing to build in China and RMB businesses will become

major battle field. China Securities Journal (November 17).

Xie, P., & Chen, R. (2001). The causes and resolutions of non-performing loans in state-owned

banks. PBC Research Institution Paper, June 17th.

Xu, J. M. (2004). QFII starts to invest heavy, taking lead in volatile domestic share market.

Securities Daily (September 2).

Zhang, P. J. (2006). NPL ratios of Chinese banks are still very high. Beijing Business Today

(September 4).

Zhong, H. (2006a). Foreign banks borrow RMB to lend and grab quality loan customers.

Capital Week (November 18).

Zhong, T. (2006b). Foreign banks will take 30% of market shares in China. Securities Time

(September 28).

Zhu, M. (2006). Chinese banks face with intellectual rights crisis. Bank Decision Makers Forum,

Beijing, December 8th.


CHAPTER 22

MARKET DISCIPLINE BY CD

HOLDERS: EVIDENCE FROM

JAPAN WITH A COMPARISON

TO THE US$

Ayami Kobayashi

ABSTRACT

Certificates of deposit (CDs) are uninsured deposits that have not been

protected by the Japan Deposit Insurance Corporation (DIC) since the

beginning of the issuance in May 1979. Thus, CDs should reflect

exceedingly well banks’ failure probabilities and the risk perception of

market participants among many types of depositors in Japan. Because

of this, CDs issued by Japanese banks may enhance the market discipline of

banking organizations. This is the first chapter to test the depositor

discipline hypothesis using Japanese bank data from the financial year 1998

to the financial year 2003 . The chapter develops reduced-form models that

describe how interest rates and the quantity of CDs may be related to

banks’ financial measures. Among the Japanese CAMEL ratings, the

chapter finds that CD interest rates are sensitive to the capital adequacy

$Any opinion, findings, and conclusions of recommendations expressed in this study are the

sole responsibility of the author and do not necessarily reflect the views of Tokai Tokyo

Research Center Co., Ltd., Japan.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00022-2

471

dx.doi.org/10.1016/S1569-3767(07)00022-2.3d

ratio (CAR) and that CD quantities are sensitive to ROA. The chapter

also insists that CD holders in Japan are sensitive to bank risks and

exercise disciplinary power to impose market discipline that compliments

regulatory discipline.

1. INTRODUCTION

According to Berger (1991), market discipline in the banking sector can bedescribed as a situation in which bank stakeholders face increasing costs asbanks undertake risks, and take action on the basis of these cost.

The utilization of subordinated debt yield spread and the rates ofcertificates of deposit (CDs) as means of imposing market discipline hasbeen discussed as a way to complement regulatory discipline since the 1980sin the USA. Since the beginning of the 1980s, there have been arguments inthe USA and European countries over whether bank holding companies(BHCs) or banks themselves should use CDs to enhance market discipline.To this end, a number of studies have analyzed the relationship between CDinterest rates, CD quantity, and bank-specific risks for US bankingorganizations and have found evidence suggesting that issuing CDs has apositive effect on market discipline. Up to now there have been no empiricalstudies regarding the effectiveness of the issuing of CDs by Japanese banksas a way to increase the market discipline of Japanese banks. Further, usingdata from Japanese banks, I (Kobayashi, 2003) found no evidence ofsubordinated debt holders imposing market discipline on banks. Thus, inthis study, I analyze whether there is a relationship between CD interestrates, CD quantity, and bank-specific risks in the Japanese CD market.

This chapter is organized as follows. In the second section, the reasonswhy depositors are needed to share the role of disciplining banks areexplained. In the third section, the previous literature related to the depositordiscipline hypothesis is reviewed. The fourth section is an empirical analysisin which the data are described, the hypotheses and reduced-form model arestated, and the variables utilized in this study are discussed. In the fifthsection, the empirical results for the model are reported. Finally, in the sixthsection, conclusions are drawn and policy implications are discussed.

2. HOW DEPOSITORS DISCIPLINE IS SIGNIFICANT?

Certificates of deposit are defined as unsecured deposits issued by banks andnot fully guaranteed by the Deposit Insurance Corporation (DIC) of Japan.

AYAMI KOBAYASHI472

Thus, if a failed bank is liquidated, CD holders would receive payment onlyif all insured depositors are paid in full.

By definition how depositors discipline banks is as follows: all depositorsmay penalize riskier banks by requiring higher interest rates or bywithdrawing their deposits because they are exposed to bank risk-taking,but those who hold demand deposits or CDs would be especially prone tosuch behavior.

In theory, insured depositors should be insensitive to bank risk becausethey are fully covered, and uninsured depositors would be the primarymonitors of banks because they are exposed to banks’ risk-taking behaviorand can lose their deposits above the deposit-insurance ceiling when banksfail. However, empirical studies have found that insured depositors areresponsive to the financial conditions of banks as well, suggesting thatinsured depositors are concerned about the solvency of banks. For example,Dewatripont and Tirole (1994) found that numerous small depositorscannot effectively identify or control bank risks because of informationcosts and coordination problems. On the other hand, Martinez-Peria andSchmukler (2001) found that even insured depositors were able to disciplinebanks in developing countries such as Argentina, Chile, and Mexico duringthe 1980s and 1990s. Cook and Spellman (1994) also found that Savings andLoan Association (S&L) offering rates on small-CDs generally rise and falltogether with banks’ financial conditions.1 Likewise, Park and Peristiani(1998) found that insured depositors at riskier thrifts are uneasy aboutfinancial risk as well, but the evidence for market discipline being imposedby insured depositors is less profound than that for uninsured depositors.Moreover, Kane (1987) found that small depositors have been able todiscriminate between solvent and insolvent depository institutions even incrises. In short, depositors are concerned not only about the solvency ofindividual banks but also about the solvency of the DIC and the willingnessof the government to support the insurer (Flannery, 1998), and nodepositors perceive insured deposits as being perfectly safe. Thus, there ispotential for depositors as a whole, in addition to government regulators, todiscipline banks.

There are two advantages for banks in adopting the practice of issuingCDs: (1) to provide market signals about troubled banks’ level of risk on adaily basis, signals that could be missed by surveillance tools based onquarterly, semi-annual, or annual financial statements; and (2) to reducemoral hazard incentives. First, market participants such as CD holders mayinterpret a rise in interest rates in the secondary market for CDs as a signalof an increased risk to the issuing banks if the market is rational. Likewise,

Market Discipline by CD Holders 473

the price of a firm’s traded securities is the most obvious public signal bywhich the evaluations of bank stakeholders may enforce management toimprove efficiency by pressuring some of the relatively inefficient banks tobecome more efficient or to exit the banking industry. Moreover, if theregulators cede greater control to market forces in the sense that moretimely or accurate market signals would reflect the emerging problems ofbanks before conventional surveillance tools do, the cost to regulators ofsupervising banks could decrease. Second, market signals may reduce themoral hazard incentives of banks undertaking excessive risks that arecreated by deposit insurance guarantees. Adopting the practice of utilizingCDs would provide ex ante incentives for banks to reduce their risk becauseCD holders are more risk-averse in that they may lose their funds whenbanks fail. Therefore, utilizing CDs would compel banks to disclose theircurrent financial condition and prospects to the market, thereby refreshingsecondary market prices and enhancing market mechanisms. In otherwords, CD holders with their own funds at risk would face strong incentivesto invest in information regarding the true characteristics of banks’portfolios and prospects.

In summary, all depositors may penalize riskier banks by requiring higherinterest rates or by withdrawing their deposits because they are exposed tobank risk-taking, but those who hold demand deposits or CDs would beespecially prone to such behavior. Thus, there are several advantages forbanks to utilize CDs in order to enhance the efficiency of marketmechanisms’ influence on Japanese banks.


Most of the previous studies support the depositor discipline hypothesis.The depositor discipline hypothesis is as follows: If there were norelationship between interest rates paid on CDs or the quantity of CDsand bank-specific risks, these estimates would not be affected by bank risks,implying no market discipline.

A common place to find evidence of enhanced depositor discipline is inthe market for Jumbo CDs because they are time deposits with balancesabove the deposit-insurance ceiling (e.g., the $100,000 deposit-insuranceceiling in the USA and the f10,000,000 ceiling in Japan). The studies employcross-sectional and/or time-series regressions of Jumbo-CD variables on avariety of bank risk proxies (e.g., CAMELS or BOPEC ratings) with marketrisk factors as control variables.2

AYAMI KOBAYASHI474

Most of the previous empirical studies analyze US banks or BHCs.Analyses are divided into three categories: the interest rate effect, thequantity effect, and an interaction effect.

First, with regard to the interest rate effect, a number of studies haveinvestigated whether the interest rates paid on uninsured Jumbo CDs issuedby US banks reflect bank default risk or measures of market risk. They findevidence that uninsured deposits respond to bank risk captured by balancesheets and some measures of market risk (Brewer & Mondschen, 1994;Cargill, 1989; Cook & Spellman, 1994; Goldberg & Lloyd-Davies, 1985;Hannan & Hanweck, 1988; Herzig-Marx & Weaver, 1979; Keely, 1990).Although most studies regress CD interest rates on the financial data ofbanks, Baer and Brewer (1986) and James (1988, 1990) utilized measures ofmarket risk in lieu of some accounting ratios; they found a significantpositive effect of equity volatility on CD interest rates. Also, Ellis andFlannery (1992) estimated time series models for a daily CD risk premium asa function of innovations in the stock return of banks during the sampleperiod, and they found evidence of market discipline.

However, there are two exceptions among the majority of studies. Crane(1976) concluded that in the 1974 CD market, traditional bank riskmeasures were less important than other factors such as location, eventhough he finds that bank risk influenced CD rates positively. In summary,it is concluded that riskier banks are forced to pay higher interest rates onJumbo CDs.

Second, with regard to the quantity effect, most previous studiesempirically examine the relationship between quantities of Jumbo CDs andbank-specific risks. For example, Goldberg and Hudgins (1996, 2002) andCalomiris and Wilson (2004) found that uninsured deposit growth falls asbank risk-taking increases. Moreover, Billet, Garfinkel, and O’Neal (1998)examined the relationship between changes in bank credit risk rated byMoody’s and the CD quantity of insured deposits, and they found evidencethat risky banks increase their use of insured deposits following adowngrade. They argued that when banks become riskier, they shift theuse of uninsured deposits to insured deposits in order to reduce the cost ofuninsured deposits. Further, Jagtiani and Lemieux (2000) found that banksincrease their reliance on insured deposits when facing financial difficulties,even though they argued that there is little evidence of market discipline onbanks in the uninsured CD market. Finally, Crabbe and Post (1994) foundno evidence of market discipline by depositors. In conclusion, it is assumedthat the quantity of CDs and deposits are moderately sensitive to bank-specific risks.


Third, with regard to the interaction effect, most studies estimate theeffects of bank-specific risks on the equilibrium interest rate and theequilibrium quantity resulting from the interaction between the banks’demand and the depositors’ supply of deposits/CDs. A common metho-dology for testing the interaction effect uses reduced-form equations;parameter estimates of the interaction effect by employing reduced-formequations are different from those used in unilateral analysis of the interest-rate effect and the quantity effect. For example, Park (1995), Park andPeristiani (1998), Jordan (2000), Martinez-Peria and Schmukler (2001)empirically examined the interaction effect, and they found evidence that theinterest rates paid on Jumbo CDs and the quantity of Jumbo CDs respondto changes in bank-specific risk. Particularly, Gilbert, Meyer, and Vaughan(2003) and Hall, King, Meyer, and Vaughan (2003) focused on pre-FDICIA(Federal Deposit Insurance Corporation Improvement Act) and post-FDICIA samples. They found that interest rates paid on uninsured CDsand the quantity of uninsured CDs are sensitive to bank-specific risksduring a financial crisis. Interestingly, they found evidence that the coeffi-cients in both periods are not statistically or economically different acrossthe two sample periods. This finding implies that the enforcement ofFDICIA in 1991 did not affect market discipline in the CD market.3

Using cross-country data over 1990–1997, Demirguc--Kunt and Huizinga(2004) empirically examined whether bank interest rates and the growthrates of bank deposits to bank-specific risk are affected by depositinsurance. They found that riskier banks pay higher interest rates, whereasthey found no evidence that higher or lower deposit growth is affectedby deposit insurance. In summary, interest rates paid on deposits/CDsare sensitive to bank-specific risks, whereas the findings are unclearwith respect to sensitivity of the quantity of deposits/CDs to bank-specificrisks.

In conclusion, the studies examining the interest rates effect, thequantity effect, and an interaction effect find evidence of a relationshipbetween the interest rates paid on CDs/deposits, the quantity of CD/deposits, and bank-specific risk factors. This finding suggests thatdepositors’ discipline should result in higher interest rates on CDs andsmaller quantities of CDs with an increase in a bank’s risk profile.In addition, it is assumed that insured deposits are less sensitive to bankrisks than uninsured deposits are. Overall, all depositors, but especiallyholders of Jumbo-CDs, have disciplinary power to impose marketdiscipline on banks by withdrawing their funds or by requiring higherinterest rates.

AYAMI KOBAYASHI476

4. EMPIRICAL ANALYSES

4.1. Sample Selection and Data

The study uses unbalanced panel data from Japanese banks covering theperiod from fiscal 1998 through fiscal 2003. This period is chosen because itis apparent that financial measures of banks are considered to be morereliable since disclosure by banks became compulsory in 1998.

In order to discuss the effectiveness of issuing CDs in a panel analysisof Eqs. (3) and (4), it is assumed that the six years of the sample period(1998–2003) were stable. However, Japan has undergone significanteconomic transformation during that period.4 For example, in 1994, theliberalization of interest rates was completed. Since then, interest rates havebeen able to affect banks’ performance and depositors’ perceptions. Then, in1998, the Financial Function Early Strengthening Law and the FinancialReconstruction Law were enforced; moreover, in April 2002, a partialpayoff system was introduced. During the time period for which the payoffsystem was gradually released ending in April 2005, it was observed that atremendous amount of deposits have shifted among banks, especiallyaround 2002, from Shinkin banks to city banks. This shift implies thatdepositors perceived that Shinkin banks were financially weak, and so theyshifted their deposits to city banks or regional banks. Thus, it is notreasonable to judge that there has been no structural change during thesample period. Consequently, the sample period should be divided into twoperiods as follows: (1) a pre-partial payoff period in 1998–2000 and (2) apost-partial payoff period in 2001–2003.5

This study excludes banks that did not issue CDs and banks that issued inan amount less than one million yen and did not report them in theirfinancial statements. Further, I treat existing banks and merging banksdifferently, with the merging bank treated as one bank before the mergerand another bank after the merger. Thus, the sample size for city banks andregional banks varies as follows: (1) 19 and 102 in 1998, (2) 19 and 83 in1999, (3) 18 and 82 in 2000, (4) 15 and 88 in 2001, (5) 13 and 80 in 2002, and(6) 14 and 74 in 2003, respectively.

The quantity and interest rates as of March are taken from the Disclosure

of each bank and the Annual Nikkin Data Book published by the JapanFinancial News Co., Ltd. Financial measures are taken from Japan

Company Quarterly (IV) by Toyo Keizai Shimposha, and the Analysis of

Financial Statements of All Banks (March) by the Japanese BankersAssociation.


4.2. Methodology

4.2.1. Model Specification

To simplify the analyses in this study, suppose that the main sources of theliability in banking firms are deposits and revenues from banks’ lending.Suppose that demand from banks’ lending is given, since the aim of thechapter is to examine the depositor discipline hypothesis (shifts in supply).Thus, this chapter tests the interaction effect of interest rates and thequantity of uninsured CDs in the Japanese CDs market.

The standard method of testing the depositor discipline hypothesis is toanalyze the relationship between interest rates on CDs, the quantity of CDs,and accounting measures of risk specific to banks. Ideally, it is best toestimate a simultaneous equation model specified as Eqs. (1) and (2).

Regression model 1:

I i;t ¼ a1;i;t þ b1;tBRi;t�1 þ g1;tMRm;t þ d1;tQi;t þ u1;i;t (1)

Regression model 2:

Qi;t ¼ a2;i;t þ b2;tBRi;t�1 þ g2;tMRm;t þ d2;tI i;t þ u2;i;t (2)

where I i;t ¼ ðInterest expenses on CDs=Average amount of CDsÞ � 100%;Qi;t ¼ lnðquantity of CDsÞi;t; BRi,t�1= CAMEL, a vector of bank-specificrisk factors with a lag because balance sheet information is an ex postmeasure for the public, MRm,t=market risk, a risk that cannot bediversified away (a vector of financial market factors affecting each bank’sinterest rate or growth), and ui,t is random error term.

In the case of banking organizations, it is difficult to identify exogenousvariables (i.e., I and Q) that affect either banks’ demand or depositors’supply only in the structural-form equation.6 Thus, in this study, the linearrelationship between interest rates paid on CDs, the quantity of CDs, andCAMEL accounting measures follows the approach of Park (1995) andMartinez-Peria and Schmukler (2001). The following reduced-form model isestimated:

I i;t ¼ a1 þ mi þ dt þ b3CAMELi;t�1 þ b1Sizei;t�1 þ �i;t (3)

Qi;t ¼ a2 þ mi þ dt þ b4CAMELi;t�1 þ b2Sizei;t�1 þ oi;t (4)

where i=1, y, N and t=1, y, T. N is the number of banks in eachyear. The panel is unbalanced, so N varies across the sample period. mi is

AYAMI KOBAYASHI478

bank-specific or fixed effect, dt is time-specific effect included to controlfor macroeconomic and banking sector developments, CAMELi,t�1 is avector of bank-specific risk factors with a lag because balance sheetinformation is an ex post measure for the public, and ei,t and oi,t are randomerror terms.

In order to prove the effectiveness of issuing CDs in the Japanese CDmarket, I assert the following depositor discipline hypothesis:

� Null hypotheses, b=0� Alternative hypotheses, b 6¼ 0

4.2.2. Variables

Three types of data as of March are required: interest rates, quantities, andbank-specific risks. Table 1 gives the definitions of the variables used in theregressions.

4.2.3. Dependent Variables

There are two types of dependent variables: interest rates on CDs andquantities of CDs.

Table 1. Variable Definition.

Dependent variables

I, Q=Domestic operations of banks in yen-denominated trade

Explanatory variables (CAMEL)

Capital adequacy risk

CAR=capital adequacy ratio (%)

DCAR=1 for ISCAR, 0 otherwise

Asset quality risk

BADLOAN=disclosed nonperforming loans/total assets (%)a

Management risk

ROE=net income/stockholder’s equity� 100 (%)

Earnings risk

ROA=net income/total assets� 100 (%)

Liquidity risk

LIQ=(cash+due from banks)/total assets (%)

Control variable

Size

LOGTA=ln (total assets (million))

aNonperforming loans are defined as risk-monitored loans, bankrupt loans, nonaccrual loans,

past due loans (three months or more), and restructured loans.


Interest Rates on CDs. As sources available to the public do not explicitlyreport interest paid on deposits/CDs, an implicit rate is constructed.According to James (1988), this measure fails to account for differences inthe maturity of deposits/CDs outstanding, and it may reflect the rate offeredon deposits/CDs in previous periods as well as the rate on newly issuedliabilities. Moreover, uninsured CDs are not actively traded in secondarymarkets; although, in theory, the uninsured CDs could improve banks’surveillance. However, in this study, due to the unavailability of CDmaturities data, it is impossible to adjust the figures to use a weightedaverage maturity of deposits following the approach of Baer and Brewer(1986), James (1988), Keely (1990), and Martinez-Peria and Schmukler(2001). Fortunately, a survey conducted by James (1988) concludes that thedifference between the average interest on CDs as constructed above andthat from the Innerline Survey (the explicit rate) is not statisticallysignificant at the 1% level. Moreover, Gilbert, Meyer, and Vaughan(2001) showed that the constructed implicit rate serves as an acceptableproxy for default premiums though the model fit is poor. Therefore, theimplicit rate is employed as a proxy.

Quantities of CDs. The total amounts of CD outstanding are employed,and the natural logs of figures are taken considering the possibility ofheteroscedasticity.

4.2.4. Explanatory Variables

Because depositors can distinguish safe and sound banks by using CAMELproxies as bank-specific risks through the financial statements, Disclosure,and Japan Company Quarterly, deteriorating CAMEL values would be thebest signals of an increase in the risk profiles of banks.

Supposing that banks are adequately controlling for proficiency, risk, orother factors in the analyses, the following CAMEL proxies are selected.7

Capital Adequacy. The capital adequacy ratio (CAR) is included because itis the best measure for the public to distinguish safe and sound banksthrough financial statements. Higher CAR should raise the quantity of CDsand decrease interest rates on CDs. However, Japanese banks report twokinds of CAR: the international standard CAR (ISCAR) and the domesticCAR (DCAR). The major differences in calculating them are that (1) theISCAR allows banks to include unrealized profits in their securitiesportfolios into TIER2, while the DCAR does not; (2) 1.25% of loan lossreserves for ISCAR are allowed to be included as TIER2 and 0.625% for

AYAMI KOBAYASHI480

DCAR; and (3) the ISCAR adds market risks to the denominator (i.e., riskassets) and can include TIER3, but the DCAR does not (Yamori andKobayashi, forthcoming). Consequently, DCAR is included to control forthis difference as a dummy variable because it is impossible to compareISCA and DCAR directly. Although lower CAR banks are more likely to berisky, this point is controversial because CAR includes deferred tax assetsand may be subject to accounting manipulation in Japan. For example,Yamori and Kobayashi (forthcoming), using recent stock data of Japanesebanks, found that CAR is not economically but statistically significant.Hence, to examine whether CAR is an appropriate bank-specific riskmeasure merits further research. The expected sign on interest rates isnegative and the expected sign on the quantity of CDs is positive.

Asset Quality. BADLOAN measures the percentage of loans a bank mighthave to write off as losses. As a measure of asset quality, risky banks mighthave a higher ratio, and they ultimately charge off a relatively highpercentage of nonperforming loans.

In the regressions of Jumbo-CD yields on the ratio of BADLOAN,Herzig-Marx and Weaver (1979) and Hall et al. (2003) found no evidence ofa relationship. Furthermore, Hall et al. (2003) found no evidence in thegrowth rate of Jumbo CDs. However, using Japanese bank data, Hosono(2003) found that nonperforming loan share is negatively related to depositsgrowth for regional banks, and the share is significantly positively related tointerest rates on deposits for regional banks and negatively related for majorbanks. He concluded that this is because major banks are protected by aTBTF policy,8 and because there are discretionary accounting practices inthe disclosure of nonperforming loans and roll over bad loans by majorbanks.

In short, due to the TBTF policy, BADLOAN may not be a goodmeasure for major banks, whereas it is an acceptable measure for regionalbanks. The expected sign on the interest rate is positive and the expectedsign on the quantity is negative.

Asset Management. ROA measures the efficiency of banks assetmanagement. In theory, as banks increase their capital by accumulatingprofits, banks with large profits have more capital and could meet anunexpected loss. Among others, Martinez-Peria and Schmukler (2001)found that ROA in developing countries is significantly negatively related tointerest rates and positively related to all sizes of both insured and uninsureddeposits growth. Further, Hosono (2003) used Japanese bank data and


showed that ROA is positively correlated with the growth rate of deposits inboth major and regional banks. Further, he showed that ROA is negativelycorrelated with interest rates for regional banks. The expected sign on theinterest rate is negative and the expected sign on the quantity is positive.

Earnings. ROE measures the efficiency and profitability of bankstockholders’ equity. Among many previous studies, only Crane (1976)used ROE. He argued that ROE and ROA are interrelated measuresbecause it is impossible to identify the effect of any one of the measures onthe dependent variables. Yet, ROE is included because it is a reliable indexby which the public can evaluate banks’ performance. In theory, theexpected sign on the interest rate is negative and the expected sign on thequantity is positive.

Liquidity. LIQ is included to gauge banks’ liquidity risk. Since banks witha large volume of liquid assets are perceived to be safe, a higher ratioindicates a greater ability to meet maturing deposit obligations to pay outdepositors or unexpected withdrawals. Park (1995) and Gilbert et al. (2001,2003) found that higher LIQs enable banks to manage financial problemsmore flexibly. Furthermore, Martinez-Peria and Schmukler (2001) andDemirguc- -Kunt and Huizinga (2004) showed that the ratio is significantlynegatively associated with deposit interest rates. This finding suggests thatsafe banks with higher ratios do not pay higher interest rates. Thus, higherLIQ should lower interest rates, and therefore the expected sign on theinterest rate is negative.

Martinez-Peria and Schmukler (2001) found that LIQs are significantlypositively associated with medium uninsured deposits growth in Argentinaand with all sizes of uninsured deposits in Chile. In short, higher LIQ shouldraise quantities of CDs, and therefore the expected sign on the quantity ispositive.

Size. LOGTA, the natural logarithms of total assets, are used in theestimation to consider the possibility of heteroscedasticity.9 This proxy isincluded to control for the effect of bank size, because city banks areprotected by the TBTF policy. Many previous studies argue that depositors’belief that regulators are more tolerant of larger banks could cause marketdiscipline to be weakened due to the TBTF policy. Indeed, larger banks mayhave well-diversified loan profiles, but they are prone to extend or roll overbad loans because they benefit from the TBTF policy. Further, when citybanks fail, a systemic risk to other banking firms may result.

AYAMI KOBAYASHI482

Interestingly, Crane (1976), Hannan and Hanweck (1988), Brewer andMondschen (1994), Park (1995), and Jagtiani and Lemieux (2000) showedthat larger banks tend to offer lower rates on uninsured deposits becausethey are protected by the TBTF policy. Further, Park (1995) found thatuninsured deposits grow or fall when large banks offer higher or lowerinterest rates. On the other hand, Baer and Brewer (1986) and James (1990)found that the coefficient of bank size is insignificant in the analysis of therates effect. Thus, LOGTA is included to gauge whether major banks areprotected by the TBTF policy.


This section examines whether a relationship exists between interest rates,the quantity of CDs, and bank-specific risk.

5.1. Regression Results10

To assess the impact of utilizing CDs in enforcing market discipline, Iexamine whether deposits and the interest rates of CDs are indeed affectedby bank risk characteristics. First, I estimate the reduced-form Eqs. (3) and(4) using panel least squares. Although I estimate the model using plainOLS, the error terms may have bias in the estimated results. Thus, Eqs. (3)and (4) are reestimated using panel analysis (i.e., a fixed effect model and arandom effect model). Each bank has many branches in one prefecture ofJapan, thereby producing bank-specific characteristics in each area.Considering this heteroscedasticity, a fixed effect model is employed in theanalysis.11

Supposing, then, that there is no correlation between explanatoryvariables and bank-specific characteristics; the reduced-form equations arereestimated using a random effect model.

Finally, focusing on financial data, I conducted a robustness check toensure the results derived from the fixed effect model using the test ofequality of means for banks issuing CDs and banks not issuing CDs.

Parameter estimates are provided in Tables 2 and 3.

5.1.1. Interest Rates Analysis

To study the effects of bank characteristics on CD interest rates, I estimatethe reduced-form model Eq. (3) above; parameter estimates are presented inTable 2.


City Banks in the Pre-Payoff Period (1998–2000). In Japan, deposits areclassified into fully insured demand deposits (i.e., current, ordinary, savings,and notice) and uninsured time deposits (i.e., time, installment savings, andCDs). In May 1979, Japanese banks started to issue CDs with the intention

Table 2. Reduced-Form Model Analyses for Interest Rates, I, onCAMEL Factors in the Pre-Partial Payoff Period (1998–2000) and the

Post-Partial Payoff Period (2001–2003).

Variables Fixed Effect Model Random Effect Model

1998–2000 [1] 2001–2003 [2] 1998–2000 [3] 2001–2003 [4]

City banks

Constant 2.166 (0.71) �2.273 (2.00) 3.431 (3.317) �0.424 (0.77)

Size (LOGTA) �0.070 (0.40) 0.132 (2.02) �0.134 (2.42)� 0.025 (0.82)

C (CAR) �0.044 (3.26)�� 0.0004 (0.05) �0.043 (3.56)�� 0.003 (0.47)

A (BADLOAN) �0.081 (4.92)�� 0.001 (0.34) �0.061 (4.46)�� 0.002 (0.78)

M (ROA) 0.050 (1.56) �0.014 (1.02) 0.027 (1.07) �0.006 (0.48)

E (ROE) �3.97E�05 (4.53)�� 0.0001 (1.00) �3.25E�05 (4.05)�� 4.38E�05 (0.38)

L (LIQ) 0.027 (1.97) 0.007 (1.30) �0.0091 (0.17) 4.42E�05 (0.01)


Adjusted R2 0.41 �0.33 0.33 0.01

F-value 2.477�� 0.609


1998–2000 [5] 2001–2003 [6] 1998–2000 [7] 2001–2003 [8]

Regional banks

Constant 2.713 (1.15) 0.924 (0.31) 1.028 (2.74) 0.384 (4.42)

Size (LOGTA) �0.150 (0.92) �0.057 (0.28) �0.024 (0.87) �0.025 (4.34)��

C (CAR) �0.001 (0.13) �0.003 (0.34) �0.025 (2.70)�� 0.004 (1.34)

A (BADLOAN) �0.087 (10.35)�� 0.002 (0.45) �0.056 (8.10)�� 0.001 (0.50)

M (ROA) �0.011 (0.40) �0.003 (0.18) �0.057 (2.43)� 0.0008 (0.64)

E (ROE) �5.10E�05 (0.56) �0.0005 (1.39) 5.42E�05 (0.65) �0.0008 (3.92)��

L (LIQ) 0.015 (1.50) 0.003 (1.40) 0.006 (0.76) 0.002 (1.67)


Adjusted R2 0.43 0.35 0.28 0.09

F-value 3.35�� 2.32��

Note: Estimation of the regression model, Eq. (3). A year dummy is included in the plain OLS

analysis but is not reported here. Estimators for time dummies, fixed effects, and the DCAR

variable are not reported here even though they are included in the regressions. t-statistics are

reported in parentheses beside each coefficient estimate.�Significant at 5% level.��Significant at 1% level.

The Wansbeek–Kapteyn method is used to perform calculations in the random effects method.

AYAMI KOBAYASHI484

of encouraging the liberalization of interest rates in the Japanese secondaryCD market, the short-term money market of less than one year, the buyersof which are not limited to financial institutions. According to statistics inThe Analysis of Financial Statements of All Banks by the Japanese Bankers

Table 3. Reduced-Form Model Analyses for Quantity, Q, on CAMELFactors in the Pre-Partial Payoff Period (1998–2000) and the Post-Partial

Payoff Period (2001–2003).


1998–2000 [9] 2001–2003 [10] 1998–2000 [11] 2001–2003 [12]

City banks

Constant 17.893 (1.50) 16.774 (1.42) �4.248 (0.62) 6.277 (1.27)

Size (LOGTA) �0.188 (0.28) �0.036 (0.05) 1.070 (2.81)�� 0.541 (1.95)

C (CAR) �0.020 (0.38) �0.148 (1.79) 0.020 (0.40) �0.136 (1.93)

A (BADLOAN) �0.017 (0.27) 0.0007 (0.01) 0.028 (0.46) �0.009 (0.24)

M (ROA) 0.060 (0.47) 0.574 (4.03)�� 0.188 (1.69) 0.599 (4.68)��

E (ROE) �1.86E�05 (0.54) 0.0008 (0.64) �3.75E�06 (0.11) 0.001 (0.95)

L (LIQ) �0.083 (1.54) �0.021 (0.36) �0.101 (2.07)� 0.017 (0.45)


Adjusted R2 0.90 0.92 0.34 0.83

F-value 22.361�� 19.422��


1998–2000 [13] 2001–2003 [14] 1998–2000 [15] 2001–2003 [16]

Regional banks

Constant 5.454 (0.40) �11.302 (0.20) �21.069 (5.55) �20.269 (12.29)

Size (LOGTA) 0.204 (0.22) 1.408 (0.37) 2.020 (7.40)�� 1.776 (15.98)��

C (CAR) �0.079 (1.02) �0.051 (0.29) �0.065 (0.95) 0.406 (6.14)��

A (BADLOAN) �0.097 (2.00)� 0.152 (1.30) �0.074 (1.63) 0.083 (2.09)

M (ROA) 0.140 (0.89) 0.086 (2.84)�� 0.079 (0.53) 0.052 (2.11)�

E (ROE) �0.0002 (0.52) 0.001 (0.17) �0.0003 (0.71) �0.005 (1.23)

L (LIQ) 0.021 (0.37) �0.096 (1.96) 0.0004 (0.00) �0.139 (4.23)��


Adjusted R2 0.86 0.80 0.12 0.44

F-value 15.887�� 10.817��

Note: Estimation of the regression model, Eq. (4). A year dummy is included in the plain OLS

analysis but is not reported here. Estimators for time dummies, fixed effects, and the DCAR

variable are not reported here even though they are included in the regressions. t-statistics are

reported in parentheses beside each coefficient estimate.�Significant at 5% level.��Significant at 1% level.

The Wansbeek–Kapteyn method is used to perform calculations in the random effects method.


Association, CDs of all banks in Japan account for only 5% of bankliabilities. However, CDs represent the only bank deposit not protected bythe Japan DIC explicitly since the beginning of the issuance. Thus, CDsshould reflect banks’ failure probabilities or performance and depositors’risk perception. Indeed, CD holders are the most risk-averse among all bankclassifications of depositors.

Considering that CDs exceedingly reflect banks’ failure probabilities, it isnecessary to distinguish between fully insured deposits and uninsureddeposits and CDs, when conducting an analysis of the depositors’ disciplinehypothesis. Hosono (2002, 2003) analyzed whole deposits, but the presentchapter distinguishes between fully insured deposits and uninsured depositsand CDs. Thus, the findings of the present chapter exceedingly reflect theevidence of market discipline by CD holders, who are uninsured depositors.

Column [1] shows that in 1998–2000 CAR is significantly negativelyrelated to interest rates for city banks. For example, the coefficient of CARin city banks is �0.044. This means that a 1% increase in CAR reducesinterest rates by 0.04% in city banks. In city banks, the mean and standarddeviation of CAR are 11.011 and 2.241, respectively. Given that the averageinterest rate is 0.387% in city banks, the effect of CAR on the interest ratesis economically significant. In other words, the interest rates of CDs declinewhen the CAR ratio becomes higher; this suggests that safe and soundbanks with adequate CAR ratios will offer lower interest rates. This findingcan be explained by the following: (1) CAR has been a well-known index forthe public to select safe and sound banks since the BIS requirement, CAR,was introduced in the 1988 Accord; (2) market participants became cautiousabout bank safety and soundness since the financial system crisis was seriousduring this period; and (3) the effect appeared only for city banks becausecity banks account for 80% of CDs issued by banks.

This is consistent with the finding of Yamori and Kobayashi (forth-coming), using Japanese bank stock data, and Hori, Ito, and Murata (2005),using recent data for city banks, first and second regional banks, shinkinbanks, and credit cooperatives, and the finding that Japanese depositors inthe 1990s responded to the risks of financial institutions. In contrast, this isinconsistent with the findings of Hosono (2002, 2003). Using Japanese bankdata from the 1990s, Hosono (2002, 2003) concretely examined therelationship between interest rates, growth of deposits, and factors affectingbank risk.12 He found that for regional banks, both interest rates andgrowth of deposits are significantly correlated with some bank riskmeasures. Yet for major Japanese banks, the interest rates of deposits arenot significantly correlated and the growth rates of deposits are weakly

AYAMI KOBAYASHI486

correlated with some bank risk measures. Thus far, Hosono (2002, 2003),Hosono, Iwaki, and Tsuru (2004), Tsuru (2003), Murata and Hori (2006),and Hori et al. (2005) have investigated the hypothesis for Japanese banks.Most of the previous studies insist that interest rates of deposits and growthrates of deposits have no disciplinary power to discipline major Japanesebanks, but they may discipline regional Japanese banks. However, thepresent chapter finds that for major banks, CD holders may exercisedisciplinary power to impose market discipline.

City Banks and Regional Banks in the Pre-Payoff Period (1998–2000). Columns[1] and [5] document that BADLOAN is significantly negatively related tointerest rates. For example, the coefficients of BADLOAN are �0.081 and�0.087 in city banks and regional banks, respectively. This suggests that a 1%increase in BADLOAN reduces interest rates by 0.08%. The mean and standarddeviation of BADLOAN are 5.286, 3.170 and 4.308, 2.107 for city banks andregional banks, respectively. Given that the average interest rates are 0.387 and0.316% for city banks and regional banks, respectively, the effect of BADLOANon the interest rates is economically significant.

This finding is inconsistent with those of previous studies, yet it can beexplained as follows. In theory, higher BADLOAN ratios induce interestrates to grow, meaning that banks have more nonperforming loans to writeoff, and thereby their financial conditions are deteriorating. Yet, if ailingbanks cannot afford to offer higher interest rates and there exists girigashibetween the firms and ailing banks, girigashi may induce interest rates todecline.13

Column [1] shows that ROE is significantly negative at the 1% level.However, since the coefficient of ROE is almost zero, it is not economicallysignificant.

In summary, CD interest rates are sensitive to CAR in the pre-partialpayoff period, 1998–2000, yet this effect is not observed in the post-partialpayoff period, 2001–2003. This chapter finds that city banks offer lower CDinterest rates because the issuing banks are mostly city banks, and city banksare considered to be safe and sound banks. Thus, CAR is an appropriatemeasure of bank-specific risk for determining interest rates on CDs.

5.1.2. Quantity Analysis

To study the effects of bank characteristics on CD quantities, I estimate thereduced-form model Eq. (4) above; parameter estimates are presented inTable 3.


City Banks and Regional Banks in the Post-Payoff Period (2001–2003). Columns[10] and [14] in the post-payoff period show that coefficients of ROA in both citybanks and regional banks are positive and significant at the 1% critical level.Higher ROA implies that banks manage their assets efficiently and have adequateassets to meet unexpected losses. For example, the coefficients of ROA in citybanks and regional banks are 0.574 and 0.086, respectively. This means that a 1%increase in ROA increases the quantity of CDs by 0.57 and 0.08%, respectively.The mean and standard deviation of ROA are �0.763, 1.458 and �0.207, 0.642for city banks and regional banks, respectively. Given that the average quantities incity banks and regional banks are 13.503 and 7.588, respectively, the effect of ROAis economically significant.

This finding can be explained by the following:

� There exists a huge deposit shift in banks during the post-payoff period.Concretely, this implies that the risk perception of market participants,who transfer their deposits, has changed between the pre-partial and thepost-partial payoff period. For example, during the pre-partial payoffperiod, market participants, especially depositors, are sensitive to CARbecause CAR has been the disseminated, well-known index for selectingsafe and sound banks. Yet, since the payoff system has been graduallyreleased, market participants have begun to have strong incentives toevaluate banks’ performance in asset management. ROA is related todeposits and is an appropriate measure for the public to evaluate theprofitability of the banking business. Thus, during the post-payoff period,depositors became more cautious, so they began to start to utilize ROAinstead of CAR.� Since depositors impose market discipline on banks by withdrawing theirdeposits, depositors may withdraw their deposits based on ROA, which isthe only measure related to the efficiency of managing deposits. Thisfinding regarding the quantity of CDs is consistent with the findings ofHosono (2003) and Hori et al. (2005) using Japanese bank data and thoseof Martinez-Peria and Schmukler (2001).

For regional banks, column [13] in 1998–2000 shows that the coefficientof BADLOAN is significant and negatively related to quantity. A higherBADLOAN induces the quantity to decrease. This is consistent with thefinding of Hosono (2003).

In summary, quantities of CDs for city and regional banks are sensitive toROA among the Japanese CAMEL indices. Since ROA is an appropriateindex for determining the quantities of CDs, market participants and

AYAMI KOBAYASHI488

depositors might withdraw their deposits based on the ROA index becauseROA is the only index related to the effectiveness of managing deposits.

5.2. Robustness Check

I conducted a robustness check to validate the result for the fixed effectmodel presented in Eqs. (3) and (4) of Tables 2 and 3, respectively. To checkwhether the results for the fixed effect model are robust, I reestimated themodels using a random effect approach, and I then conducted a test ofequality of means of banks issuing CDs and banks not issuing CDs.

As in the results for interest rates on the CAMEL factors, columns [3]and [7] show that CD interest rates are sensitive to CAR and BADLOAN inthe pre-partial payoff period, 1998–2000. Furthermore, as in the resultsregarding CD quantity on CAMEL factors, columns [12] and [16] showthat CD quantities are sensitive to ROA in the post-partial payoff period,2001–2003.

Although columns [3] and [8] for ROE on interest rates are significantlynegative at the 1% level, the coefficients are not economically significant.Further, columns [11] and [16] for CD quantities show that LIQ issignificantly negative in both periods. The findings in LIQ are inconsistentwith those of previous studies. One possible explanation for this is asfollows. In theory, LIQ is an index used to measure banks’ ability to meetunexpected deposit withdrawals; thus, raising LIQ induces the quantities togrow. However, only banks can issue CDs. In their balance sheets,banks issuing CDs count CDs as Certificates of Deposit in their liabilityaccounts, whereas banks holding CDs count CDs as Cash and Cash Due intheir assets. Accordingly, an increase in the quantities of CDs issued byissuing banks corresponds to an increase in the Cash and Cash Due incalculating the LIQ variable. Therefore, holding banks have alreadyperceived issuing banks as risky banks, and raising LIQ thus induces CDquantities to decline.

Next, to determine the robustness from the perspective of financial data,I conducted a test of the equality of means of banks issuing CDs andbanks not issuing CDs.14 The test is conducted each year. Both CARand BADLOAN are significant at the 1% critical level. Thus, the resultsindicate that CAR and BADLOAN support the effectiveness ofissuing CDs.

In summary, the basic conclusions about the imposition of marketdiscipline by CDs appear to be very robust when different tests are applied.


6. CONCLUSION

This chapter empirically investigates the depositor discipline hypothesis, inparticular CD holders’ discipline in city banks and first and second regionalbanks from the financial year 1998 to the financial year 2003. The objectiveof this study was to examine the relationship between CD interest rates,quantities issued by Japanese banks, and bank-specific risks in the JapaneseCD market. I am concerned with a question that is relevant to the design ofa 21st century financial system based on a new regulatory framework forJapanese banks: namely, whether depositors, especially CD holders, candiscipline banks.

Some previous studies have analyzed whether all depositors can disciplinebanks in Japan. Even though the CD deposits of all banks in Japan accountfor only 5% of bank liabilities, CD holders are the most risk-averseamong bank depositors because only CDs are not explicitly protected by theJapan DIC. Although some previous studies have analyzed marketdiscipline in Japanese banks, those chapter focused on deposits as a wholeand did not distinguish the classifications of deposit types. To date, there isno empirical chapter in which CD holders’ discipline was tested. This is thefirst chapter to examine the effectiveness of issuing CDs in the Japanese CDmarket.

The results obtained in this chapter show that the Japanese CD market issensitive to bank-specific risks. Particularly, among the CAMEL indices,CD interest rates are sensitive to CAR and CD quantities are sensitive toROA. Since the BIS requirement, CAR, was introduced in 1988, it hasbecome a well-known, disseminated index by which market participants canselect safe and sound banks. Also, ROA is the only measure among theCAMEL indices related to deposits showing the effectiveness of managingloans and funds collected from bank deposits; thus, depositors maywithdraw their deposits based on ROA. Bank-specific risks such as CARand ROA are considered to be appropriate financial measures to determineCD interest rates and quantities. In other words, the results of this studyindicate that CD interest rates and quantities are sensitive to some bank-specific risks in the Japanese CD market if those risks include the followingcharacteristics: (1) measures should be taken from financial statements,(2) measures should be permeative enough among market participants, and(3) measures should be deeply related to the effectiveness of managingdeposits. Thus, the findings suggest that the utilization of CDs issued byJapanese banks to improve market discipline in banks is effective in theJapanese CD market.

AYAMI KOBAYASHI490

Moreover, the Bank of Japan (BOJ) maintained their quantitativemonetary easing policy, overnight call rate remained close to zero andFinance Bill (FBs) yield also remained at around 0.000–0.010%, duringMarch 2001 through March 2006. Post-partial payoff period (2001–2003) isconsistent with BOJ’s easing policy period and there are almost nosignificant results in post-partial payoff period. Thus, it is concluded thata credit risk does not reflect interest rates in post-partial payoff period dueto the introduction of BOJ’s quantitative monetary easing policy.

In conclusion, CDs issued by Japanese banks have improved the marketdiscipline of Japanese banks. CD holders are sensitive to bank risks andexercise disciplinary power that complements regulatory discipline. Thisfinding suggests that a policy requiring Japanese banks to issue CDs wouldbe likely to help achieve market discipline in banks to some degree. Forexample, a requirement to keep outstanding CDs above a required amountby the bank authorities would have a similar disciplinary function. In thatscheme, CD holders would be eager to discipline banks.

NOTES

1. S&L is a depository financial institution, federally or state chartered, holding itsassets mostly in residential mortgages and collecting its deposits from consumers.2. CAMEL(S) rating is used to evaluate bank performance and is composed of

capital adequacy, asset quality, management, earnings, liquidity, and the bank’ssensitivity to market risk. BOPEC measures the safety and soundness ratings for aBHC and is composed of the conditions of the BHC’s bank subsidiaries, other(nonbank) subsidiaries, parent company, earnings, and the capital adequacy.3. The Federal Deposit Insurance Corporation (DIC) was created in the Glass-

Steagall Act of 1933. FDICIA of 1991 represents fundamental deposit insurance andprudential regulatory reform to strengthen the financial condition of the banking andthrift industries.4. See Kobayashi (2004) for a more detailed discussion regarding reforms in the

Japanese financial system.5. The structural change analysis is conducted for each sample period, 1998–2000

and 2001–2003. F values for each test are significant at the 1% level, and thus, thenull hypotheses are rejected. Eqs. (3) and (4) are estimated using panel least squaresand panel analysis (i.e., a fixed effect model and a random effect model), respectively,during the whole sample period, 1998–2003. However, the results are random and,thus, not reported here.6. See Kobayashi and Bremer (2005) for a more detailed discussion of the

reduced-form equations models in this study.7. CAMEL had been used before the ‘‘S’’ component of CAMELS was

introduced in January 1997. And the sample period of the present chapter starts


in 1998. Thus, CAMEL, rather than CAMELS, would have been used during theperiod of the present chapter.8. TBTF is an acronym of too-big-to-fail, and TBTF policy is defined as follows:

A bank having a tremendous amount of total asset is recognized as a large or bigbank. If the large bank fails, a systemic risk to other banking firm may result. In thiscase, the government has to avoid a big bank’s failing in order not to spread theinfluence of a systemic risk in an economy.9. Demirguc--Kunt and Huizinga (2004) found no relationship between growth

rate and bank size. Likewise, Hall et al. (2003) found no relationship between banksize and deposits growth rates or yields.10. Descriptive statistics are available from the author upon request.11. By using a fixed effect model, variance will be adjusted; as a result, it is

possible to estimate by considering bank-specific characteristics that are notobserved.12. Concretely, Hosono (2002) developed a market discipline model of bank exit

behavior, utilizing the following variables: (1) the bank exit measures are the depositinterest rates, the rates of increase in deposits, BIS ratios, and the ratios of capital tototal assets; and (2) the bank risk measures are bank failure probabilities, real estateloan proportions, disclosed nonperforming loan ratios, and market value capitaladequacy ratios.13. Girigashi is the traditional lending system in Japan. For example, ailing banks

may not be able to afford to offer higher interest rates on CDs. However, if the bankhas a long, deep relationship with the firm, the ailing bank is allowed by the firm tooffer a CD account with lower interest rates. However, according to an interviewwith one mega bank, there is no girigashi in the CD account trade, but it may exist inlending for firms.14. The table of test for equality of means for banks issuing CDs and banks not

issuing CDs for regional banks is available from the author upon request.

ACKNOWLEDGMENTS

The author is grateful for the helpful comments and suggestions received onearlier drafts of this chapter from professors Nobuyoshi Yamori, WataruOhta, Marc Bremer, Kenya Fujiwara, Eiji Okuyama, Katsutoshi Shimizu,Thoru Nakazato, and anonymous referees. Earlier versions of the chapterwere presented at the Summer Institute in Modern Monetary Economics atKobe University in August 2004 and the Yamori-Ohta joint seminar atNagoya University in November 2004, the Chubu Japan Society ofMonetary Economics in November 2005, the Nagoya City UniversityWednesday Workshop in January 2006, the Nanzan Modern AccountingWorkshop in March 2006, and the Nippon Finance Association AnnualMeeting in June 2006. Japan Society of Monetary Economics in September2007, The Japan Society of Household Economics Chubu-District in

AYAMI KOBAYASHI492

October 2007, Japan Finance Association in October 2007. Special thanksare due to the participants for productive discussions.

REFERENCES

Baer, H., & Brewer, E. (1986). Uninsured deposits as a source of market discipline: Some new

evidence. Federal Reserve Bank of Chicago, Economic Perspectives (September), 23–31.

Berger, A. (1991). Market discipline in banking. In: Proceeding of a conference on bank structure

and competition (pp. 419–437). Chicago: Federal Reserve Bank of Chicago.

Billet, M., Garfinkel, J., & O’Neal, E. (1998). The cost of market versus regulatory discipline in

banking. Journal of Financial Economics, 48, 333–358.

Brewer, E., & Mondschen, T. (1994). An empirical test of the incentive effects of deposit

insurance. Journal of Money, Credit, and Banking, 26, 146–164.

Calomiris, C. W., & Wilson, B. (2004). Bank capital and portfolio management: The 1930s

‘‘Capital Crunch’’ and the scramble to shed risk. Journal of Business, 77(3), 421–455.

Cargill, T. (1989). CAMEL ratings and the CD market. Journal of Financial Services Research,

3(4), 347–358.

Cook, D., & Spellman, L. (1994). Repudiation risk and restitution costs: Toward understanding

premiums on insured deposits. Journal of Money, Credit, and Banking, 26, 439–459.

Crabbe, L., & Post, A. M. (1994). The effect of a rating downgrade on outstanding commercial

paper. Journal of Finance, 49, 39–56.

Crane, D. B. (1976). A study of interest rate spreads in the 1974 CD market. Journal of Bank

Research, 7, 213–224.

Demirguc- -Kunt, A., & Huizinga, H. (2004). Market discipline and deposit insurance. Journal of

Monetary Economics, 51, 375–399.

Dewatripont, M., & Tirole, J. (1994). The prudential regulation of banks. Cambridge: The MIT

Press.

Ellis, D., & Flannery, M. (1992). Does the debt market assess large banks’ risk? Time series

evidence from money center CDs. Journal of Monetary Economics, 30, 481–502.

Flannery, J. M. (1998). Using market information in prudential bank supervision: A review of

the US empirical evidence. Journal of Money, Credit, and Banking, 30, 273–305.

Gilbert, A. R., Meyer, P. A., & Vaughan, D. M. (2001). The use of market information in bank

supervision: Interest rates on large time deposits. Working Paper. Federal Reserve Bank

of St. Louis, St. Louis, MO.

Gilbert, A. R., Meyer, P. A., & Vaughan, D. M. (2003). Can feedback from the Jumbo-CD

market improve bank surveillance? Working Paper no. 2003-041A. Federal Reserve Bank

of St. Louis, St. Louis, MO.

Goldberg, L., & Hudgins, S. (1996). Response of uninsured depositors to impending S&L

failures: Evidence of depositor discipline. Quarterly Review of Economics and Finance,

36, 311–325.

Goldberg, L., & Hudgins, S. (2002). Depositor discipline and changing strategies for regulating

thrift institutions. Journal of Financial Economics, 63, 263–274.

Goldberg, M., & Lloyd-Davies, P. (1985). Standby letters of credit: Are banks overextending

themselves? Journal of Bank Research, 16, 29–39.


Hall, J. R., King, B. T., Meyer, P. A., & Vaughan, D. M. (2003). Do Jumbo-CD holders care

about anything? Federal Reserve Bank of St. Louis Supervisory Policy Analysis Working

Paper No. 2002-05.

Hannan, T. H., & Hanweck, A. G. (1988). Bank insolvency risk and the market for large

certificates of deposit. Journal of Money, Credit, and Banking, 20, 203–211.

Herzig-Marx, C., & Weaver, S. A. (1979). Bank soundness and the market for large negotiable

certificates of deposit. Research Paper No. 79-1. Federal Reserve Bank of Chicago,

Chicago.

Hori, M., Ito, Y., & Murata, K. (2005). Do depositors respond to bank risks as expected?

Evidence from Japanese financial institutions in the banking crisis. ESRI Discussion Paper

Series No. 151. Economic and Social Research Institute, Cabinet Office, Tokyo.

Hosono, K. (2002). Naniga Ginko o kiritsudukerunoka? [Stock market, depositors, and the

government]. Discussion Papers in Economics no. 315. Nagoya City University, Nagoya,

Japan.

Hosono, K. (2003). Market discipline and forbearance policy to banks. Discussion Paper in

Economics No. 339. Nagoya City University, Nagoya, Japan.

Hosono, K., Iwaki, H., & Tsuru, K. (2004). Bank regulation and market discipline around the

world. RIETI Discussion Paper Series 04-E-31. Research Institute of Economy, Trade

and Industry, Japan.

Hosono, K., & Sakuragawa, M. (2003). Soft budget problems in the Japanese credit market.

Discussion Papers in Economics No. 345. Nagoya City University, Nagoya, Japan.

Jagtiani, J., & Lemieux, C. (2000). Stumbling blocks to increasing market discipline in the banking

sector: A note on bond pricing and funding strategy prior to failure. Emerging Issues Series

S&R-99-8R. Federal Reserve Bank of Chicago, Chicago.

James, C. (1988). The use of loan sales and standby letters of credit by commercial banks.

Journal of Monetary Economics, 22, 395–422.

James, C. (1990). Heterogeneous creditors and the market value of bank LDC loan portfolios.

Journal of Monetary Economics, 25, 325–346.

Jordan, J. S. (2000). Depositor discipline at failing banks. New England Economic Review

(March/April), 15–28.

Kane, E. (1987). Who should learn what from the failure and delayed bailout of the ODGF? In:

Proceedings from a conference on bank structure and competition (pp. 306–326). Chicago:

Federal Reserve Bank of Chicago.

Keely, M. C. (1990). Deposit insurance, risk, and market power in banking. American Economic

Review, 80, 1183–1200.

Kobayashi, A. (2003). Is issuing subordinated debt by Japanese banks effective in the Japanese

market? International Finance Review, 4, 303–323.

Kobayashi, A. (2004). Market discipline and financial system reforms in Japan: Have the

reforms been successful in establishing the sound financial system? Economic Science, 52,

105–116.

Kobayashi, A., & Bremer, M. (2005, October). The depositors discipline hypothesis: A review of

the empirical evidence in the US and Japan. Working Paper Series. Nanzan University

Center for Management Studies, Nagoya, Japan.

Martinez-Peria, M. S., & Schmukler, L. S. (2001). Do depositors punish banks for bad

behavior? Market discipline, deposit insurance, and banking crises. Journal of Finance,

56, 1029–1051.

AYAMI KOBAYASHI494

Murata, K., & Hori, M. (2006). Do small depositors exit from bad banks? Evidence from small

financial institutions in Japan. Japanese Economic Review, 57(2), 260–278.

Park, S. (1995). Market discipline by depositors: Evidence from reduced-form equations.

Quarterly Review of Economics and Finance, 35, 497–514.

Park, S., & Peristiani, S. (1998). Market discipline by thrift depositors. Journal of Money,

Credit, and Banking, 30, 347–364.

Tsuru, K. (2003). Depositors’ selection of banks and the deposit insurance system in Japan:

Empirical evidence and its policy implications. RIETS Discussion Paper Series 03-E-024

Research Institute of Economy, Trade and Industry, Japan.

Yamori, N., & Kobayashi, A. (forthcoming). Wealth effect of public fund injections to ailing

banks: Do deferred tax assets and auditing firms matter? Japanese Economic Review.


CHAPTER 23

WHAT ARE THE NEXT STEPS FOR

BOND MARKET DEVELOPMENT

IN THAILAND?

Jonathan A. Batten and Pongsak Hoontrakul

ABSTRACT

Recently East Asian policymakers have focused on facilitating corporate

bond market development through a host of financial market reforms

including greater foreign participation in the domestic markets as issuers

and investors. However, the alternate approach – the encouragement of

domestic issuers to further tap international markets – remains largely

ignored. The objective of this study is to investigate these issues in the

context of reform undertaken by Thailand following the Asian Crisis of

1997. As a small and open economy, Thailand was forced to become more

receptive to foreign investment and capital market participation. We raise

the significance of bond return volatility and skewness as an impediment

to greater involvement by international investors. Empirical analysis

highlights the time-varying nature of both variance and skewness of bond

returns, which can only be overcome through government policy that

focuses upon stabilizing the macroeconomic environment and not simply

enhancing domestic and regional financial market infrastructure.





ISSN: 1569-3767/doi:10.1016/S1569-3767(07)00023-4

497

dx.doi.org/10.1016/S1569-3767(07)00023-4.3d

1. INTRODUCTION

One of the well-documented successes of the Asia-Pacific region in the post-crisis period has been the significant development of local financial markets:credit extended in domestic markets, stock market capitalization, and thesize of both the government and corporate bond sectors have all increasedat a regional level over the last 10 years. In absolute terms, these increasesrepresent commendable successes in terms of achieving national policyoutcomes as well as those mandated by regional initiatives.1

Though noteworthy, these regional successes disguise the considerablevariation that exists in the achievement of these outcomes within the region,and especially when compared with the levels evident in some financialmarkets in Europe and America. For example, the corporate bond sectorremains undeveloped in many countries (less than 5% of GDP in 2005/2006in India, Indonesia and Philippines); the market for domestic credit fromthe banking sector is considerably smaller in India, Indonesia, Philippinesand Thailand than Australia, China, Korea and Japan, while stock andgovernment bond markets also display significant variation.

The overriding policy response to financial market reform in the Asia-Pacific region has focused on developing alternatives to bank-intermediatedfinancing in the corporate sector, by facilitating corporate bond marketdevelopment (Herring & Chatusripitak, 2000; IMF, 2005), with consider-able progress being made to the improvement of the scale and scope ofmany domestic bond markets in recent years. However, McCauley and Park(2006) noted that domestic bond market development is one of threeseparate policy perspectives that could be adopted by government with eachrequiring different responses and strategies: first, an enhanced but notnecessarily integrated series of domestic markets; second, a regional bondmarket denominated in regional currencies; and finally, a global marketwhere East Asian borrowers and possibly investors are minor players. Theseauthors favour the third image of bond market development, where nationalbond markets are developed with the ultimate objective of integration intoa global market. Enhancing global integration requires greater foreignparticipation in the domestic markets (Burger & Warnock, 2006a) thatincludes the issuance activities of multilateral development banks (Hoschka,2006), foreign corporations (Batten & Szilagyi, 2007) and foreign investors(Bae, Yun, & Bailey, 2006), as well as an expanded role for domestic issuersin international markets.

The key objective of this study is to investigate some of the key empiricalfeatures of Thai international bond issues that may enhance or impede their

JONATHAN A. BATTEN AND PONGSAK HOONTRAKUL498

appeal in international markets. This study therefore builds upon Burgerand Warnock (2006a), who previously have highlighted the importance ofkey statistical characteristics of bond returns as well as the more frequentlydiscussed investor protection and governance factors that may act asinvestor impediments to bond market development.

Thailand is a natural candidate for an examination of these issues. Asa small and open economy, after the 1997 crisis Thailand was requiredto be more receptive to foreign investment and capital market participationthan ever before. The key condition stipulated by the InternationalMonetary Fund (IMF) for financial support was that Thailand had tofurther develop its domestic bond markets to avoid the historic mismatchin the maturity and currency of cooperate borrowing. Improving legalrights and investor/lending protection, emphasizing corporate governanceand enhancing regulatory supervisions were also among IMF recommen-dations.

Thailand was also encouraged to adopt a more free-market approach toeconomic management, which resulted in lifting the permissible levels offoreign ownership in many strategic industries (e.g. bank, insurance,property, automobile, etc.). As a result, over the past decade, considerabledisintermediation was undertaken in Thailand’s financial markets, whichalso saw the blossoming of the domestic bond markets. For example, thedomestic debt market grew more than 10-fold from 1997 to 2006, while totalbanking lending was reduced to less than 80% of GDP compared with128% in 1997.

In many ways Thailand has also been a victim of its own success withanti-foreign sentiment resurfacing after the successful coup in late 2006, aperiod also characterized by extensive intervention by the Bank of Thailand(BoT) in foreign exchange markets to stabilize the Thai baht (THB). Theserecent actions by the BoT have distorted prices in bond (and FX) marketsdue to the extensive issues in the THB government bond market required topurchase incoming US dollar (USD).2 To impede the flow of foreign capital,controls were imposed in December 2006, with changes then made to theForeign Business Act in April 2007 to tighten foreign engagement in variousindustries. In short, it is a very interesting time to examine the Thailandbond market, 10 years after the financial crisis.

This chapter is set out as follows. Initially, a succinct review is made ofthe expanding literature on domestic bond market development. Then, aperspective of the scale and scope of the Thai international bond marketis briefly undertaken. Finally, some key points to expanding internationalbond issuance are discussed and conclusions provided.

What are the Next Steps for Bond Market Development in Thailand? 499

2. BOND MARKET DEVELOPMENT

There is a rich literature – and opinions – from authors investigating the lackand variation of domestic corporate bond markets, especially in the Asia-Pacific or East Asian markets (e.g. Benzie, 1992; Emery, 1997; Schinasi &Smith, 1998; Kim, 1999; Herring & Chatusripitak, 2000; Batten & Kim,2001; Rhee, 2003, 2004; Eichengreen & Luengnaruemitchai, 2004; Jiang andMcCauley, 2004; McCauley and Jiang, 2004; IMF, 2005; Burger &Warnock, 2006b; amongst others). These authors highlight many of theobvious technical obstacles: supply side impediments (providing an enablingenvironment, maintaining the reform of corporate governance), demandside impediments (strengthening the role of institutional investors andmutual funds, considering private placement as a short-term option), andinfrastructure impediments (ensuring that credit ratings are reliable, creatingbenchmark yield curves, ensuring that there is an effective and enforceableregulatory framework, providing quality settlement and risk managementsystems and technology). The latter cannot be underestimated since theabsence of deep and rich derivatives markets prevents the hedging of interestand exchange rate risk of international investors and issuers (Burger &Warnock, 2006a, 2006b).

Issues concerning the inadequacy of investor protection and governanceremain a major concern with investors and issuers within Thailand andmust act as a major impediment to further bond market development.An investigation of these issues by the IMF highlights many of thesedeficiencies, by providing information on the relative position of borrowersand lenders’ legal rights in the Asia-Pacific region compared with matureand more developed markets.3 Table 1 provides a summary.

Overall, Thailand’s score of 5 compares favourably with other countries inthe region but is below that of countries of more mature markets (score of 7).However, this overall score disguises some key externalities that require theattention of government. First, the time taken for contract enforcement isgenerally long and specifically in the case of Thailand (390 days) is secondlongest only to India (425 days). Second, while the cost of contractenforcement in Thailand (13.4% of the face value of debt) is not the highestin the region where the average is 22%, the length of the bankruptcy process(2.6 years) and the cost (38% of the estate) are the highest in the region. Inaddition, there is poor judicial efficiency (3.3 is the lowest in the region)despite reasonable accounting standards (6.4 compared with the G3measure of 6.6) and rule of law measures (6.3 compared with the G3average of 9.4). These areas all need to be addressed along with


improvements in the efficiency of the judicial process more generally andthrough specific measures such as the adoption of ISDA master agreementsin standardizing the diverse contractual agreements – and interpretations –that may otherwise exist.

A region-wide perspective on the obstacles, omissions and policydisparities has already been undertaken by Lejot, Arner, and Qiao (2006).The authors provide a comprehensive coverage of the legal, fiscal,regulatory and systematic reforms necessary for regional and domesticbond market development. Relevant reforms will be discussed in more detailin the conclusions.

The simultaneous development of domestic corporate and governmentbond markets has been the initial focus for many countries (Batten & Kim,2001; Fernandez & Klassen, 2006). This focus included strategies to buildinfrastructure, including settlement systems (Park & Rhee, 2006), establishreputable credit ratings (Kisselev & Packer, 2006) and establish benchmark

Table 1. The Importance of Legal Rights and Investor Protection.

Country Borrowers’

and Lenders’

Legal Rights

Index

Contract

Enforcement

Time (Days)

Contract

Enforcement

Cost (as

Percentage of

Debt)

Length of

Bankruptcy

Process

(Years)

Bankruptcy

Costs (as

Percentage of

Estate)

China 2 241 25.5 2.4 18

India 4 425 43.1 10 8

Korea 6 75 5.4 1.5 4

Malaysia 8 300 20.2 2.3 18

Thailand 5 390 13.4 2.6 38

Asia 5 286 22 4 17

Mature

markets

7 165 9 2 7

Country Accounting

Standards

Rule of Law Judicial

Efficiency

Contract

Repudiation

Expropriation

Risk

India 5.7 4.2 8 6.1 7.8

Korea 6.2 5.4 6 8.6 8.3

Malaysia 7.6 6.8 9 7.4 8

Thailand 6.4 6.3 3.3 7.6 7.4

Asia 6.5 5.7 6.6 7.4 7.9

G-3 6.6 9.4 9.7 9.5 9.9

Source: IMF (2005) (Chapter IV on ‘‘Recent Trends in Corporate Finance’’).


yield curves (Wooldridge, 2001).4 In part this need was driven by reluctanceon the part of local and regional issuers to issue in foreign currency due toexchange rate revaluation implications; many governments and corpora-tions suffered massive increases in the market value of their liabilitiesfollowing the depreciation of local currencies in the post–Asian Crisisperiod. Nonetheless, given the more stable macroeconomic environmentthat pervades the region in recent times, it is appropriate to now reconsiderthis view.

Such an approach is in effect an extension of regional bond marketinitiatives such as the Changmai Proposal at the meetings of the AsianCooperation Dialogue on the 23rd June 2003 in Changmai, Thailand, whichcalled for the development of an Asian or regional bond market (Pei, 2006)and the establishment of an Asian Bond Fund (Leung, 2006). Theseinitiatives have established the credentials of local governments as beingcommitted to the ongoing reform of local and regional financial markets.However, the failure of these regional markets to take-off is likely due to thepresence of non-viable domestic bond markets (Park & Park, 2004). Thus,building domestic bond markets through the simultaneous development ofglobal markets may prove to be a better overall strategy than tackling themore difficult task of coordinating regional bond market development.

There have been other initiatives within the region, which fit more clearlywith the global strategy advocated by McCauley and Park (2006). Initially,under the working group of the ASEAN+3 Asian Bond Market Initiative(ABMI), an additional focus was determining the effects of bond issuanceby foreign corporations and supranational institutions on the developmentof local markets. Multilateral development banks have a significant presencein the Australian foreign bond market and have issued in other regionalmarkets, especially Korea (Hoschka, 2006). Following this lead, foreigncorporations have also become more involved in domestic bond markets asissuers (Batten & Szilagyi, 2007), while foreign investors (Bae et al., 2006)are now beginning to build a presence.

Foreign bond markets have now been established in Australia and theSamurai market has been revived in Japan, while smaller markets areunderway in Korea, Hong Kong and Singapore. Foreign bond markets haveclear advantages for multinationals since they allow better risk managementby allowing the matching of foreign currency assets (from foreign direct orportfolio investment) with foreign currency denominated liabilities. Froma balance sheet and revaluation viewpoint, this strategy ‘‘termed naturalhedging’’ minimizes net foreign translation (currency) exposure. This is


especially important in Thailand for many multinational firms currentlyinvesting or considering further investment in manufacturing capability inareas such as automobiles and electronics.

Benefits also accrue to local investors who obtain more diverse yield andrisk offerings to what was previously available in local markets, while thereare clear macroeconomic benefits in the form of the alternate channellingof domestic savings. Investment corporations such as provident funds,social security and life insurance organizations are especially interested inthe quality and fixed rate side of the market.

This last point is especially important for many Asian countries (such asChina) where the savings build-up (mostly in bank deposits) may be linked tothe formation of bubbles in other asset markets, such as stock and propertymarkets. Finally, there are positive exchange rate benefits, especially forcountries suffering from excessive foreign reserve build-up: by borrowing indomestic currencies, and not in foreign currency that otherwise requiresconversion, multinational corporations alleviate balance of paymentspressures. The recent decision (2007) by China to allow foreign companiesto sell yuan denominated bonds accommodates both of these concerns. Next,we examine the state of international bond markets in Thailand.

3. THE THAI DOMESTIC AND INTERNATIONAL

BOND MARKETS

The scale and scope of international bond issuance can be regarded as abarometer of general development in the respective local bond market,although admittedly they are frequently substitutes for domestic issues.There is also the improved diversification provided to both local andinternational investors through the offer of quality, longer dated securitiesthan may be currently available in domestic markets. International bondsbeing denominated in non-local currency may also be more appealing toinvestors who require higher yield but are reluctant, or unable, to take onlocal currency risk. They also offer the advantage of globalizing a country’sbond market without necessarily needing the authorities to internationalizethe currency, which occurred in Australia (McCauley, 2006) and is nowtaking place in Korea (Batten & Szilagyi, 2007). Overall, the lessons andexperiences of these initially high-quality issuers can then serve as a templatefor issuance by less-credit worthy local issuers in international markets.


Burger and Warnock (2006b) report the scale and scope of domestic andinternational bonds. Table 2, taken from their study, reports the scale andscope of domestic and international bond issues in emerging Asia comparedwith other key emerging and world markets. Taken from this globalperspective, the total bonds outstanding in the emerging markets representonly a small component of the world bond market (about 7%). The ratio of

Table 2. Domestic and International Bond Markets in Thailand and theAsia-Pacific Region.

Total Bonds Outstanding Local Currency Bonds Outstanding

$

Billions

% in

World

Bond

Market

% of

Country’s

GDP

$

Billions

% in

World

Bond

Market

% of

Country’s

GDP

% of

Country’s

Total

Bonds

Emerging

markets

2183 7 38 1652 5.3 28 76

Latin America 596 1.9 34 314 1.0 18 53

Emerging

Asia

1124 3.6 40 1013 3.3 36 90

China 329 1.1 28 316 1.0 27 96

India 141 0.5 29 137 0.4 28 97

Indonesia 50 0.2 34 48 0.2 33 97

Korea 325 1.0 77 281 0.9 66 86

Malaysia 89 0.3 101 73 0.2 82 82

Pakistan 27 0.1 44 27 0.1 44 100

Philippines 32 0.1 45 16 0.1 22 50

Thailand 43 0.1 37 35 0.1 30 81

Taiwan 89 0.3 32 82 0.3 29 92

Financial

centres

91 0.3 36 55 0.2 22 61

Hong Kong 44 0.1 27 23 0.1 14 53

Singapore 46 0.1 54 32 0.1 37 69

Emerging

Europe

227 0.7 31 138 0.4 19 61

Other

emerging

markets

146 0.5 56 132 0.4 51 90

World 31168 100 105 28711 92 97 92

Note: The table reports the total domestic and local currency bonds issued by various emerging

and developed countries in US dollar billions and as a percent relative to each countries GDP

and relative to the total world bond market.

Source: Burger and Warnock (2006a).


total bonds outstanding relative to GDP of 38% for emerging markets,compared with 105% for the world, highlights the degree of their under-development – the key exceptions in the table are Korea and Malaysiawhose bond markets are almost the same size as the country’s GDP.However, there is an interesting variation due in part to the success of recentpolicy initiatives that have encouraged domestic bond market developmentin some countries and regions. For example, the emerging markets of Asiaare unique in that most bonds (90%) are issued in local currency (and indomestic markets), whereas significantly less are issued in local currency inemerging Europe (61%) and Latin America (53%). That is, while domesticbond markets have become well developed in Asia, issuers in emergingEurope and Latin America rely heavily on international bond markets forfunding.

The alternative to bond market financing, either in domestic orinternational markets, is financing through stock market issues and banks.To illustrate the changes in financing in the Asian region, it is informative toconsider the case of Thailand.5 Bank loan origination in Thailand remainsa main funding source for the economy (about 80% of GDP in 2006),although financing by bonds has increased steadily (from about 16%of GDP in 1997 to about 50% of GDP in 2006). This process ofdisintermediation highlights the success of regional initiatives aimed atenhancing bond market development that have been discussed by McCauleyand Park (2006), even though it is widely recognized that still much moremust be done to fully maximize the regions’ potential. As Eichengreen andLuengnaruemitchai (2004) noted: corruption, poor regulatory quality andthe absence of quality accounting standards continue to slow development.However, it is precisely for these reasons that it is important to pursue apolicy of both bank and bond market reform, since the intermediation roleof banks is especially important in developing economies.6

Developing better bond markets enables banks to find additional funding(other than deposits) and enable them to focus on risk transformation andintermediation activities such as asset securitization or property funds. Also,the financing of mega-projects can only be done through bank loans thatmay be guaranteed by third party MTBs (ADB, World Bank) or othercommercial banks. Overall, the presence of banking markets facilitatesthe development of a mature financial market by adding depth and diversity.

Despite these reservations, the Thai experience of bond market develop-ment has been impressive. Fig. 1 shows the growth in domestic bondissuance by government, corporate and financial sectors for the period from1990 to 2005. Beginning at less than USD 10 billion in 1990, the domestic


market now approaches USD 80 billion. Importantly, growth has beensignificant across all three sectors, although the government sector nowdominates the domestic bond market (and has done so since 1998), in partbecause of the need to finance domestic fiscal deficits.

A sectoral breakdown of Thai international bond issues for the period1990–2005 is provided in Fig. 2. Thai international bond issuance is nowrarely discussed or considered but was very important as a funding source inthe period prior to the Asian Crisis period of 1997–1998, when total issuespeaked at about USD 15 billion. International issues by financialcorporations remain the dominant sector, although issuance is now almost50% less than the USD 8 billion issued in 1999. While the government sectormaintains a steady program of international bond issuance, the interna-tional market now represents less than 10% of total outstandings.

This fact is clearer when one considers Fig. 3. This figure showsinternational bond issuance as a percentage of total issues (includingdomestic issues) by each of the three sectors. For example, in 2000 thefinancial sector issued almost 100% of its bonds in domestic markets. Until1997–1998, as much as 80% of government bonds issued were ininternational markets, while today the proportion is less than 10%. In fact

0.0

10.0

20.0

30.0

40.0

50.0

60.0

70.0

80.0

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Total Domestic Government Corporate Financial

Fig. 1. Domestic Bond Issuance by Sector (1990–2005) (Billions of USD). Note:

The figure shows total domestic bond issuance in US dollar billions (y-axis) by three

segments of the Thai economy: the government, corporate and finance sectors.

A total is also provided. The period covered is from 1994 to 2005 and is sourced from

the Bank for International Settlements (Table 16) BIS Quarterly Review (2006).


0.0

2.0

4.0

6.0

8.0

10.0

12.0

14.0

16.0

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Total International Government Corporate Financial

Fig. 2. Thai International Bond Issuance: Total Issuance (1990–2005). Note:

The figure shows international bond issuance in US dollar billions (y-axis) by

three segments of the Thai economy: the government, corporate and finance

sectors. A total is also provided. The period covered is from 1994 to 2005

and is sourced from the Bank for International Settlements (Tables 11–15) BIS

Quarterly Review (2006).

0.0

20.0

40.0

60.0

80.0

100.0

120.0

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

% Government % Corporate % Financial

Fig. 3. Thai International Bond Issuance: Sector as a Percentage of Total

Issues (1990–2005). Note: The figure shows international bond issuance as a

percentage of total international issues by three segments of the Thai economy: the

government, corporate and finance sectors. The period covered is from 1994 to 2005

and is sourced from the Bank for International Settlements (Tables 11–15)

BIS Quarterly Review (2006).


the proportion of international bonds issued by all three sectors has declinedas domestic bond markets have developed.

While some of these international bond issues (usually denominated in USdollars or yen) were unhedged with Thai issuers suffering the subsequenteffect of USD appreciation after the Asian Crisis, many were not and reliedupon the simultaneous development of the cross-currency USD/THB swapmarket (usually fixed rate USD to fixed rate THB). These swaps enabled thecreation of fixed rate THB borrowings at a time when there was littledomestic investor appetite for fixed long-term debt. We argue that the nextround of reforms should focus on the development of floating rate domesticinstruments. These could be arbitraged against forward foreign exchangecontracts to ensure market completeness.

4. THE BEHAVIOUR OF THAI BONDS AND

CREDIT SPREADS

4.1. Thai Bond Returns

The time series properties of individual bonds and their credit spreads(the spread of a risky bond over a riskless bond) has been the subject ofrecent empirical investigation. Especially noteworthy in recent times is thereduction in credit spreads to historic lows, not just in the Asian region, butacross all emerging markets, despite a variety of economic and politicalshocks. Addressing the first of these issues, Burger and Warnock (2006a)argued the importance that international investors place on the skewnessand volatility of bond returns. High variance and negative skewness areconsidered unattractive features and are present in both the unhedged andhedged USD returns from most emerging market bonds.

Key results from Burger and Warnock (2006a) are provided in Table 3.Looking at these data from a regional perspective, the variance of unhedgedUSD returns is greatest in Latin America (1.048), the emerging Asia(0.926) and other emerging markets such as Africa and the Middle East(0.567), while the lowest variance is in emerging Europe. Returns are mostnegatively skewed in Latin America (�1.62), other emerging markets(�0.62) and emerging Asia (�0.59). Hedging for USD exchange risksignificantly reduces the variance (e.g. a reduction from 0.926 to 0.399 in thecase of emerging Asia), although it has the opposite effect on skewness (e.g.an increase of negative skewness from �0.59 to �1.00 for emerging Asia).


It is especially noteworthy that within the Asian region there is extensivevariation in both variance and skewness, which underpins both the difficultyof regional market integration, while also highlighting the potential benefitsto investors who are able to adequately diversify their portfolios.

Looking at these data from the perspective of Thailand, the variance ofunhedged Thai bonds returns in terms of US dollars (1.048) is secondhighest in the region to Indonesia (3.245), although the skewness is morefavourable (0.07). Hedged US dollar returns have considerably less variance –suggesting that Thai baht–US dollar currency risk is a major factor thatmust be overcome – although the skewness is now negative (�0.35). Overall,Thai bonds have less appeal to international investors than bonds issued byother similarly rated countries (such as China).

While this may be true, there is considerable variation in the varianceand skewness of individual bonds, which is also linked to the currency ofdenomination. Table 4 reports the results of average monthly returns,variance and skewness for six Thai government bonds denominated in both

Table 3. The Mean, Variance and Skewness of Historical Returns inThailand and the Asia-Pacific Region.

Unhedged USD Returns Hedged USD Returns

Mean Variance Skewness Mean Variance Skewness

Emerging markets 0.004 0.809 �0.95 0.076 0.431 �1.35

Latin America �0.041 1.048 �1.62 0.049 0.665 �1.89

Emerging Asia 0.073 0.926 �0.59 0.088 0.399 �1.00

China 0.096 0.043 0.70 0.096 0.043 0.68

India 0.077 0.037 �1.43 0.119 0.024 �0.92

Indonesia �0.168 3.245 0.28 �0.081 1.370 �0.94

Korea 0.208 0.753 �2.86 0.144 0.227 �1.76

Malaysia 0.104 0.615 0.13 0.098 0.352 �0.38

Philippines 0.037 0.739 �1.05 0.100 0.321 �2.28

Thailand 0.160 1.048 0.07 0.143 0.454 �1.44

Financial centres 0.052 0.041 0.12 0.092 0.027 �0.35

Hong Kong 0.107 0.042 �0.02 0.109 0.041 �0.01

Singapore �0.003 0.040 0.26 0.076 0.013 �0.69

Emerging Europe �0.077 0.567 �0.62 0.084 0.233 �0.78

Other emerging markets �0.007 0.696 �1.46 0.092 0.427 �2.55

World ex US 0.006 0.483 �0.20 0.072 0.240 �0.92

Note: The table reports the mean, variance and skewness of hedged and unhedged returns for

bonds issued by various emerging and developed countries.

Source: Burger and Warnock (2006b).


US dollars and yen. The data available cover the period from September2002 to March 2006 and were sourced from either the Reuters or BloombergFixed Income databases. There were a maximum of 910 observations foreach bond, though on some days there were no prices available from eitherReuters or Bloomberg. There is very little trading in any of these bonds insecondary markets and most prices were indicative and based on dealers’quotes rather than actual trades.

Interestingly, and possibly not surprisingly, the bonds of longer maturityhave higher variance. However, there does not appear to be a significantdifference in variance between the US dollar and the yen denominatedbonds. Importantly for international investors, the skewness is positive andfor these six bonds ranges from 0.494 to 1.049. Thus, key features of Thaiinternational bonds for international investors – irrespective of currencydenomination – are that (1) they are positively skewed and (2) the degree ofpositive skewness of these international bonds is significantly higher thanthe skewness reported for either hedged or unhedged domestic Thai bonds(Table 3).

Table 4. The Variance and Skewness of Thai Government InternationalBonds Denominated in Yen and US Dollars.

Foreign Bond Issue

(Maturity Date, Coupon and

Currency)

Average

Monthly

Return

Average

Monthly

Variance

Average

Monthly

Skewness

Source:

Bloomberg

(B)/Reuters

(R)

THAILAND KINGDOM

03/27/2006 3.35% JPY

0.000 0.858 0.737 B

THAILAND KINGDOM

12/20/2006 2.85% JPY

0.000 0.793 0.629 B

THAILAND KINGDOM

12/21/2006 1.13% JPY

0.000 0.816 0.666 B

THAILAND KINGDOM

12/19/2008 1.70% JPY

0.001 0.703 0.494 B

THAILAND KINGDOM

4/15/2007 7.75% USD

0.002 0.769 0.592 B

THAILAND KINGDOM

9/30/2013 7.07% USD

0.003 1.024 1.049 R

Note: The table reports three key moments (mean return, variance and skewness) for six Thai

government international bonds denominated in yen and US dollars. The results were averages

calculated over a rolling one-month (22-day) period. The period is from September 2002 to

March 2006 and was sourced from either the Bloomberg (B) or the Reuters (R) Fixed Income

databases.


While this feature of international bonds is attractive for internationalinvestors, it is important to also note that both variance and skewness aretime-varying. This is very clear in Figs. 4 and 5, which plot the monthlyvariance and skewness of the 7.75% coupon USD Eurobond returns over

0

0.001

0.002

0.003

0.004

0.005

1 101 201 301 401 501 601 701

Fig. 4. Time-Varying Monthly Volatility (Standard Deviation) of the THAILAND

KINGDOM 4/15/2007 7.75% USD Eurobond. Note: The figure shows the daily

standard deviation calculated over a rolling one-month (22-day) period. The period

is from September 2002 to March 2006 and was sourced from the Bloomberg Fixed

Income database.

-3

-2

-1

0

1

2

3

1 101 201 301 401 501 601 701

Fig. 5. Time-Varying Monthly Skewness of the THAILAND KINGDOM 4/15/

2007 7.75% USD Eurobond. Note: The figure shows the daily skewness calculated

over a rolling one-month (22-day) period. The period is from September 2002 to

March 2006 and was sourced from the Bloomberg Fixed Income database.


the sample period (all the bonds displayed similar patterns). It is interestingto note that while the standard deviation tended to oscillate down overthe sample period (from around 0.004 to about 0.003), the correspondingskewness (both negative and positive) tended to increase (from below 71to above 72). Hence the time-varying variance and skewness will affectthe performance of international portfolios, and while some of this riskmay be diversifiable, the portfolio returns will be clearly unstable overtime.

4.2. Credit Spreads on International Bonds

Batten, Fetherston, and Hoontrakul (2002, 2006) have previously reportedthe results from modelling the credit spreads of a group of nine Asianinternational bonds. These bonds included those issued by China (threebonds), Korea (one bond), Malaysia (one bond), Philippines (three bonds)and Thailand (one bond).7 The dependent variable was the change in thecredit spread (yield) for each of these nine bonds, where the spread wasestimated by matching the maturity of each Asian bond with an equivalentmaturity US Treasury benchmark bond. Credit spread theory suggests thatcredit spreads on risky bonds are negatively related to the underlying risk-free interest rate (in this case a US Treasury bond) and an asset factor,proxied by the return on the local stock market index (see Batten & Hogan,2003; Batten, Hogan & Jacoby, 2005). Exchange rate variables are alsoconsidered to proxy macroeconomic stability.

Tables 5a and 5b report the key results for all bonds; although particularattention is paid to the 7.75% coupon 2007 Thai Eurobond. The resultsfrom these tables may be summarized as:

� The intercept term (a) reflects a risk premium that could exist on thespread return. However, the value was less than 0.01 (0.005 in the case ofthe Thai bond) suggesting that it was not important in explaining thespread return.� The interest rate factor (Y) was in fact both statistically and economicallysignificant in all nine cases. In the case of Thailand, the value of �2.39suggested that changes in the underlying US Treasury bond were thesingle most important factor that affected credit spreads. The negativevalue also suggested that the spread would increase or decrease oppositeto changes in the underlying US yield change (i.e. if US Treasury bondyields increased then the credit spread would decrease and vice versa).


� The interest rate variable to accommodate the change in the shape of theyield curve due to inflation expectations between a 30- and 2-year USTreasury bond (Y30�Y2) was generally statistically significant, with apositive sign suggesting that a steepening of the yield curve results in a risein the credit spread. In the case of the Thai bond, the value was 0.110.

Table 5a. The Credit Spreads of Key Asian International Bond Issuesagainst US Treasury Bonds.

Credit Spread Pair Credit Spread Equation (Independent Variables)

a bDYt cD(Y30�Y2)t dðDY Þ2t eDIt fDet

CHU04–US2 �0.004 �1.234 0.107

0.002 0.000 0.026

CHU06–US5 �0.005 �2.242 0.319 0.099 0.363

0.000 0.000 0.000 0.000 0.010

CHU08–US5 �0.007 �2.458 0.157 0.063

0.000 0.000 0.000 0.000

KOG08–US5 �0.006 �2.520 0.157 0.067

0.001 0.000 0.000 0.000

MYG09–US5 �0.001 �0.681 0.046 0.016

0.002 0.000 0.000 0.000

PHG08–US5 �0.412 �0.206 0.217

0.000 0.000 0.033

PHG19–US10 �3.639

0.000

PHU24–US10 �0.003 �3.726 �0.479 0.702

0.057 0.000 0.001 0.037

THU07–US5 �0.005 �2.389 0.110 0.084

0.001 0.000 0.072 0.000

Note: The table reports the key regression results from Batten et al. (2006, Table 3) for the credit

spread of East Asian issuers with US Treasury bonds (China 2004, 2006, 2008 maturities

(CHU04, CHU06, CHU08), Korea 2008 maturity (KOG080), Malaysia 2009 maturity

(MYG09), Philippines 2019 and 2024 maturities (PHG19 and PHU24) and Thailand 2007

maturity (THU07)) with near maturity US government Treasury bonds with 2- (US2), 5- (US5)

and 10-year (US10) maturities. Note that the designation ‘‘G’’ or ‘‘U’’ in the bond code refers to

whether the bond was a global bond or a Yankee bond, respectively. The data were sourced

from the Reuters Fixed Income database. A simplified model of their model is

DSt ¼ aþ bDY t þ cDðY 30 � Y 2Þt þ dðDY Þ2t þ eDI t þ fDet þ �t, where DSt is the change in the

credit spread (for the various Asian bond issues as described in Table 5b) at time t, DYt is the

change in the risk-free interest rate (identical in maturity to the riskless bond used to calculate

the spread), D(Y30�Y2)t is the change in the slope of the yield curve for 30- and 2-year

maturities, ðDY Þ2t is the change in the squared spot rate (rates with the same maturity as the

riskless bond), DIt is the change in the logarithm of the stock market index and Det is the change

in the spot exchange rate.


� The interest rate variable to accommodate possible convexity in the yieldcurve (Y2) was statistically significant in five of the nine cases, althoughthe actual coefficient was small and generally not economically significant.The value for the Thai bond was 0.084.� The asset factor (I) was only significant in three of the nine cases.A positive relationship suggests that a rise in the stock market is associa-ted with an increase in the spread and vice versa. This suggests thatthe linkage between stock and international bond markets was not soclear-cut. In the case of the Thai bond this variable was statisticallyinsignificant, suggesting there was no relationship between changes in theThai credit spread and the stock market.� The exchange rate variable (e) was not significant with the exception ofPhilippine bonds where a positive value suggests that a rise in theexchange rate (a depreciation of the peso) was also associated with anincrease in the credit spread and a fall in the stock index. In the case of theThai bond this variable was also not statistically insignificant, suggestingthere was no relationship between changes in the Thai credit spread andthe exchange rate.

These two sets of results suggest that Thai international bonds benefitfrom positive skewness, which should make them more attractive tointernational investors than bonds issued within Thailand. From the issuerperspective, the single most important factor affecting the change inthe credit spreads on international bonds – for all Asia-Pacific issuers not

Table 5b. Information on Sovereign Bonds of Asian Issuers.

Issuer Code Coupon Issued Maturity Rating

CHINA, PEOPLE’S REPUBLIC OF CHG08 7.3 12/9/1998 12/15/2008 BBB

CHINA, PEOPLE’S REPUBLIC OF CHU06 7.75 7/1/1996 7/5/2006 BBB

CHINA, PEOPLE’S REPUBLIC OF CHU04 6.5 2/2/1994 2/17/2004 BBB+

FEDERATION OF MALAYSIA MYG09 8.75 5/27/1999 6/1/2009 BBB�

KOREA, REPUBLIC OF KOG08 8.875 4/7/1998 4/15/2008 A�

PHILIPPINES, REPUBLIC OF PHU24 9.5 10/14/1999 10/21/2024 BB+

PHILIPPINES, REPUBLIC OF PHG19 9.875 1/6/1999 1/15/2019 BB+

PHILIPPINES, REPUBLIC OF PHG08 8.875 4/2/1998 4/15/2008 BB+

THAILAND, KINGDOM OF THU07 7.75 4/10/1997 4/15/2007 BBB�

Note: The table shows key East Asian issues in international markets, sourced from the Reuters

Fixed Income database, used by Batten et al. (2006). It also reports the coupon of the bond, the

date of issue and maturity and the credit rating. Note that the designation ‘‘G’’ or ‘‘U’’ in the

bond code refers to whether the bond was a global bond or a Yankee bond, respectively.


just Thailand – is the negative relationship with changes in US Treasurybond yields. Also, exchange rate changes are not especially importantfactors. This is especially important for international investors sincethey would hope that buying a sovereign bond denominated in a foreigncurrency (such as USD and yen) would avoid local currency volatilityspillover effects.

5. CONCLUSIONS

After the 1997 financial crisis, to reduce the dependence by the corporatesector on the bank funding channel and stock market ventures for refinancingand investment, East Asian policymakers developed a financial market visionthat encompassed three key dimensions for bond market development:

(1) an enhanced, but not necessarily integrated series of domestic bondmarket;

(2) a regional bond market denominated in regional currencies for regionalintegration; and

(3) a global market where national bond markets are developed withultimate goal of global financial integration and foreign participation.

Enhancing global integration requires greater foreign participation in thedomestic market (Burger &Warnock, 2006a), which incorporates additionalissuance activities of multilateral development banks (Hoschka, 2006),foreign cooperations (Batten & Szilagyi, 2007), the engagement of foreigninvestors (Bae et al., 2006) and an expanded role for domestic issuers ininternational markets. This contribution focuses on the international bondmarket, which we believe holds important benefits for both issuers andinvestors in the form of better risk and maturity management than thatcurrently exists in many domestic markets.

Specifically, we focus on Thailand’s issuance activities in internationalmarkets. Thailand is an excellent candidate for such a study since after the1997 crisis it was required under IMF guidelines to become more receptiveto foreign investment and capital market participation. Specifically, we raisethe importance of bond return volatility and skewness as an impediment togreater involvement by international investors. We also highlight the time-varying nature of both variance and skewness of bond returns, which canonly be overcome through government policy that focuses upon stabilizingthe macroeconomic environment and not simply relying upon theenhancement of domestic and regional financial market infrastructure.


Nonetheless, we do not want to understate the importance of qualityinfrastructure in encouraging participation in bond markets by investorsand issuers and do recognize the accomplishments made in recent year,including the establishment of better dealing, trading and settlement systems(Park & Rhee, 2006; Wooldridge, 2001) and enhanced regional cooperationsuch as the development of the ASEAN+3 bond fund (Leung, 2006; Pei,2006). However, much more still needs to be done, including improving thereliability of credit ratings, enhancing derivatives markets to facilitate riskand maturity management and better investor protection and governance(Kisselev & Packer, 2006; Bae et al., 2006). Enhanced risk managementcapability can only arise through the development of deeper markets forfloating rate instruments, such as Forward Rate Agreements (FRAs) andthe ability to swap cash-flow type and currency, though instruments such ascross-currency swaps. These are invariably traded over-the-counter (OTC)and remain the province of leading foreign banks. Reform to facilitatetrading is therefore essential and requires an ongoing commitment bygovernment in conjunction with industry. Eventually, floating rate futuresmarkets could be developed modelled on the Singapore and Hong Kongexperience. However, it is important to recognize that these markets areused by financial rather than non-financial institutions for hedging andtrading purposes.

The need for developing alternate investment vehicles to house the vastbuild-up of savings in each individual country and across the region cannotbe underestimated. Possibilities include the development of better, deeperand more diverse markets in asset-backed securities – as has occurred inKorea – where the assets are traditional mortgages or infrastructure such asthose in the German Pfandbrief market (see Fernandez & Klassen, 2006).One possibility to better motivate reform would be to set a national targetfor each domestic bond market to approach a size that is equal to 100% ofGDP – a ratio common in many developed markets.

In conclusion, we argue that the next step necessary for bond marketdevelopment throughout the East Asia region is to direct policy towards theinternationalization of domestic markets, instead of a single policy aimed atregionalization. This proposal is consistent with McCauley and Park (2006).The two important steps towards this goal involve making domestic marketsmore attractive to foreign issuers, in line with the recommendations ofHoschka (2006), Bae et al. (2006) and Batten and Szilagyi (2007), and toencourage local issuance in international markets. Both these steps willinvariably lead to further internationalization of domestic currency(McCauley, 2006), as has occurred in Australia and is developing in Korea.


Consequently, exchange rate policies must be more accommodating (such asminimizing capital flow restrictions and providing withholding tax exemp-tions). Overall, this positive economic climate will lead to positive nationalbenefits such as the reduction in the risk attached to investment, whichshould further encourage foreign direct investment.

NOTES

1. For example, the ASEN Bond Fund, based on a basket currency incorporatingthe yen, the euro, and the US dollar was proposed and implemented in 2004. ASEMtask force for closer economic partnership between Asia and Europe, Final reportand recommendations presented to the ASEM V summit in Hanoi, October 8–9,2004.2. When the BoT buys USD in foreign exchange markets, it must sell the

equivalent amount of THB in spot or forward markets. When THB is sold in spotmarkets, the authorities must issue domestic bonds to fund the short THB cashposition and also then invest the US dollar proceeds. Detailed information on thescale and scope of the Thai foreign exchange market is provided in BIS (2005).3. See Lu and Batten (2001) and Batten and Szilagyi (2004) for further discussion.4. The importance of establishing benchmark yield curves is critical for corporate

bond pricing although liquidity related distortions exist even in developed marketssuch as Japan (In, Batten & Kim, 2003).5. Batten and Szilagyi (2003) and Szilagyi and Batten (2004) provide a detailed

discussion of corporate bond market development in Japan.6. As Eichengreen (2006) noted: Banks provide underwriting services for

prospective domestic issuers, advising the issuer on the terms and timing ofthe offer; they provide bridge finance in the period when the marketing of bondsis still underway; they provide distribution channels for government bonds andform an important part of the primary dealer network; their institutional supportmay also be conducive to secondary-market liquidity; and finally, and most directly,banks owing to their relatively large size can be major issuers of domestic bondsthemselves.7. See Table 5b for details of each bond.

REFERENCES

Bae, K.-H., Yun, Y.-S., & Bailey, W. (2006). Determinates of bond holdings by foreign investors

(pp. 102–128). BIS Papers No. 30. Bank for International Settlements, Basle.

Batten, J., Fetherston, T., & Hoontrakul, P. (2006). Factors affecting the yields of Asian issues

in international bond markets. Journal of International Financial Markets, Institutions

and Money, 16, 57–60.

Batten, J., & Hogan, W. (2003). Time-variation in the credit spreads on Australian Eurobonds.



Batten, J., Hogan, W., & Jacoby, G. (2005). Measuring credit spreads: Evidence from

Australian Eurobonds. Applied Financial Economics, 15, 651–666.

Batten, J., & Kim, Y.-H. (2001). Expanding long-term financing through bond market

development: A post crisis policy task. In: Y.-H. Kim (Ed.), Government bond markets in

Asia. Manila: Asian Development Bank (http://www.adb.org/Documents/Conference/

Govt_Bond_Market/default.asp).

Batten, J., & Szilagyi, P. (2003). Why Japan needs to develop its corporate bond market.

International Journal of the Economics of Business, 11, 85–110.

Batten, J., & Szilagyi, P. (2007). Domestic bond market development: The Arirang bond

experience in Korea. World Bank Research Observer, 22, 165–195.

Batten, J. A., Fetherston, T., & Hoontrakul, P. (2002). Modelling emerging market issues in

international bond markets: Evidence from Thailand. Asian Economic Journal, 16(4),

379–397.

Benzie, R. (1992). The development of the international bond markets. Basle, Switzerland: Bank

for International Settlements, Monetary and Economic Department.

BIS (Bank for International Settlements). (2005). Triennial Central Bank survey of foreign

exchange and derivatives market activity. Basle: BIS.

Burger, J., & Warnock, F. (2006a, September). Local currency bond markets. NBER Working

Paper 12552. National Bureau of Economic Research, Cambridge, MA.

Burger, J., & Warnock, F. (2006b, September). Foreign participation in local currency bond

markets. NBER Working Paper 12548. National Bureau of Economic Research,

Cambridge, MA.

Eichengreen, B. (2006, August). The state of bond markets in Latin America. Working Paper,

Department of Economics, University of California, Berkeley, USA.

Eichengreen, B., & Luengnaruemitchai, P. (2004). Why doesn’t Asia have bigger bond markets?

NBERWorking Paper 10576. National Bureau of Economic Research, Cambridge, MA.

Emery, R. F. (1997). The bond markets of developing East Asia. Colorado: Westview Press

Boulder.

Fernandez, D. G., & Klassen, S. (2006). Choice of currency by East Asia bond issuers (pp. 129–143).

BIS Papers no. 30. Bank for International Settlements, Basle, Switzerland.

Herring, R. J., & Chatusripitak, N. (2000). The case of the missing market: The bond market

and why it matters for financial development. Paper Presented at the Asian Development

Bank Institute/Wharton Financial Institutions Center seminar on financial structure for

sustainable development in post-crisis Asia, May 26, Tokyo.

Hoschka, T. C. (2006). Local currency financing – The next frontier for MTBs. ERD Working

Paper 078. Asian Development Bank, Economics and Research Department, Manila.

IMF (International Monetary Fund). (2005). Development of corporate bond markets in

emerging markets. In: Global financial stability report: Market development and issues

(September). Washington, DC, USA.

In, F., Batten, J., & Kim, S. (2003). What drives the Japanese yen Eurobond term structure.

Quarterly Review of Economics and Finance, 43, 518–541.

Jiang, G., & McCauley, R. (2004). Asian local currency bond markets. In: BIS Quarterly Review

(June), Basle, Switzerland: Bank for International Settlements.

Kim, Y.-H. (1999). Creating long-term mortgage-backed bond markets in Asian developing

economies – A postcrisis reform agenda. Manila, Philippines: Asian Development Bank.

Kisselev, K., & Packer, F. (2006). Minding the gap in Asia: Foreign and local currency ratings

(pp. 19–39). BIS Papers no. 30. Bank for International Settlements, Basle.


Lejot, P., Arner, D., & Qiao, L. (2006). Missing links: Regional reforms for Asia’s bond

markets. Asia Pacific Business Review, 12(3), 309–331.

Leung, J. (2006). Developing bond markets in Asia: Experience with ABF2 (pp. 74–79). BIS

Papers no. 26. Bank for International Settlements, Basle, Switzerland.

Lu, J., & Batten, J. (2001). The implementation of OECD corporate governance principles in

post crisis Asia. The Journal of Corporate Citizenship, 4, 47–62.

McCauley, R. (2006). Internationalising a currency: The case of the Australian dollar.

In: BIS Quarterly Review (December), 41–54, Basle: Bank for International Settlements.

McCauley, R., & Jiang, G. (2004). Diversifying with Asian local currency bonds.

In: BIS Quarterly Review (September), Basle: Bank for International Settlements.

McCauley, R., & Park, Y.-C. (2006). Developing the bond markets of East Asia: Global, regional

or national (pp. 19–39). BIS Papers no. 30. Bank for International Settlements, Basle.

Park, D. K., & Park, Y.-C. (2004). Creating regional bond markets in East Asia: Rationale and

strategy. In: T. Ito & Y. C. Park (Eds), Developing Asian bond markets. Canberra: Asia

Pacific Press.

Park, D. K., & Rhee, G. (2006). Building infrastructure for Asian bond markets: Settlement and

credit rating (pp. 202–221). BIS Papers no. 30. Bank for International Settlements, Basle.

Pei, C. (2006). Asian financial cooperation: Priority to develop bilateral bond markets.

Emerging Markets Finance and Trade, 41, 75–82.

Rhee, G. S. (2003). Regional bond markets: Rationale, concerns and roadmap. ADB Working

Paper. Asian Development Bank, Manila.

Rhee, G. S. (2004). The structure and characteristics of East Asian bond markets. In: T. Ito &

Y. C. Park (Eds), Developing Asian bond markets. Canberra: Asia Pacific Press.

Schinasi, G. J., & Smith, T. R. (1998). Fixed income markets in the United States, Europe, and

Japan: Some lessons for emerging markets. IMF Working Paper 173. International

Monetary Fund, Washington, DC, USA.

Szilagyi, P., & Batten, J. (2003). Disintermediation and bond market development in Japan. In:

J. J. Choi & T. Hiraki (Eds), Japanese finance: Corporate finance and capital markets in

changing Japan [special issue]. International Finance Review, 4, 255–281.

Szilagyi, P., & Batten, J. (2004). Corporate governance and financial system development: Asia-

Pacific in comparative perspective. Journal of Corporate Citizenship, 11, 49–64.

Wooldridge, P. D. (2001). The emergence of new benchmark yield curves. BIS Quarterly Review

(December), 48–57.


Asia-Pacific Financial Markets, Volume 8: Integration, Innovation and Challenges

Documents

Transcript of Asia-Pacific Financial Markets, Volume 8: Integration, Innovation and Challenges