University of Western Sydney

Intraday Return, Volatility and Liquidity: An Investigation of the Market Microstructure of the Chinese Stock Market

A thesis

submitted to the School of Economics and Finance of University of Western Sydney

in fulfilment of the requirements for the degree of

Doctor of Philosophy (Economics and Finance)

By

Mingyuan Guo

Thesis Supervisor: Dr. Gary Tian

Thesis Co-supervisor: Dr. Craig Ellis

Sydney, Australia

May 2006

DEDICATION

This thesis is dedicated to the people who bring this thesis to completion My father Shixian Guo Who passed away during my study and My mother Baozhu Xian

Special dedication to My wife Ying Bao and My daughter Yiran Guo

ii ACKNOWLEDGEMENTS

Firstly, I must give my deepest gratitude to my outstanding and long suffering supervisor, Dr. Gary Tian. I am really fortunate to have the chance to learn from him and work with him. He has been an absolute inspiration and the epitome of encouragement, patience, talent and helpfulness. The time he has spent assisting me through discussions, suggestions and advice has been priceless. I appreciate the freedom I was allowed and his perpetual support and belief in my ability throughout the years. It has been overwhelming. I am extremely indebted to him for his generosity in providing part of his own research fund for my financial assistance. My deepest admiration also goes to my co-supervisor, Dr. Craig Ellis, for his time, valuable comments, great help and advice in my study, especially for providing detailed comments on my dissertation. Thank you both for your suggestions, critiques and guidance throughout my Ph.D. research. I couldn’t have done it without you.

In addition, I would like to thank Dr. Roger Ham from the University of

Western Sydney for his class note on applied econometrics, which has given me a deeper understanding of financial time series. To Dr John Ablett, Dr Sudhir Lodh and Mr Edward Mariyani-Squire, thank you for providing me with teaching opportunities in corporate financial management and microeconomics throughout the past two years. The experiences I’ve gained have been invaluable.

I want to express thanks to the College of Law and Business of UWS, which has provided me with a Fee Relief Scholarship for the past three years. To other research and administration staff at the School of Economics and Finance, but especially Professor P.N. (Raja) Junankar, Dr Kaven Daly, Ms Jo Roger, Ms Leanne

iii Falconer, Ms Patricia Fournier and Ms Olga Becroft, thanks for making life easier throughout the whole research program. I also need to thank Ms Claire Aitchison at the Office of the Dean of Students for her reading and suggestions on my thesis structure, Mr Mark Reed for editing and proof-reading assistance, and the research office at UWS for its continued support during the past three years.

My family has been the source of my energy throughout this sometimes difficult journey. I am grateful to my parents for their support and encouragement in practically every aspect of my life. I am especially indebted to my father for priceless encouragement and moral support during all of my education. The boundless moral support, faith, patience and love of my wife Ying Bao and my daughter Yiran Guo have been equally heartening. Thank you for standing by me and making the sacrifices you have made to create the warm, familial environment that has sustained me during this chapter of my life.

Lastly, I want to take this opportunity to thank all the gifted staff at the

School of Economics and Finance at the University of Western Sydney. I’ve been very fortunate to have the opportunity to work with you.

iv Intraday Return, Volatility and Liquidity: An

Investigation of the Market Microstructure of

the Chinese Stock Market

ABSTRACT

This thesis examines the characteristics of market microstructure on the

Chinese stock exchanges (the Shanghai Stock Exchange and the Stock

Exchange). Analysis is based on using intraday 5-minute data covering a three-year period (2000 to 2002). The study focuses on empirical analysis and statistical testing of the Chinese stock market in three parts.

In the first part (Chapter 3), I study the behaviour of intraday returns and return volatility. The statistical properties and systematic characteristics of intraday returns and return volatility of the Shanghai Composite Index and the Shenzhen

Component Index are explored. Several parametric and non-parametric tests are used to investigate if there are systematic patterns in the Chinese stock market. Overall findings indicate that the Chinese stock market returns data is non-normal and has a fat-tailed distribution. The intraday returns broadly follow a U-shaped pattern

(Wood, McInish and Ord, 1985) while the volatility of returns broadly follows an L- shaped pattern (Tang and Lui, 2002). Examinations by subperiods, wider and finer than 5-minute data, indicate that there exist intraday variations for each subperiod, and each 1-minute and 10-minute period, which are consistent with the intraday 5- minute return characteristics. The results also provide the evidence that the return variance in the active trading period (open-to-close) of the stock market is larger than

v that in the nontrading period (close-to-open). In this chapter, volatilities of interday

24-hour returns based on 5-minute intervals are calculated and the variance ratio test is used to show that high volatility of intraday returns at the market opening is not completely due to the trading mechanism (call auction) but the overnight trading halt.

In the second part (Chapter 4), I examine the intraday behaviours of the bid/ask spread and depth and their determinants on the limited order-driven market.

The analyses are based on a set of sixty individual stocks from the Shanghai Stock

Exchange using intraday 5-minute data. Since spread and depth are two dimensions of market liquidity, both variables should play important roles in explaining the market liquidity and their components. My study shows, firstly, that the intraday 5- minute bid/ask spreads display an L-shaped pattern and the depths exhibit an inverted L-shaped pattern. This is in striking contrast to the findings of the overall U- shape of the spread by McInish and Wood (1992) and Madhavan et al. (1997) in the

US market and Brockman and Chung (1998) and Ahn and Cheung (1999) in the

Hong Kong market. The existence of the L-shaped spread pattern (inverse L-shaped depth) on the Shanghai Stock Exchange means that specialist activity alone cannot account for the widening of spreads (lowest depth) at the opening of the call auction and continuous trading. This existence is attributed to the factor of asymmetric information and plays an important role in the patterns for liquidity. The findings are consistent with the CBOE (Chan et al., 1995).

Secondly, the chapter documents the interday liquidity patterns that still show differences in spreads and depths across days of the week but the strength of these differences is much less than across intervals of the day, similar to previous studies based on US data. The simultaneous occurrence of relatively wide spreads and lower depth on Mondays is consistent with the predictions of Foster and Viswanathan

vi (1990). Higher Monday opening spreads are probably due to information accumulation of non-trading.

Thirdly, analysis shows that the striking L-shaped pattern of the bid/ask spreads has a positive relationship with volatility but negative relationship with stock price and trading volume while the reverse L-shaped pattern of the depths has a negative relationship with stock price and volatility but a positive relationship with trading volume in the regression model.

Finally, there is an evident negative relationship between the spread and the depth at the opening of the trading day based on the statistical analysis.

Overall, the study suggests that the determinant of information asymmetries, through time and across traders, plays a key role in generating observed liquidity variations.

In the third part (Chapter 5), I apply the GARCH model to the dynamic volatility of intraday return using a set of twenty-six individual stocks in my study.

The main purpose of this chapter is to explore the issue of whether the return volatility in high-frequency data can be described by the GARCH(1,1) specification, and whether the GARCH modeling captures the effects of temporal dependence in trading volume or bid/ask spreads for individual stocks on the Chinese stock market.

The study indicates that all individual stocks in the conditional variance model exhibit the strong persistence of the GARCH effects. Similar to Rahman, Lee and

Ang (2002), the AR(1)-GARCH(1,1) model best describes the volatility of intraday returns. For most stocks, the tests on the squared standard residuals are not significant at the five percent level. Current volatility can be explained by past volatility that tends to persist over time. The findings of this chapter further indicate that the AR(1)-GARCH(1,1) model successfully accounts for the nonlinear

vii dependencies (volatility clustering) in the individual stocks. However, the AR(1)-

GARCH(1,1) model may not account for the linear dependencies (return clustering) in the high-frequency data.

Further, based on the model of MDH, the inclusion of intraday contemporaneous/lagged trading volume or bid/ask spread as a mixing variable for information arrival in the conditional variance model does help in explaining the

GARCH effects in the stock return series. Most of the coefficients for contemporaneous/lagged volume or bid/ask spread are positive but small for the

GARCH(1,1). Thus, the GARCH effects do not disappear as a result of this inclusion, which means that the persistence in volatility remains strong. The observed results are contradictory to the findings of Lamoureux and Lastrapes (1990) but in accordance with those of Rahman, Lee and Ang (2002).

viii

Contents

Dedication …………………………………………………………………….. ii

Acknowledgements …………………………………………………………... iii

Abstract ……………………………………………………………………….. v

List of Abbreviations …………………………………………………….…... xii

List of Tables …………………………………………………….…………… xiii

List of Figures ………………………………………………….…………….. xv

Certification …………………………………………………………………... xvii

Chapter 1 Introduction …………………………………………………….. 1

1.1 Background ……………………………………………………………. 1 1.2 Motivation ……………………………………………………………... 4 1.3 The Purpose and Contribution …………………………………………. 6

Chapter 2 China Stock Market and Market Microstructure …………… 13

2.1 Introduction ……………………………………………………………. 13 2.2 Stock Market Indices …………………………………………………... 17 2.3 Listing Requirements ………………………………………………….. 19 2.4 Special Features ………………………………………………………... 20 2.5 Settlement ……………………………………………………………… 22 2.6 Trading Systems and Market Microstructure ………………………….. 22 2.6.1 Trading Systems ……………………………………………….. 23 2.6.2 Two Trading Sessions …………………………………………. 23 2.6.3 Price Limit and Trading Rules ………………………………… 25

Figures ………………………………………………………………………. 28

Chapter 3 Behaviour of Intraday Return and Return Volatility ……….. 29

3.1 Introduction …………………………………………………………… 29 3.2 Literature Review ………………………………………………….….. 32 3.3 Dataset and Trading Systems …………………………………………. 37 3.4 Methodology ………………………………………………………….. 38 3.4.1 Definition of the Returns ……………………………………… 38 3.4.2 Hypotheses Development ……………………………………... 40 3.4.3 Testing for the Intraday Patterns ……………………………… 44 3.5 Empirical Results of Intraday Return Patterns ………………………... 47

ix 3.5.1 Intraday Return during the Period …………………………….. 47 3.5.2 Intraday Return during the Weekdays ………………………… 51 3.5.3 Test Results …………………………………………………… 52 3.6 Empirical Results of Intraday Return Volatility ……………………… 55 3.6.1 Intraday Return Volatility during the Period …………………. 56 3.6.2 Intraday Return Volatility during the Weekdays ……………... 57 3.6.3 Test Results of the Volatilities ………………………………… 58 3.7 Explanations of the L-shaped Volatility Pattern ……………………… 60 3.8 Volatility in Trading and non-Trading Periods ……………………….. 62 3.9 Call Auction and Continuous Trading ………………………………... 65 3.10 Conclusion ………………………………………………………….… 67

Tables ………………………………………………………………………... 70 Figures ……………………………………………………………………….. 78

Chapter 4 Inter-Temporal Behaviour and the Determinants of Bid/Ask Spread and Depth ……………………………………………… 89

4.1 Introduction ……………………………………………………………. 89 4.2 The Background and Literature Review ………………………………. 93 4.3 Data and Methodology ………………………………………………… 100 4.3.1 Data …………………………………………………………….. 100 4.3.2 Methodology …………………………………………………… 102 4.4 Empirical Results ………………………………………………………. 107 4.4.1 Summary Statistics …………………………………………….. 107 4.4.2 Regression Results ……………………………………………... 111 4.4.3 Discussion of the Bid/Ask Patterns and Depth ………………… 116 4.5 Summary and Conclusion ……………………………………………… 119

Tables ………………………………………………………………………... 122 Appendix …………………………………………………………………….. 128 Figures ……………………………………………………………………….. 129

Chapter 5 Intraday Return Volatility, Trading Volume and Bid/Ask Spread: GARCH Model Application ………………….……… 130

5.1 Introduction ……………………………………………………………. 130 5.2 Background and Data ………………………………………………….. 134 5.2.1 Background ……………………………………………………. 134 5.2.2 Data ……………………………………………………………. 137 5.3 Research Hypotheses and Methodology ………………………………. 140 5.3.1 Hypotheses …………………………………………………….. 140 5.3.2 Methodology …………………………………………………... 143 5.4 Statistical Analysis and Model Identification …………………………. 149 5.4.1 Return Series …………………………………………………... 149 5.4.2 Trading Volume Series ……………………………………… 152 5.4.3 Bid/Ask Spread Series ………………………………………… 153 5.5 Empirical Results ……………………………………………………… 154 5.5.1 Estimation of GARCH Model without Volume and Spread Series …………………………………………………………... 154

x 5.5.2 Estimation of GARCH Model with Trading Volume …………. 156

5.5.3 Estimation of GARCH Model with Lagged-one Trading Volume ………………………………………………………… 158 5.5.4 Estimation of GARCH Model with Bid/Ask Spread ………….. 161 5.5.5 Estimation of GARCH Model with Lagged-one Bid/Ask Spread ………………………………………………………….. 163 5.6 Conclusion …………………………………………………………….. 165

Tables ………………………………………………………………………... 168 Appendix …………………………………………………………………….. 181

Chapter 6 Summary and Conclusions ……………………………………. 182

Bibliography …………………………………………………………………... 187

xi

List of Abbreviations

ANOVA Analysis of variance AR Auto Regressive ARCH Auto Regressive Conditional Heteroskedasticity ARIMA Auto Regressive Integrated Moving Average CCPCC the Chinese Communist Party Central Committee CSCCRC China Securities Central Clearing & Registration Corporation CSRC China Securities Regulatory Commission EMH Efficient Market Hypothesis GARCH Generalized Autoregressive Conditional Heteroskedasticity ID Identification IPO Initial Public Offer IQR Inter-quartile Range MDH Mixture of Distributions Hypothesis NYSE New York Stock Exchange OLS Ordinary Least-Squares PRC The Peoples Republic of China QFII Qualified Foreign Institutional Investors RMB Renminbi SDCC Shenzhen Depository and Clearing Corporation SHSE Shanghai Securities Exchange SHZE Shenzhen Securities Exchange SIRCA Securities Industry Research Centre of Asia-Pacific, Australia SMEB Small and Medium Enterprise Board WTO World Trade Organization

xii

List of Tables

Chapter 3

Table 3.1 Descriptive statistics of the intraday 5-minute returns in percentages of the Shanghai Composite Index, 2000 – 2002 70 Table 3.2 Descriptive statistics of the intraday 5-minute returns in percentages of the Shenzhen Component Index, 2000 – 2002 ……………………………………………………………….. 71 Table 3.3 Mean intraday 5-minute returns by trading time and weekday in percentages of the Shanghai Composite Index, 2000 – 2002 …………………………………………………………. 72 Table 3.4 Mean intraday 5-minute returns by trading time and weekday in percentages of the Shenzhen Component Index, 2000 – 2002 …………………………………………………………. 73 Table 3.5 Mean 5-minute return standard deviations of the Shanghai Composite Index, 2000 – 2002 ……………………………... 74 Table 3.6 Mean 5-minute return standard deviations of the Shenzhen Composite Index, 2000 – 2002 ……………………………... 75 Table 3.7 Descriptive statistics for the returns of the trading and non- trading period of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002 (in percentages) ... 76 Table 3.8 Interday 24-hour returns, return volatility and variance ratio test, 2000 – 2002 ……………………………………………. 77

Chapter 4

Table 4.1 Sixty Individual Stocks from the Shanghai Stock Exchange, 2000 – 2002 …………………………………………………. 122 Table 4.2 Summary statistics of the average intraday 5-minute spreads and depths, 2000 – 2002 …………………………………….. 123 Table 4.3 Descriptive statistics and correlation matrix of variables of 5- minute intervals, 2000 – 2002 ………………………………. 124 Table 4.4 Regression results for spread of the 5-minute intervals …….. 125 Table 4.5 Regression results for depth of the 5-minute intervals ……... 126 Table 4.6 Correlation between spreads and depths of 60 individual stocks, 2000 – 2002 ………………………………………… 127 Appendix List of Sixty Individual Stocks from the Shanghai Stock 4.1 Exchange, 2000 – 2002 …………………………………….. 128

Chapter 5

Table 5.1 Descriptive statistics of intraday 5-minute log returns (in percentages) for twenty-six individual stocks, 2000 – 2002 ... 168 Table 5.2 Tests of intraday 5-minute log returns (in percentages) for twenty-six individual stocks, 2000 – 2002 …………………. 169 Table 5.3 Unit root tests of intraday 5-minute log returns (in percentages), log volumes and log bid/ask spreads for

xiii twenty-six individual stocks, 2000 – 2002 …………………. 170 Table 5.4 Descriptive statistics of intraday 5-minute log trading volumes for twenty-six individual stocks, 2000 – 2002 ……. 171 Table 5.5 Tests of intraday 5-minute log trading volumes for twenty- six individual stocks, 2000 – 2002 …………………………. 172 Table 5.6 Descriptive statistics of intraday 5-minute log bid/ask spreads for twenty-six individual stocks, 2000 – 2002 …….. 173 Table 5.7 Tests of intraday 5-minute log bid/ask spreads for twenty-six individual stocks, 2000 – 2002 ……………………………... 174 Table 5.8 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) for Equation 5.4 …………... 175 Table 5.9 AR(1)-GARCH(p,q) model estimation of intraday 5-minute returns for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) …………………………….. 176 Table 5.10 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with contemporaneous volume for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) for Equation 5.5 ……………………………………... 177 Table 5.11 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with lagged-one volume for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) for Equation 5.6 ………………………………………………... 178 Table 5.12 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with contemporaneous bid/ask spread for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) for Equation 5.7 …………………………………….... 179 Table 5.13 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with lagged-one bid/ask spread for twenty-six individual stocks on the Shanghai Stock Exchange (2000 – 2002) for Equation 5.8 ……………..……………………….. 180 Appendix Twenty-six Chinese Individual Stocks on the Shanghai Stock 5.1 Exchange …………………………………………………… 181

xiv List of Figures

Chapter 2

Figure 2.1 The Daily movement of the Chinese Stock market indices: Shanghai Composite Index and Shenzhen Component Index from 1991 to 2003 ………………………………….. 28

Chapter 3

Figure 3.1 Time plots of intraday 5-minute index closes and log returns (in percentages) of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ……….. 78 Figure 3.2 Intraday mean 5-minute returns and cumulative mean 5- minute returns (in percentages) of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 … 79 Figure 3.3 Mean 5-minute returns (in percentages) by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ………………………………………... 80 Figure 3.4 Intraday cumulative mean 5-minute returns (in percentages) by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 … 81 Figure 3.5 Intraday 5-minute return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ………………... 82 Figure 3.6 Intraday 5-minute return volatility based on standard deviations by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ……….. 83 Figure 3.7 Intraday mean 5-minute returns of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002 (in percentages) …………………….. 84 Figure 3.8 Intraday cumulative mean 5-minute returns of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002 (in percentages) ………….. 85 Figure 3.9 Intraday mean 5-minute return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002 ……………………………………………………….. 86 Figure 3.10 Intraday mean 1-minute returns, cumulative mean returns (in percentages) and return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ………………... 87 Figure 3.11 Intraday mean 10-minute returns, cumulative mean returns (in percentages) and return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002 ………………... 88

xv Chapter 4

Figure 4.1 Intraday 5-minute and interday bid/ask spread patterns and bid/ask depths ……………………………………………... 129

xvi

Certification

I hereby declare that the PhD thesis entitled “Intraday Return, Volatility and

Liquidity: An Investigation of the Market Microstructure of the Chinese Stock

Market” which I am submitting as well as the research reported in this thesis is original and a product of my own exertion, except where otherwise acknowledged.

This thesis has not been previously submitted, in total or in part, to the University of

Western Sydney or any other University for a degree, diploma, or any other qualification.

Mingyuan Guo

May 2006

xvii

Chapter 1

Introduction

1.1 Background

There exist many studies, both theoretical and empirical, on the effects of market structure and individual behaviour on the process of price formation in stock markets. Market microstructure analyses the behaviour of trading and structure of a stock exchange. It explores specific trading mechanisms to identify and understand the patterns, if existing, inherent in the trading practices of stock markets and their intermediaries. The study of the systems, processes and principles that dictate the handling of orders, the introduction of trades and the determination of prices is an important aspect of finance and one that answers the queries of researchers regarding the nature of price formulation and the performance of specific market structures1.

There are two prevalent kinds of trading systems: the continuous trading market (continuous matching) and the call market (call auction). In the continuous trading system, trades are executed immediately when orders match in price and stock is available for liquidity. The best bid and ask quotes are generally divulged to

1 O'Hara (1995) suggests that the study of market microstructure is the study of the process and outcomes of exchanging assets under explicit trading rules and analysis of how specific trading mechanisms affect the price formation process. The key components of this process include the generation and dissemination of information, the arrival of orders, the rules and the institutional features of a stock market that determine how orders are transformed into trades.

1 investors enabling them to see portfolio adjustment instantaneously. The alternate call market has orders batched together for simultaneous clearance at set times

(usually once or twice) in each trading day. In an auction, submitted buy and sell orders form demand and supply schedules and the intersection of these curves determines the market clearing price at which all trades are executed.

A second criterion that classifies the varying topology of structures refers to the type of market agent - whether an agency market or a dealer market. In the agency market, the public submits orders that are then processed by a broker’s broker who matches them with other public orders. Public investors are really not a part of the exchange in the dealer market. Their orders are purchased or sold by the dealers, who serve as intermediaries, for their own portfolios. They can also be differentiated in the price discovery process into an order driven market which directly matches buy and sell quotes, and a price (quote) driven market where investors trade at bid or ask prices provided by a market maker (specialist) who continuously adjusts them according to supply and demand. The trading system under the quoted driven market is known as a specialist system. The system of trade used in the U.S. market, such as the NYSE, is an example of the specialist system. In order-driven markets, many exchanges open their trading sessions with call market auctions and then switch to continuous matching trading.

In the financial market, high-frequency data changes the nature of time series when taking intraday data, such as 15-minute, 5-minute or finer time intervals. The microstructure approach involves the use of short-run dynamics of high-frequency data to test statistical properties such as returns, return volatility and trading volume.

Therefore, high-frequency financial data provide a study of the market microstructure with an important facility in a variety of issues related to the trading

2 process and market structure. With the use of high-frequency data, researchers are able to answer more difficult questions concerning market behaviour, over short time intervals, which can’t be answered by analysis based on daily or monthly data.

The development of the capital market accompanies China’s economic development. As more and more foreign investors and institutions are involving themselves in Chinese high-speed economic growth and diversifying their investment portfolios, it is necessary to understand the Chinese stock market and how its market microstructure functions. Market microstructure analysis considers several factors affecting the price-forming rules, indicating the architectural structure of a market, the time of news releases, investors’ behaviour and the intraday trading patterns of institutional investors. Unlike other mature markets in the Western countries, the Chinese stock market is a limited order-driven market that uses electronic trading without market makers. This particular structural feature of an emerging stock market will obviously test the well-established principles of mature

Western markets. Thus, an exploration of the behaviour of the Chinese stock market from a microstructure approach is urgently needed. This study will provide a new contribution to understanding how an emerging order-driven market behaves and to what extent it differs from a price-driven market and other order-driven markets. It examines the behaviour of intraday returns and return volatility, inter-temporal behaviour of intraday liquidity (bid/ask spread and depth) and their determinants from a market microstructure approach. It also examines the functioning of the trading system, call trading and continuous matching, based on Amihud and

Mendelson (1987) and the intraday dynamic application in the Chinese stock market.

3 1.2 Motivation

Given that a financial market displays high speeds of adjustment, studies based on daily or weekly observations may fail to capture instantaneous information contained in intraday market movements. For example, stock returns and return volatility are two factors in the financial market based on daily or weekly observations which feature the behaviour of a stock market, but may ignore critical information concerning intraday price effects. Thus, an examination of the Chinese stock market based on intraday data is a very good starting point. The choice of the

Chinese stock market for this study is motivated by the following reasons:

Since its establishment in 1990, the Chinese stock market has experienced rapid growth. Although it has been subject to many influential investigations, its asset price behaviour and trading mechanism are not well understood. As studies on the effects of the market microstructure have been extended to stock markets of many countries, most previous research has only been conducted using daily or monthly data on the Chinese stock market. The Chinese stock market is developing rapidly so a comprehensive exploration from the microstructure approach on the behaviour of the stock market will not only benefit market investors but also policy makers.

The Chinese stock market is quite complex compared to most Western markets. Its complexity is due to its unique trading systems and its share-consisting structure. In China, firstly, the trading system is electronic call auction at the opening and continuous order-driven trading market during the trading day. Secondly, most listed companies are state-owned enterprises and a large proportion of shares of the listed companies are non-tradable. Shares are divided into two broad categories: non-

4 tradable and tradable shares. Generally, the non-tradable shares consist of existing assets assessed and computed before listing, and tradable shares are new issues to the public. A second way to differ between shares is that there are two groups of shares:

A-shares and B-shares. The Chinese residents can only trade the A-shares. Like many emerging stock markets, foreign investors face restrictions on owning domestic shares in China2. The B-shares are Chinese incorporated companies and can only be traded by overseas investors3. Both A-shares and B-shares are traded on China’s two stock exchanges: the Shanghai Stock Exchange (SHSE) and the Shenzhen Stock

Exchange (SZSE). Thus, this complexity is the environment for the present study, which focuses on the market microstructure.

Intraday behaviour of the trading and market structure can reflect the market efficiency. This study is searching for evidence of trading behaviour based on the

Chinese trading system and market structure. There are two continuous trading sessions in the two Chinese stock exchanges: the morning session (09:30 - 11:30) and the afternoon session (13:00 - 15:00). Before the morning session, there is a 10- minute open call auction session (09:15 - 09:25), the pre-opening period, to decide the opening price. There is also a one-and-half-hour lunch break (11:30 - 13:00) between the morning and afternoon trading sessions. In this aspect, the Chinese stock market is unique and different from Western markets and other emerging Asian markets. However, so far there is no existing research and empirical analysis of the

Chinese stock market. Thus, this study should contribute to the recent international research in using high-frequency stock market data.

2 Since April 2003, the Chinese stock market has opened to foreign investors gradually. New policy has been released that Qualified Foreign Institutional Investors (QFII) can trade domestic shares in China. 3 In February 2001, this restriction to B-shares for foreign investors was, however, relaxed when it became permissible for domestic investors.

5

1.3 The Purpose and Contribution

The purpose of this study is to perform an empirical analysis on the Chinese stock market. Specifically, this study examines issues of the behaviour of intraday returns and return volatility, the trading mechanism, liquidity characteristics and their determinants, and the dynamic application of the GARCH methodology to the intraday volatility using the high-frequency data from the years 2000 to 2002. In exploring these issues, each of them details relevant hypothesis that guide the study, statistical testing and analysis of, and application to, the Chinese market. Market microstructure studies are focused on three different aspects (three parts), presented in Chapters 3, 4 and 5. To help readers better understand the Chinese stock market, a special Chapter 2 is included in this study to introduce the development of stock market in China, and explains the stock market indices, listing requirements, special features, settlement rules, the trading system and market microstructure. Chapter 6 is the conclusion.

Chapter 3 examines the behaviour of intraday return and return volatility in the Chinese stock market using two market indices: the Shanghai Composite Index and the Shenzhen Component Index. These two indices are chosen because they are more frequently quoted corresponding to the Dow Jones Index and NASDAQ index in the US market, which mostly reflect the information in the markets.

The study of intraday return and return volatility is concerned with intraday stock price variation while the intraday stock price variation depends on the trading system and market structure of the stock exchange. Similar studies have been conducted in Western markets despite the differing market structures and

6 institutional frameworks between the different markets. Empirical studies on the behaviour of intraday return and return volatility reported a U-shaped pattern in the

US market, (for instance, the higher points of these variables occurs at the opening and closing of the trading day (Harris, 1986 and Jain and Joh, 1988)). Theoretical models developed by Kyle (1985), Admati and Pfleiderer (1988) and Foster and

Viswanathan (1990) have been used to explain the U-shaped patterns. They suggest that traders can be classified into four classes: private information traders, random liquidity traders, market makers and discretionary liquidity traders. Their findings suggest that public/private information and trading noise cause a systematic pattern in return volatilities.

This chapter first explores the statistical properties and systematic characteristics of intraday returns and volatility using 5-minute intraday data of the

Shanghai Composite Index and the Shenzhen Component Index over a three-year period. It then examines the volatilities of the trading period and the non-trading period. Finally, it examines the trading structure at the opening session, call trading and continuous matching, by comparing the interday 24-hour return and return volatility according to Amihud and Mendelson (1987). While there is a certain controversy about the advantage of the call market or continuous trading for the trading system, my study uses variance ratio tests to verify the trading mechanism and its effects on the volatility in the market. The methodologies discussed in this chapter follow the hypothesis tests suggested by Harris (1986), Cheung (1995),

Bildik (2001), and Tang and Lui (2002). Based on the non-normal distribution of intraday stock returns, besides parametric tests such as the mean equality test, the non-parametric Kruskal-Wallis test, Levene test and modified Levene test (Brown and Forsythe test) are applied.

7 The findings in the chapter indicate that intraday return series of the Chinese stock market are non-normal and have a fat-tailed distribution. The intraday returns broadly follow a U-shaped pattern (Wood, McInish and Ord, 1985) while the volatility of returns broadly follows an L-shaped pattern. This result is consistent with previous findings in the Hong Kong market (Lam and Tong, 1999 and Tang and

Lui, 2002). Further examinations by subperiods, using wider and finer than 5-minute data, indicate that there exists an intraday variation for each subperiod, and each 1- minute and 10-minute period. The conclusions are consistent with the intraday 5- minute return characteristics. The results also show that the return variance in the active trading period (open-to-close) of the stock market is larger than that in the nontrading hours (close-to-open). High volatility of intraday returns at the market opening is not due to the trading mechanism (call auction) but the overnight trading halt.

Chapter 4 explores evidence on inter-temporal behaviour and the determinants of liquidity proxies, bid/ask spread and depth in the Shanghai Stock

Exchange. Liquidity studies decide the formation of stock price, which is related to the effects of market trading procedures and price variation. Intraday liquidity patterns contain different information about price variation at different times in the trading day. The determinants of the two liquidity proxies are analysed for a better understanding of the stock price formation. The chapter provides unique empirical findings to date on the systematic patterns of interday and intraday bid/ask spreads and depths for companies traded on the SHSE. Based on the 5-minute intraday data over a three-year period (2000-2002), a set of sixty most active stocks is used to examine and analyse the inter-temporal behaviour and affecting factors of intraday

8 and interday liquidity and the relationship between the liquidity, trading price, risk, and information in the Chinese stock market.

This chapter follows the research of McInish and Wood (1992) and

Brockman and Chung (1998). The empirical analysis based on statistical results of the intraday 5-minute data with about five million observations in both bid/ask spread and depth is first performed. Statistical analysis and graphs of relative spread and depth of the time-of-day are plotted to examine intraday patterns and interday patterns. It then uses the linear regression model to further test hypotheses and verify the patterns of liquidity and their determinants of two liquidity proxies. In the model, price, the risk, and information flow are considered as the determinants for both bid/ask spread and depth proxies. The correlations between spread and depth for each

5-minute interval are investigated by the statistic analysis. Based on the theoretical points and implications, inter-temporal relationship and patterns between intraday bid/ask spread and depth are explained.

The results in this chapter reveal that, firstly, the spread in the limit-order market of Shanghai exhibits an L-shaped intraday pattern while the depth displays a reverse L-shape. The spread is the largest at the market call opening and the second largest at the open of continuous trading. The depth, on the other hand, shows the opposite pattern to the bid/ask spread. It is lowest at the call opening and the second lowest at the open of continuous trading. This is in striking contrast to the findings of the overall U-shape of the spread by McInish and Wood (1992) and Madhavan et al.

(1997) in the US market and Brockman and Chung (1998) and Ahn and Cheung

(1999) in the Hong Kong market. The findings are consistent with those documented on the multi-dealer systems (Chan et al., 1995). The widening of spreads (lowest depth) at the market opening is attributed to the factor of asymmetric information.

9 Secondly, a similar L-shaped intraweek pattern in the spread and a reverse V- shaped intraweek pattern in the depth were identified. The simultaneous occurrence of relatively wider spreads and lower depths on Mondays is consistent with the predictions of Foster and Viswanathan (1990). Higher Monday opening spreads are due to the information accumulation during non-trading hours. Thirdly, the striking

L-shaped pattern of the bid/ask spreads has a positive relationship with the volatility and a negative relationship with stock prices, while the striking reversed L-shaped pattern of the depths has a negative relationship with stock prices and the volatility but a positive relationship with trading volumes. The findings show that stock price, risk and information asymmetries are significant determinants of the bid/ask spreads and depths. Finally, there is a negative relationship between the spreads and the depths at the opening of the trading day. Overall, the results suggest that the determinants of information asymmetries, through time and across traders, play a key role in generating observed liquidity variations.

In Chapter 5, the GARCH model is applied to investigate the dynamic relationship between intraday volatility, trading volume and bid/ask spread based on high-frequency data by focusing on a set of most active stocks. Most financial time series show time-varying variance (volatility) and do not follow a normal distribution. However, recently, the GARCH model has been widely used in the financial time series because it takes into account the role of time-varying variance

(excess kurtosis such as fat tail behaviour and volatility clustering). In this chapter, a set of individual Shanghai stock returns is used for the analysis. Following the important paper by Lamoureux and Lastrapes (1990), the study will add evidence to the GARCH model for the high-frequency data application. Based on 5-minute intraday data over a three-year period, the chapter investigates the issues: (i) whether

10 model AR(1)-GARCH(1,1) can capture the GARCH effects of temporal dependence in the intraday return series and (ii) whether intraday trading volume and bid/ask spread can be used as variables of information flow to remove the GARCH effects in the Chinese stock market. Examination is made of the return–volume relationship and return-bid/ask spread relationship using individual stocks on the Shanghai Stock

Exchange to determine whether there are differences between developed markets and one emerging market.

The findings indicate that all individual stocks in the conditional variance model exhibit the strong persistence of the GARCH effect. Similar to Rahman, Lee and Ang (2002), the GARCH (1,1) model best describes the volatility of intraday returns. Current volatility can be explained by past volatility that tends to persist over time. The findings of this chapter further indicate that the AR(1)-GARCH(1,1) model successfully accounts for the nonlinear dependencies in the individual stocks.

However, the AR(1)-GARCH(1,1) model does not account for the linear dependencies in the high-frequency data.

Based on possible theoretical explanations of the mixture of distributions hypothesis (MDH), the inclusion of intraday trading volume as a mixing variable for information arrival in the conditional variance model helps in explaining the

GARCH effects on the stock return series. The coefficient for volume is positive and small for the GARCH(1,1). But the GARCH effects do not disappear entirely as a result of this inclusion; in other words, the persistence in volatility remains in intraday return series even after lagged-one volume is included in the model as an explanatory variable. Similar to the proxy of trading volume, the inclusion of intraday bid/ask spread as a mixing variable for information arrival in the conditional variance model helps in explaining the GARCH effects. Most of the coefficients for

11 bid/ask spreads are positive and small for the GARCH(1,1). The GARCH effects remain strongly significant after the introduction of the lagged-one bid/ask spreads.

The findings in this chapter suggest that the volume or bid/ask spread as an information variable has quite a limited effect on the volatility of intraday returns in the Shanghai stock market, which is similar to the results in the US market.

Information does not explain the volatility that may result from noise trading in the market. These findings are quite important because they could indicate that the

Shanghai stock market, after a decade’s development, has experienced some improvement in its quality of trading, becoming more similar to more mature markets, such as the NASDAQ market, at least in this aspect. It could also suggest that the volume or the spread as an information variable is not a very important factor for predicting the volatility of stock returns in intraday levels, irrespective of how mature or efficient a market is. If the latter is true, based on the MDH hypothesis, there should be other major factors that affect the volatility change in intraday returns and should be rewarded with further research.

12

Chapter 2

Chinese Stock Market and Market Microstructure

2.1 Introduction

This chapter addresses the development and current structure status of the Chinese stock market. The chapter introduces stock market development, and explains stock market indices, listing requirements, special features, settlement rules, the trading system and market microstructure.

The development of the Chinese stock market responded to the opening up of the economy with its major economic reforms. After the Third Plenary Session of the

Eleventh Chinese Communist Party Central Committee (CCPCC), convened in

December 1978, the Chinese government made the strategic decision of shifting the focus of work to socialist modernization and defined the guiding policy of revitalizing the domestic economy and opening up to the outside world. In order to achieve the desired economic modernization, the Chinese government has continued to search for a way of simulating privatization while retaining public ownership in a large number of state-owned enterprises. Since then, by incorporating a more market- oriented economic system, the government has gradually recognized that the share holding system could be incorporated into a socialist economy without negatively affecting socialism. Thus, the development of the capital market is an efficient way

13 to satisfy the need for large amounts of funds, which not only speeds up national economic growth, encourages the independence of the economic units but also solves the problem of rights, responsibility and interest of the state-owned enterprise. So far, the evolution of the Chinese stock market has mirrored that of China’s fast economic development in the past two decades. China has established a nation-wide equity market with two stock exchanges located in Shanghai and Shenzhen. The Shanghai

Stock Exchange and the were founded in December 1990 and April 1991 respectively1.

Both stock exchanges have similar organizations, structures and operating mechanisms. They are open for trading from Monday to Friday and have two trading sessions in each day. The China Securities Regulatory Commission (CSRC) governs the two stock exchanges and cross listing is not allowed. Each company by itself can only select one of the two exchanges and be listed on it. In general, large state-owned enterprises choose to list on the Shanghai Stock Exchange and smaller export- oriented companies choose to list on the Shenzhen Stock Exchange but there is no set rule for company listing on either exchange. In fact, some large state-owned enterprises are also listed on the SZSE.

The securities listed at the SHSE and the SZSE fall into three categories: stocks, bonds and funds. Both the SHSE and SZSE have been committed to creating an open, fair and efficient market since their establishment (the Fact books of the

Shanghai Stock Exchange and the Shenzhen Stock Exchange, 2003). Over the past decade, the SHSE and the SZSE have developed from regional markets into national markets with listed companies and members throughout the country. Both exchanges enjoy an important share of the gross national market in terms of listed companies,

1 For details, see the Fact Books (2003) from the Shanghai Stock Exchange and the Shenzhen Stock Exchange. Website: http://www.sse.com.cn/sseportal/en_us/ps/home.shtml and http://www.szse.cn/main/en/catalog_1697.aspx.

14 market capitalization and trading turnover. According to the Fact Book 2003 from the SHSE and the SZSE, the Chinese stock market has become one of the fastest growing markets in the world in terms of the number of listed companies, number of shares listed, total market value, tradable market value, and securities turnover in value.

By the end of 2003, the Shanghai and Shenzhen Stock Exchanges had a total of 1439 listed securities (824 and 615 respectively) and 1285 listed companies (780 and 505 respectively). Their combined total market capitalization was RMB2 4245.8 billion (about US$513.75 billion3). The total number of investors in the two markets exceeds 70 million people4, of which more than 19,000 were institutional investors.

Shares in China are divided into two broad categories: non-tradable and tradable shares of a listed company. Generally, the non-tradable shares consist of existing assets assessed and computed before listing, and tradable shares are new issues to the public. The funds from the new issue are used to develop the company.

Most of these listed firms are state-owned enterprises. The non-tradable shares have a capitalization of about RMB 2927.92 billion ($US354.29 billion) when computed using the market price. Only the tradable shares are allowed to trade, which means the float shares with the capitalization of about RMB 1317.85 billion ($US159.46 billion). The government and the legal-entity own most of the non-tradable shares and the legal-entity is any form of corporate identities, including privately owned companies, state-owned enterprises or a mixture of both such as banks and listed companies. In 2003, the total non-tradable equity of the listed companies was about

2 The RMB is the abbreviation for RenMinBi, which is the basic unit of the Chinese currency (Chinese Dollar). 3 The exchange rate used is 8.2643, which was the rate on December 31, 2003 according to the Bank of China (http://www.bank-of-china.com/info/qpindex.shtml). 4 Though some individual investors and institutional investors may open more than one account for the purpose of applying for a new issue of shares and other reasons, the exact accounts are not known and not large by estimation.

15 414.42 billion shares, 64.47% of the total equity of the listed companies, in which the government owns most of the non-tradable shares (73.51%) and the legal-entity owns (16.89%), others own about 10%. For details, see website5 of the CSRC.

The Chinese stock market has some unique features compared with those of the developed countries such as the US market and the British market. Tradable shares are further classified into two types: the A-shares and B-shares. A company with both the A-shares and B-shares is listed in the same Stock Exchange. Both the

A-shares and B-shares are divided into two broad categories: non-tradable and tradable. The local investors and foreign investors are separated in the A-shares and

B-shares markets. One important reason behind this is that the local currency is not convertible. The A-shares are denominated in RMB. Foreign individuals or institutions are not allowed to buy and sell A-shares. The B-shares are issued to, and traded by, foreign investors only. The B-shares are denominated in US dollars on the

Shanghai Exchange and Hong Kong dollars on the Shenzhen Exchange. The total capitalization of B-shares is much smaller than that of A-shares. At the end of 2003, there were only 111 B-shares with a total of capitalization about RMB 166.3 billion

($US20.12 billion).

However, in February 2001, the restriction to B-shares for foreign investors was relaxed. It became permissible for domestic investors with foreign currency holdings to trade B-shares. In 2002, a new policy was released to the Chinese stock market that Qualified Foreign Institutional Investors (QFII) could trade Chinese A- shares. This policy change reflects the further openness and the ongoing globalization efforts of the country. In June 2004, the small and medium enterprise board (SMEB) was established in the Shenzhen Stock Exchange for high and new

5 http://www.csrc.gov.cn/en/homepage/index_en.jsp

16 technology companies for which the listing conditions are somewhat different from other companies.

2.2 Stock Market Indices

The Shanghai Stock Exchange and the Shenzhen Stock Exchange have established several market indices to provide a comprehensive measure of market trends. The SHSE and the SZSE index series are compiled and published by the

SHSE and the SZSE, respectively. The Shanghai Stock Exchange index series includes the SHSE Composite index, the SHSE 180 index, the SHSE 50 index, the

A-share index, the B-share index, the Bond index and the Fund index. Among them, the SHSE Composite index is the main market index and is composed of all common stocks listed on the SHSE. The SHSE 180 index is based on 180 blue chip stocks.

Similarly, the Shenzhen Stock Exchange index series include the SZSE Composite index, the SZSE Component index, the SZSE 100 index, the A share index, the B share index, the Bond index and the Fund index. The SZSE Composite index is, likewise, the main market index and is composed of all common stocks listed on the

SZSE. But currently, the SZSE Component index is the main index indicator used by investors of the SZSE. It consists of about 40 of the most representative shares6. The

SZSE 100 index is counterpart of the SHSE 180 index. Figure 2.1 shows the historical daily movement of the Shanghai Composite Index and the Shenzhen

Component Index from 1991 – 2003 (Data source: Sohu com Inc, Web Site: http://www.sohu.com).

6 As dated May 23, 2005

17 All the market indices reflect the movement resulting from overall market activities and various different aspects, and provide investors with benchmarks for their different investment portfolios. Both the SHSE and the SZSE index series are market capitalization-weighted composite price indices. Since the SHSE Composite index and the SZSE Composite index are market weighted composite price indices, a large and widely held company will have a greater impact on the Chinese stock market performance than a small company under an equal stock price change.

Generally, the SHSE and the SZSE index series are calculated simply by the following formula:

Current Market Capitalization × Base Point (2.1) Base Market Capitalization

The base point is either of 100 or 1000. There is no fixed rule about using a base point as 100 or 1000. In general, the indices made earlier from the stock exchanges use a base point of 100. The SHSE composite index took December 19, 1990 as its base day and the SHZE Composite index took July 3, 1991 as its base day to calculate the base market capitalization. The market capitalization of an individual stock is calculated by multiplying the common stock price by the number of outstanding shares. The total market capitalization is the sum of the market capitalization of all constituent stocks. The SHZE Component Index takes January

23, 1995 as the base day and the base point is 1000. The SHSE 180 index, the SZSE

Component index and the SZSE 100 index are calculated in the same way as the composite indices but the SHSE 180 index and the SZSE 100 index use a sample of

180 and 100 individual stocks in the SHSE and the SZSE, respectively. The included

18 stocks are selected as the most representative according to total shares, trading value, trading volume and other conditions among all A shares.

In order to gauge the performance of various industries within the overall index, the SHSE and the SZSE also publish industry group indices. All the industry group indices are calculated in a similar way to the composite index; these include the Property Index, Public Utility Index, Comprehensive Index, Industrial Index,

Commercial Index, T-Bond Index and Fund Index.

2.3 Listing Requirements

Currently, the CSRC is responsible for examining the eligibility of Initial

Public Offer (IPO) applicants. The listing requirements are the same for both exchanges. Companies need the approval of the CSRC to offer stocks to the public.

Before equities can be publicly traded, the companies need to apply to the relevant

Exchange for listing. Generally, considering the location, development, listed fees and other factors, more large, state-owned enterprises are likely to be listed on the

Shanghai Stock Exchange while more smaller export-oriented companies are likely to be listed on the Shenzhen Stock exchange. According to the regulations of the

Securities Law of the People’s Republic of China and Company Law of the People’s

Republic of China, enterprises shall meet the following main requirements in order to be listed:

• The shares must have been publicly issued following approval by the CSRC;

19 • The company’s total share capital must be RMB 50 million (about

US$6.0501 million) or above;

• Minimum three-year operating history and positive earnings in the last three

years;

• Minimum of 1,000 shareholders, each holds 1,000 shares or more of the

outstanding shares;

• At least 25% public float, or 15% when the share capital exceeds RMB 400

million (about US$48.4 million);

• No serious misconduct or fraudulent records in the three years prior to listing.

• Other conditions stipulated by the State Council.

• The companies involved in high and new technology for different application

conditions can be considered to list on the small and medium enterprises

board in Shenzhen Stock Exchange.

2.4 Special Features

The China Securities Regulatory Commission is responsible for giving business licenses to engage in the securities business. Companies that have obtained a business license from the CSRC can apply for membership on both Exchanges and are allowed to access the market with approval. The exchange members need trading seats to gain material access. As the Chinese stock market uses electronic equipment, an unlimited number of trading seats are available for offer.

20 Both the Shanghai Stock Exchange and the Shenzhen Stock Exchange are non-profit membership institutions directly governed by the CSRC. They have the same features as follows:

• Cross listing is not allowed

• Same trading time

• Automated trading

• Order driven, price and time priority

• Dematerialized, book entry delivery

• 10% price limit for ordinary stocks on every trading day

• T+1 settlement for A-share traded securities and T+3 settlement for B-shares;

• Floating broker’s commission.

But there are some different features in the two stock exchanges:

• More state-owned enterprises are listed on the Shanghai Stock Exchange,

many of them monopoly suppliers to the domestic market.

• More smaller export-oriented companies are listed on the Shenzhen Stock

exchange, many of them are joint ventures (Kim and Shin, 2000)

• The Shenzhen Stock Exchange has established a Small and Medium

Enterprise Board (SMEB) for different application conditions or share listed

companies involved in high and new technology7.

• The B-shares are denominated in US dollars on the Shanghai Exchange and

Hong Kong dollars on the Shenzhen Exchange.

7 For further details refer to the website of the Shenzhen Stock Exchange (http://www.szse.cn)

21

2.5 Settlement

Although the SHSE and the SZSE have separate companies for settlements, the mechanism for clearing and settlement in the two stock exchanges is almost identical. The China Securities Central Clearing & Registration Corporation

(CSCCRC) is responsible for the central depository, registration and clearing of securities on the SHSE, and the Shenzhen Depository and Clearing Corporation

(SDCC) is responsible for the SZSE. A unique feature of both clearing systems is that it keeps the records of both brokers and investors. In terms of settlement, each clearing member shall designate one of its trading seats as the legal person to handle the settlement of securities between the clearing house and all its trading branches on a netting basis. It then settles internally and each of its broking branches settles with their respective clients. The settlement period is T+1 for A-Share traded securities and T+3 for B-shares. Finally, it must be noted that all transactions in the two exchanges are in cash and short selling is not allowed.

2.6 Trading Systems and Market Microstructure

The trading system and trading characteristics of a stock market can be an important factor in explaining market microstructure. This section gives a brief description of the market microstructure of China’s stock market.

22

2.6.1 Trading Systems

Both the SHSE and the SZSE have modern trading systems supporting paperless trading at a high speed and they have also established a nation-wide satellite telecommunication network with the most sophisticated equipment, the most complete range of functions, the largest number of users and the widest coverage.

For example, the SHSE owns a huge 3600-square-meter trading floor, which is the largest in the Asia-Pacific area. The SHSE host computer is capable of executing 29 million orders and settling 60 million transactions at a speed of 8000 transactions per second. The SHSE comprises the largest domestic satellite and optical communication network, which can disseminate real-time transaction information to the whole country and abroad.

Orders are matched automatically by a computer system according to the principle of time-price priority on the Shanghai Exchange, and price-time-order priority on the Shenzhen Exchange. Orders can be sent to the two exchanges' main frameworks through terminals either on the floor or from member firms. The two stock exchanges are now performing real-time monitoring of market activities under a sophisticated surveillance system that meets the demand of market operation.

2.6.2 Two Trading Sessions

The SHSE and the SZSE have a call market (call auction) at the opening and continuous order-driven trading market in the trading day. There are no designated

23 dealers (specialists) to intervene in trading. Trading is continuous and conducted through transparent computer-assisted systems in each exchange’s trading hall, and through terminals at the members’ offices. All of the order flows on the two stock exchanges must be displayed on computer terminals viewable by investors on and off both exchanges.

The trading mechanism of the Shenzhen Stock Exchange is very similar to that of the Shanghai Stock Exchange. Trading days for the two exchanges are from every Monday to Friday. There are two trading sessions at the two stock exchanges: the morning session and the afternoon session. Before the morning session, there is a

10-minute open call auction session (the pre-opening period) to decide the opening price. There is also a one-and-half-hour lunch break between the morning and afternoon trading sessions. Because both exchanges are in the same time zone and all areas in China use Beijing time as standard time, the trading time of the trading day is exact in the same for the two exchanges. The opening call auction starts at 09:15 and ends at 09:30 for the determination of the centralized competitive opening price.

During the 10 minutes between 09:15 and 09:25, investors can place limit orders and participate in the opening auction. The criterion used for the determination of the opening price is the maximization of trading volume. At 09:25, the market is cleared at a single price that maximizes the transaction volume. Orders that are not executed in the opening auction are automatically transferred to continuous trading. The determined opening price at 09:25 is continued to 09:30. The continuous trading in the morning session starts at 09:30 and ends at 11:30, and the afternoon trading session is from 13:00 to 15:00.

During the two continuous trading sessions, the electronic system is based on the matching of orders for consecutive bidding according to price and time priorities.

24 Closing prices of the stocks for a trading day on the Shanghai and the Shenzhen stock exchanges are generated by the weighted average trading price during the final

1-minute of each trading day8. The market is closed on Saturdays and Sundays and other public holidays announced by the two stock exchanges such as Chinese new- year’s spring festival and national holidays.

2.6.3 Price Limit and Trading Rules

During trading sessions on the SHSE and the SZSE, a stock is allowed to trade at a price plus or minus 10% from the previous day’s closing price in order to avoid sharp price increases caused by ‘buy manias’ and sharp declines caused by

‘sell panics’. The current continuous trading on China’s stock market is a pure limit order book with the price and the time as the first and second priorities. Investors can only place simple and good-to-day limit orders. Besides the buy and sell limit orders, no other sophisticated order types, such as trading-at-open, trading-at-close, stop orders, buy-at-minus and sell-at-plus, are supported by the trading system. The minimum tick sizes for all stocks are RMB 0.01.

Investors place their orders with the brokers and only limit orders are accepted by the trading system. A buy limit order must give the bid price and number of shares to be purchased. A sell limit order must give the ask price and number of shares to be sold. The trading system prioritizes orders first by price and then by time. In a continuous order-driven market, investors’ orders can be executed immediately upon arrival. Bid prices are arranged in priority from highest to lowest,

8 Before December 1, 2001, the Shanghai Stock Exchange took the final trading price as the closing price.

25 while ask prices are arranged in priority from lowest to highest. As from December

2002, the bid/ask prices and the number of shares of the five latest orders are revealed continuously to traders on the screens.

During the continuous trading session, investors can change or reverse their orders if they feel that their orders cannot be executed at the given price. All trades may be continuously confirmed by recording the transaction, stating the exact order, the price at which the transaction was effected, the precise time of the trade and the transaction volume, thus ensuring full transparency of the market.

If a buyer requires an immediate fill, they will submit a limit bid that is high enough to touch the lowest posted ask. This buy order will then be executed at the best ask. This is called a buyer-initiated trade. If a seller requires an immediate fill, they will submit a limit ask that is low enough to touch the highest posted bid. This sell order will then be executed at the best bid. This is called a seller-initiated trade.

For each buyer-initiated or seller-initiated trade, the party which initiates the trade bears the execution cost, while the counter party gains from the bid/ask spread as compensation for their expected loss to the traders and as compensation for providing liquidity.

This chapter provided a brief introduction of the securities market development in China, particularly with respect to the history, regulation and market microstructure of the trading mechanism. Chapter 3 will review the related literature on the effects of intraday return and return volatility, and explore the behaviours of intraday returns and return volatility using 5-minute high-frequency data on the

Chinese stock market. Several different tests will be used to investigate if there are systematic patterns according to the efficient market hypothesis (EMH). The finer (1- minute) and wider (10-minute) data are used to further verify empirical results. In

26 addition, the volatilities in the trading period, non-trading period and interday 24- hour period will be examined with respect to the trading mechanism of the Chinese stock market.

27

Figure 2.1 The Daily movement of the Chinese Stock market indices: Shanghai Composite Index and Shenzhen Component Index from 1991 to 2003

Shanghai Composite Index From 1991 to 2003

2400 2200 2000 1800 1600 1400 1200 1000 800 600 400 200 0 91 92 93 94 95 96 97 98 99 00 01 02 03 04 Chinese Stock Index

Shenzhen Component Index From 1991 to 2003

6500 6000 5500 5000 4500 4000 3500 3000 2500 2000 1500 1000 500

91 92 93 94 95 96 97 98 99 00 01 02 03 04 Chinese Stock Index

Data source: Sohu.com Inc (Web Site: http://www.sohu.com)

28

Chapter 3

Behaviour of Intraday Return and Return Volatility

3.1 Introduction

Chapter 2 introduced the development of the Chinese stock market and its trading mechanism. This chapter examines the behaviour of intraday returns and return volatility of the market indices on the Shanghai Stock Exchange and the

Shenzhen Stock Exchange. The objective of this chapter is to provide an initial understanding of the behaviours of intraday return and return volatility in the

Chinese stock market, which is a limited order-driven market using electronic trading without market makers. The chapter will investigate the following questions: firstly, are there systematic patterns in the Chinese stock market; secondly, is there any difference in the volatilities between the trading and non-trading periods; finally, is the call auction, or continuous matching of the trading mechanism, the source of high volatility in the morning opening?

In the financial market, microstructure models have been invoked to explain some well-established intraday seasonality, which suggests intraday stock returns exhibit systematic patterns (Kyle, 1985; Admati and Pfleiderer, 1988 and Foster and

Viswanathan, 1990). Possible explanations for these patterns are related to the flow of information and market microstructure. Fama (1976) suggests an information

29 hypothesis that a stock market is efficient if prices rationally, fully and instantaneously reflect all relevant available information and no profit opportunities are left unexploited. Thus, according to the EMH, equities markets should quickly react and fully reflect all available information, but since this is unpredictable by definition, asset price changes or asset returns cannot be predicted.

High-frequency financial data provide important conditions in empirical research on a variety of issues related to the trading process and market microstructure. Goodhart and O’Hara (1997) address the summary advent of high- frequency data for the model estimation and market microstructure research. There are no empirical studies on the microstructure of the Chinese stock market using intraday data. The availability of high-frequency data allows a better understanding of the intraday behavior of the market.

Empirical analyses of this chapter are based on 5-minute high-frequency data over a three-year period from 2000 to 2002. The intraday patterns of the two main stock indices are characterized by the intraday returns and return volatilities both in trading day and weekdays (such as Monday and Tuesday). The Monday effects and

Friday effects are also examined. Several parametric and non-parametric tests will be used to investigate if there are systematic patterns according to the efficient market hypothesis. The three subperiods of year 2000, year 2001 and year 2002, and finer

(1-minute) and wider (10-minute) data are used to further verify empirical results.

Furthermore, volatilities between the trading and non-trading periods are examined based on the statistical analysis. Finally, this chapter compares the trading mechanism: call market and continuous matching for the effect on volatility. The interday 24-hour return and return volatility are explored and variance ratio tests are

30 used to verify the trading mechanism for high volatility in the market opening according to Amihud and Mendelson (1987).

Empirical findings indicate that the Chinese stock market returns data is non- normal and has a fat-tailed distribution. The intraday returns broadly follow a U- shaped pattern (Wood, McInish and Ord, 1985) while the volatility of returns broadly follows an L-shaped pattern. This result is consistent with previous findings in the

Hong Kong market (Lam and Tong, 1999 and Tang and Lui, 2002). Further examination by subperiods, and wider and finer than 5-minute data indicate that there exists an intraday variation for each subperiod, and each 1-minute and 10-minute period. The conclusions are consistent with the intraday 5-minute return characteristics.

This chapter also discusses possible reasons for the noticeable L-shaped intraday volatility. The main reasons are due to flow of information and market microstructure, which are related to the generation and dissemination of information, the arrival of orders and the rules and institutional features of a stock market that determine how orders are transformed into trades.

The results further provide evidence that the return variance in the active trading period (open-to-close) of the stock market is larger than that in the nontrading hours (close-to-open). The volatilities of interday 24-hour returns are calculated and the variance ratio test is used to show that high volatility of intraday returns at the market opening is not due to the trading mechanism (call auction) but to the overnight trading halt.

The chapter is organized as follows. Section 3.2 presents the literature review on intraday return and volatility. Section 3.3 discusses the data and trading system.

Section 3.4 discusses methodology. In Section 3.5 and next Section 3.6, the empirical

31 analysis is performed on the intraday patterns based on the period of the trading day and weekdays. Section 3.5 examines the characteristics of intraday returns. The

Monday effects and Friday effects are also examined in this Section. Section 3.6 examines the characteristics of volatilities of intraday returns. Section 3.7 discusses the L-shaped volatility pattern of the Chinese stock market. Section 3.8 discusses the volatility of the trading and non-trading periods. Section 3.9 conducts an exploration of the trading mechanism between the call auction and continuous trading. In Section

3.10, the conclusion is made.

3.2 Literature Review

Intraday data has been used to study return and return volatility for stock price variation. The intraday stock price variation depends on the market microstructure characteristics of the stock exchanges. Microstructure analysis of intraday patterns allows the researcher to contrast empirical findings on intraday patterns of returns and volatility and examine the market efficiency of different exchanges. In an efficient market, the basic value of a security fluctuates randomly.

The fundamental idea of the EMH is that security prices reflect optimal use of all relevant and available information. Thus, in an efficient market, return and return volatility should be consistent during the trading time and no profit opportunities are left unexploited. Wood, McInish and Ord (1985) were the first to explore and identify the distinct U-shaped return and return volatility pattern over the trading day.

They found that return and return volatility are higher at the open and close of trading and lower in the middle of the day using 15-mintue return data. Their

32 findings are contrary to the assumption of the EMH that returns and return volatilities should be equal across every 15-minute interval of the day.

Harris (1986) examined weekly and intradaily return patterns in common stock prices using transaction data on the NYSE and reported significant positive returns during the first 45 minutes (except Mondays) and during the last 15 minutes of the trading day. Jain and Joh (1988) found that common stock intraday returns differ across trading hours of the day. The largest stock returns occur during the first

(except Mondays) and the last trading hours and the lowest return is earned in the fifth hour of the day. They report a significant U-shaped intraday return and return volatility patterns in stock trading on the NYSE.

Return volatility can be considered as a direct measure of risk and an indirect measure of the level of information. Recently, Ozenbas, Schwartz and Wood (2002) have examined intraday share price volatility over the year 2000 for five markets: the

New York Stock Exchange, Nasdaq, the London Stock Exchange, Euronext Paris and Deutsche Borse. They observed a U-shaped intraday volatility pattern, a particularly sharp spike for the opening half hour, and a general level of intraday volatility that is accentuated vis-à-vis volatility over longer differencing intervals in each of these markets. They suggest that the volatility accentuation is attributable to spreads, market impact, price discovery and momentum trading.

Other studies also support the existence of a U-shaped pattern in the intraday patterns of the US stock market (French and Roll, 1986 and Lockwood and Linn,

1990). The theoretical explanations for the behaviour of these variables of the U- shaped patterns are not easy. Kyle (1985) suggests that traders can be classified into three classes: private information traders, random liquidity traders, and market makers. Admati and Pfleiderer (1988) and Foster and Viswanathan (1990) add a

33 fourth type of trader to Kyle’s model: discretionary liquidity traders. They take into account the impact, the costs, the size and time of the trade. They suggest that public and private information, and trading noise, are the causes of a systematic pattern in return volatilities, which lead traders to minimize trading during the periods of the open and close of the market.

When public information is released during the non-trading period

(overnight), the liquidity traders (random liquidity traders and discretionary liquidity traders) would be more active at the open of the market. When the market is near to close, the informed traders, who get access to private information during the trading hours, would be more active towards the close of trading before the information becomes public during the overnight non-trading periods. This public/private information, and noise trading hypotheses, are well supported by the empirical evidence of a U-shaped pattern in return volatilities (French and Roll, 1986 and Stoll and Whaley, 1990).

In addition to the evidence from the US market, there are similar studies conducted by Chang et al. (1993) on the Japanese market and Cheung et al. (1994) on the Hong Kong stock market. Chang et al. (1993) analyzed intraday returns and return volatilities for the index of the Tokyo Stock Exchange. The Japanese market has the morning session and the afternoon session with a lunch break. Instead of the single U-shaped pattern for returns and volatility observed on the US exchanges, they found a double U-shaped return pattern corresponding to the two trading sessions.

This result has been attributed to the fact that the market has two daily trading sessions, separated by a lunch break. Cheung et al. (1994) also reported a double U- shaped volatility pattern for 15-minute intervals using the Hang Seng Index during the period April 1986 to December 1990. The double U-shaped pattern is reflective

34 of the two daily trading sessions of the Stock Exchange of Hong Kong, consistent with that reported for the Tokyo Stock Exchange.

Bildik (2001) examines the intra-daily seasonalities of the stock returns in the

Turkish Stock Market in the period from 1996 to 1999, by using 15-minute and also

5-minute and 1-minute interval data. He finds that stock returns follow a U-shaped or more precisely a W-shaped pattern over the trading day at the Istanbul Stock

Exchange. Tang and Lui (2002) examine the intraday and interday volatility patterns and test the wait-to-trade hypothesis using 24-hour interday returns and 15-minute intraday returns on the Hang Seng Index and Hang Seng Index Futures. They find that the intraday variance forms an L-shaped pattern during the morning session on all weekdays in both indices. Copeland and Jones (2002) provide further evidence that for the intraday effects in the Korean market, there exists an U-shaped pattern over the day. They find that both volume and volatility are found to be consistently higher at the start of the trading day.

Furthermore, empirical studies find that the return variance in the active trading hours (open-to-close) of the security market is larger than that in the nontrading hours (close-to-open) (French and Roll, 1986; Amihud and Mendelson,

1987; Barcaly, Litzenberger and Warner, 1990 and Stoll and Whaley, 1990). Three possible explanations have been proposed for this phenomenon in the literature.

Firstly, more public information reaches the marketplace during normal trading periods than the non-trading periods. Secondly, trading activity from the informed investors reveals their private information inducing greater return variance (French and Roll, 1986). Thirdly, the process of trading itself introduces more noise into stock prices and returns when investors overreact to others’ trades, which leads to larger volatility over the trading periods (Shiller, 1981 and Black, 1986).

35 Amihud and Mendelson (1987) and Stoll and Whaley (1990) compare the effects of a call auction market and continuous trading market on return volatility.

The opening procedure of the New York Stock Exchange (NYSE) is essentially a call auction mechanism while all other trades are by the continuous trading mechanism. In the call auction mechanism, the buy and sell orders accumulated overnight are executed at a single price at the market opening. They reported that the inter 24-hour daily stock returns computed using open-to-open prices have greater variance and show more evidence of reversals than comparable returns computed from close-to-close prices. Amihud and Mendelson (1987) attribute the result to differences in trading mechanisms between the opening and closing transactions and

Stoll and Whaley (1990) attribute the result to the monopoly power of the specialist system.

Amihud and Mendelson (1991) examine the Japanese market where there are two trading sessions: the morning session and the afternoon session. Each session is based on the call auction for the opening and then the continuous trading session to the close. They find that the morning open-to-open volatility is high, but not the afternoon open-to-open volatility, and conclude that it is caused by the preceding long period of non-trading rather than the trading mechanism.

Gerety and Mulherin (1994) estimate transitory volatility throughout the trading day based on forty years of hourly Dow Jones sixty-five Composite price index data. They hypothesize that if the opening auction is responsible for higher open-to-open volatility, a sudden drop in interdaily variance after the open should be observed. However, their findings are in contrast with their hypothesis and the interday 24-hour volatilities decline steady, which reflects information processing.

They attribute the result to halt of trade.

36 Choe and Shin (1993) study intraday and interday volatilities on the Korea

Stock Exchange. They show that open-to-open volatilities are larger than close-to- close volatilities. Because the KSE is not a specialist system, they rule out monopoly power and the lack of specialist stabilizing activity as two potential explanations for their findings.

3.3 Dataset and Trading Systems

This chapter uses intraday data of the Shanghai Composite Index and the

Shenzhen Component Index. Like the Dow Jones Index of the NYSE and the

NASDAQ Index of the NASDAQ in the US market, the Shanghai Composite Index in the SHSE and Shenzhen Component Index in the SZSE are the most quoted and representative indices used by the SHSE, the SZSE, the mass media and investors

(hereafter, I will refer to them as the Shanghai index and the Shenzhen index, respectively). As discussed in Section 2.6 of chapter 2, the Shanghai and the

Shenzhen stock exchanges are electronic call auctions at the opening and continuous order-driven trading markets in the trading day. Orders are matched automatically by a computer system according to the principle of time-price priority in the two stock exchanges. The current continuous trading on the Chinese stock market is a pure limit order book with the price and the time as the first and second priorities. There are no designated dealers (specialists) to intervene in trading.

The morning session is from 09:30 to 11:30 and the afternoon session from

13:00 to 15:00. Before the morning session, a call auction starts at 09:15 and ends at

09:25 for the determination of the centralized competitive opening price. At 09:25,

37 the market is cleared at a single price that maximizes the transaction volume. The determined opening price at 09:25 is continued to 09:30. Closing prices of the stocks are generated by the weighted average trading price during the final 1-minute of each trading day.

With the objective of analysing the intraday patterns, the 5-minute data of the

Shanghai index and the Shenzhen index were collected. The data used for the two indices covers the three-year trading period from January 1, 2000 to December 31,

2002, a total of 716 trading days with more than 30,000 observations for each index.

The data were provided by the Securities Industry Research Centre of Asia-Pacific,

Australia (SIRCA) 1.

The dataset of the Shanghai and the Shenzhen indices records information on the time-stamped transactions, including the code, the order of time interval, the trading date, the day, trading time, the opening value, the highest value, the lowest value and the closing value in each time interval in each trading day.

3.4 Methodology

3.4.1 Definition of the Returns

In the financial market, stock return instead of price is most of the used variable. Campbell, Lo and MacKinlay (1997, page 9) identify two main reasons for using returns. First, the stock return is a complete and scale-free summary of the

1 I am grateful to Mr. Patrick Huang from the SIRCA for providing us with all index and individual stock data series.

38 investment opportunity. Second, statistical properties of returns are more attractive and easy to process than the price series. In this chapter intraday interval returns are computed by taking the first difference of the natural logarithm of the successive interval values of the indices; that is,

Rt = ln(Pt /Pt-1) (3.1)

where, for each interval during the trading day, Pt is the close value of the interval at time t and Pt-1 is the closing value of the proceeding interval at time t-1, which is also the opening value of the next interval. Rt is the return on stock indices at the interval.

As there are two sessions in the trading day, the 5-minute returns consist of forty-eight observations over the two sessions from 09:30 to 11:30 and from 13:00 to

15:00 in a trading day with 34368 observations over the three years. The first and last

5-minute intervals for the morning session are from 09:30 to 09:35 and from 11:25 to

11:30, respectively. The first and last 5-minute intervals for the afternoon session are from 13:00 to 13:05 and from 14:55 to 15:00, respectively. Thus, the forty-eight 5- minute intervals throughout the trading day break up into 24 intervals for the morning session and 24 intervals for the afternoon session.

In determining the return for a 5-minute interval, if there is no trading at the end of the 5 minutes, the closest trading price to the end of that 5 minutes is the closing value of that 5-minute interval. The return for the first 5-minute interval of the trading day is calculated by comparing the closing value at t1 (09:35) with the closing value at time t0 (09:30) when the trading starts. The close-to-open (overnight or non-trading) return is computed by comparing the closing value at time t0 (09:30) with the last value (closing value) at time t48 (15:00) of the previous day-end. Open-

39 to-close returns are computed by comparing the last value (closing value) at time t48

2 (15:00) with the opening value (closing value at time t0) at 09:30 of the trading day .

Other intraday returns for different time intervals such as 1-minute and 10-minute are calculated using the same method. A similar method can be used to compute a 1- minute return series. When considering larger time intervals, for example, Equation

3.1 can be changed to Rt = ln(Pt /Pt-2) for the 10-minute interval based on a 5-minute computation.

3.4.2 Hypotheses Development

According to the EMH, an asset market should react only to new information.

Information release is related to the asset price. Prices move rapidly to reflect all new information as it flows into the marketplace. Further, buy and sell prices are promptly reflected the price of an asset with minimal impact and consequently for stock returns (LeRoy, 1980 and Bernstein, 1987). If information arrives equally during the four hours of the trading day (09:30-11:30 and 11:00-15:00) and is immediately incorporated in the price, then the intraday returns, return medians and return volatility (standard variations or variances) across the time intervals during the period should be the same and not significantly different from zero for all forty-eight

5-minute intervals. Thus, to investigate the impact of seasonalities in the Chinese stock market, the hypotheses can be made that there are no intraday effects in the

Chinese stock market if the market is efficient.

2 Another way is to standardise the return series by subtracting each observation from the mean of the sum of all corresponding intervals and then dividing the difference by the standard deviation of its corresponding 5-minute interval. The results may be obtained upon request.

40 As for the Shanghai index and the Shenzhen index, the tests of the hypotheses can be separated for the full data set by the period of the three years, and by the each day of the week, for example, Monday, Tuesday and Friday. If a systematic pattern is discovered on the stock exchange, any observed differences in the time intervals of the return series reflects differences in value discovery and in the transitory price disturbances among them.

First, according to the EMH, intraday returns should be constant across 5- minute time intervals during the period, as would be the return medians and return volatilities. Hence, the following hypothesis should hold. If the tests reject this hypothesis, there is evidence of systematic patterns of intraday returns or return medians or return volatilities. These are called the time-of-day effects.

H0A: The means, medians and variances of 5-minute returns are equal across

all intervals.

H1A: At least one 5-minute interval has a significantly different mean return,

median and variance across all intervals.

Second, in any day of the week, such as Monday or Tuesday, intraday returns should be constant across 5-minute time intervals by the day of the week, as would be the return medians and return volatilities. Hence, these considerations lead to the following hypothesis. If the tests reject this hypothesis, there is evidence of systematic patterns of intraday return patterns or return medians or return volatilities in the weekday. These are called the time-of-day effects by the day of the week.

41 H0B: The means, medians and variances of 5-minute returns are equal across

all intervals by the day of the week.

H1B: At least one 5-minute interval has a significantly different mean return,

median and variance across all intervals by the day of the week.

Third, intraday returns should be constant across days of the week, as would be the return medians and return volatilities. Hence, these considerations lead to the third hypothesis. If the tests reject this hypothesis, there is evidence of systematic interday return pattern or return median pattern or return volatility pattern, which are called the day-of-week effects.

H0C: The means, medians and variances of 5-minute returns are equal across

days of the week.

H1C: At least one 5-minute return day has a significantly different mean

return, median and variance across days of the week.

Fourth, when considering each interval in the trading day, intraday returns should be constant across days of the week, as are the return medians and return volatilities. Hence, the following hypothesis is expected to hold. If the tests reject this hypothesis, there is evidence of systematic patterns of interday interval returns or return medians or return volatilities in the weekday. These are called the day-of-week effects by the interval.

H0D: The means, medians and variances of 5-minute returns are equal across

days of the week by the interval.

42 H1D: At least one 5-minute interval has a significantly different mean return,

median and variance across days of the week by the interval.

Fifth, the mean return on Monday or Friday should be equal to the means of other days of the week. Hence, the following hypothesis is expected to hold. If the tests reject this hypothesis, there is evidence of Monday or Friday (weekend) effects.

H0E: The means of 5-minute interval returns on Monday (Friday) are equal

to the means of other days of the week.

H1E: The means of 5-minute interval returns on Monday (Friday) are not

equal to the means of other days of the week.

Sixth, the mean return in each 5-minute interval on Monday or Friday should be equal to the means of other days of the week. Hence, the following hypothesis is expected to hold. If the tests reject this hypothesis, there is evidence of Monday or

Friday (weekend) effects by the interval.

H0F: The means of 5-minute interval returns on Monday (Friday) are equal

to the means of other days of the week by the interval.

H1F: The means of 5-minute interval returns on Monday (Friday) are not

equal to the mean returns of the other days of the week by the interval.

43 3.4.3 Testing for the Intraday Patterns

The methodologies used in this study follow the hypothesis tests suggested by

Harris (1986) in the US market, Cheung (1995) and Tang and Lui (2002) in the Hong

Kong market and Bildik (2001) in the Turkish market, which investigated the statistical properties of intraday returns of the stock index and individual shares.

Statistical tests are used to verify: (1) if the intraday stock returns are normally distributed, and (2) if the Chinese stock exchanges follow a U-shaped intraday pattern. Andersen and Bollerslev (1997), Mian and Adam (2001) and Bildik (2001) have demonstrated that distribution of the returns in high-frequency data is non- normal due to high-kurtosis and negative-skewness. In this chapter, the tests of normality used are the Jarque-Bera normality test and Anderson-Darling test. The

Jarque-Bera statistic is known to follow the chi-squares distribution with two degrees of freedom. The Anderson-Darling test compares the sample CDF to the normal CDF

2 with mean equal to X and variance equal to sX . Parametric tests, such as the mean equality test, and non-parametric tests such as the Kruskal-Wallis test, Levene test and modified Levene test (Brown and Forsythe test) are applied. Nonparametric tests are shown to be statistically superior to parametric tests in detecting abnormal price reactions for small time intervals (Mucklow, 1994). The variance ratio test is also used to test return volatility.

In order to test mean differences of the intraday returns across different intervals for the market indices, the following econometric model is considered:

rDt1122=+β ttββ D +++… n Dntεt (3.2)

44 rt is the intraday return at interval between time t-1 and time t. The Dit is a dummy variable for each interval of the trading day and εt is the disturbance term. The coefficients (βt) are the expected mean returns at each interval t. For the 5-minute interval data, n equals 48 in Equation (3.3), so there are forty-eight dummy variables:

D1t, …, D48t. β1 represents the expected return of the first interval (first 5-minutes) of the trading day. D1t is a dummy variable equal to 1 if the mean occurs between the first interval (first 5-minutes) of the trading day and 0 otherwise. If the expected mean return is the same for each interval, the estimates of β1, …, βn of Equation (3.2) will be close to zero and the F-statistic measuring the joint significance of the dummy variables should be insignificant.

However, in the financial market, the intraday returns can be serially dependent, especially the index return, as a result of thin or non-synchronous trading or even slow reaction of the market to news (Niarchos and Alexakis, 2003). In addition, mis-specifying the dynamics of the intraday returns can produce slow decay of the autocorrelation. As a result of this, the model was to adjusted to include additional lagged returns as the explanatory variables as follows:

k rD D… D r (3.3) t1122=+β ttβββ +++ nnti∑ t−i +εt i=1

Equation (3.3) can be used to test the day of the week effect by the time interval when using five dummy variables (Monday, Tuesday, Wednesday, Thursday and Friday) to represent the days of the week. In addition, Equation (3.2) could also be changed to test the Monday/Friday effect when using dummy variables as

Monday or Friday as follows:

45 rt = β0 + β1DMonday + εt ; (3.4)

rt = β0 + β1DFriday + εt ; (3.5)

Rank-based non-parametric test, such as the Kruskal-Wallis’ one-way analysis of variance (ANOVA), is used to test the hypothesis that the subgroups of a sample have the same general distribution, against the alternative that at least one subgroup has a different distribution (Sheskin, 1997). This is a generalization of the

Mann-Whitney test for more than two subgroups. The idea behind the Mann-

Whitney test is to rank the series from smallest value (rank 1) to largest, and to compare the sum of the ranks from subgroup 1 to the sum of the ranks from subgroup 2. If the groups have the same median, the values should be similar.

Kruskal-Wallis tests report the asymptotic normal approximation to the U-statistic

(with continuity and tie correction) and the p-values for a two-sided test.

The Levene test (Levene, 1960) and modified Levene test (Brown and

Forsythe, 1974) are used for the equality of variances. The Levene test is based on an analysis of variance of the absolute difference from the mean. The modified Levene test (Brown and Forsythe test) employs absolute deviations from the class means.

When empirical distributions are not normal, the Brown and Forsythe test can provide robust results based on deviations from the median. For further details refer to Brown and Forsythe (1974) and Conover et al. (1981).

The variance ratio (VR test) test is to test the null hypothesis of equal variance between the two series by the F-test. F-ratio which is greater than one indicates variance of first series is larger than that of the second one. For further details refer to Amihud and Mendelson (1987).

46 3.5 Empirical Results of Intraday Return Patterns

In this section and the next section, the empirical findings on the intraday patterns will be investigated based on the period of the trading day and weekdays

(such as Monday and Tuesday). Section 3.5 examines the characteristics of intraday returns. The Monday effects and Friday effects are also examined. Section 3.6 examines the characteristics of volatilities of intraday returns. Parametric and non- parametric tests will be used to investigate if there are the systematic patterns according to the EMH. The three subperiods of year 2000, year 2001 and year 2002, and finer (1-minute) and wider (10-minute) data are used to further verify empirical results. Section 3.7 will present theoretical explanations for the intraday patterns in

Chinese stock market.

3.5.1 Intraday Return during the Period

This section focuses on analysis of the 5-minute return series of the Shanghai

Composite Index and the Shenzhen Component Index over the three-year period.

Figure 3.1 shows the time plot of intraday 5-minute sample index closes and logarithmic returns for the Shanghai Index and the Shenzhen Index from January 1,

2000 to December 31, 2002. The left plots are for the Shanghai Index and the right plots are for the Shenzhen Index. As the plots show in the upper side for both indices, the basic trend for the index close went up in year 2000 and first half of year

2001, and went down in the second half of year 2001 and year 2002. Comparing the index closes in December 2002 with January 2000, both the Shanghai index and the

47 Shenzhen index have negative returns in 3 years. The bottom plots are for the index returns of the 5-minute interval. As the plot shows, the movements of the intraday 5- minute indices and returns in the two stock exchanges are consistent, and verified by the coefficients of the two intraday 5-minute indices and returns (0.9031 and 0.8157).

Because local investors could participate in the two markets at the same time, high correlations between the two markets reflect the common prediction for the market’s movements. In general, Figure 3.1 shows consistently high and low positions during the period, which means that the two markets have the same ability to assimilate information such as political, economic or other related situations in China.

Table 3.1 and Table 3.2 provide the descriptive statistics of the 5-minutes returns (in percentage) by trading time for the year 2000-2002 for the Shanghai index and the Shenzhen index, respectively. From the bottom lines (All) of Table 3.1 and

Table 3.2, negative mean 5-minute returns are observed for both indices during the 3- year period and they are close to zero, equal to -0.0014% and -0.0018%, respectively and the tests further show that the mean returns of both the Shanghai and Shenzhen indices are not significantly different from zero at the 5% level (P-values are 0.0783 and 0.0526, respectively). The skewness is generally not obvious but the values of excess kurtosis are larger during the three-year period. The Jarque-Bera and

Anderson-Darling tests of normality have p-values less than 0.01, thus each test would reject the null hypothesis of normality. These indicated that the returns data is non-normal and has a fat-tailed distribution on the Chinese stock market.

In the trading day, more negative mean 5-minute returns are observed than positive mean returns per 5-minute interval period, but positive returns are observed at the beginning and at the ending of the trading day. For example, during the mean return at the opening of the Shanghai Composite Index, which is represented by the

48 first 5-minute interval (09:30-09:35) of the continuous trading of the day in the morning session, is positive and with an average of 0.0306%. During the next eleven

5-minute intervals, however, the mean returns are all negative. This implies that high opening returns quickly converge to negative. The mean of closing returns representing the last 5-minute interval (14:55-15:00) of the continuous trading is the largest and positive 0.1017%. It should be noted that mean returns over the three 5- minute intervals before the closing return (last 5-minute interval) in the afternoon session are negative. The Shenzhen Component Index has the similar situation for the mean return 0.0285% at the opening in the morning session and 0.083% (largest) at the close of the afternoon session.

Furthermore, for the two stock indices, the mean returns at the closing

(11:25-11:30) of the morning session are positive 0.0125% and 0.0130%, respectively, but not as strong as the closing returns in the afternoon session. A positive but small mean return (0.0036%) is also observed at the beginning of the afternoon session (13:00-13:05) for during the Shanghai index but a large mean return for the Shenzhen index (0.0219%). To further analyze the intraday seasonality of during the two-index return series, the upper plots of Figure 3.2 display the movement of the 5-minute returns throughout the day for the Shanghai Composite

Index and the Shenzhen Component Index. The two graphs show that there are large, positive mean returns at the beginning of the trading day and a big upward increase in returns at the end of the trading day, which presents typical U-shaped patterns.

These intraday effects are consistent with previous literature in the US market

(Wood, McInish and Ord, 1985; Harris, 1986; Jain and Joh, 1988 and Lockwood and

Linn, 1990) and other countries for the Hong Kong market (Cheung, 1995) and for the Turkish market (Bildik, 2001).

49 It is interesting to see these price changes (high returns) in the morning opening. Possible reasons are the overnight halt and effects of information released, which reflect the new price-discovered process, and the high returns at the trading closing will be attributed to informed traders with private information for the next day’s opening. Different informed and uninformed traders may lead to new price discovery. The theoretical discussion that intraday returns are basically consistent with intraday volatilities will be put in Section 3.7.

Cumulative returns can be used to calculate and analyze differences in the open-to-close returns, which is the return by comparing the closing price at 15:00 with the opening price at 09:30. From this method, the real-time intraday lowest and highest returns can be found in the trading day. This means that during the trader may make a profit by daily trading. The lower plots of Figure 3.2 display the cumulative 5-minute returns throughout the day for the Shanghai Composite Index and the Shenzhen Component Index. For the Shanghai index, the cumulative returns begin to increase to 09:35 with a return of 0.031% and then fall sharply with little fluctuation to 11:25 to -0.153% and then rise to -0.141% in the last 5-minute interval of the morning session. So far, the intraday cumulative average return is negative for the first session. In the afternoon session, the cumulative returns continue to move up in the first 5-minute interval to -0.137% at time 13:05, and then again go down to the

28th 5-minute interval, to record the lowest level of the trading day at 13:20 (-

0.185%). After this time, it begins its rising trend, which starts to increase gradually to -0.124% at 14:20 and then decreases again to –0.168% at 14:55. Finally, in the last

5-minute interval at 15:00, there is a sharp increase in the mean 5-minute return

(0.1017%), at which time the cumulative mean return reaches -0.067%. The

Shenzhen index has a similar cumulative moving trend to the Shanghai index from

50 Figure 3.2 with the highest cumulative return at 09:35 (0.0285%), the lowest at 13:20

(-0.184%) and ending at 15:00 (-0.087%).

To sum up, the two indices have similar characteristics in return patterns, which show striking U-shaped patterns. The findings are consistent with those in the other markets with different trading systems. The patterns of the cumulative returns imply profit making by the daily traders and institutions for trading and applications of derivative instruments, but this is outside the area to explore for this study. Also, currently, the Chinese stock market does not allow trading derivative instruments.

3.5.2 Intraday Return during the Weekdays

Table 3.3 and Table 3.4 show the 5-minute mean returns by time interval and by the day of the week (Monday to Friday) for the Shanghai Composite Index and the Shenzhen Component Index over the 3-years period. Statistical parametric, and non-parametric tests for the equality of intraday returns for the trading period are also provided (the details will be in the next subsection 3.5.3). Figure 3.3 plots the mean

5-minute return patterns against the time interval by the day of the week and Figure

3.4 shows the cumulative mean 5-minute returns by the day of the week.

First, there exists an intraday variation in each weekday for both indices.

Second, there is a striking difference in the first 20 minutes between Monday and the other weekdays. Third, the cumulative mean return is positive on Tuesday for both indices and is negative on other days. Fourth, a spike in mean returns at the end of each weekday is observed and this phenomenon appears consistent with all trading days during the period. Thus, except on Monday, the U-shaped patterns are observed

51 on the weekdays. Negative opening returns for the two indices are observed on

Monday, which is consistent with the previous findings by Harris (1986) for the weekend effect or Monday effect in the anomalies literature for stock markets.

3.5.3 Test Results of the Returns

On the right side of Table 3.3 and Table 3.4, column F-test 1 shows the F-test for the equality of intraday returns across days of the week, and across days of the week by the 5-minute interval. Row F-test 2 contains the F-test values for the equality of intraday returns across the 5-minute intervals and across the 5-minute intervals by the day of the week. Column KW-test 1 shows the test values for the equality of intraday medians across days of the week and across days of the week by the 5-minute interval. Row KW-test 2 contains test values for the equality of intraday medians across 5-minute intervals and across the 5-minute intervals by the day of the week. Column FMon shows the test result of the equivalence of the Monday mean to the average of the other weekday means during the period and by the 5-minute interval, and column FFri is the test result of the equivalence of the Friday mean to the average of the other weekday means during the period and by the 5-minute interval.

For the equality of mean return tests (F-tests), firstly, the null hypotheses H0A and H0B (F-test 2) for equality of intraday returns across the 5-minute intervals during the period, and by the day of the week for each index, are rejected at the 1% significance level. For the equality of mean return median tests (KW-test), the null hypothesis H0A and H0B (KW-test 2) for equality of intraday return medians across the 5-minute intervals during the period, and by the day of the week for each index,

52 are rejected at the 1% significance level. The F-tests and KW-tests have consistent results. Thus, the time-of-day effect on the intraday returns of the Shanghai index and the Shenzhen index can be supported, which means that there are systematic intraday return patterns existing on the Chinese stock market. It should be noted that the first two intervals from 09:30 to 09:35 and from 09:35 to 09:40 of the morning session are significantly important at the 1% level for both indices.

Secondly, the null hypothesis H0C (F-test 1) for the equality of intraday returns across days of the week for each index cannot be rejected, and the null hypothesis H0C (KW-test 1) of equality of intraday return medians across days of the week for each index cannot be rejected. Again, F-tests and KW-tests have consistent results. Thus, the day-of-week effect on the intraday returns of the Shanghai index and the Shenzhen index cannot be supported.

Thirdly, the null hypothesis H0D (F-test 1) for the equality of intraday returns across days of the week by the 5-minute interval for each index is rejected in the 2nd interval at 09:40 at the 1% significance level, and is rejected for the Shanghai index

nd in the 22 interval at 10:50 at the 5% significance level. And the null hypothesis H0D

(KW-test 1) of equality of intraday return medians across days of the week by the 5- minute interval is rejected at the 1st, 2nd, 3rd, 5th and 46th intervals of the Shanghai index and at the 1st, 2nd, 4th, 44th and 46th intervals of the Shenzhen index at the 5% significance level. There are mixed results between the F-tests and KW-tests. At least, there is a day-of-week effect in interval 2 according to the two tests.

Finally, for most of the time intervals across days of the week by the 5- minute interval, there is no variation in both the intraday returns and the medians, and the null hypothesis is not rejected. So, there is no the day-of-week effect by the time interval except for the interval 2 for both markets.

53 Monday effects are not obvious in the two indices. The null hypothesis H0E

(FMonday-test) that the Monday mean return is equal to the average of the means of the other days of the week cannot be rejected for the Shanghai index but is rejected for the Shenzhen index at the 5% significance level. The null hypothesis H0F (FMonday- test) that the Monday mean is equal to the average of the means of the other days of the week by the 5-minute interval is rejected for each index for the 1st and 2nd intervals (the first two 5-minute intervals) at the 1% significance level, and is rejected for the Shanghai index at the 16th interval at the 5% significance level. Thus, there is the Monday effect on the Shenzhen Stock Exchange during the period and the Monday effect on both stock exchanges in the first two intervals.

For the Monday effect, the longer the non-trading period the stronger the impact on stock prices in the opening, which supports the closed-market hypothesis that explains the negative Monday effect and relatively higher volatility on Mondays in anomalies literature (Harris, 1986). Contrary to the previous literature, the highest opening volatility among the days of the week occurred on Thursdays (0.460% for the Shanghai index and 0.473% for the Shenzhen index) but not on Mondays, which strengthens the explanation of characteristics of Chinese stock market settings.

The null hypothesis H0E (FFriday-test) that the Friday mean return is equal to the average of the means of the other days of the week cannot be rejected for either index. The null hypothesis H0F (FFriday-test) that the Friday mean is equal to the average of the other weekday means by the 5-minute interval is rejected at the 5% significance level for each index for the 32nd interval and for Shenzhen index at 42nd and 44th intervals. In general, there is no Friday effect on the two stock markets except for interval 32.

54 To sum up, the intraday return during the period exhibits the striking U- shaped patterns in the Chinese stock market. Test results highlight the intraday return

(F-test 2) and medians (KW-test 2) variations. Thus, the mean returns (medians) are not equivalent across the 5-minute intervals during the period, which further verifies the existing systematic patterns during the period and by the day of the week. There is a strong time-of-day effect on the Shanghai index and the Shenzhen index but the day-of-week effects are not obvious. In addition, large closing returns in the two stock exchanges imply that the specialist-related explanation of Miller (1989) should be refuted in the Chinese stock market since there is no specialist system. The findings show that the large price changes at market openings and closings need to be stabilized in the two Chinese exchanges, which underscores the importance of making the trading systems more efficient.

3.6 Empirical Results of Intraday Return Volatility

In this section, the characteristics of intraday volatility will be investigated based on the period of the trading day and weekdays. The Levene-test and Brown-

Forsythe test are used for the equality of return variance across days of the week, and across days of the week by the 5-minute interval. Again, Section 3.7 will present theoretical explanations for the intraday patterns in the Chinese stock market.

55 3.6.1 Intraday Return Volatility during the Period

Table 3.5 and Table 3.6 present the standard deviation of the mean 5-minute returns during the period and by the day of the week for the Shanghai and Shenzhen indices. To further analyze the intraday seasonality of two indices volatility, Figure

3.5 plots the mean volatility in terms of the standard deviation against the 5-minute interval. With the largest volatility at the beginning of the morning session, the two indices have distinct L-shaped patterns, inconsistent with the U-shaped patterns observed in the US market by Wood, McInish and Ord (1985), Harris (1986), Jain and Joh (1988) and Goodhart and O’Hara (1997).

The mean volatility during the opening, or the first 5-minute interval, is the largest (0.363% and 0.368% for the Shanghai and the Shenzhen indices, respectively), and is almost triple that of the rest of the day. Volatility goes down sharply during the next 25 minutes. The lowest volatility of 0.096% occurs before the

5-minute ending of the morning session in the Shanghai market and the lowest volatility occurs in Shenzhen market (0.109%) at 11:10 and 11:25. The volatility begins to rise in the last 5-minutes of the morning session to 0.123% and 0.145% for the Shanghai and the Shenzhen markets, respectively. At the beginning of the afternoon session volatility continues to rise to a higher level (0.194% and 0.187% for the Shanghai and the Shenzhen indices, respectively) in the first 5 minutes, which represents the largest volatility throughout the afternoon session. After that it comes down again (with lowest 0.093% at 13:20 in Shanghai market) and fluctuates until the end of the day to 0.116% and 0.17%, respectively, which is 1/3 and 1/2 of its value at the opening of two indices, respectively.

56 To sum up, it is very obvious that the intraday volatility follows a striking L- shaped pattern on the Chinese stock market. The large opening volatility of the

Shanghai index and the Shenzhen index also support the fact that the mean return at the opening is significantly higher than for most other intervals of the day.

3.6.2 Intraday Return Volatility during the Weekdays

Figure 3.6 plots the volatility in terms of the standard deviation of the

Shanghai index and the Shenzhen index against the 5-minute intervals by the day of the week. For all days of the week the indices have similar intraday volatility patterns and these patterns are also very similar to the pattern during the period, which is distinctly L-shaped.

Volatility is higher at the opening of the two sessions during the weekdays for both indices. In general, the highest volatility occurs in the first 10-minutes of each day for both indices. The opening volatility for the first 5-minute period of each weekday is higher for both indices except for Monday and Tuesday of the Shenzhen index, where it occurs in the second 5-minute period. Volatilities at the opening of

Thursdays (0.460% and 0.473% for the Shanghai and Shenzhen indices, respectively) are the highest, and are approximately 51% and 64% higher than the lowest value of Wednesday’s opening. The L-shaped volatility patterns on each weekday are consistent with the findings of Tang and Lui (2002) in the Hong Kong market.

Finally, in order to verify the stability of the results, the entire 3-year period is divided into three subperiods: year 2000, year 2001 and year 2002. Moreover, the

57 intraday intervals use relatively finer and wider frequency data of 1-minute and 10- minute interval returns. Figure 3.7, Figure 3.8, Figure 3.9 show the mean 5-minute returns, cumulative mean returns and return volatility based on standard deviations against the 5-minute interval during year 2000, year 2001 and year 2002, separately.

These are consistent with the patterns during the period of three years. Figure 3.10,

Figure 3.11 show the mean 1-minute and 10-minute returns, cumulative mean returns and return volatility based on standard deviations against the 5-minute interval during the 3-year period for the Shanghai and Shenzhen indices, which further support the results that the findings are not the chance results of the particular calendar period studied. There exists an intraday variation for each subperiod, and each interval of 1-minute and 10-minute of the 3-year period. The mean intraday returns are the highest in the day-end at 15:00. The cumulative intraday returns are the lowest between 13:15 and 13:30. For the 1-minute mean return data, the exact time is at 13:23. Furthermore, the highest volatility appears at the opening of the morning session.

3.6.3 Test Results of the Volatilities

On the right side of Table 3.5 and Table 3.6, the column Levene-test 1 and column Brown-Forsythe 1 contain the test-statistic for the equality of return variance across days of the week, and across days of the week by the 5-minute interval. The row Levene-test 2 and row Brown-Forsythe 2 contain the test-statistic for testing the equality of intraday return variances across the 5-minute intervals, and across the 5- minute intervals by the day of the week.

58 Firstly, the null hypothesis H0A and H0B (Levene-test 2) of equality of volatility of mean 5-minute returns across the 5-minute intervals during the period and by the day of the week for each index is rejected at the 1% significance level.

The Brown-Forsythe 2 tests have the same results as Levene-test 2. Thus, intraday volatility in terms of the variance of 5-minute returns is not distributed equally across the 5-minute intervals in a given day and the time-of-day effect on the intraday return volatilities of the Shanghai index and the Shenzhen index can not be rejected, which means that there are systematic intraday volatility patterns existing on the Chinese stock market.

Secondly, the null hypothesis H0C (Levene-test 1 and Brown-Forsythe 1) of equality of volatility of mean 5-minute returns across days of the week for each index is rejected at the 1% significance level. Thus, there are systematic volatility patterns existing across days of the week on the Chinese stock market. The day-of- week effect on the intraday return volatilities of the Shanghai index and the

Shenzhen index can be supported.

Thirdly, the null hypothesis H0D (Levene-test 1 and Brown-Forsythe 1) of equality of volatility of mean 5-minute returns across days of the week by the 5- minute interval for each index can not be rejected except for the 40th interval at

14:20 at the 5% significance level for both indices. Thus, most of the 5-minute intervals across days of the week have no variation in the return variance. The

Levene test and Brown-Forsythe statistics show that intraday volatility in terms of the variance of 5-minute returns is distributed equally across days of the week by the

5-minute interval except where the null hypothesis is rejected at 14:20 for both indices at the 5% significance level.

59 To summarize, the intraday return volatility during the period exhibits the striking L-shaped volatility pattern on the Chinese market, which is in contrast to the

U-shaped volatility patterns documented in previous literature in other markets. The tests further verify the existing systematic patterns during the period and by the day of the week. There is a strong time-of-day effect of volatility on both the Shanghai index and the Shenzhen index. There are also day-of-week effects of volatility on the two markets. The findings show that the high volatility at market openings needs to be reduced in the two Chinese exchanges, which underscores the importance of making the trading systems more efficient.

3.7 Explanations of the L-shaped Volatility

The noticeable L-shaped volatility pattern is due to flow of information and market microstructure, which are related to the generation and dissemination of information, the arrival of orders and the rules and institutional features of a stock market that determine how orders are transformed into trades.

Kyle (1985) developed a model that classifies traders into three classes: (1) private information traders, (2) random liquidity traders, and (3) market makers.

Admati and Pfleiderer (1988) extended Kyle’s model further by adding a fourth type of traders: (4) discretionary liquidity traders, to explain the volatility patterns in the

U.S. market. They then argued that both the liquidity (random and discretionary) traders and the informed traders would converge to trade during the period of the open and close of the market, causing a systematic pattern in return volatilities (the

U-shaped patterns).

60 These explanations can be also applied to the Chinese market and the L- shaped volatility pattern. Overnight information has an enormous impact on stock volatility. This implies that accumulated information (public and private) from overnight is reflected in prices immediately at the opening of the day. In addition, noise trading may be another source of higher volatility in the market opening.

Therefore, in contrast to other 5-minute intraday price changes that reflect the news released during the corresponding 5-minute interval, the first few 5-minute return volatilities reflect the assimilating information that was released and accumulated due to a much longer overnight trading halt. These abnormal trading activities of different traders in the market’s opening are related to their different information

(public and private) that causes the L-shaped volatility in the markets.

As such, the high volatility at the open would be evidence of price change during the opening processing period. During the trading halts, information released by the central government, related management agencies and the companies, about macroeconomic, social or political news, and trading information, significantly affect the market’s outlook as well as investment decisions3.

Furthermore, private information is most likely asymmetric among investors.

Private information accumulated over the trading halt is incorporated in the stock prices when the trading opens. Traders who have no access to private information would trade on random factors but smart traders may try to trade along with the trading activities of private information traders in order to obtain the private information indirectly. Moreover, the Chinese stock market is not a specialist system and does not have market makers. The short history of the Chinese market is dominated by a high number of inexperienced domestic investors in share trading.

3 In China, most important events, including government policy information and firm-specific information, are released in the evening.

61 They tend to show severe ‘herd behaviour’ in that they follow the tendencies of other traders when buying or selling. These noise-traders may be the second source of high volatility in the morning opening.

The volatilities of the morning and the afternoon closing in both the Shanghai and the Shenzhen markets, although slightly higher, are not as dramatic, as documented in the US market. One possible reason is that the market portfolio’s rebalancing force which causes higher volatility at the close, does not exist in the

Chinese market, and this is consistent with the Hong Kong market (Lam and Tong,

1999 and Tang and Lui, 2002).

In conclusion, high intraday volatilities observed during the opening on the

Chinese stock market are costly to the market in aggregate, which discourages trade and makes investors’ returns more uncertain. The high volatility at the opening during the period and weekdays in the Chinese market will underscore the market efficiency and also make investors reduce their trading while their investing returns are eroded by trading costs.

3.8 Volatility in Trading and non-Trading Periods

Some studies focused on the volatility in trading and non-trading periods

(French and Roll, 1986; Barcaly, Litzenberger and Warner, 1990 and Stoll and

Whaley, 1990). Studies have found that the return variance in the active trading hours (open-to-close) of the security market is larger than that in the nontrading hours (close-to-open). The large difference between variances of trading and non- trading periods is mainly attributable to the trading activities of informed investors.

62 Table 3.7 shows the descriptive statistics of mean return and variance for the overnight period (close-to-open), morning session, lunch break, afternoon session, and daily trading period (open-to-close) for the Shanghai and the Shenzhen indices.

In Table 3.7, volatility in terms of the variance of the open-to-close period is more than four times that of the close-to-open (overnight) period in the Shanghai index (1.703% versus 0.388%) and triple in the Shenzhen index (1.858% versus

0.627%). This is consistent with the U.S. results as reported by Lockwood and Linn

(1990) in that the open-to-close return variance exceeds that of close-to-open return by a factor of 2.34 to 3.37. The higher the volatility, the larger the information flow affected in the Chinese stock market. The results are also consistent with the hypothesis that private information is disclosed during trading hours. French and

Roll (1986) test the private information hypothesis and report that return variances are reduced by both the election day closing and the exchange holidays, whereas

Barclay, Litzenberger and Warner (1990) support the same hypothesis using the opposite case in which return variances are increased by Saturday trading at the

Tokyo Stock Exchange. It should be noted that the return volatility in the morning trading session (0.604% and 0.734% for the Shanghai and the Shenzhen indices, respectively) is greater than the volatility measured during the afternoon trading session (0.529% and 0.537% for the Shanghai and the Shenzhen indices, respectively). This confirms further that private information is not produced at a constant rate even during the trading period. It also implies fewer price reversals as trade proceeds.

Table 3.7 also shows that the overnight volatility is much larger than after the lunch break in the Shanghai index (0.388% vs. 0.006%, or 0.021% per hour vs

0.004% per hour) and the Shenzhen index (0.627% vs. 0.010%, or 0.034% per hour

63 vs 0.007% per hour), consistent with the findings of Amihud and Mendelson (1991) for the Japanese market. Stock return volatility is due to the arrival of new information or the trading process itself. If trading contributes to volatility either through the revelation of private information or due to pricing errors, the return variance over the 1-hour-and-half nontrading period (lunch break) should be lower.

A lower volatility during the 11:30 to 13:00 period could also reflect a lower intensity of new information arriving over these one and half hours. In addition, the increase in the volatility at the opening of the afternoon session supports Brock and

Kleidon’s (1992) view that no trade is the reason. It can be said that volatility follows a broadly L-shaped pattern if the increase in volatility at the opening of the second session is ignored.

In the Shanghai market, the sum (1.527) of the mean variances of the four periods: returns morning (0.604), lunch (0.006), afternoon (0.529) and overnight

(0.388) is less than 25.73% of the mean daily variance of close-to-close return

(2.056) and the Shenzhen index further verifies this with the total of the four mean variances (1.908) less than 14.32% of the mean daily variance of close-to-close return (2.277)4. These results are consistent with Chang et al (1993) and Lam and

Tong (1999) but not with Amihud and Mendelson (1991) and Bildik (2001). Amihud and Mendelson (1991) found that the sum of variances of the four separate time intervals, including overnight, morning session, lunch break, and afternoon session, exceeds the close-to-close return variance by 21.7%.

In summary, the return variances in the active trading hours (open-to-close) on the Chinese stock market are larger than those in the nontrading hours (close-to- open). This is consistent with the hypothesis that more public information reaches the

4 The computation for these return series are for Morning: 09:30 – 11:30, for lunch break: 11:30 – 13:00 (afternoon opening price), for afternoon: 13:00 – 15:00 and for overnight: 15:00 – 09:30 next day.

64 marketplace during normal trading periods than the non-trading periods. In addition, the variances of overnight are larger than these of lunch break and variances of the morning session are larger than those of the afternoon session.

3.9 Call Auction and Continuous Trading

Amihud and Mendelson (1991) compare the variability of interday 24-hour returns on the Japanese stock market, such as morning open to previous morning open and afternoon open to afternoon open. They note that the return volatility at the afternoon open is not different from the volatility observed at the afternoon close, even though the same Itayose method (call auction) is employed both in the morning and the afternoon opening transaction. They conclude that it is caused by the preceding long period of non-trading rather than by the Itayose method. The trading mechanisms on the Chinese stock market are different from those of other countries.

There is call auction for the morning opening and then continuous trading sessions.

Between the morning opening (09:25) and the beginning continuous trading sessions

(09:30), there is a five-minute waiting period. In addition, there is only one call auction for the opening in the morning, and no call auction in the afternoon opening.

This section will not discuss the advantages of the two trading mechanisms: call auction and continuous matching. However, this section discusses the question: is the call auction the source of high volatility at the morning opening? For this reason, interday 24-hour returns are computed by the 5-minute interval. Table 3.8 shows the interday descriptive statistics of mean interday 24-hour returns, variances and the tests of variance ratio for call auction opening (09:25), first interval at 09:35

65 (in the continuous trading), the second interval at 09:40, and other related 5-minute intervals.

For the purpose of comparison, variance ratios for each interday 24-hour returns compared with the afternoon close-to-close are reported. According to

Amihud and Mendelson (1987), one would expect that if the variance ratio test of the call auction is larger and significantly different but the variance ratio test of all other interday returns in the continuous trading period are not larger and significantly different compared with the close-to-close interday return, the market volatility in the call auction may indeed be affected by the call auction method.

The findings show that the interday 24-hour open-to-open (call auction) volatilities (2.535% and 2.983%) are higher than and significantly different to the interday 24-hour close-to-close (C-C) volatilities (2.056% and 2.278%) for the

Shanghai and Shenzhen indices, respectively. The variance ratio tests of the first interday 24-hour interday return at 09:35 are also larger and significantly different to the interday 24-hour close-to-close interday return, which is the same as the call auction’s variance ratio test value in both the Shanghai and the Shenzhen markets.

Thus, the hypothesis is rejected because no sudden drop is observed in interdaily variance after the market opening according to the hypothesis suggested by Gerety and Mulherin (1994).

The results provide further evidence on the Chinese stock market that information accumulation due to the overnight halt of trade is the driving force for higher return volatility at the market opening. The trading mechanism (call auction) before the morning session is not the source of high volatility in the beginning of the continuous trading session in the morning. Markets are mainly dominated by accumulated overnight news and information released in the halt of trade. The result

66 is basically consistent with the variance ratio of 1.20 found by Amihud and

Mendelson (1987) and of 1.13 found by Stoll and Whaley (1990), who showed that the interday 24-hour open-to-open return variance is higher than the interday 24-hour close-to-close return variance in the US stock market.

3.10 Conclusion

The object of this chapter is to examine the behaviour of intraday return and return volatility in the Chinese stock market, which is a limited order-driven market using electronic trading without market makers. Similar results have been found in

Western markets despite the differing microstructures, and institutional framework between the different markets. The statistical properties and systematic characteristics of intraday returns and return volatility using 5-minute data of the

Shanghai Composite Index and the Shenzhen Component Index are explored over a three-year period. Several parametric and non-parametric tests are used to investigate if there are systematic patterns according to the EMH. Furthermore, volatilities between the trading and non-trading periods are examined. Finally, interday 24-hour return and return volatility are explored and variance ratio tests are used to verify the trading mechanism on call auction and continuous trading for high volatility in the market opening in accordance with Amihud and Mendelson (1987).

The results of this study are very interesting. For both the Shanghai and the

Shenzhen indices, first, the distributions of the 5-minute returns are distinctly fat- tailed, non-normal and leptokurtic during the three-year period.

67 Second, the mean intraday 5-minute returns are negative and not significantly different from zero, but overnight (close-to-open) returns are positive and significantly different from zero during the three-year period. The closing mean returns (14:55-15:00) are highest and larger opening returns (09:30-09:35) are also observed in the trading day. Both of these are statistically significantly different from zero.

Third, over the trading day, intraday returns broadly follow a U-shaped pattern as reported by Wood, McInish and Ord (1985) while the volatility of returns broadly follows an L-shaped pattern. This result is consistent with the previous findings in the Hong Kong market (Lam and Tong, 1999 and Tang and Lui, 2002).

Fourth, over the weekdays, intraday returns also broadly follow the U-shaped patterns except on Monday and volatility of the intraday returns broadly follows an

L-shaped pattern on each weekday. Negative and large mean returns during the

Monday opening in the Chinese market are consistent with the previous finding of

Harris (1986).

Fifth, in order to verify the stability of the results, the entire period is divided into three subperiods: year 2000, year 2001 and year 2002. The relatively finer and wider frequency data of 1-minute and 10-minute interval return series are also used.

These five different processes indicate that there exists an intraday variation for each subperiod, and each 1-minute and 10-minute period. The conclusions are consistent with the intraday 5-minute return characteristics.

Sixth, for the volatility of trading and non-trading period, the results also show that the return variance in the active trading period (open-to-close) of the stock market is larger than that in the nontrading hours (close-to-open). The volatility of interday 24-hour returns is calculated and the variance ratio test is used to show that

68 high volatility of intraday returns for the market opening is not mainly due the trading mechanism (call auction in the market opening) but the overnight trading halt.

This chapter examines the behaviour of intraday return and return volatility of the market indices on the Shanghai Stock Exchange and the Shenzhen Stock

Exchange. Analyses are based on 5-minute high-frequency data over a three-year period. The next chapter will explore liquidity characteristics (bid/ask spread and depth) and its determinants on the Shanghai Stock Exchange using sixty individual stocks.

69

Table 3.1 Descriptive statistics of the intraday 5-minute returns in percentages of the Shanghai Composite Index, 2000 - 2002 No Interval Size Mean Max Min. Std. Skew. ExKurt. J.Bera Cum. No Interval Size Mean MaxMin. Std. Skew. ExKurt. J.Bera Cum. 1 9:30-9:35 716 0.0306 2.39 -2.39 0.363 -0.02 12.24 2545.1 0.031 25 13:00-13:05 716 0.0036 1.06 -2.27 0.194 -2.047 32.27 26061 -0.137 2 9:35-9:40 716 -0.0403 1.62 -1.37 0.266 0.615 9.93 1477.7 -0.010 26 13:05-13:10 716 -0.0252 0.73 -0.49 0.091 1.114 15.19 4584.1 -0.162 3 9:40-9:45 716 -0.0110 1.18 -1.09 0.201 0.774 9.65 1390.4 -0.021 27 13:10-13:15 716 -0.0179 1.3 -0.54 0.106 2.93 40.33 42598 -0.18 4 9:45-9:50 716 -0.0025 0.88 -0.83 0.151 0.252 8.19 812.5 -0.023 28 13:15-13:20 716 -0.0049 0.95 -0.71 0.093 0.952 28.76 19911 -0.185 5 9:50-9:55 716 -0.0070 0.74 -0.73 0.151 -0.33 7.78 694.4 -0.030 29 13:20-13:25 716 -0.0002 0.72 -0.42 0.097 0.581 10.48 1711.3 -0.185 6 9:55-10:00 716 -0.0103 0.73 -0.82 0.141 -0.11 9.31 1189.1 -0.040 30 13:25-13:30 716 0.0027 0.68-0.49 0.109 0.376 9.07 1116.6 -0.182 7 10:00-10:05 716 -0.0080 0.85 -0.76 0.125 -0.22 10.11 1515.1 -0.048 31 13:30-13:35 716 0.0056 0.83-0.46 0.117 0.463 9.44 1263.7 -0.177 8 10:05-10:10 716 -0.0090 0.87 -1.25 0.153 -0.91 14.45 4013.7 -0.057 32 13:35-13:40 716 0.008 0.9 -0.630.114 0.668 12.94 3003.3 -0.169 9 10:10-10:15 716 -0.0062 0.95 -0.67 0.135 0.159 9.51 1268.4 -0.064 33 13:40-13:45 716 0.0124 0.65 -0.58 0.114 0.283 8.41 883 -0.156 10 10:15-10:20 716 -0.0114 0.93 -0.88 0.15 -0.13 11.59 2203.4 -0.075 34 13:45-13:50 716 0.0088 1.06-1.02 0.134 0.663 18.5 7215.8 -0.148 11 10:20-10:25 716 -0.0071 1.09 -0.74 0.138 0.428 12.22 2560.5 -0.082 35 13:50-13:55 716 0.005 0.72 -0.72 0.124 -0.013 11.38 2095.3 -0.143 12 10:25-10:30 716 -0.0024 0.59 -0.79 0.129 -0.75 9.81 1450.2 -0.085 36 13:55-14:00 716 0.0052 0.69-0.59 0.126 0.336 9.41 1241.1 -0.137 13 10:30-10:35 716 0.0064 0.82 -1.21 0.136 -0.86 16.43 5472 -0.078 37 14:00-14:05 716 0.0029 0.57 -0.7 0.125 -0.394 8.69 985.1 -0.135 14 10:35-10:40 716 0.0017 0.86 -1.09 0.137 -0.09 14.68 4070.3 -0.077 38 14:05-14:10 716 0.003 0.57 -0.7 0.136 -0.25 7.8 694.6 -0.132 15 10:40-10:45 716 -0.0032 0.98 -0.57 0.131 1.103 12.21 2674.9 -0.080 39 14:10-14:15 716 0.0043 1.12 -0.7 0.136 0.903 15.78 4967.7 -0.127 16 10:45-10:50 716 -0.0088 0.86 -0.65 0.13 0.396 10.74 1807.2 -0.089 40 14:15-14:20 716 0.0032 1.14 -0.65 0.142 0.939 13.29 3266.1 -0.124 17 10:50-10:55 716 -0.0051 0.72 -0.52 0.116 0.524 9.16 1165.5 -0.094 41 14:20-14:25 716 -0.0034 1.01 -1.02 0.144 -0.427 13.06 3038.3 -0.127 18 10:55-11:00 716 -0.0068 0.86 -0.52 0.119 0.94 12.32 2699.3 -0.100 42 14:25-14:30 716 -0.0044 0.57 -0.76 0.137 -0.435 7.81 713.3 -0.132 19 11:00-11:05 716 -0.0141 0.68 -0.98 0.117 -0.19 15.61 4746.4 -0.115 43 14:30-14:35 716 0.0052 0.9 -0.750.15 0.245 8.91 1050.9 -0.127 20 11:05-11:10 716 -0.0135 0.57 -0.69 0.111 -0.11 8.75 987.7 -0.128 44 14:35-14:40 716 0.0019 0.88 -0.87 0.154 -0.218 9 1079.1 -0.125 21 11:10-11:15 716 -0.0072 0.53 -0.9 0.112 -0.62 12.42 2692.7 -0.135 45 14:40-14:45 716 -0.0149 0.75 -0.97 0.158 -0.718 9.1 1172.2 -0.14 22 11:15-11:20 716 -0.0107 0.47 -0.76 0.1 -0.59 10.38 1668.6 -0.146 46 14:45-14:50 716 -0.0208 0.67 -1.45 0.148 -1.452 18.15 7096.5 -0.16 23 11:20-11:25 716 -0.0072 0.48 -0.65 0.096 -1.11 12.52 2852 -0.153 47 14:50-14:55 716 -0.0078 0.74 -0.95 0.128 -0.94 13.55 3427.7 -0.168 24 11:25-11:30 716 0.0125 0.68 -1.76 0.123 -4.67 69.55 134743 -0.141 48 14:55-15:00 716 0.1017 0.77 -0.44 0.116 0.193 7.96 738.5 -0.067 Overnight 715 0.0655 8.70 -2.26 0.623 8.22 101.477 296966 All 34368 -0.0014 2.39 -2.39 0.146 -0.024 22.422 540173 3 2 Note: Time of the trading day effect: OLS Results (R = 0.016, Q1 = 0.011, Q5 = 2.14), rDt1122=+ββtt D +++…… β n Dnti βε rti +t,1 n =,2,,49 ∑ − i=1 Mean = 0 test of significance: (1) Mean returns of intervals: 1, 2, 26, 46 and 48 are different than zero at the 1% level of significance (2) Mean return of intervals: 4, 19, 24 and 45 at the 5% level of significance. (3) Mean return of intervals: 10, 20, 22 and 33 at the 10% level of significance. (4) Mean return of interval 49 (Overnight): at the 1% level of significance. (5) All does not include overnight returns 70

Table 3.2 Descriptive statistics of the intraday 5-minute returns in percentages of the Shenzhen Component Index, 2000 - 2002 No Interval Size Mean Max Min. Std. Skew. ExKurt. J.Bera Cum. No Interval Size Mean MaxMin. Std. Skew. ExKurt. J.Bera Cum. 1 9:30-9:35 716 0.0285 2.31 -2.51 0.368 -0.325 13.58 3352.4 0.029 25 13:00-13:05 716 0.0219 1.17 -1.28 0.187 0.052 13.43 3247.8 -0.112 2 9:35-9:40 716 -0.0355 2.24 -2.49 0.342 -0.093 14.29 3803.1 -0.007 26 13:05-13:10 716 -0.0243 0.73 -1.09 0.130 -0.279 15.16 4420.8 -0.136 3 9:40-9:45 716 -0.0234 1.32 -1.08 0.251 0.844 7.69 740.9 -0.030 27 13:10-13:15 716 -0.0270 0.94 -0.91 0.129 0.058 14.94 4254.6 -0.163 4 9:45-9:50 716 -0.0088 1.35 -1.13 0.201 0.275 9.61 1311.8 -0.039 28 13:15-13:20 716 -0.0132 1.15 -1.21 0.138 -0.165 24.90 14314.8 -0.177 5 9:50-9:55 716 -0.0099 0.97 -1.00 0.189 0.092 8.04 759.1 -0.049 29 13:20-13:25 716 -0.0077 1.36 -0.70 0.134 1.548 21.54 10546.2 -0.184 6 9:55-10:00 716 -0.0066 0.88 -1.16 0.173 -0.117 8.59 935.0 -0.056 30 13:25-13:30 716 -0.0001 0.88 -0.57 0.131 0.684 9.75 1416.3 -0.184 7 10:00-10:05 716 -0.0031 0.95 -0.88 0.138 -0.344 12.04 2451.9 -0.059 31 13:30-13:35 716 0.0049 0.74 -0.72 0.135 0.073 7.79 683.9 -0.180 8 10:05-10:10 716 -0.0059 1.18 -1.26 0.199 -0.137 11.06 1938.9 -0.065 32 13:35-13:40 716 0.0061 1.11 -0.93 0.140 0.082 15.02 4310.0 -0.174 9 10:10-10:15 716 -0.0093 0.80 -0.86 0.159 -0.338 7.86 717.4 -0.074 33 13:40-13:45 716 0.0093 0.67 -0.68 0.122 0.436 9.32 1214.9 -0.164 10 10:15-10:20 716 -0.0085 1.25 -1.11 0.172 0.279 13.39 3232.7 -0.083 34 13:45-13:50 716 0.0155 2.10 -0.70 0.190 3.076 32.44 26983.9 -0.149 11 10:20-10:25 716 -0.0062 1.27 -0.76 0.169 0.710 12.16 2562.5 -0.089 35 13:50-13:55 716 0.0029 0.78 -1.18 0.152 -0.472 12.98 2999.5 -0.146 12 10:25-10:30 716 -0.0026 0.82 -1.04 0.154 -0.856 11.49 2237.9 -0.091 36 13:55-14:00 716 0.0071 0.83 -0.88 0.151 0.208 10.30 1596.6 -0.139 13 10:30-10:35 716 0.0129 1.08 -0.89 0.159 0.752 10.34 1675.8 -0.079 37 14:00-14:05 716 0.0064 1.33 -0.91 0.159 0.952 18.43 7210.3 -0.132 14 10:35-10:40 716 0.0015 1.09 -1.93 0.176 -0.947 28.79 19943.7 -0.077 38 14:05-14:10 716 0.0011 0.78 -0.76 0.160 0.136 7.63 642.2 -0.131 15 10:40-10:45 716 -0.0027 0.92 -0.66 0.152 0.762 8.81 1074.9 -0.080 39 14:10-14:15 716 0.0049 1.16 -1.00 0.161 0.640 14.71 4139.7 -0.126 16 10:45-10:50 716 -0.0119 0.70 -0.82 0.147 -0.153 8.24 821.2 -0.092 40 14:15-14:20 716 0.0027 0.94 -1.00 0.154 0.178 10.27 1579.6 -0.124 17 10:50-10:55 716 0.0000 0.73 -0.62 0.138 0.355 7.84 714.3 -0.092 41 14:20-14:25 716 -0.0055 0.89 -2.05 0.178 -2.036 28.91 20529.3 -0.129 18 10:55-11:00 716 -0.0030 0.86 -0.61 0.137 0.884 9.65 1412.6 -0.095 42 14:25-14:30 716 -0.0073 0.76 -0.96 0.162 -0.904 10.57 1806.7 -0.136 19 11:00-11:05 716 -0.0150 0.78 -1.05 0.139 -0.042 12.77 2850.8 -0.110 43 14:30-14:35 716 0.0061 0.79 -0.74 0.165 0.318 7.09 510.7 -0.130 20 11:05-11:10 716 -0.0097 0.50 -0.48 0.109 0.395 7.48 617.3 -0.119 44 14:35-14:40 716 0.0013 0.91 -1.01 0.174 -0.109 8.41 873.3 -0.129 21 11:10-11:15 716 -0.0095 1.36 -0.87 0.158 0.882 15.24 4559.5 -0.129 45 14:40-14:45 716 -0.0129 1.32 -1.13 0.175 -0.084 13.50 3287.9 -0.142 22 11:15-11:20 716 -0.0143 0.62 -1.08 0.127 -1.283 13.86 3713.7 -0.143 46 14:45-14:50 716 -0.0194 0.81 -2.09 0.158 -3.157 46.45 57499.7 -0.161 23 11:20-11:25 716 -0.0038 0.73 -0.70 0.109 -0.330 11.13 1984.8 -0.147 47 14:50-14:55 716 -0.0085 0.67 -1.16 0.169 -1.260 13.14 3255.3 -0.170 24 11:25-11:30 716 0.0130 0.74 -1.74 0.145 -2.605 37.43 36169.8 -0.134 48 14:55-15:00 716 0.0830 1.04 -1.09 0.170 0.246 11.25 2037.5 -0.087 Overnight 715 0.0586 9.02 -8.04 0.792 2.633 74.122 151522 All 34368 -0.0018 2.31 -2.51 0.173 -0.027 21.227 475768 3 2 Note: Time of the trading day effect: OLS Results (R = 0.094, Q1 = 0.005, Q5 = 3.658), rDt1122=+ββtt D +++…… β n Dnti βε rti +t,1 n =,2,,49 ∑ − i=1 Mean = 0 test of significance: (1) Mean returns of intervals: 1, 2, 26 and 48 are different than zero at the 1% level of significance. (2) Mean return of intervals: 25, 27, 34 and 46 at the 5% level of significance. (3) Mean return of intervals: 4, 13, 19, 22 and 28 at the 10% level of significance. (4) Mean return of interval 49 (Overnight): at the 1% level of significance (5) Area All does not include overnight returns 71 Table 3.3 Mean intraday 5-minute returns by trading time and weekday in percentages of the Shanghai Composite Index, 2000 - 2002 F-Test KW- F-Test KW- Time Mon Tue Wed Thu Fri Overall 1 Test 1 F-Mon F-Fri Time Mon Tue Wed Thu Fri Overall 1 Test 1 F-Mon F-Fri 9:30-9:35 -0.0364 b 0.0506 c 0.0291 b 0.0465 c 0.0626 c 0.0306 c 1.67 10.79** -0.084** 0.04 13:00-13:05 -0.0033 0.0180 -0.0067 -0.0135 0.0233 a 0.0036 0.99 2.50 -0.009 0.025 9:35-9:40 -0.1092 c -0.0046 -0.0532 c -0.0174 -0.0179 -0.0403 c 3.69*** 21.81*** -0.086*** 0.028 13:05-13:10 -0.0248 a -0.0327 a -0.0231 a -0.0301 b -0.0151 -0.0252 c 0.81 6.01 0.001 0.013 9:40-9:45 -0.0353 b 0.0188 -0.0257 b -0.0241a 0.0107 -0.0110 a 2.08* 9.57** -0.03 0.027 13:10-13:15 -0.0220 -0.0247 -0.0104 -0.0088 -0.0234 a -0.0179 c 0.73 1.51 -0.005 -0.007 9:45-9:50 -0.0134 0.0162 -0.0033 0.0000 -0.0122 -0.0025 0.90 7.24 -0.014 -0.012 13:15-13:20 -0.0041 -0.0077 0.0030 -0.0020 -0.0135 -0.0049 0.63 3.00 0.001 -0.011 9:50-9:55 -0.0209 0.0106 -0.0047 0.0038 -0.0237 a -0.0070 1.42 12.65** -0.017 -0.021 13:20-13:25 0.0113 0.0038 0.0016 -0.0076 -0.0098 -0.0002 1.13 8.32* 0.014 -0.012 9:55-10:00 -0.0082 0.0021 -0.0205 -0.0165 -0.0085 -0.0103 0.55 2.52 0.003 0.002 13:25-13:30 0.0129 0.0087 -0.0016 0.0014 -0.0078 0.0027 0.81 6.78 0.013 -0.013 10:00-10:05 -0.0090 0.0047 -0.0156 -0.0113 -0.0089 -0.0080 0.53 0.59 -0.001 -0.001 13:30-13:35 0.0074 0.0030 0.0053 0.0073 0.0048 0.0056 0.04 1.10 0.002 -0.001 10:05-10:10 -0.0092 -0.0035 -0.0122 -0.0011 -0.0191 -0.0090 0.31 0.21 0 -0.013 13:35-13:40 0.0087 -0.0093 0.0046 0.0103 0.0258 a 0.0080 1.76 5.27 0.001 0.022** 10:10-10:15 0.0001 -0.0032 -0.0162 0.0024 -0.0141 -0.0062 0.56 1.19 0.008 -0.01 13:40-13:45 0.0077 0.0132 0.0073 0.0078 0.0257 a 0.0124 b 0.68 3.99 -0.006 0.017 10:15-10:20 0.0051 -0.0225 -0.0104 -0.0051 -0.0239 a -0.0114 a 0.94 1.46 0.021 -0.016 13:45-13:50 0.0030 0.0126 0.0023 0.0134 0.0125 0.0088 0.25 3.59 -0.007 0.005 10:20-10:25 -0.0078 -0.0016 -0.0059 0.0042 -0.0245 a -0.0071 0.87 2.32 -0.001 -0.022* 13:50-13:55 0.0069 0.0124 0.0002 -0.0051 0.0107 0.0050 0.50 2.49 0.002 0.007 10:25-10:30 -0.0025 0.0018 0.0045 -0.0027 -0.0128 -0.0024 0.37 1.71 0 -0.013 13:55-14:00 0.0107 0.0191 -0.0033 -0.0083 0.0074 0.0052 1.08 7.76 0.007 0.003 10:30-10:35 0.0140 0.0096 0.0122 -0.0010 -0.0029 0.0064 0.47 3.06 0.01 -0.012 14:00-14:05 -0.0005 0.0193 -0.0106 0.0139 -0.0074 0.0029 1.61 5.08 -0.004 -0.013 10:35-10:40 -0.0006 0.0025 -0.0003 0.0067 0.0000 0.0017 0.07 0.54 -0.003 -0.002 14:05-14:10 0.0035 -0.0012 -0.0065 0.0157 0.0035 0.0030 0.52 2.44 0.001 0.001 10:40-10:45 -0.0131 -0.0026 -0.0102 -0.0013 0.0110 -0.0032 0.73 2.95 -0.012 0.018 14:10-14:15 0.0204 -0.0085 0.0113 0.0114 -0.0129 0.0043 1.59 6.68 0.02 -0.022 10:45-10:50 0.0149 -0.0129 -0.0208 -0.0283 b 0.0030 -0.0088 2.66** 9.76 0.030** 0.015 14:15-14:20 0.0146 -0.0054 0.0045 -0.0030 0.0054 0.0032 0.44 4.35 0.014 0.003 10:50-10:55 0.0103 -0.0169 0.0014 -0.0092 -0.0107 -0.0051 1.25 5.38 0.019 -0.007 14:20-14:25 -0.0073 -0.0093 -0.0013 -0.0086 0.0091 -0.0034 0.41 2.77 -0.005 0.016 10:55-11:00 -0.0078 -0.0232 0.0171 -0.0084 -0.0117 -0.0068 2.21* 8.11* -0.001 -0.006 14:25-14:30 -0.0123 0.0039 0.0045 0.0061 -0.0239 a -0.0044 1.35 3.06 -0.01 -0.024* 11:00-11:05 -0.0198 -0.0180 -0.0101 -0.0176 -0.0051 -0.0141 b 0.41 1.16 -0.007 0.011 14:30-14:35 -0.0116 0.0177 -0.0011 0.0216 -0.0005 0.0052 1.23 2.33 -0.021 -0.007 11:05-11:10 -0.0183 -0.0042 -0.0266 b -0.0158 -0.0029 -0.0135 b 1.17 2.56 -0.006 0.013 14:35-14:40 -0.0023 0.0056 -0.0102 -0.0042 0.0202 0.0019 0.84 9.20* -0.005 0.023 11:10-11:15 -0.0134 -0.0040 -0.0157 -0.0027 -0.0002 -0.0072 0.54 2.98 -0.008 0.009 14:40-14:45 -0.0141 -0.0167 -0.0048 -0.0322 b -0.0068 -0.0149 c 0.67 6.00 0.001 0.01 11:15-11:20 -0.0181 -0.0154 -0.0059 -0.0084 -0.0059 -0.0107 a 0.46 0.30 -0.009 0.006 14:45-14:50 -0.0256 a -0.0158 0.0069 -0.0277 b -0.0415 c -0.0208 c 2.13* 13.42*** -0.006 -0.026 11:20-11:25 -0.0031 -0.0064 -0.0114 -0.0117 -0.0034 -0.0072 0.27 1.48 0.005 0.005 14:50-14:55 0.0009 -0.0111 -0.0011 -0.0086 -0.0190 -0.0078 0.57 5.50 0.011 -0.014 11:25-11:30 0.0209 0.0197 -0.0023 0.0104 0.0139 0.0125 b 0.83 2.04 0.01 0.002 14:55-15:00 0.1089 c 0.1104 c 0.1015 c 0.0984 c 0.0895c 0.1017 c 0.77 4.60 0.009 -0.015 Overnight 0.1257 c 0.0707 c 0.0331 c 0.0201 0.0782 c 0.0655 c 0.63 1.09* 0.075 0.016 All -0.0040 b 0.0021 -0.0028 -0.0013 -0.0011 -0.0014 1.73 8.58* -0.003 0 F-test 2 5.82*** 3.16*** 2.94*** 2.45*** 3.20*** 12.073*** KW-test 2 329.3*** 301.7*** 239.7*** 256.5*** 236.7*** 1119*** Notes: (1) F-test 1 is the F statistic testing the equality of a given intraday return across all days of the week. (2) F-test 2 is the F statistic testing the equality of the intraday returns for a given weekday and all days. (3) KW-test 1 is the non-parametric Kruskal Wallis statistic testing the equality of a given intraday return across all days of the week. (4) KW-test 2 is the non-parametric Kruskal Wallis statistic testing the equality of the intraday returns for a given weekday and all days. (5) The * for 10% level of significance. The ** for 5% level of significance. The *** for 1% level of significance. (6) The a for 10% level of significance. The b for 5% level of significance. The c for 1% level of significance.

72

Table 3.4 Mean intraday 5-minute returns by trading time and weekday in percentages of the Shenzhen Component Index, 2000 - 2002 F-Test KW- F-Test KW- Time Mon Tue Wed Thu Fri Overall 1 Test 1 F-Mon F-Fri Time Mon Tue Wed Thu Fri Overall 1 Test 1 F-Mon F-Fri 9:30-9:35 -0.0498 c 0.0606c 0.0311 b 0.0598c 0.0399 b 0.0306 c 2.19* 15.98*** -0.098*** 0.014 13:00-13:05 0.0000 0.0345 a 0.0116 0.0207 0.0422 c 0.0036 c 1.19 7.15 -0.027 0.025 9:35-9:40 -0.1322 c 0.0090 -0.0462 c -0.0017 -0.0079 -0.0403 c 4.15*** 32.58*** -0.121*** 0.035 13:05-13:10 -0.0303 a -0.0150 a -0.0303 b -0.0322 a -0.0141 -0.0252 c 0.69 1.76 -0.007 0.013 9:40-9:45 -0.0595 c 0.0011 -0.0103 -0.0429 c -0.0059 -0.0110 c 1.57 7.84* -0.045* 0.022 13:10-13:15 -0.0237 -0.0321 a -0.0090 -0.0271 -0.0431c -0.0179 c 1.34 5.55 0.004 -0.020* 9:45-9:50 -0.0365 b 0.0268 -0.0053 -0.0051 -0.0240 -0.0025 2.04* 13.6*** -0.035* -0.019 13:15-13:20 -0.0222 -0.0125 -0.0037 -0.0114 -0.0163 -0.0049 a 0.35 1.30 -0.011 -0.004 9:50-9:55 -0.0166 0.0115 -0.0183 -0.0019 -0.0246 -0.0070 0.85 7.01 -0.008 -0.018 13:20-13:25 0.0033 -0.0148 -0.0097 -0.0087 -0.0087 -0.0002 0.36 4.20 0.014 -0.001 9:55-10:00 -0.0021 0.0011 -0.0170 -0.0153 0.0001 -0.0103 0.37 2.04 0.006 0.008 13:25-13:30 0.0119 0.0028 -0.0006 -0.0012 -0.0131 0.0027 0.67 2.62 0.015 -0.016 10:00-10:05 -0.0008 0.0077 -0.0046 -0.0022 -0.0157 -0.0080 0.53 1.84 0.003 -0.016 13:30-13:35 0.0117 -0.0010 0.0065 0.0086 -0.0012 0.0056 0.26 1.26 0.009 -0.008 10:05-10:10 -0.0122 0.0081 0.0017 -0.0042 -0.0231 -0.0090 0.53 1.65 -0.008 -0.021 13:35-13:40 0.0098 -0.0112 -0.0006 0.0024 0.0299 a 0.0080 1.73 7.77 0.005 0.030** 10:10-10:15 -0.0085 -0.0039 -0.0241 -0.0074 -0.0028 -0.0062 0.42 0.62 0.001 0.008 13:40-13:45 0.0117 -0.0017 0.0086 0.0064 0.0217 0.0124 0.70 3.43 0.003 0.015 10:15-10:20 0.0136 -0.0231 -0.0074 -0.0079 -0.0171 -0.0114 0.94 2.53 0.028 -0.011 13:45-13:50 0.0054 0.0273 0.0036 0.0160 0.0251 0.0088 b 0.47 6.08 -0.013 0.012 10:20-10:25 -0.0078 -0.0158 0.0073 0.0075 -0.0220 -0.0071 0.90 4.24 -0.002 -0.02 13:50-13:55 0.0071 -0.0091 -0.0059 0.0002 0.0222 0.0050 0.96 2.59 0.005 0.024* 10:25-10:30 -0.0083 0.0060 0.0060 0.0044 -0.0213 -0.0024 0.87 3.93 -0.007 -0.023 13:55-14:00 0.0179 0.0248 0.0001 -0.0093 0.0017 0.0052 1.22 5.37 0.014 -0.007 10:30-10:35 0.0210 0.0245 0.0189 -0.0014 0.0016 0.0064 a 0.80 1.39 0.01 -0.014 14:00-14:05 0.0083 0.0314 -0.0172 0.0184 -0.0087 0.0029 2.24* 7.09 0.002 -0.019 10:35-10:40 -0.0069 0.0000 0.0099 0.0008 0.0037 0.0017 0.17 1.55 -0.01 0.003 14:05-14:10 -0.0010 0.0103 -0.0160 0.0093 0.0031 0.0030 0.63 3.19 -0.003 0.002 10:40-10:45 -0.0054 -0.0151 -0.0012 -0.0034 0.0116 -0.0032 0.57 4.60 -0.003 0.018 14:10-14:15 0.0183 -0.0059 0.0179 0.0060 -0.0114 0.0043 1.01 4.69 0.017 -0.02 10:45-10:50 -0.0032 -0.0091 -0.0213 -0.0255 -0.0005 -0.0088 0.80 2.56 0.011 0.014 14:15-14:20 0.0212 -0.0106 0.0025 -0.0011 0.0016 0.0032 0.81 4.03 0.023 -0.001 10:50-10:55 0.0158 -0.0109 0.0036 -0.0030 -0.0055 -0.0051 0.79 4.35 0.02 -0.007 14:20-14:25 -0.0138 -0.0209 -0.0133 0.0031 0.0174 -0.0034 1.10 2.31 -0.01 0.029* 10:55-11:00 0.0016 -0.0233 0.0200 -0.0030 -0.0103 -0.0068 1.96* 8.50* 0.006 -0.009 14:25-14:30 -0.0074 0.0039 0.0114 -0.0104 -0.0339 b -0.0044 1.64 6.07 0 -0.033** 11:00-11:05 -0.0263 -0.0272 0.0068 -0.0212 -0.0070 -0.0141 b 1.58 6.72 -0.014 0.01 14:30-14:35 -0.0093 0.0234 0.0141 0.0158 -0.0138 0.0052 1.44 1.87 -0.019 -0.025 11:05-11:10 -0.0154 -0.0024 -0.0176 -0.0095 -0.0037 -0.0135 0.56 2.31 -0.007 0.007 14:35-14:40 -0.0044 0.0030 -0.0227 0.0007 0.0298 a 0.0019 1.70 13.19** -0.007 0.036** 11:10-11:15 -0.0157 -0.0068 -0.0235 -0.0049 0.0034 -0.0072 0.61 3.37 -0.008 0.016 14:40-14:45 -0.0137 -0.0041 -0.0130 -0.0264 -0.0071 -0.0149 a 0.34 4.06 -0.001 0.007 11:15-11:20 -0.0250 -0.0144 -0.0106 -0.0082 -0.0135 -0.0107 a 0.36 1.75 -0.013 0.001 14:45-14:50 -0.0156 -0.0350 0.0071 -0.0236 -0.0296 a -0.0208 b 1.55 9.99** 0.005 -0.013 11:20-11:25 0.0001 -0.0064 -0.0088 -0.0054 0.0014 -0.0072 0.23 1.39 0.005 0.007 14:50-14:55 -0.0196 -0.0097 0.0092 -0.0090 -0.0134 -0.0078 0.58 2.72 -0.014 -0.006 11:25-11:30 0.0178 0.0093 -0.0016 0.0072 0.0320* 0.0125 a 1.11 4.91 0.006 0.024* 14:55-15:00 0.0860 c 0.0910 c 0.0799 c 0.0822 c 0.0760 c 0.1017 c 0.16 1.77 0.004 -0.009 Overnight 0.1179 c 0.0565 c 0.0172 -0.0035 0.1049 c 0.0655 c 0.64 1.32 0.074 0.058 All -0.0040 c 0.0021 -0.0028 -0.0013 -0.0011 -0.0014 1.95* 5.18 -0.006** 0.001 F-test 2 4.57*** 2.42*** 1.82 1.84*** 2.26 7.35*** KW-test 2 243.7*** 187.4*** 166.02 154.6*** 192.6*** 686.7*** Notes: (1) F-test 1 is the F statistic testing the equality of a given intraday return across all days of the week. (2) F-test 2 is the F statistic testing the equality of the intraday returns for a given weekday and all days. (3) KW-test 1 is the non-parametric Kruskal Wallis statistic testing the equality of a given intraday return across all days of the week. (4) KW-test 2 is the non-parametric Kruskal Wallis statistic testing the equality of the intraday returns for a given weekday and all days. (5) The * for 10% level of significance. The ** for 5% level of significance. The *** for 1% level of significance. (6) The a for 10% level of significance. The b for 5% level of significance. The c for 1% level of significance. 73

Table 3.5 Mean 5-minute return standard deviations of the Shanghai Composite Index, 2000 - 2002 Brown- Brown- Levene- Forsythe Levene- Forsythe No Time Mon Tue Wed Thu Fri Overall Test 1 1 No Time Mon Tue Wed Thu Fri Overall Test 1 1 1 9:30-9:35 0.331 0.347 0.305 0.460 0.351 0.363 0.56 0.527 25 13:00-13:05 0.157 0.175 0.191 0.253 0.179 0.194 0.36 0.342 2 9:35-9:40 0.262 0.301 0.235 0.264 0.252 0.266 0.78 0.729 26 13:05-13:10 0.075 0.091 0.083 0.097 0.106 0.091 1.00 0.986 3 9:40-9:45 0.183 0.231 0.204 0.182 0.198 0.201 0.77 0.575 27 13:10-13:15 0.081 0.097 0.110 0.133 0.103 0.106 1.00 0.918 4 9:45-9:50 0.145 0.144 0.147 0.145 0.173 0.151 0.26 0.276 28 13:15-13:20 0.084 0.090 0.116 0.099 0.072 0.093 0.87 0.723 5 9:50-9:55 0.134 0.171 0.138 0.154 0.154 0.151 0.88 0.895 29 13:20-13:25 0.083 0.092 0.092 0.125 0.089 0.097 1.32 1.368 6 9:55-10:00 0.141 0.131 0.128 0.146 0.155 0.141 0.22 0.216 30 13:25-13:30 0.093 0.123 0.116 0.117 0.093 0.109 0.85 0.86 7 10:00-10:05 0.127 0.128 0.126 0.119 0.127 0.125 0.37 0.367 31 13:30-13:35 0.087 0.121 0.118 0.135 0.119 0.117 1.31 1.206 8 10:05-10:10 0.136 0.156 0.184 0.132 0.154 0.153 0.80 0.752 32 13:35-13:40 0.094 0.109 0.142 0.108 0.110 0.114 0.80 0.779 9 10:10-10:15 0.136 0.158 0.134 0.111 0.132 0.135 0.94 0.993 33 13:40-13:45 0.114 0.121 0.107 0.111 0.118 0.114 0.07 0.074 10 10:15-10:20 0.158 0.127 0.144 0.149 0.168 0.150 0.78 0.767 34 13:45-13:50 0.090 0.133 0.135 0.166 0.136 0.134 1.81 1.78 11 10:20-10:25 0.137 0.122 0.138 0.165 0.124 0.138 0.85 0.846 35 13:50-13:55 0.094 0.145 0.116 0.124 0.134 0.124 1.24 1.224 12 10:25-10:30 0.119 0.127 0.132 0.117 0.149 0.129 0.27 0.246 36 13:55-14:00 0.112 0.110 0.123 0.140 0.144 0.126 0.83 0.781 13 10:30-10:35 0.136 0.153 0.105 0.157 0.125 0.136 0.66 0.643 37 14:00-14:05 0.094 0.132 0.144 0.131 0.115 0.125 1.52 1.355 14 10:35-10:40 0.131 0.138 0.113 0.160 0.141 0.137 0.36 0.356 38 14:05-14:10 0.114 0.114 0.156 0.149 0.142 0.136 0.94 0.94 15 10:40-10:45 0.124 0.118 0.111 0.137 0.162 0.131 0.82 0.859 39 14:10-14:15 0.098 0.113 0.143 0.159 0.154 0.136 1.90 1.791 16 10:45-10:50 0.134 0.106 0.118 0.156 0.126 0.130 0.91 0.806 40 14:15-14:20 0.103 0.127 0.141 0.169 0.161 0.142 2.64** 2.62** 17 10:50-10:55 0.115 0.109 0.123 0.111 0.121 0.116 0.04 0.042 41 14:20-14:25 0.115 0.162 0.148 0.133 0.158 0.144 0.62 0.575 18 10:55-11:00 0.093 0.113 0.109 0.127 0.144 0.119 1.46 1.483 42 14:25-14:30 0.120 0.146 0.131 0.126 0.156 0.137 1.62 1.595 19 11:00-11:05 0.122 0.101 0.122 0.133 0.106 0.117 0.30 0.295 43 14:30-14:35 0.130 0.153 0.141 0.154 0.170 0.150 1.22 1.139 20 11:05-11:10 0.110 0.116 0.119 0.092 0.115 0.111 0.48 0.419 44 14:35-14:40 0.132 0.163 0.159 0.162 0.151 0.154 0.78 0.749 21 11:10-11:15 0.104 0.143 0.110 0.098 0.097 0.112 1.78 1.726 45 14:40-14:45 0.145 0.204 0.143 0.151 0.140 0.158 1.45 1.354 22 11:15-11:20 0.105 0.109 0.088 0.094 0.104 0.100 1.23 1.14 46 14:45-14:50 0.126 0.180 0.137 0.144 0.145 0.148 0.17 0.139 23 11:20-11:25 0.080 0.109 0.095 0.107 0.088 0.096 0.21 0.22 47 14:50-14:55 0.102 0.154 0.131 0.126 0.121 0.128 0.49 0.493 24 11:25-11:30 0.078 0.088 0.132 0.171 0.122 0.123 1.27 1.128 48 14:55-15:00 0.097 0.121 0.120 0.115 0.125 0.116 1.11 1.13 Overnight 0.864 0.769 0.411 0.334 0.573 0.623 2.05* 1.576 All 0.132 0.149 0.141 0.158 0.148 0.146 5.46*** 5.64*** Levene-test Brown- 2 15.1*** 9.32*** 7.10*** 7.29*** 9.43*** 43.8*** Forsythe 2 14.3*** 8.90*** 7.01*** 7.20*** 9.18*** 43.0*** Notes: *: 10% level of significance. **: 5% level of significance. ***: 1% level of significance.

74

Table 3.6 Mean 5-minute return standard deviations of the Shenzhen Composite Index, 2000 - 2002 Brown- Brown- Levene- Forsythe Levene- Forsythe No Time Mon Tue Wed Thu Fri Overall Test 1 1 No Time Mon Tue Wed Thu Fri Overall Test 1 1 1 9:30-9:35 0.346 0.333 0.289 0.473 0.366 0.368 1.12 1.06 25 13:00-13:05 0.126 0.161 0.244 0.188 0.193 0.187 2.23 2.161 2 9:35-9:40 0.384 0.351 0.292 0.322 0.338 0.342 0.66 0.598 26 13:05-13:10 0.101 0.143 0.122 0.150 0.128 0.130 0.91 0.972 3 9:40-9:45 0.241 0.261 0.256 0.244 0.251 0.251 0.24 0.199 27 13:10-13:15 0.109 0.127 0.124 0.126 0.155 0.129 1.52 1.384 4 9:45-9:50 0.231 0.170 0.225 0.172 0.197 0.201 1.39 1.525 28 13:15-13:20 0.124 0.170 0.152 0.122 0.114 0.138 0.26 0.269 5 9:50-9:55 0.201 0.195 0.170 0.181 0.199 0.189 0.36 0.364 29 13:20-13:25 0.109 0.132 0.141 0.168 0.110 0.134 0.89 0.882 6 9:55-10:00 0.164 0.193 0.153 0.167 0.184 0.173 0.84 0.841 30 13:25-13:30 0.113 0.146 0.146 0.127 0.119 0.131 0.47 0.513 7 10:00-10:05 0.144 0.160 0.136 0.121 0.129 0.138 0.67 0.656 31 13:30-13:35 0.117 0.131 0.145 0.142 0.137 0.135 0.57 0.504 8 10:05-10:10 0.170 0.199 0.231 0.179 0.212 0.199 0.70 0.7 32 13:35-13:40 0.110 0.154 0.174 0.118 0.133 0.140 0.93 0.938 9 10:10-10:15 0.154 0.185 0.163 0.148 0.141 0.159 0.97 0.96 33 13:40-13:45 0.123 0.113 0.126 0.134 0.115 0.122 0.37 0.363 10 10:15-10:20 0.181 0.142 0.182 0.170 0.181 0.172 0.30 0.297 34 13:45-13:50 0.133 0.214 0.167 0.185 0.234 0.190 1.62 1.552 11 10:20-10:25 0.174 0.146 0.150 0.204 0.165 0.169 0.59 0.594 35 13:50-13:55 0.125 0.191 0.154 0.135 0.147 0.152 0.86 0.818 12 10:25-10:30 0.143 0.147 0.167 0.145 0.168 0.154 0.62 0.643 36 13:55-14:00 0.131 0.134 0.138 0.169 0.176 0.151 1.42 1.426 13 10:30-10:35 0.169 0.186 0.131 0.153 0.152 0.159 0.55 0.495 37 14:00-14:05 0.115 0.145 0.172 0.167 0.182 0.159 1.33 1.212 14 10:35-10:40 0.135 0.168 0.163 0.222 0.183 0.176 0.49 0.49 38 14:05-14:10 0.124 0.159 0.177 0.168 0.167 0.160 0.42 0.417 15 10:40-10:45 0.153 0.141 0.142 0.158 0.165 0.152 0.14 0.097 39 14:10-14:15 0.131 0.132 0.167 0.190 0.178 0.161 0.89 0.853 16 10:45-10:50 0.148 0.129 0.153 0.166 0.138 0.147 1.03 0.993 40 14:15-14:20 0.114 0.137 0.156 0.160 0.191 0.154 2.52** 2.613** 17 10:50-10:55 0.143 0.119 0.144 0.143 0.138 0.138 0.66 0.692 41 14:20-14:25 0.142 0.217 0.167 0.161 0.193 0.178 1.55 1.535 18 10:55-11:00 0.110 0.134 0.134 0.141 0.158 0.137 1.34 1.382 42 14:25-14:30 0.151 0.161 0.150 0.155 0.188 0.162 2.03* 1.874 19 11:00-11:05 0.130 0.120 0.137 0.172 0.131 0.139 1.25 1.297 43 14:30-14:35 0.129 0.159 0.176 0.161 0.192 0.165 1.65 1.594 20 11:05-11:10 0.106 0.115 0.096 0.109 0.117 0.109 0.69 0.714 44 14:35-14:40 0.148 0.186 0.172 0.177 0.179 0.174 0.68 0.684 21 11:10-11:15 0.179 0.186 0.145 0.134 0.142 0.158 1.04 1.039 45 14:40-14:45 0.157 0.221 0.185 0.153 0.153 0.175 0.78 0.741 22 11:15-11:20 0.160 0.123 0.098 0.121 0.125 0.127 0.88 0.85 46 14:45-14:50 0.128 0.227 0.137 0.140 0.136 0.158 1.07 0.671 23 11:20-11:25 0.103 0.132 0.095 0.106 0.107 0.109 0.25 0.282 47 14:50-14:55 0.164 0.183 0.174 0.154 0.168 0.169 0.17 0.167 24 11:25-11:30 0.103 0.113 0.159 0.186 0.145 0.145 1.07 1.078 48 14:55-15:00 0.146 0.212 0.142 0.165 0.178 0.170 1.36 1.337 Overnight 0.908 1.100 0.506 0.660 0.645 0.173 0.99 0.852 All 0.162 0.177 0.169 0.178 0.177 0.173 4.27*** 4.359*** Levene-test Brown- 2 11.7*** 6.13*** 5.95*** 6.64*** 6.89*** 33.7*** Forsythe 2 11.1*** 6.02*** 5.85*** 6.61*** 6.75*** 33.6*** Notes: *: 10% level of significance. **: 5% level of significance. ***: 1% level of significance.

75

Table 3.7 Descriptive statistics for the returns of the trading and non-trading period of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002 (in percentages)

Shanghai Composite Index Shenzhen Component Index

Variance Variance Time Mean Variance Ratio Mean Variance Ratio

1 Overnight 0.0655 0.388 0.733 0.0586 0.627 1.168 2 Morning -0.1407 0.604 1.142 -0.1340 0.734 1.367 Lunch 3 Break 0.0083 0.006 0.011 0.0198 0.010 0.019 4 Afternoon 0.0705 0.529 1.000 0.0255 0.537 1.000 (1)+(2)+ 5 (3)+(4) 1.527 1.908

6 Open-close -0.0665 1.703 3.219 -0.0866 1.858 3.460 7 Close-close -0.0049 2.056 3.887 -0.0331 2.277 4.240 Note: The computation for these return series are for morning: 09:30 – 11:30, for lunch break: 11:30 – 13:00 (afternoon opening price), for afternoon: 13:00 – 15:00 and for overnight: 15:00 – 09:30 next day.

76 Table 3.8 Interday 24-hour returns, return volatility and variance ratio test, 2000 - 2002 Shanghai Composite Index Shenzhen Component Index Shanghai Composite Index Shenzhen Component Index Variance Variance Variance Variance No Time Mean Variance Ratio Test Mean Variance Ratio Test No Time Mean Variance Ratio Test Mean Variance Ratio Test 0 Auction Call -0.0002 2.535 1.233*** -0.0279 2.983 1.310*** 1 09:35-09:35 -0.0003 2.517 1.224*** -0.0281 3.122 1.371*** 25 13:05-13:05 -0.0025 2.189 1.065 -0.0304 2.555 1.122 2 09:40-09:40 0.0002 1.997 0.971 -0.0276 2.547 1.118 26 13:10-13:10 -0.0030 2.190 1.065 -0.0313 2.657 1.166** 3 09:45-09:45 0.0006 1.824 0.887 -0.0270 2.286 1.004 27 13:15-13:15 -0.0031 2.080 1.012 -0.0318 2.534 1.113 4 09:50-09:50 0.0006 1.879 0.914 -0.0270 2.312 1.015 28 13:20-13:20 -0.0030 1.995 0.97 -0.0318 2.366 1.039 5 09:55-09:55 0.0006 1.908 0.928 -0.0271 2.365 1.038 29 13:25-13:25 -0.0031 1.991 0.968 -0.0314 2.339 1.027 6 10:00-10:00 0.0006 1.926 0.937 -0.0271 2.390 1.049 30 13:30-13:30 -0.0035 2.024 0.985 -0.0314 2.381 1.045 7 10:05-10:05 0.0004 1.961 0.954 -0.0273 2.423 1.064 31 13:35-13:35 -0.0038 2.024 0.984 -0.0318 2.372 1.041 8 10:10-10:10 0.0000 1.971 0.959 -0.0281 2.445 1.074 32 13:40-13:40 -0.0040 2.010 0.978 -0.0321 2.329 1.022 9 10:15-10:15 -0.0003 2.015 0.98 -0.0285 2.499 1.097 33 13:45-13:45 -0.0038 2.004 0.975 -0.0321 2.336 1.026 10 10:20-10:20 -0.0003 2.046 0.995 -0.0286 2.502 1.098 34 13:50-13:50 -0.0038 1.980 0.963 -0.0319 2.342 1.028 11 10:25-10:25 -0.0002 2.051 0.998 -0.0283 2.454 1.077 35 13:55-13:55 -0.0038 1.956 0.951 -0.0319 2.337 1.026 12 10:30-10:30 -0.0002 2.014 0.98 -0.0282 2.455 1.078 36 14:00-14:00 -0.0041 1.915 0.932 -0.0322 2.224 0.976 13 10:35-10:35 -0.0004 1.996 0.971 -0.0284 2.435 1.069 37 14:05-14:05 -0.0040 1.947 0.947 -0.0324 2.242 0.984 14 10:40-10:40 -0.0005 1.971 0.959 -0.0285 2.405 1.056 38 14:10-14:10 -0.0038 1.991 0.968 -0.0322 2.295 1.008 15 10:45-10:45 -0.0009 1.971 0.959 -0.0290 2.397 1.052 39 14:15-14:15 -0.0037 1.977 0.962 -0.0320 2.294 1.007 16 10:50-10:50 -0.0012 1.922 0.935 -0.0295 2.329 1.022 40 14:20-14:20 -0.0039 1.935 0.942 -0.0319 2.243 0.985 17 10:55-11:55 -0.0014 1.875 0.912 -0.0296 2.270 0.997 41 14:25-14:25 -0.0044 1.930 0.939 -0.0323 2.182 0.958 18 11:00-11:00 -0.0012 1.846 0.898 -0.0296 2.237 0.982 42 14:30-14:30 -0.0047 1.926 0.937 -0.0324 2.174 0.955 19 11:05-11:05 -0.0011 1.832 0.891 -0.0291 2.227 0.978 43 14:35-14:35 -0.0047 1.924 0.936 -0.0327 2.183 0.958 20 11:10-11:10 -0.0011 1.872 0.911 -0.0290 2.245 0.985 44 14:40-14:40 -0.0053 1.944 0.946 -0.0340 2.212 0.971 21 11:15-11:15 -0.0012 1.884 0.916 -0.0291 2.331 1.023 45 14:45-14:45 -0.0057 1.982 0.964 -0.0345 2.235 0.981 22 11:20-11:20 -0.0013 1.865 0.907 -0.0292 2.287 1.004 46 14:50-14:50 -0.0056 2.000 0.973 -0.0344 2.237 0.982 23 11:25-11:25 -0.0015 1.867 0.909 -0.0296 2.273 0.998 47 14:55-14:55 -0.0056 2.024 0.984 -0.0340 2.271 0.997 11:30-11:30 15:00-15:00 24 (MorningClose) -0.0018 1.916 0.932 -0.0299 2.323 1.020 48 (Close-Close) -0.0049 2.056 1 -0.0331 2.278 1 Notes: **: 5% level of significance. ***: 1% level of significance.

77

Figure 3.1 Time plots of intraday 5-minute index closes and log returns (in percentages) of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002. The left plots are for the Shanghai Composite Index and the right plots are for the Shenzhen Component Index. The upper plots are for index closes and the bottom plots are for the index returns.

3000 6000

2500 2000 2001 2002 5000

2000 4000

1500 3000

1000 2000 2000 2001 2002 500 1000

0 0 10000 20000 30000 10000 20000 30000

5-minute Intraday Shanghai Index 5-minute Intraday Shenzhen Index

3 3

2001 2001 2 2000 20022 2000 2002

1 1

0 0

-1 -1

-2 -2

-3 -3 10000 20000 30000 10000 20000 30000

Shanghai 5-minute Intraday Return Shenzhen 5-minute Intraday Return

78

Figure 3.2 Intraday mean 5-minute returns and cumulative mean 5-minute returns (in percentages) of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002.

.12 .10

.08 .08 .06

.04 .04

.02 .00 .00 -.04 -.02

-.08 -.04 0 4 8 12 16 20 24 28 32 36 40 44 48 0 4 8 12 16 20 24 28 32 36 40 44 48

Mean 5-minute intraday return of Shanghai Mean 5-minute intraday return of Composite Index Shenzhen Component Index

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12

-.16 -.16

-.20 -.20 0 4 8 12 16 20 24 28 32 36 40 44 48 0 4 8 12 16 20 24 28 32 36 40 44 48

Cumulative 5-minute intraday return of Cumulative 5-minute intraday return of Shanghai Composite Index Shenzhen Component Index

79

Figure 3.3 Mean 5-minute returns (in percentages) by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002

.15 .15

.10 .10

.05 .05

.00 .00

-.05 -.05

-.10 -.10

-.15 -.15 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Monday Mean 5-minute Return of Shang hai Index Monday Mean 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Tuesday Mean 5-minute Return of Shang hai Index Tuesday Mean 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Wednesday Mean 5-minute Return of Shang hai Index Wednesday Mean 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Thursday Mean 5-minute Return of Shang hai Index Thursday Mean 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Friday Mean 5-minute Return of Shanghai Index Friday Mean 5-minute Return of Shenzhen Index

80

Figure 3.4 Intraday cumulative mean 5-minute returns (in percentages) by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 – 2002.

.3 .3

.2 .2

.1 .1

.0 .0

-.1 -.1

-.2 -.2

-.3 -.3

-.4 -.4

-.5 -.5 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Monday Cumulative 5-minute Return of Shanghai Index Monday Cumulative 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Tuesday Cumulative 5-minute Return of Shanghai Index Tuesday Cumulative 5-minute Return of Shenzhen Index

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0

-.1 -.1

-.2 -.2 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Wednesday Cumulative 5-minute Return of Shang hai Index Wednesday Cumulative 5-minute Return of Shenzhen Index

.15 .15

.10 .10

.05 .05

.00 .00

-.05 -.05

-.10 -.10

-.15 -.15

-.20 -.20 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Thursday Cumulative 5-minute Return of Shanghai Index Thursday Cumulative 5-minute Return of Shenzhen Index

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12

-.16 -.16 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Friday Cumulative 5-minute Return of Shanghai Index Friday Cumulative 5-minute Return of Shenzhen Index

81

Figure 3.5 Intraday 5-minute return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002

.40 .40

.35 .35

.30 .30 .25 .25 .20 .20 .15

.10 .15

.05 .10 0 4 8 12 16 20 24 28 32 36 40 44 48 0 4 8 12 16 20 24 28 32 36 40 44 48

Volatility of 5-minute intraday Volatility of 5-minute intraday return of Shanghai Composite Index return of Shenzhen Component Index

82

Figure 3.6 Intraday 5-minute return volatility based on standard deviations by weekdays of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Monday 5-minute Return Volatility of Shang hai Index Monday 5-minute Return Volatility of Shenzhen Index

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Tuesday 5-minute Return Volatility of Shanghai Index Tuesday 5-minute Return Volatility of Shenzhen Index

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Wednesday 5-minute Return Volatility of Shang hai Index Wednesday 5-minute Return Volatility of Shenzhen Index

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Thursday 5-minute Return Volatility of Shanghai Index Thursday 5-minute Return Volatility of Shenzhen Index

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Friday 5-minute Return Volatility of Shang hai Index Friday 5-minute Return Volatility of Shenzhen Index

83

Figure 3.7 Intraday mean 5-minute returns of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002 (in percentages)

.15 .15

.10 .10

.05 .05

.00 .00

-.05 -.05

-.10 -.10

-.15 -.15 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Mean 5-minute Return of Shanghai Index in 2000 Mean 5-minute Return of Shenzhen Index in 2000

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Mean 5-minute Return of Shanghai Index in 2001 Mean 5-minute Return of Shenzhen Index in 2001

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Mean 5-minute Return of Shanghai Index in 2002 Mean 5-minute Return of Shenzhen Index in 2002

84

Figure 3.8 Intraday cumulative mean 5-minute returns of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002 (in percentages)

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0

-.1 -.1

-.2 -.2 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Cumulative 5-minute Return of Shanghai Index in 2000 Cumulative 5-minute Return of Shenzhen Index in 2000

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0

-.1 -.1

-.2 -.2

-.3 -.3 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Cumulative 5-minute Return of Shanghai Index in 2001 Cumulative 5-minute Return of Shenzhen Index in 2001

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0

-.1 -.1

-.2 -.2 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

Cumulative 5-minute Return of Shanghai Index in 2002 Cumulative 5-minute Return of Shenzhen Index in 2002

85

Figure 3.9 Intraday mean 5-minute return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index in each year 2000, 2001 and 2002

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

5-minute Return Volatility of Shanghai Index in 2000 5-minute Return Volatility of Shenzhen Index in 2000

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

5-minute Return Volatility of Shanghai Index in 2001 5-minute Return Volatility of Shenzhen Index in 2001

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

5-minute Return Volatility of Shanghai Index in 2002 5-minute Return Volatility of Shenzhen Index in 2002

86

Figure 3.10 Intraday mean 1-minute returns, cumulative mean returns (in percentages) and return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component index, 2000 - 2002

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08 25 50 75 100 125 150 175 200 225 25 50 75 100 125 150 175 200 225

Mean 1-minute Intraday Return of Shanghai Index Mean 1-minute Intraday Return of Shenzhen Index

.10 .15

.05 .10 .05 .00 .00 -.05 -.05 -.10 -.10 -.15 -.15

-.20 -.20

-.25 -.25 25 50 75 100 125 150 175 200 225 25 50 75 100 125 150 175 200 225

Cumulative 1-minute Intraday Return of Shanghai Index Cumulative 1-minute Intraday Return of Shenzhen Index

.5 .7

.6 .4 .5

.3 .4

.3 .2

.2 .1 .1

.0 .0 25 50 75 100 125 150 175 200 225 25 50 75 100 125 150 175 200 225

1-minute Intraday Return Volatility of Shanghai Index 1-minute Intraday Return Volatility of Shenzhen Index

87

Figure 3.11 Intraday mean 10-minute returns, cumulative mean returns (in percentages) and return volatility based on standard deviations of the Shanghai Composite Index and the Shenzhen Component Index, 2000 - 2002

.12 .12

.08 .08

.04 .04

.00 .00

-.04 -.04

-.08 -.08 2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24

Mean 10-minute Intraday Return of Shanghai Index Mean 10-minute Intraday Return of Shenzhen Index

.00 .00

-.04 -.04

-.08 -.08

-.12 -.12

-.16 -.16

-.20 -.20

-.24 -.24 2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24

Cumulative 10-minute Intraday Return of Shanghai Index Cumulative 10-minute Intraday Return of Shenzhen Index

.5 .5

.4 .4

.3 .3

.2 .2

.1 .1

.0 .0 2 4 6 8 10 12 14 16 18 20 22 24 2 4 6 8 10 12 14 16 18 20 22 24

10-minute Intraday Return Volatility of Shanghai Index 10-minute Intraday Return Volatility of Shenzhen Index

88

Chapter 4

Inter-Temporal Behaviour and the Determinants of

Bid/Ask Spread and Depth

4.1 Introduction

The previous chapter examines the behaviours of intraday returns and return volatility using 5-minute high-frequency data over a three-year period on the Chinese stock market. Analyses are based on the portfolio indices of the Shanghai and the

Shenzhen stock exchanges. Several different tests were used to investigate if there are systematic patterns according to the efficient market hypothesis. This chapter examines the inter-temporal behaviours and the determinants of liquidity proxies, bid/ask spread and depth, on the limit order-driven market. Analysis is based on a set of sixty individual stocks from the Shanghai Stock Exchange over the three-year period with about five million observations. The chapter will investigate the systematic patterns in both bid/ask spread and depth as market liquidity proxies of the Shanghai stock market. While the trading price, asymmetric information and risk may be the determinants of the bid/ask spreads and the depths, the regression model

89 is used to further test and verify the statistical analysis. The chapter also discusses the possible reasons from the results.

The purpose of this chapter is to provide evidence on the inter-temporal behaviour of bid/ask spreads and depths using intraday data from the Shanghai Stock

Exchange and to contribute to the existing literature in the following ways: First, in contrast to the large quantity of empirical studies based on the US market and other countries, no studies so far have used high-frequency data to analyze China’s emerging market. Second, this chapter provides the unique empirical findings to date on the systematic patterns of interday and intraday bid/ask spreads and depths for companies traded on the SHSE. Third, if the results on behavior of liquidity are similar to those of the other markets, it will provide additional evidence on the benefits of the order-driven market. If there is no similarity on liquidity, an analysis of the major differences may enhance the understanding of the Chinese stock market.

The sixty individual stocks from the Shanghai Stock Exchange are used to examine and analyse the behaviour and affecting factors of intraday and interday liquidity and the relationship between liquidity, trading price, risk, and information in the Chinese stock market.

There exists a substantial amount of microstructure literature, both empirical and theoretical, on liquidity. In Chan et al. (1995) the bid/ask spread in the stock market is not constant, but varies through time, and is affected by a number of variables. Liquidity is generally measured by a stock’s trading frequency, bid/ask spread and by the depth of the market for that stock, though liquidity is multi- dimensional in its nature. Liquidity affects the formation of stock price, market trading procedures and price variation. Intraday liquidity patterns contain different information about price variation at different times in the trading day. Large amounts

90 of literature reported the U-shaped or L-shaped (or reverse J-shaped) patterns of the bid/ask spreads, and inverse U-shaped or L-shaped of depths in the stock markets (cf.

Wood, McInish and Ord, 1985; Brock and Kleidon, 1992; Lehman and Modest,

1994; Chan et al., 1995; Madhavan, Richardson and Roomans, 1997; Chan, 2000 and

McInish and Ness, 2002).

Empirical analyses of this chapter are based on 5-minute high-frequency data over a three-year period from 2000 to 2002. The chapter first characterizes the intraday bid/ask spread and depth measured as the market liquidity proxies from sixty most actively traded stocks. Both intraday and intraweek patterns are analysed.

The behaviour of the inter-temporal relationship between intraday bid/ask spread and depth is also examined while they are considered as the market liquidity proxies.

Second, based on the approaches of McInish and Wood (1992) and Brockman and

Chung (1998), the hypotheses are made and dummy variables are used to verify the intraday and interday patterns, as well as the liquidity determinants of bid/ask spread and depth. In the model, intraday price, return variance as the risk variable and trading volume as the variable of information flow, are considered as the determinants in the regression model for both bid/ask spread and depth proxies.

Third, the correlations between spread and depth for each 5-minute interval are investigated by statistical analysis. Finally, based on the theoretical points and implications, the chapter discusses the inter-temporal relationship and patterns between intraday bid/ask spread and depth.

Firstly, the studies reveal that the spread in the limit-order market of

Shanghai exhibits an L-shaped intraday pattern while the depth displays a reverse L- shape. The spread is the largest at the market call opening and the second largest at the open of the continuous trading and then declines almost monotonically

91 throughout the trading day until the market close. The depth, measured as the dollar amount of bid and ask orders submitted at the best bid and best offer prices, on the other hand, shows the opposite pattern to the bid/ask spread. It is the lowest at the call opening and the second lowest at the open of the continuous trading and then rises monotonically until the close. A similar L-shaped intraweek pattern in the spread and a reverse V-shaped intraweek pattern in the depth are identified. The bid/ask spread is lowest on Mondays and highest on Wednesdays. The depth is lowest on Fridays and highest on Wednesdays. The findings show that interday and intraday bid/ask spreads display an L-shaped pattern similar to that documented on the multi-dealer systems (Chan et al., 1995) and dissimilar to that documented on the specialist systems. These findings also support Foster and Viswanathan’s (1990) prediction of the asymmetric information theory of intraday spread patterns and interday variations.

Secondly, the inter-temporal liquidity proxies, spread and depth, associated with the transition from non-trading to trading periods and from trading to non- trading periods, still reveal strong variations even after controlling for price level, volatility and trading volume. These findings, combined with spread and depth results, tell a consistent story with respect to two dimensions of the measure of liquidity. The beginning of trading coincides with a period of acute illiquidity due to the long trading halt. As trading activity continues over the session, liquidity rises steadily until the market close. An important implication is that spread and depth patterns combine in such a way as to magnify inter-temporal variations in liquidity.

Analyzing both spreads and depths independently would lead to understating the strength and persistence of systematic liquidity changes.

92 Thirdly, the striking L-shaped pattern of the bid/ask spreads has a positive relationship with the volatility, and a negative relationship with stock prices, while the striking reversed L-shaped pattern of the depths has a negative relationship with stock prices and the volatility but a positive relationship with trading volumes. The findings show that the stock price, risk and information asymmetries are significant determinants of the bid/ask spreads and depths.

Finally, there is a negative relationship between the spreads and the depths at the opening of the trading day. Overall, my results suggest that the determinants of information asymmetries, through time and across traders, play a key role in generating observed liquidity variations.

The chapter is organized as follows. Section 4.2 presents the background and literature review on liquidity theories. Section 4.3 illustrates the data and the methodology. Section 4.4 performs empirical analyses on the bid/ask spread and depth, as well as the regression approaches for the robustness of the results. Section

4.4 also provides the theoretical explanation on liquidity patterns and the inter- temporal relationship between the bid/ask spread and depth. In Section 4.5, the summary and conclusion are made.

4.2 The Background and Literature Review

There are two types of stock markets: the price (quote) driven market or the order-driven market. The price driven market is characterized by the presence of a market maker (specialist), who acts as an intermediary between buyers and the sellers. The trading system under the quoted driven market is known as the specialist

93 system. The system of trade used in the U.S. market such as the NYSE is an example of the specialist system. Under this system, the market maker provides liquidity; that is, stand by to buy or sell an asset at any time, regardless of the quantity of shares.

The market maker is therefore required to maintain an inventory of stocks, which entails significant inventory risk. For example, the spread is affected by information uncertainty (Hasbrouck, 1988). Investors demand liquidity through the submission of market orders that are subsequently matched against the market makers' bid or ask prices (and depths). To ensure liquidity and fair prices, the dealers have to manage the asset inventories to the best of their ability. In the case of inventory maladjustments, they are obliged to sell or buy an asset from other dealers on the market. Such inter-dealer trades account for a large proportion of market transactions.

Market liquidity means the ability to buy or sell significant quantities of a security quickly, anonymously, and with little price impact. In return for providing liquidity, market makers are granted monopoly rights by the stock exchange to post different prices for purchases and sales of a security. They buy at the bid price Pb and sell at a higher ask price Pa thus providing liquidity (for investors, Pb is the sale price and Pa is the purchase price). The difference Pa–Pb is called the bid/ask spread, which is the primary source of compensation to market makers for providing liquidity while the depth, which can be measured as the dollar amount of bid and ask orders submitted at the best bid and best offer prices, is determined by the willingness of investors to provide immediacy through the submission of limit orders.

In contrast to the quote-driven market, the order-driven market has no dealer intermediation. Prices and quantities are set altogether. In order-driven systems,

94 public orders provide liquidity to the market and establish the bid/ask spread and depth. The orders submitted by traders are directly entered into an order book monitored by a computerized system and wait for execution. All transactions are visible to all market participants. The order book registers all movements that reflect changes in the list of queued orders. Trades occur whenever orders are matched through an electronic medium according to the price and timing priority criteria.

Thus, trades result directly from transactions concluded automatically between investors. These explicit trading rules can be imposed and monitored to minimize transaction costs and also let traders track the market price which reflects asymmetric information. There is no obligation on the part of any market participant to submit such orders.

Order-driven trading mechanisms allow for two different procedures of order matching: call auction and continuous matching (trading). A call auction takes orders batched together and takes place at predetermined times of day. In the auction, submitted buy and sell orders form demand and supply schedules and the intersection of these curves determines the market clearing price at which all trades are executed.

Continuous matching refers to ongoing trading throughout the trading time in a trading day. In principle, a continuous trading mechanism provides immediate execution of trades. When liquidity is low, immediate execution of a trade may be at high cost due to wider bid/ask spreads and low depths. An average intertrade duration is typically much larger on a call auction market than on a continuous trading market.

The large amount of research on bid/ask spread, depth and its components has been documented in both quoted-driven markets and order-driven markets. Demsetz

(1968) is the earliest study on spread research but Wood, McInish and Ord (1985)

95 report the earliest evidence of inter-temporal liquidity patterns based on the US’s quote-driven or hybrid quote-driven equity markets. Using NYSE stocks in the quoted-driven market, McInish and Wood (1992) identify factors that determine the inter-stock differences in bid/ask spread and show that spreads exhibit a distinct pattern over the trading day. They further show that differences in bid/ask spreads over the trading day can be explained in terms of four classes of determinants, namely are activity, risk, information and competition. They not only find that the spread is inversely related to the number of transactions in a given time period, number of shares per trade and competition from regional exchange, but also find that the bid/ask spreads are directly related to cross-section risk, time-series risk during the trading day, and unusually large or small trades. Moreover, McInish and

Wood (1992) find that both the proportional bid/ask spreads and the coefficients of the time-of-day dummy variables exhibit a reverse J-shaped pattern over the trading.

Brock and Kleidon (1992) find a U-shaped intraday spread on the NYSE, which was attributed to the market power of the specialists. In their model, a specialist can observe the order imbalances and has information on who is trading before deciding on the price of trade; thus, he charges higher prices to earn monopoly profits at the beginning and at the end of the day. However, as the transaction demand is lower and less elastic in the rest of the day he cannot make profits. Lee et al. (1993) perform an intraday analysis on the spreads, depths and the impact of earnings information but they do not use control variables or statistical testing since inter-temporal depth variation is not the focus of their study. They find a negative association between spread and depth on the NYSE where wide spreads are associated with small depths and narrow spreads are associated with large depths.

96 Chan et al. (1993) find that the daily pattern of bid/ask spread for the

NASDAQ is relatively stable throughout the trading day but narrows near the end of the day. They attribute the patterns to the price discovery process, where initially dealers avoid committing to large trades, but as the day progresses, they become more confident about the equilibrium values.

Chan et al. (1995) find that the results from multi-dealer markets have revealed a distinctively different pattern with relatively wide spreads only at the opening. Bid/Ask spreads do not widen toward the close of trading in these dealer markets but tend to narrow somewhat (L-shaped). Most of the intraday studies rely solely on bid/ask spreads as the measure of liquidity and show that specialists’ spreads follow the same pattern as that of returns, volume and volatility: a U-shaped pattern with the highest spreads at the open and close of trading. Recently, Chung et al. (1999) report a U-shaped pattern of spreads on 144 stocks on the NYSE, which largely reflects the intraday variation in spreads established by limit-order traders.

In the emerging markets, Brockman and Chung (1998) examine interday and intraday liquidity patterns in the Hong Kong equity market, which is a purely limit order-driven mechanism. The market making system of the Stock Exchange of Hong

Kong is highly transparent and operates with minimal third party intervention. They find that the patterns of bid/ask spreads on the Stock Exchange of Hong Kong also follow a U-shaped pattern, which is similar to those of specialist systems reported in the NYSE, the AMEX, and the Tokyo Stock Exchange markets but dissimilar to that reported for the NASDAQ and CBOE markets. Brockman and Chung (1999) investigated inter-temporal and cross-sectional depth patterns on the Stock Exchange of Hong Kong. Based on over six million observations, they report an inverted U- shaped pattern that mirrors the commonly reported U-shaped spread pattern.

97 Ahn and Cheung (1999) examine the temporal behaviour of the spread and depth using common stocks listed on the Stock Exchange of Hong Kong. They provide further evidence that there are U-shaped intraday and intraweek patterns in the spread and reverse U-shaped patterns in the depth. They suggest that the negative association between spread and depth in Hong Kong’s market implies that limit order traders actively manage both price and quantity dimensions of liquidity by adjusting the spread and depth.

All these findings provide further evidence that stylised spread and depth patterns should not be solely attributed to specialist market making activities.

Bid/Ask spreads are compensations to market makers (dealers) for providing liquidity. Dealers quote different prices for the bid and the ask to cover trading costs.

Three microstructure models that deal with order arrival and quote revision have been suggested to explain the intraday liquidity patterns in the US market: order-processing, asymmetric information and inventory holding costs. Chan et al.

(1995) provide the detailed descriptions of these three theoretical models. The order- processing theory suggests that dealers incur substantial costs in providing immediacy. The asymmetric information theory is based on the existence of adverse selection and suggests that changes in the bid/ask spread reflect changes in the level of asymmetric information.

The inventory theory suggests that the trading process is a matching problem in which the market maker, facing an unbalanced risk, uses the price to balance supply and demand across time. Market makers achieve inventory control by shifting the quotes (bid and ask) to elicit the imbalance of buy and sell orders. The existence of the liquidity patterns on different markets means that specialist activity alone in the US market cannot account for the widening of spreads at the open and close of

98 trading. The key factors in the order-driven market are the asymmetric information and inventory holding costs because there is no market maker. The overall findings of intraday liquidity patterns are not consistent with the EMH, which assumes that stock prices incorporate all information so that changes in prices are only due to news or unanticipated events. In addition, unanticipated information is incorporated instantaneously.

Interday theories are based on the strategic interaction of informed and uninformed traders such as discretionary liquidity traders and noise traders. In the

Chapter 3, I discuss the microstructure model developed by Kyle (1985), who suggests that traders can be classified into three classes: private information traders, random liquidity traders, and market makers. Admati and Pfleiderer (1988) and

Foster and Viswanathan (1990) add the fourth type of traders to Kyle’s model: discretionary liquidity traders. They take into account the impact and costs of their trade, and choose the size or time of the trade. They suggest that public and private information, and trading noise, would be the reasons causing a systematic pattern in return volatilities, and leads traders to converge their trading to the period of the open and close of the market. The theories provide hypothesized relationships among such variables as volume, volatility, and trading costs. Admati and Pfleiderer (1988) propose that trading costs are lowest in periods when the trading volume is highest because market makers compete to lower spreads during such periods of concentrated liquidity. Foster and Viswanathan (1990) predict that the adverse- selection component of the bid/ask spread will be at its widest on Mondays due to the accumulation of information over the weekend non-trading period.

99 4.3 Data and Methodology

4.3.1 Data

The data set in this study contains the history of trades and orders of sixty highly liquid individual stocks from the Shanghai Stock Exchange. Some are chosen from the stocks with the best trading value and best trading volume in 2003, and most of the sixty stocks selected are from the Shanghai 180 index (previously the

Shanghai 30 index). The SHSE 180 index is based on 180 blue chip stocks.

Appendix 4.1 provides the codes and names of the sixty individual stocks from the

Shanghai Stock Exchange. The sample of sixty individual stocks records information on the time-stamped transactions. They are the code, the order of time interval, the trading date, the day, trading time, open value, close value, trading volume, best bid price, best bid volume, best ask price, and best ask volume in each 5-minute interval in each trading day. The data for the sixty individual stocks covers the trading period of three years: from January 1, 2000 to December 31, 2002, for a total of 716 trading days. Data were adjusted for stock splits and provided by the SIRCA1.

The bid/ask spreads and other transaction data are computed at 5-minute intervals throughout the trading day. In determining the bid/ask spreads and other transaction data for a 5-minute interval, the closest observed data to the end of that 5 minutes are the values of that 5-minute interval. The first and last 5-minute intervals for the morning trading session are from 09:30 to 09:35 and from 11:25 to 11:30, respectively, and the first and last 5-minute intervals for the afternoon trading session are from 13:00 to 13:05 and from 14:55 to 15:00, respectively. The ending bid value

1 I am grateful to Mr. Patrick Huang from the SIRCA for providing us with all sixty individual stock data series.

100 and ask value of the intervals recorded are used to calculate the relative bid/ask spread. As there is a call auction at the opening (09:25) of the trading day, the bid/ask spread at 09:25 is included as the interval from 09:25 to 09:30. In total, there are forty-nine 5-minute intervals throughout the day broken up into one call auction at the opening, twenty-four intervals for the morning session and twenty-four intervals for the afternoon session. Due to the lack of data of the trading frequency of the stock, the bid/ask spreads and depths are only used to measure the liquidity in the study. The mathematical expressions for the spread and the depth follow the study of the McInish and Wood (1992), Brockman and Chung (1998) and Brockman and

Chung (1999). For all sixty individual stocks, the average intraday bid/ask spreads

(relative spreads) and depths (dollar depth measure) of the intervals in the study are calculated as an equally weighted average of the individual stocks’ relative spreads and depths over a given 5-minute interval. Each relative spread is calculated as follows,

best ask pricet− best bid pricet spdt = (4.1) ()best ask pricett+ best bid price /2

where, for each 5-minute interval of the trading day, best ask pricet and best bid pricet are the ending values of the 5-minute interval. The depths are calculated at the ending time of the 5-minute interval as the sum of the dollar amounts of the buy and sell orders submitted at the best bid and ask prices as follows:

depthtt= best ask price× best ask volumet (4.2) +×best bid pricett best bid volume

101 where best ask volumet and best bid volumet are the ending values of the 5-minute interval. For each interval of the trading day, the price is the stock close price recorded at the end of the 5-minute interval. Trading volume is the total trading volume during the 5-minute interval. The intraday 5-minute return volatility of the interval is calculated by taking the variance of the previous ten returns of 5-mintue intervals of the individual stock. Finally, if there are no observations of bid/ask spread, trading volume and price for all given individual stocks over a given interval, then that interval is deleted from the analysis.

4.3.2 Methodology

The chapter follows the study of McInish and Wood (1992) and Brockman and Chung (1998) to present the null hypothesis as follows. The EMH assumes that stock prices incorporate all information such that changes in prices are only due to news or unanticipated events. In addition, unanticipated information is incorporated instantaneously. According to the EMH, the bid/ask spreads (depths) should be constant both across 5-minute time intervals within the trading day and across days of the week. Hence, the first and second hypotheses are developed:

Hypothesis 1:

H0: The spreads (depths) are constant across 5-minute time

intervals within the trading day.

H1: At least one 5-minute interval has a significantly different

spread (depth) within the trading day.

102

Hypothesis 2:

H0: The spreads (depths) are constant across days of the week.

H1: At least one day of the week has a significantly different

spread (depth).

Rejection of the first hypothesis is evidence of systematic intraday liquidity patterns and rejection of the second hypothesis is evidence of systematic interday liquidity patterns.

Previous research also shows that liquidity is affected by cross-sectional variation in price level, risk and information flow (Benston and Hagerman, 1974;

Stoll, 1978; Barclay and Smith, 1988; Admati and Pfleiderer, 1988; Franz et al.,

1995; McInish and Wood, 1992; Brock and Kleidon, 1992 and Noronha et al., 1996).

Franz et al. (1995) show that bid/ask spreads are systematically affected by price level, price volatility and trading volume (information flow). Greater trading activity can lead to lower spreads due to economies of scale in trading costs. To extend prior work, this chapter also examines whether these variables are the determinants of liquidity after controlling the intraday and interday dummy variables during the trading day. While liquidity can be measured as the bid/ask spread and depth, the depth should also have a systematic relationship with price level, risk and information flow. A further hypothesis test is made of the intraday pattern of bid/ask spreads2. Demsetz (1968), Tinic (1972), Benston and Hagerman (1974) and Stoll

2 I omit the relative test hypotheses on depth but provide test results in Table 5.6, which is consistent with the liquidity theory.

103 (1978) have shown that there is an inverse relationship between a stock's spread and its price. Hence, the following hypothesis is believed to hold.

Hypothesis 3:

H0: There is an inverse relationship between spread and trading

price.

H1: There is not an inverse relationship between spread and

trading price.

Studies have suggested that spreads increase with increasing price volatility

(Hamilton, 1978; Stoll, 1978 and McInish and Wood, 1992). Information affects trading activity and Greater trading activity can lead to lower spreads due to economies of scale in trading costs. The positive relationship between spreads and price volatility arises because market markers are risk averse. Greater trading activity can also lead to large price variation, which results in high volatility. Thus, volatility is positively correlated with information asymmetry. The positive effect of price volatility on spreads can also be found from the observation that greater price volatility implies larger inventory holding costs for specialists. These considerations lead to the fourth hypothesis:

Hypothesis 4:

H0: There is a direct relationship between the level of risk and

spreads.

H1: There is not a direct relationship between the level of risk and

spreads.

104

Glosten and Milgrom (1985) develop a model that deals with how spreads respond to market-generated, and other, public information. Since dealers perceive that there is a positive probability that orders originate from informed traders, orders convey information and affect quotations. Schwartz (1988) reports that the spread widens at times of substantial informational change and the effect of information on spreads could be tested by considering the effect of current trading volume on spreads while controlling for a stock's normal trading volume. Hasbrouck (1988) argues that the spread is related to the specialist’s perceived exposure to private information so that large trades convey more information than small trades. Hence, the following hypothesis is expected to hold.

Hypothesis 5:

H0: There is an inverse relationship between spreads and the

amount of information coming to the market.

H1: There is not an inverse relationship between spreads and the

amount of information coming to the market.

By examining the determinants of liquidity, this provides indirect evidence of the validity of these interday and intraday day predictions. In addition to testing the hypothesis of the existence of interday and intraday liquidity patterns, the results also provide evidence of the validity of prevailing market microstructure theories.

The empirical statistics on the intraday 5-minute bid/ask spreads and depth are first performed and then graphs of relative spread and depth of the time-of-day are plotted in order to examine intraday patterns and interday patterns. The linear

105 regression model is finally used to test Hypotheses 1-5 and verify the patterns of liquidity. The following regression model is estimated in order to investigate interday and intraday patterns after controlling for price effects, risk and information flow.

spdit,01= α +++ααα price it ,2,3var it vol it, 48 4 (4.3) time day ++∑∑γ jj,,tβ iitt+ε ji==11

where spd is the relative bid/ask spread at the end of each 5-minute interval, vol is the trading volume during each 5-minute interval, var is the variance of ten previous returns over each 5-minute interval. Returns are calculated on a continuously compounded basis over 5-minute intervals and price is the recorded stock price at the end of each 5-minute interval. For all observations during the 5-minute interval, the variance (var) is used as the risk variable and the trading volume (vol) as the variable of information flow, in the regression model. All variables of the spread, trading price, variance and trading volume have been taken with a natural logarithm. The null hypothesis 1, that intraday spreads are constant throughout the trading day, is

tested for equality across γ i coefficients, and null hypothesis 2, that interday spreads are constant throughout the trading week, is investigated by testing for equality

across βi coefficients. No dummy variable is included for the 24th time interval

(11:25–11:30) or for Wednesdays in order to avoid perfect collinearity within each set of dummy variables. The null hypotheses 3, 4 and 5 are examined by testing the

αi coefficients.

When considered bid/ask spreads are systematically affected by price effect, risk and information flow, depth can also be a function of these same variables. The

106 regression model that specifically accounts for the effects of these three independent variables in depth is also estimated.

depthit,01=+α ααα price it ,2,3 +var it + vol it, 48 4 (4.4) time day +++∑∑γβjj,,t iittε ji==11

4.4 Empirical Results

This section performs empirical analyses on liquidity proxies, bid/ask spread and depth on the Chinese stock market. From statistical and graphical analyses,

Section 4.4.1 examines the characteristics of the bid/ask spread and depth using 3- year intraday data of sixty stocks. Section 4.4.2 performs the regression for bid/ask spread and depth on the stock price, risk, trading and dummy variables of intraday intervals and interday. The robustness of the results is provided and hypothetical tests are examined. Section 4.4.3 provides the theoretical explanation on liquidity patterns and the inter-temporal relationship between the bid/ask spread and depth.

4.4.1 Summary Statistics

Table 4.1 provides descriptive statistics for the selected sixty stocks covering

716 trading days over the sample period. The average daily trading volume is about

4.98 million. Average spread of 5-minute intraday intervals is 0.001776 and average depth of 5-minute intraday intervals is RMB 538,975.8 (US$65,217.36).

107 Approximately 68.69% of sixty companies are traded on a given day, and the percentage of 5-minute intervals with trading is over 78.22%. For each stock, the number of 5-minute intervals with trading divided by the total number of possible 5- minute intervals is calculated (the total number of 5-minute intervals for a stock with

716 trading days would be 716 times 49, or 35,084). I then calculate the last two summary statistics for the average over the sixty individual stocks.

Figure 4.1 graphs the intraday and interday mean patterns for all 4,988,485 observations of bid/ask spread and bid/ask depth. The upper-left graph of Figure 4.1 shows that the intraday spread pattern is a pronounced L-shaped pattern in spreads across the time of the day with the largest spreads at the auction open and second largest spreads at the opening of continuous order trading. The upper-right graph of

Figure 4.1 displays a broadly L-shaped pattern in spreads across the days of the week with the largest spreads on Mondays and smallest spreads on Wednesdays. The bottom-left graph of Figure 4.1 shows that the intraday depth is a pronounced reverse

L-shaped pattern across the time of the day with the lowest depth at the auction open and second lowest at the opening of continuous order trading. The bottom-right graph of Figure 4.1 displays an inverted V-shaped pattern in depth across the days of the week with the highest depth on Wednesdays and a relative low depth on

Mondays.

Corresponding to Figure 4.1, Table 4.2 provides summary statistics on the average of relative spreads and depth for the five days of the week and 49 5-minute intervals of the day of the sixty stocks. The mean relative spread and depth are the equally weighted average of individual firm’s relative spreads and depths over a given time interval. Panel A provides size, mean, median, maximum and minimum

108 of the days of the week and Panel B provide size, mean, median, maximum and minimum of time interval of the trading day.

In the Panel A of Table 4.2, the mean (median) values of spreads for

Mondays, Tuesdays, Wednesdays, Thursdays, and Fridays are 0.00179 (0.00138),

0.00178 (0.00138), 0.00177 (0.00137), 0.00177 (0.00137) and 0.00177 (0.00138), respectively. The results show that Mondays are associated with the highest spreads and Wednesdays are associated with the lowest spreads. The mean (median) values of depths for Mondays, Tuesdays, Wednesdays, Thursdays, and Fridays are 524444

(363671), 539306 (389633), 566748 (411258), 549299 (399725) and 513628

(367216), respectively. The results show that Mondays are associated with lower depth, Fridays is the lowest in depth and Wednesdays are associated with the highest.

The overall weekly spread pattern is broadly L-shaped. This pattern of the spread is the mirror image of the V-shaped depth pattern in Figure 4.1 (except for the lowest depth on Fridays).

High spreads and low depth on Mondays are consistent with the findings of

Foster and Viswanathan (1990) since non-trading over the weekend allows for a build up of information yet to be impounded into prices. When information is accumulated over the weekend, the liquidity traders would reduce their Monday trading activities due to the higher uncertainty of encountering informed traders. The spread should be wider while depth should be lower on Mondays so that traders will reduce the number of shares against which they are willing to trade. The steady increase in spreads and the decrease in depths from Wednesday to Friday could be explained by the argument that when the market is near to close for the weekend, the informed traders, who get access to public/private information earlier, would want to gather greater profit/compensation towards the close of trading before the

109 information becomes public during the weekend. But this situation may not be obvious because, for the spread and the depth magnitudes across the days of the week, only on Monday, are both statistically significant (Table 4.4 and Table 4.5) from the regression results.

Panel B of Table 4.2 provides the summary statistics of the intraday spreads and depths for auction open and each of 48 intraday intervals. The mean spread is highest at the auction open while the depth is the lowest. The mean spread declines monotonically from the auction time (0.00424) and then the first 5-min interval of the morning open (0.00282) until the morning close (0.00163), which is lowest in the morning session, while mean depth generally rises from the auction open (228,098) to close of the morning session (571,854). In the afternoon session, the mean spread is a bit high at the opening during the 13:00-13:05 period with a mean value of

0.00172 and then steadily declines until rising over the last two 5-minute intervals immediately before the close (0.00159 and 0.001667), while mean depth shows consistent fluctuation during trading time with almost the same level of the morning close. The spread and depth patterns are mirror images of one another.

These results are explained as evidence that the underlying cause(s) of liquidity changes must be related to the opening of the trading session due to the long overnight/lunch-break trading halt. After the open of trading, both in the morning and afternoon sessions, spreads continue to decrease while depths continue to increase until the close of trading in the morning except for the afternoon sessions in the last three intervals from Figure 4.1. This is the pure trading act that appears to increase liquidity. It is interesting to see that these findings confirm the model predicted by Easley and O'Hara (1992) that bid/ask spread depends on time between trades, with spread decreasing when this time increases.

110 To summarize, overall the intraday spread exhibits an L-shaped pattern while the depth displays a reverse L-shape. The spread is the largest and depth is the lowest at the call opening and spread is the second largest and depth is the second lowest at the opening of the continuous trading period. The spread and depth patterns are mirror images of one another. A similar L-shaped intraweek pattern in the spread and a reverse V-shaped intraweek pattern in the depth are also identified. The spread is lowest on Mondays and highest on Wednesdays. The depth is lowest on Fridays and highest on Wednesdays.

4.4.2 Regression Results

This section formally tests the null hypotheses 1-5. The resulting spread pattern will compared with the depth pattern in order to determine whether the two liquidity dimensions work to amplify, or cancel out, the effect of the other. The overall results provide evidence on the validity of the prevailing market microstructure theories.

Table 4.3 presents descriptive statistics for the dependent and independent variables used in the regression, which represent the 5-minute intraday bid/ask spread, depth, trading price, variance and trading volume, respectively. A total of

1,646,749 observations is made for each variable. Panel A of Table 4.3 shows that the maximum and minimum values of all five variables possess very wide dispersions, as expected in such cross-sectional data. Mean (median) values are

0.001739 (0.001343), 12.24 (10.95), 0.120546 (0.05), 67662.8 (20000.0) and

266057.5 (97812.0) for relative spread, prices, variances, trading volume and depth,

111 respectively, corresponding to the same positive skewness of interest. Mean trading volumes of 67,662 shares and share prices of Chinese Dollar $12.24 (US$1.48) suggest that typical trading behavior consists of a relatively large volume of low- priced shares. Mean bid/ask spreads (depths) value of 0.001739 (266057.5) suggest that order-driven bid/ask spreads (depths) are comparable in magnitude to those generated by quote-driven systems (0.0018 for McInish and Wood (1992)).

Panel B of Table 4.3 provides correlation values for all five variables of interest. The spread is positively correlated with variance (0.1421) and trading volume (0.0402) but negatively associated with price (-0.1599) while depth is positively correlated with variance (0.0007) and trading volume (0.3283) but negatively associated with price (-0.1112). The very small correlation (0.0007) between depth and variance is found. Correlation between the independent variables is relatively weak (highest value is only 0.1599 between variance and trading volume), suggesting negligible problems from multi-collinearity when the regression model is used.

Regression models (Equation 4.3 and Equation 4.4) are estimated. The results of the regression of bid/ask spread (depth) against the explanatory and dummy variables are presented in Table 4.4 and Table 4.5. Analysis is only made of the regression results of the spread against all control variables and dummy variables in

Table 4.4. The results of depth in Table 4.5 are basically consistent with these explanations of the spread in the liquidity analyses. In the results of the spread regression in Table 4.4, the F-statistics for the null hypotheses 1 and 2, showing the results of 1819.62 and 9.58, are each significant at the 1% level. Thus, the null hypotheses 1 and 2 are rejected, which provides direct evidence of statistically significant patterns for the time-of-day periods and day-of-week, respectively.

112 The overall L-shaped intraday spread pattern, visible in Figure 4.1, is still shown to be robust with respect to the control variables. Spread magnitudes and significance levels in Table 4.4 are relatively high at the start of trading (call auction,

0.8794) and fall from interval 1 (0.5977) to interval 10 (0.1507) and then fluctuate throughout the rest of the day. The spread is also unusually lowest (0.0629) at the close of the afternoon session. Hence, after controlling the spread determinants used in this analysis, an L-shaped time-of-day component remains.

Following the method of Brockman and Chung (1998), columns five and six provide the sign and F-statistics for testing the null hypothesis of equality for coefficients between consecutive 5-minute intervals. For example, the first entries in columns five and six (a minus sign and 4432.488***, respectively) in Table 4.4 include that the coefficient of the first 5-minute interval (continuous trading) minus that of the call auction is negative and significantly different from zero at the 1% level. Focusing on these last two columns reveals that the mean spreads from the 2nd to the 12th interval display significant declines from their preceding 5-minute interval. An explanation of this is that, after the market opening, information accumulated during the overnight trading halt affects the market, quickly decreasing during the first hour in the morning session and the spread narrows sharply and significantly from its previous 5-minute magnitude.

It is possible that the time-of-day dummy variables are capturing a time-of- day preference for trading (such as structural changes in the trading process across the trading day). The results of depth regression in Table 4.5 also support the existing patterns of liquidity theories. This is evidence that significant differences exist among interday and intraday spreads and depths and that these differences cannot be

113 explained solely by related variations in trading prices, volatilities, or trading volume.

The day-of-week dummy coefficients (βi) of the relative spread for Monday,

Tuesday, Thursday, and Friday in Table 4.4 are estimated at 0.0068, 0.0014, 0.0008 and 0.0001, respectively and the only significant coefficient is for Monday at the 1% level of significance. But the coefficients for Monday and Thursday are significant at the 1% significance level in the depth regression. When variations in the control variables are taken into account, the wider open-of-week spread and depth become even more pronounced. The coefficients reveal that the same weekly L-shaped pattern of spread and reverse V-shaped pattern of depth observed in the raw data remains intact. In addition, the Monday effects are obvious.

The higher spreads at the open of trading appear very robust. The depth patterns documented in Table 4.5 are the mirror images of the patterns reported. Each dimension of liquidity magnifies the effect of the other such that when bid/ask spreads are wide (narrow), depths are shallow (deep). These findings are consistent with the results from the other markets. Thus, other specific market characteristics, such as designated dealers, the pure order-driven system, pre-mandatory quotation periods and price limitations, do not uniquely explain observed bid/ask spread and depth variations.

Estimation results of Equation 4.3 indicate that coefficients for trading price, return variance and trading volume are –0.5626, 0.1280 and –0.0709 in the spread regression in Table 4.4. All controlling variables are significant at the 1% level and the null hypotheses 3, 4 and 5 are accepted. The coefficient (-0.5626) of trading price is significantly negative confirming hypothesis 3. The negative relationship between bid/ask spreads and prices can be explained by the existence of fixed costs in the

114 order-processing component of the spread. If order-processing costs constitute a relatively fixed dollar amount, then higher price stocks will produce lower proportional spreads. The coefficient (0.1280) of variance is significantly positive, demonstrating that differential spreads across stocks can be explained, in part, by differences in a stock’s risk. Hence, the result supports hypothesis 4.

The positive relationship (0.1280) between bid/ask spreads and variance

(risk) is usually associated with the inventory-holding cost component, which can be explained as liquidity interaction based on the suggestion by Brock and Kleidon

(1992). Investors hold inventory in order to even out order flow imbalances. Wider spread leads to higher risk, and higher levels of price volatility lead to higher inventory management risks, which, in turn, lead to higher required compensation in the form of wider spreads.

The coefficient (-0.0709) of trading volume, the measure of information flow, is significantly negative supporting hypothesis 5. This result is consistent with that of

McInish and Wood (1992) but in contrast to Brock and Kleidon (1992). Higher volumes are expected to put downward pressure on order processing costs and higher variances are expected to put upward pressure on adverse selection and/or inventory holding costs. At the same time, the result is in agreement with the predictions of current information based models (Admati and Pfleiderer, 1988).

In the depth regression of Equation 4.4, the coefficients of trading price, variance and trading volume are -0.3720, -0.1877, 0.3796, respectively, and all of them are significant at the 1% level. Thus, the spread increases with higher return variance (0.1280) and decreases with higher trading volumes (-0.0709) while the depth increases with higher trading volume (0.3796) and decreases with higher return variance (-0.1877). These are very consistent with liquidity theories suggesting that

115 spread is a decreasing function of trading price and trading volume, and an increasing function of return variance, and vice versa for depth. However, one result from the depth regression shows that depth has a negative relationship with trading price (-0.3720). This negative trading price relationship in depth is hard to explain, but its effect is smaller on depth than on spread. Overall, the consistency between the findings with respect to spreads and depths is reconfirming that both spread and depth are used interchangeably to measure liquidity.

Table 4.6 reports the correlations between spread and depth for each 5-minute interval. All correlation coefficients in intraday 5-minute intervals are negative. The results provide further evidence of the existence of the negative relationship between the spread and the depth. The negative association between spread and depth on the

Shanghai Stock Exchange implies that limit order traders actively manage both the price and quantity dimensions of liquidity by adjusting the spread and depth.

To conclude, all coefficients of the dummy variables for intraday intervals in the spread and depth regressions are significant at the 1% level, further providing evidence of the striking L-shaped spread pattern and reverse L-shaped depth pattern.

All controlling variables for the trading price, risk and trading volume are the determinants of the liquidity and are significant at the 1% level. The spread increases with higher return variance (0.1280) and decreases with higher trading volumes (-

0.0709) while the depth increases with higher trading volume (0.3796) and decreases with higher return variance (-0.1877). The coefficient for Monday is significant at the 1% level, which suggests the striking Monday effects. There is a negative correlation between spread and depth by statistical analysis in each 5-minute interval.

116 4.4.3 Discussion of the Bid/Ask Patterns and Depth

The findings reveal that the spread in the limit-order market of Shanghai exhibits an L-shaped intraday pattern while the depth displays a reverse intraday L- shaped pattern. These findings conflict with the EMH that the bid/ask spreads

(depths) should be constant both across 5-minute time intervals within the trading day and across days of the week. The question arises, why do these intraday patterns exist in the Shanghai market? The potential results of liquidity variation at the opening of the trading session are interpreted. The order-driven market in China allows us to isolate these opening effects from other complicating factors embedded in other stock exchanges’ designs and procedures.

So far, three models of microstructure theories have been proposed to explain the intraday liquidity patterns: order-processing, asymmetric information, and inventory holding costs. The first theory (order-processing) is based on differential liquidity demand elasticity during the trading day and the ability of liquidity suppliers to exploit these differentials. Brock and Kleidon (1992) suggest that liquidity demand is more inelastic at the open and close of the market, thus making it possible for monopolistic suppliers to extract rents at these times. The second theory

(asymmetric information) is based on the existence of adverse selection and suggests that changes in the liquidity reflect changes in the level of asymmetric information

(Kyle, 1985; Amihud and Mendelson, 1988 and Foster and Viswanathan, 1990).

Higher spreads following a period of non-trading (Mondays and daily openings), for example, are consistent with the view that accumulated private information is impounded during these periods. The third theory (inventory holding costs) claims that the spread exists to compensate market makers for bearing the risk of undesired

117 inventory (Stoll, 1978 and Ho and Stoll, 1981). The market makers will adjust the bid/ask spread to attract orders and move back to their optimal inventory positions when there is an order imbalance that moves them away from their desired inventory positions.

In considering the characteristics of the China’s stock market, it does not have market makers, and orders from the investors and trading transactions are very transparent. In contrast to NYSE and AMEX studies, the findings demonstrate that designated market makers cannot be solely responsible for observed patterns. Similar to Chan et al. (1995), the findings are able to rule out the monopolist market maker hypothesis as the underlying cause since the liquidity patterns documented in China’s stock market are based on an order-driven trading mechanism with no designated market makers.

Based on the above three theoretical models, the explanation of asymmetric information (the second theory) would be a main factor in the formation of the patterns for liquidity. The asymmetric information hypothesis is consistent with a relatively illiquid opening of trading followed by improvements over subsequent information-revealing trading. Due to its short history, China’s stock market is dominated by a high number of inexperienced and uninformed individual investors.

Due to the long trading halt overnight, the accumulated information release and private information will affect the market’s opening on the next day. The combination of a wider spread and smaller depth at the opening on the SHSE is consistent with the trading strategy adopted by uninformed investors. Most investors with less information attempt to minimize losses from trading with the informed investors when they face a severe adverse selection problem around market opening.

118 On the other hand, investors may have a difference in understanding of accumulating overnight news; this results in the stock price reflecting some “noise- trading”. So a wider bid/ask spread and lower depth occur in the market opening. As for the market closing, investors have gradually understood and assimilated information during the market, so the bid/ask spreads are small and depths are high in their normal positions. Another possible explanation is that investors choose a small bid/ask spread and high depth in order to reduce the possible loss due to the overnight halt. In China’s market, the mechanism of public information release and private information are not completely monitored. In addition, systems do not exist for the complete monitoring the effect of asymmetric information, and at the opening, are less attractive to uninformed investors. Higher spread with low depth implies less liquidity as less order flow is absorbed with large changes in prices.

Thus, policy-making and market monitoring need further attention.

4.5 Summary and Conclusion

This chapter examines the intraday behaviors of the bid/ask spread and depth and their determinants on the order-driven market in the Shanghai Stock Exchange.

To date, most evidence of interday and intraday liquidity patterns has been obtained from either a specialist or multi-dealer system in the US market, or an order-driven market such as in the Japan and Hong Kong markets. But none of the literature so far, to my knowledge, has examined China’s stock market using high-frequency data.

The trading system in China is different from any of these exchanges mentioned above in that it is an electronic call auction at the opening and continuous order-

119 driven trading market in the trading day, and the market relies solely on limit-order placement.

Since spread and depth are two dimensions of market liquidity, both variables should play important roles in explaining the market liquidity and components. Three competing theories have been proposed in the US market to explain the intraday liquidity patterns: order-processing, asymmetric information, and inventory holding costs. This chapter analyzes unique high-frequency 5-minute data for sixty individual stocks over a three-year period from January 1, 2000 to December 31, 2002 on the

Shanghai Stock Exchange. The analysis shows unique findings that are different from those for other major exchanges such as the US and Hong Kong markets. In particular, the intraday 5-minute bid/ask spread displays an L-shaped pattern and depth exhibits an inverted L-shaped pattern. This is in striking contrast to the findings of the overall U-shape of the spread by McInish and Wood (1992) and

Madhavan et al. (1997) in the US market and Brockman and Chung (1998) and Ahn and Cheung (1999) in the Hong Kong market.

First, the existence of the L-shaped spread pattern (inverse L-shaped depth) on the Shanghai Stock Exchange means that specialist activity alone cannot account for the widening of spreads (lowest depth) at the opening of the call auction and continuous trading. The evidence is, therefore, inconsistent with the monopolistic market maker hypothesis from the first theory. The widening of spreads (lowest depth) at the opening of the call auction and continuous trading is attributed to the factor of asymmetric information, which plays an important role in the patterns for liquidity. The findings are consistent with the Chicago Board of Exchange (Chan et al., 1995).

120 Second, the chapter documents the interday liquidity patterns that still show differences in spreads and depths across days of the week but the strength of these differences is much less than across intervals of the day, similar to previous studies based on US data. The simultaneous occurrence of relatively wide spreads and lower depth on Mondays is consistent with the predictions of Foster and Viswanathan

(1990). Higher Monday opening spreads are probably due to information accumulation of trading halt.

Third, the findings of the striking L-shaped pattern of the bid/ask spreads have the positive relationship with volatility but negative stock price and trading volume while the striking reverse L-shaped pattern of the depths have the negative relationship with stock price, volatility but positive trading volume in the regression model.

Finally, there is an evident negative relationship between the spread and the depth at the opening of the trading day. Overall, the study suggests that the determinant of information asymmetries, through time and across traders, plays a key role in generating observed liquidity variations.

This chapter performs an empirical analysis on the inter-temporal liquidity characteristics of bid/ask spread and depth and their determinants in the Chinese stock market during the three-year period. The next chapter will explore the time series characteristics of the Chinese stock market using a set of individual stocks from the Shanghai Stock Exchange. The GARCH methodology will be used for the volatility analysis, and the relationship between the volatility, trading volume and bid/ask spread, which does not require the assumption of normality but can capture the dynamic characteristics of the volatility.

121

Table 4.1 Sixty Individual Stocks from the Shanghai Stock Exchange, 2000 - 2002

No Data Description Size

1 Number of companies Listed 60 2 Number of trading days over the sample period 716 3 Average daily trading volume in the number of stocks 4988485 4 Average spread of 5-minute intraday intervals 0.001776 5 Average depth of 5-minute intraday intervals RMB $538975.8 6 Average company-days with one or more shares traded 68.69% 7 Average 5-minute intraday intervals of spread (for all companies over the sample period) with one or more shares traded 78.23%

122 Table 4.2 Summary statistics of the average intraday 5-minute spreads and depths, 2000 – 2002 Panel A Spread Depth Spread Depth Interval Size Mean Median Max Min Mean Median Max Min Interval Size Mean Median Max Min Mean Median Max Min Monday 324787 0.00179 0.00138 0.0295 0.00083 524444 363671 13669352 4527 Thursday 333254 0.00177 0.00137 0.027 0.00083 549300 399725 11602800 8511 Tuesday 330369 0.00178 0.00138 0.0277 0.00083 539306 389633 11802895 5052 Friday 332221 0.00177 0.00138 0.027 0.00082 513628 367216 10364878 17410 Wednesday 326118 0.00177 0.00137 0.0266 0.00083 566748 411259 13017153 12591 Panel B All 1646749 0.00178 0.00138 0.0431 0.00080 538976 388714 20685948 3696 Call Auction 33008 0.00424 0.00296 0.0402 0.00083 228098 122028 5434584 5188 09:30-09:35 29985 0.00282 0.00204 0.0204 0.00084 308396 190161 4260181 12540 13:00-13:05 34661 0.00172 0.00139 0.0105 0.00084 545599 401961 5977913 45789 09:35-09:40 32722 0.00246 0.00182 0.0163 0.00084 362341 256541 5253181 18487 13:05-13:10 34173 0.00167 0.00136 0.0104 0.00084 544414 405017 5804175 56751 09:40-09:45 33351 0.00220 0.00166 0.0142 0.00084 397305 256630 4527547 21066 13:10-13:15 34129 0.00165 0.00134 0.0097 0.00084 552533 400727 6596488 52146 09:45-09:50 33472 0.00206 0.00158 0.0134 0.00084 436850 322241 4769648 30949 13:15-13:20 34032 0.00165 0.00135 0.0095 0.00084 557520 427120 6068111 54755 09:50-09:55 33744 0.00195 0.00150 0.0122 0.00084 477833 323030 5698036 37090 13:20-13:25 34120 0.00165 0.00134 0.0095 0.00084 571079 439716 5982705 52320 09:55-10:00 33782 0.00188 0.00147 0.0121 0.00084 485949 337446 6427286 46193 13:25-13:30 34249 0.00164 0.00134 0.0095 0.00084 596339 450311 5737799 54594 10:00-10:05 33925 0.00183 0.00144 0.0106 0.00084 518081 358261 5049395 59760 13:30-13:35 34354 0.00163 0.00134 0.0094 0.00084 567937 427543 6543729 61164 10:05-10:10 33973 0.00179 0.00141 0.0111 0.00084 520111 345048 5520964 38185 13:35-13:40 34429 0.00163 0.00133 0.0101 0.00084 595497 424209 6357055 27760 10:10-10:15 33942 0.00177 0.00140 0.0111 0.00083 536658 354667 5044765 43349 13:40-13:45 34458 0.00163 0.00133 0.0095 0.00084 594571 429794 6344955 58719 10:15-10:20 33924 0.00174 0.00139 0.0098 0.00084 534648 363861 5334947 42627 13:45-13:50 34476 0.00164 0.00134 0.0105 0.00084 583341 429831 6661226 52575 10:20-10:25 33927 0.00172 0.00139 0.0102 0.00084 535123 389742 6148231 40998 13:50-13:55 34595 0.00162 0.00133 0.0096 0.00084 572284 413508 6897123 41574 10:25-10:30 33876 0.00171 0.00138 0.0102 0.00084 548366 363518 5978172 38837 13:55-14:00 34472 0.00162 0.00133 0.0091 0.00084 585413 420433 5982779 37938 10:30-10:35 33869 0.00170 0.00137 0.0104 0.00084 550499 390188 5638414 40155 14:00-14:05 34381 0.00161 0.00132 0.0091 0.00084 583574 415092 7327358 44200 10:35-10:40 34011 0.00169 0.00137 0.0105 0.00084 539971 379340 5929730 40268 14:05-14:10 34488 0.00162 0.00133 0.009 0.00084 580690 430506 7177610 16141 10:40-10:45 34012 0.00167 0.00136 0.0096 0.00084 530854 379796 6025225 40087 14:10-14:15 34576 0.00161 0.00133 0.0087 0.00084 614631 447410 7845002 50495 10:45-10:50 33981 0.00167 0.00136 0.0097 0.00084 540884 396213 6156388 10147 14:15-14:20 34562 0.00161 0.00132 0.0094 0.00084 593521 422315 8015008 61337 10:50-10:55 33935 0.00166 0.00135 0.0096 0.00084 542772 420245 6162921 20594 14:20-14:25 34590 0.00161 0.00132 0.0104 0.00084 600346 425041 7457465 29351 10:55-11:00 33902 0.00166 0.00135 0.01 0.00084 548362 404417 6575774 38919 14:25-14:30 34536 0.00161 0.00133 0.0105 0.00083 574522 429402 7069436 34925 11:00-11:05 33878 0.00166 0.00136 0.0094 0.00084 566635 439873 6131504 46663 14:30-14:35 34536 0.00160 0.00132 0.0089 0.00083 576025 416916 6743168 29926 11:05-11:10 33916 0.00166 0.00136 0.0097 0.00084 542929 410453 5687431 50777 14:35-14:40 34611 0.00160 0.00132 0.0096 0.00083 576308 417509 8028948 40140 11:10-11:15 33942 0.00166 0.00136 0.0098 0.00084 553358 401003 6327052 41495 14:40-14:45 34663 0.00159 0.00131 0.0097 0.00083 591084 444425 6602199 18913 11:15-11:20 33884 0.00166 0.00136 0.0096 0.00084 571836 424630 5913466 49579 14:45-14:50 34728 0.00159 0.00131 0.01 0.00083 584123 442063 6294306 51131 11:20-11:25 33831 0.00167 0.00136 0.0099 0.00084 557850 424651 5359727 57039 14:50-14:55 34692 0.00159 0.00131 0.0102 0.00084 568646 401739 6011801 34018 11:25-11:30 17938 0.00163 0.00132 0.0088 0.00084 571854 416705 5210784 59912 14:55-15:00 29508 0.00167 0.00129 0.0165 0.00084 547657 343754 7841949 53663

123

Table 4.3 Descriptive statistics and correlation matrix of variables of 5-minute intervals, 2000 - 2002

Panel A: Descriptive statistics Excess Variable Size Mean Median Max Min Skewness Kurtosis spd 1646749 0.001739 0.001343 0.103943 0.000147 7.332282 161.6475 Trading price 1646749 12.24 10.95 70.00 2.55 1.997653 10.70235 Variance 1646749 0.120546 0.05 27.66 0.00 18.01618 775.6224 Trading volume 1646749 67662.8 20000.0 99200000.0 1.0 62.29658 14772.69 depth 1646749 266057.5 97812.0 82800000.0 149.0 23.14345 843.0973

Panel B: Correlation matrix Trading Trading Variable spd price Variance volume depth spd 1 Trading price -0.1599 1 Variance 0.1421 0.0328 1 Trading volume 0.0402 -0.1047 0.1599 1 depth -0.1112 0.0007 0.3283 1

Note: • The statistical data in this table is from the stacked dataset. • Spd is the relative bid/ask spread at the end of each 5-minute interval. Vol is the trading volume during each 5-minute interval. Var is the variance of return over each 5-minute interval, and returns are calculated on a continuously compounded basis over 5-minute intervals. Price is the recorded stock price at the end of each 5-minute interval. Depths are calculated at the ending time of the 5-minute interval as the sum of the dollar amounts of the buy and sell orders submitted at the best bid and ask prices

124 Table 4.4 Regression results for spread of the 5-minute intervals for Equation 4.3 Overall model statistics Additional tests for equality of coefficients Coefficient t-Statistic Coefficient t-Statistic *** *** Sample size 1,646,749 Intercept α0 -4.3925 -823.69 Monday β1 0.0068 5.22 *** *** F-ratio 13402.35 F-Statistic Degrees of freedom Trading price α1 -0.5626 -698.01 Tuesday β2 0.0014 1.06 Degrees of freedom 55 and *** *** 1646693 0=1=…=23=25=…=48=0 1819.62 48 and 1646693 Return Variance α2 0.1280 356.77 Thursday β4 0.0008 0.6 2 *** *** Adjusted R 0.3092 1 = 2 = 4 = 5=0 9.58 4 and 1646693 Trading volume α3 -0.0709 -254.59 Friday β5 0.0001 0.05

No Time Coefficient t-Statistic Sign of (j -j-1) F-Statistic H0: j = j-1 No Time Coefficient t-Statistic Sign of (j -j-1) F-Statistic H0: j = j-1 *** *** + *** Call 09:15-09:25 γ0 0.8794 178.79 25 13:00-13:05 γ25 0.2559 52.08 221.662 *** - *** *** *** 1 09:30-09:35 γ1 0.5977 118.69 4432.488 26 13:05-13:10 γ26 0.2110 42.93 - 124.303 *** - *** *** *** 2 09:35-09:40 γ2 0.4545 91.74 1147.265 27 13:10-13:15 γ27 0.2003 40.74 - 6.981 *** - *** *** + 3 09:40-09:45 γ3 0.3539 71.67 597.746 28 13:15-13:20 γ28 0.2033 41.33 0.555 *** - *** *** + 4 09:45-09:50 γ4 0.2913 59.01 234.084 29 13:20-13:25 γ29 0.2064 41.94 0.57 *** - *** *** + 5 09:50-09:55 γ5 0.2459 49.87 123.663 30 13:25-13:30 γ30 0.2104 42.78 1.001 *** - *** *** 6 09:55-10:00 γ6 0.2136 43.32 62.919 31 13:30-13:35 γ31 0.2083 42.37 - 0.265 *** - *** *** 7 10:00-10:05 γ7 0.1923 39.02 27.526 32 13:35-13:40 γ32 0.2040 41.51 - 1.123 *** - *** *** 8 10:05-10:10 γ8 0.1663 33.76 40.911 33 13:40-13:45 γ33 0.2019 41.08 - 0.286 *** - ** *** 9 10:10-10:15 γ9 0.1581 32.1 4.036 34 13:45-13:50 γ34 0.2056 41.84 + 0.86 *** - * *** 10 10:15-10:20 γ10 0.1507 30.6 3.332 35 13:50-13:55 γ35 0.2008 40.89 - 1.418 *** * *** 11 10:20-10:25 γ11 0.1582 32.12 + 3.384 36 13:55-14:00 γ36 0.1982 40.34 - 0.413 *** + * *** 12 10:25-10:30 γ12 0.1654 33.58 3.118 37 14:00-14:05 γ37 0.1959 39.85 - 0.325 *** + *** 13 10:30-10:35 γ13 0.1697 34.47 1.153 38 14:05-14:10 γ38 0.1956 39.78 - 0.009 *** + *** 14 10:35-10:40 γ14 0.1756 35.68 2.089 39 14:10-14:15 γ 39 0.1938 39.44 - 0.189 *** *** 15 10:40-10:45 γ15 0.1730 35.15 - 0.42 40 14:15-14:20 γ40 0.1892 38.5 - 1.294 *** *** 16 10:45-10:50 γ16 0.1747 35.51 + 0.187 41 14:20-14:25 γ41 0.1884 38.33 - 0.047 *** *** 17 10:50-10:55 γ17 0.1743 35.43 - 0.009 42 14:25-14:30 γ42 0.1941 39.47 + 2.049 *** + *** 18 10:55-11:00 γ18 0.1787 36.31 1.127 43 14:30-14:35 γ43 0.1891 38.44 - 1.556 *** + *** 19 11:00-11:05 γ19 0.1819 36.96 0.619 44 14:35-14:40 γ44 0.1924 39.1 + 0.677 *** + *** 20 11:05-11:10 γ20 0.1836 37.33 0.193 45 14:40-14:45 γ45 0.1932 39.25 + 0.042 *** + *** + 21 11:10-11:15 γ21 0.1892 38.46 1.863 46 14:45-14:50 γ46 0.1958 39.74 0.396 *** + *** + *** 22 11:15-11:20 γ22 0.1930 39.23 0.876 47 14:50-14:55 γ47 0.2064 41.85 6.998 *** + *** *** 23 11:20-11:25 γ23 0.1957 39.79 0.457 48 14:55-15:00 γ48 0.0629 12.55 - 1152.597 Note: * stands for 10% significance level, ** stands for 5% significance level and *** stands for 1% significance level 125 Table 4.5 Regression results for depth of the 5-minute intervals for Equation 4.4. Overall model statistics Additional tests for equality of coefficients Coefficient t-Statistic Coefficient t-Statistic *** *** Sample size 1,646,749 Intercept α0 9.1486 851.2 Monday β1 -0.0124 -4.7 *** *** F-ratio 12414.49 F-Statistic Degrees of freedom Trading price α1 -0.3720 -229 Tuesday β2 -0.0022 -0.85 Degrees of freedom 55 and *** *** *** 1646693 0=1=…=23=25=…=48=0 1081.41 48 and 1646693 Return Variance α2 -0.1877 -259.63 Thursday β4 -0.0074 -2.81 2 *** *** ** Adjusted R 0.2931 1 = 2 = 4 = 5=0 6.64 4 and 1646693 Trading volume α3 0.3796 676.05 Friday β5 -0.0059 -2.24

No Time Coefficient t-Statistic Sign of (j -j-1) F-Statistic H0: j = j-1 No Time Coefficient t-Statistic Sign of (j -j-1) F-Statistic H0: j = j-1 *** *** - *** Call 09:15-09:25 γ0 -1.4118 -142.41 25 13:00-13:05 γ25 -1.1179 -112.87 300.557 *** - *** *** + *** 1 09:30-09:35 γ1 -1.5296 -150.71 190.978 26 13:05-13:10 γ26 -0.9847 -99.43 268.534 *** + *** *** + 2 09:35-09:40 γ2 -1.3783 -138.04 315.26 27 13:10-13:15 γ27 -0.9797 -98.89 0.375 *** + *** *** + 3 09:40-09:45 γ3 -1.2846 -129.09 127.453 28 13:15-13:20 γ28 -0.9742 -98.27 0.46 *** + *** *** - * 4 09:45-09:50 γ4 -1.2078 -121.42 86.716 29 13:20-13:25 γ29 -0.9881 -99.66 2.914 *** + *** *** - ** 5 09:50-09:55 γ5 -1.1614 -116.88 31.892 30 13:25-13:30 γ30 -1.0050 -101.38 4.298 *** + *** *** 6 09:55-10:00 γ6 -1.1125 -111.96 35.503 31 13:30-13:35 γ31 -1.0054 -101.46 - 0.002 *** + *** *** 7 10:00-10:05 γ7 -1.0825 -109.02 13.398 32 13:35-13:40 γ32 -1.0066 -101.61 - 0.021 *** + *** *** 8 10:05-10:10 γ8 -1.0524 -106.01 13.532 33 13:40-13:45 γ33 -1.0045 -101.41 + 0.067 *** + *** *** 9 10:10-10:15 γ9 -1.0244 -103.17 11.738 34 13:45-13:50 γ34 -1.0004 -101 + 0.252 *** + ** *** + 10 10:15-10:20 γ10 -1.0039 -101.11 6.26 35 13:50-13:55 γ35 -0.9980 -100.81 0.089 *** *** + 11 10:20-10:25 γ11 -1.0102 -101.77 - 0.584 36 13:55-14:00 γ36 -0.9893 -99.87 1.151 *** - *** 12 10:25-10:30 γ12 -1.0141 -102.15 0.223 37 14:00-14:05 γ37 -0.9896 -99.85 - 0.001 *** - *** 13 10:30-10:35 γ13 -1.0149 -102.26 0.011 38 14:05-14:10 γ38 -0.9945 -100.38 - 0.363 *** - *** 14 10:35-10:40 γ14 -1.0235 -103.18 1.098 39 14:10-14:15 γ 39 -0.9861 -99.57 + 1.05 *** *** 15 10:40-10:45 γ15 -1.0187 -102.7 + 0.347 40 14:15-14:20 γ40 -0.9947 -100.42 - 1.113 *** *** 16 10:45-10:50 γ16 -1.0149 -102.33 + 0.215 41 14:20-14:25 γ41 -0.9926 -100.21 + 0.069 *** ** *** 17 10:50-10:55 γ17 -0.9966 -100.49 + 4.966 42 14:25-14:30 γ42 -0.9968 -100.57 - 0.271 *** + *** 18 10:55-11:00 γ18 -0.9873 -99.55 1.298 43 14:30-14:35 γ43 -0.9985 -100.71 - 0.047 *** + *** - *** 19 11:00-11:05 γ19 -0.9831 -99.13 0.266 44 14:35-14:40 γ44 -1.0278 -103.64 13.025 *** + *** - * 20 11:05-11:10 γ20 -0.9764 -98.49 0.664 45 14:40-14:45 γ45 -1.0426 -105.08 3.351 *** - *** - ** 21 11:10-11:15 γ21 -0.9785 -98.71 0.064 46 14:45-14:50 γ46 -1.0604 -106.81 4.792 *** - *** - *** 22 11:15-11:20 γ22 -0.9831 -99.16 0.315 47 14:50-14:55 γ47 -1.1081 -111.49 34.783 *** + *** + *** 23 11:20-11:25 γ23 -0.9766 -98.5 0.621 48 14:55-15:00 γ48 -0.2283 -22.59 11000 Note: * stands for 10% significance level, ** stands for 5% significance level and *** stands for 1% significance level 126 Table 4.6 Correlation between spreads and depths of 60 individual stocks, 2000 - 2002 3rd 1st 3rd 1st No Interval Mean Std. Max. Quartile Median Quartile Min. No Interval Mean Std. Max. Quartile Median Quartile Min. 0 Call Auction -0.0561 0.0839 0.1408 -0.0202 -0.0592 -0.0989 -0.4468 1 09:30-09:35 -0.0447 0.1066 0.4838 -0.0032 -0.0538 -0.0893 -0.4863 25 13:00-13:05 -0.0563 0.0883 0.2012 -0.0045 -0.0642 -0.0920 -0.3180 2 09:35-09:40 -0.0663 0.0951 0.2251 -0.0230 -0.0702 -0.1045 -0.5332 26 13:05-13:10 -0.0601 0.0895 0.2206 -0.0154 -0.0623 -0.1121 -0.2723 3 09:40-09:45 -0.0646 0.1136 0.3016 -0.0255 -0.0797 -0.1175 -0.5276 27 13:10-13:15 -0.0653 0.0947 0.2394 -0.0240 -0.0700 -0.1148 -0.2655 4 09:45-09:50 -0.0574 0.1071 0.2719 -0.0136 -0.0682 -0.1105 -0.5510 28 13:15-13:20 -0.0616 0.0872 0.2089 -0.0222 -0.0695 -0.0952 -0.3223 5 09:50-09:55 -0.0627 0.0974 0.1519 -0.0147 -0.0679 -0.1012 -0.5913 29 13:20-13:25 -0.0672 0.0958 0.2430 -0.0191 -0.0669 -0.1189 -0.3400 6 09:55-10:00 -0.0607 0.1074 0.1898 -0.0101 -0.0596 -0.1110 -0.5681 30 13:25-13:30 -0.0729 0.0923 0.2475 -0.0240 -0.0739 -0.1194 -0.3913 7 10:00-10:05 -0.0689 0.1053 0.2496 -0.0233 -0.0803 -0.1190 -0.5240 31 13:30-13:35 -0.0802 0.0841 0.2300 -0.0370 -0.0836 -0.1174 -0.3515 8 10:05-10:10 -0.0736 0.1006 0.2396 -0.0313 -0.0728 -0.1090 -0.5110 32 13:35-13:40 -0.0762 0.0950 0.2889 -0.0358 -0.0894 -0.1221 -0.4335 9 10:10-10:15 -0.0645 0.0995 0.2236 -0.0194 -0.0683 -0.1201 -0.4758 33 13:40-13:45 -0.0695 0.1063 0.2990 -0.0282 -0.0763 -0.0963 -0.3989 10 10:15-10:20 -0.0609 0.1060 0.3080 -0.0326 -0.0692 -0.1046 -0.4617 34 13:45-13:50 -0.0782 0.0943 0.3073 -0.0369 -0.0890 -0.1267 -0.4343 11 10:20-10:25 -0.0678 0.1138 0.3838 -0.0266 -0.0639 -0.1105 -0.4820 35 13:50-13:55 -0.0687 0.0991 0.2323 -0.0277 -0.0616 -0.1156 -0.4385 12 10:25-10:30 -0.0671 0.1083 0.2991 -0.0185 -0.0784 -0.1162 -0.4791 36 13:55-14:00 -0.0735 0.1103 0.2340 -0.0294 -0.0777 -0.1173 -0.5028 13 10:30-10:35 -0.0728 0.0911 0.1841 -0.0281 -0.0731 -0.1090 -0.4761 37 14:00-14:05 -0.0747 0.0994 0.2913 -0.0346 -0.0795 -0.1183 -0.4733 14 10:35-10:40 -0.0680 0.1090 0.1754 -0.0309 -0.0742 -0.1038 -0.4653 38 14:05-14:10 -0.0759 0.0862 0.2012 -0.0366 -0.0746 -0.1091 -0.4198 15 10:40-10:45 -0.0790 0.1145 0.2892 -0.0394 -0.0798 -0.1303 -0.5346 39 14:10-14:15 -0.0756 0.1001 0.2660 -0.0367 -0.0676 -0.1289 -0.4870 16 10:45-10:50 -0.0708 0.1020 0.2362 -0.0250 -0.0666 -0.1107 -0.5121 40 14:15-14:20 -0.0791 0.0928 0.2171 -0.0331 -0.0800 -0.1168 -0.4940 17 10:50-10:55 -0.0786 0.0922 0.1448 -0.0280 -0.0633 -0.1129 -0.4602 41 14:20-14:25 -0.0584 0.1096 0.3187 -0.0205 -0.0634 -0.0980 -0.4495 18 10:55-11:00 -0.0683 0.0964 0.2148 -0.0217 -0.0708 -0.1055 -0.4624 42 14:25-14:30 -0.0630 0.1067 0.2510 -0.0154 -0.0581 -0.1166 -0.4244 19 11:00-11:05 -0.0771 0.1015 0.2709 -0.0269 -0.0812 -0.1218 -0.4674 43 14:30-14:35 -0.0667 0.1030 0.2668 -0.0221 -0.0549 -0.1147 -0.4527 20 11:05-11:10 -0.0717 0.1035 0.2458 -0.0369 -0.0707 -0.1210 -0.3870 44 14:35-14:40 -0.0628 0.1017 0.1991 -0.0099 -0.0526 -0.1011 -0.4712 21 11:10-11:15 -0.0688 0.0967 0.2782 -0.0284 -0.0686 -0.1101 -0.3403 45 14:40-14:45 -0.0668 0.1126 0.3243 -0.0186 -0.0583 -0.1238 -0.4297 22 11:15-11:20 -0.0672 0.1011 0.2807 -0.0326 -0.0648 -0.1071 -0.4272 46 14:45-14:50 -0.0633 0.1010 0.2864 -0.0192 -0.0633 -0.1187 -0.3520 23 11:20-11:25 -0.0649 0.1043 0.2779 -0.0271 -0.0772 -0.1124 -0.4442 47 14:50-14:55 -0.0539 0.0957 0.2411 -0.0152 -0.0579 -0.0989 -0.2747 24 11:25-11:30 -0.0646 0.1069 0.2699 -0.0116 -0.0627 -0.1259 -0.3868 48 14:55-15:00 -0.0316 0.0958 0.2594 0.0125 -0.0347 -0.0894 -0.3671 All -0.0669 0.0769 0.1656 -0.0359 -0.0578 -0.0953 -0.4279 Note: (1) The correlation between spreads and depths is calculated for each stock and for 5-minute interval of the trading day of the total 60 stocks from the Shanghai Stock Exchange. (2) Cross-sectional summary statistics are provided.

127

Appendix 4.1 List of Sixty Individual Stocks from the Shanghai Stock Exchange, 2000 - 2002 Code Company Name Code Company Name 600001 Handan Iron & Steel Co., Ltd. 600609 Shenyang Jinbei Automotive Company Limited 600002 SINOPEC QILU Company Ltd. 600618 Shanghai Chlor-alkli Chemical Co., Ltd. 600005 Wuhan Steel Processing Co., Ltd. 600623 Shanghai Tyre & Rubber Co., Ltd. 600006 Dongfeng Automobile Co., Ltd. 600637 SVA Information Industry Co., Ltd. 600008 Beijing Capital Co., Ltd. 600639 Shanghai Jinqiao Export Processing Zone Development Co., Ltd. 600009 Shanghai International Airport Co., Ltd. 600642 Shenergy Company Limited 600011 Huaneng Power International, Inc. 600648 Shanghai Wai Gaoqiao Free Trade Zone Development Co., Ltd. 600016 China Minsheng Banking Corporation Limited 600649 Shanghai Municipal Raw Water Co., Ltd. 600018 Shanghai Port Container Co., Ltd 600653 Shanghai Shenhua Holdings Co., Ltd. 600019 Baoshan Iron & Steel Co., Ltd. 600663 Shanghai Lujiazui Finance & Trade Zone Development Co., Ltd. 600026 China Shipping Development Company Limited 600664 Harbin Pharmaceutical Group Co., Ltd. 600028 China Petroleum & Chemical Corporation 600679 Phoenix Co., Ltd. 600036 China Merchants Bank Co., Limited 600688 Sinopec Shanghai Petrochemical Company Limited 600050 China United Telecommunications Corporation Limited 600690 Qingdao Haier Co., Ltd. 600072 Jiangnan Heavy Industry Co., Ltd. 600695 Shanghai Dajiang (Group) Stock Co., Ltd. 600098 Guangzhou Development Industry (Holdings) Co., Ltd. 600726 Heilongjiang Electric Power Company Limited 600100 Tsinghua Tongfang Co., Ltd. 600736 Suzhou New District Hi-Tech Industrial Co., Ltd 600104 Shanghai Automotive Co., Ltd 600737 Xinjiang Tunhe Investment Co., Ltd. 600115 China Eastern Airlines Corporation Limited 600776 Eastern Communications Co., Ltd. 600171 Shanghai Belling Corp., Ltd 600799 Heilong Jiang Clever Net Corp., Ltd. 600188 Yanzhou Coal Mining Company Limited 600808 Maanshan Iron & Steel Company Ltd 600236 Guangxi Guiguan Electric Power Co., Ltd. 600816 Anshan Trust & Investment Co., Ltd 600332 Guangzhou Pharmaceutical Company Limited 600832 Shanghai Oriental Pearl Co., Ltd 600350 Shandong Infrastructure Company Ltd. 600839 Sichuan Changhong Electric Co., Ltd 600362 Jiangxi Copper Company Limited 600863 Inner Mongolia MengDian HuaNeng Thermal Power Co., Ltd 600377 Jiangsu Expressway Company Limited 600867 Tonghua Dongbao Medicines Co., Ltd 600548 Shenzhen Expressway Company Limited 600871 Sinopec Yizheng Chemical Fibre Company Limited 600585 Anhui Conch Cement Company Limited 600874 Tianjin Capital Environmental Protection Company Limited 600598 Heilongjiang Agriculture Company Limited 600886 SDIC Huajing Power Holdings CO., Ltd 600600 Tsingtao Brewery Company Limited 600890 Cred Holding Co., Ltd

128

Figure 4.1 Intraday 5-minute and interday bid/ask spread patterns and bid/ask depths

.0045 .0017950

.0040 .0017900

.0035 .0017850

.0030 .0017800

.0025 .0017750

.0020 .0017700

.0015 .0017650 0 4 8 12 16 20 24 28 32 36 40 44 48 M T W T F

Intraday pattern of spread Interday pattern of spread

570000 600000

560000 500000

400000 550000

300000 540000

200000 530000

100000 520000

0 510000 0 4 8 12 16 20 24 28 32 36 40 44 48 M T W T F

Intraday pattern of depth Interday pattern of depth

129

Chapter 5

Intraday Return Volatility, Trading Volume and Bid/Ask

Spread: GARCH Model Application

5.1 Introduction

Most financial time series show time-varying variance (volatility) and do not follow a normal distribution. The Generalized AutoRegressive Conditional

Heteroskedasticity (GARCH) model has been widely used in the financial time series because it takes into account the role of time-varying variance (excess kurtosis such as fat tail behaviour and volatility clustering). The focus of this chapter is on investigating empirically dynamic characteristics of intraday return volatility using the AR(1)-GARCH(1,1) model in a set of twenty-six individual stocks. The trading volume or bid/ask spread are used as a mixing variable for information arrival in the conditional variance model. The main purpose of the chapter is to examine if the intraday return volatility can be described by a AR(1)-GARCH(1,1) specification, and if trading volume and bid/ask spreads can capture the GARCH effects. Attempts are made to add to the existing rich literature on the GARCH model by examining the intraday volatility in the Chinese stock market.

130 Although the GARCH methodology has been used extensively in modelling financial time series, there are only a few applications of the GARCH model on intraday data for individual stocks (Rahman, Lee and Ang, 2002 and Darrat, Rahman and Zhong, 2003). Rahman, Lee and Ang (2002) have applied the GARCH model to the NASDAQ market using thirty individual stocks based on 5-minute intraday data.

Their results show that the persistence in volatility remains in the intraday return series even after trading volume or bid/ask spread is included in the model as an explanatory variable, which is consistent with the results of Najand and Yung (1991) and contrary to those of Lamoureux and Lastrapes (1990).

Furthermore, though there are some empirical studies (for example, Andersen and Bollerslev, 1997 and Mian and Adam, 2001) on the intraday volatility process for developed, highly liquid stock markets in industrial countries, no attention has been given to this issue for emerging markets. Empirical study on intraday volatility clustering for emerging markets in this chapter is mainly motivated by the following reasons.

First, as some of the emerging stock markets were newly established, the exchange used state-of-the-art technology for its trading platforms and included insights from financial research into the design and regulation of the trading process, these markets provide natural experimental areas to research (Bohl and Henke,

2003).

Second, China’s stock exchanges are new amongst the world’s financial markets and their trading system is different from those in other countries as documented in Section 2.6 of Chapter 2. Thus, the order-driven market with call auction and continuous matching of trading system in China has attracted great

131 attention in empirical and academic research, which provides a new platform for the market microstructure study.

Third, the study employs a set of individual stocks, which provides an opportunity to examine both the micro and macro behaviours of stock returns.

Trading volumes and bid/ask spreads are determined by both firm specific factors and market wide factors for individual stocks. Thus, intraday trading volume and bid/ask spread become a good proxy for information that contributes to conditional heteroscedasticity for individual stocks.

Fourth, there are no empirical studies on the microstructure of the Chinese stock market using intraday data1. Even in the US and European capital markets, most studies are based on daily data in application of the GARCH model, and there are only a few research studies based on the intraday data.

Hence, this chapter provides initial evidence for one of the developing stock markets in Asia, China, which is an order driven electronic stock market without market makers. The data includes information on the most actively traded stocks. It contains the transaction data, best bid and ask prices and their corresponding order volumes at 5-minute intervals during the three-year period .The intraday relationships between volatility and trading volume, and volatility and bid/ask spread of Shanghai stocks are examined to determine whether or not there are differences between developed markets and emerging markets in terms of this relationship. Since the Chinese stock market is developing rapidly, a comprehensive analysis with a microstructure approach on the stock return volatility process will not only make a contribution to academic literature on microstructure theory but also benefit investors and policy makers.

1 Liu et al (2000) performs an intraday research on the Chinese stock market but in Chinese.

132 In this chapter, the GARCH model is applied to a set of individual Shanghai stock returns. The empirical analysis is done in three steps. First, the chapter examines the GARCH model without the trading volume and bid/ask spread in the variance equation to determine: (1) if the GARCH model captures the

ARCH/GARCH effects, that is, if the GARCH model can capture the temporal dependence (volatility clustering) in the high-frequency data when information arrives in clustering, and (2) if the ARCH/GARCH effects are significant in the volatility equation. To investigate how these effects change during the trading day, individual stocks are examined based on 5-minute intraday data.

Second, based on the information model of the Mixture of Distributions

Hypothesis (MDH) suggested by Clark (1973) and Lamoureux and Lastrapes (1990), the chapter investigates (1) whether contemporaneous or lagged-one intraday trading volume can be used as information variables in the volatility model and (2) whether the ARCH/GARCH effects disappear when the information variable of trading volume is added in the volatility model. Following the Raman et al. (2002) method, examinations are made for the volatility–volume relationship when the volume variable is in the volatility equation.

Third, a similar analysis is carried out on the volatility-bid/ask spread relationship when contemporaneous or lagged-one intraday bid/ask spreads are considered as information variables to remove the ARCH/GARCH effects.

The rest of the chapter is organized as follows. The following Section discusses the institutional characteristics and data in the Chinese stock market.

Section 5.3 presents the research hypothesis and methodology. Section 5.4 provides a statistical analysis and GARCH model identification. Section 5.5 presents the

133 empirical results and discusses the findings and their implications. The final Section concludes the chapter.

5.2 Background and Data

5.2.1 Background

In financial markets, most time series have a distribution with a fatter tail than the normal distribution, and exhibit strong time-varying variance, volatility clustering and volatility persistence. This is a logical application of the autoregressive conditional heteroskedasticity (ARCH) model introduced by Engle

(1982) and Generalized ARCH model by Bollerslev (1986). The GARCH model allows past conditional variances to be used in the current conditional variance equation.

Studies on the volatility of stock returns have been dominated by time series models of conditional heteroscedasticity and have found strong support for

ARCH/GARCH effects. These findings are important to the field of applied finance for at least three reasons. First, the estimated return variances are used as risk measures and enter directly into Black–Scholes’ derivative pricing formulas. Second, heteroscedasticity must be taken into account for tests of market efficiency to produce reliable test statistics. Third, most asset pricing theories relate expected returns to the joint second-order movements of returns as well as to other stochastic processes; therefore, efficient estimating and testing must take into account the heteroscedasticity property of returns.

134 The dynamics intraday return volatility model has attracted a large amount of market microstructure research. This is because the arrival of news and the resolution of its informational impact can be directly related to the dynamics of the return volatility process. More recently, there has been an emphasis on inter-temporal dependence models to explain empirical observations of volatility clustering.

As many studies have found evidence in favor of ARCH/GARCH effects in stock returns, there is no consensus on the underlying economic explanations for the autoregressive effect on the conditional variance. One of the possible theoretical explanations is the Mixture of Distributions Hypothesis (MDH) put forward by Clark

(1973), Epps and Epps (1976), and Tauchen and Pitts (1983) and more recently by

Lamoureux and Lastrapes (1990). Among them, the MDH model suggested by Clark

(1973) plays a prominent role in the empirical finance literature. The MDH argues that the variance of returns at a given interval is proportional to the rate of information arrival. As a result, volatility clustering could be a reflection of the serial correlation of information arrival frequencies. All traders simultaneously receive the new price signals and the shift to a new equilibrium is immediate and there will be no intermediate partial equilibrium2. According to the MDH, a serially correlated mixing variable measuring the rate at which information arrives at the market explains the ARCH/GARCH effects in the returns.

Lamoureux and Lastrapes (1990) were the first to investigate the GARCH effects in stock returns by adding the trading volume in the GARCH model. They use trading volume as the information variable to eliminate the GARCH effects in the model. Their results are based on the daily data of 20 individual stocks in the US market. Following Lamoureux and Lastrapes, a number of studies provide empirical

2 Alternative explanations include the sequential information arrival hypothesis (Copeland, 1976 and Smirlock and Starks, 1985), the existence of autocorrelation in the news arrival process (Diebold and Nerlove, 1989), and market microstructure effects (Bollerslev and Domowitz, 1991).

135 evidence concerning the relationship between return volatility and information flow in the U.S. stock market (Najand and Yung, 1991; Kim and Kon, 1994; Andersen,

1996 and Gallo and Pacini, 2000) and in other stock markets (Melvin and Yin, 2000;

Omran and McKenzie, 2000 and Miyakoshi, 2002). In general, empirical studies has found evidence that the inclusion of trading volume in GARCH models for returns results in a decrease of the estimated persistence, or even causes it to vanish.

Given that financial markets display high speeds of adjustment, studies based upon daily observations may fail to capture information contained in intraday market movements. The dynamic behaviour of the intraday return volatility in the application of high-frequency data has received much attention in using ARCH and

GARCH models. The question is whether the use of high-frequency data has yet had much particular influence on the study of volatility. Andersen and Bollerslev (1997) summarized that the studies on dynamics of intraday return volatility in time series built on the ARCH methodology fall into one of three categories. First, some literature investigates the interrelation between returns in geographically separated financial markets that trade sequentially, with a focus on the transmission of information as measured by the degree of spill-over in the mean returns and/or volatility from one market to the others (Engle, Ito and Lin, 1990 and Hamao et al,

1990). Second, some literature is concerned with the lead-lag relations between two or more markets that trade simultaneously (Baillie and Bollerslev, 1991 and Chan et al, 1991). Third, others explore the role of information flow and other microstructure variables as determinants of intraday return volatility (Bollerslev and Domowitz,

1993 and Locke and Sayers, 1993).

Andersen and Bollerslev (1998) characterize the relative explanatory power of intraday patterns, macroeconomic announcements and long-term volatility factors

136 for intraday and daily return volatility in the foreign exchange market. Mian and

Adam (2001) study the behaviour of volatility for intraday high-frequency returns in the Australian market. They find that volatility of Australian equities follows an L- shaped curve over the trading day that is distinct from the U-shaped pattern commonly documented by previous studies on other markets. They further suggest that the GARCH model remains useful in capturing volatility clustering for high- frequency returns.

5.2.2 Data

In this chapter, a set of individual stocks is used for the application of the

GARCH model. The trading period covers in three years: from January 1, 2000 to

December 31, 2002. Data were adjusted for stock splits and provided by the SIRCA3.

Lamoureux and Lastrapes (1990) used twenty individual stocks and Rahman, Lee and Ang (2002) used thirty individual stocks in their research. In this study, twenty- six individual stocks are used for GARCH application. The criterion for the selection of these twenty-six stocks is based on trading activity in order to reduce problems due to errors of measurement. These individual stocks collectively represent some of best trading value, best trading volume stocks in 2003 from the Shanghai 180 index, which consists of the 180 blue chip stocks. Stocks from the Shanghai Stock

Exchange were selected considering that the Shanghai market is more recognised by investors, both local and overseas, and is also relatively larger then the Shenzhen market, just like NYSE compared with NASDAQ.

3 I am grateful to Mr. Patrick Huang from the SIRCA for providing us with all sixty individual stock data series.

137 The sample of the twenty-six individual stocks records information on time- stamped transactions. This information includes the code, the order of time interval, the trading date, the day, trading time, open value, highest value, lowest value, close value, trading volume, bid price, bid volume, ask price, and ask volume in each 5- minute interval in each trading day.

Return series in 5-minute intervals t are calculated by rt = ln (pt / pt-1), where

p and p represent the last price on interval t and the last price on interval t-1, t t−1 respectively. Due to significant changes in the level of prices from time to time, it is more appropriate to base volatility measures on percentage return as just discussed, rather than absolute price movements4. From an investment point of view, it is also obvious that comparing rates of return is more meaningful than comparing absolute movements (Rahman, Lee and Ang, 2002).

The trading volume series in 5-minute intervals t is calculated by taking the natural logarithm of the individual stock, that is, volt = ln (volumet), where volumet is the trading volume of the interval t. The reason for using the natural logarithm of trading volume is to improve the normality of the series in order to better fit into the

GARCH model estimated in a later Section of this chapter5. This process enables us

4 Another consideration is that data series incorporates the effect of the intraday periodicity (intraday pattern). This could standardize the data series for the analysis (Andersen and Bollerslev, 1997). I tried a de-seasonalising treatment of these data and estimated the GARCH family model based on these de-seasonalised data, but the results showed no significant improvement from the original data without de-seasonalisation. A further consideration is that of 10% price limits in the Chinese stock market. Price limit is used to reduce stock market volatility but no individual stock with 5-minute return in my sample reaches this 10% limit. So there is no major influence on this empirical study. However, some researchers (Kyle, 1988 and Kim, 2001) argue that price limits may increase the stock volatility, and this needs further research. 5 The descriptive statistics of intraday 5-minute raw volumes and spread will be available upon request. I also estimated the GARCH-cum-volume model by including the raw volume series into it compared with the results when a log-volume series is included. I found the results of GARCH (1,1) with log-volume are better than those with a raw volume series, which provides further evidence that log-volume series fit the GARCH (1,1) model best. The GARCH-cum-spread model has similar results.

138 to stabilize the variance of the error terms, and approximate error terms toward a symmetric distribution.

The intraday bid/ask spread series in 5-minute intervals t is calculated by as

ask pricett− bid price spdt =×ln(2 ) (5.1) ask pricett+ bid price

where, for each interval of the trading day, ask pricett and bid price are the ending values of the interval t. The reason for the bid/ask spread being transformed into log form is the same as for the trading series.

In the papers of Stoll and Whaley (1990) and Rahman, Lee and Ang (2002), the first two 5-minute returns of the trading day are excluded from the analysis. The reasons are considered that first, Stoll and Whaley (1990) reported that the average opening time (i.e., the average time elapsed between the exchange opening and opening transaction) for stocks in the S&P 500 Index is between five and seven minutes. Prices during these intervals may reflect the stale closing price of the previous day. Andersen and Bollerslev (1997) also reported that the first 5-minute return for the trading day constitutes an unusual time interval, which incorporates the adjustments to the information accumulated overnight and consequently displays a much higher than average return variability compared to most other intervals.

Second, trading on the Shanghai Stock Exchange starts at 09:15, with the call auction market for the next 10-minutes. The information on bid prices and ask prices, as well as the corresponding volumes, is not available to the public in the morning call auction process. Thus very few people participated in call auctions; even in the first 10-minutes of continuous trading after 09:30 due to the larger price variation.

139 Normal risk-averse investors will wait a few minutes until the trading volume has picked up. Therefore, the first 10-minutes in Shanghai stock trading largely reflects the adjustments to information accumulated overnight, and consequently displays a much higher than average return variability compared to any other 5 minute interval.

Following the methods of Stoll and Whaley (1990) and Rahman, Lee and

Ang (2002), the first two 5-minute returns of each trading day, 09:30-09:40 were excluded from the analysis in this chapter. This leaves us with a sample of each trading day consisting of forty-six intraday 5-minute returns. Therefore, disregarding the first two 5-minute return observations each day mitigates the effects of the stale price information. In addition, because returns are computed within each day using only intraday prices, overnight returns are not included in any of these series.

5.3 Research Hypotheses and Methodology

5.3.1 Hypotheses

A substantial amount of research on both theoretical and empirical aspects shows evidence in favour of the volatility clustering in stock return series. Engle, Ito and Lin (1990) suggest two possible explanations for the volatility clustering.

First, if information arrives in clusters, returns volatility may exhibit clustering. Nominal interest rate, dividend yield, money supply, oil price, margin requirement, business cycles, and information patterns are the sources of volatility clustering.

140 Second, if traders have different prior beliefs and take time to digest the information shocks and resolve the expectation differences, market dynamics can lead to volatility clustering. Empirical researchers have found that the GARCH model can capture the GARCH effects in the other countries’ markets (Engel, 1982;

Lee, Chen and Rui, 2001; Mian and Adam, 2001 and Rahman, Lee and Ang, 2002).

Thus, the hypothesis is made that the GARCH model can capture the temporal dependence (volatility clustering) in high-frequency data for the Chinese stock market.

Hypothesis 1:

H0A: The GARCH model can capture the temporal dependence

(volatility clustering) in 5-minute data.

H1A: The GARCH model cannot capture the temporal dependence

(volatility clustering) in 5-minute data.

An increased trading volume and/or variability of prices following the arrival of new information into the market could lead to the occurrence of time-dependent conditional heteroskedasticity. Clark (1973) and Lamoureux and Lastrapes (1990) suggest the MDH and recognize this aspect of conditional volatility in their analysis of daily returns. According to the MDH, a serially correlated mixing variable measuring the rate at which information arrives to the market explains the GARCH effect in the returns, thus volatility is closely related to the rate of information arrival into the market. While there is a substantial amount of empirical research on the relationship between the volatility and trading volume (Copeland, 1976; Epps and

Epps, 1976; Kim and Kon, 1994; Andersen, 1996 and Rahman, Lee and Ang, 2002),

141 no study in the current literature, to my knowledge, has been made in using the high- frequency data in the Chinese stock market. Therefore, the second hypothesis is made that the GARCH model can eliminate the GARCH effects in high-frequency data in the Chinese stock market while trading volume as a mixed variable for information arrival.

Hypothesis 2:

H0B: The volume as an information variable in the GARCH model

can remove the GARCH effect.

H1B: The volume as an information variable in the GARCH model

cannot remove the GARCH effect.

The relationship between the return volatility and variations of bid/ask spreads has also been reported in the literature (Copeland and Galai, 1983; Grossman and Miller, 1988; McInish and Wood, 1992; Walsh and Quek, 1999; Wang and Yau,

2000 and Rahman, Lee and Ang, 2002). The sizes of bid/ask spreads are thought to carry information about the existence of information flow and bid/ask spreads tend to become wider as the probability increases of informed agent trading. Thus, the third hypothesis is that the bid/ask spread is another measure of information that flows into the mark et and the GARCH model can eliminate the GARCH effects in high- frequency data for the Chinese stock market.

142 Hypothesis 3:

H0C: The bid/ask spread as an information variable in the GARCH

model can remove the GARCH effect.

H1C: The bid/ask spread as an information variable in the GARCH

model cannot remove the GARCH effect.

5.3.2 Methodology

First used by Engle (1982), the ARCH model models the mean and the variance of a time series. Engle found that in many time series, particularly those involving the financial data, large and small residuals tend to come in clusters, suggesting that the variance of an error may depend on the size of the preceding error. The ARCH(q) model specifies that the conditional variances are dependent on elements in the information set in an autoregressive manner, while the conditional mean equation can be any form of ARMA model. The variable q represents lags of

the squared residuals. An example of the ARCH model can be as follows: let rt denote the return of stock, then, specifically, the AR(1)-ARCH(q) process assumes that

rbbra tt=+011− +t

awttt==σε,.here ε tii.d.N(0,1) (5.2) q 2 a2 σαt0=+ α it−i i1∑=

143

where {εt} is a sequence of independent and identically distributed random variables with mean zero and variance 1, 2 is a conditiona l variance, a2 equals squared σ t ti−

innovations, α0 > 0 and α1,…, αq ≥ 0. In practice, at is often assumed to follow the standard normal distribution while the distribution function for the asset return is fat- tailed in the most financial asset.

2 By using the ARCH model, a series of squared residuals at can be used to derive the conditional variance. Unlike the OLS assumption of a constant variance of

{ at }, the ARCH model assumes that { at } has a non-constant variance or heteroscedasticity, denoted by σt. The problem in applying the ARCH model is the non-negativity constraint on the coefficient parameters of {αt} to ensure the positivity of the conditional variance.

To avoid the problem of long lag structure and the non-negativity of the

ARCH(q), Bollerslev (1986) presents a generalized ARCH model, the GARCH(p,q), which allows the current conditional variance to be a function of past conditional variances as well. In the GARCH model, a lagged conditional variance introduces long memory to the ARCH model. The dependence in the second moments is made explicit by the introduction of parameters that allow for the conditional variance

(conditional on its past) to vary through time and parameters that allow the variance to depend on its near past. Thus, the GARCH(p,q) specifies the conditional variance to be a linear combination of (q) lags of the squared residuals ( a2 ) from the t−1 conditional returns Equation (5.2) and (p) lags from the conditional variance ( 2 ). σ t−1

144 An example of the following generalized autoregressive conditional heteroscedasticity AR(1)-GARCH(p,q) model of order p, q with AR(1) is specified as follows:

rbbra tt=+011− +t

awttt==σε,.here ε tii.d.N(0,1) (5.3) qp 2 a22 σαt0=+∑∑ α it−−i + βσ jtj i1== j1

2 where σ t is the conditional variance (volatility) of at at time t, The βj coefficients

capture the persistence of volatility over time. α0 is a constant and αi measures the presence of volatility clustering. This model is viewed as a reduced form of more complicated dynamic structures for time varying conditional second order moments.

When the parameter p is defined as zero, the GARCH model is named as the generalized version of the ARCH model.

This chapter focuses on the estimation for a set of twenty-six stock series using the AR(1)-GARCH(1,1) model. Bollerslev (1987), Engle (1993) and

Alexander (2001) have shown that the GARCH(1,1) model would adequately fit many high-frequency time series. As statistical analysis, reported later in Table 5.2, indicates, the 5-minute returns exhibit statistically significant serial correlation at the first lag, thus an AR(1) structure is fitted on the returns6. The conditional mean and conditional variance equations of the AR(1)-GARCH(1,1) model are defined as:

6 Some return series actually need higher lag components, such as AR(2) and AR(3) to remove the autocorrelation; the estimate results, however, are not reported in Table 3 and may be obtained by request.

145 rbbra tt=+011− +t

awttt==σε,.here ε tii.d.N(0,1) (5.4) 2 a22 σααt01=+t−−11 + βσ 1t

where αα01 > 0, ,β1≥+ 0; and αβ 11 < 1

The residual series, at , can be expected to be uncorrelated. The main interest in this

chapter is to model the variance of a return innovation, at , not the return itself, with the GARCH process. In the model, α1 reflects the influence of random deviations in the previous period on σ t , whereas β1 measures the part of the realized variance in the previous period that is carried over into the current period. The sizes of the parameters α1 and β1 determine the short-run dynamics of the resulting volatility time series. Large GARCH error coefficients, α1, mean that volatility reacts intensely to market movements. Large GARCH lag coefficients, β1, indicate that shocks to conditional variance take a long time to die out, so volatility is ‘persistent’. If α1 is relatively high and β1 is relatively low then volatilities tend to be more ‘spiky’

(Alexander, 2001).

As discussed in the previous Section, the occurrence of time-dependent conditional heteroskedasticity could be due to an increased volume of trading and/or variability of prices following the arrival of new information into the market.

Lamoureux and Lastrapes (1990) used trading volumes as measures for the amount of information that flows into the market and empirically verified that including volume as an additional variable in the conditional variance equation leads to a very significant decrease in the autoregressive coefficient of the variance equation, i.e., α

+ β of the GARCH process, to almost zero. However, Lamoureux and Lastrapes

(1990) also recognized that the use of contemporaneous trading volumes to explain

146 volatility raises the issue of simultaneity bias, in that the volume will not be an exogenous variable. To solve this problem, Najand and Yung (1991) use a lagged volume variable in their volatility equation. Thus, the hypothesis is made in this study that the sizes of trading volume and bid/ask spreads are thought to carry information about the existence of information flow, so an information arrival would be also expected to affect market volatility. Trading volume, lagged volume, bid/ask spread and lagged bid/ask spread are used in the conditional variance equation to the

GARCH model.

The adequacy of the GARCH model can be examined by the standardized

at residuals, ( at = ), where (σ t ) is the conditional standard deviation as calculated σ t

by the GARCH model, and at is the residual of the conditional return Equation (5.4).

The Ljung-Box-Pierce Portmanteau Q-statistic at lag can be used to check at for the

2 adequacy of the mean equation and at for the adequacy of the variance equation. If the Ljung-Box-Pierce Portmanteau tests (Q(m) and Q2(m)) of the standardized residuals and squared standardized residuals are found to be insignificant, the AR(1)-

GARCH(1,1) model can fully account for the linear dependencies (serial correlation) and nonlinear dependencies (volatility clustering) in the intraday return series.

Under the assumption that trading volume and bid/ask spread can be used as mixing variables, these variables can be incorporated into the variance equation of the GARCH model. Four alternative augmente d GARCH models of volatility - one with contemporaneous trading volumes, the second with lagged trading volumes, the third with contemporaneous bid/ask spread, the fourth with lagged bid/ask spread are estimated in the study:

147 with contemporaneous trading volume series:

2 a22log( Volume ) σααt01=+t1t1−−11 + βσθ + t (5.5) with ααβ011 >0, ,≥+ 0; and αβ 11 < 1

with lagged-one trading volume series:

2 a22log( Volume ) σααt01=+t1t1−−11 + βσθ + t −1 (5.6) with ααβ011 >0, ,≥+ 0; and αβ 11 < 1

with contemporaneous bid/ask spread series:

2 aS22P σααt01=+t−−11 + βσγ 1t1t + (5.7) with ααβ011 >0, ,≥+ 0; and αβ 11 < 1

with lagged-one bid/ask spread series:

2 aS22P σααt01=+t−−11 + βσγ 1t1t + − 1 (5.8) with ααβ011 >0, ,≥+ 0; and αβ 11 < 1

If the MDH is relevant in explaining the ARCH/GARCH effect of stock returns, then the inclusion of volume or bid/ask spread series in the conditional variance should absorb volatility persistence in the conditional variance process of the GARCH(1,1). In Equation 5.4, the volatility persistence is captured by the sum of the coefficients α1 + β1. The greater this sum, the greater is the persistence of shocks to volatility. Since information flow explains the variance persistence represented and approximated by the ARCH/GARCH effect, if the trading volume or bid/ask spread is incorporated into the model, it is expected that θ > 0 or γ > 0, and α1 and β2 should become smaller and insignificant due to this inclusion in Equations 5.5, 5.6,

5.7 and 5.8.

148 5.4 Statistical Analysis and Model Identification

Before the application of the GARCH model, in this section, I discuss the essential features of time series of return, log-volume and bid/ask spread for the set of twenty-six individual stocks. First, a statistical summary of data series is provided.

Second, the return and the squared return of serial correlation and ARCH effects are examined. Third, the stationarity condition is verified through the augmented

Dickey-Fuller test.

5.4.1 Return Series

Appendix 5.1 provides the stock identification numbers, trading codes and names of the twenty-six individual stocks. The trading period covers the three years: from January 1, 2000 to December 31, 2002, a total of 716 trading days from the

Shanghai Stock Exchange.

Table 5.1 provides the statistical properties of all twenty-six stock intraday return series. All the return series of all these selected companies display a significant departure from normality as reflected by the extreme excess kurtosis. The intraday mean returns of twenty-six individual stocks in the Shanghai market range from –0.00359% to 0.00881% and close to zero. Seventeen out of twenty-six return series are not statistically different from zero at the 5% level, except for ID 3, 4, 6,

11, 16, 19, 21, 22 and 25. This indicates the data of individual stocks are basically reasonable. The statistics show that all twenty-six return series are positively skewed, ranging from 0.302 to 1.519. The positive skewness indicates a relatively long right

149 tail compared to the left, which implies that the intraday return distributions of the stocks traded on the Chinese market have a higher probability of earning positive returns. In addition, all the kurtosis values are much larger than 3, ranging from

10.655 to 36.444. This shows that for all series, the distribution of the intraday return has fat tails compared with the normal distribution. This behavior is known to frequently occur in financial markets (for example, Andersen and Bollerslev, 1997 and Mian and Adam, 2001). The Jarque-Bera test statistics are found to have high values and be significant at the 1% level, which provides clear evidence to reject the null hypothesis of normality for the unconditional distribution of all intraday 5- minute returns series. Though the interquartile ranges (0.169% - 0.312%) are small, the returns are clearly not normally distributed.

Table 5.2 provides statistics tests for all twenty-six stock intraday return series, including Ljung-Box-Pierce Portmanteau test for the return series and squared return series, first-order autocorrelation, maximum likelihood function value when a normal distribution is fitted to the data, and Engle's ARCH-test. The first-order autocorrelations (Rho) of most series are large and statistically significant. Nearly half of the twenty-six individual stocks have negative first-order autocorrelations.

This highly significant autocorrelation may be explained by a non-synchronous trading effect in the financial market (Chan et al. (1991) and Lo and MacKinlay

(1990)). The values of Q(10) and Q(20) in Table 5.2 are the Ljung-Box-Pierce

Portmanteau tests for the first ten and twenty order serial correlations in the intraday returns, respectively. The Q(10) and Q(20) test statistics reject the null hypothesis of uncorrelated price changes at the 1% significance level, suggesting a slowly decaying autoregressive effect. Thus, the null hypothesis of strict white noise is rejected in almost all cases. These findings are comparable to those documented by

150 Mian and Adam (2001) for the Australian market and Rahman, Lee and Ang (2002) for the NASDAQ market.

Many financial time series exhibit volatility clustering, that is, autoregressive conditional heteroscedasticity. These can imply a strong autocorrelation in squared returns. The ARCH-test was also performed before an estimation of the GARCH model. The values of Q2(10) and Q2(20) in Table 5.2 are the Ljung-Box-Pierce

Portmanteau tests for the first ten and twenty order serial correlations in the squared intraday returns, respectively. The values of the Q2(10) and Q2(20) test statistics of all twenty-six individual stocks are large and statistically significant at the 1% level and indicate a strong persistence for intraday volatility and existence of the ARCH effect. The slow decline of the autocorrelation function of the squared intraday returns suggests that a GARCH(1,1) process may be adequate for describing the errors, although a low order ARCH process may not fully capture the time-varying volatility in the data.

The results of Engle’s ARCH LM-test for all return series are also presented in Table 5.2. ARCH effects are tested on lags 1-10 and 1-20. This test supports the existence of heteroscedasticity. With these high values of LM, the hypothesis that no

ARCH effects exist is rejected. Thus, the LM tests for ARCH(10) and ARCH(20) errors further confirm the presence of ARCH effects for all of the return series.

The augmented Dickey-Fuller tests reported in Table 5.3 show that all return series are free of unit root problems, which means that all series are stationary. The null hypothesis of a unit root is rejected at the 1% significance level for all return series. These wide ranges of statistics and tests provide a conclusive rejection of the hypothesis that the 5-minute return series are strict white noise processes.

151 In summary, the return series are characterized by inter-temporal dependence in both mean and variance. These results are evidence that the 5-minute returns tend not to be independent but to exhibit volatility clustering. Thus, the data display all the previously documented characteristics of the unconditional distribution of returns that are used to justify the GARCH specifications.

5.4.2 Trading Volume Series

In addition to the return time series, the data set contains time series on trading volumes for the twenty-six stocks. Table 5.4 provides descriptive statistics of

5-minute log-volume series of all twenty-six individual stocks. These statistics include the following distributional parameters: size, mean, variance, skewness, kurtosis, Jarque-Bera test to a normal distribution, maximum, minimum and interquartile range (IQR). The first two 5-minute trading volumes have been removed in the series for the same reasons as in the return series. The means of the trading volume series of twenty-six individual stocks in the Shanghai market range from 7.532 to 10.420. This indicates the data of individual stocks are reasonable while considering the variances of the intraday trading volume series. The statistics show that the values of the intraday trading volume series are negatively skewed ranging from -6.607 to -3.689. In addition, all the kurtosis values are larger than 3, ranging from 16.330 to 76.858. This shows that the distributions of the intraday trading volume series have fat tails compared with the normal distribution. The

Jarque-Bera test statistics are also found to have high values and be significant at the

1% level and provide clear evidence to reject the null hypothesis of normality for the

152 unconditional distribution of all the intraday 5-minute trading volume series. Thus, the intraday 5-minute trading volume series are not normally distributed, although the results are much better than those of the raw volume series.

The ARCH effect for the log-volumes is also examined in Table 5.5, suggesting the clear presence of ARCH effects for all log-volume series. The autocorrelation coefficients and the Ljung-Box-Pierce Portmanteau statistics indicate that all log-volume time series show serial correlation and are significant at the 1% level. These show that the flow of information arrival is serially correlated when it is measured by the log-volume. Similar to the return series in this study, and unlike most daily volume series in previous studies, the log-volume series are stationary, i.e. integrated of order 1, according to the Augmented Dickey-Fuller test; these results are also reported in Table 5.3.

5.4.3 Bid/Ask Spread Series

Table 5.6 provides descriptive statistics of the 5-minute log bid/ask spread series of all twenty-six individual stocks. These statistics include the following distributional parameters: size, mean, variance, skewness, kurtosis, Jarque-Bera test to a normal distribution, maximum, minimum and interquartile range. Again, the first two 5-minute bid/ask spreads have been removed in the series for the same reasons as in the return series. The means of the bid/ask spread series of twenty-six individual stocks in the Shanghai market range from –7.018 to –6.29. This indicates the data of individual stocks are reasonable while considering the variances of the intraday bid/ask spread series. The statistics shows that the intraday bid/ask spread

153 series are mostly positively skewed, ranging from 0.462 to 2.364. In addition, most of the kurtosis values are larger than 3, ranging from 2.427 to 9.566. This shows that the distributions of the intraday bid/ask spread series have fat tails compared with the normal distribution. The Jarque-Bera test statistics are also found to have high values and be significant at the 1% level and provide clear evidence to reject the null hypothesis of normality for the unconditional distribution of all the intraday 5-minute bid/ask spread series. Thus, the intraday 5-minute bid/ask spread series are not normally distributed.

The ARCH effect for the bid/ask spread series is also examined in Table 5.7, suggesting the clear presence of ARCH effects for all bid/ask spread series. The autocorrelation coefficients and the Ljung-Box-Pierce Portmanteau statistics indicate that all bid/ask spread time series show serial correlation and are significant at the

1% level. These show that the flow of information arrival is serially correlated when it is measured by the bid/ask spread. Similar to the trading volume series in this study, all bid/ask spread series are stationary, which is also reported in Table 5.3.

5.5 Empirical Results

5.5.1 Estimation of GARCH M odel without Volume and Spread Series

This section provides empirical results for the twenty-six individual stock returns from the Shanghai Stock Exchange according to Equation 5.4. The parameters are estimated jointly using numerical techniques to quasi-maximize the

154 log-likelihood functions 7 . The iteration is carried out until convergence to the optimum is obtained. Table 5.8 includes the results of fitting a pure AR(1)-

GARCH ( 1,1) process to the 5-minute intraday return series.

First, the estimates of α0 are all positive and considerably smaller than the sample variances shown in Table 5.1. This is due to changing conditional variance over time and their eventual contribution to unconditional variances.

Second, all coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 1% level. The persistence in volatility as measured by the sum of

α1 and β1 , ranging from 0.8847 to 0.9925 w ith a average of 0.9514 for the twenty- six individual stocks, is closer to one, indicating a strong presence of the ARCH and

GARCH effects (Engle and Bollerslev, 1986) and greater persistence of shocks to volatility (Lamoureux and Lastrapes, 1990 and Najand and Yung, 1991). This result is consistent with the findings of Rahman, Lee and Ang (2002) with thirty stocks in the NASDAQ. The results imply that current volatility is relevant in predicting future volatility over a very long horizon.

Third, to examine how well the AR(1)-GARCH(1,1) model adequately captures the dependencies in the individual stocks, Table 5.8 provides the results of diagnostic tests on the standardized residuals for the twenty-six stocks after estimation of AR(1)-GARCH model. The Q(20) statistics indicate there is still serial correlation in the standardized residuals but Q2(20) statistics indicate no serial correlation in the squared standardized residuals for seventeen out of the twenty-six individual stocks. This suggests that the GARCH(1,1) model is successful in

7 Andersen and Bollerslev (1997) assume the innovations to be conditionally normally distributed in their applications. They further suggest that although it usually represents a reasonable approx imatio n , the GARCH(1,1) model is not necessarily the preferred specification for the return generating process . However, estimating the same model across different return series facilitates meaningful comparisons of the findings. Moreover, the theoretical aggregation results are available by the model estimation. I use this specification for all individual stocks in my sample.

155 explaining the time-varying volatility in the data, and can fully capture the volatility clustering in the intraday return series. Thus, the estimated models fit the data very well in seventeen individual stocks, which supports hypothesis 1.

Fourth, to further consider higher orders of GARCH (p, q) models, several different order of p and q in GARCH(p, q) models are estimated. These results are reported in Table 5.9. However, there is not much significant improvement in the linear and nonlinear dependencies as well as the value of log-likelihood function.

In summary, the persistence in volatility as measured by the sum of α1 and

β1 is larger and closer to one in the high-frequency return series. These results provide strong evidence that the 5-minute return series can be characterized by a

GARCH(1,1) specification. The model estimation indicates that the GARCH (1,1) model is moderately successful in accounting for the nonlinear dependencies

(volatility clustering) but not in accounting for the linear dependencies (serial correlation) in the Shanghai stock market.

5.5.2 Estimation of GARCH Model with Trading Volume

According to Lamoureux and Lastrapes (1990), if the trading volume series is serially correlated, adding the trading volume variable to the conditional volatility

equation causes the coefficients α1 and β1 to become statistically insignificant if the

coeffici en t θ1 of trading volumes is positive and significant. Table 5.10 provides the maximum likelihood estimation of coeff icients and asymptotic t-statistics of the volatility equation with log-volume (Equation 5.5).

156 First, in all twenty-six stocks estimated, the coefficients of log-volumes (θ1 ) except ID 9 were found to be positive and all are statistically significant at the 1% level. Thus, a positive relationship exists between intraday volatility and trading volume for all twenty-five individual stocks.

Second, all coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 1% level. The persistence in volatility as measured by the sum of

α1 and β1 ranges from 0.5231 to 0.9924 with an average of 0.8210 for the twenty-

six individual stocks. The average sum of α1 and β1 of the twenty-six individual stocks decreases from 0.9514 to 0.8210 compared with the Table 5.8. These findings

show a drop in the volatility persistence (α1 + β1 ) with the contemporaneous volume as the mixing variable.

These results are similar to those of Najand and Yung (1991), Foster (1995) and Rahman, Lee and Ang (2002), but differ from those of Lamoureux and Lastrapes

(1990) in that the GARCH effect remains strongly significant for the intraday return series. However, the average of coefficients of trading volumes is fairly small, only

0.0012, compared with 0.8210 for that of the sum of α1 and β 1 of the twenty-six individual stocks. Thus, volatility is better explained by previous volatility than by log-volume, and the results do not support hypothesis 2.

Third, the positive contribution of the contemporaneous trading volume variable in terms of its ability to explain the volatility can be noticed by comparing the log-likelihood of the estimations in Table 5.10 with Table 5.8. The average absolute value of log-likelihood function of the twenty-six individual stocks slightly increases from 4166.19 to 4424.43 due to the inclusion of the contemporaneous trading volume in the conditional variance equation.

157 Fourth, diagnostic tests on the standardized residuals are provided in Table

5.10. The Q(20) statistics indicate there is still serial correlation in the standardized residuals but Q2(20) statistics indicate no serial correlation in the squared standardized residuals for thirteen out of twenty-six individual stocks. The test results are not better than the GARCH estimation without log-volume (seventeen out of twenty-six). But the results can suggest that the GARCH(1,1) model has some success in explaining the time-varying volatility in the data, and can capture the volatility clustering in the intraday return series. Thus, the estimated models fit the data for the volatility clustering in the thirteen individual stocks, which supports hypothesis 1.

To sum up, the inclusion of intraday log-volume as a mixing variable for information arrival in the conditional variance model helps in explaining the

GARCH effects in the stock return series. But the GARCH effects do not disappear as a result of this inclusion. The average persistence in volatility as measured by the

sum of α1 and β1 has a large decrease from 0.9514 to 0.8210 compared with Table

5.8. The model estimation indicates that the GARCH (1,1) model has some success in accounting for the nonlinear dependencies (volatility clustering) but not in accounting for the linear dependencies (serial correlation) in the Shanghai stock market.

5.5.3 Estimation of GARCH Model with Lagged-one Trading Volume

Foster (1995) estimates two alternative GARCH models of volatility: one with contemporaneous trading volume and the other with lagged trading volume. He

158 finds that in no case does trading volume remove the GARCH effects, thus contradicting the results of Lamoureux and Lastrapes (1990) but supporting those of

Najand and Yung (1991). Table 5.11 provides the maximum likelihood estimate of coefficients and asymptotic t-statistics of the volatility equation with lagged-one log- volume (Equation 5.6). According to Lamoureux and Lastrapes (1990), if the lagged- one log-volume series is serially c orrelat ed, adding the lagged-one log-volume

variable to the conditional volatility equation causes the coefficients α1 and β1 to

become s tatistically insignificant if the coefficient θ1 of trading volumes is positive and significant.

First, in all twenty-six stocks estimated, the coefficients of lagged-one log-

volumes (θ1 ) were found to be positive except for two stocks (ID 9, 10, 15 21 and

24). All of the coefficients are statistically significant at the 1% level. Thus, a positive relationship exists between intraday volatility and lagged-one log-volumes for all twenty-one individual stocks.

Second, all coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 1% level. The persistence in volatility as measured by the sum of

α1 and β1 ranges from 0.7303 to 0.9912 with an average of 0.9365 for the twenty-

six individual stocks. The average sum of α1 and β1 of the twenty-six individual stocks slightly decreases from 0.9514 to 0.9365 compared with Table 5.8 but these

findings show an increase in the volatility persistence ( α1 + β1 ) with the contemporaneous log-volume as the mixing variable (0.8210).

These results are similar to those of Rahman, Lee and Ang (2002). The

GARCH effect remains strongly significant for the intraday return series. However, the average coefficient of lagged-one log-volumes is fairly small, only 0.0003,

159 compared with 0.9365 for that of the sum of α1 and β1 of the twenty-six individual stocks. Thus, volatility is better explained by previous volatility than by lagged-one log-volume, and the results reject hypothesis 2.

Third, the positive contribution of the lagged-one trading volume variable, in terms of its ability to explain the volatility, can be noticed by comparing the log- likelihood of the estimations in Table 5.11 with Table 5.8. The average absolute value of the log-likelihood function of the twenty-six individual stocks slightly increases from 4166.19 to 4210.52 due to the inclusion of the lagged-one trading volume in the conditional variance equation.

Fourth, diagnostic tests on the standardized residuals are provided in Table

5.11. The Q(20) statistics indicate there is still serial correlation in the standardized residuals but the Q2(20) statistics indicate no serial correlation in the squared standardized residuals for nineteen out of twenty-six individual stocks. The test results are better than the GARCH estimation without the trading volume (seventeen out of twenty-six) and with conte m poraneous log-volume (thirteen out of twenty- six). Thus, the results suggest that the GARCH(1,1) model is successful in explaining the time-varying volatility in the data, and can capture the volatility clustering in the intraday return series. Thus, the estimated models fit the data for the volatility clustering in the nineteen individual stocks, which supports hypothesis 1.

In summary, the inclusion of intraday lagged-one log-volume as a mixing variable for information arrival in the conditional variance model helps in explaining the GARCH effects in the stock return series. The GARCH effects do not completely disappear as a result of this inclusion. The average persistence in volatility as

measured by the sum of α1 and β1 has a slight decrease from 0.9514 to 0.9365 compared with Table 5.8. The model estimation indicates that the GARCH (1,1)

160 model is successful in accounting for the nonlinear dependencies (volatility clustering) but not in accounting for the linear dependencies (serial correlation) in the

Shanghai stock market.

5.5.4 Estimation of GARCH Model with Bid/Ask Spread

Rahman, Lee and Ang (2002) estimate two alternative GARCH models of volatility with bid/ask spread: one with contemporaneous bid/ask spread and the other with lagged bid/ask spread. They find that bid/ask spread does not remove the

GARCH effects, thus contradicting the results of Lamoureux and Lastrapes (1990) but supporting those of Najand and Yung (1991). Table 5.12 provides the maximum likelihood estimate of coefficients and asymptotic t-statistics of the volatility equation with log-volume (Equation 5.7). According to Lamoureux and Lastrapes

(1990), if the bid/ask spread series is serially correlated, adding the bid/ask spread

variable to the conditional volatility equation causes the coefficients α1 and β1 to become s tatistically insignificant if the coefficient γ of bid/ask spreads is positive and significant.

First, in all twenty-six stocks estimated, the coefficients of bid/ask spreads (γ) were found to be positive and statistically significant at the 1% level. Thus, a positive relationship exists between intraday volatility and bid/ask spread for all twenty-six individual stocks. This finding can be interpreted as indicating that an increase in liquidity, which narrows bid/ask spreads, will reduce return volatility and vice versa.

Second, all coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 1% level. The persistence in volatility as measured by the sum of

161 α1 and β1 ranges from 0.6927 to 0.9606 with an average of 0.8231 for the twenty-

six individual stocks. The average sum of α1 and β1 of the twenty-six individual stocks decreases from 0.9514 to 0.8231 compared with Table 5.8. These findings

show a drop in the volatility persistence (α1 + β1 ) with the contemporaneous bid/ask spread as the mixing variable.

These results are similar to those of Rahman, Lee and Ang (2002). The

GARCH effect remains strongly significant for the intraday return series. However, the average coefficient of bid/ask spreads is fairly small, only 0.0236, compared with

0.8231 for that of the sum of α1 and β1 of the twenty-six individual stocks. Thus, volatility is better explained by previous volatility than by bid/ask spread, and the results do not support hypothesis 3

Third, the positive contribution of the contemporaneous bid/ask spread variable in terms of its ability to explain the volatility can be noticed by comparing the log-likelihood of the estimations in Table 5.12 with Table 5.8. The average absolute value of the log-likelihood function of the twenty-six individual stocks slightly decreases from 4166.19 to 3906.21 due to the inclusion of the contemporaneous bid/ask spread in the conditional variance equation.

Fourth, diagnostic tests on the standardized residuals are provided in Table

5.10. The Q(20) statistics indicate there is still serial correlation in the standardized residuals b ut the Q2(20) statistics indicate no serial correlation in the squared standardized residuals for ten out of twenty-six individual stocks. The test results are not better than the GARCH estimation without log-volume (seventeen out of twenty- six). However, the results can suggest that the GARCH(1,1) model is partly successful in explaining the time-varying volatility in the data, and can capture the volatility clustering in the intraday return series. Thus, the estimated models fit the

162 data for the volatility clustering in the ten individual stocks, which supports hypothesis 1.

In summary, the inclusion of the intraday bid/ask spread as a mixing variable for information arrival in the conditional variance model helps in explaining the

GARCH effects in the stock return series. But the GARCH effects do not disappear as a result of this inclusion. The average persistence in volatility, as measured by the

sum of α1 and β1 , decreases from 0.9514 to 0.8231, compared with Table 5.8. The model estimation indicates that the GARCH (1,1) model is moderately successful in accounting for the nonlinear dependencies (volatility clustering) but not in accounting for the linear dependencies (serial correlation) in the Shanghai stock market.

5.5.5 Estimation of GARCH Model with Lagged-one Bid/Ask Spread

Table 5.13 provides the maximum likelihood estimate of coefficients and asymptotic t-statistics of the volatility equation with lagged-one bid/ask spread

(Equation 5.8). According to Lamoureux and Lastrapes (1990), if the lagged-one bid/ask spread series is serially correlated, adding the lagged-one bid/ask spread

variable to the conditional volatility equa t ion causes the coefficients α1 and β1 to

become statistically insignificant if the coefficient γ 1 of bid/ask spreads is positive and sign i ficant.

First, in all twenty-six stocks estimated, t he coefficients of lagged-one bid/ask

spreads (θ1 ) were found to be positive except for ID 4 and 22. The coefficients are statistically significant at the 1% level. In general, a positive relationship exists

163 between intraday volatility and lagged-one bid/ask spreads for all twenty-four individual stocks.

Second, all coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 1% level. The persistence in volatility as measured by the sum of

α1 and β1 ranges from 0.8163 to 0.9820 with an average of 0.9118 for the twenty-

six individual stocks. The average sum of α1 and β1 of the twenty-six individual stocks slightly decreases from 0.9514 to 0.9118 compared with Table 5.8 but these

findings show an increase in the volatility persistence ( α1 + β1 ) with the contemporaneous bid/ask spread as the mixing variable (0.8231).

These results are similar to those of Rahman, Lee and Ang (2002). The

GARCH effect remains strongly significant for the intraday return series. However, the average coefficient of lagged-one bid/ask spreads is fairly small, only 0.0082,

compared with 0.9118 for that of the sum of α1 and β1 of the twenty-six individual stocks. Thus, volatility is better explained by previous volatility than by lagged-one bid/ask spread, and the results reject hypothesis 3.

Third, the positive contribution of the lagged-one bid/ask spread variable, in terms of its ability to explain the volatility, can be noticed by comparing the log- likelihood of the estimations in Table 5.13 with Table 5.8. The average absolute value of the log-likelihood function of the twenty-six individual stocks slightly decreases from 4166.19 to 4111.60 due to the inclusion of the lagged-one bid/ask spread in the conditional variance equation.

Fourth, diagnostic tests on the standardized residuals are provided in Table

5.11. The Q(20) statistics indicate there is still serial correlation in the standardized residuals but the Q2(20) statistics indicate no serial correlation in the squared standardized residuals for eighteen out of twenty-six individual stocks. The test

164 results are better than the GARCH estimation without the bid/ask spread (seventeen out of twenty-six) and with bid/ask spread (ten out of twenty-six). Thus, the results suggest that the GARCH(1,1) model is successful in explaining the time-varying volatility in the data, and can capture the volatility clustering in the intraday return series. Thus, the estimated models fit the data for the volatility clustering in the eighteen individual stocks, which supports hypothesis 1.

To sum up, the inclusion of the intraday the lagged-one bid/ask spread as a mixing variable for information arrival in the conditional variance model can help in explaining the GARCH effects in the stock return series but is not obvious. The

GARCH effects do not disappear as a result of this inclusion. The average

persistence in volatility as measured by the sum of α1 and β1 has a slight decrease from 0.9514 to 0.9118 compared with Table 5.8. The model estimation indicates that the GARCH (1,1) model is successful in accounting for the nonlinear dependencies

(volatility clustering) but not in accounting for the linear dependencies (serial correlation) in the Shanghai stock market.

5.6 Conclusion

This chapter applies the GARCH model to investigate the dynamic relationship between intraday volatility, trading volume and bid/ask spread by focusing on a subset of twenty-six stocks from the Shanghai Stock Exchange 180 Index. In financial markets, most time series have a distribution with a fatter tail than the normal distribution, and exhibit strong time-varying variance, volatility clustering and volatility persistence. Thus, the traditional OLS estimation technique cannot be

165 used for intraday volatility. Recently, the GARCH model has been widely used in the financial time series while the variances in such series change through time, which does not require the normal distribution of stock data.

The main purpose of this chapter is to explore the issue of whether the return volatility in high-frequency data can be described by the GARCH(1,1) specification, and whether GARCH modeling captures the effects of temporal dependence in trading volume or bid/ask spreads for individual stocks in the Chinese stock market.

The findings indicate that all individual stocks in the conditional variance model exhibit the strong persistence of the GARCH effects. Similar to Rahman, Lee and

Ang (2002), the GARCH (1,1) model best describes the volatility of intraday returns.

For most stocks, the Q2(20) statistics on the squared standard residuals are not significant at the five percent level. Current volatility can be explained by past volatility that tends to persist over time. The findings of this chapter further indicate that the AR(1)-GARCH(1,1) model successfully accounts for the nonlinear dependencies in the individual stocks. However, the AR(1)-GARCH(1,1) model may not account for the linear dependencies in the high-frequency data.

The inclusion of intraday contemporaneous/lagged trading volume as a mixing variable for information arrival in the conditional variance model helps in explaining the GARCH effects in the stock return series. The coefficient for volume is positive and small for the GARCH(1,1). But the GARCH effects do not disappear entirely as a result of this inclusion; in other words, the persistence in volatility remains in intraday return series even after contemporaneous/lagged volume is included in the model as an explanatory variable. The observed results contradict the findings of Lamoureux and Lastrapes (1990). This inconsistency is based on the fact that the GARCH effect did not disappear when an attempt was made to account for

166 the uneven flow of information by the introduction of the contemporaneous/lagged trading volume as a proxy into the conditional variance equation.

Similar to the proxy of trading volume, the inclusion of intraday contemporaneous/lagged bid/ask spread as a mixing variable for information arrival in the conditional variance model helps in explaining the GARCH effects. Most of the coefficients for bid/ask spreads are positive and small for the GARCH(1,1). The

GARCH effects remain strongly significant after the introduction of the contemporaneous/lagged bid/ask spreads. The findings are in accordance with those of Rahman, Lee and Ang (2002).

The findings in this chapter suggest that the volume and bid/ask spread as an information variable has quite a limited effect on the volatility of intraday returns in the Shanghai stock market, which is similar to the results in the US market.

Asymmetric information does not explain the volatility that may result from noise trading in the market. These findings are quite important because it could indicate that the Shanghai stock market, after a decade’s development, has experienced some improvement in its quality of trading, becoming closer to more mature markets, such as the NASDAQ market, at least in this aspect. It could also suggest that the volume or bid/ask spread as an information variable is not a very important factor for predicting the volatility of stock returns in intraday levels, irrespective of how mature or efficient a market is. If the latter is true, based on the MDH hypothesis, there should be other major factors that affect the volatility change in intraday returns, which should be found with further research.

167

Table 5.1 Descriptive statistics of intraday 5-minute log returns (in percentages) for twenty-six individual stocks, 2000 – 2002.

Jarque ID Size Mean Variance Skewness Kurtosis Bera Test Max Min IQR

1 30947 -0.00174 0.07230 1.030 19.315 348686.6 5.539 -2.453 0.256 2 30733 0.00013 0.08358 0.793 16.666 242382.7 4.175 -3.363 0.312 3 30546 0.00603* 0.08752 0.485 14.241 162025.7 4.240 -4.085 0.262 4 27614 -0.00398* 0.09827 0.597 20.845 368052.5 4.340 -5.399 0.247 5 30974 -0.00083 0.06006 0.849 34.377 1274355.5 6.188 -5.129 0.204 6 25096 -0.00367* 0.06581 0.898 28.267 670934.7 5.632 -3.871 0.200 7 30986 -0.00269 0.10934 0.845 20.017 377543.8 5.318 -4.920 0.245 8 30474 -0.00093 0.09811 1.484 34.444 1266631.2 9.122 -2.653 0.225 9 27712 -0.00273 0.08090 0.670 22.871 458006.9 5.108 -3.424 0.169 10 19214 -0.00286 0.07453 0.533 17.052 158990.7 3.885 -3.622 0.183 11 19949 0.00881* 0.08148 0.302 21.135 273658.6 3.727 -4.883 0.214 12 9330 0.00051 0.06840 0.493 10.655 23159.9 2.535 -2.578 0.281 13 30572 -0.00115 0.08727 1.000 25.350 641410.8 6.748 -4.256 0.242 14 30353 -0.00309 0.13384 0.792 16.237 224790.5 5.419 -3.895 0.282 15 29663 -0.00303 0.15123 1.012 17.336 259067.3 6.044 -4.214 0.274 16 30571 -0.00565* 0.14259 0.987 23.929 562903.7 5.907 -4.556 0.268 17 31134 -0.00329 0.11107 0.711 17.035 258147.6 3.953 -5.043 0.296 18 30852 -0.00056 0.10262 1.519 31.675 1068837.9 8.471 -2.794 0.261 19 30453 -0.00508* 0.12368 1.334 30.654 979372.0 8.224 -4.405 0.255 20 29796 0.00090 0.10657 0.569 21.972 448474.0 5.022 -5.745 0.259 21 30264 -0.00379* 0.06526 0.833 20.961 410286.2 4.948 -2.656 0.199 22 30874 -0.00417* 0.13634 0.457 36.444 1439963.4 7.050 -8.300 0.282 23 31210 -0.00359 0.13348 0.883 14.027 162175.0 5.081 -3.815 0.309 24 30659 0.00063 0.07994 0.770 12.683 122792.9 3.366 -2.688 0.234 25 30789 -0.00418* 0.10217 0.846 19.046 333985.2 4.678 -3.690 0.234 26 30272 -0.00075 0.13629 0.397 13.177 131433.4 4.908 -4.643 0.265 Note: * represents different from zero at the 5% significance level.

168

Table 5.2 Tests of intraday 5-minute log returns (in percentages) for twenty-six individual stocks, 2000 – 2002.

ID Size Q(10) Q(20) Q2(10) Q2(20) LLF Rho Rho2 LM(10) LM(20)

1 30947 633.93 645.13 4064.99 5004.04 -3262.57 -0.0171 0.2638 2601.05 2714.54 2 30733 485.70 511.45 4056.51 4853.09 -5468.05 -0.0278 0.2645 2591.69 2707.60 3 30546 237.46 248.98 3309.98 3699.19 -6139.13 -0.0415 0.2669 2398.98 2462.76 4 27614 332.04 362.41 4077.22 5197.57 -7149.41 0.0058 0.1881 2201.32 2302.43 5 30974 492.58 516.81 5370.82 5667.65 -393.64 -0.0671 0.3841 4983.44 5027.11 6 25096 360.30 373.66 708.58 882.42 -1465.85 0.0051 0.0812 659.26 719.92 7 30986 903.66 922.28 5316.37 6254.90 -9676.77 0.0545 0.2405 2749.77 2815.69 8 30474 408.58 426.12 425.00 482.94 -7864.60 0.0324 0.0677 328.96 351.18 9 27712 251.95 264.11 2375.71 2694.99 -4480.23 -0.0373 0.1878 1524.09 1598.73 10 19214 282.16 308.38 1768.86 1997.61 -2318.01 0.0117 0.2346 1241.90 1287.25 11 19949 89.87 124.65 2476.59 3244.76 -3295.31 -0.0243 0.1965 1379.37 1506.85 12 9330 222.97 241.26 1184.02 1333.08 -724.71 -0.0918 0.2289 736.17 786.77 13 30572 456.92 481.51 2128.60 2763.45 -6100.58 0.0272 0.1917 1477.68 1610.43 14 30353 485.06 496.49 2944.37 3279.03 -12546.39 0.0288 0.2031 1843.67 1891.63 15 29663 198.03 219.85 1812.35 1947.96 -14073.62 -0.0045 0.2030 1389.16 1428.70 16 30571 727.07 768.99 5575.77 8247.90 -13605.24 0.0034 0.2233 2771.33 3053.35 17 31134 655.82 665.19 2958.54 3643.60 -9967.41 0.0535 0.1748 1705.55 1773.47 18 30852 397.13 415.43 631.84 754.42 -8656.37 0.0007 0.1024 867.49 910.84 19 30453 497.77 524.13 1084.13 1372.71 -11386.38 0.0279 0.1140 769.28 857.70 20 29796 205.19 219.52 2330.10 2405.64 -8922.74 -0.0376 0.2528 2009.21 2028.94 21 30264 217.70 235.54 2915.72 3271.09 -1640.21 0.0291 0.2314 1966.60 2016.67 22 30874 656.06 709.53 2817.71 3737.14 -13048.05 0.0451 0.1733 1717.32 1904.04 23 31210 694.01 732.70 4794.02 6293.49 -12858.93 0.0125 0.2191 2497.07 2643.04 24 30659 288.35 303.67 3779.00 4783.74 -4773.85 0.0344 0.2166 2139.78 2268.28 25 30789 443.26 468.33 4884.02 5814.56 -8570.04 0.0424 0.2677 2844.96 2944.18 26 30272 307.04 319.35 4587.42 5707.49 -12788.01 0.0173 0.2296 2468.68 2641.63 Average of absolute value (twenty-six individual stocks) 7352.93 Note: • Q(10) and Q(20) are the Ljung-Box-Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the return series. Q2(10) and Q2(20) are the Ljung-Box-Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the squared return series. • LLF is log-likelihood function. Rho and Rho2 are first-order autocorrelation of return series and squared return series, respectively. The first-order autocorrelation of return series of ID 4, 6, 10, 15, 16 and 18 are not significant at the 5% level • LM(10) and LM(20) are the Lagrange multiplier test values up to lag 10 and lag 20.

169

Table 5.3 Unit root tests of intraday 5-minute log returns (in percentages), log volumes and log bid/ask spreads for twenty-six individual stocks, 2000 – 2002.

ID Return Series Trading Volume Series Bid/Ask Spread Series

1 -13.275 -6.771 -4.719 2 -14.172 -6.437 -3.701 3 -14.034 -7.910 -6.342 4 -13.968 -5.860 -7.095 5 -13.381 -6.757 -5.683 6 -12.387 -7.171 -8.276 7 -14.398 -6.747 -7.672 8 -14.716 -6.964 -5.223 9 -16.137 -6.102 -6.321 10 -11.479 -6.381 -6.772 11 -13.112 -6.520 -5.759 12 -10.404 -5.752 -6.927 13 -13.442 -6.475 -4.469 14 -13.974 -6.898 -6.047 15 -13.404 -6.771 -6.037 16 -14.148 -5.036 -8.130 17 -12.842 -6.389 -5.231 18 -12.364 -6.473 -5.312 19 -13.443 -6.946 -6.615 20 -13.225 -6.468 -6.102 21 -12.997 -7.513 -5.168 22 -14.191 -8.296 -6.583 23 -13.586 -5.725 -4.099 24 -13.131 -6.056 -4.142 25 -13.247 -5.082 -4.859 26 -14.555 -7.454 -4.719 Note: The augmented Dickey-Fuller regression equation to test the unit root is: p with trend: rrtr ∆=+ttα0112αα−− ++∑ γ j ∆+tjεt j=1 where, rt denotes returns and ∆ denotes the first-difference operator. Critical values for t with ∞ observations are: -2.57 at the 10% significance level, -2.86 at 5% significance level and –3.43 at 1% significance level.

170

Table 5.4 Descriptive statistics of intraday 5-minute log of trading volumes for twenty-six individual stocks, 2000 – 2002.

Jarque ID Size Mean Variance Skew. Kurtosis Bera Test Max Min IQR

1 30947 10.138 4.824 -6.205 61.536 4616795.3 15.376 -11.513 1.714 2 30733 9.904 8.348 -5.154 38.795 1776834.4 15.742 -11.513 2.060 3 30546 10.093 10.378 -4.886 32.949 1263150.8 14.989 -11.513 2.131 4 27614 9.635 5.923 -5.849 51.385 2851132.6 15.556 -11.513 1.731 5 30974 10.029 5.060 -5.897 57.013 3944626.9 15.480 -11.513 1.777 6 25096 9.153 7.904 -5.545 41.158 1651121.7 16.233 -11.513 1.707 7 30986 9.478 4.289 -6.004 61.896 4664575.4 15.765 -11.513 1.654 8 30474 9.521 7.101 -5.620 44.867 2386068.6 15.083 -11.513 1.742 9 27712 7.709 19.501 -3.700 16.428 271449.7 14.926 -11.513 1.872 10 19214 7.944 18.215 -3.778 17.609 216561.9 15.176 -11.513 2.180 11 19949 9.796 6.524 -5.620 47.479 1749468.9 15.558 -11.513 1.834 12 9330 9.636 11.636 -4.499 28.565 285560.5 16.273 -11.513 2.327 13 30572 9.590 4.242 -6.389 66.151 5288064.9 15.019 -11.513 1.625 14 30353 8.111 13.120 -4.486 24.691 696863.8 14.059 -11.513 1.813 15 29663 7.532 19.236 -3.689 16.330 286863.9 14.276 -11.513 1.964 16 30571 9.140 6.898 -5.407 43.217 2209181.1 15.368 -11.513 1.891 17 31134 9.378 4.685 -6.354 62.026 4729112.0 14.579 -11.513 1.553 18 30852 10.053 4.767 -5.905 58.876 4192792.6 16.151 -11.513 1.749 19 30453 8.109 13.029 -4.495 24.799 705537.6 13.754 -11.513 1.825 20 29796 8.213 16.919 -4.036 19.694 426909.1 14.815 -11.513 1.938 21 30264 8.380 10.784 -5.039 30.826 1104417.2 13.676 -11.513 1.642 22 30874 9.194 5.129 -6.310 58.054 4103853.3 14.463 -11.513 1.608 23 31210 10.420 3.600 -6.607 76.858 7320860.7 15.077 -11.513 1.542 24 30659 9.364 8.145 -5.521 40.820 1982931.6 15.589 -11.513 1.766 25 30789 8.959 9.294 -5.049 34.603 1412030.0 15.022 -11.513 1.900 26 30272 9.171 11.766 -4.788 29.254 985046.1 14.985 -11.513 1.906

171

Table 5.5 Tests of intraday 5-minute log trading volumes for twenty-six individual stocks, 2000 – 2002.

ID Size Q(10) Q(20) Q2(10) Q2(20) LLF Rho Rho2 LM(10) LM(20)

1 30947 23475 40772 110839 196880 -68262 0.312 0.672 103.68 122.12 2 30733 29070 52070 129688 237222 -76216 0.342 0.712 509.16 646.97 3 30546 31949 52537 102432 170787 -79076 0.400 0.693 1361.64 1544.93 4 27614 19391 31206 85158 141844 -63743 0.323 0.652 410.37 475.02 5 30974 24551 42770 100033 175262 -69059 0.327 0.655 187.38 274.67 6 25096 12979 21357 45340 71771 -61550 0.278 0.539 621.28 748.66 7 30986 23459 36947 97729 159966 -66527 0.331 0.675 114.70 154.95 8 30474 20954 32457 75538 119183 -73108 0.327 0.620 714.62 767.50 9 27712 16464 27587 29027 48153 -80480 0.295 0.390 2172.33 2395.16 10 19214 13219 22681 41702 74698 -55145 0.317 0.523 1101.73 1260.46 11 19949 13450 21655 60550 99191 -47013 0.337 0.669 243.12 301.28 12 9330 6118 10878 33366 59483 -24687 0.311 0.683 122.30 224.84 13 30572 20509 33582 100329 170753 -65467 0.293 0.663 66.48 70.66 14 30353 17417 28220 52327 87868 -82135 0.297 0.497 1434.40 1604.21 15 29663 13991 23593 27683 45823 -85943 0.268 0.378 1381.15 1611.14 16 30571 24474 39806 95802 163031 -72897 0.333 0.654 701.91 796.33 17 31134 22237 35488 104935 176885 -68218 0.306 0.677 184.80 209.43 18 30852 23543 38653 95941 160541 -67869 0.344 0.669 165.08 199.78 19 30453 18394 31135 49728 84342 -82300 0.285 0.475 1258.89 1456.48 20 29796 11832 19300 36586 61255 -84417 0.249 0.423 1072.49 1218.98 21 30264 16351 27190 56319 98418 -78927 0.274 0.495 1244.95 1391.23 22 30874 16572 24911 67073 105462 -69047 0.292 0.584 369.78 450.89 23 31210 27161 43153 108488 176963 -64274 0.338 0.707 89.60 110.81 24 30659 22678 39403 103162 180489 -75655 0.320 0.659 784.52 1005.23 25 30789 23337 39810 109452 195248 -78006 0.306 0.664 721.82 834.74 26 30272 23611 40192 75409 123530 -80267 0.357 0.619 1718.75 2032.82 Note: • Q(10) and Q(20) are the Ljung-Box-Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the return series. Q2(10) and Q2(20) are the Ljung-Box- Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the squared return series. • LLF is log-likelihood function. Rho and Rho2 are first-order autocorrelation of return series and squared return series, respectively. • LM(10) and LM(20) are the Lagrange multiplier test values up to lag 10 and lag 20.

172

Table 5.6 Descriptive statistics of intraday 5-minute log bid/ask spreads for twenty-six individual stocks, 2000 – 2002.

Jarque ID Size Mean Variance Skew. Kurtosis Bera Test Max Min IQR

1 30947 -6.458 0.096 1.640 6.626 30821.9 -4.182 -6.909 0.371 2 30733 -6.290 0.092 2.364 9.566 83831.3 -3.757 -6.676 0.213 3 30546 -6.512 0.258 1.014 4.197 7055.3 -3.822 -7.304 0.656 4 27614 -6.912 0.391 0.737 2.953 2503.4 -3.555 -7.839 0.885 5 30974 -6.713 0.138 2.186 7.865 55211.2 -3.588 -7.047 0.179 6 25096 -7.018 0.459 0.834 2.904 2919.3 -4.080 -7.842 1.011 7 30986 -7.004 0.440 0.624 2.734 2100.9 -4.145 -8.004 1.049 8 30474 -6.630 0.250 1.484 4.910 15819.1 -3.660 -7.177 0.694 9 27712 -6.673 0.581 0.874 3.106 3539.6 -2.512 -7.613 1.264 10 19214 -6.659 0.428 1.024 3.413 3496.0 -3.476 -7.421 0.857 11 19949 -6.638 0.201 1.732 5.768 16341.7 -3.842 -7.120 0.531 12 9330 -6.404 0.142 2.043 6.629 11610.9 -4.288 -6.728 0.097 13 30572 -6.587 0.171 1.589 5.330 19779.6 -3.746 -7.102 0.292 14 30353 -6.540 0.498 0.696 2.827 2489.2 -3.100 -7.688 1.099 15 29663 -6.378 0.605 0.515 2.562 1548.3 -3.182 -7.583 1.310 16 30571 -6.817 0.452 0.593 2.706 1900.2 -3.321 -7.978 0.970 17 31134 -6.759 0.287 1.105 3.552 6725.6 -3.040 -7.438 0.774 18 30852 -6.847 0.336 0.904 3.476 4491.1 -3.010 -7.740 0.802 19 30453 -6.677 0.548 0.465 2.427 1515.2 -3.465 -7.908 1.237 20 29796 -6.425 0.406 0.691 3.356 2526.8 -3.013 -7.552 0.899 21 30264 -6.695 0.346 1.088 3.779 6736.0 -3.655 -7.424 0.920 22 30874 -6.837 0.525 0.462 2.656 1250.5 -3.693 -8.049 1.016 23 31210 -6.668 0.191 1.121 4.291 8697.3 -4.363 -7.227 0.593 24 30659 -6.548 0.197 1.688 5.706 23918.2 -3.824 -6.993 0.373 25 30789 -6.620 0.301 1.095 3.812 6995.2 -3.073 -7.342 0.802 26 30272 -6.488 0.342 1.194 4.131 8810.8 -3.293 -7.234 0.811

173

Table 5.7 Tests of intraday 5-minute log bid/ask spreads for twenty-six individual stocks, 2000 – 2002.

ID Size Q(10) Q(20) Q2(10) Q2(20) LLF Rho Rho2 LM(10) LM(20)

1 30947 44832 85578 50442 96892 -7658 0.435 0.454 1724.38 1824.49 2 30733 31374 53783 35312 61515 -6869 0.435 0.449 3122.87 3209.68 3 30546 23668 34110 24730 36286 -22665 0.431 0.433 8162.02 8387.27 4 27614 9420 13620 9635 14202 -26215 0.2940.292 2339.58 2381.82 5 30974 12424 19691 13068 20998 -13257 0.303 0.304 2826.76 2978.16 6 25096 13595 19915 13222 19522 -25830 0.350 0.342 3645.98 3723.29 7 30986 9503 13678 9717 14195 -31251 0.285 0.283 2353.99 2425.91 8 30474 14042 21048 14553 22169 -22144 0.337 0.337 3329.81 3425.56 9 27712 33957 51748 32512 50322 -31786 0.5130.497 9131.62 9177.02 10 19214 23025 34455 23554 36139 -19105 0.511 0.505 5278.08 5293.47 11 19949 7264 9987 7153 9955 -12324 0.325 0.320 3470.21 3543.43 12 9330 4855 7237 4903 7387 -4133 0.402 0.398 1935.69 1990.13 13 30572 12767 20594 13772 22585 -16400 0.306 0.309 2429.87 2517.88 14 30353 16318 22471 15562 21754 -32481 0.3960.382 5991.45 6009.80 15 29663 22618 32899 20607 30104 -34638 0.449 0.429 7897.53 7981.54 16 30571 8669 11661 8613 11709 -31252 0.293 0.290 3150.63 3284.83 17 31134 9231 14441 9407 14883 -24749 0.279 0.277 2568.67 2671.62 18 30852 12826 20684 13626 22339 -26938 0.302 0.305 2601.11 2697.31 19 30453 11618 16451 11043 15827 -34049 0.3380.328 3546.36 3568.48 20 29796 27225 41936 28188 44560 -28845 0.468 0.464 6895.43 6987.90 21 30264 40014 66289 41399 69625 -26903 0.502 0.499 6687.27 6753.14 22 30874 10780 15343 10694 15405 -33854 0.301 0.295 3072.14 3267.06 23 31210 19697 35399 22324 40637 -18473 0.308 0.322 1432.00 1514.44 24 30659 34057 57684 35799 61541 -18581 0.463 0.465 6836.73 6972.44 25 30789 24910 41003 26108 43650 -25213 0.403 0.403 3961.31 4036.24 26 30272 28481 46215 28586 47124 -26726 0.441 0.435 8312.25 8510.21 Note: • Q(10) and Q(20) are the Ljung-Box-Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the return series. Q2(10) and Q2(20) are the Ljung-Box-Pierce Portmanteau statistics to test the null hypothesis of no serial correlation up to lag 10 and lag 20 for the squared return series. • LLF is log-likelihood function. Rho and Rho2 are first-order autocorrelation of return series and squared return series, respectively. • LM(10) and LM(20) are the Lagrange multiplier test values up to lag 10 and lag 20.

174 Table 5.8 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns for twenty-six individual stocks on the Shanghai Stock Exchange in percentage (2000 – rbbra, tt=+011− +t

2002) for Equation 5.4: awttt==σε,.here ε tii.d.N(0,1) 2 a22 σααt01=+t−−11 + βσ 1t 2 2 ID α0 α1 1 α1+1 LLF Q(20) Q (20) ID α0 α1 1 α1+1 LLF Q(20) Q (20)

1 0.002667 0.146445 0.825449 0.9719 1405.5 253.68 73.313* 14 0.01661 0.240802 0.659772 0.9006 -9113.1 160.18 13.73 (43.55) (88.14) (513.4) (72.23) (68.92) (169.6) 2 0.005748 0.19212 0.755114 0.9472 -1386.5 222.29 36.165* 15 0.017024 0.227708 0.689274 0.9170 -11011.5 85.743 23.78 (63.05) (82.81) (299.9) (64.43) (75.19) (209.8) 3 0.006179 0.209646 0.743373 0.9530 -2534.6 126.96 42.413* 16 0.00322 0.123712 0.860368 0.9841 -6742.0 229.79 31.964* (57.96) (80.5) (291.2) (89.32) (83.91) (913.6) 4 0.004724 0.147875 0.808449 0.9563 -2723.8 224.27 27.413 17 0.004447 0.153229 0.818622 0.9719 -5628.6 235.63 56.117* (55.42) (67.23) (345.3) (55.61) (68.75) (415.4) 5 0.003478 0.227622 0.742833 0.9705 4934.4 146.39 9.063 18 0.004861 0.16528 0.807101 0.9724 -4874.7 181.64 1.443 (71.39) (122.4) (418.2) (59.04) (68.62) (342.8) 6 0.002714 0.131519 0.83808 0.9696 1693.7 137.52 5.997 19 0.003746 0.099273 0.876653 0.9759 -7521.9 178.88 16.906 (70.04) (64.89) (635.1) (82.53) (101.4) (1613) 7 0.007455 0.228352 0.723831 0.9522 -4146.0 298.44 15.706 20 0.010641 0.215186 0.71144 0.9266 -5687.9 76.519 14.383 (91.4) (93.86) (310.7) (68.14) (71.8) (242) 8 0.011621 0.255807 0.658541 0.9143 -4595.5 168.47 1.53 21 0.005356 0.217879 0.719017 0.9369 2797.0 72.489 13.327 (78.41) (80.31) (206.4) (81.77) (70.9) (234.7) 9 0.00643 0.200444 0.743415 0.9439 -622.9 70.461 17.114 22 0.00641 0.179151 0.787383 0.9665 -7378.6 291.28 12.242 (81.95) (81.09) (341.8) (72.78) (75.17) (332.6) 10 0.008153 0.228801 0.682025 0.9108 15.6 141.33 25.363 23 0.001736 0.078375 0.911365 0.9897 -8273.3 308.48 171.87* (58.49) (59.61) (165.6) (43.61) (96.55) (1385) 11 0.002078 0.160032 0.832508 0.9925 447.1 55.342 46.335* 24 0.001096 0.084707 0.907305 0.9920 -550.1 86.436 127.94* (39.77) (67.13) (454.6) (40.27) (103.4) (1663) 12 0.007898 0.152153 0.73256 0.8847 26.8 118.63 11.755 25 0.005433 0.198082 0.767881 0.9660 -3330.9 141.46 34.271* (26.03) (24.82) (80.95) (104) (94.18) (492.8) 13 0.005624 0.190425 0.763881 0.9543 -2030.4 173.56 10.161 26 0.014206 0.238704 0.677621 0.9163 -8848.6 126.93 19.252 (57.39) (68.35) (300.9) (64.55) (68.84) (194.3) Average of absolute value (twenty-six individual stocks) 0.9514 4166.19

Note: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parentheses are t-statistics. (3) * represents significance levels of 5%

175

Table 5.9 AR(1)-GARCH(p,q) model estimation of intraday 5-minute returns for twenty-six individual stocks on the Shanghai Stock Exchange in rbbra tt=+011− +t percentag e (2000 – 200 2 ) by awtt==σεt,.hereε tii.d.N(0,1) 2 aaa222 2 2 2 σ t01=+ααt−123 + α 2t−−++ α 3t βσ 1t2t3t −1 +β σ − 2 + β σ − 3

α1+ α2+ α3+Log- 2 ID GARCH(p,q) α0 α1 α2 α3 1 2 3 1+2 +3 Likelih ood Q( 2 0) Q (20)

1 GARCH (2,2) 0.00008 0.25007 -0.24036 1.42410 -0.43479 0.99 9 10 200 0. 59 26 5. 33 35. 0 1* (16.60) (56.65) (-56.43) (134.82) (-42. 42) 7 GARC H (2,2) 0.00005 0.31060 -0.30589 1.46084 -0.4 6 572 0.99 9 87 -34 41 .65 29 1. 90 13 .2 0 (25.74) (85.68) (-84.92) (251.44) (-81.29) 15 GARCH (2,1) 0.00938 0.30065 -0.160622 0.813925 0.96 3 33 -109 1 9.78 11 0. 55 16 .9 4 (45.70) (62.26) (-34. 99) (255.62) 15 GARC H (2,3) 0.00015 0.27656 -0.2 7 045 1.34934 -0.21476 -0.1 4 121 0.99 9 63 -105 5 5.53 10 6. 02 14 .7 7 (21.09) (60.82) (-60.67) (90.92) (-8.53) (-12. 67) 17 GARCH (1,2) 0.00527 0.19298 0.41241 0.36121 0.97 1 87 -55 45 .86 25 8. 12 51. 5 0* (47.62) (57.15) (26.90) (27.59) 17 GARC H (2,2) 0.00008 0.23093 -0.22334 1.52505 -0.53302 0.99 9 70 -51 42 .64 23 6. 29 11 .5 4 (1 6.07) (54.05) (-54.14) (187.16) (-67.55) 17 GARCH(2,3) 0 .0 0007 0.25064 -0.24363 1.34135 -0.17593 -0.17276 0.99974 -5114.65 252.31 10.69 (15.41) (52.01) (-51.88) (95.38) (-7.30) (-15.12)

Notes: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parenthesis are t-statistics. (3) * represents significance levels of 5%. (4)

176

Table 5.10 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with contemporaneous volume for twenty-six individual stocks on the Shanghai Stock Exchange rbbra, tt= 011++− t in percentage (2000 – 2002) for Equation 5.5: awttt==σε,.here ε tii.d.N(0,1) 2 aV22ol σααt01=+t−−11 + βσθ 1t + t 2 2 ID α0 α1 1 θ α1+1 LLF Q(20) Q (20) ID α0 α1 1 θ α1+1 LLF Q(20) Q (20)

1 0.00214 0.21381 0.70056 0.00045 0.9144 1457.5 255.94 41.59* 14 0.01396 0.27389 0.56527 0.00121 0.8392 -8919.1 169.22 17.375 (39.5) (89.48) (748.5) (81.18) (90.46) (81.22) (183) (117.9) 2 0.00973 0.33360 0.47289 0.00102 0.8065 -1425.6 236.33 145.37* 15 0.01674 0.23809 0.64005 0.00073 0.8781 -10942 91.63 20.54 (77.73) (75) (132.7) (1151) (76.44) (74.23) (200.1) (50.9) 3 0.00697 0.28729 0.56201 0.00082 0.8493 -2494.0 137.85 59.765* 16 0.02338 0.28508 0.46161 0.00205 0.7467 -8042.0 342.43 100.18* (58.8) (69.41) (136.1) (91.18) (72.47) (52.36) (96.13) (252.4) 4 0.00592 0.22670 0.64498 0.00066 0.8717 -2660.9 233.74 40.129* 17 0.00268 0.19306 0.73968 0.00061 0.9327 -5527.3 245.2 27.14 (16.01) (68.1) (244.3) (24.5) (23.75) (78.4) (536.5) (60) 5 0.01838 0.62111 0.12639 0.00161 0.7475 3555.3 239.71 242.9* 18 0.03455 0.15100 0.37840 0.00332 0.5294 -7063.5 307.73 97.494* (164.4) (83.18) (46.7) (3543) (40.36) (30.6) (30.84) (68.49) 6 0.00933 0.23539 0.49986 0.00109 0.7353 1424.9 192.43 30.193 19 0.04882 0.17395 0.41026 0.00441 0.5842 -9881.2 320.74 189.02* (80.71) (48.21) (137.6) (3049) (106.3) (41.99) (87.54) (106.9) 7 0.00325 0.25417 0.66527 0.00077 0.9194 -3939.7 313.59 13.235 20 0.00919 0.23541 0.62840 0.00093 0.8638 -5552.3 84.001 10.469 (38.9) (88.09) (239.2) (71.58) (94.96) (74.27) (269.2) (247) 8 0.03722 0.17727 0.34587 0.00346 0.5231 -6373.2 297.55 48.705* 21 0.00449 0.22292 0.70949 0.00014 0.9324 2807.3 72.552 14.177 (103.9) (28.36) (68.93) (1156) (62.94) (70.86) (230.3) (24.36) 9 0.00768 0.20459 0.74538 -0.00020 0.9500 -599.8 69.338 20.389 22 0.01009 0.28469 0.59917 0.00113 0.8839 -7519.5 324.42 20.742 (78.83) (76.58) (328.5) (-27.99) (85.52) (74.66) (209.1) (120.7) 10 0.00643 0.23467 0.65327 0.00041 0.8879 43.4 142.1 20.971 23 0.00162 0.16638 0.77909 0.00060 0.9455 -8244.0 308.47 22.764 (61.3) (57.97) (169.8) (35.98) (28.73) (55.96) (226.2) (33.41) 11 0.00062 0.24567 0.70456 0.00054 0.9502 441.3 66.158 36.279* 24 0.00078 0.08125 0.91116 0.00003 0.9924 -548.9 87.491 132.38* (8.092) (60.13) (211.2) (66.52) (13.32) (79.03) (1201) (4.933) 12 0.00501 0.16786 0.61558 0.00090 0.7834 97.5 122.48 31.423 25 0.01479 0.34796 0.41786 0.00138 0.7658 -3509.6 174.59 171.48* (15.91) (24.07) (54.7) (50.44) (247) (64.82) (135.2) (38000) 13 0.01858 0.21360 0.42513 0.00246 0.6387 -3260.5 274.55 111.39* 26 0.00837 0.23326 0.64049 0.00118 0.8737 -8704.6 136.13 26.354 (56.81) (45.96) (94.24) (1093) (44.18) (61.1) (152.3) (88.01) Average of absolute value (twenty-six individual stocks) 0.8210 4424.43

Note: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parentheses are t-statistics. (3) * represents significance levels of 5%

177

Table 5.11 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with lagged-one volume for twenty-six individual stocks on the Shanghai Stock Exchange in rbbra, t= 011++t− t percentage (200 0 – 200 2) for Equ ation 5.6 : awtt==σεt,.hereε tii.d.N(0,1) 2 aV22ol σααt01=+t−−11 + βσ 1t+θ t− 1 2 2 ID α0 α1 1 θ α1+1 LLF Q(20) Q (20) ID α0 α1 1 θ α1+1 LLF Q(20) Q (20)

1 0.00172 0.14611 0.82020 0.00012 0.9663 1422.6 255.11 68.977* 14 0.01339 0.22758 0.67094 0.00037 0.8985 -9093.6 160.25 14.26 (34.34) (85.61) (454.8) (21.95) (52.92) (67.87) (175.6) (19.94) 2 0.00282 0.19370 0.74234 0.00038 0.9360 -1348.3 225.92 34.246* 15 0.01763 0.23023 0.68890 -0.00010 0.9191 -11010 85.265 24.072 (18.57) (82.35) (287.5) (21.09) (62.15) (74.69) (209) (-4.952) 3 0.00525 0.20664 0.74321 0.00011 0.9498 -2528.6 126.91 41.602* 16 0.00310 0.12391 0.85954 0.00002 0.9835 -6741.6 230.03 31.352 (38.73) (81.51) (289.9) (8.803) (77.35) (81.06) (906.5) (5.027) 4 0.00247 0.15953 0.78051 0.00037 0.9400 -2628.4 223.83 24.953 17 0.00250 0.18886 0.74921 0.00059 0.9381 -5589.8 242.53 27.808 (43.63) (62.4) (283.9) (72.44) (31.93) (66.89) (284) (54.81) 5 0.00235 0.23835 0.72740 0.00015 0.9657 4940.5 145.19 8.858 18 0.01672 0.15296 0.57735 0.00220 0.7303 -6559.7 262.03 28.657 (16.24) (112.6) (338.2) (8.892) (21.91) (35.79) (62.98) (98.31) 6 0.00131 0.13123 0.83184 0.00019 0.9631 1721.4 139.51 4.871 19 0.00340 0.10790 0.85812 0.00019 0.9660 -7499.0 179.61 10.542 (18.07) (61.26) (565.1) (20.42) (70.86) (98.84) (1284) (30.1) 7 0.00430 0.22211 0.72155 0.00039 0.9437 -4104.2 302.81 14.254 20 0.01002 0.21354 0.71262 0.00008 0.9262 -5687.0 76.746 14.459 (62.05) (93.37) (301.3) (53.01) (38.52) (71.49) (242.9) (3.06) 8 0.00609 0.23319 0.67012 0.00060 0.9033 -4496.2 174.01 2.438 21 0.00761 0.22574 0.71126 -0.00026 0.9370 2822.2 74.036 12.696 (47.66) (68.07) (195.4) (68.11) (45.83) (70.63) (227.4) (-15.84) 9 0.01028 0.22616 0.72617 -0.00049 0.9523 -518.8 68.899 22.482 22 0.00473 0.18122 0.77827 0.00026 0.9595 -7351.8 291.08 11.254 (86.68) (78.35) (315.6) (-45.35) (66.91) (72.82) (308.8) (39.17) 10 0.00939 0.23597 0.67831 -0.00017 0.9143 22.5 141.54 26.558 23 0.00012 0.06969 0.92046 0.00014 0.9901 -8266.9 312.72 187.65* (55.86) (60.12) (164.7) (-11.77) (1.059) (76.38) (1391) (12.29) 11 0.00072 0.16175 0.82944 0.00015 0.9912 453.3 55.535 44.326* 24 0.00250 0.10942 0.87862 -0.00009 0.9880 -544.2 81.606 88.85* (3.877) (67.16) (445.5) (7.79) (46.14) (76.56) (811.6) (-19.34) 12 0.00469 0.14360 0.73746 0.00035 0.8811 39.1 118. 54 11.667 25 0.00364 0.19619 0.76311 0.00025 0.9593 -3307.3 1 40.68 3 3.035* (13.31) (24.45) (80.07) (12.0 4) (41.57) (87.88) (444.9) (25) 13 0.00056 0.19190 0.73938 0.00069 0.9313 -1929.7 175.14 9.363 26 0.01293 0.23519 0.68116 0.00013 0.9164 -8846.8 126.4 19.097 (7.969) (64.83) (259.7) (44.63) (37.37) (65.23) (185.8) (4.759) Average of absolute value (twenty-six individual stocks) 0.9365 4210.52

Note: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parentheses are t-statistics. (3) * represents significance levels of 5%

178

Table 5.12 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with contemporaneous bid/ask spread for twenty-six individual stocks on the Shanghai Stock rbbra tt= 011++− t

Exchange in percentage (2000 – 2002) for Equation 5.7: awttt==σε,.here ε tii.d.N(0,1) 2 aS22P σααt01=+t−−11 + βσγ1t + t 2 2 ID α0 α1 1 α1+ 1 LLF Q(20) Q (20) ID α0 α1 1 α1+ 1 LLF Q(20) Q (2 0)

1 0.04167 0.18863 0.75778 0.00573 0.9464 1502.3 255.35 46.715* 14 0.19420 0.30847 0.51627 0.02555 0.8247 -7998.3 204.83 15.778 (29.04) (80.33) (273.1) (26.97) (10750) (89.52) (214.3) (1557) 2 0.20268 0.27185 0.58049 0.03003 0.8523 -612.5 252.67 115.58* 15 0.24251 0.29156 0.48827 0.03206 0.7798 -9761.3 118.3 29.16 (71.75) (72.88) (159.6) (70.44) (119.5) (65.66) (143) (120.1) 3 0.15242 0.29579 0.53677 0.02076 0.8326 -1743.2 139.63 66.559* 16 0.04822 0.16690 0.79371 0.00610 0.9606 -6481.4 230.58 16.2 (84.65) (64.45) (153.3) (82.95) (107.2) (90.26) (935.1) (103.2) 4 0.13823 0.29350 0.56040 0.01765 0.8539 -2231.8 246.09 121.91* 17 0.12725 0.21939 0.67174 0.01687 0.8911 -5180.0 276.36 42.513* (100.7) (60.62) (162.7) (95.58) (78.12) (59.27) (195) (78.05) 5 0.21117 0.29226 0.52315 0.02973 0.8154 6234.0 185.92 111.32* 18 0.13335 0.24538 0.63282 0.01731 0.8782 -4100.4 204.2 8.246 (111.2) (60.87) (136.3) (109.2) (134.4) (62.92) (186.7) (135.4) 6 0.14716 0.28664 0.47941 0.01860 0.7660 2746.5 195.26 32.453* 19 0.20059 0.29440 0.50496 0.02584 0.7994 -6542.8 195.06 43.475* (130.1) (49.23) (112.2) (128.8) (206.2) (66.3) (155.6) (205.2) 7 0.13414 0.34343 0.50317 0.01640 0.8466 -3453.7 349.14 29.323 20 0.20007 0.28642 0.43766 0.02641 0.7241 -4331.3 96.675 30.795 (114.9) (78.76) (168.6) (111.9) (123.6) (69.97) (129.3) (121.8) 8 0.21179 0.32422 0.49356 0.02859 0.8178 -3908.8 197.35 0.931 21 0.19162 0.29183 0.40087 0.02570 0.6927 4769.6 147.94 48.932* (62.75) (69.59) (102.9) (60.35) (98.1) (71.41) (84.24) (96.24) 9 0.20346 0.32486 0.37709 0.02670 0.7020 1723.8 109.12 82.019* 22 0.12440 0.25402 0.65020 0.01572 0.9042 -6801.3 326.16 6.821 (107) (62) (80.14) (106.4) (141.6) (85.55) (416.9) (137.5) 10 0.16597 0.31595 0.42463 0.02179 0.7406 817.1 179.24 60.116* 23 0.19303 0.22114 0.65755 0.02641 0.8787 -7691.6 326.18 63.089* (71.65) (50.89) (72.97) (69.08) (60.84) (54.13) (149) (59.83) 11 0.11704 0.23303 0.68867 0.01645 0.9217 896.5 77.025 26.469 24 0.27989 0.27582 0.48413 0.03961 0.7600 681.5 129.95 134.59* (59.44) (67.75) (269.7) (58.68) (76.58) (59.34) (99.43) (75.79) 12 0.17928 0.19654 0.57359 0.02555 0.7701 180.1 132.87 39.877* 25 0.17723 0.35287 0.51765 0.02410 0.8705 -2365.4 169.04 32.276* (23.64) (23.93) (52.4) (22.88) (100.3) (91.09) (175.7) (97.2) 13 0.20844 0.28118 0.54438 0.02907 0.8256 -1325.5 2 01.88 17.083 26 0.32159 0.33415 0.41122 0.04385 0.7454 -7480.8 176.41 119.37* (74.35) (70.99) (193.7) (71.96) (90.01) (55.13) (77.2) (88.46) Average of absolute value (twenty-six individual stocks) 0.8231 3906.21

Note: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parentheses are t-statistics. (3) * represents significance levels of 5%

179

Table 5.13 AR(1)-GARCH(1,1) model estimation of intraday 5-minute returns with lagged-one bid/ask spread for twenty-six individual stocks on the Shanghai Stock Exchange rbbra tt= 011++− t in percenta ge (200 0 – 20 02) f or Equation 5.8: awttt==σε ,.hereε tii.d.N(0,1) 2 a22SP σt0=+αα1 t−−11 +βσ1t +γ t−1 2 2 ID α0 α1 1 α1+1 LLF Q(20) Q (20) ID α0 α1 1 α1+1 LLF Q(20) Q (20)

1 0.01384 0.15234 0.81536 0.00168 0.9677 1420.9 253.58 69.566* 14 0.08264 0.28070 0.57810 0.00909 0.8588 -9011.5 166.65 12.456 (13.53) (84.63) (425.4) (10.98) (46.9) (68.07) (124.6) (39.91) 2 0.08927 0.21031 0.70219 0.01282 0.9125 -1224.9 227.06 37.378* 15 0.11150 0.25394 0.61344 0.01352 0.8674 -10748 87.719 19.665 (39.91) (75.79) (242.7) (37.45) (52.5) (69.99) (167.3) (46.69) 3 0.05020 0.22449 0.69548 0.00632 0.9200 -2404.1 123.96 35.861* 16 0.00960 0.12351 0.85848 0.00090 0.9820 -6731.1 229.93 31.741* (42.86) (72.03) (234.4) (38.56) (27.37) (82.76) (924.4) (18.44) 4 0.00055 0.14638 0.81254 -0.00057 0.9589 -2721.4 225.05 28.781 17 0.03215 0.15715 0.80447 0.00394 0.9616 -5579.9 240.34 48.022* (0.7412) (64.38) (351.2) (-5.611) (25.26) (68.44) (392.6) (21.95) 5 0.08378 0.23508 0.68744 0.01158 0.9225 5181.0 154.73 16.529 18 0.05279 0.17218 0.77361 0.00663 0.9458 -4700.1 187.16 2.463 (61.06) (78.83) (272.5) (57.87) (54.39) (69.36) (295.8) (52.13) 6 0.03487 0.18029 0.74570 0.00412 0.9260 1788.6 139.99 3.334 19 0.01020 0.10075 0.87319 0.00093 0.9739 -7511.6 177.64 15.56 (44.13) (56.94) (264.5) (40.41) (20.59) (98.86) (1513) (12.77) 7 0.02201 0.23452 0.70708 0.00193 0.9416 -4128.3 301.93 12.802 20 0.08318 0.24013 0.62541 0.01025 0.8655 -5469.3 69.853 9.591 (30.86) (87.73) (290.1) (21.11) (52.36) (73.72) (192.3) (47.31) 8 0.09720 0.27511 0.60025 0.01227 0.8754 -4471.7 173.36 1.631 21 0.11246 0.26877 0.54752 0.01482 0.8163 3568.2 94.238 15.265 (32.24) (76.21) (141.6) (28.76) (69.58) (67.93) (122.3) (67.1) 9 0.10747 0.28039 0.55512 0.01369 0.8355 199.6 70.871 25.989 22 0.00090 0.17546 0.79532 -0.00073 0.9708 -7375.3 290.23 13.604 (97.68) (67.6) (183) (93.91) (1.351) (76.39) (335.2) (-8.483) 10 0.07332 0.25638 0.59797 0.00911 0.8543 192.3 146.66 20.222 23 0.07130 0.15086 0.79443 0.00950 0.9453 -8120.7 297.57 40.44* (41.16) (53.88) (116) (37.39) (37.56) (57.41) (263.2) (35.61) 11 0.04307 0.16844 0.80277 0.00594 0.9712 566.2 58.425 38.334* 24 0.10817 0.19020 0.70969 0.01512 0.8999 -164.0 88.963 34.959* (35.63) (67.61) (434.2) (34.09) (57.01) (61.99) (235) (55.45) 12 0.07401 0.16498 0.68244 0.00993 0.8474 59.9 121. 98 17.097 25 0.05961 0.23023 0.70261 0.00764 0.9328 -3152.7 143.8 25.298 (14.89) (24.05) (70.14) (13.5 7) (57.19) (100.1) (398.9) (52.4) 13 0.07610 0.23071 0.68239 0.01010 0.9131 -1909.6 168.34 10.907 26 0.17386 0.30491 0.53611 0.02276 0.8410 -8500.2 140.09 28.971 (53.88) (67.69) (218.1) (50.36) (63.36) (60.59) (106.2) (59.2) Average of absolute value (twenty-six individual stocks) 0.9118 4111.60

Note: (1) All coefficients of α1 and β1 of the twenty-six individual stocks are significant at the 5% level. (2) Numbers in parentheses are t-statistics. (3) * represents significance levels of 5%

180

Appendix 5.1 Twenty-six Chinese Individual Stocks on the Shanghai Stock Exchange

ID Code Name

1 600001 Handan Iron & Steel Co., Ltd. 2 600005 Wuhan Steel Processing Co., Ltd. 3 600006 Dongfeng Automobile Co., Ltd. 4 600008 Beijing Capital Co., Ltd. 5 600009 Shanghai International Airport Co., Ltd. 6 600018 Shanghai Port Container Co., Ltd 7 600171 Shanghai Belling Corp., Ltd 8 600188 Yanzhou Coal Mining Company Limited 9 600236 Guangxi Guiguan Electric Power Co., Ltd. 10 600332 Guangzhou Pharmaceutical Company Limited 11 600377 Jiangsu Expressway Company Limited 12 600585 Anhui Conch Cement Company Limited 13 600600 Tsingtao Brewery Company Limited 14 600618 Shanghai Chlor-alkli Chemical Co., Ltd. 15 600623 Shanghai Tyre & Rubber Co., Ltd. 16 600637 SVA Information Industry Co., Ltd. 17 600639 Shanghai Jinqiao Export Processing Zone Development Co., Ltd. 18 600642 Shenergy Company Limited 19 600648 Shanghai Wai Gaoqiao Free Trade Zone Development Co., Ltd. 20 600726 Heilongjiang Electric Power Company Limited 21 600736 Suzhou New District Hi-Tech Industrial Co., Ltd 22 600776 Eastern Communications Co., Ltd. 23 600816 Anshan Trust & Investment Co., Ltd 24 600863 Inner Mongolia MengDian HuaNeng Thermal Power Co., Ltd 25 600867 Tonghua Dongbao Medicines Co., Ltd 26 600874 Tianjin Capital Environmental Protection Company Limited

181

Chapter 6

Summary and Conclusions

This thesis focuses on market microstructure issues in the Chinese stock market in three Chapters (three parts): 3, 4 and 5. Studies based on intraday stock market indices and individual stock data are applied for the limited order-driven market in China. Though literature on the market microstructure in the developed countries has been widely documented, the thesis provides a uniquely complete exploration of the emerging market of China. Discussions in the thesis may be of interest to local investors, foreign investors and academic researchers and government regulators.

Chapter 3 (the first part) examines the behavior of intraday return and return volatility. Based on the unique trading system, the statistical properties and systematic characteristics of intraday returns and return volatility are explored using

5-minute data of the Shanghai Composite Index and the Shenzhen Component Index over a three-year period. The three subperiods: year 2000, year 2001 and year 2002, and the relatively finer and wider frequency data of 1-minute and 10-minute interval return series, are also used. These five different processes indicate that there exists a systematic intraday variation for each subperiod, and each 1-minute and 10-minute period, consistent with the intraday 5-minute return characteristics. The chapter also discusses the return variance in the active trading period (open-to-close) of the stock

182 market, which is larger than that in the nontrading period (close-to-open). In final section of the chapter, volatilities of interday 24-hour returns based on 5-minute intervals are calculated and the variance ratio test is used to show that high volatility of intraday returns at the market opening is not completely due to the trading mechanism (call auction) but the overnight trading halt.

Chapter 4 (the second part) examines the intraday behaviors of the bid/ask spread and depth and their determinants on the order-driven market in the Shanghai

Stock Exchange. As spread and depth are two dimensions of market liquidity, both variables play important roles in explaining the market liquidity and their components. Three competing theories have been proposed in the US market to explain the intraday liquidity patterns: order-processing, asymmetric information, and inventory holding costs. The analysis of the Chinese stock market shows empirical findings that are different from those of other major exchanges such as the

US market. In particular, the intraday 5-minute bid/ask spread displays an L-shaped pattern and depth exhibits an inverted L-shaped pattern. The striking L-shaped pattern of the bid/ask spreads has a positive relationship with volatility but a negative relationship with stock price and trading volume while the striking reverse L-shaped pattern of the depths has a negative relationship with stock price, volatility but a positive relationship with trading volume in the regression model. The findings also indicate that there is an evident negative relationship between the spread and the depth at the opening of the trading day. Overall, the study suggests that the factor of information asymmetries, through time and across traders, plays a key role in generating observed liquidity variations.

Chapter 5 (the third part) explores the issue of whether the return volatility in high-frequency data can be described by the GARCH(1,1) specification, and whether

183 GARCH model captures the effects of temporal dependence in trading volume or bid/ask spreads for individual stocks in the Chinese stock market after the inclusion of intraday trading volume or bid/ask spread as a mixing variable for information arrival in the conditional variance model. The findings indicate that the AR(1)-

GARCH(1,1) model successfully accounts for the nonlinear dependencies (volatility clustering) in the individual stocks. However, the AR(1)-GARCH(1,1) model may not account for the linear dependencies (serial correlation) in the high-frequency data.

Based on the MDH hypothesis, the inclusion of intraday contemporaneous and lagged trading volume or bid/ask spread as a mixing variable for information arrival in the GARCH model helps in explaining the GARCH effects, but the

GARCH effects do not disappear. In other words, the persistence in volatility remains in the intraday return series. Thus, there should be other major factors affecting the volatility change in intraday returns, which are areas for further research. The findings in this chapter suggest that volume, or bid/ask spread, as an information variable, has quite a limited effect on the volatility of intraday returns in the Shanghai stock market, which is similar to the results in the US market.

Information does not explain the volatility that may result from noise trading in the market and the observed results contradict the findings of Lamoureux and Lastrapes

(1990). It could also suggest that the volume or bid/ask spread as an information variable is not a very important factor for predicting the volatility of stock returns in intraday levels, irrespective of how mature or efficient a market is.

The study in this thesis is subject to some limitations. First, due to problems limiting data collection, this study lacks detailed transaction data when I explore the intraday characteristic of the Chinese stock market. Intraday trading volume data for

184 the Shanghai index and the Shenzhen index was also unavailable. This made it impossible to compare simultaneous relationship between index and volume, as well as the GARCH effects with individual stock and market indices.

The second weakness is from the analysis of the GARCH model used in this study. Currently, many different GARCH models such as IGARCH, EGARCH and

APARCH are discussed in the financial literature. I used AR(1) in the mean equation for the GARCH model. Actually, the autoregressive integrated moving average

(ARIMA) models can be used in the GARCH mean equation. These give us a number of choices to analyse intraday return volatility. A comparison of these different models can be useful for analysis of the Chinese stock market. While considering the modelling volatility using GARCH, seasonal adjustments to the return serials should be used to improve the GARCH effects. Two ways to process the data can be found from the literature (Andersen and Bollerslev, 1997 and Cho,

Russell, Tiao and Tsay, 2003). Furthermore, extensions of the GARCH model can be used to analyse the possible asymmetric effects, and the effects of stock price regulation.

It should be noted that the Chinese stock market is a relatively underdeveloped market due to its short history (about fifteen years). Speculative activities and price manipulations, government interventions and policy effects in stock markets may distort market behaviour. What is more, most of the companies listed are state-owned, and a large part of the outstanding shares are controlled by the government and are not tradable. This is always an important factor that affects the behaviour of the market.

Given these weaknesses, there is a need for ongoing research is concerned, using intraday transaction data. As far as market microstructure, further research on

185 (1) the trading mechanism of the call auction and continuous trading, (2) the effects of converting non-tradable shares into tradable shares, and (3) the effect of changing the price limit, are directions for further research. Furthermore, the trading behaviour of institutional investors, and its effect on the intraday temporal relations of intraday volatility, bid/ask spread and trading volume should be explored. Finally, examining components of bid/ask spread would be a third direction for liquidity research of market microstructure.

186

Bibliography

Admati, A. and Pfleiderer, P. (1988) A theory of intraday patterns: volume and price

variability, Review of Financial Studies, 1, 3-40

Ahn, H. J. and Cheng, Y. (1999) The intraday patterns of spreads and depth in a

market without market makers. The stock exchange of Hong Kong, Pacific-

Basin Finance Journal, 17, 539-556

Alexander, C. (2001) Market Models: A Guide to Financial Data Analysis, John

Wiley & Sons Publisher

Amihud, Y. and Mendelson, H. (1987) Trading mechanism and stock returns: an

empirical investigation, Journal of Finance, 42, 533-555

Amihud Y, and Mendelson, H. (1988) Liquidity and Asset Prices: Financial

Management Implications, Financial Management, 17, 5-15

Amihud, Y. and Mendelson, H. (1991) Efficiency and trading: evidence from the

Japanese Stock Market, Journal of Finance, 46 1765-1790

Andersen, T. G. (1996) Return Volatility and Trading Volume: An Information Flow

Interpretation of Stochastic Volatility, Journal of Finance, 51(1), 169-204

Andersen, T. G. and Bollerslev, T. (1997) Intraday periodicity and volatility

persistence in financial markets, Journal of Empirical Finance, 4, 115-158

Andersen, T. G. and Bollerslev, T. (1998) Deutschemark-dollar volatility: intraday

activity patterns, macroeconomic announcements, and longer-run

dependencies, Journal of Finance, 53, 219-265

187 Baillie, R. T. and Bollerslev, T. (1991) Intra-day and inter-market volatility in

foreign exchange rates, Review of Economic Studies, 58, 568-585

Barclay, M., Litzenberger, R. and Warner J. (1990) Private Information, Trading

Volume, and Stock-Return Variances, The Review of Financial Studies, 3,

233-253

Barclay, M. and Smith, C., Jr. (1988) Corporate payout policy: Cash dividends

versus open market repurchases, Journal of Financial Economics, 22, 61-82

Benston, G. and Hagerman, R. (1974) Determinants of bid-ask spreads in the over-

the-counter market, Journal of Financial Economics, 1, 353-364

Bernstein, P. (1987) Liquidity, Stock Markets and Market Makers, Financial

Management, Summer, 1987, pp. 54-62

Bildik, R. (2001) Intra-day seasonalities on stock returns: evidence from the Turkish

Stock Market, Emerging Markets Review, 2, 387-417

Black, F. (1986) Noise, Journal of Finance, 3, 529-543

Bohl, M. T. and Henke, H. (2003) Trading volume and stock market volatility: The

Polish case, International Review of Financial Analysis, 12, 513-525

Bollerslev, T. (1986) Generalized autoregressive conditional heterscedasticity,

Journal of Econometrics, 31, 307-27

Bollerslev, T., (1987) A Conditionally Heteroskedastic Time Series Model for

Speculative Prices and Rates of Return, Review of Economics and Statistics,

69, 542-547

Bollerslev, T. and Domowitz, I. (1991) Price Volatility, Spread Variability and the

Role of Alternative Markets Mechanisms, Review of Futures Markets, 10, 78-

102

188 Bollerslev, T. and Domowitz, I. (1993) Trading patterns and prices in the interbank

foreign exchange market, Journal of Finance, 48, 1421-1443

Brock, W. A. and Kleidon, A. W. (1992) Periodic market closure and trading

volume, Journal of Economic Dynamics and Control, 16, 451-489

Brockman, P. and Chung, D. Y. (1998) Inter- and intra-day liquidity patterns on the

Stock Exchange of Hong Kong, Journal of International Financial Markets,

Institutions and Money, 8, 279-300

Brockman, P. and Chung, D. Y., (1999) An analysis of depth behavior in an

electronic, order-driven environment, Journal of Banking & Finance, 23,

1861-1886

Brown, M. B. and Forsythe, A. B. (1974) Robust Tests for the Equality of Variances,

Journal of the American Statistical Association, 69, 364–367

Campbell, J. Y., Lo, A. W. and MacKinlay, A. C. (1997) The Econometrics of

Financial Markets, Princeton University Press

Chan K., Chan, K. C. and Karolyi G. A. (1991) Intraday Volatility in the Stock Index

and Stock Index Futures Markets, Review of Financial Studies, 4, 657-684

Chan, K., Chung, Y. P. and Johnson, H. (1995) The intraday behavior of bid-ask

spreads for NYSE stocks and CBOE options, Journal of Financial and

Quantitative Analysis, 30, 329-346

Chan, L. K. C. and Lakonishok, J. (1993) Institutional Trades and Intraday Stock

Price Behavior, Journal of Financial Economics, 33, 173-200

Chan, Y. C. (2000) The price impact of trading on the Stock Exchange of Hong

Kong, Journal of Financial Markets, 3, 1-16

189 Chang, R. P., Fukuda, T., Rhee, S. G. and Takano, M. (1993) Intraday and interday

behavior of the TOPIX, Pacific-Basin Finance Journal, North-Holland, 1,

67-95

Cheung, Y., (1995) Intraday returns and the day-end effect: evidence from the Hong

Kong Equity Market, Journal of Business Finance and Accounting, October,

1023-1035.

Cheung, Y. L., Ho, R., Yan-ki, P. P., and Draper, P. (1994) Intraday stock return

volatility: The Hong Kong evidence, Pacific-Basin Finance Journal, 2, 261-

276

China Securities Regulatory Committee (CSRC)

http://www.csrc.gov.cn/en/homepage/index_en.jsp

Choe, H. and Shin, H. K. (1993) An analysis of interday and intraday return volatility

evidence from the Korea Stock Exchange, Pacific-Basin Finance Journal, 1,

175-188

Chung, K. H., Van Ness, F. and Van Ness, R. (1999) Limit orders and the bid-ask

spread, Journal of Financial Economics, 53, 255-287

Clark, P. K., (1973) A Subordinated Stochastic Process Model with Finite Variance

for Speculative Prices, Econometrica, 41, 135-156

Conover, W. J., Johnson, M. E. and Johnson, M. M. (1981) A comparative study of

test for homogeneity of variances, with applications to the outer continental

shelf bidding data, Technometrics, 23, pp. 351-61

Copeland, T. (1976) A Model of Asset Trading under the Assumption of Sequential

Information Arrival, Journal of Finance, 31, 1149-1168

190 Copeland, T. and Galai, D. (1983) Information effects on bid/ask spread, Journal of

Finance, 38, 1457-1469

Copeland, L. and Jones, S. A. (2002) Intradaily Patterns in the Korea Index Futures

Market, Asian Economic Journal, 16, 2, 153-174

Darrat, A., Rahman, S. and Zhong, M. (2003) Intraday trading volume and return

volatility of the DJIA stocks: A note, Journal of Banking & Finance, 27,

2035-2043

Demsetz, H., (1968) The Cost of Transacting, Quarterly Journal of Economics, 82,

33-53

Diebold, F. and Nerlove, M. (1989) The dynamics of exchange rate volatility: a

multivariate latent factor ARCH model, Journal of Applied Econometrics,

4(1), 1-22

Easley, D. and O'Hara, M. (1992) Adverse selection and large trade volume: the

implications for market efficiency, Journal of Financial and Quantitative

Analysis, 27, 185-208

Engel R. F. (1982) Autoregressive conditional heteroscedasticity with estimates of

variables of UK inflation, Econometrica, 50, 987-1008

Engle, R. F., (1993) Statistical Models for Financial Volatility. Financial Analysts

Journal, 49, 72-78

Engle, R. F. and Bollerslev, T. (1986) Modeling the Persistence of Conditional

Variances, Econometric Reviews, 5, 1-50

Engle, R. F., Ito, T. and Lin, W. L. (1990) Meteor Showers or Heat Waves?

Heteroskedastic Intra-Daily Volatility in the Foreign Exchange Market,

Econometrica, 58, 525-542

191 Epps, T. and Epps, M. (1976) The Stochastic Dependence of Security Price Changes

and Transaction Volumes: Implications for the Mixture of Distributions

Hypothesis, Econometrica, 44, 305-321

Fama, E. F. (1976) Efficient capital markets: reply, Journal of Finance, 3, 143-145

Flood, M. D. (1991) Microstructure theory and foreign exchange market, Federal

Reserve Bank of St. Louis, 73(6), 52-70

Foster, A. (1995) Volume-Volatility Relationships for Crude Oil Futures Markets,

Journal of Futures Markets, 15, 929-951

Foster, F. D. and Viswanathan, S. (1990) A theory of interday variations in volumes,

variances and trading costs in securities markets, Review of Financial Studies,

3, 593-624

Franz, D. R., Rao, R. P. and Tripathy, N. (1995) Informed trading risk and bid-ask

spread changes around open market stock repurchases in the NASDAQ

market, Journal of Financial Research, 18, 311-327

French, K. R. and Roll, R. (1986) Stock Return Variances; the Arrival of Information

and the Reaction of Traders, Journal of Financial Economics, 17, 5-26

Gallo, G. M., and Pacini, B. (2000) The effects of trading activity on market

volatility, The European Journal of Finance, 6, 163-175

Gerety M. S. and Mulherin J. H. (1994) Price Formation on Stock Exchanges: the

Evolution of Trading within the Day, Review of Financial Studies, 7, 609-629

Glosten, L. R., and Milgrom, P. (1985) Bid/Ask and Transaction Prices in a

Specialist Market with Heterogeneously Informed Agents, Journal of

Financial Economics, 14, 71-100

192 Goodhart, C. A. E. and O’Hara, M. (1997) High frequency data in financial markets:

issues and applications, Journal of Empirical Finance, 4, 73-114

Grossman S. J. and Stiglitz J. E. (1980) On the Impossibility of Informationally

Efficient Markets, American Economic Review, 70, 393-408

Grossman, S. and Miller, M. (1988) Liquidity and Market Structure, Journal of

Finance, 43, 617-633

Hamao, Y., Masulis, R. W. and Ng, V. (1990) Correlations in price changes and

volatility across international stock markets, Review of Financial Studies, 3,

281-307

Hamilton, J. L. (1978) Marketplace organization and marketability: NASDAQ, the

stock exchange and the national market system, Journal of Finance, 33, 487-

503.

Harris, L. (1986) A Transaction Data Study of Weekly and Intraday Patterns in Stock

Returns, Journal of Financial Economy, 16, 99-117

Hasbrouck, J. (1988) Trades, quotes, inventories and information, Journal of

Financial Economics, 22, 229-252

Ho, T. and Stoll, H. R. (1981) Optimal dealer pricing under transaction and return

uncertainty, Journal of Financial Economics, 9, 47-73

Jain, P. C. and Joh, G. H. (1988) The Dependence between Hourly Prices and

Trading Volume, Journal of Financial and Quantitative Analysis, 23, 269-

283

Jarque, C. M. and Bera, A. K. (1982) Model Specification Tests: A Simultaneous

Approach. Journal of Econometrics, 20 (1), 59-82

193 Kim, K. A. (2001) Price Limits and Stock Market Volatility, Economics Letters, 71,

131-136

Kim, D. and Kon, S. (1994) Alternative models for the conditional heteroskedasticity

of stock returns, Journal of Business, 67, 563-588

Kim, Y. and Shin, J. (2000) Interactions among China-related stocks, Asia-Pacific

Financial Markets, 7, 97-115

Kyle, A. S. (1985) Continuous auctions and insider trading, Econometrica, 53(6),

1315-1335

Kyle, A. S. (1988) Trading halts and price limits, The Review of Futures Markets, 7,

426-434

Lam, P. H. L. and Tong, W. H. S. (1999) Interdaily Volatility in a Continuous Order-

Driven Market, Journal of Business Finance and Accounting, 26, 7/8, 1013-

1036

Lamoureux C. G. and Lastrapes W. D. (1990) Heteroskedasticity in Stock Return

Data: Volume versus GARCH Effects, Journal of Finance, 45, 221-229

Lee, C., Mucklow, B. and Ready, M. (1993) Spreads, depths, and the impact of

earnings information: an intraday analysis. Review of Financial Studies, 6,

345-374

Lee, C. F., Chen, G. M. and Rui, O. M. (2001) Stock returns and volatility on

China’s stock markets, The Journal of Financial Research, Vol. XXIV, No. 4

Winter, 523-543

Lehman, B. N. and Modest, D. M. (1994) Trading and liquidity on the Tokyo Stock

Exchange: A bird's eye view, Journal of Finance, 44, 951-984

194 LeRoy, S. F. (1980) Efficient capital markets and martingales, Journal of Economic

Literature, XXVII, 1583-1621

Levene, H. (1960) Robust Tests for the Equality of Variances, in I. Olkin, S. G.

Ghurye, W. Hoeffding, W. G. Madow, and H. B. Mann (eds.), Contribution

to Probability and Statistics, Stanford University Press

Liu, L, Liu, C. X., Li, W. G. and Zhao, Y. G. (2000) An empirical study on intraday

price and volume models of A share between Shanghai and Shenzhen stock

market, Economic Science, Number 1, 52-65

Lo, A. W. and MacKinlay, A. C. (1990) An econometric analysis of non-

synchroneous trading, Journal of Econometrics, 45, 181-212

Locke, P. R. and Sayers, C. L. (1993) Intra-day futures price volatility: Information

effects and variance persistence, Journal of Applied Econometrics, 8, 15-30

Lockwood, L. J. and Linn, S. C. (1990) An examination of stock market return

volatility during overnight and intraday periods, 1964-1989, Journal of

Finance, 45, 591-602

Madhavan, A., Richardson, M. and Roomans, M. (1997) Why do security prices

change? A transaction-level analysis of NYSE stocks, Review of Financial

Studies, 10, 1035-1064

McInish, T. and Wood, R. (1992) An Analysis of Intraday Patterns in Bid-Ask

Spreads for NYSE Stocks, Journal of Finance, 47, 753-764

McInish, T. H and Ness, B. F. (2002) An Intraday Examination of the Components

of the Bid-Ask Spread, The Financial Review, 37, 507-524

Melvin, M. and Yin, X. X. (2000) Public Information Arrival, Exchange Rate

Volatility and Quote Frequency, Economic Journal, 110, (465), July, 644-661

195 Mian, G. M. and Adam, C. M. (2001) Volatility dynamics in high frequency

financial data: an empirical investigation of the Australian equity returns,

Applied Financial Economics, 11, 341-352

Miller, E. (1989) Explaining intraday and overnight price behaviour, Journal of

Portfolio Manager, Summer, 10-99

Miyakoshi T.(2002) ARCH versus information-based variances: evidence from the

Tokyo Stock Market, Japan and the World Economy, 14, 215-231

Mucklow, B. (1994) Market microstructure: An Examination of the Effects on

Intraday Event Studies, Contemporary Accounting Research, Vol. 10 No. 2

Spring 355-382

Najand, M. and Yung, K. (1991) A GARCH Examination of the Relationship

between Volume and Price Variability in Futures Markets, Journal of Futures

Markets, 11, 613-621

Niarchos, N. A. and Alexakis, C. A. (2003) Intraday stock price patterns in the Greek

stock exchange, Applied Financial Economics, 13, 13-22

Noronha, G., Sarin, A. and Saudagaran, S. (1996) Testing for liquidity effects of

international dual listings using intraday data, Journal of Banking and

Finance 20, 965-983

O’Hara, M. (1995) Market Microstructure Theory, Cambridge: Blackwell Publishers

Omran, M. F., and McKenzie, E. (2000) Heteroskedasticity in stock returns data

revisited: volume versus GARCH effects, Applied Financial Economics, 10,

553-560

196 Ozenbas, D., Schwartz, R. A, and Wood, R. A. (2002) Volatility in US and European

Equity Markets: An Assessment of Market Quality, International Finance,

5:3, 437-461

Rahman, S., Lee, C. F. and Ang K. P. (2002) Intraday return volatility process:

evidence from NASDAQ stocks, Review of Quantitative Finance and

Accounting, 19: 155-180

Schwartz, R. A. (1988) Equity markets: Structure, trading and performance, Harper

and Row, New York

Shanghai Stock Exchange, (2003) Fact book,

http://www.sse.com.cn/sseportal/en_us/ps/home.shtml

Sharma, J. and Mougoue, M. (1996) Heteroskedasticity in stock market indicator

return data: volume versus GARCH effects, Applied Financial Economics, 6:

337-342

Shenzhen Stock Exchange, (2003) Fact book,

http://www.szse.cn/main/en/catalog_1697.aspx

Sheskin, D. J. (1997) Parametric and Nonparametric Statistical Procedures, CRC

Press

Shiller, R. (1981) Do Stock Prices Move Too Much to be Justified by Subsequent

Changes in Dividends?, American Economic Review, 3, 421-436

Smirlock, M. and Starks, L. (1985) A further examination of stock price changes and

transaction volume, Journal of Financial Research, 8, 217-225

Sohu.com Inc, http://www.sohu.com

Stoll, H. (1978) The pricing of security dealer services: an empirical study of

NASDAQ stocks, Journal of Finance, 33, 1153-1172

197 Stoll, H. R. and Whaley, R. E. (1990) Stock Market Structure and Volatility, Review

of Financial Studies, 3, 37-71

Tang, G. Y. N. and Lui, D. T. W. (2002) Intraday and intraweek volatility patterns of

Hang Seng Index and index futures, and a test of the wait-to-trade hypothesis,

Pacific-Basin Finance Journal, 10, 475-495

Tauchen, G. and Pitts, M. (1983) The Price Variability-Volume Relationship on

Speculative Markets, Econometrica, 51, 485-505

Tinic, S. M. (1972) The economics of liquidity services, Quarterly Journal of

Economics 66, 79-97

Walsh, D. and Quek, J. (1999) An Empirical Examination of the SIMEX Nikkei 225

Futures Contract around the Kobe Earthquake and the Barings Bank

Collapse, Journal of Futures Markets 19, 1-29

Wang, G. and Yau, J. (2000) Trading Volume, Bid-Ask Spread, and Price Volatility

in Futures Markets, Journal of Futures Markets 20, 943-970

Wood, R. A., McInish, T. H. and Ord, J. K. (1985) An investigation of transactions

data for NYSE stocks, Journal of Finance, 40, 723-739

198