Market Microstructure Studies: Liquidity, Price Discovery and Manipulation Ching (Jane) Chau University of Wollongong

University of Wollongong Research Online

University of Wollongong Thesis Collection University of Wollongong Thesis Collections

2013 Market microstructure studies: liquidity, price discovery and manipulation Ching (Jane) Chau University of Wollongong

Recommended Citation Chau, Ching (Jane), Market microstructure studies: liquidity, price discovery and manipulation, Doctor of Philosophy thesis, School of Accounting and Finance, University of Wollongong, 2013. http://ro.uow.edu.au/theses/3921

Research Online is the open access institutional repository for the University of Wollongong. For further information contact the UOW Library: [email protected]

MARKET MICROSTRUCTURE STUDIES:

LIQUIDITY, PRICE DISCOVERY AND MANIPULATION

A thesis submitted in fulfilment of the requirements for the award of the degree

DOCTOR OF PHILOSOPHY

From

UNIVERSITY OF WOLLONGONG

Ching (Jane) Chau

Bachelor of Commerce Honours (Class 1) in Accountancy

School of Accounting and Finance, Faculty of Commerce Australia June 2013

CERTIFICATION

I, Ching Chau, declare that this thesis, submitted in partial fulfilment of the requirements for the award of Doctor of Philosophy, in the School of Accounting and Finance of the Faculty of Commerce, University of Wollongong, is wholly my own work unless otherwise referenced or acknowledged. The document has not been submitted for qualifications at any other academic institution.

Ching (Jane) Chau

June, 2013

DEDICATION

To the memory of my mother, Yuduo Huang (January 1944 – May 2013)

To my father, who cares for my mother with his whole heart and love.

ACKNOWLEDGEMENTS

Many people have made valuable contributions to this research. Without their support and encouragement, it would have been difficult for me to complete this thesis. I would like to take this opportunity to gratefully acknowledge those whose contributions have been instrumental in the successful completion of this thesis.

First and foremost, I would like to express my sincere thanks to my supervisors,

Professor Gary Tian and Dr Shiguang Ma. Their constant support, humble attitude, deep knowledge and professional advice throughout my PhD candidature have been invaluable to the completion of this thesis.

I am deeply indebted to Professor Alex Frino, who introduced me to the Capital

Market CRC PhD Research Program. His immense support and encouragement during the initial years of my PhD have enabled me to pursue and complete this challenging journey. I would also like to thank Professor Helen Irvine and Dr

Aelee Jun for suggestions, motivation and encouragement.

This research has also benefited greatly from the thoughtful insights and feedback provided by Rick Harris, Sean Foley, George Li, Angelo Aspris, Jing Gao,

Andrew Lepone, Wangchun Wei, Dionigi Gerace, Bart Frijns and Aaron Gilbert. I appreciate very much the valuable input of each of them into my journey.

I gratefully acknowledge financial support provided by the Capital Market

Cooperative Research Centre for a four-year full scholarship. Without this

financial support I would have been unable to help my brother with study and my mother with hospital expenses. I also thank the Securities Industry Research

Centre of Australia (SIRCA) for providing the data used in this thesis.

A special thank you goes to my family and friends. I am especially grateful to

Noeline Wiggins for her love and wholehearted support. I appreciate Xinjun

Wang for his regular C# programming help and for his warmth and friendliness during my PhD process. Finally, I have dedicated this thesis to my parents who have been a constant source of encouragement.

SYNOPSIS

This thesis offers an original way to examine three specific issues in market microstructure: liquidity, price discovery and price manipulation. The purpose of this thesis is to provide empirical evidence on these issues of significance to exchange regulators in designing market structure.

This thesis is structured with an introductory chapter, a theoretical chapter, three empirical analysis chapters and a concluding chapter. Chapter 1 discusses the importance of the study, and identifies research motivation and contributions. It also provides a brief description of the structure and functions of financial markets.

Chapter 2 presents a theoretical framework of the study by looking at several key areas in market microstructure theory.

Chapters 3 to 5 address three research questions relating to market structure effects on liquidity, price discovery and price manipulation, respectively. First,

Chapter 3 examines the liquidity impact of market structure change from a transparent market to an anonymous market, in the trading of cross-listed stocks on both the Australian Stock Exchange (ASX) and New Zealand Stock Exchange

(NZX). Results show that spreads decline, quoted depth and trading volume increase with the introduction of an anonymous market, after controlling for both stock-specific and market-wide liquidity factors. Anonymity attracts the trading of cross-listed stocks from the foreign counterparty.

Chapter 4 further examines the impact of market structure change on the price

discovery process in the context of ASX and NZX. It finds compelling evidence that trader anonymity improves the price discovery process. Information share improves on ASX, but deteriorates on NZX, after ASX switched to anonymous trading. On the other hand, information share increases on NZX, but decreases on

ASX, after NZX adopted anonymous trading. These results also add evidence to the prior literature on the choice of anonymous market by informed traders.

Chapter 5 examines price manipulation with a specific reference to the impact of the trading mechanism in the context of the Hong Kong Stock Exchange (HKEx).

It is found that the trading mechanism determines the techniques used to manipulate price. A new form of closing price manipulation, quote-based manipulation, is facilitated by the closing mechanism of HKEx. Closing price can be manipulated solely through quotes without trading, and hence without cost.

The manipulator is able to inflate (deflate) closing prices through placing orders to buy (sell) small quantities of shares at prices higher (lower) than the prevailing market prices near the market close.

TABLE OF CONTENTS

CHAPTER 1

INTRODUCTION ...... 17

1.1 Background and Motivation ...... 17

1.1.1 Market Microstructure Study ...... 18

1.1.2 Why Market Microstructure Matters ...... 19

1.2 Purpose and Contributions ...... 21

1.3 Structure of This Thesis ...... 25

References ...... 27

CHAPTER 2

THEORETICAL FRAMEWORK OF THIS THESIS ...... 30

2.1 Introduction ...... 30

2.2 Market Liquidity ...... 32

2.3 Price Discovery ...... 34

2.4 Price Manipulation ...... 37

2.5 Summary ...... 41

References ...... 42

CHAPTER 3

IMPACT OF ANONYMITY ON LIQUIDITY IN LIMIT ORDER BOOKS: EVIDENCE FROM CROSS-LISTED STOCKS ...... 47

3.1 Abstract ...... 47

3.2 Introduction ...... 48

3.3 Literature Review and Hypotheses ...... 50

3.4 ASX versus NZX and Market Structure Change ...... 57

3.5 Data and Summary Statistics ...... 60

3.6 Research Methodology ...... 70

3.6.1 Univariate Analysis ...... 70

3.6.2 Fixed Effect Instrumental Variable 2SLS Estimation ...... 74

3.7 Results ...... 80

3.7.1 Liquidity Impact of ASX Anonymity ...... 80

3.7.2 Liquidity Impact of NZX Anonymity...... 86

3.8 Robustness Check ...... 91

3.8.1 Length of Event Window...... 91

3.8.2 Difference-in-difference Estimation ...... 98

3.9 Summary ...... 107

References ...... 110

Appendices ...... 115

CHAPTER 4

IMPACT OF ANONYMITY ON PRICE DISCOVERY: A NATURAL EXPERIMENT ...... 117

4.1 Abstract ...... 117

4.2 Introduction ...... 117

4.3 Literature Review and Hypotheses ...... 120

4.4 Data and Sampling ...... 124

4.4.1 Data and Study Period ...... 124

4.4.2 Pairing ASX and NZX Trades ...... 130

4.4.3 Unit Roots and Cointegration ...... 132

4.5 Research Methodology ...... 135

4.5.1 Hasbrouck Information Share ...... 135

4.5.2 The ECM and the Measure of Contribution to Price Discovery ...... 138

4.6 Results ...... 141

4.6.1 Change in Hasbrouck Information Share ...... 141

4.6.2 Change in Information Share from ECM ...... 147

4.7 Summary ...... 152

References ...... 154

Appendices ...... 159

CHAPTER 5

QUOTE-BASED PRICE MANIPULATION: THE HONG KONG EVIDENCE ...... 162

5.1 Abstract ...... 162

5.2 Introduction ...... 163

5.3 Hong Kong Market and the Closing Procedure ...... 166

5.3.1 Market Structure ...... 167

5.3.2 The Closing Procedure and Quote-based Manipulation ...... 169

5.4 Data ...... 171

5.4.1 A Typical Example of Quote-based Manipulation Scheme ...... 172

5.4.2 Trading Characteristics of Manipulated Stocks ...... 176

5.5 Literature Review and Hypotheses ...... 177

5.6 Research Methodology ...... 184

5.6.1 Difference-in-difference Estimation ...... 184

5.6.2 Variables Measuring Trading and Quoting Characteristics...... 187

5.6.3 Further Analysis: A Benchmark Approach ...... 188

5.7 Results ...... 193

5.7.1 Manipulation Effects on Trading Characteristics ...... 193

5.7.2 Manipulation Effects on Quoting Characteristics...... 198

5.7.3 Long-run Effects of Quote-based Manipulation ...... 209

5.8 Implications ...... 213

5.8.1 The Persistent Risk of Quote-based Manipulation in HKEx ...... 213

5.8.2 Potential Solutions for Illiquid Securities ...... 215

5.9 Summary ...... 218

References ...... 221

Appendices ...... 225

CHAPTER 6

CONCLUSION ...... 229

6.1 How Anonymous Trading Affects Liquidity Migration of Cross-listed Stocks ...... 229

6.2 How Anonymous Trading Affects the Price Discovery Process ...... 230

6.3 How Trading Mechanism Affects Price Manipulation ...... 231

6.4 Avenues for Future Research ...... 232

6.5 Implications for Economic Growth and Policy ...... 233

References ...... 236

LIST OF TABLES

Table 2-1: The ASX’s SEATS Screen (prior to 28 November 2005)...... 31

Table 3-1: Sample Selection for ASX Anonymity ...... 62

Table 3-2: Sample Selection for NZX Anonymity ...... 63

Table 3-3: Trading Statistics of Final Sample Stocks for ASX Anonymity ...... 66

Table 3-4: Trading Statistics of Final Sample Stocks for NZX Anonymity ...... 67

Table 3-5: Univariate Analysis of ASX Anonymity ...... 72

Table 3-6: Univariate Analysis of NZX Anonymity...... 73

Table 3-7: Summary Statistics for the Sample and Matched Stocks ...... 79

Table 3-8: 2SLS Results – Impact of ASX Anonymity on ASX Liquidity ...... 81

Table 3-9: 2SLS Results – Impact of ASX Anonymity on NZX Liquidity ...... 85

Table 3-10: 2SLS Results – Impact of NZX Anonymity on ASX Liquidity ...... 88

Table 3-11: 2SLS Results – Impact of NZX Anonymity on NZX Liquidity ...... 90

Table 3-12: Sensitivity to Event Window – Impact of ASX Anonymity on ASX Liquidity ...... 93

Table 3-13: Sensitivity to Event Window – Impact of ASX Anonymity on NZX Liquidity ...... 94

Table 3-14: Sensitivity to Event Window – Impact of NZX Anonymity on ASX Liquidity ...... 96

Table 3-15: Sensitivity to Event Window – Impact of NZX Anonymity on NZX Liquidity ...... 97

Table 3-16: Univariate Analysis of ASX Anonymity during Normal Trading Hours...... 101

Table 3-17: Univariate Analysis of NZX Anonymity during Normal Trading Hours...... 102

Table 3-18: DID Estimation of the Impact of ASX Anonymity ...... 104

Table 3-19: DID Estimation of the Impact of NZX Anonymity ...... 106

Table 4-1: Trading Frequency of Cross-listed New Zealand Companies ...... 126

Table 4-2: Trading Frequency of Cross-listed Australian Companies ...... 127

Table 4-3: Trading Characteristics of Final Sample Stocks...... 128

Table 4-4: Pairing ASX and NZX Trades ...... 131

Table 4-5: Cointegration Tests ...... 134

Table 4-6: Hasbrouck Information Share for Cross-listed Stocks on ASX and NZX ...... 142

Table 4-7: Information Share over a Period of Staggered Anonymity Regime Change ...... 145

Table 4-8: ECM Estimation Results ...... 148

Table 4-9: Information Share from ECM...... 151

Table 5-1: Examples of HKEx Closing Price Determination ...... 170

Table 5-2: Three Instances of Quote-based Manipulation for Stock 0385 ...... 174

Table 5-3: Descriptive Statistics for the Sample of 123 Stock-days of Closing Price Manipulation...... 175

Table 5-4: Characteristics of Manipulated Stocks Compared to HSI Stocks ..... 176

Table 5-5: DID Estimation of the Impact of Quote-based Manipulation on Trading Behaviour ...... 195

Table 5-6: DID Estimation of the Impact of Quote-based Manipulation on Bid Behaviour ...... 200

Table 5-7: DID Estimation of the Impact of Quote-based Manipulation on Ask Behaviour ...... 206

Table 5-8: Suspected Instances of Quoted-based Manipulation ...... 214

LIST OF FIGURES

Figure 3-1: Plot Stock Price and Trading Volume for TEL ...... 69

Figure 4-1: Price Plot for AIA ...... 130

Figure 4-2: Hasbrouck Information Share over Time ...... 147

Figure 5-1: Cumulative Abnormal Return of Stock Price ...... 197

Figure 5-2: Abnormal Trading Volume ...... 198

Figure 5-3: Cumulative Abnormal Return of Bid Price ...... 201

Figure 5-4: Bid Order Frequency Ratio ...... 202

Figure 5-5: Abnormal Bid Depth ...... 203

Figure 5-6: Spread Ratio ...... 204

Figure 5-7: Cumulative Abnormal Return of Ask Price ...... 207

Figure 5-8: Ask Order Frequency ...... 208

Figure 5-9: Abnormal Ask Depth ...... 209

Figure 5-10: Daily Cumulative Abnormal Return of Stock Price ...... 211

Figure 5-11: Daily Price Returns of Manipulate Stocks Compared to HSI Stocks ...... 211

Figure 5-12: Daily Abnormal Trading Volume ...... 212

LIST OF APPENDICES

Appendix 3-1: 1-2 Matched Stocks for ASX Anonymity ...... 115

Appendix 3-2: 1-2 Matched Stocks for NZX Anonymity ...... 116

Appendix 4-1: Unit Root Test for New Zealand Stocks ...... 159

Appendix 4-2: Unit Root Test for Australian Stocks ...... 160

Appendix 4-3: Hasbrouck Information Share over Time ...... 161

Appendix 5-1: Summary of Quote-based Closing Price Manipulation Cases .... 225

Appendix 5-2: Impact of Quote-based Manipulation on Trading Behaviour Using Benchmark Analysis ...... 226

Appendix 5-3: Impact of Quote-based Manipulation on Bid Behaviour Using Benchmark Analysis ...... 227

Appendix 5-4: Impact of Quote-based Manipulation on Ask Behaviour Using Benchmark Analysis ...... 228

LIST OF PUBLICATIONS

The following publications are derived from the early version of the chapters in this thesis:

Journals

Chau, C., Foley, S., and Wang, J., 2014. Costless Closing Price Manipulation in the Hong Kong Stock Exchange. Australasian Accounting Business and Finance

Journal (Forthcoming).

Conferences

Chau, C., 2012. Quote-based Manipulation: The Hong Kong Evidence. 2012

CMCRC Annual Research Conference, 26th June 2012.

Chau, C., and Frino, A., 2012. Impact of Anonymity on Price Discovery: A

Natural Experiment. 2012 Accounting and Finance Association of Australia and

New Zealand Conference, 1-3 July, Melbourne, Australia.

Chau, C., Frino, A., Tian, G., and Ma, S.G., 2012. Impact of Anonymity on

Liquidity: Evidence from Cross-listed Stocks. 25th Australasian Banking and

Finance Conference, 16-18 December, Sydney, Australia.

Chau, C., Frino, A., Tian, G., and Ma, S.G., 2012. Impact of Anonymity on

Liquidity: Evidence from Cross-listed Stocks. 2012 Auckland Finance Meeting,

19-21 December, Auckland, New Zealand.

CHAPTER 1

INTRODUCTION

Market microstructure is the study of the process and outcomes of

exchanging assets under explicit trading rules. While much of economics

abstracts from the mechanics of trading, the microstructure literature

analyses how specific trading mechanisms affect the price formation

process…. Whatever the specific mechanism, however, prices emerge

and buyers and sellers trade (O’Hara, 1995, p1).

1.1 Background and Motivation

The financial market is the place where buyers and sellers gather to trade securities. Traders can meet at a physical trading floor, or they can communicate through an electronic trading platform. Trades can be arranged by financial intermediaries, such as dealers and brokers, or transacted directly through interaction of buyers and sellers without the involvement of intermediaries.

According to O’Hara (1995), whatever the setting, there are rules either explicit or implicit that govern the trading mechanisms and define the market structure. This organizational structure of trading determines traders’ behaviour – what, when, where and how they can trade – and is the origin of market liquidity and price formation. Market microstructure is thus defined as the study of the process and outcomes of exchanging assets under the explicit trading structures used for

financial securities (O’Hara, 1995). Madhavan (2000) also describes market microstructure as the process by which investors' latent demands are ultimately translated into prices and volumes. An important implication drawn from these definitions is that market microstructure is shaped by market structure and trading rules.

1.1.1 Market Microstructure Study

O’Hara (1995) indicates that market microstructure study exploits the role of specific market structure to characterise how different trading protocols affect price formation, and why prices exhibit particular time series. Hasbrouck (2007) lists the three main themes in empirical microstructure analysis. First, microstructure seeks to identify the sources of value and reasons for trade, in a setting where a wide range of market participants with different information decide to trade. The second theme is the understanding of trade mechanisms used to accomplish trade, such as limit order book, continuous and non-continuous auction trading. The third is the process of equilibrium price setting. At any given time, there may be many prices depending on the direction of trade (buying or selling), the trade quantity, the required speed for the trade, and the trader’s identity. For example, transparent markets require the trader’s identify to be disclosed allowing other market participants to see all orders, quotes, and trades as they occur. Price information can thus be learned through the knowledge of informed and uninformed order flow. In an anonymous market, where broker identifiers are concealed, traders are not able to see this information (Harris, 2003).

O’Hara (1995) indicates that the information structure of the market could also affect liquidity and thus the market equilibrium price.

1.1.2 Why Market Microstructure Matters

Research into the significance of market structure has been the subject of considerable interest in microstructure analysis. Market structure is important in that it affects market outcomes through its impact on the motivations, opportunities and decisions of market participants. As indicated by O’Hara (2003), the behaviour of prices and even the capability of markets depend on the ability of the trading structures to match the trading desires of sellers and buyers. On the micro level, the structure of trading matters to investors because fair pricing encourages confidence and attracts market participants to trade stocks that are fairly priced. Market structure matters to company managers because equity prices in a well-functioning market will incorporate the effect of decisions aimed to enhance shareholder wealth. Market structure also matters to exchange regulators because the process and outcomes under designed trading mechanisms indicate whether the market is liquid and competitive, whether prices are informative, and which traders trade profitably. On the macro level, there are wider implications for the economy as a whole, because the sophisticated structure and depth of securities markets provide a crucial link between the functioning of the financial system and economic growth, which makes it of considerable importance to policy makers. For these reasons, conducting microstructure analysis has important applications in market regulation, market design and formulation of

new trading mechanisms, and is of great importance to academics, exchanges and regulators.

Despite the significant interest in microstructure study, many areas of market microstructure are not well understood. For example, in the field of anonymous trading, existing studies of liquidity impact yield vague and conflicting conclusions. Some studies show that informed traders prefer anonymous trading venues, while uninformed traders prefer transparent trading venues (see, e.g.,

Forster and George, 1992; Barclay et al., 2003; Theissen, 2003). The reason is that in an anonymous trading venue, informed traders are able to conceal their trading intention, while in a transparent market liquidity suppliers are able to identify the counterparty and thus whether the order or trade is likely informed or uninformed.

Liquidity suppliers will thus charge informed traders a higher bid-ask spread and provide uninformed market participants with lower trading costs. In contrast to these studies, other researchers show that an anonymous trading system is able to attract both informed and uninformed traders resulting in overall narrower spreads

(see, e.g., Foucault et al., 2007; Comerton-Forde and Tang, 2009). The argument for this is the inability to discriminate between informed and uninformed parties and to pick off uninformed orders or free-ride informed orders in an anonymous market. Traders will therefore place more aggressive limit orders, and not behave differently on informed and uninformed trades (Foucault et al., 2007). Another example is the market structure effect on price manipulation. Exchanges monitor the trading process and develop new trading structures in order to promote market liquidity, price discovery and reduce manipulation instances. Though the liquidity impact of different trading methods has been widely studied, market manipulation

has not. The existing microstructure literature on market manipulation focuses mainly on modelling multiple forms of manipulation, while different modelling frameworks result in different predictions. For example, Aggarwal and Wu’s

(2006) market manipulation model demonstrates the possibility of profitable manipulation, while Hanson and Oprea (2009) model market manipulation in a prediction market and show that manipulation induces more traders to be better informed, resulting in improved price accuracy.

This thesis touches upon all of the above issues and seeks to offer an original way in the study of three specific microstructure areas: market liquidity, price discovery and price manipulation, all of which lack compelling and conclusive research evidence in the existing literature. This thesis considers specifically the effects of different market structures because different market structures affect market trading behaviour, and consequently market liquidity and price formation process in different ways.

1.2 Purpose and Contributions

This thesis aims to enhance our understanding of market structure. It provides evidence precisely on liquidity, price discovery and price manipulation – in the context of specific market structure. Results can be used by researchers and exchange regulators in the regulation of markets, and in the design and formulation of new trading mechanisms.

This thesis examines firstly the liquidity impact of change in market structure from transparent to anonymous trading in Chapter 3. Comerton-Forde and Tang

(2009) indicate that comparing trading on separate anonymous and transparent trading platforms poses problems in isolating the effects from market structure differences. Chapter 3 circumvents these problems in the dissimilarities in market structures, by utilising a unique natural experiment created by the staggered movement to limit order anonymity by the Australian Stock Exchange (ASX) and

New Zealand Stock Exchange (NZX). It examines the impact of anonymous trading on liquidity for cross-listed stocks on ASX and NZX and shows that anonymity attracts the trading of cross-listed stocks from the foreign counterparty, and yields significant benefits to both exchanges.

The main contributions of Chapter 3 are twofold. The first contribution is the natural experimental methodology used in this study. Eom et al. (2007) argue that the existing event studies have been econometrically flawed, because endogenous variables such as volume and volatility are used as controls in the model. Maher et al. (2008) suggest using 2SLS instrumental variables estimation to overcome this endogeneity issue. Majois (2007) argues that a “global liquidity factor” should also be taken into account in a natural experiment study to assess the impact of a change in design on market liquidity. By incorporating these views, Chapter 3 applies the instrumental variable 2SLS techniques and controls for both individual determinants of liquidity and market-wide commonality in order to isolate anonymity effects from other compounding factors. The analysis uses the same liquidity measure of the same cross-listed stock in the home (foreign) market as a natural control for individual determinants of liquidity in the foreign (home)

market. To control for market-wide factors, a sample of control stocks matched 1-

2 with each cross-listed stock in each market is constructed. Second, this study investigates anonymity effects from a new perspective, by observing liquidity migration of cross-listed stocks between the home and foreign markets. This natural experiment allows the study of anonymity effects simultaneously in both

ASX and NZX during periods of staggered regulatory changes. It investigates whether the observed anonymity effects are consistent across the two discrete changes. Moreover, similar market structure between the Australian and New

Zealand exchanges ensures the comparability of results. Chapter 3 thus enhances our ability to understand trading behaviour and markets, and makes a contribution to the field of financial market design.

Chapter 4 investigates market structure effects on the price discovery process by making further use of the natural experiment created by the staggered move to anonymity undertaken by ASX and NZX. O’Hara (2003) indicates that markets have two key functions: liquidity and price discovery – these functions are important for market structure design. Despite the significant research from the liquidity aspect, implications for price discovery due to market structure change have been largely missing from the existing literature. To fill this research gap,

Chapter 4 investigates how price discovery changes over time through the transition from trading transparently to anonymously. It is found that anonymity contributes significantly to the price discovery process for cross-listed stocks.

Using Minspan (Harris et al., 1995) synchronous transactions data of cross-listed stocks, both the Hasbrouck (1995) information share and the error correction model of Harris et al. (1995) show that information share improves on ASX, but

deteriorates on NZX, after ASX switched to anonymous trading. On the other hand, information share increases on NZX, but decreases on ASX, after NZX adopted anonymous trading. These results also add evidence to the prior literature on the choice of anonymous market by informed traders.

Chapter 4 is distinguished by three main contributions. The first is a road map for our understanding of the role of anonymity in security trading. This road map forms the link between two issues of broad interest in market microstructure, anonymity and price discovery, by observing directly market trading behaviour given the choice between transparent and anonymous markets. The second contribution is the unique setting of the natural experiment. This data set allows for: (1) comparing markets in similar screen-based limit order book systems; (2) studying the change in price discovery simultaneously in both ASX and NZX during the staggered regulatory change; (3) conducting a clean test of the impact of anonymous trading on the price discovery process. Third, this study has important implications for academics, practitioners and regulators in analysing the influence of anonymous trading structure on the behaviour of market participants, especially in the light of existing anonymity literature, which has no uniform view on what structures offer the greater liquidity.

Finally, Chapter 5 broadens the focus to include the impact of trading mechanism on price manipulation in the Hong Kong Stock Exchange (HKEx). Studying the

Hong Kong market is motivated by two factors. First, the majority of market manipulation literature focused on the US markets and European markets, while studies on the Hong Kong market have seldom been conducted. Second, HKEx is

unique among world exchanges in adopting a closing price procedure and differs significantly from other jurisdictions. It uses the median of five nominal prices from 15:59 to 16:00 at 15-second intervals with no reference to the last trade price.

Based on a sample of 123 prosecuted cases from HKEx, Chapter 5 examines the impact of the closing price mechanism on market manipulation with two key contributions. First, it complements prior literature on market manipulation by providing initial evidence of quote-based manipulation, a new form of closing price manipulation facilitated by the closing mechanism of the HKEx. Closing price can be manipulated solely through quotes without trading, and hence without cost. The manipulator is able to inflate (deflate) closing prices through placing orders to buy (sell) small quantities of shares at prices higher (lower) than the prevailing market prices near the market close. Second and the most important, this study has immediate applications in the design and formulation of new trading mechanisms. Results suggest that quote-based manipulation is associated with inflated closing prices, low trading activity and depth, narrower spreads, and much less mean reversion after the manipulation. The closing mechanism of the

HKEx is particularly susceptible to price manipulation for illiquid stocks. The introduction of call auctions and/or market-makers may be necessary to minimise this kind of manipulative conduct.

1.3 Structure of This Thesis

The structure of this thesis is organised as follows. The next chapter introduces the theoretical framework applied to this thesis. Chapters 3, 4 and 5 present the three studies discussed in this chapter. Each chapter contains sections including

literature review, data description, research design, empirical results, additional tests and conclusions. Chapter 6 concludes by highlighting how the evidence presented in this thesis can be used by academics and exchange regulators in understanding market liquidity, price discovery and price manipulation under various market structures.

References

Aggarwal, R.K., and Wu, G., 2006. Stock Market Manipulations. The Journal of

Business, 79(4), pp.1915-1953.

Barclay, M.J., Hendershott, T., and McCormick, D.T., 2003. Competition among

Trading Venues: Information and Trading on Electronic Communications

Networks. The Journal of Finance, 58(6), pp.2637-2665.

Comerton-Forde, C., and Tang, K.M., 2009. Anonymity, liquidity and fragmentation. Journal of Financial Markets, 12(3), pp.337-367.

Eom, K.S., Ok, J., and Park, J.-H., 2007. Pre-trade transparency and market quality. Journal of Financial Markets, 10(4), pp.319-341.

Forster, M.M., and George, T.J., 1992. Anonymity in securities markets. Journal of Financial Intermediation, 2(2), pp.168-206.

Foucault, T., Moinas, S., and Theissen, E., 2007. Does Anonymity Matter in

Electronic Limit Order Markets? Review of Financial Studies, 20(5), pp.1707-

1747.

Hanson, R., and Oprea, R., 2009. A Manipulator Can Aid Prediction Market

Accuracy. Economica, 76(302), pp.304-314.

Harris, L., 2003. Trading and exchanges: market microstructure for practitioners,

Oxford, New York.

Harris, F.H., McInish, T.H., Shoesmith, G.L., and Wood, R.A., 1995.

Cointegration, Error Correction, and Price Discovery on Informationally Linked

Security Markets. The Journal of Financial and Quantitative Analysis, 30(4), pp.563-579.

Hasbrouck, J., 1995. One Security, Many Markets: Determining the Contributions to Price Discovery. The Journal of Finance, 50(4), pp.1175-1199.

Hasbrouck, J., 2007. Empirical Market Microstructure, Oxford University Press,

New York.

Madhavan, A., 2000. Market microstructure: A survey. Journal of Financial

Markets, 3(3), pp.205-258.

Maher, O., Swan, P.L., and Westerholm, P.J., 2008. Twilight falls on the limit order book: Endogeneity and the demise of broker identity. Working Paper,

University of New South Wales.

Majois, C., 2007. Natural experiments methodology and global liquidity in financial markets. Working paper, Louvain School of Management and FUCaM.

O’Hara, M., 1995. Market Microstructure Theory, Blackwell Publishing,

Cambridge.

O’Hara, M., 2003. Presidential Address: Liquidity and Price Discovery. The

Journal of Finance, 58(4), pp.1335-1354.

Theissen, E., 2003. Trader Anonymity, Price Formation and Liquidity. European

Finance Review, 7(1), pp.1-26.

CHAPTER 2

THEORETICAL FRAMEWORK OF THIS THESIS

2.1 Introduction

Market microstructure deals with the trading of financial assets and the evolution of asset prices by taking into account liquidity, price discovery, trading integrity and market design. This chapter presents select parts of the theory of market microstructure, and lays the theoretical framework for the empirical analysis in

Chapters 3 to 5. What follows is a brief description of the two types of trading systems: a quote-driven market and an order-driven market.

In a quote-driven market, buyers and sellers trade through financial intermediaries, known as market-makers, or dealers or specialists. Stock prices are determined from bid and ask quotations made by market-makers. It is used in some US exchanges, such as NYSE, AMEX, and NASDAQ (e.g., Aggarwal and Wu (2006) study prosecuted manipulation cases in these markets).

Today the main market mechanism in the world’s stock exchanges is order-driven, with no designated market-makers (e.g., Euronext, Hong Kong, Australia and

New Zealand). Buyers and sellers submit the prices and quantities at which they are willing to trade a security through an electronic trading platform. These buy and sell orders are displayed and aggregated in a limit order book. A buy limit order is an instruction to buy at a price, which must be at or below the specified price. For a sell limit order, the price must be at or above the limit price. Table 2-1

presents an example of a limit order book from the SEATS (Stock Exchange

Automated Trading System) screen of the Australian Stock Exchange (ASX).

Table 2-1: The ASX’s SEATS Screen (prior to 28 November 2005) Stock AAA Bid (Buy Orders) Ask (Sell Orders)

Broker ID Quantity Price Quantity Price Broker ID

111 5250 20.24 1450 20.28 999

222 350 20.23 1000 20.28 888

333 7000 20.22 500 20.29 444

444 1200 20.21 1000 20.30 999

555 27 20.20 500 20.30 777

666 300 20.20 1000 20.32 666

777 350 20.20 50 20.36 222

888 1000 20.20 500 20.36 111 Source: Frino et al. (2009)

By consolidating all such orders in a limit order book, the highest buying price

(20.24) is known as the best bid, and lowest selling price (20.28) is the best ask.

The quantity of shares available at each price is called market depth. The difference between the bid and ask is known as the bid-ask spread. When the ask is close to the bid, the spread is narrow. When the ask is much higher than the bid, the spread is wide. Trades take place when a trader is willing to pay the bid-ask spread (i.e., to buy at the ask price or sell at the bid price). The execution of trades is immediately based on price and then time priority. This is known as a continuous auction. Some markets also conduct non-continuous call auctions, in which all trades are arranged at a particular time. An unmatched or unexecuted order is known as a quote. Once an order is executed, the quote is removed from the limit order book and replaced by the next best available bid or ask.

Table 2-1 also displays the broker identifier (ID) beside each quote. This is the indication of a transparent market, where brokers are able to identify the parties of other limit orders and the counterparties to trades after transactions have occurred. In Australia, ASX had displayed the full limit order book including broker identification number, until 28 November 2005 when its broker regime changed to be anonymous. World exchanges which have adopted an anonymous market include New Zealand, Paris, and Tokyo.

In a well-functioning market, buyers and sellers can easily find each other, and trade without significant adverse effect on prices. This is known as a liquid market where there are many standing limit orders and small bid-ask spreads all the time.

The price of the security being quoted at any given point in time will thus reflect the information held by market participants. This price formation process is known as price discovery. In a fair and manipulation-free market, prices obtained on the market are a reflection of genuine supply and demand. This concept refers to market integrity. The following section reviews each of these principal issues in market microstructure theory.

2.2 Market Liquidity

In market microstructure literature, liquidity is defined as “the willingness of some traders to take the opposite side of a trade that is initiated by someone else at low cost” (Harris, 1990). Thus, as indicated by Lee et al. (1993), market liquidity has two dimensions: the price dimension, represented by the bid-ask spread, and the quantity dimension, represented by market depth. A large body of

microstructure research focuses on market liquidity based on a theory of asymmetric information, which assumes that one party has more or better information than the other in a transaction. Well-informed traders profit at the expense of less-informed traders. Less-informed traders therefore try to avoid well-informed traders (Harris, 2003).

Lee et al. (1993) and Benveniste et al. (1992) argue that if specialists believe that there is a chance of informed trading, they will respond by increasing the bid-ask spread and or reduce the depth at the quoted prices. This also implies a negative relationship between the spread and depth. Kavajecz (1999) shows that specialists manage quoted depth to deal with risks associated with an information event.

Specifically, Kavajecz finds that liquidity providers, both market-maker and limit order traders, reduce depths around earning announcements to decrease adverse selection costs.

By extending this reasoning to the case of the anonymous limit order book,

Foucault et al. (2007) develop a theoretical model which enables them to conclude that anonymous quotes can lead to overall tighter bid-ask spreads. Their model explains that in a transparent market where broker identification codes are displayed, uninformed traders estimate the proportion of informed trades in the market before submitting orders. If they believe that the participation rate of informed traders is small, they will actively set the best quotes, as there is a relatively low chance that informed traders will pick off their limit orders. This leads to narrower spreads. Conversely, when informed traders’ participation rate is high, wide spreads from uninformed traders are predicted. However, in an

anonymous market, traders generally are unable to discriminate between informed and uninformed parties, and to pick off uninformed orders or free-ride informed orders. They will therefore place more aggressive limit orders, and not behave differently on informed and uninformed trades. This is consistent with the study by Garfinkel and Nimalendran (2003), who investigate the impact of insider trading on market-maker behaviour for anonymous NASDAQ and transparent

NYSE. They find that NASDAQ dealers do not adjust to the presence of insider trading by raising effective spreads. The effective spreads of stocks traded in the anonymous NASDAQ dealer system are narrower than in transparent NYSE specialist system.

Comerton-Forde et al. (2005) also document narrower spreads in Paris, Tokyo, and Korea, which moved to anonymous trading. Motivated by prior studies,

Chapter 3 of this thesis examines the liquidity migration effects of the change in anonymous market structure in the trading of cross-listed stocks, which has not been previously considered.

2.3 Price Discovery

Like liquidity, price discovery is another central function of financial markets

(O’Hara, 2003). While the former refers to the ability of an asset to be sold, the latter refers to the ability of the market to find the efficient price (O’Hara, 2003).

Price discovery has been defined as “the incorporation of new information into security prices” (Hasbrouck, 1995), and as “ the process by which markets

attempt to find equilibrium prices from new information” (Schreiber and Schwartz,

1986).

In a multiple markets setting, the concept of price discovery drawn from these definitions is that the prices for the same security in different markets should tend to converge in the long run but might deviate from one another in the short run.

Each observable price of an asset in multiple markets can be conceived as an information-based common efficient price shared by all these markets (Gonzalo and Granger, 1995), plus a transitory liquidity/noise trading shock such as bid-ask bounce and order imbalances on liquidity trades. Following Hasbrouck (1995 and

1996), this concept can be expressed in a random walk model, when considering a security traded in two separate markets at potentially different prices and :

(2.1)

( ) ( )

Here, the common underlying implicit efficient price is , which follows random walk; reflects new information; is the observed security price; shows the non-informational features, i.e. transitory liquidity shocks, and is assumed a zero- mean covariance stationary process.

Following the assumption that an implicit unobservable efficient price is common to all markets, Hasbrouck (1995) initiates the information share method as an explicit measure of relative contribution to price discovery by a particular market to the innovation in this common efficient price. The application of this method

can be found in Booth et al. (2002), Eun and Sabherwal (2003), and Huang (2002).

A detailed description of Hasbrouck information share is provided in Chapter 4 of this thesis, which investigates the impact of trader anonymity on the price discovery process in a multiple markets setting.

Market microstructure theory suggests that the evolution of asset prices depends on the nature of players in the market. An early model by Lintner (1969) on asset price formation shows that financial markets aggregate the beliefs of individual traders, and the market equilibrium price is a weighted average of these beliefs with the weights being determined by the investors’ risk aversion. Grossman

(1976) considers a rational expectations equilibrium model of a stock market in which there are two types of traders: informed and uninformed. Informed traders know the true value of traded asset, and take positions in the market based on their information. Uninformed traders invest no resources in collecting information, but know that prices will reflect the information of the informed traders. Under this framework, when informed traders trade, the security price will reflect all of the information to all traders, and private information is transmitted from the informed to the uninformed. A limitation of this result is that since traders take the price as given, they have no incentives to collect information when the market is free from “noise” (uninformed noise traders who do not trade on material information, but trade randomly, see Kyle, 1985).

In a noisy rational expectations model by Grossman and Stiglitz (1980) and

Diamond and Verrecchia (1981), an unobserved “noise” term is introduced so as to prevent identifying the price information by simply observing the equilibrium

price. They show that private information is always valuable and the aggregation of this information plays a prominent role in the price formation process. Kyle

(1985) develops a model of speculative trading in which a monopolist insider trades sequentially in the asset market against uninformed noise traders, who trade randomly without information. Kyle's main result is that the insider trades slowly, so that his private information is incorporated into prices gradually.

Despite the development of various models of price formation, one feature in common is that they assume that market prices are based on private information.

Uninformed investors act as price takers and price discovery occurs through trading with informed traders.

2.4 Price Manipulation

According to Allen and Gorton (1992), in a market with asymmetric information, when uninformed traders face uncertainty about the existence of informed traders, manipulators will have an incentive to manipulate the equilibrium price. Market manipulation distorts the price discovery process. O’Hara (2001) indicates that price integrity is the main factor that affects how well and how quickly prices adjust to fundamental values. Margotta (2011) also indicates that market integrity exists when stock prices are set in a market free from manipulation.

Allen and Gale (1992) categorise market manipulation based on the types of action taken to distort the security price: action-based manipulation, information- based manipulation, and trade-based manipulation. In action-based trading a

manipulator takes actions to affect the actual or perceived value of the assets. For example, directors may short the firm’s stock and shut down the factory to depress the share price, and then cover their short positions and reopen the factory when the stock price rises to its previous level. Information-based trading involves releasing false information or spreading false rumours. Trade-based manipulation occurs when a manipulator attempts to manipulate a stock by buying and then selling.

Trade-based manipulation is mostly examined in the microstructure literature. It involves a wide variety of manipulative practices, for example, engaging in a series of truncations reported on a public display facility to give the impression of activity of price movement in a security is known as “painting the tape” or

“spoofing”. “Front-running” takes place when brokers place orders ahead of client orders for the same security. “Wash sales” occur when manipulators buy and sell a specific security among themselves without genuine change in ownership.

“Pump-and-dump” or “ramping” involves artificially inflating a stock’s price in order to sell the stocks at a higher price. This manipulation practice often occurs at the close of the market, known as “marking the close”1.

Many studies on market manipulation attempt to model the possibility of profitable price manipulation. Theoretical models of trade-based manipulation often examine the possibility and profitability of manipulation. Some argue against the profitability of manipulation, while others build models to derive

1 A detailed description of a wide variety of manipulative practices is provided in Cumming and Johan (2008).

conditions under which trade-based manipulation can be profitable. Most models show that the manipulator tries to take advantage of the presence of asymmetric information where one side of the market is unsure whether the other side is informed or manipulating the market. It is often assumed that the manipulator pretends to be informed in order to deceive the market. Fischel and Ross (1991) argue that successful manipulation is difficult to achieve in an efficient market.

The argument is that when a trader tries to buy a stock, he drives up the price.

However, when he tries to sell it, he drives down the price. Because of the liquidity cost associated with trading, a manipulator must buy high and sell low, which makes manipulation unprofitable. Aggarwal and Wu (2006) refer to this as the “unravelling problem”.

Jarrow (1992) argues that profitable manipulation is possible if the manipulator can establish a price trend and trade against it. Jarrow models market manipulation by large traders who have no information. Their trade moves prices either because of trade size or because other market participants believe that the large trader is informed. Due to this information asymmetry, Allen and Gale

(1992) model market manipulation in a rational expectations framework, and show that an uninformed manipulator can make a profit, as long as the other investors believe that the manipulator is informed. Aggarwal and Wu (2006) extend this model and consider a market with information seekers (or arbitrageurs, who seek out information about the stock’s true value). They show that the

“unravelling problem” can be solved when information seekers compete for shares, pushing the stock price up. The competition from information seekers makes it easier for manipulators to enter the market and make a profit. In contrast

to many of these theoretical studies, Hanson and Oprea (2009) model traded- based manipulation in a prediction market, and find that manipulators are unable to mislead the market and distort prices.

A few studies specifically analyse closing price manipulation. Kumar and Seppi

(1992) model a manipulator who firstly takes a large position in the futures market, and then artificially bids up the stock price before the close in the spot market to profit from the improved settlement price. A model of mutual fund managers by Carhart et al. (2002) suggests that fund managers have incentives to use short-term price impacts to manipulate closing prices at the end of the reporting period. Hillion and Suominen (2004) develop a model in which brokers manipulate closing prices to alter customers’ perceptions of their execution quality. They show that closing call auctions reduce the instances of manipulation and enhance price efficiency.

These theoretical models provide insights about the conditions under which manipulation is possible and identify circumstances in which profitable manipulation opportunities may exist. However, given specific and complex trading mechanisms that differ from one market to another, manipulation practices in real markets are too complicated to be generalised into one theoretical model.

For this reason, Chapter 5 examines the manipulation techniques and motivations and effects of manipulation on liquidity and price behaviour with reference to a particular trading system in the context of the Hong Kong market.

2.5 Summary

Market microstructure literature analyses how specific trading mechanisms affect price behaviour and the price formation process. This chapter presents the theoretical framework of this thesis in the fields of market liquidity, price discovery and price manipulation. It also provides a brief description of quote- driven and order-driven markets, the types of market infrastructure (such as transparent versus anonymous market, and continuous versus non-continuous trading), and key features of financial markets (e.g., information asymmetry, bid- ask spread and depth). Market structure defines the quality of a market. In a well- functioning market, there are many standing limit orders and small bid-ask spreads all the time. Buyers and sellers have equal access to information and the security price is set free from manipulation and is a reflection of a genuine supply and demand.

References

Aggarwal, R.K., and Wu, G., 2006. Stock Market Manipulations. The Journal of

Business, 79(4), pp.1915-1953.

Allen, F., and Gale, D., 1992. Stock-Price Manipulation. The Review of Financial

Studies, 5(3), pp.503-529.

Allen, F., and Gorton, G., 1992. Stock price manipulation, market microstructure and asymmetric information. European Economic Review, 36(2-3), pp.624-630.

Benveniste, L.M., Marcus, A.J., and Wilhelm, W.J., 1992. What’s special about the specialist? Journal of Financial Economics, 32(1), pp.61-86.

Booth, G.G., Lin, J.-C., Martikainen, T., and TSE, Y., 2002. Trading and Pricing in Upstairs and Downstairs Stock Markets. The Review of Financial Studies, 15(4), pp.1111-1135.

Carhart, M., Kaniel, R., Musto, D.K., and Reed, A.V., 2002. Leaning for the tape:

Evidence of gaming behaviour in equity mutual funds. Journal of Finance, 57(2), pp.661-693.

Comerton-Forde, C., Frino, A., and Mollica, V., 2005. The impact of limit order anonymity on liquidity: Evidence from Paris, Tokyo and Korea. Journal of

Economics and Business, 57(6), pp.528-540.

Cumming, D., and S. Johan, 2008. Global market surveillance. American Law and

Economics Review, 10(2), pp.454-506.

Diamond, D.W., and Verrecchia, R.E., 1981. Information aggregation in a noisy rational expectations economy. Journal of Financial Economics, 9(3), pp.221-235.

Eun, C.S., and Sabherwal, S., 2003. Cross-Border Listings and Price Discovery:

Evidence from U.S.-Listed Canadian Stocks. The Journal of Finance, 58(2), pp.549-575.

Foucault, T., Moinas, S., and Theissen, E., 2007. Does Anonymity Matter in

Electronic Limit Order Markets? Review of Financial Studies, 20(5), pp.1707-

1747.

Fischel, D., and Ross, D. 1991. Should The Law Prohibit Manipulation in

Financial Markets? Harvard Law Review, 105(2), pp.503-553.

Frino, A., Chen, Z., and Hill, A. L., 2009. Introduction to Corporate Finance,

Pearson, Australia.

Gonzalo, J., and C. Granger, 1995. Estimation of common long-memory components in cointegrated systems. Journal of Business and Economic Statistics,

13 (1), pp.27-35.

Garfinkel, J.A., and Nimalendran, M., 2003. Market Structure and Trader

Anonymity: An Analysis of Insider Trading. Journal of Financial and

Quantitative Analysis, 38 (3), pp.591-610.

Grossman, S., 1976. On the Efficiency of Competitive Stock Markets Where

Trades Have Diverse Information. The Journal of Finance, 31(2), pp.573-585.

Grossman, S.J., and Stiglitz, J. E., 1980. On the impossibility of informationally efficient markets. American Economic Review, 70(3), pp.393-407.

Hanson, R., and Oprea, R., 2009. A Manipulator Can Aid Prediction Market

Accuracy. Economica, 76(302), pp.304-314.

Hasbrouck, J., 1995. One Security, Many Markets: Determining the Contributions to Price Discovery. The Journal of Finance, 50(4), pp.1175-1199.

Hasbrouck, J., 1996. Modeling Market Microstructure Time Series. Working

Paper Series, New York University.

Harris, L. E., 1990. Liquidity, Trading Rules, and Electronic Trading Systems.

Monograph Series in Finance and Economics, Monograph No. 1990-4, New York

University.

Harris, L., 2003. Trading and exchanges: market microstructure for practitioners,

Oxford, New York.

Hillion, P., and Suominen, M., 2004. The manipulation of closing prices. Journal of Financial Markets, 7(4), pp.351-375.

Huang, R.D., 2002. The Quality of ECN and Nasdaq Market Maker Quotes. The

Journal of Finance, 57(3), pp.1285-1319.

Jarrow, R.A., 1992. Market Manipulation, Bubbles, Corners, and Short Squeezes.

Journal of Financial and Quantitative Analysis, 27(3), pp.311-336.

Kavajecz, K.A., 1999. A Specialist’s Quoted Depth and the Limit Order Book.

The Journal of Finance, 54(2), pp.747-771.

Kumar, P., and Seppi, D.J., 1992. Futures Manipulation with “Cash Settlement”.

The Journal of Finance, 47(4), pp.1485-1502.

Kyle, A.S., 1985. Continuous Auctions and Insider Trading. Econometrica, 56(6), pp.1315-35.

Lee, C.M., Mucklow, B., and Read, M.J., 1993. Spreads, Depths, and the Impact of Earnings Information: An Intraday Analysis. The Review of Financial Studies,

23(2), pp.345-374.

Lintner, J., 1969. The Aggregation of Investors Diverse Judgments and

Preferences in Purely Competitive Stock markets. Journal of Financial and

Quantitative Analysis, 4(4), pp.347-400.

Margotta, D., 2011. Market Integrity, Market Efficiency, Market Accuracy. The

Business Review, 17(2), pp.14-20.

O’Hara, M., 2001. Designing Markets for Developing Countries. International

Review of Finance, 2(4), pp.205-215.

O’Hara, M., 2003. Presidential Address: Liquidity and Price Discovery. The

Journal of Finance, 58(4), pp.1335-1354.

Schreiber, P. S., and Schwartz, R.A., 1986. Price Discovery in Securities Markets.

Journal of Portfolio Management, 12(4), pp.43-48.

CHAPTER 3

IMPACT OF ANONYMITY ON LIQUIDITY IN LIMIT ORDER

BOOKS: EVIDENCE FROM CROSS-LISTED STOCKS

3.1 Abstract

This chapter examines the liquidity impact of market structure change in the trading of cross-listed stocks using a natural experiment created by the staggered move to an anonymity regime by the Australian Stock Exchange and New

Zealand Stock Exchange. Estimation with both 2SLS instrumental variable and difference-in-difference shows two interesting trends. When considering the liquidity impact after ASX switched to anonymous trading, the bid-ask spreads, quoted depth and trading volume improve on ASX, but deteriorate on NZX. On the other hand, when considering NZX’s adoption of anonymous trading, liquidity decreases on ASX, but increases on NZX. Results suggest that anonymity attracts the trading of cross-listed stocks from the foreign counterparty and yields significant benefits to both exchanges. The existence of commonality in liquidity in financial markets seems also apparent, and the inclusion of this commonality in natural experiment studies may be necessary.

3.2 Introduction

On 6 July 2007 the New Zealand Stock Exchange (NZX) adopted anonymous trading by removing broker IDs 2 , following the Australian Stock Exchange’s

(ASX) move to anonymity on 28 November 2005. NZX, which is overshadowed by ASX, aimed to incentivise traders and improve market liquidity. The fear of losing competitiveness to the new ASX anonymous system was seen as an apparent motive behind NZX’s adoption of anonymity.

This staggered movement to anonymous markets by ASX and NZX provides an ideal natural experiment to examine the impact of anonymous trading on liquidity for cross-listed stocks on ASX and NZX. The small body of existing evidence on the liquidity impact of market structure change is mixed and inconclusive, largely due to the difficulties in employing an appropriate research method in the study of natural experiments. For example, Foucault et al. (2007) study the change in bid- ask spreads on the Paris Bourse before and after the change to anonymous trading, and their ordinary least squares (OLS) regression analysis shows that trader anonymity results in narrower spreads. However, Maher et al. (2008) argue that the methodology employed by Foucault et al. (2007) is flawed, because endogenous variables such as stock price, trading volume and volatility are used to control for the bid-ask spread determinants. They obtain the opposite conclusion when using instrumental variable two-stage least squares regression model (2SLS). Majois (2007) argues that a “global liquidity factor” should be taken into account in natural experiment studies. Majois finds that the decrease in

2 Anonymous trading started on 6 July 2007 when NZX’s new electronic trading platform, Trayport, was implemented. This date was verified by NZX.

spread in the Paris Bourse also appears at the same time on NYSE, which does not experience a change in market design. After adding the spread on NYSE as an additional control variable for a market-wide factor in the regression of Foucault et al. (2007), Majois finds that the decrease in spreads in the Paris Bourse completely vanishes.

By overcoming this methodological issue in prior literature, this chapter is distinguished by two contributions. First, it contributes to the limited literature on the effects of broker anonymity by providing additional evidence using a natural experiment on the two exchanges and takes into account endogenetiy. The 2SLS instrumental variables technique is applied, with controls for both stock-specific and overall market movements. The regression uses the same liquidity measure of the same cross-listed stock in the home (foreign) market as a natural control for individual determinants of liquidity in the foreign (home) market. To control for market-wide factors, a sample of control stocks matched 1-2 with each cross- listed stock in each market is constructed. This natural experiment methodology circumvents problems inherent in previous studies which either: (1) use endogenous variables such as stock price, trading volume and volatility as control for the spread determinants (e.g., Foucault et al., 2007; Comerton-Forde et al.

2005); or (2) do not control for market-wide factors (e.g., Maher et al., 2008); or

(3) do not control for stock-specific issues (e.g., Majois, 2007). Second, this chapter investigates anonymity effects from a new perspective. It observes directly liquidity migration of cross-listed stocks between the home and foreign markets during the two discrete regulatory changes, which has not been previously considered.

The results of 2SLS estimation and difference-in-difference analysis show that the bid-ask spreads, depth and trading volume improve on ASX, but deteriorate correspondingly in NZX, after ASX introduced anonymous trading. On the other hand, the adoption of anonymous trading on NZX leads to an opposite finding, that is, liquidity increases on NZX, but decreases on ASX. These results have an immediate implication for market structure design, suggesting that the anonymity regime attracts the trading of cross-listed stocks from the foreign counterparty, and yields significant benefits to both exchanges.

The remainder of this chapter is organised as follows. The next section reviews the literature pertaining to the liquidity impact of anonymous trading, and outlines several hypotheses which are tested in this study. Section 3.4 describes the relevant institutional details for ASX and NZX markets and Section 3.5 presents the data. Section 3.6 sets out the research methodology. Section 3.7 reports the results and Section 3.8 provides a robustness check. Section 3.9 concludes.

3.3 Literature Review and Hypotheses

Anonymity enables traders to execute transactions on the market without displaying their identifiers, allowing them to conceal trading intentions. Many theoretical models predict that informed traders choose to trade in an anonymous venue, so that their trading activity is less likely to be detected (O’Hara, 1995;

Madhavan, 1995; Forster and George, 1992). They fear that revealing their presence will drive liquidity traders out of the marketplace and increase their execution costs (Harris, 2003; Benveniste et al., 1992). Theoretically, this

reasoning assumes market participants’ ability to discriminate between informed and uninformed traders in transparent markets by observing brokers’ identities.

In the model of Rindi (2008), liquidity suppliers can be either uninformed or informed. She shows that when the number of informed traders is given, transparency increases liquidity by reducing adverse selection costs. Uninformed traders learn about order flows by observing identification codes. Uninformed traders get information at no cost and thus become more informed themselves.

They will therefore make more aggressive limit orders and thus increase liquidity.

However this diminishes the informed traders’ incentive to acquire information, so that increasing transparency can reduce the number of informed traders, hence reduce liquidity. She concludes that the overall liquidity effects depend on the endogenous entry of informed traders.

The preference for anonymity by informed traders is evident in many empirical studies. For example, Grammig et al. (2001) analyse trader anonymity in the

German stock market, where a transparent floor trading system coexists with an anonymous electronic market. They estimate the probability of informed trading for these two parallel markets, and find that a higher level of anonymity is associated with a higher probability of informed trading. Heidle and Huang (2002) document similar findings in their study on information-based trading between anonymous competing dealers (electronic screen-based NASDAQ), and transparent auction markets (specialist systems such as NYSE and AMEX). They find that the probability of informed trading declines for the firms that transfer their exchange listing from an anonymous to a transparent venue, and increases

with a change from a transparent to an anonymous venue. Their comparison of the two trading structures shows that the probability of informed trading is more pronounced in an anonymous system than in a transparent auction market.

The informed trader’s preference for an anonymous market is clear, whereas the empirical evidence on the liquidity impact is mixed. Early research, such as

Grammig et al. (2001) and Heidle and Huang (2002), finds that the spreads in an anonymous trading platform are wider than on a transparent platform. However,

Simaan et al. (2003) find that market-makers quote narrower spreads on anonymous Electronic Communication Networks (ECN) than on the transparent

NASDAQ dealer system, because anonymity reduces the probability of collusion among quote setters. Since these studies conduct comparisons between different market structures (i.e., anonymous electronic systems versus transparent floor or dealer systems), Comerton-Forde and Tang (2009) point out that such comparisons pose inherent problems in isolating the effects of anonymity from other market structure effects.

Recent research uses natural experiments. Comerton-Forde and Tang (2009) point out that natural experiments control more effectively for the dissimilarities in market structures. Foucault et al. (2007) use a natural experiment on the Paris

Bourse, which switched to an anonymous broker regime on 23 April 2001. They develop a theoretical model which concludes that anonymous quotes can lead to tighter bid-ask spreads. They argue that in anonymous markets market participants are unable to discriminate between informed and uninformed traders by observing traders’ identities, and therefore will place more aggressive limit

orders. This argument is consistent with the study by Garfinkel and Nimalendran

(2003), who show that anonymous NASDAQ quotes exhibit smaller changes in the proportional effective spreads when compared to the transparent NYSE specialist system.

On this basis, Foucault et al. (2007) conclude that anonymous quotes lead to tighter bid-ask spreads. They test this proposition by using data on large stocks from the Paris Bourse. Their OLS regression analysis shows the decrease in bid- ask spreads following the switch to anonymous broker IDs, after controlling for the changes in the spread determinants. Following this line of thought, Comerton-

Forde et al. (2005) conduct a natural experiment and employ the same multivariate methodology to include markets in Paris, Tokyo, and Korea. They find higher liquidity in markets that move to anonymous trading. Comerton-Forde and Tang (2009) replicate the Foucault et al. (2007) spreads model, but with the addition of a trend term, for investigating the market quality of ASX after the removal of broker identifiers. They obtain similar results with lower spreads, and greater depth in anonymous markets. They also examine the direction of order flow in ASX and NZX (a transparent market at that time), and find evidence for the migration of trading activity from NZX to ASX for large cross-listed stocks after the removal of broker identifiers on ASX. In line with earlier results,

Hachmeister and Schiereck (2010) study the impact of post-trade anonymity on liquidity and informed trading in the German stock market after the introduction of the Central Counterparty in March 2003. They find a significant increase in liquidity measured through a reduction of 25% in implicit transaction costs. They

also find that informed traders change their behavior in providing liquidity more aggressively in an anonymous environment.

These studies provide strong empirical evidence for believing that the adoption of an anonymous market should lead to an improvement of bid-ask spreads (thus liquidity). However, Maher et al. (2008) claim that those results are sensitive to the econometric specification employed. They replicate the above studies for the five exchanges including Paris, Brussels, Tokyo, Australia, and Korea, but using fixed effect instrumental variables estimation. They show the opposite findings, that anonymity leads to an increase in the effective bid-ask spread, intraday volatility and a decrease in overall trading volume. They argue that this is because these studies do not control for endogenous variables such as volume and price, and ignorance of endogeneity issues can seriously affect results.

Eom et al. (2007) adopt a panel-data approach to control for endogeneity. They show that increasing pre-trade transparency in the form of an increasing level of quote disclosure in the Korean Exchange (KRX) limit order book improves market quality. Poskitt et al. (2011) study anonymity effects on the liquidity of

NZX-listed stocks. Their OLS and 2SLS estimations point to the same conclusion, that NZX’s market share improves in the trading of cross-listed stocks in relation to ASX (also an anonymous market at that time), although they find that effective spreads increase following the switch to an anonymity regime on NZX.

Majois (2007) points out that a global liquidity factor should also be taken into account in a natural experiment study to assess the anonymity effects on market

liquidity. This is because market-wide factors can influence liquidity in financial markets as a whole (Chordia et al., 2000; Brockman, 2009). Majois replicates the regression analysis of Foucault et al. (2007) and shows that the decrease in spreads also appears on the NYSE, which did not experience any broker regime change. Majois further shows that after adding the spread on NYSE as an additional control variable in the regression when analysing the spread in Paris, the post-event dummy variable becomes non-significant. Majois attributes this to a global phenomenon that would affect liquidity around the world.

Motivated by previous studies, this chapter uses a natural experiment, in the form of staggered regulatory change by ASX and NZX, to examine the effects of anonymity on liquidity for both Australian and New Zealand cross-listed stocks.

Incorporating the views of Maher et al. (2008) and Majois (2007), this chapter uses instrumental variables 2SLS estimation with controls for both stock-specific and market-wide liquidity factors.

Based on the theoretical rationale outlined in Foucault et al. (2007), that limit order book traders are more willing to trade aggressively in an anonymous trading venue, as well as the evidence obtained by Comerton-Forde and Tang (2009) and

Poskitt et al. (2011) on the increased cross-listed shares trading in the home market, this chapter tests whether trader anonymity attracts liquidity migration in the trading of cross-listed stocks.

Specifically, it is expected that the change from a transparent market to an anonymous market in ASX will lead to an improvement in liquidity on ASX, but a decline on NZX. This leads to the first pair of hypotheses (H3.1A and H3.1B):

Hypothesis3.1A: Liquidity of cross-listed stocks on ASX will increase after ASX introduced anonymous trading.

Hypothesis3.1B: Liquidity of cross-listed stocks on NZX will decrease after ASX introduced anonymous trading.

The measure of liquidity encompasses the spread (both quoted and effective spreads), quoted depth and trading volume. Lee et al. (1993) identify that both spread and depth are needed to draw conclusions about the changes in liquidity. A widening (narrowing) of the spread, combined with a decrease (increase) in depth, infers a decrease (increase) in liquidity unambiguously. It is surprising that much of the anonymity literature focuses on the spread while evidence on depth is limited. For example, Foucault et al. (2007) find an ambiguous effect on depth in the Paris Bourse after the switch to pre-trade anonymity. For this reason, this chapter extends the work examining the liquidity impact of the adoption of anonymous trading on bid-ask spreads to depth. It is expected that bid-ask spreads and depth will improve on ASX, and deteriorate correspondingly on NZX, after

ASX introduced anonymous trading.

The third liquidity measure is volume, which examines anonymity effects on trading activity. Comerton-Forde and Tang (2009) find an increase in ASX’s trading volume of cross-listed stocks relative to NZX after the removal of broker

identifiers on ASX. Poskitt et al. (2011) also find that the switch to anonymous trading improves NZX’s share of trading of cross-listed stocks relative to ASX.

It is therefore expected that all these three principal measures of liquidity will improve on ASX, and deteriorate on NZX after ASX’s adoption of anonymous trading. On the other hand, the second pair of hypotheses (H3.2A and H3.2B) anticipates the opposite findings after NZX introduced trader anonymity.

Hypothesis3.2A: Liquidity of cross-listed stocks on ASX will decrease after NZX introduced anonymous trading.

Hypothesis3.2B: Liquidity of cross-listed stocks on NZX will increase after NZX introduced anonymous trading.

3.4 ASX versus NZX and Market Structure Change

Both Australian and New Zealand exchanges operate analogous open electronic limit order books and do not have market-makers in their equities markets.

Investors place orders into a trading platform through licensed operators within stockbroking firms. Buy and sell orders are then matched and executed on a price and time priority basis. ASX is currently using a trading platform called ASX

Trade in the trading of equity securities, and goes through a number of market phases each trading day. The pre-opening phase takes place from 7:00 to 10:00

(Australian Eastern Standard Time), where orders are entered and queued according to price-time priority and will not trade until the market opens. In the opening phase between 10:00 and 10:10, the trading system calculates the opening

prices, and securities open in accordance with alphabetical order in their ASX code. Normal trading hours are from 10:00 to 16:00, brokers enter orders into the trading platform, and all trades are matched in price and time priority continuously. At 16:00, a 10-minute period of pre-close begins, followed by the single-price closing auction.

On NZX, the New Zealand Stock Market (NZSX) is the premier equities market.

The trading of equity securities is executed online through the GlobalVision trading system provided by Trayport. Each trading day, NZSX operates through several sessions. During the pre-opening period from 9:00 to 10:00 (New Zealand

Standard Time), orders are entered into the system but no matching takes place.

Prior to the commencement of the normal trading session at 10:00, the trading system matches the orders and establishes the opening prices to maximise the total volume traded. The majority of trading takes place during the normal trading phase, which lasts until 16:45. Orders entered during normal trading hours are matched, resulting in trades, or are stored in the order book automatically. At

16:45, a 15-minute period of pre-close session begins. At 17:00, the trading system matches the orders and calculates the closing prices, and the market is closed.

The difference in time zones between Australia and New Zealand is small, with

New Zealand usually two hours ahead of Australia. Trading hours on ASX and

NZX overlap from 10:00 to 14:45 on ASX and 12:00 to 16:45 on NZX (both in local time). For most of the year, this results in 4.75 hours of overlapping operation between the two markets. However, due to differences in the start and

end dates of daylight savings, this overlap can range between 3.75 hours to 5.75 hours.

Both markets are highly integrated and competitive, and market participants can easily trade in either market. In terms of market capitalisation, the New Zealand

Stock Exchange is very small and less liquid compared to Australian and many other overseas markets. As of December 2011, ASX has 2079 companies listed, including 1983 domestic companies and 96 foreign companies, while NZX has

151 companies listed on the main board NZSX, in which 108 are domestic securities. In Australia, the top 500 stocks by market capitalisation are included in the broad-based All Ordinaries Index (All Ords), while in New Zealand the NZX

50 Index comprises the 50 largest and most liquid companies.

Prior to the change in broker anonymity regime, both markets displayed the full limit order book, including individual broker identification numbers. Brokers were able to identify the parties of other limit orders and the counterparties to trades after transactions occurred. As part of Equity Market Reform, ASX removed broker identifiers from the trading screens on 28 November 2005. The

ASX was of the view that all market participants, including investors and brokers, should have equal access to broker identification details (prior to the change broker identifiers were only visible to ASX participating organizations). ASX believed that trader anonymity would provide investors with a greater range of trading options and thus attract market liquidity and improve price discovery to the benefit of all market participants (ASX, 2005).

Following ASX’s introduction of anonymous trading, NZX started its anonymous trading on 6 July 2007, when GlobalVision was implemented. The new system allows market participants to post orders anonymously. The counterparty is not displayed on the trading screen to the market participants, nor disclosed on the trade confirmation. NZX stated that the change to anonymous trading would bring more flexibility and open up opportunities for the market. A fear of losing competitiveness to the new ASX anonymous system seems to be the potential motive behind the NZX’s adoption of anonymity.

3.5 Data and Summary Statistics

Thirty-nine stocks were traded on both ASX and NZX when ASX introduced the broker anonymity regime. This includes 21 Australian companies and 18 New

Zealand companies. On the other hand, 35 cross-listed stocks were traded on both exchanges when NZX went to anonymous trading, including 20 Australian- incorporated stocks and 15 New Zealand-incorporated stocks. Table 3-1 and

Table 3-2 list these stocks and calculate the common trading days in both markets during a six-month period around the regulatory changes. Clearly, the level of trading activity in these stocks varies considerably. Approximately half of them trade on all or most days in both markets, while the remainder trade infrequently or not at all.

To be included in the final sample, stock data needs to be available throughout the six-month study period around the regulatory change in broker regime. To avoid any potential data errors associated with the switch and allow for a period of

learning, one week before and after the exact broker identification change date has been excluded. This is defined as the “learning period” (Goldstein and Kavajecz,

2000; Maher et al., 2008). Therefore, a period of three months before and after the regulatory change is used in the analysis, with the exclusion of the learning period.

Separate analysis on anonymity effects is conducted for the regulatory change in

ASX and NZX. “ASX anonymity” is the label for studying the liquidity impact after ASX introduced the broker anonymity regime on 28 November 2005. The three-month period prior to the change is from 29 August 2005 to 20 November

2005, and the three-month period after the change is from 5 December 2005 to 27

February 2006. “NZX anonymity” is the label for studying the liquidity impact after NZX adopted anonymous trading on 6 July 2007. The pre-change period is from 6 April 2007 to 28 June 2007, and the post-change period is from 13 July

2007 to 4 October 2007.

Since some stocks are rarely traded and would not provide reliable observations, a stock is required to be traded at least once every five trading days. In addition, stocks are excluded if they undergo acquisitions during the sample period to circumvent any possible problems caused by structural changes. After all this filtering, Table 3-1 shows 24 cross-listed stocks selected for studying ASX anonymity. This includes 10 Australian companies and 14 New Zealand companies. Table 3-2 shows 22 stocks selected for studying NZX anonymity, including 10 Australian companies and 12 New Zealand companies.

Table 3-1: Sample Selection for ASX Anonymity This table reports summary trading statistics for 39 stocks listed on both ASX and NZX. “Common Trading Days” reports the number of days the company’s stocks were traded on both ASX and NZX during a six- month period around the introduction of an anonymous market by ASX (excluding one week before and after the change). “Final Sample” is “Yes” if the company is included in the final sample, otherwise it is “No”.

Final Common Company Name Code Sample Trading Days Reasons for Exclusion

Panel A: Australian Companies Australian Foundation Investment Co. Ltd AFI Yes 91 AMP Ltd AMP Yes 112 Australia and New Zealand Banking Group Ltd ANZ Yes 106

APN News & Media Ltd APN No 43 Acquisition on 20/12/2005 AXA Asia Pacific Holdings Limited AXA Yes 111

Babcock & Brown Infrastructure BBI No 0 Thin trading on NZX Downer EDI Limited DOW No 0 Thin trading on NZX Energy World Corporation Limited EWC No 0 Thin trading on NZX Lend Lease Corporation Limited LLC No 0 Thin trading on NZX Lion Nathan Ltd LNN Yes 106

National Australia Bank Limited NAB No 0 Thin trading on NZX Pacific Brands Limited PBG Yes 64

People Telecom Limited PEO No 0 Thin trading on NZX Pan Pacific Petroleum NL PPP Yes 56

Rio Tinto Limited RIO No 0 Thin trading on NZX RMG Limited RMG No 0 Thin trading on NZX Summit Resources Limited SMM Yes 99

Tag Pacific Limited TAG No 12 Thin trading on NZX Telstra Corporation Ltd TLS Yes 108

Transpacific Industries Group Ltd TPI No 0 Thin trading on NZX Westpac Banking Corporation WBC Yes 113

Panel B: New Zealand Companies Auckland International Airport Ltd AIA Yes 110 Air New Zealand Ltd AIR Yes 106 Carter Holt Harvey Limited CAH Yes 110 Fletcher Building Ltd FBU Yes 113 Fisher & Paykel Appliances Holdings Ltd FPA Yes 109 Fisher & Paykel Healthcare Corporation Ltd FPH Yes 94 Gensis Research and Development

Corporation Ltd GEN No 0 Thin trading on ASX Heritage Gold NZ Ltd HGD No 19 Thin trading on ASX Nuplex Industries Ltd NPX Yes 51 New Zealand Oil & Gas Ltd NZO Yes 68 Kathmandu Holdings Limited KMD No 0 Thin trading on ASX Sky City Entertainment Group Ltd SKC Yes 110 Sky Network Television Ltd SKT No 0 Thin trading on ASX Telecom Corporation of New Zealand Ltd TEL Yes 113 Tower Ltd TWR Yes 113 Vea Advantage Limited VEA Yes 104 Waste Management NZ Limited WAM Yes 42

The Warehouse Group Limited WHS Yes 87

Table 3-2: Sample Selection for NZX Anonymity This table reports summary trading statistics for 35stocks listed on both ASX and NZX. “Common Trading Days” reports the number of days the company’s stocks were traded on both ASX and NZX during a six- month period around the introduction of an anonymous market by NZX (excluding one week before and after the change). “Final Sample” is “Yes” if the company is included in this data sample, otherwise it is “No”.

Final Common Company Name Code Sample Trading Days Reasons for Exclusion Panel B: Australian Companies

Australian Foundation Investment Co. Ltd AFI Yes 102

AMP Ltd AMP Yes 113

Australia and New Zealand Banking Group Ltd ANZ Yes 111

APN News & Media Ltd APN No 23 Missing data Babcock & Brown Infrastructure BBI No 0 Thin trading on NZX Downer EDI Limited DOW No 0 Thin trading on NZX Energy World Corporation Limited EWC No 0 Thin trading on NZX Goodman Fielder Limited GFF Yes 109

Lend Lease Corporation Limited LLC No 0 Thin trading on NZX L & M Petroleum Limited LMP Yes 76

Lion Nathan Ltd LNN Yes 100

National Australia Bank Limited NAB No 0 Thin trading on NZX Pacific Brands Limited PBG Yes 37

People Telecom Limited PEO No 0 Thin trading on NZX Pan Pacific Petroleum NL PPP Yes 96

Rio Tinto Limited RIO No 0 Thin trading on NZX Tag Pacific Limited TAG No 6 Thin trading on NZX Telstra Corporation Ltd TLS Yes 107

Transpacific Industries Group Ltd TPI No 0 Thin trading on NZX Westpac Banking Corporation WBC Yes 113

Panel B: New Zealand Companies

Auckland International Airport Ltd AIA Yes 110

Air New Zealand Ltd AIR Yes 113

Fletcher Building Ltd FBU Yes 112

Fisher & Paykel Appliances Holdings Ltd FPA Yes 106

Fisher & Paykel Healthcare Corporation Ltd FPH Yes 111

Gensis Research and Development Corporation Ltd GEN No 0 Thin trading on ASX Heritage Gold NZ Ltd HGD Yes 74

Nuplex Industries Ltd NPX Yes 36

New Zealand Oil & Gas Ltd NZO Yes 91

Kathmandu Holdings Limited KMD No 0 Thin trading on ASX Sky City Entertainment Group Ltd SKC Yes 110

Sky Network Television Ltd SKT No 0 Thin trading on ASX Telecom Corporation of New Zealand Ltd TEL Yes 104

Tower Ltd TWR Yes 113

The Warehouse Group Limited WHS Yes 64

This final sample covers all industry sectors, including consumer, energy, financials, health care, information technology, materials, telecommunication services and utilities. Seven cross-listed stocks are the components of the

S&P/ASX 200, among which six are Australian-incorporated (AMP, ANZ, GFF,

PBG, TLS, and WBC), and one is New Zealand-incorporated (TEL). Five stocks are included in NZX 10 (AIA, FBU, FPH, SKC and TEL), which are all New

Zealand-incorporated.

For each stock, trade and quote data are obtained from the Reuters DataScope

Tick History Database, provided by the Securities Industry Research Centre of

Asia Pacific (SIRCA). Two data sets are used in this study. The first data set is calculated from overlapping trading hours only, in order to observe liquidity change in ASX and NZX simultaneously. This intraday trade data is in one- minute intervals, containing fields with the security code, date, time, price and volume. In each interval, the last trade price and the total volume traded are

1 th calculated. The market depth data is time-stamped to the nearest /100 of a second.

Each quote consists of the best bid and ask price along with the number of shares willing to be transacted at that respective price. For consistency, a one-minute quote data set is generated from this data. The prevailing quotes and respective depth levels are recorded at the end of each minute. To reduce the data set to a more manageable size, a single daily observation is calculated from the one- minute data set. The average price and depth, and the total trading volume are computed during common trading hours each day for each stock on NZX and

ASX respectively. As a result, for each stock and each liquidity measure, the

working sample consists of at most 120 daily observations during the six-month study period.

Since NZX opens and closes two hours before ASX, this may build in potential bias such as wider spreads at the open and close. The second data set thus considers normal trading hours in a robustness check. This data set is based on time-and-sales data, which includes the quoted price and size, the trade price and trade amount, and the time the quote or trade was made. A single daily observation during normal trading hours is also calculated from this data set. All intraday data thus has been summarised to a daily basis. Table 3-3 and Table 3-4 present brief summaries of the trading activity for the sample stocks during the study period of ASX and NZX anonymity respectively.

Table 3-3: Trading Statistics of Final Sample Stocks for ASX Anonymity This table presents sample stocks’ average price, number of shares traded, trade size and number of trades per day over the six-month study period of ASX anonymity. Average NZX and ASX prices are expressed in their respective currencies.

Trading on ASX Trading on NZX Company Name Price ($) Shares Traded Trade Size Trades/Day Price ($) Shares Traded Trade Size Trades/Day

Panel A: Australian Companies Australian Foundation Investment Co. Ltd 4.2 151,655 2,334 65 4.58 8,975 3,213 3 AMP Ltd 7.6 3,198,212 3,981 805 8.28 14,444 1,283 9 Australia and New Zealand Banking Group Ltd 23.8 2,540,856 1,922 1,298 25.81 7,738 1,598 5 AXA Asia Pacific Holdings Limited 5.0 970,496 3,529 276 5.39 8,843 1,424 6 Lion Nathan Ltd 7.8 484,480 2,219 217 8.45 11,726 2,503 5 Pacific Brands Limited 2.6 1,823,388 7,138 252 2.85 6,853 3,609 2 Pan Pacific Petroleum NL 0.1 257,570 39,435 6 0.13 71,407 22,807 3 Summit Resources Limited 0.7 640,162 8,119 74 0.75 59,957 8,691 6 Telstra Corporation Ltd 4.1 21,422,834 12,935 1,627 4.47 46,789 5,149 9 Westpac Banking Corporation 21.8 2,676,927 2,166 1,195 23.66 27,019 1,526 17

Panel B: New Zealand Companies Auckland International Airport Ltd 1.9 43,242 6,525 6 2.02 974,549 16,588 59 Air New Zealand Ltd 1.1 31,194 4,420 6 1.21 254,487 15,494 14 Carter Holt Harvey Limited 2.4 142,020 5,654 24 2.55 1,635,395 120,255 13 Fletcher Building Ltd 7.0 42,681 2,499 16 7.59 390,401 6,950 54 Fisher & Paykel Appliances Holdings Ltd 3.1 20,277 2,842 6 3.41 227,375 7,481 29 Fisher & Paykel Healthcare Corporation Ltd 3.3 27,436 8,945 4 3.63 327,175 8,778 37 Nuplex Industries Ltd 4.1 6,579 2,787 3 4.49 46,590 2,693 16 New Zealand Oil & Gas Ltd 0.9 22,026 6,992 3 0.93 135,309 8,826 15 Sky City Entertainment Group Ltd 4.3 38,742 2,682 13 4.70 570,692 10,234 51 Telecom Corporation of New Zealand Ltd 5.4 1,255,801 7,943 157 5.85 3,475,368 26,414 129 Tower Ltd 1.9 477,621 4,725 96 2.09 337,119 11,018 26 Vea Advantage Limited 3.5 342,944 4,482 82 3.79 53,541 5,944 7 Waste Management NZ Limited 5.6 6,756 2,855 2 6.08 98,961 3,724 27 3.5 10,142 2,554 4 3.82 192,355 6,397 30 The Warehouse Group Limited

Table 3-4: Trading Statistics of Final Sample Stocks for NZX Anonymity This table presents sample stocks’ average price, number of shares traded, trade size and number of trades per day over the six-month study period of NZX anonymity. Average NZX and ASX prices are expressed in their respective currencies.

Trading on ASX Trading on NZX Company Name Price ($) Shares Traded Trade Size Trades/Day Price ($) Shares Traded Trade Size Trades/Day

Panel B: Australian Companies 5.7 201,208 1,980 103 6.52 6,619 1,943 3 Australian Foundation Investment Co. Ltd 10.4 3,667,692 2,178 1,683 11.76 13,729 1,071 11 AMP Ltd 29.3 3,310,134 1,115 2,957 33.27 11,392 1,679 6 Australia and New Zealand Banking Group Ltd 2.5 2,432,382 4,932 514 2.80 33,089 6,023 5 Goodman Fielder Limited 0.2 242,013 23,741 10 0.18 82,759 16,152 5 L & M Petroleum Limited 9.0 697,434 1,126 630 10.24 9,679 2,100 4 Lion Nathan Ltd 3.4 1,662,999 3,058 602 3.80 4,133 2,395 2 Pacific Brands Limited 0.2 1,252,546 25,801 47 0.28 236,378 33,618 7 Pan Pacific Petroleum NL 4.6 19,949,531 13,330 1,514 5.23 17,570 2,998 5 Telstra Corporation Ltd 26.6 2,990,328 1,140 2,622 30.22 24,025 1,447 11 Westpac Banking Corporation

Panel B: New Zealand Companies 2.6 55,999 4,626 13 2.94 1,708,632 23,548 72 Auckland International Airport Ltd 2.3 54,236 4,289 15 2.57 647,372 18,265 39 Air New Zealand Ltd 10.7 110,165 1,413 90 12.17 671,205 8,247 81 Fletcher Building Ltd 3.2 45,363 13,928 11 3.63 499,994 13,090 37 Fisher & Paykel Appliances Holdings Ltd 3.1 50,001 4,013 15 3.50 828,225 10,800 76 Fisher & Paykel Healthcare Corporation Ltd 0.1 371,205 50,804 7 0.08 118,601 27,449 4 Heritage Gold NZ Ltd 6.2 3,001 1,490 2 7.10 29,007 1,907 15 Nuplex Industries Ltd 0.9 63,119 7,800 8 1.06 219,459 9,750 22 New Zealand Oil & Gas Ltd 4.2 55,290 2,876 19 4.82 958,539 11,148 73 Sky City Entertainment Group Ltd 4.1 1,975,694 6,273 323 4.62 4,061,509 35,028 119 Telecom Corporation of New Zealand Ltd 2.0 289,998 2,450 129 2.30 282,338 7,569 35 Tower Ltd 5.6 17,725 2,598 6 6.33 287,291 7,603 36

The Warehouse Group Limited

In general, the trading of cross-listed stocks is more active in the home market.

For example, the trading activity of Australian companies (Panel A) is substantially higher in ASX. The average number of trades per day on ASX ranges from 6 to 1627, compared to a range from 2 to 17 on NZX. The daily trading volume is typically hundreds of times higher on ASX. On the other hand, the trading in New Zealand companies (Panel B) is generally higher on NZX. The average trade size and volume are several times higher on NZX than on ASX.

This similar trading pattern is also observed during the period of NZX anonymity.

The total number of trades on ASX is 207,130, with 94% from Australian- incorporated stocks, while the total number of trades on NZX is 69,992, of which

91% are from New Zealand stocks. The discrepancy in the average price at which the stocks trade across the two markets reflects the New Zealand-Australia exchange rate, which was approximately equal to 1.09 on average over the ASX anonymity period, and 1.13 over the NZX anonymity period.

Figure 3-1 shows a representative section of the price and volume behaviour of

TEL (Telecom Corporation of New Zealand) in both markets. Clearly, price series track each other closely, and trading volume in the two markets also follows a similar pattern. This implies a very close relationship between the home and cross-listed market. Using the same liquidity measure of a cross-listed stock in the other market can be seen as a good natural control for stock-specific movements.

Figure 3-1: Plot Stock Price and Trading Volume for TEL

Stock Price 6.5

5.5

4.5 ASX Price (AUD) NZX Price (NZD)

4 29/08/2005 28/09/2005 28/10/2005 27/11/2005 27/12/2005 26/01/2006

Trading Volume 17

12 Natural Natural logarithm 11 ASX Volume NZX Volume

10 29/08/2005 28/09/2005 28/10/2005 27/11/2005 27/12/2005 26/01/2006

3.6 Research Methodology

As observed earlier, the sample stocks trade more frequently in their home market.

Hasbrouck (1991) points out that illiquid stocks tend to exhibit higher information asymmetry. Easley et al. (1996) indicate that inactive stocks are subject to greater informed trading. Literature on home bias also finds information differences between the foreign and domestic investors. Therefore, information effects would be greater for foreign stocks, which are less liquid compared to domestic stocks.

Given the differences in trading characteristics between domestic and foreign stocks, separate analysis is conducted for Australian and New Zealand companies.

This also allows the observation of whether anonymity effects are consistent across the two groups of stocks. This section conducts both the univariate statistical analysis and multivariate regression using instrumental variables 2SLS, followed by a robustness check using difference-in-difference estimation in

Section 3.8.

3.6.1 Univariate Analysis

The univariate analysis is performed by using the first set of overlapping data. For each market, liquidity is examined over a six-month period around the event date, and measured in three principal ways. First, the bid-ask spreads are calculated as:

( ) ( )

( ) ( ) ( ) ( )

where Spread ($) is denoted as dollar spread, and Relative Spread is the spread calculated in percentage. Ask and bid are the best ask and bid prices, respectively.

Second, quoted depth is the average dollar value of shares on offer at the best bid and ask price:

( ) ( ) ( )

Third, trading volume is measured as the number of shares traded multiplied by the price per share:

( ) ( )

Table 3-5 and Table 3-6 report the univariate analysis for these liquidity measures during the regulatory change in ASX and NZX, respectively. Cross-sectional means of bid-ask spreads, depth and trading volume for NZX and ASX companies are calculated respectively in Panel A and B. Separate figures are reported for the pre-event period Pre, post-vent period Post, and the difference between the Pre period and Post period. Difference is calculated as Post less Pre. Figures within parentheses are t-statistics of the differences (from unequal variance) 3. Dollar spread, depth and volume on ASX and NZX are specified in their respective currencies.

Consistent with the previous tables, greater trading activity with narrower spread on the home market is evident in Tables 3-5 and 3-6. Average depth and volume

3 Two-sample t-test on the mean difference is based on Selvanathan et al. (2006).

for Australian companies are many times higher on ASX than on NZX, while liquidity of New Zealand companies is higher on NZX. Spreads of Australian stocks are lower on ASX and spreads of New Zealand stocks are lower on NZX, reflecting a better liquidity on the home market.

Table 3-5: Univariate Analysis of ASX Anonymity The table reports cross-sectional average for bid-ask spreads, dollar depth and volume for ASX anonymity. Pre is the pre-event period, and Post is the post-event period. Difference is calculated as Post less Pre. t- statistics of the differences (from unequal variance) are estimated in parentheses. Dollar spread, depth and volume on ASX and NZX are specified in their respective currencies.

Trading on ASX Trading on NZX Pre Post Difference Pre Post Difference Panel A: Australian companies Spread ($) 0.012 0.012 0.000 0.096 0.086 -0.010 (-0.9) (-0.9) Relative Spread (%) 0.705 0.783 0.078 1.978 1.876 -0.102 (1.0) (-0.8) Depth ($000) 489 453 -37 43 44 1 (-0.5) (0.3) Volume ($000) 27,639 25,681 -1,958 153 118 -35 (-0.8) (-2.0)** Panel B: New Zealand Companies Spread ($) 0.052 0.043 -0.009 0.020 0.021 0.001 (-3.0)*** (1.2) Relative Spread (%) 1.493 1.344 -0.149 0.628 0.670 0.042 (-2.1)** (1.4) Depth ($000) 69 52 -17 178 103 -75 (-3.2)*** (-4.0)*** Volume ($000) 784 877 92 2,777 3,194 417 (0.7) (1.1)

Table 3-6: Univariate Analysis of NZX Anonymity The table reports cross-sectional average for bid-ask spreads, dollar depth and volume for NZX anonymity. Pre is the pre-event period, and Post is the post-event period. Difference is calculated as Post less Pre. t- statistics of the difference in liquidity (from unequal variance) are estimated in parentheses. Spread ($), depth ($) and volume ($) on ASX and NZX are specified in their respective currencies.

Trading on ASX Trading on NZX Pre Post Difference Pre Post Difference Panel A: Australian companies Spread ($) 0.011 0.012 0.001 0.215 0.245 0.030 (4.9)*** (1.5) Relative Spread (%) 0.811 1.075 0.264 2.731 3.752 1.021 (2.4)** (4.2)*** Depth ($000) 520 554 34 23 23 1 (0.4) (0.80) Volume ($000) 34,082 38,144 4,061 99 201 102 (1.4) (3.2)*** Panel B: New Zealand Companies Spread ($) 0.054 0.062 0.008 0.018 0.020 0.001 (1.4) (1.7)* Relative Spread (%) 1.680 2.143 0.463 0.859 1.147 0.288 (3.7)*** (2.7)*** Depth ($000) 40 30 -10 70 60 -9 (-3.1)*** (-2.8)*** Volume ($000) 1,064 924 -140 3,811 4,535 724 (-0.8) (1.5)

This univariate analysis is not very indicative of anonymity effects. The simple statistics in the tables may ignore some factors that also affect liquidity in the stocks of interest. For example, comparing Pre to Post from Table 3-5, quoted depth of New Zealand companies decreases significantly on both exchanges, indicating stock-specific movements of cross-listed stocks.

Similarly, after the introduction of anonymity on NZX, Table 3-6 shows that bid- ask spreads and depth deteriorate in both exchanges. Clearly, the overall market trends either upturn or downturn would make it difficult to identify any anonymity effects on the trading of cross-listed stocks. In fact, during middle August 2007, the global financial crisis struck financial markets. Within the study period, the

Australian All Ordinaries Index (New Zealand NZ50 Index) was 6392.2 (4236.7) on 6 July 2007, and fell to 5670.3 (3894.3) on 17 August 2007, a decline of 11.3%

(8.1%), before recovering to 6579.9 (4279.8) points on 4 October 2007. Chordia et al. (2000) attribute this to the commonality in liquidity in financial markets. In their study of NYSE stocks, they show that quoted spreads and depth co-move with market and industry-wide liquidity. They argue that market liquidity is affected not only by a structural change, but also by market-wide factors. This confirms the need for a multiple regression analysis that is able to properly isolate the anonymity effects. Brockman et al. (2009) use methodology of Chordia et al.

(2000) to measure commonality in spread and depth in an international context, which includes 47 stock exchanges. They find that firm-level changes in liquidity are significantly influenced by both exchange-level changes and global commonality.

3.6.2 Fixed Effect Instrumental Variable 2SLS Estimation

Multivariate analysis is conducted by employing the instrumental variable 2SLS technique using the first set of overlapping data. The structure of the model is presented in equation (3.5).

퐿 𝑞 𝑖𝑦

훽0 훽 𝐶 𝑛 𝑖 훽 𝑛 𝑖 훽3푀 𝐶 𝑛

훽4𝐶 𝑛𝑔 훼𝑖 𝜀𝑖 ( 5)

where the subscript i denotes individual firms, t denotes the day, 훼𝑖 is a firm-

specific parameter, Change=1 if day t is after the introduction of anonymity, and zero otherwise. This variable is the key variable for assessing the impact of the broker identification policy change. Liquidity is the natural logarithm of

ASX/NZX liquidity (spreads, depth and volume), Stock Control, Price Difference and Market Control are three log-transformed control variables. Gujarati (2007) comments that the log transformation compresses the scales in which the variables are measured, and increases the model accuracy.

Stock Control is the natural logarithm of the same liquidity measures of the same stock i on NZX (ASX), when the dependent variable is the liquidity of that cross- listed stock i on ASX (NZX). Instead of using volume, price and volatility as individual determinants, this variable is an ideal natural control for a stock- specific shift in liquidity. However, introducing the NZX (ASX) liquidity as an independent variable in the ASX (NZX) liquidity regression may give the impression of two-way or mutual causality. A 2SLS using instrumental variables is implemented, if the Hausman specification test of simultaneity indicates that

Stock Control is indeed endogenous in the regression.

In accordance with Gujarati (2007), the instrumental variable must be theoretically justified, and is required to be correlated with the endogenous variable, but uncorrelated with the error term of the original regression.

Instruments used in this study include three lags of Stock Control and the natural logarithm of trade size. The theoretical basis of using a lagged explanatory variable is that the past (that is, lagged) variables are not systematically correlated with current condition (Harris, 1994). For instance, yesterday’s liquidity of cross-

listed stocks on NZX may affect today’s liquidity on NZX; however, it is less likely to be influenced by current liquidity on ASX. Trade size is also included in the first-stage regression. Maher et al. (2008) indicate that when predicting the liquidity impact of policy change, trade size can serve as a strong instrument.

All these instrumental variables have been used in related literature (e.g., Harris,

1994; Maher et al., 2008). Various tests of the relevance and validity of various combinations of instruments are performed. The 2SLS analysis is conducted separately for each of the three liquidity measures during each study period. Since not every instrumental variable satisfies the required conditions of relevance and exogeneity in every analysis, the 2SLS implements instruments if they satisfy these conditions and uses fitted values from the first-stage regressions as the explanatory variables in the second stage.

The second control variable is Price Difference, which is defined as the logarithmic price difference between NZX and ASX prices (NZX price minus

ASX price), in which ASX price is converted into New Zealand dollar by using the prevailing exchange rate. As indicated by Grammig et al. (2005), the foreign exchange rate influences the differential between prices on home and foreign markets. Liquidity from the foreign market is then crucial for performing arbitrage trading, bringing prices to fundamental values and keeping markets efficient

(Shleifer and Vishny, 1997). It is likely that a source of liquidity in the trading of cross-listed stocks comes from their foreign counterparts, where traders seek the cheapest trading location or price advantages. Including Price Difference thus captures price difference effects.

The last control variable is Market Control for a market-wide factor. Indeed, without controlling for a market-wide factor, the analysis from ordinary least squares regression shows deteriorated spreads and depth in both ASX and NZX during the period of NZX anonymity, due to the impact of global financial crisis.

Therefore, to ensure independence of possible broad market movements, Market

Control is introduced by constructing a control sample of non-cross-listed stocks matched 1-2 with each cross-listed stock in each market. Following an approach similar to Bacidore and Sofianos (2002), these control stocks are matched based on the priority of trading volume, price, then market capitalisation and industry if possible. The order of these selection criteria is due to the trading characteristics of the sample stocks and the constraints on selecting NZX-listed matched stocks.

NZX is a very small market. During the study period, only 63 New Zealand domestic stocks are available to be chosen as control stocks (after excluding the sample stocks and overseas stocks). These stocks have relatively smaller market capitalisation, lower trading prices and volumes compared to the sample stocks.

The majority is middle and small-cap stocks, and only three out of 63 are included in NZX 10 index. Moreover, the trading of some of ASX’s large-cap stocks on

NZX may be hundreds of times less than the trading on ASX. For example, TLS

(Telstra Corporation Ltd) has a market capitalisation of $AUD 55 billion, while its daily average trading volume on NZX is only about $AUD110,000. Matching priority is thus in order of trading volume, price, and then market capitalisation and industry. Appendices 3-1 and 3-2 list these control stocks matched with the sample cross-listed stock for ASX and NZX anonymity respectively.

Table 3-7 provides summary statistics for the sample and matched stocks. It shows that NZX control stocks are relatively smaller compared to the sample stocks, as discussed earlier. For the trading on ASX, the sample stocks have average trading volume and price similar to their matched sample, suggesting that the matching procedure is effective in identifying suitable control stocks.

The average liquidity variable across these matched stocks is calculated as a market liquidity proxy, i.e., Market Control for controlling the effect of broad market movements. To ensure that the results are not subject to stock matching bias, a 1-1 matched sample is also constructed, as well as the use of all 63 NZX stocks as a proxy of market liquidity. The results are robust to these alternative comparison samples4.

4 Those results are available on request.

Table 3-7: Summary Statistics for the Sample and Matched Stocks This table contains summary statistics for the sample of cross-listed stocks and matched non-cross-listed stocks during the study periods of ASX and NZX anonymity. The mean dollar volume, price and market capitalisation are calculated, and all are specified in their own currencies.

Trading on ASX Trading on NZX

Mean Median Min Max Mean Median Min Max Panel A: ASX Anonymity

Volume ($000)

24 Sample firms 10,358,618 385,292 19,024 83,887,016 1,559,467 263,002 9,888 20,099,244 10,074,113 1,186,351 13,856 209,750,036 161,424 59,503 6,810 1,381,952 48 Matched firms Price ($)

24 Sample firms 5.24 3.81 0.12 23.78 5.69 4.15 0.13 25.81 6.80 3.84 0.06 40.71 3.23 2.39 0.36 20.41 48 Matched firms Market capitalisation ($000)

24 Sample firms 8,152,041 1,758,482 52,585 51,095,740 8,848,185 1,908,296 57,679 55,579,094

48 Matched firms 5,301,219 1,105,947 5,520 80,046,243 495,397 207,209 8,442 3,975,042

Panel B: NZX Anonymity

Volume ($000)

22 Sample firms 14,989,281 462,615 14,926 94,172,447 2,113,681 296,032 9,251 18,716,873

44 Matched firms 20,552,633 3,060,492 14,189 391,265,530 228,197 83,623 5,980 2,350,977

Price ($)

22 Sample firms 6.22 3.72 0.06 29.32 7.06 4.20 0.08 33.27

44 Matched firms 10.17 5.97 0.13 54.55 3.50 2.55 0.65 10.38

Market capitalisation ($000)

22 Sample firms 10,168,900 2,081,437 14,353 57,484,652 11,536,301 2,359,558 17,344 65,140,856

44 Matched firms 8,369,724 2,128,814 11,374 117,832,559 643,100 251,916 34,362 5,231,508

3.7 Results

Using instrumental variables 2SLS of equation (3.5), this section tests the hypotheses for the liquidity impact on the trading of cross-listed stocks after the introduction of anonymous trading by ASX and NZX respectively. The equation is conducted separately for Australian and New Zealand stocks and uses robust standard errors. The first data set of overlapping trading hours is used in this analysis.

3.7.1 Liquidity Impact of ASX Anonymity

Table 5-8 reports the results of 2SLS estimation. The dependent variable is the liquidity of cross-listed stock i on ASX, the stock control variable is the same liquidity measure of same stock i on NZX (NZX Liquidity), and the control variable for market commonality is the average liquidity of 48 matched stocks on

ASX (Market Control). When analysing this change in ASX liquidity, equation

(3.5) can be expressed as:

𝑋 퐿 𝑞 𝑖𝑦

훽0 훽 𝑍𝑋 퐿 𝑞 𝑖𝑦 훽 𝑛 𝑖 훽3 𝑋 푀 𝐶 𝑛

훽4𝐶 𝑛𝑔 훼𝑖 𝜀𝑖 ( 6)

Table 3-8: 2SLS Results – Impact of ASX Anonymity on ASX Liquidity This table reports the 2SLS estimation of equation (3.6). The instruments include three lags of NZX liquidity measures and the natural logarithm of trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (NZX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread ($) Depth ($) Volume ($) Constant -1.64 -2.70 1.90 0.39 -0.06 -1.44 1.20 -2.36 (-4.7)*** (-7.6)*** (1.2) (0.2) (-0.1) (-2.2)** (0.6) (-0.9)

NZX Liquidity 0.00 -0.02 0.06 0.69 0.79 0.32 0.72 0.83 (0.2) (-0.7) (2.1)** (4.1)*** (5.0)*** (8.5)*** (8.8)*** (5.6)***

Price Difference 0.82 0.35 -0.83 5.21 -1.14 0.69 -5.29 -6.62 (1.9)* (0.9) (-0.5) (2.1)** (-1.1) (0.6) (-4.8)*** (-2.4)**

Market Control 0.55 0.53 0.66 0.39 0.07 0.27 0.04 0.07 (8.3)*** (6.6)*** (4.8)*** (2.5)** (0.5) (1.8)* (0.2) (0.4)

Change 0.01 -0.02 -0.02 0.11 -0.10 -0.10 0.06 0.10 (0.6) (-2.1)** (-0.6) (1.6) (-3.7)*** (-3.9)*** (1.7)* (1.3)

R -squared 0.97 0.78 0.92 0.90 0.73 0.78 0.71 0.64

Hausman Specification Test 2.01** 0.26 0.88 2.38** 0.85 1.00 1.29 1.20 Durbin-Wu-Hausman Chi-sq Test 3.36* 7.00*** 2.28 21.69*** 11.24*** 2.09 28.39*** 16.87*** First-stage Coefficients Lag 1 of NZX Liquidity 0.32*** 0.32*** 0.12*** 0.23*** 0.21*** Lag 2 of NZX Liquidity 0.13*** 0.14*** 0.07** 0.23*** 0.07*** Lag 3 of NZX Liquidity 0.05* 0.07*** Trade Size ($) 0.18*** Partial R-sq 0.15 0.15 0.02 0.06 0.16 0.06 Partial F-statistic 52.05*** 54.43*** 8.89*** 19.79*** 67.28*** 24.86*** Anderson canon. corr. LM statistic 151.41*** 155.23*** 22.06*** 76.34*** 217.18*** 71.37*** Sargan Test 0.15 0.01 0.41 0.88 0.09 1.21 LM Test of Redundancy Lag 1 of NZX Liquidity 91.88*** 93.04*** 14.96*** 76.15*** 53.48*** Lag 2 of NZX Liquidity 17.52*** 18.54*** 4.72** 181.97*** 7.06*** Lag 3 of NZX Liquidity 3.13* 74.29*** Trade Size ($) 163.76***

The first-stage diagnostics and various statistics reported in Table 3-8 show that a

2SLS model is necessary for all liquidity measures except depth ($) regression in

Panel A and spread ($) in Panel B. The instruments include three lags of NZX liquidity measures, log-transformed trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether IV is the preferred estimator. The null hypothesis is that there are no endogenous variables or that endogeneity does not affect the OLS estimator.

If both tests indicate no endogeneity issue, as well as no preferred IV estimates,

OLS estimates will be used. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (NZX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity.

The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified.

Consistent with Figure 3.1, the NZX Liquidity variable in Table 3-8 indicates that liquidity of cross-listed stocks in the home and foreign markets is closely related.

All ASX liquidity measures including bid-ask spreads, depth and volume are directly related to the liquidity on NZX with six of eight measures statistically significant at least at the 5% level. Moreover, the Market Control variable shows that all ASX liquidity measures are positively related to the liquidity of matched stocks, with four of eight liquidity measures statistically significant at least at the

5% level, and one of eight at the 10% level. This probably constitutes evidence of

commonality in liquidity.

The evidence of price difference impacts seems to appear in spread (%) and volume ($) regressions in Panel A, and depth ($) and volume ($) regressions in

Panel B. In this daily time-series analysis, it may present an arbitrage opportunity whereby traders try to take advantage of a price difference between the markets, by buying an asset when the price is low and reselling it when the price is high.

The extent and direction of price difference effects depend on various factors, such as trade direction (buy or sell-initiated) and transaction costs of buying, holding and reselling. An estimation of price impact on liquidity is beyond the scope of this study. The 2SLS equation (3.6) aims to examine anonymity effects using the multiple regression procedure that is able to control for other factors that simultaneously affect the liquidity5.

After controlling for all these confounding factors, Table 3-8 shows that ASX’s move to an anonymous market leads to an improved liquidity on ASX in the trading of both Australian and New Zealand companies, supporting Hypothesis3.1A.

In the trading of Australian stocks (Panel A), the change dummy indicates a remarkable decrease in dollar spread, which is statistically significant at the 5% level. This liquidity effect seems more apparent in the trading of New Zealand stocks (Panel B). There is a significant (at the 1% level) decrease in both dollar and relative spreads, and a significant increase (at the 10% level) in depth. In the volume regression, the change variable is positive but not significant. Overall, the

5 Regression analysis is also conducted without including price difference as a control variable, and converting the AUD price series to New Zealand dollars using the prevailing exchange rate. The same conclusion is obtained. These results are available on request.

R-square is approximately 80% on average, indicating that the independent variables explain a significant portion of the variation in each liquidity measure.

Table 3-9 presents the 2SLS results for the liquidity impact on NZX after ASX switched to anonymous trading. When analysing this change in NZX liquidity, equation (3.5) can be expressed as:

𝑍𝑋 퐿 𝑞 𝑖𝑦

훽0 훽 𝑋 퐿 𝑞 𝑖𝑦 훽 𝑛 𝑖 훽3 𝑍𝑋 푀 𝐶 𝑛

훽4𝐶 𝑛𝑔 훼𝑖 𝜀𝑖 ( 7)

The dependent variable is the liquidity of a cross-listed stock i on NZX, the control variable Stock Control is the same liquidity measure of the same stock i on

ASX (ASX Liquidity), and Market Control is the average liquidity variable across

48 matched samples on NZX.

The 2SLS results provide evidence that ASX’s move to an anonymous market adversely affects the liquidity on NZX, supporting Hypothesis3.1B. Table 3-9 shows that the trading of Australian stocks on NZX deteriorates, with a decline in volume at the 10% significance level. The deterioration in liquidity is more evident in New Zealand stocks. There is a considerable increase in both percentage and dollar spreads, and decrease in depth, which are all statistically significant at the 1% level. These results also find evidence of liquidity commonality. NZX liquidity of cross-listed stocks co-moves with ASX liquidity and market liquidity.

Table 3-9: 2SLS Results – Impact of ASX Anonymity on NZX Liquidity This table reports the 2SLS estimation of equation (3.7). The instruments include three lags of ASX liquidity measures and the natural logarithm of trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (ASX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread Depth ($) Volume ($) Constant -1.8 -0.63 9.67 -4.8 -2.82 -3.24 5.08 1.44 (-2.5)** (-0.9) (6.1)*** (-1.8)* (-6.4)*** (-8.5)*** (3.7)*** (0.9)

ASX Liquidity 0.42 0.36 0.13 1.10 0.26 0.23 0.56 0.65 (4.7)*** (4.1)*** (2.6)*** (4.1)*** (3.8)*** (3.5)*** (7.5)*** (4.9)***

Price Difference -1.75 -1.59 -5.86 -6.58 3.8 4.11 4.15 7.25 (-1.2) (-1.0) (-4.1)*** (-1.5) (2.4)** (2.6)*** (4.3)*** (3.3)***

Market Control -0.07 0.36 -0.15 0.08 0.28 0.12 0.08 0.57 (-0.5) (2.3)** (-0.9) (0.6) (3.6)*** (1.4) (0.5) (4.6)***

Change -0.01 -0.02 -0.05 -0.15 0.06 0.07 -0.09 -0.05 (-0.4) (-0.5) (-1.3) (-1.8)* (3.5)*** (4.1)*** (-2.8)*** (-0.8)

R-squared 0.6 0.74 0.6 0.43 0.751 0.713 0.737 0.617

Hausman Specification Test 0.63 1.42 0.6 0.59 2.14** 1.70* 1.12 1.52 Durbin-Wu-Hausman Chi-sq Test 0.29 0.38 0.44 18.63*** 2.03 1.06 21.62*** 13.03*** First-stage Coefficients Lag 1 of ASX Liquidity 0.23*** 0.26*** 0.26*** 0.29*** 0.16*** Lag 2 of ASX Liquidity 0.14*** 0.05* 0.05*** 0.16*** 0.08*** Lag 3 of ASX Liquidity Trade Size ($) -0.06*** -0.06*** Partial R-sq 0.09 0.11 0.11 0.15 0.04 Partial F-statistic 18.92*** 43.04*** 43.61*** 87.28** 20.99*** Anderson canon. corr. LM statistic 86.88*** 143.22*** 145.40*** 196.22*** 50.72*** Sargan Test 0.23 0.05 0.68 0.84 0.19 LM Test of Redundancy Lag 1 of ASX Liquidity 47.08*** 86.82*** 89.67*** 110.62*** 35.15*** Lag 2 of ASX Liquidity 20.44*** 4.26** 4.22** 37.40*** 8.63*** Lag 3 of ASX Liquidity Trade Size ($) 25.09*** 25.13***

In summary, the results from testing the first pair of hypotheses show that liquidity increases on ASX and decreases on NZX, after ASX switched to anonymous trading. In addition to an improvement in dollar spread on ASX, and a decline in volume on NZX in the trading of Australian stocks, liquidity migration effect is more apparent in the trading of New Zealand stocks. Both dollar and relative spreads and quoted depth improve significantly on ASX, and deteriorate drastically on NZX. These results provide support to Foucault et al. (2007) that anonymity makes limit order book traders more willing to trade. Strong evidence is also found for market-wide commonality in liquidity. Liquidity of cross-listed stocks in domestic and foreign markets and the overall market movement are directly related, providing support to Chordia et al. (2000), Brockman (2009) and

Majois (2007).

Moreover, there is some evidence of an increase in ASX’s market share of

Australian domestic stocks (generally large and liquid stocks) after ASX adopted an anonymity regime. ASX’s volume has marginally increased (see Table 3-8) while NZX’s volume has decreased at the 10% significance level (see Table 3-9).

These results are somewhat consistent with Comerton-Forde and Tang (2009), who find that ASX’s market share increases in the large stocks that are cross-listed on NZX after ASX adopted an anonymity regime.

3.7.2 Liquidity Impact of NZX Anonymity

This section presents the results for Hypothesis3.2A and Hypothesis3.2B and provides evidence of liquidity impact after NZX’s adoption of a broker anonymity

regime. The 2SLS estimation equation (3.6) examines the change in ASX liquidity, where Market Control is the average liquidity variable across 44 matched stocks on ASX.

Consistent with Tables 3-8 and 3-9, Table 3-10 shows again the commonality in liquidity, with liquidity of cross-listed stocks on ASX co-moving with NZX liquidity and market liquidity.

After controlling for stock-specific and market-wide factors, the post-anonymity dummy variable shows decreased liquidity of cross-listed stocks on ASX after

NZX introduced anonymous trading, supporting Hypothesis3.2A. Panel A reveals that Australian stocks experience a significant increase in percentage and dollar spreads, and a significant decline in dollar depth and volume. These declines are statistically significant at least the 5% level. Panel B indicates that the liquidity impact is less significant in the trading of New Zealand stocks. There is an increase in the percentage spread, which is statistically significant at the 10% level, while no significant change is observed in depth and volume regressions.

Table 3-10: 2SLS Results – Impact of NZX Anonymity on ASX Liquidity This table reports the 2SLS estimation of equation (3.6). The instruments include three lags of NZX liquidity measures and the natural logarithm of trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (NZX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread Depth ($) Volume ($) Constant -0.22 -3.09 1.98 -3.71 2.53 1.50 -0.70 -5.92 (-0.6) (-11.2)*** (1.8)* (-1.9)* (2.7)*** (1.6)* (-0.4) (-1.8)*

NZX Liquidity 0.07 0.02 0.04 0.49 0.79 0.67 0.67 1.31 (2.3)** (2.6)*** (1.8)* (3.9)*** (4.1)*** (3.1)*** (6.8)*** (4.8)***

Price Difference 0.26 0.81 0.92 2.44 0.14 -0.51 2.54 5.43 (0.4) (1.9)* (0.8) (1.3) (0.1) (-0.5) (2.1)** (1.4)

Market Control 0.50 0.48 0.60 0.62 0.51 0.54 0.28 -0.18 (7.9)*** (7.3)*** (6.3)*** (5.1)*** (3.2)*** (2.9)*** (2.1)** (-0.7)

Change 0.10 0.05 -0.18 -0.14 0.06 0.02 -0.03 -0.08 (7.7)*** (4.6)*** (-5.7)*** (-2.3)** (1.7)* (0.5) (-0.7) (-0.9)

R-squared 0.98 0.84 0.93 0.92 0.68 0.98 0.66 0.51

Hausman Specification Test 2.71*** 1.23 1.22 0.84 1.73* 1.91* 1.96** 2.62*** Durbin-Wu-Hausman Chi-sq Test 0.19 0.48 1.95 17.75*** 9.09*** 4.51** 15.60*** 23.37*** First-stage Coefficients Lag 1 of NZX Liquidity 0.30*** 0.14*** 0.18*** 0.12*** Lag 2 of NZX Liquidity 0.12*** 0.13*** 0.18*** 0.13*** 0.14*** 0.11*** Lag 3 of NZX Liquidity 0.14*** 0.15*** Trade Size ($) 0.17*** Partial R-sq 0.13 0.04 0.07 0.05 0.16 0.03 Partial F-statistic 55.54*** 15.51*** 24.39*** 17.67*** 54.88*** 13.86*** Anderson canon. corr. LM statistic 128.55*** 38.97*** 82.05*** 63.35*** 183.75*** 35.98*** Sargan Test 0.00 1.51 0.06 0.05 0.91 0.35 LM Test of Redundancy Lag 1 of NZX liquidity 85.52*** 18.98*** 42.07*** 18.71*** Lag 2 of NZX Liquidity 14.44*** 14.70*** 38.27*** 27.60*** 25.75*** 12.76*** Lag 3 of NZX Liquidity 20.42*** 20.77*** Trade Size ($) 95.09***

Table 3-11 presents the 2SLS results of the anonymity effects for the trading of cross-listed stocks on NZX. The estimation is based on equation (3.7), where market control is the average liquidity variable of 44 matched stocks on NZX.

Table 3-11 shows again the evidence of liquidity commonality in the Australian and New Zealand markets. The coefficient of Market Control variable is statistically significant in most regressions. The post-anonymity Change variable shows that the switch to an anonymous market by NZX leads to a significant improvement in NZX market share in the trading of cross-listed stocks, supporting

Hypothesis3.2B. Trading volume increases drastically on NZX in the trading of both Australian and New Zealand companies, which is statistically significant at least at the 5% level, while, as shown in the previous table, trading volume decreases considerably in ASX in the trading of Australian stocks. These findings provide support to Poskitt et al. (2011), who show an increase in NZX’s share of trading in cross-listed stocks after the switch to anonymous trading.

There is no significant change in spreads and depth, which could be due to the fact that at the time when NZX adopted an anonymity regime, ASX had already been operating anonymously. These results stand in contrast to Poskitt et al. (2011), whose univariate and multivariate analyses show an increased effective spread associated with the removal of broker identifiers in NZX. This can be explained by the methodology in Poskitt et al. (2011). They use OLS and 2SLS without controlling for liquidity commonality. Given the global financial crisis in August

2007, results obtained by Poskitt et al. (2011) are probably not surprising.

Table 3-11: 2SLS Results – Impact of NZX Anonymity on NZX Liquidity This table reports the 2SLS estimation of equation (3.7). The instruments include three lags of ASX liquidity measures and trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (ASX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread Depth ($) Volume ($) Spread (%) Spread Depth ($) Volume ($) Constant 2.12 2.78 6.82 -2.06 -2.87 -2.54 2.75 4.50 (2.7)*** (2.2)** (3.4)*** (-0.9) (-1.2) (-7.2)*** (1.7)* (2.6)***

ASX Liquidity 0.87 0.96 0.29 0.75 0.35 0.30 0.53 0.52 (4.6)*** (3.0)*** (3.1)*** (4.5)*** (7.0)*** (5.7)*** (6.1)*** (4.9)***

Price Difference -0.18 0.28 -0.25 -2.13 -0.89 -1.15 -1.80 -2.25 (-0.2) (0.2) (-0.2) (-0.8) (-1.0) (-1.3) (-2.1)** (-0.8)

Market Control 0.49 0.40 -0.06 0.37 0.24 0.23 0.34 0.42 (2.9)*** (2.1)** (-0.3) (2.7)*** (2.5)** (2.3)** (2.6)*** (4.2)***

Change -0.04 0.02 -0.01 0.22 -0.02 -0.01 0.00 0.14 (-0.8) (0.5) (-0.2) (2.6)*** (-0.6) (-0.4) (0.1) (2.3)**

R-squared 0.50 0.84 0.38 0.41 0.81 0.71 0.75 0.73

Hausman Specification Test 2.48** 2.76*** 1.81* 2.11** 2.49** 1.55 1.56 3.58*** Durbin-Wu-Hausman Chi-sq Test 4.00** 3.15* 10.84*** 5.12** 23.36*** 15.75*** 14.48*** 7.77*** First-stage Coefficients Lag 1 of ASX Liquidity 0.39*** 0.98*** 0.28*** 0.33*** 0.31*** 0.30*** 0.22*** 0.15*** Lag 2 of ASX Liquidity 0.22*** 0.15*** 0.22*** 0.13*** 0.19*** 0.18*** 0.11*** 0.16*** Lag 3 of ASX Liquidity Trade Size ($) 0.33*** Partial R-sq 0.27 0.13 0.34 0.17 0.2 0.18 0.08 0.06 Partial F-statistic 72.19*** 31.90*** 126.13*** 30.95** 125.32*** 109.75*** 43.03** 30.58*** Anderson canon. corr. LM statistic 271.92*** 130.32*** 338.65*** 162.50*** 232.06*** 205.96*** 89.47*** 62.27*** Sargan Test 0.11 0.00 0.77 1.26 0.01 0.08 0.13 1.38 LM Test of Redundancy Lag 1 of ASX Liquidity 133.82*** 73.53*** 82.85*** 99.25*** 109.27*** 101.54*** 55.65*** 25.68*** Lag 2 of ASX Liquidity 47.99*** 21.83*** 64.91** 17.14** 46.65*** 39.91** 15.89*** 27.41*** Lag 3 of ASX Liquidity Trade Size ($) 135.24*** 22.04***

3.8 Robustness Check

Maher et al. (2008) find that using different econometric models, such as the method of OLS and instrumental variables, estimation comes to opposite conclusions for spread measures. In order to obtain an unbiased conclusion for anonymity effects, this section provides additional tests to examine the robustness of the reduction in bid-ask spreads and increase in quoted depth and volume after the change to an anonymous market structure. It firstly employs the 2SLS using a different length of event window during overlapping trading hours, and secondly conducts difference-in-difference estimation using the second set of non- overlapping data.

3.8.1 Length of Event Window

To examine whether the improved liquidity still holds in a longer period after the change to an anonymous market structure, it would be of interest to have a more extended post-anonymity period to further analyse the findings. The instrumental variables 2SLS re-estimates all liquidity measures using data from three months pre-event to four months post-event.

Table 3-12 reports the results of equation (3.6) for the liquidity impact on ASX trading, while Table 3-13 reports the results of equation (3.7) for the liquidity impact on NZX trading after ASX’s switch to a broker anonymity regime. The results show an improvement of bid-ask spreads and depth on ASX, and a deterioration of bid-ask spreads, depth and volume on NZX. This indicates a significant liquidity migration from NZX to ASX, in particular for New Zealand 91

stocks. All coefficient estimates are consistent with original results, suggesting the robustness of the results to the length of the event window around ASX’s anonymity regime change.

Table 3-12: Sensitivity to Event Window – Impact of ASX Anonymity on ASX Liquidity This table presents the results from equation (3.6), using data from three months pre-event and four months post-event. The instruments include three lags of NZX liquidity measures and trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (NZX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread ($) Depth ($) Volume ($) Constant -1.18 -2.73 1.74 0.48 -2.10 -1.80 0.26 -1.76 (-3.3)*** (-8.6)*** (1.3) (0.3) (-1.7)* (-3.0) (0.2) (-0.9)

NZX Liquidity 0.01 -0.02 0.06 0.73 0.32 0.32 0.70 0.75 (0.3) (-0.6) (2.4)** (5.4)*** (9.2)*** (8.9)*** (7.9)*** (5.7)***

Price Difference 0.64 0.26 -1.35 4.10 0.91 0.76 -4.60 -6.18 (1.7)* (0.7) (-0.9) (1.7)* (0.9) (0.7) (-4.4)*** (-2.4)**

Market Control 0.61 0.52 0.68 0.35 0.17 0.19 0.15 0.12 (9.6)*** (7.2)*** (5.7)*** (2.4)** (1.4) (1.4) (1.0) (0.8)

Change -0.01 -0.02 -0.01 0.09 -0.06 -0.10 0.08 0.11 (-0.1) (-2.3)** (-0.2) (1.5) (-3.1)*** (-4.4)*** (2.4)** (1.6)

R -squared 0.98 0.79 0.92 0.88 0.74 0.77 0.70 0.66

Hausman Specification Test 2.45** 0.10 0.12 2.54** 1.05 1.51 0.74 1.18 Durbin-Wu-Hausman Chi-sq Test 3.09* 7.36*** 1.88 38.58*** 1.26 0.89 36.30*** 19.51*** First-stage Coefficients Lag 1 of NZX Liquidity 0.32*** 0.32*** 0.14*** 0.21*** Lag 2 of NZX Liquidity 0.14*** 0.14*** 0.09*** 0.21*** 0.10*** Lag 3 of NZX Liquidity 0.09*** Trade Size ($) 0.18*** Partial R-sq 0.15 0.16 0.03 0.16 0.06 Partial F-statistic 65.86*** 71.75*** 16.30*** 67.28*** 27.52*** Anderson canon. corr. LM statistic 180.62** 192.14*** 35.25*** 256.25*** 98.58*** Sargan Test 0.19 0.01 0.00 0.04 0.61 LM Test of Redundancy Lag 1 of NZX Liquidity 107.88*** 112.93*** 21.01*** 63.67*** Lag 2 of NZX Liquidity 22.63*** 24.51*** 9.63*** 198.02*** 16.31*** Lag 3 of NZX Liquidity 105.44*** Trade Size ($) 175.99***

Table 3-13: Sensitivity to Event Window – Impact of ASX Anonymity on NZX Liquidity This table presents the results from equation (3.7), using data from three months pre-event and four months post-event. The instruments include three lags of ASX liquidity measures and trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (ASX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread ($) Depth ($) Volume ($) Constant -2.27 -1.23 9.65 -3.31 -2.85 -3.44 5.10 0.57 (-3.4)*** (-1.9)* (6.6)*** (-1.8)* (-7.1)*** (-10.2)*** (3.9)*** (0.4)

ASX Liquidity 0.41 0.36 0.13 0.93 0.21 0.17 0.52 0.75 (5.1)*** (4.4)*** (3.1)*** (5.4)*** (3.4)*** (2.9)*** (7.6)*** (5.1)*** Price Difference -1.57 -1.62 -5.07 -4.82 3.03 3.35 3.78 7.40 (-1.2) (-1.2) (-3.8)*** (-1.3) (2.0)** (2.5)** (4.2)*** (3.3)*** Market Control -0.16 0.19 -0.15 0.15 0.32 0.13 0.12 0.53 (-1.2) (1.3) (-1.1) (1.1) (4.5)*** (1.7)* (0.9) (4.8)***

Change 0.01 0.03 0.00 -0.11 0.04 0.05 -0.08 -0.11 (0.2) (0.83) (-0.1) (-1.5) (2.4)** (3.5)*** (-2.9)*** (-1.7)*

R-squared 0.59 0.76 0.62 0.43 0.75 0.71 0.72 0.57

Hausman Specification Test 0.07 0.87 0.61 0.85 3.38*** 1.76* 0.84 0.83 Durbin-Wu-Hausman Chi-sq Test 0.13 0.38 1.18 14.98*** 0.46 0.01 21.32*** 18.62*** First-stage Coefficients Lag 1 of ASX Liquidity 0.29*** 0.27*** 0.27*** 0.29*** 0.15*** Lag 2 of ASX Liquidity 0.16*** 0.06** 0.06*** 0.17*** 0.07*** Lag 3 of ASX Liquidity Trade Size ($) -0.06*** -0.06*** Partial R-sq 0.17 0.11 0.11 0.16 0.03 Partial F-statistic 15.49*** 53.82*** 55.23*** 87.28** 21.10*** Anderson canon. corr. LM statistic 31.32*** 174.26*** 175.74*** 143.64*** 49.41*** Sargan Test 0.63 0.37 0.32 1.24 0.69 LM Test of Redundancy Lag 1 of ASX Liquidity 93.96*** 107.02*** 107.93*** 134.92** 34.61*** Lag 2 of ASX Liquidity 30.02*** 5.92** 6.33** 47.11*** 8.51*** Lag 3 of ASX Liquidity Trade Size ($) 29.78*** 29.80***

Table 3-14 and Table 3-15 further examine the liquidity impacts after NZX’s change to an anonymous broker regime and show that trading on ASX is seriously affected after NZX became anonymous. Bid-ask spreads, depth and trading volume have significantly deteriorated, which is more evident in Australian- incorporated stocks. Table 3-15 shows that NZX market share in the trading of cross-listed stocks has improved. These results provide further evidence on liquidity migration of cross-listed stocks from ASX to NZX after the anonymity regime change in NZX. All coefficient estimates are consistent with the original results in Section 3.7. The qualitatively similar conclusion using a different length of event window indicates that the results are robust.

Table 3-14: Sensitivity to Event Window – Impact of NZX Anonymity on ASX Liquidity This table presents the results from equation (3.6), using data from three months pre-event and four months post-event. The instruments include three lags of NZX liquidity measures and trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (NZX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread ($) Depth ($) Volume ($) Constant -0.50 -3.37 1.16 -4.75 2.54 2.11 0.31 -3.13 (-1.6) (-12.5)*** (1.1) (-2.4)** (3.0)*** (2.6)*** (0.2) (-1.1)

NZX Liquidity 0.06 0.04 0.04 0.54 0.97 0.91 0.64 1.11 (5.6)*** (4.5)*** (1.8)* (4.1) (5.4)*** (4.4)*** (7.5)*** (5.7)***

Price Difference -0.40 0.68 1.25 3.05 0.36 -0.04 2.60 4.51 (-0.6) (1.3) (1.3) (1.8)* (0.4) (-0.1) (2.4)** (1.4)

Market Control 0.43 0.40 0.67 0.65 0.33 0.43 0.22 -0.17 (7.9)*** (6.1)*** (6.9)*** (5.3)*** (1.9)* (2.2)** (1.7)* (-0.7)

Change 0.11 0.06 -0.14 -0.17 0.05 0.00 -0.07 -0.09 (10.3)*** (5.6)*** (-4.4)*** (-3.0)*** (1.7)* (0.1) (-1.9)* (-1.1)

R-squared 0.98 0.82 0.92 0.91 0.66 0.73 0.66 0.57

Hausman Specification Test 1.08 0.25 1.12 0.45 1.70* 2.94** 0.89 1.87* Durbin-Wu-Hausman Chi-sq Test 2.41 0.11 1.61 22.51*** 9.09*** 15.71*** 19.92*** 21.54*** First-stage Coefficients Lag 1 of NZX Liquidity 0.13*** 0.15*** 0.17*** 0.20*** 0.14*** Lag 2 of NZX Liquidity 0.10*** 0.19*** 0.13*** 0.15*** 0.12*** Lag 3 of NZX Liquidity Trade Size ($) 0.17*** Partial R-sq 0.04 0.08 0.06 0.17 0.04 Partial F-statistic 15.51*** 28.97*** 22.05*** 71.13*** 22.30*** Anderson canon. corr. LM statistic 33.52*** 106.41*** 79.53*** 234.53*** 57.00*** Sargan Test 1.34 1.21 1.18 0.56 0.61 LM Test of Redundancy Lag 1 of NZX Liquidity 17.47*** 56.96*** 41.67*** 59.01*** 28.87*** Lag 2 of NZX Liquidity 12.03*** 32.06*** 23.72*** 34.30*** 19.53*** Lag 3 of NZX Liquidity Trade Size ($) 106.88*** 96

Table 3-15: Sensitivity to Event Window – Impact of NZX Anonymity on NZX Liquidity This table presents the results from equation (3.7), using data from three months pre-event and four months post-event. The instruments include three lags of ASX liquidity measures and trade size. The Hausman specification test of simultaneity and Durbin-Wu-Hausman chi-sq test of endogeneity are used to determine whether a 2SLS procedure is necessary. Instruments are implemented if they passed the tests of relevance and validity. The condition of relevance is tested by examining the fit of the first-stage endogenous regressor (ASX Liquidity) on the full set of instruments. The first-stage coefficients, partial R-square and partial F statistics on the relevance of instruments are reported. The Sargan test of over-identifying restrictions and LM IV test of redundancy are used for the instruments’ validity. The Anderson canon.corr. LM statistic is to examine whether the equation is adequately identified. *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Panel A: Australian Stocks Panel B: New Zealand Stocks Spread (%) Spread ($) Depth ($) Volume ($) Spread (%) Spread ($) Depth ($) Volume ($) Constant 2.02 1.79 7.85 -0.82 -2.70 -2.40 3.85 6.05 (2.8)*** (1.4) (4.6)*** (-0.4) (-7.2)*** (-8.2)*** (2.7)*** (4.2)***

ASX Liquidity 0.98 1.08 0.17 0.63 0.32 0.26 0.52 0.45 (5.4)*** (4.2)*** (2.5)** (4.4)*** (7.0)*** (5.5)*** (6.3)*** (4.8)***

Price Difference 0.19 0.56 -0.24 -1.73 -0.93 -1.03 -1.83 -0.85 (0.2) (0.6) (-0.2) (-0.7) (-1.2) (-1.4) (-2.4)** (-0.4)

Market Control 0.40 0.29 -0.06 0.35 0.30 0.30 0.24 0.36 (2.9)*** (2.0)** (-0.4) (2.7)*** (3.6)*** (3.5)*** (2.0)** (4.0)***

Change -0.02 0.05 -0.06 0.24 -0.03 -0.03 0.00 0.10 (-0.4) (1.2) (-1.3) (3.0)*** (-1.4) (-1.3) (-0.0) (1.8)*

R-squared 0.49 0.84 0.37 0.40 0.82 0.71 0.73 0.79

Hausman Specification Test 1.5 1.78* 0.45 1.61 2.99*** 1.93 2.08** 4.58*** Durbin-Wu-Hausman Chi-sq Test 5.53** 5.03** 4.83** 2.78* 18.78*** 15.75*** 16.31*** 5.55** First-stage Coefficients Lag 1 of ASX Liquidity 0.43*** 0.34*** 0.31*** 0.35*** 0.31*** 0.30*** 0.22*** 0.16*** Lag 2 of ASX Liquidity 0.15*** 0.11*** 0.23*** 0.13*** 0.19*** 0.17*** 0.13*** 0.15*** Lag 3 of ASX Liquidity Trade Size ($) 0.35*** Partial R-sq 0.25 0.15 0.41 0.18 0.19 0.17 0.08 0.05 Partial F-statistic 75.75*** 46.54*** 161.24** 37.83** 139.14*** 120.35*** 53.55** 33.01*** Anderson canon. corr. LM statistic 292.68*** 114.96*** 471.54*** 198.01*** 263.98*** 231.28*** 113.96*** 57.55*** Sargan Test 0.01 0.08 1.07 0.82 0.05 0.26 0.11 1.41 LM Test of Redundancy Lag 1 of ASX Liquidity 177.76*** 112.87*** 123.43*** 124.83*** 128.99*** 118.91*** 66.68*** 30.32*** Lag 2 of ASX Liquidity 26.38*** 13.98*** 74.38** 17.75** 49.14*** 41.03** 24.25*** 29.47*** Lag 3 of ASX Liquidity Trade Size ($) 170.75***

3.8.2 Difference-in-difference Estimation

The difference-in-difference (DID) estimation is another approach that accounts for endogeneity, but without relying on instrumental variables. Gujarati (2007) indicates that identifying strong and valid instruments is not easy in practice. The

DID method compares the difference in outcomes in the treatment group before and after the change to the difference in outcomes in the untreated control group in the same period (Ashenfelter and Card, 1985). Blundell and Costa Dias (2000) indicate that the DID estimator can provide a robust estimate of the impact of the treatment if the data is in a longitudinal or repeated cross-section format. The basic logic behind the DID estimator is that if the liquidity of cross-listed stocks is just compared before and after the regulatory change, the effects of other factors around the time of the change cannot be excluded. By using a control group, outcomes are observed for two groups for two time periods, so as to difference out these confounding factors and isolate the effects of trader anonymity.

Using the DID technique, the liquidity impact can be estimated in the trading of cross-listed stocks first through time, before and after the broker regime change, and second across subjects, between cross-listed stocks and non-cross-listed control stocks. Based on Meyer (1995) model, the DID regression is constructed in the following format:

𝑖 𝑖 퐿 𝑞 𝑖𝑦 훼0 훼 훼 𝑖 훼3 𝜀𝑖 ( )

where 퐿 𝑞 𝑖𝑦 denotes the liquidity outcome for an individual stock i in period t, t=1 denotes the post-period of broker regime change and t=0 denotes the

pre-period, i=1 is indexed for the cross-listed stocks, i=0 for non-cross-listed and

1-2 matched control stocks as previously identified (see Appendices 3-1 and 3-2).

is a dummy variable taking value 1 in the post-period and 0 otherwise,

𝑖 𝑖 𝑖 =1 if i=1 and 0 otherwise, and =1 if i=1 and t=1 and 0 otherwise.

The key idea behind this regression is that 훼 summarises the way that both groups, cross-listed stocks and non-cross-listed control stocks, are influenced by time, and 훼 captures the time-invariant difference in overall means between the two groups. The estimated impact of change in a broker anonymity regime on liquidity for cross-listed stocks is 훼3, which is the coefficient on the interaction

𝑖 between and 𝑖 . This DID estimator not only accounts for the effects of stock-specific characteristics, but also removes trends in the market, such as overall market downturn due to the global financial crisis.

Slightly different from the analysis in Section 3.7, Equation (3.8) uses the second data set of normal trading hours over a six-month period around the event date.

퐿 𝑞 𝑖𝑦 is measured in four ways: effective spread, depth, volume and price volatility.

The effective spread is similar to the quoted bid-ask spread; however, it allows for trades to occur at different prices within or outside the quotes. As indicated by

Bessembinder et al. (2009), many electronic exchanges including ASX allow traders to hide some or all of the order size, implying that there are limit orders offering more attractive prices that differ from quoted prices. When trades occur

outside the quotes, the effective spread, which is based on the actual trade price, is a better measure of trading cost. The effective spread can be computed on a dollar basis and a percentage basis:

( ) | | ( )

( ) ( ) ( )

These effective spreads are measured as the difference between the actual price at which a trade occurs and the midpoint of the prevailing quoted spread at the time of the trade.

Quoted depth is measured in both quantity (number of shares offered at the best bid and ask) and in dollar value. Trading volume is the number of shares traded as well as their dollar amount. These two measures are transformed into natural logarithms when conducting the difference-in-difference analysis. Price volatility measures the amount of variability of dispersion of intraday prices around the average. It is calculated as the standard deviation of intraday traded prices according to Ederington and Lee (1993). Tables 3-16 and 3-17 present the univariate analysis for these liquidity measures during normal trading hours over a six-month period around the anonymity regime change. Consistent with Tables 3-

5 and 3-6, effective spreads and price volatility are narrower, and volume and depth are higher at the home market.

100

Table 3-16: Univariate Analysis of ASX Anonymity during Normal Trading Hours The table reports the cross-sectional average for the effective spreads, depth, volume and price volatility for ASX anonymity. Pre is the pre-event period, and Post is the post-event period. Difference is calculated as Post less Pre. t-statistics of the differences (from unequal variance) are estimated in parentheses6. Dollar values on ASX and NZX are specified in their respective currencies.

Trading on ASX Trading on NZX Pre Post Difference Pre Post Difference Panel A: Australian companies

Effective Spread ($) 0.011 0.011 0.000 0.060 0.063 0.003

(-0.6) (0.6)

Effective Spread (%) 0.344 0.365 0.021 0.626 0.594 -0.031

(0.5) (-0.7)

Price Volatility 0.028 0.026 -0.002 0.039 0.039 0.000

(-1.2) (-0.0)

Depth (000) 114 132 18 11 10 -2

(1.2) (-1.4)

Depth ($000) 471 421 -49 35 35 -1

(-0.8) (-0.3)

Volume (000) 4,853 3,423 -1,430 31 22 -10

(-2.4)** (-2.4)**

Volume ($000) 33,442 27,773 -5,669 188 135 -52

(-2.2)** (-2.0)**

Panel B: New Zealand Companies

Effective Spread ($) 0.033 0.028 -0.005 0.015 0.017 0.002

(-2.8)*** (3.4)***

Effective Spread (%) 0.526 0.453 -0.073 0.241 0.267 0.026

(-2.9)*** (3.7)***

Price Volatility 0.015 0.015 0.000 0.016 0.015 -0.001

(0.3) (-1.1)

Depth (000) 23 16 -7 55 31 -24

(-4.1)*** (-4.1)***

Depth ($000) 66 51 -15 148 86 -62

(-3.1)*** (-4.2)***

Volume (000) 384 400 16 758 699 -58

(0.4) (-0.8)

Volume ($000) 1,532 1,598 66 2,975 3,075 100

(0.3) (0.3)

6 Two-sample t-test on the mean difference is based on Selvanathan et al. (2006).

101

Table 3-17: Univariate Analysis of NZX Anonymity during Normal Trading Hours. The table reports the cross-sectional average for the effective spreads, depth, volume and price volatility for NZX anonymity. Pre is the pre-event period, and Post is the post-event period. Difference is calculated as Post less Pre. t-statistics of the differences (from unequal variance) are estimated in parentheses. Dollar values on ASX and NZX are specified in their respective currencies.

Trading on ASX Trading on NZX

Pre Post Difference Pre Post Difference Panel A: Australian companies

Effective Spread ($) 0.010 0.011 0.001 0.141 0.163 0.022

(3.2)*** (1.7)*

Effective Spread (%) 0.410 0.468 0.058 0.913 1.218 0.305

(1.1) (3.8)***

Price Volatility 0.029 0.039 0.010 0.049 0.063 0.014

(4.5)*** (2.7)***

Depth (000) 129 139 10 14 11 -2

(0.6) (-1.2)

Depth ($000) 581 584 3 26 27 0

(0.0) (0.3)

Volume (000) 4,933 6,360 1,427 49 48 -1

(1.8)* (-0.1)

Volume ($000) 48,032 51,585 3,554 140 235 95

(0.9) (1.7)*

Panel B: New Zealand Companies

Effective Spread ($) 0.036 0.041 0.005 0.016 0.021 0.005

(1.5) (6.4)***

Effective Spread (%) 0.689 0.964 0.274 0.354 0.468 0.114

(3.8)*** (2.4)**

Price Volatility 0.018 0.021 0.003 0.018 0.020 0.002

(3.0)*** (2.4)**

Depth (000) 12 10 -2 18 16 -2

(-2.9)*** (-2.6)***

Depth ($000) 41 31 -11 63 52 -11

(-3.6)*** (-4.5)***

Volume (000) 629 586 -43 914 1,202 288

(-0.6) (2.7)***

Volume ($000) 2,448 2,301 -147 4,105 5,317 1,211

(-0.5) (2.4)**

The t-statistics in Table 3-16 on the difference between the pre- and post- period show a marked decrease in the effective spreads on ASX, and a significant increase on NZX in the trading of New Zealand companies, after the introduction

102

of the broker anonymity regime by ASX. Table 3-16 further shows that liquidity of cross-listed stocks moves together in the home and foreign markets. Quoted depth decreases on both markets in the trading of New Zealand companies; while

Australian companies also record a decrease in trading volume on both markets.

This commonality in liquidity is also shown in Table 3-17. There is an increase in effective spreads and price volatility in the trading of both Australian and New

Zealand companies, after the introduction of anonymity on NZX. These results are consistent with the findings in Table 3-6.

Table 3-18 presents the results of DID equation (3.8) for the liquidity impact after

ASX switched to an anonymity regime. The Effect dummy variable in Panel A suggests that ASX anonymity has had a positive impact on ASX market.

Australian stocks have a significant reduction in price volatility (at the 5% significance level), while New Zealand stocks have a significant decrease in both dollar and relative effective spreads (at the 1% significance level), though there is no significant change in depth and volume. Panel B further shows that ASX’s anonymous trading adversely affects NZX liquidity of cross-listed stocks. For depth and volume liquidity measures, there is a strongly significant and deteriorated coefficient for Effect in the trading of Australian companies. In the trading of New Zealand companies, effective spread ($) and price volatility increase, quoted depth decreases on NZX. These results provide further support to the first pair of hypotheses (Hypothesis3.1A and Hypothesis3.1B), that the change to anonymous market in ASX leads to an improvement in liquidity on ASX, but a decline on NZX.

103

Table 3-18: DID Estimation of the Impact of ASX Anonymity This table presents the results for the DID estimation using the second data set calculated from normal trading hours during a six-month period around the broker regime change. Panels A and B report the results for the liquidity impact on ASX and NZX, respectively. Depth and trading volume have been transformed into natural logarithms. The variable of interest is Effect, which captures the effect of the change in trader anonymity by ASX on the various liquidity measures *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Australian Stocks New Zealand Stocks Price Volume Price Volume EF ($) EF (%) Depth Depth ($) Volume EF ($) EF (%) Depth Depth ($) Volume Volatility ($) Volatility ($) Panel A: Trading on ASX Constant 0.013 0.002 0.72 10.04 11.56 14.20 15.72 0.023 0.005 0.46 9.21 10.30 11.71 12.80

(86.5)*** (16.8)*** (21.1)*** (231.8)*** (328.7)*** (349.9)*** (228.0)*** (24.4)*** (26.9)*** (18.7)*** (294.2)*** (383.2)*** (214.5)*** (209.5)***

Dcross-listed -0.001 0.001 -0.04 0.31 0.16 -0.10 -0.26 0.010 0.000 -0.28 0.04 0.02 -0.63 -0.65

(-4.4)*** (4.1)*** (-0.7) (4.1)*** (2.5)** (-1.3) (-2.2)** (5.7)*** (1.4) (-6.4)*** (0.6) (0.4) (-6.6)*** (-6.1)***

TTime 0.000 0.000 0.12 -0.14 -0.04 -0.13 -0.04 0.004 0.000 0.02 -0.07 -0.02 -0.19 -0.15

(1.9)* (-0.2) (2.5)** (-2.2)** (-0.8) (-2.2)** (-0.4) (2.8)*** (0.5) (0.4) (-1.5) (-0.6) (-2.4)** (-1.7)*

Effect -0.001 0.000 -0.20 0.09 -0.01 -0.02 -0.12 -0.009 -0.001 -0.02 0.02 -0.02 0.13 0.09

(-1.4) (0.7) (-2.4)** (0.9) (-0.2) (-0.2) (-0.7) (-3.8)*** (-2.0)** (-0.3) (0.3) (-0.2) (0.9) (0.5)

Panel B: Trading on NZX Constant 0.033 0.007 0.03 9.07 9.80 9.61 10.35 0.033 0.006 0.34 9.04 10.14 10.26 11.36

(20.8)*** (42.0)*** (0.7) (285.6)*** (402.4)*** (192.1)*** (199.5)*** (55.8)*** (50.1)*** (14.6)*** (296.3)*** (429.3)*** (232.6)*** (247.9)***

Dcross-listed 0.031 -0.002 0.97 -0.39 0.40 -0.25 0.54 -0.018 -0.004 -0.13 0.74 0.82 2.16 2.24

(11.3)*** (-8.1)*** (17.7)*** (-7.2)*** (9.6)*** (-2.9)*** (6.1)*** (-18.4)*** (-18.0)*** (-3.6)*** (14.5)*** (20.6)*** (29.3)*** (29.2)***

TTime 0.001 0.001 -0.02 0.08 0.02 0.02 -0.04 -0.001 0.000 -0.16 0.04 -0.07 -0.17 -0.27

(0.4) (2.7)*** (-0.4) (1.7)* (0.4) (0.3) (-0.5) (-1.0) (2.8)*** (-4.7)*** (0.8) (-2.1)** (-2.6)*** (-4.1)***

Effect 0.003 0.000 0.06 -0.21 -0.10 -0.29 -0.19 0.003 0.000 0.15 -0.16 -0.06 -0.13 -0.04

(0.7) (-0.6) (0.7) (-2.6)*** (-1.7)* (-2.3)** (-1.4) (1.9)* (-0.7) (2.9)*** (-2.1)** (-1.1) (-1.2) (-0.3)

104

Table 3-19 presents the difference-in-difference analysis for NZX anonymity. The TTime coefficient shows that the Australian and New Zealand markets experience an overall downturn. There is generally an increase in effective spreads, and a decrease in depth and volume. This confirms again liquidity commonality in financial markets. Moreover, the Effect dummy in Panel A shows positive and significant coefficients of relative effective spreads for Australian and New Zealand companies, which also record marginally deteriorated dollar depth. On the other hand, Panel B shows an improved liquidity on NZX after becoming anonymous. In general, there is an increase in quoted depth and dollar trading volume for both Australian and New Zealand stocks. The effective spreads decrease in the trading of New Zealand companies. These improvements are statistically significant at least at the 10% level. These results further support the second pair of hypotheses, Hypothesis3.2A and Hypothesis3.2B, that liquidity deteriorates on ASX, and improves on NZX, after the switch to an anonymous regime on NZX.

105

Table 3-19: DID Estimation of the Impact of NZX Anonymity This table presents the results for the DID estimation using the second data set calculated from normal trading hours during a six-month period around the broker regime change. Panels A and B report the results for the liquidity impact on ASX and NZX, respectively. Depth and trading volume have been transformed into natural logarithms. The variable of interest is Effect, which captures the effect of the change in trader anonymity by NZX on the various liquidity measures *. **. *** denote statistical significance at the 10%, 5% and 1% levels.

Australian Stocks New Zealand Stocks Price Volume Price Volume EF ($) EF (%) Depth Depth ($) Volume EF ($) EF (%) Depth Depth ($) Volume Volatility ($) Volatility ($)

Panel A: Trading on ASX Constant 0.012 0.002 1.09 9.83 11.71 14.68 16.57 0.018 0.005 0.61 9.20 10.31 11.46 12.45

(86.7)*** (13.1)*** (30.9)*** (198.7)*** (308.0)*** (359.7)*** (240.7)*** (18.5)*** (17.9)*** (21.3)*** (260.2)*** (399.7)*** (285.2)*** (239.4)***

Dcross-listed -0.002 0.002 -0.54 0.56 0.01 -0.22 -0.77 0.018 0.002 -0.34 -0.08 -0.32 0.27 -0.68

(-8.3)*** (5.7)*** (-8.8)*** (6.5)*** (0.2) (-3.1)*** (-6.4)*** (10.2)*** (4.4)*** (-6.5)*** (-1.3) (-7.1)*** (4.1)*** (-7.3)***

TTime 0.001 0.000 0.19 -0.23 -0.25 0.00 -0.01 0.002 0.000 0.24 -0.21 -0.11 -0.13 0.08

(3.6)*** (-1.0) (3.7)*** (-3.4)*** (-4.6)*** (-0.0) (-0.1) (1.4) (-0.6) (5.9)*** (-4.2)*** (-2.9)*** (-2.3)** (1.1)

Effect 0.000 0.001 0.01 0.00 -0.02 0.05 0.03 0.003 0.003 -0.10 0.06 -0.11 -0.14 -0.09

(-0.2) (2.0)** (0.2) (-0.0) (-0.2) (0.4) (0.2) (1.2) (4.4)*** (-1.3) (0.6) (-1.8)* (-1.5) (-0.7)

Panel B: Trading on NZX Constant 0.032 0.007 0.06 8.94 9.78 9.72 10.54 0.034 0.005 0.26 8.90 10.07 10.44 10.29

(8.2)*** (22.4)*** (1.4) (241.6)*** (395.3)*** (186.3)*** (203.9)*** (35.6)*** (26.4)*** (10.6)*** (336.4)*** (488.8)*** (230.1)*** (221.7)***

Dcross-listed 0.109 0.002 0.94 -0.56 0.08 -0.35 0.32 -0.018 -0.001 -0.04 0.69 0.58 2.37 2.53

(16.5)*** (4.7)*** (14.9)*** (-9.0)*** (2.0)** (-3.8)*** (3.4)*** (-10.9)*** (-4.4)*** (-0.9) (15.2)*** (16.4)*** (30.4)*** (32.6)***

TTime 0.013 0.004 0.20 -0.27 -0.36 -0.06 -0.16 0.010 0.002 0.18 -0.21 -0.26 -0.08 -0.13

(2.4)** (8.3)*** (3.6)*** (-5.1)*** (-10.4)*** (-0.8) (-2.2)** (7.8)*** (6.1)*** (5.1)**** (-5.6)*** (-8.9)*** (-1.2) (-1.9)**

Effect 0.009 0.000 0.06 0.17 0.32 0.15 0.31 -0.006 0.000 0.00 0.01 0.09 0.15 0.20

(0.9) (-0.6) (0.6) (1.9)* (5.4)*** (1.1) (2.4)** (-2.5)** (-1.1) (0.1) (0.1) (1.9)* (1.4) (1.9)*

106

3.9 Summary

This chapter examines the liquidity impact of the limit order anonymity change in

Australia and New Zealand markets using both the 2SLS and difference-in- difference techniques. The 2SLS estimation uses the first data set of overlapping trading hours. After controlling for both stock-specific and market-wide liquidity factors, results show that spreads decline, quoted depth and trading volume increase with the introduction of an anonymous market. Trader anonymity attracts the trading of cross-listed stocks from the foreign counterpart. ASX’s introduction of anonymous trading leads to a significant improvement of bid-ask spreads and quoted depth in ASX, and a significant deterioration of liquidity in the foreign market, NZX. On the other hand, NZX’s adoption of an anonymity regime leads to a significant improvement of spreads and quoted depth in NZX, and a significant deterioration in ASX, the foreign market. These results support

Foucault et al. (2007) that limit order book traders are more willing to trade aggressively and reduce bid-ask spreads in an anonymous venue. Moreover, results indicate that the anonymity impact on liquidity is more apparent in the trading of foreign cross-listed stocks, which are less liquid compared to their home market trading. This difference may be due to the higher probability of information-based trading in less liquid foreign stocks (Easley et al., 1996).

Stocks with greater information asymmetry seem more likely to be traded in an anonymous market. These results are consistent with Garfinkel and Nimalendran

(2003), who show that traders in an anonymous trading venue do not actively adjust to the presence of informed trading by raising effective spreads.

107

The difference-in-difference estimation uses the second data set of normal trading hours. It compares liquidity of cross-listed stocks before and after the broker regime change after controlling for a market-wide factor. Results show that effective spreads and price volatility improve on ASX, but deteriorate on NZX.

Depth and volume also decrease on NZX, after ASX introduced anonymous trading. On the other hand, when considering the liquidity impact after NZX adopted anonymous trading, quoted depth and relative effective spreads deteriorate on ASX, but improve on NZX. Slightly different from the 2SLS estimation, there is no apparent difference in anonymity effects between the trading on home and foreign markets. This is probably because the 2SLS equation

(3.5) controls the individual determinants of liquidity in the home (foreign) market by using the same liquidity measure of the same stock in the foreign

(home) market. Nevertheless, the DID estimation is consistent with 2SLS results, showing that anonymous trading has a positive impact on the liquidity migration of cross-listed stocks.

Interestingly, without controlling for commonality in liquidity, the univariate analysis shows that liquidity deteriorates in both Australian and New Zealand markets after NZX adopted an anonymity regime. In fact, the global financial crisis caused the equity markets to crash in early August 2007, resulting in a worldwide market downturn. Consistent with Chordia et al. (2000), Brockman

(2009) and Majois (2007), the inclusion of market-wide commonality may be necessary in natural experimental studies.

These findings hold an important implication for market design. The adoption of

108

an anonymity regime has yielded significant benefits for both ASX and NZX in the trading of cross-listed stocks. NZX’s switch to anonymous trading not only halts the migration of trading from NZX to ASX, but also increases NZX’s liquidity in the trading of cross-listed stocks. Future research on cross-listed stocks could examine whether these results hold on other stock exchanges.

109

References

Ashenfelter, O., and Card, D., 1985. Using the Longitudinal Structure of Earnings to Estimate the Effect of Training Programs. The Review of Economics and

Statistics, 67(4), pp.648-660.

ASX, 2005. Enhancing the Liquidity of Australian Equity Market, accessed

1/11/2012. https://www.asxonline.com/intradoc- cgi/groups/participant_services/documents/communications/asx014870.pdf

Bacidore, J.M., and Sofianos, G., 2002. Liquidity provision and specialist trading in NYSE-listed non-U.S. stocks. Journal of Financial Economics, 63(1), pp.133-

158.

Benveniste, L.M., Marcus, A.J., and Wilhelm, W.J., 1992. What’s special about the specialist? Journal of Financial Economics, 32(1), pp.61-86.

Bessembinder, H., Panayides, M., and Venkataraman, K., 2009. Hidden Liquidity:

An analysis of order exposure strategies in electronic stock markets. Journal of

Financial Economics, 94(3), pp.361-383.

Blundell, R., and Costa Dias, M., 2000. Evaluation Methods for Non-

Experimental Data. Fiscal Studies, 21(4), pp.427-468.

Brockman, P., Chung, D.Y., and Perignon, C., 2009. Commonality in liquidity: A global perspective, Journal of Financial and Quantitative Analysis, 44(4), pp.851-

882.

110

Chordia, T., Roll, R., and Subrahmanyam, A., 2000. Commonality in liquidity.

Journal of Financial Economics, 56(1), pp.3-28.

Comerton-Forde, C., Frino, A., and Mollica, V., 2005. The impact of limit order anonymity on liquidity: Evidence from Paris, Tokyo and Korea. Journal of

Economics and Business, 57(6), pp.528-540.

Comerton-Forde, C., and Tang, K.M., 2009. Anonymity, liquidity and fragmentation. Journal of Financial Markets, 12(3), pp.337-367.

Easley, D., Kiefer, N.M., O’Hara, M., and Paperman, J.B., 1996. Liquidity,

Information, and Infrequently Traded Stocks. The Journal of Finance, 51(4), pp.1405-1436.

Ederington, L.H., and Lee, Jae-ha, 1993. How Markets Process Information:

News Releases and Volatility. The Journal of Finance, 48(4), pp.1161-1191.

Eom, K.S., Ok, J., and Park, J.-H., 2007. Pre-trade transparency and market quality. Journal of Financial Markets, 10(4), pp.319–341.

Forster, M.M., and George, T.J., 1992. Anonymity in securities markets. Journal of Financial Intermediation, 2(2), pp.168-206.

Foucault, T., Moinas, S., and Theissen, E., 2007. Does Anonymity Matter in

Electronic Limit Order Markets? Review of Financial Studies, 20(5), pp.1707-

1747.

111

Garfinkel, J.A., and Nimalendran, M., 2003. Market Structure and Trader

Anonymity: An Analysis of Insider Trading. Journal of Financial and

Quantitative Analysis, 38(3), pp.591-610.

Goldstein, M.A., and A. Kavajecz, K., 2000. Eighths, sixteenths, and market depth: changes in tick size and liquidity provision on the NYSE. Journal of

Financial Economics, 56(1), pp.125-149.

Grammig, J., Schiereck, D., and Theissen, E., 2001. Knowing me, knowing you:

Trader anonymity and informed trading in parallel markets. Journal of Financial

Markets, 4(4), pp.385-412.

Grammig, J., Melvin, M., and Schlag, C., 2005. Internationally cross-listed stock prices during overlapping trading hours: price discovery and exchange rate effects.

Journal of Empirical Finance, 12(1), pp.139-164.

Gujarati, D.N., 2007. Basic Econometrics, McGraw, Australia.

Hachmeister, A., and Schiereck, D., 2010. Dancing in the dark: post-trade anonymity, liquidity and informed trading. Review of Quantitative Finance and

Accounting, 34(2), pp.145-177.

Harris, L.E., 1994. Minimum price variations, discrete bid-ask spreads, and quotation sizes. The Review of Financial Studies, 7(1), pp.149-178.

Harris, L., 2003. Trading and exchanges: market microstructure for practitioners,

Oxford, New York.

112

Hasbrouck, J., 1991. Measuring the information content of stock trades, The

Journal of Finance, 46(1), pp.179-207.

Heidle, H.G., and Huang, R.D., 2002. Information-Based Trading in Dealer and

Auction Markets: An Analysis of Exchange Listings. The Journal of Financial and Quantitative Analysis, 37(3), pp.391-424.

Lee, C.M., Mucklow, B., and Read, M.J., 1993. Spreads, Depths, and the Impact of Earnings Information: An Intraday Analysis. The Review of Financial Studies,

23(2), pp.345-374.

Madhavan, A., 1995. Consolidation, Fragmentation, and the Disclosure of

Trading Information. The Review of Financial Studies, 8(3), pp.579-603.

Maher, O., Swan, P.L., and Westerholm, P.J., 2008. Twilight falls on the limit order book: Endogeneity and the demise of broker identity. Working Paper,

University of New South Wales.

Majois, C., 2007. Natural experiments methodology and global liquidity in financial markets. Working paper, Louvain School of Management and FUCaM.

Meyer, B. D., 1995. Natural and quasi-experiments in economics. Journal of

Business & Economic Statistics,13(2), pp.151-161.

O’Hara, M., 1995. Market Microstructure Theory, Blackwell Publishing,

Cambridge.

113

Poskitt, R., Marsden, A., Nguyen, N., and Shen, J., 2011. The introduction of broker anonymity on the New Zealand Exchange. Pacific Accounting Review,

23(1), pp.34-51.

Rindi, B., 2008. Informed Traders as Liquidity Providers: Anonymity, Liquidity, and Price Formation. Review of Finance, 12(3), pp.497–532.

Selvanathan, A., Selvanathan, S., Keller, G., and Warrack, B., 2006. Australian

Business Statistics, Thomson, Australia.

Simaan, Y., Weaver, D.G., and Whitcomb, D.K., 2003. Market Maker Quotation

Behavior and Pretrade Transparency. The Journal of Finance, 58(3), pp.1247-

1268.

Shleifer, A., and Vishny, R.W., 1997. The Limits of Arbitrage. The Journal of

Finance, 52(1), pp.35-55.

114

Appendices

Appendix 3-1: 1-2 Matched Stocks for ASX Anonymity

Sample Stocks Matched Stocks from NZX Matched Stocks from ASX Panel A: Australian Companies Australian Foundation Investment Co. Ltd NZX Limited Sydney Airport Team Talk Limited Bendigo and Adelaide Bank Limited AMP Ltd Pyne Gould Corporation Limited Insurance Australia Group Limited Hellaby Holdings Limited Alumina Limited Australia and New Zealand Banking Group Ltd Tourism Holdings Limited Newcrest Mining Limited Rubicon Limited Commonwealth Bank of Australia AXA Asia Pacific Holdings Limited Turners & Growers Limited GPT Group Lyttelton Port Company Limited Qantas Airways Limited Lion Nathan Ltd Property for Industry Limited Aristocrat Leisure Limited Methven Limited Gunns Limited Pacific Brands Limited NZF Group Limited Boral Limited The Colonial Motor Company Limited Paladin Energy Ltd Pan Pacific Petroleum NL Comvita Limited Cue Energy Resources Limited Smiths City Group Limited Molopo Energy Limited Summit Resources Limited Renaissance Corporation Limited Indophil Resources NL Kingfish Limited Kingsgate Consolidated Limited Telstra Corporation Ltd Dorchester Pacific Limited BHP Billiton Limited Horizon Energy Distribution Limited Tabcorp Holdings Limited Westpac Banking Corporation Pumpkin Patch Limited QBE Insurance Group Limited Scott Technology Limited Suncorp Group Limited Panel B: New Zealand Companies Auckland International Airport Ltd Goodman Property Trust Australian Infrastructure Fund Infratil Limited Virgin Australia Holdings Limited Air New Zealand Ltd Ryman Healthcare Limited Coffey International Limited Tenon Limited Data3 Limited Carter Holt Harvey Limited Kiwi Income Property Trust Investa Office Fund Northland Port Corporation (NZ) Limited Primary Health Care Limited Fletcher Building Ltd Vector Limited Flight Centre Limited The New Zealand Refining Company Limited WorleyParsons Limited Fisher & Paykel Appliances Holdings Ltd Skellerup Holdings Limited REA Group Ltd Cavalier Corporation Limited Blackmores Limited Fisher & Paykel Healthcare Corporation Ltd Ebos Group Limited Redflex Holdings Limited Briscoe Group Limited Australand Property Group Nuplex Industries Ltd PGG Wrightson Limited Zimplats Holdings Limited Sanford Limited Discovery Metals Limited New Zealand Oil & Gas Ltd Restaurant Brands New Zealand Limited Amalgamated Holdings Limited Abano Healthcare Group Limited RuralAus Investments Limited Sky City Entertainment Group Ltd Port of Tauranga Limited (NS) GUD Holdings Limited Freightways Limited CPT Global Limited Telecom Corporation of New Zealand Ltd Contact Energy Limited Transurban Group TrustPower Limited Amcor Limited Tower Ltd Michael Hill International Limited InvoCare Limited Steel & Tube Holdings Limited Transfield Services Limited Vea Advantage Limited CDL Investments New Zealand Limited David Jones Limited Wellington Drive Technologies Limited Ten Network Holdings Limited Waste Management NZ Limited Allied Farmers Limited Colorpak Limited Veritas Investments Limited JB Hi-Fi Limited The Warehouse Group Limited Hallenstein Glasson Holdings Limited Northern Star Resources Ltd Mainfreight Limited Allied Gold Mining PLC

115

Appendix 3-2: 1-2 Matched Stocks for NZX Anonymity

Sample Stocks Matched Stocks from NZX Matched Stocks from ASX Panel A: Australian Companies Australian Foundation Investment Co. Ltd NZX Limited Sydney Airport Team Talk Limited Bendigo and Adelaide Bank Limited AMP Ltd Pyne Gould Corporation Limited Insurance Australia Group Limited Hellaby Holdings Limited Alumina Limited Australia and New Zealand Banking Group Ltd Kiwi Income Property Trust Newcrest Mining Limited Rubicon Limited Commonwealth Bank of Australia Goodman Fielder Limited Turners & Growers Limited GPT Group Lyttelton Port Company Limited Qantas Airways Limited L & M Petroleum Limited Comvita Limited Aristocrat Leisure Limited Smiths City Group Limited Gunns Limited Lion Nathan Ltd Property for Industry Limited Boral Limited Methven Limited Paladin Energy Ltd Pacific Brands Limited Allied Farmers Limited Cue Energy Resources Limited Veritas Investments Limited Molopo Energy Limited Pan Pacific Petroleum NL Renaissance Corporation Limited Indophil Resources NL Kingfish Limited Kingsgate Consolidated Limited Telstra Corporation Ltd Dorchester Pacific Limited BHP Billiton Limited Tourism Holdings Limited Tabcorp Holdings Limited Westpac Banking Corporation Pumpkin Patch Limited QBE Insurance Group Limited Scott Technology Limited Suncorp Group Limited Panel B: New Zealand Companies Auckland International Airport Ltd Goodman Property Trust Australian Infrastructure Fund Infratil Limited Virgin Australia Holdings Limited Air New Zealand Ltd Ryman Healthcare Limited Coffey International Limited Tenon Limited Data3 Limited Fletcher Building Ltd Vector Limited Investa Office Fund The New Zealand Refining Company Limited Primary Health Care Limited Fisher & Paykel Appliances Holdings Ltd Skellerup Holdings Limited Flight Centre Limited Cavalier Corporation Limited WorleyParsons Limited Fisher & Paykel Healthcare Corporation Ltd Ebos Group Limited REA Group Ltd Briscoe Group Limited Blackmores Limited Heritage Gold NZ Ltd South Port New Zealand Limited (NS) Redflex Holdings Limited Seeka Kiwifruit Industries Limited Australand Property Group Nuplex Industries Ltd PGG Wrightson Limited Zimplats Holdings Limited Sanford Limited Discovery Metals Limited New Zealand Oil & Gas Ltd Restaurant Brands New Zealand Limited Amalgamated Holdings Limited Abano Healthcare Group Limited RuralAus Investments Limited Sky City Entertainment Group Ltd Port of Tauranga Limited (NS) GUD Holdings Limited Freightways Limited CPT Global Limited Telecom Corporation of New Zealand Ltd Contact Energy Limited Transurban Group TrustPower Limited Amcor Limited Tower Ltd Michael Hill International Limited Transfield Services Limited Steel & Tube Holdings Limited David Jones Limited The Warehouse Group Limited Hallenstein Glasson Holdings Limited Northern Star Resources Ltd Mainfreight Limited Allied Gold Mining PLC

116

CHAPTER 4

IMPACT OF ANONYMITY ON PRICE DISCOVERY: A

NATURAL EXPERIMENT

4.1 Abstract

Using Minspan synchronous transactions data of cross-listed stocks on ASX and

NZX, this chapter examines the change in price discovery on the two exchanges over the period of staggered introduction of anonymity. Both the Hasbrouck information share and the error correction model of Harris et al. (1995) show two interesting trends. Information share improves on ASX, but deteriorates on NZX after ASX switched to anonymous trading. On the other hand, information share increases on NZX, but decreases on ASX after NZX adopted anonymous trading.

These results provide further support to the conclusion in Chapter 3 that anonymity attracts the trading of cross-listed stocks from the foreign counterpart.

4.2 Introduction

This chapter makes use of the natural experiment created by the staggered move to an anonymity regime undertaken by the Australian Stock Exchange (ASX) on

28 November 2005 and the New Zealand Stock Exchange (NZX) on 6 July 2007 to examine the effects of anonymity on price discovery for cross-listed stocks.

117

O’Hara (2003) indicates that markets have two key functions: liquidity and price discovery. The first is to provide liquidity for buyers and sellers, and the latter involves the incorporation of new information into asset prices (Hasbrouck, 1995).

Surprisingly, prior literature on anonymity effects has exclusively focused on the liquidity aspect (see Chapter 3, e.g., Grammig et al., 2001; Foucault et al., 2007;

Comerton-Forde et al., 2005). Implications for price discovery have been largely missing from anonymity literature. To fill this research gap, this chapter investigates how price discovery changes over time on ASX and NZX as they moved from transparent to anonymous trading.

This chapter is distinguished by two main contributions. The first is a road map for the understanding of the role of anonymity in security trading. This road map forms the linkage between previous works on anonymity and price discovery, by observing directly the conduct of market traders, given the choice between transparent and anonymous markets. In fact, the concepts of anonymity and price discovery are closely related. Anonymity allows traders to conceal their trading intentions, in particular informed traders who do not want to be identified (Harris,

2003). Empirical evidence also suggests that anonymity attracts informed traders and enhances price competition because it provides informed traders with incentives to trade (e.g., Grammig et al., 2001; Barclay et al., 2003). On the other hand, price discovery involves the incorporation of new information into asset prices (Hasbrouck, 1995), and it requires the consideration of the role of the informed traders again. If anonymity is the preference for informed traders, anonymous markets could lead to improved informativeness of stock prices.

118

The second contribution is the unique setting of the natural experiments. The staggered move to an anonymity regime undertaken by ASX and NZX provides clean natural experiments for examining the impact of anonymous trading on the nature of the price discovery process for cross-listed stocks. Unlike previous studies which compare markets in different trading systems (e.g., electronic and dealer, see Barclay et al., 2003), both the Australian and New Zealand markets operate similar screen-based limit order book systems. This ensures the comparability of results between periods and exchanges. Moreover, this natural experiment enables the study of anonymity effects simultaneously in both ASX and NZX, during the period of staggered regulatory change.

By estimating Hasbrouck information share and error correction model of Harris et al. (1995), it is found that anonymity contributes significantly to the price discovery process in the trading of cross-listed stocks. ASX’s switch to anonymous trading increases the share of price discovery on ASX, but decreases it on NZX. Conversely, NZX’s adoption of anonymous trading leads price discovery to deteriorate on ASX, but improve on NZX. These results have important implications for market design, especially as there has been no uniform view on what structures offer the greater benefits.

The remainder of this chapter is organised as follows. The next section reviews the literature pertaining to anonymity and price discovery, and outlines several hypotheses tested in this study. Section 4.4 describes the relevant institutional details for ASX and NZX markets and presents the data. Section 4.5 sets out the research methodology by introducing Hasbrouck information share and Engle and

119

Granger’s (1987) error-correction model. Section 4.6 reports the results, and finally Section 4.7 concludes.

4.3 Literature Review and Hypotheses

Prior literature on anonymity effects has exclusively focused on the liquidity dimension, with mixed empirical evidence. One side of the argument shows that an anonymous market is associated with higher adverse selection risk, which may deter market liquidity (e.g., Barclay et al., 2003; Theissen, 2003). The other side of the argument shows that anonymous markets could improve liquidity and price competition, because anonymity makes informed traders more willing to trade aggressively (e.g., Foucault et al., 2007; Comerton-Forde et al., 2005).

Given the above contrasting conclusions, this chapter seeks to address the issue of anonymity effects by presenting new evidence from the price discovery aspect, which has not been examined in the existing literature. Though the concepts of anonymity and price discovery are not the same, they are closely related through the role of the informed traders. Anonymity enables traders to execute transactions in the market without displaying their identifiers, allowing them to conceal their trading intentions. As identified in the previous chapter, the preference for anonymity by informed traders is evident in many theoretical and empirical studies (e.g., O’Hara 1995; Forster and George, 1992; Grammig et al.,

2001; Heidle and Huang, 2002).

Market microstructure theory indicates that more information-based trading enhances price informativeness. In Grossman’s (1976) model of price formation,

120

security prices fully and immediately reveal information possessed by informed traders. Easley and O’Hara (2004) develop a rational expectations equilibrium model that explains this role of informed trading in equilibrium assets prices.

They indicate that informed traders’ information affects their demands, which is reflected in equilibrium prices, while uninformed traders make correct inferences about this private information from prices. As a result, the private information is transmitted from informed to rational uninformed traders, leading to increased price efficiency.

Empirical microstructure studies provide evidence that informed trades are the main cause of stock price changes. Barclay and Warner (1993) examine 105 tender-offer targets firms in NYSE, and find that informed trades concentrate on trades of medium-size, which account for 92.8% of the cumulative price-change during preannouncement periods. Barclay and Hendershott (2003) investigate the interaction between informed trading and price discovery. They find that after- hour trading generates significant price discovery on the anonymous electronic communication network (ECN), due to the higher frequency of informed trading in the post-close compared to market-maker trades. Barclay et al. (2003) study the competition between ECN and NASDAQ dealers. They find that ECN offers the advantages of anonymity and speed of execution, which attract informed traders.

ECN contributes to the majority of the aggregate price discovery, which explains from 60% to 100% more of the efficient price variance than market-maker trading.

The relation between price discovery and informed trading is further shown in more recent studies. Inci et al. (2010) examine the market microstructure effects

121

of informed trading based on NYSE, AMEX and NASDAQ stocks. They find that stock prices incorporate insiders’ information more quickly and more fully relative to similar trades from non-insiders. Eun and Sabherwal (2003) examine price discovery of Canadian stocks cross-listed on the US exchanges. They observe a positive relationship between price discovery and the proportion of information trades, and a negative relationship with the ratio of bid-ask spreads.

This is consistent with the conclusion of Foucault et al. (2007) that anonymous quotes lead to tighter bid-ask spreads. They argue that anonymous trading venues make traders more willing to trade aggressively, because market participants are generally unable to discriminate between informed and uninformed traders and to pick off uninformed orders or free-ride informed orders,.

With regard to price discovery in international markets, the majority of the studies have found that home markets remain dominant due to informational advantages.

Hasbrouck (1995) develops an econometric approach called the information share to measure various markets’ relative contribution to price discovery for NYSE securities that also traded in other regional markets. Based on a sample of 30 Dow stocks, Hasbrouck shows that price discovery is concentrated at NYSE with the median information share of 92.7%. Lok and Kalev (2006) study New Zealand and Australian cross-listings using an error-correction model. They find that each market contributes to price discovery, while prices in the foreign market error- correct mostly towards prices in the home market. However, they do not compute explicit measures for the relative contribution to price discovery by each market.

Frijns et al. (2010) also conduct a study on a sample of Australian and New

Zealand cross-listed stocks using Hasbrouck information share. They observe that

122

both markets contribute to price discovery but the home market is dominant. They further find the growing importance of the Australian market for both Australian and New Zealand domiciled firms.

Differing from prior studies, this chapter aims to examine the change in price discovery in the Australian and New Zealand markets after the market structure change from trader transparency to anonymity. If anonymity does attract informed trading (e.g., O’Hara, 1995; Grammig et al., 2001; Heidle and Huang, 2002;

Foucault et al., 2007), which could in turn cause stock prices to be more informative (Grossman, 1976; Easley and O’Hara, 2004; Inci et al., 2010; Eun and

Sabherwal, 2003), it is expected that an anonymous trading system would enhance an execution channel’s contribution to the price discovery process. That is, the change from a transparent market to an anonymous market by ASX would lead to an increased price discovery in ASX. This leads to the first hypothesis (H4.1).

Hypothesis4.1: ASX’s contribution to price discovery will be increasing, and

NZX’s contribution to price discovery will be decreasing, after ASX introduced anonymous trading.

On the other hand, the opposite findings can be anticipated after NZX switched to an anonymity regime. This leads to the second hypothesis (H4.2).

Hypothesis4.2: ASX’s contribution to price discovery will be deteriorating, and

NZX’s contribution to price discovery will be improving, after NZX introduced anonymous trading.

123

4.4 Data and Sampling

4.4.1 Data and Study Period

Both ASX and NZX operate analogous open electronic limit order books, with

ASX being relatively larger than NZX. Both markets are highly integrated and competitive, and market participants can easily trade in either market. As shown in Chapter 3, NZX trades from 10:00 to 16:45 (New Zealand Standard Time) and

ASX trades from 10:00 to 16:00 (Australian Eastern Standard Time), resulting in around five hours of overlapping operation between the two exchanges.

ASX removed broker identifiers on 28 November 2005, followed by NZX’s adoption of an anonymity regime on 6 July 2007. In order to analyse substantial change in price discovery before and after the staggered move to an anonymous market by ASX and NZX, this study breaks the analysis into three sub-periods.

“Transparent” is the label for the first period from 22 April 2004 to 27 November

2005, during which broker identifications were disclosed in the both markets;

Period 2 from 28 November 2005 to 5 July 2007 is labelled “transition”, during which anonymous trading took place only in the Australian market; Period 3 from

6 July 2007 to 10 February 2009 is the “anonymous” period, during which both markets became anonymous. Each sub-period contains approximately 400 trading days.

During the entire study period from 22 April 2004 to 10 February 2009, there were 32 stocks which were traded on both ASX and NZX, including 15 New

Zealand-incorporated companies and 17 Australian-incorporated companies.

124

Table 4-1 and Table 4-2 report these stocks with the total number of trades computed during overlapping trading hours in each sub-period.

To be included in the final sample, stock data needs to be available throughout the entire study period. Since accurate estimation of price discovery can only be achieved if there is sufficient transaction data, the sample is restricted to liquid stocks traded on both ASX and NZX. A stock is required to be traded at least five times per day on average during each of the three sub-periods. Stocks are eliminated if their total number of trades executed is less than 2000 (5*400 trading days). The final sample stocks include eight New Zealand companies

(AIA, AIR, FBU, FPA, FPH, SKC, TEL and TWR), and three Australian companies (AMP, TLS and WBC).

For this final sample of 11 cross-listed stocks, price discovery analysis is performed using regular transaction prices. The trade data is from the Reuters

DataScope Tick History Database, provided by the Securities Industry Research

Centre of Asia Pacific (SIRCA). The data is stamped to the nearest millisecond and contains fields identifying the security code, date, time, price and volume.

125

Table 4-1: Trading Frequency of Cross-listed New Zealand Companies This table reports summary trading statistics for 15 New Zealand companies cross-listed on ASX as at 10 February 2009. “ASX (NZX) Trades” reports the total number of trades on ASX and NZX respectively. They are calculated during overlapping hours. “Final Sample” is “Yes” if the company is ultimately included in this study, otherwise it is “No”.

Company Name Transparent Transition Anonymous Final Sample

NZX Code ASX Trades NZX Trades ASX Trades NZX Trades ASX Trades NZX Trades

Auckland International Airport Ltd AIA 2,205 20,709 2,929 21,299 3,502 26,813 Yes

Air New Zealand Ltd AIR 3,498 6,626 3,273 8,350 3,224 11,858 Yes

Fletcher Building Ltd FBU 5,520 20,530 10,681 22,031 39,856 44,981 Yes

Fisher & Paykel Appliances Holdings Ltd FPA 4,110 16,359 2,678 14,404 3,706 14,693 Yes

Fisher & Paykel Healthcare Corporation Ltd FPH 2,362 15,419 2,565 19,945 6,530 25,441 Yes

Gensis Research and Development Corporation Ltd GEN 67 548 135 381 43 137 No

Heritage Gold NZ Ltd HGD 258 279 1,531 709 846 588 No

Nuplex Industries Ltd NPX 544 7,363 455 6,199 394 5,852 No

New Zealand Oil & Gas Ltd NZO 1,185 5,510 772 5,097 5,194 13,070 No

Pike River Coal Ltd PRC 0 0 0 0 6,196 10,293 No

Sky City Entertainment Group Ltd SKC 3,797 21,095 6,715 19,680 9,091 26,686 Yes

Sky Network Television Ltd SKT 13 2,162 103 9,634 661 10,085 No

Telecom Corporation of New Zealand Ltd TEL 47,817 42,045 101,744 55,167 198,009 60,201 Yes

Tower Ltd TWR 34,107 13,367 55,104 14,274 19,529 8,044 Yes

Warehouse Group Ltd WHS 1,789 13,395 1,632 12,723 1,038 9,995 No

126

Table 4-2: Trading Frequency of Cross-listed Australian Companies This table reports summary trading statistics for 17 Australian companies cross-listed on NZX as at 10 February 2009. “ASX (NZX) Trades” reports the total number of trades on ASX and NZX respectively. They are calculated during overlapping hours. “Final Sample” is “Yes” if the company is ultimately included in this study, otherwise it is “No”.

Company Name Transparent Transition Anonymous Final Sample

ASX Code ASX Trades NZX Trades ASX Trades NZX Trades ASX Trades NZX Trades

Australian Foundation Investment Co. Ltd AFI 25,489 916 34,110 1,297 51,129 1,128 No

AMP Ltd AMP 371,098 4,300 474,932 4,302 877,326 2,935 Yes

Australia and New Zealand Banking Group Ltd ANZ 418,604 1,773 763,298 1,923 2,167,272 3,126 No

APN News & Media Ltd APN 49,941 247 92,029 296 245,433 283 No

Babcock & Brown Infrastructure BBI 46,349 0 118,081 1 347,896 0 No

Downer EDI Limited Dow 64,427 4 202,729 1 467,631 2 No

Energy World Corporation Limited EWC 1,769 0 22,516 4 70,861 3 No

Goodman Fielder Limited GFF 0 0 108,171 3,499 371,366 2,199 No

L & M Petroleum Limited LMP 0 0 1,414 611 2,561 922 No

Lion Nathan Ltd LNN 72,487 1,985 146,632 1,274 449,650 1,286 No

Pacific Brands Limited PBG 81,342 1,025 127,287 416 308,923 138 No

People Telecom Limited PEO 2,466 5 1,283 4 504 0 No

Pan Pacific Petroleum NL PPP 2,959 1,025 5,147 1,058 14,602 2,189 No

Tag Pacific Limited TAG 265 131 476 116 146 49 No

Telstra Corporation Ltd TLS 538,557 3,490 589,686 2,619 948,813 1,582 Yes

Transpacific Industries Group Ltd TPI 11,635 0 109,060 5 338,087 0 No

Westpac Banking Corporation WBC 380,009 2,269 707,676 4,054 2,086,801 3,950 Yes

127

Table 4-3 provides a summary of daily trading statistics for each of these 11 cross-listed stocks over the entire study period from 22 April 2004 to 10 February

2009. The trading of cross-listed stocks is more active in the home market. Panel

A shows that the trading in New Zealand companies is generally higher on NZX, except for the stocks of TEL and TWR, which are more actively traded on ASX.

On the other hand, Panel B shows that the trading activity of Australian companies is substantially higher on ASX. The average daily trades on ASX are hundreds of times higher than on NZX.

Table 4-3: Trading Characteristics of Final Sample Stocks This table presents the average market capitalisation, daily price, number of shares traded, and number of trades during the entire study period from 22 April 2004 to 10 February 2009. Stock prices and market capitalisation are expressed in their respective currencies.

Code Market Cap ($m) Price ($) Shares Traded Trades

Panel A: New Zealand Companies

AIA NZX 2,669.37 3.30 993,702 57 ASX 2.81 198,814 8

AIR NZX 1,582.61 1.47 454,837 22 ASX 1.32 88,640 9

FBU NZX 3,968.56 8.22 606,442 73 ASX 7.23 146,651 47

FPA NZX 936.10 3.40 386,635 38 ASX 3.05 109,630 9

FPH NZX 1,715.15 4.39 556,274 50 ASX 3.84 109,362 11

SKC NZX 2,079.84 4.59 692,087 56 ASX 4.05 189,747 17

TEL NZX 9,348.98 4.82 3,871,634 131 ASX 4.27 2,605,036 288

TWR NZX 643.93 2.20 292,084 30 ASX 1.94 391,371 92

Panel B: Australian Companies

AMP NZX 9.12 13,114 10 ASX 15,055.27 7.99 3,817,133 1,418

TLS NZX 5.02 32,082 7 ASX 54,896.32 4.41 21,911,303 1,710

WBC NZX 26.87 22,316 11 ASX 41,870.29 22.13 3,641,995 2,617

128

These 11 cross-listed stocks are traded in two different currencies, in NZD in the

New Zealand market, and in AUD in the Australian market. The law of price

(Lamont and Thaler, 2003) requires the conversion of a single currency. The AUD price series is thus converted to New Zealand dollars using the prevailing exchange rate7. The intraday NZD/AUD exchange rate is sourced from SIRCA. It is formed by selecting the last price at one-minute intervals, and adjusted for differences in time zones and overlapping time. Over the study period, the exchange rate is fairly stable, ranging from 1.09 to 1.18 with an average standard deviation of 0.037.

Figure 4-1 shows a representative section of the price behaviour of AIA

(Auckland International Airport) in both markets (converted to NZD). Clearly, the two prices track each other closely and do not depart too far from each other. This provides a preliminary indication that prices are possibly cointegrated.

7 Analysis is also conducted when prices are converted to Australian dollars, with no qualitative change in results. These results are available on request.

129

Figure 4-1: Price Plot for AIA

4 3.5 3 2.5 2 1.5 1 NZX ASX 0.5

06-Jul-07 06-Jul-08

06-Jan-08 06-Jan-09

06-Jun-08

06-Oct-07 06-Oct-08

06-Apr-08

06-Sep-07 06-Feb-08 06-Sep-08 06-Feb-09

06-Dec-07 06-Dec-08

06-Aug-07 06-Aug-08

06-Nov-08 06-Nov-07

06-Mar-08 06-May-08

4.4.2 Pairing ASX and NZX Trades

To investigate price discovery, it is necessary to pair ASX trades with NZX trades, or NZX trades with ASX trades. The Minspan procedure of Harris et al. (1995) is used to form synchronous pairs of data. The Minspan procedure starts with a slow frequency channel (less liquid market), and looks forward and back to find the minimum time span within trade tuples. More specifically, it begins with the first observation of a trade on one illiquid exchange (trade 1 exchange 1), then selects the trade from the other exchange (trade 1 exchange 2) that occurs closest in time to the trade from the first exchange. That pair is then saved, and a new matched pair is formed in the same manner. For instance, pairing the price of Australian stock TLS starts with a trade on NZX (slow frequency channel), and then the most recent trade from ASX.

The use of Minspan sampling minimises the observation time span within the pair by looking forward and back in trading time. Each observation therefore reflects

130

new prices in both markets. The frequency of Minspan tuples is thus capped by the frequency of trading in the most illiquid channel. This procedure effectively avoids the use of stale price introduced by other sampling methods, such as

Fillforward, which fill forward the observations in the illiquid channel to match the frequency of the most liquid channel (Hasbrouck, 1995). As an illustration, the frequency of tuples for stock TLS is capped by 2,077,056 observations (538557 +

589686 + 948813, see Table 4-2) when using Fillforwd sampling, while it is only

7639 (3490+2619+1582) observations using Minspan sampling. Clearly, Minspan sampling reduces a huge amount of stale fictitious prices into the less liquid NZX channel and facilitates the true adjustment dynamics.

Table 4-4 presents Minspan statistics for the resulting 11 sample stocks over the entire study period. The average number of Minspan matched pairs is 30,179, with TEL having the largest number of observations (153,787) and TLS the smallest (7,639). Within each pair, the mean time between the two trades is 3.8 minutes.

Table 4-4: Pairing ASX and NZX Trades This table presents Minspan statistics for the 11 cross-listed stocks over the entire study from 22 April 2004 to 10 February 2009.

Number of Mean Span Standard Maximum Span Sample Observations (minute) Deviation (minute) AIA 8,517 4.4 10.0 214 AIR 9,826 12.1 21.2 267 FBU 54,638 3.2 7.1 238 FPA 10,265 6.1 11.7 226 FPH 11,280 4.7 8.2 197 SKC 19,352 4.3 8.1 231 TEL 153,787 1.5 2.7 76 TWR 35,014 4.7 13.4 270 AMP 11,466 0.3 0.8 43 TLS 7,639 0.4 3.3 171 WBC 10,189 0.4 1.5 48 Average 30,179 3.8 8.0 180

131

4.4.3 Unit Roots and Cointegration

A prerequisite for conducting price discovery analysis is that the price series is cointegrated (Gujarati, 2007); it is therefore necessary to test whether the home market price and the cross-listed price are indeed cointegrated. Cointegration testing requires the individual time series to contain a unit root (Gujarati, 2007),

The augmented Dickey and Fuller (1981) unit root test is performed on price series of each stock to establish that the two price series are nonstationary and integrated of I(1). All price series are transformed by taking natural logarithms.

These results are reported in Appendices 4-1 and 4-2.

Subsequently, the reduced rank regression procedure of Johansen (1988) is used to test for possible cointegration using the following error-correction model

(ECM):

∑ 𝑖 𝑖 훼훽 𝜀 ( ) 𝑖

where is (2x1) vector of constants, ( ) is the 2x1 vector of the log prices ( is the price series in NZX and is the price series in ASX),

denotes the first difference operator (i.e., ), 𝑖 is (2x2) coefficient matrices measuring the short-run adjustment of the system to changes in , 훼 is the (2x1) vector of error correction coefficients measuring the speed of convergence to the long-run equilibrium relations, arbitrage implies that prices can never diverge without bounds and defines the cointegrating vector 훽 =(1 -1),

132

so that 훽 is a stationary process I(0), and 𝜀 is a 2x1 vector of stationary residuals.

The Johansen (1988) maximum eigenvalue (max) and trace statistics (trace) test the rank of , i.e., the number of cointegration relationships. In the case of two random walk price series, rank =1 indicates that there exists one cointegrating vector, then the number of cointegration relationships is equal to 1 (2-1=1).

The optimal lag length k is identified by the Akaike Information Criterion in the

ECM of equation (4.1). The AIC is minimised at a maximum lag length of two.

With the two ASX and NZX price series, there is at most one cointegrating vector

(rank=1). The null hypothesis that rank=0 is tested using maximum eigenvalue and trace tests. The critical values for these tests, at the 1% level obtained from

Enders (1995), are max=18.78 and trace=21.96. Table 4-5 presents the Johansen’s cointegration test for each of the 22 price series over the entire study period from

22 April 2004 to 10 February 20098. These results show that all of the series are cointegrated.

Table 4-5 also presents the estimated cointegrating vectors, i.e.,  from equation

(4.1) and the sum of their magnitudes (in brackets) in the fourth column.

According to Ding et al. (1999), the magnitudes of the cointegrating vectors suggest the availability of arbitrage profit opportunities in the trading of cross-

listed stocks. As shown from ECM (4.1), if 훽 =(1 -1), then the scalar 훽

8 Similar results are obtained for all sub-periods.

133

훽 . Examining the quantity of 훽 훽 in Table 4-5 for each price series, shows that the price of a security does not deviate far from the equilibrium and profitable arbitrage trading is largely not present in the long-run.

Table 4-5: Cointegration Tests This table presents the Johansen’s cointegration test for each of the 22 price series over the entire study period from 22 April 2004 to 10 February 2009. The null hypothesis that rank=0 is tested using maximum eigenvalue and trace tests. The critical values for these tests, at the 1% level obtained from Enders (1995), are max=18.78 and trace=21.96. All prices are converted to NZD.

Johansen Test

Price Series Maximum Eigenvalue Trace Cointegrating Vectors

AIA 19.05*** 19.25** -1.133 1.134 (0.001) AIR 25.45*** 25.59*** -0.906 0.905 (-0.001) FBU 19.51*** 19.52** -0.593 0.595 (0.001) FPA 23.35*** 23.41*** -0.963 0.966 (0.003) FPH 22.63*** 22.91*** -0.971 0.971 (0.000) SKC 24.54*** 24.55*** -0.760 0.757 (-0.003) TEL 29.86*** 29.86*** -0.332 0.332 (0.000) T WR 19.24*** 19.27** -0.710 0.713 (0.003) AMP 27.25*** 27.29*** -1.030 1.030 (0.000) TLS 23.61*** 23.81*** -1.404 1.414 (0.001) WBC 24.86*** 25.01*** -1.348 1.351 (0.003)

134

4.5 Research Methodology

Schreiber and Schwartz (1986) define price discovery as the process by which markets attempt to find equilibrium prices from new information. Two approaches are commonly used to measure this price discovery process: the error correction mechanism and the information share approach. The ECM was popularized by

Engle and Granger (1987), and first applied by Gonzalo and Granger (1995) and

Harris et al. (1995); the information share was developed by Hasbrouck (1995) using the estimation results from the ECM. This section discusses both approaches in the analysis of price discovery of cross-listed stocks on ASX and

NZX.

4.5.1 Hasbrouck Information Share

The key concept underlying price discovery metric is the decomposition of a price innovation into permanent and temporary components. The permanent effects are associated with information, while the transitory effects are related to market phenomena such as the price discreteness and inventory adjustments (Booth et al.,

2002). The Hasbrouck information share decomposes the innovation variance for cointegrated series based on a Cholesky factorization when a security is traded in multiple markets.

In a cointegrated system, Hasbrouck (1995) shows the following vector moving average (VMA) model driven from the ECM (4.1):

(퐿)𝜀 ( )

135

where 𝜀 is a zero-mean vector of serially uncorrelated disturbances with covariance matrix , and (L) is a polynomial in the lag operator. Following

Beveridge and Nelson’s (1981) random walk decomposition model, the price change can be written as:

( )𝜀 ( 퐿) (퐿)𝜀 ( )

where ( ) is the sum of all moving average coefficients ( ( )

), with ( )𝜀 being the long-run impact of an innovation on each of the prices.

Stationarity implies that all of the rows of ( ) are identical, and then this long- run impact is the same for all prices. (퐿) is a second matrix polynomial. By integrating equation (4.3), the price levels can be written as:

0 ( ) ∑ 𝜀 (퐿)𝜀 ( )

where P0 is a (2x1) vector of constant initial values (in the case of two

cointegrated price series). ( ) ∑ 𝜀 (for s=1,…,t) captures the common efficient price in the two markets. That is, any increment 𝜀 is the component of the price change that is permanently impounded into the security price and is

presumably due to new information. The third term, (퐿)𝜀 denotes the transitory portion of the price change, and is a zero-mean covariance stationary process.

Equation (4.4) thus defines the price series as the sum of an initial value, a common random walk term, and a stationary term. Letting denote the common row vector in ( ) , the information share as defined by Hasbrouck (1995) follows from the variance decomposition of the common factor innovations, i.e.,

136

( 𝜀 ) has to be decomposed. In other words, the information share of a market is the proportion of variance in the common factor that is attributable to innovations in that market. As Hasbrouck (1995) argues, if price innovations are not correlated across the two markets, the innovations’ variance covariance matrix is diagonal, then will consist of two terms, the first (second) represents the contribution to the common factor innovation from the first (second) market. The proportion of this for market j relative to the total variance is defined as market j’s information:

( 5)

th th where is the j element of , is the j diagonal element of .

However, if the price innovations are correlated across markets (i.e., the two error terms in 𝜀 are correlated), then will not be diagonal, and the information share cannot be clearly assigned. In this case, Hasbrouck (1995) uses the Cholesky factorization to define the information share of the jth market prices as:

( ) ( 6)

th where is the j element of the row matrix , and F is the lower triangular

Choleski factorization of (i.e., =FF’). The lower triangular factorization maximises the information share of the first market, and consequently minimises the information share of the second market. Therefore, the Hasbrouck information share needs to permute all possible orderings of 𝜀 , with the consequence that the

137

information share is obtained in a range (upper and lower bounds) and not a unique value for each market.

Since information share is estimated with permuting ordering of markets, and the upper and lower bounds depend on contemporaneous correlation of innovations,

Eun and Sabherwal (2003) point out that when these bounds diverge, the precision of this measure can be markedly reduced. Booth et al. (2002) and Huang (2002) find that the discrepancy between the upper and lower bounds is small when using extremely high frequency data, such as the one-second interval data in Hasbrouck

(1995). However, using intervals in the range of a few minutes, this ordering- caused discrepancy can be substantial. For example, the information share bounds found by Booth et al. (2002) for the upstairs and downstairs markets at the

Helsinki Stock Exchange diverge by about 80 percentage points.

Due to the caveats of the Hasbrouck (1995) methodology, the ECM is also employed as an additional analysis of price discovery, given that the intervals of the Minspan sample used in this study are in the range of a few minutes. Booth et al. (2002) point out that the ECM is a very flexible specification. It not only determines which market drives price discovery but also details the interaction between both markets for each stock.

4.5.2 The ECM and the Measure of Contribution to Price Discovery

The ECM is warranted by the Granger representation theorem. It describes the short-run dynamics in which each price series moves towards long-run

138

equilibrium. If two prices are cointegrated, their price changes should be represented by a vector error correction model of equation (4.1). The intuition behind the error-correction model is that a change in price is determined by the deviation from the long-term equilibrium relationship. In this study, cointegration between ASX and NZX prices of a stock provides evidence of a long-run relationship between the markets. The two prices are influenced by any deviation from equilibrium. If there is a price innovation in one of the markets, one or both prices have to respond to the magnitude of this movement.

In the case of price relationship between the trading of cross-listed stocks on ASX and NZX, prices should be the same across the two markets. As equation (4.1) shows, the long run equilibrium relationship is thus  , and the error term has the form: 𝑍 –  . The coefficients  of this error term are the key determinant of the price discovery process. They measure the price reaction to the deviation of the price difference between the two markets from zero. Let  (  ) , if (positive 𝑍 ), then  𝑍 should be negative, yielding reduced , while  𝑍 should be positive, yielding increased . In this illustration, if  is insignificant, that means does not respond to disequilibrium from the system. If ASX prices did not respond to deviations from NZX prices, and price on NZX responds to deviations from ASX prices ( will be significant), that would be evidence that price discovery is focused in ASX. The greater the coefficient, the more the particular market reacts to deviations from equilibrium, the smaller the contribution to the price discovery process.

139

From this perspective, Eun and Sabherwal (2003) indicate that the coefficients

 and  can be interpreted as the average adjustment of each series towards the other in order to restore the equality of the two prices. The proportion of the total adjustment that occurs at the one market can be considered as a measure of the relative contribution to price discovery that take place on another market. This measure was first proposed by Schwarz and Szakmary (1994), and subsequently used by Theissen (2002) and Eun and Sabherwal (2003).

Therefore, the share of price discovery that occurs in ASX or NZX market can be measured as the share of total adjustment, i.e., the sum of the two adjustment coefficients. In other words, for measuring the relative contribution to price discovery by ASX, the NZX adjustment coefficient is taken as the share of the summed coefficients:

| | ( 7) | |

Correspondingly, the ASX adjustment is then attributed relative to NZX market:

( ) | |

Hence, . A high value of the variable means the coefficient is high.

140

4.6 Results

Using the Hasbrouck information share and the error correction model of Harris et al. (1995), this section tests Hypothesis4.1 and Hypothesis4.2, to examine the effects of anonymity on the price discovery process performed by New Zealand and Australian markets.

4.6.1 Change in Hasbrouck Information Share

To have a brief overview of how price discovery is performed by ASX and NZX,

Table 4-6 reports midpoints of the Hasbrouck information share for the sample stocks over the entire study period from 22 April 2004 to 10 February 2009. It shows that the range of information share between the upper and lower bounds is quite wide when permuting ordering of markets. Perhaps these discrepancies are not surprising, given that the intervals of the Minspan sample are in the range of a few minutes. As expected, Table 4-6 shows the high correlations of NZX and

ASX price innovations, which makes the Hasbrouck method ineffective (Booth et al., 2002). Due to this presence of correlations in the price innovations across the two markets, the mean of the bounds for the information share can be used as a unique measure of the price discovery contribution (Baillie et al., 2002). These information share statistics reflect the average estimate.

141

Table 4-6: Hasbrouck Information Share for Cross-listed Stocks on ASX and NZX This table calculates the upper and lower bounds of the Hasbrouck information share for cross-listed stocks on ASX and NZX, as well as the correlation coefficient for ASX and NZX prices over the period from 22 April 2004 to 10 February 2009.

NZX (%) ASX (%) Correlation Coefficient for Upper Bound Lower Bound Midpoint Upper Bound Lower Bound Midpoint ASX and NZX Prices

Panel A: New Zealand Companies

AIA 99.70 10.20 54.95 89.80 0.30 45.05 0.93

AIR 86.50 25.40 55.95 74.60 13.50 44.05 0.82

FBU 96.10 60.20 78.15 39.80 3.90 21.85 0.48

FPA 98.70 61.40 80.05 38.60 1.30 19.95 0.54

FPH 100.00 16.70 58.35 83.30 0.00 41.65 0.91

SKC 96.90 60.10 78.50 39.90 3.10 21.50 0.49

TEL 65.70 29.30 47.50 70.70 34.30 52.50 0.36

TWR 90.10 48.70 69.40 51.30 9.90 30.60 0.46

Panel B: Australian Companies

AMP 13.90 2.00 7.95 98.00 86.10 92.05 0.66

TLS 12.40 16.00 14.20 84.00 87.60 85.80 0.54

WBC 16.90 5.90 11.40 90.50 83.10 86.80 0.70

142

Table 4-6 confirms that price discovery mainly takes place in the home market, which is in line with findings of previous studies (e.g., Frijns, et al., 2010; Lok and Kalev, 2006). In the trading of New Zealand companies, Panel A shows that the NZX market contributes most to price discovery, accounting roughly for over

65% of the information share, while ASX prices share the remainder on average.

In the trading of ASX companies, Panel B shows that ASX market leads the price discovery process with roughly 88% of the information share, while NZX contributes only about 12% of price discovery. The sole exception is Telecom

Corporation of New Zealand (TEL), for which 52.5% of the information share takes place in the foreign market ASX. This finding confirms the relationship between information share and trading activity (greater trading activity of TEL on

ASX can be seen from Tables 4-1 and 4-3) as suggested by Eun and Sabherwal

(2003).

To investigate how the price discovery process changes after the change to the broker anonymity regime, the information share during each sub-period is calculated in Table 4-7 (see Appendix 4-3 for the full results). This table reveals two interesting trends. First, when considering ASX’s adoption of anonymous trading, it can clearly be seen that the average information share decreases in NZX, and increases correspondingly in ASX. Nine out of 11 stocks record a deteriorated information share on NZX. The average information share on NZX decreases from 69.63% (during the transparent period) to 58.01% (during the transition period) in the trading of New Zealand domiciled companies, and from 18.15% to

11.87% in Australian domiciled companies. On average, the information share increases by 10.16% on ASX, and decreases correspondingly on NZX. To

143

examine whether these changes are significant, the one-tail sign test is conducted with the following null hypothesis:

H0: P (–) <= P (+) v.s. H1: P (–) > P (+), where P denotes probability, “ – ” represents the decrease in the information share on NZX, and “ + ” represents the increase in the information share on NZX during the transition period.

The testing statistic for this sign test is M=T–n/2, where T is the number of decreased information share, and n is the total number of information share. Panel

C reports the test statistics. The difference of 10.16% in the information share between the transparent and transition period is significant at the 5% significance level. These results reveal that the adoption of an anonymity regime by ASX has a significant and positive impact on the price discovery process, supporting

Hypothesis4.1.

Second, when examining the change in price discovery after NZX adopted anonymous trading, it can be seen that the average information share increases on

NZX, and decreases correspondingly on ASX in the anonymous period. Ten out of 11 stocks record an improved information share on NZX. The average information share on NZX increases from 58.01% (during the transition period) to

70.26% (during the anonymous period) in New Zealand domiciled companies, and from 11.87% to 24.48% in Australian domiciled companies. On average, the

NZX information share increases by 12.35%, and the ASX information share decreases correspondingly.

144

Table 4-7: Information Share over a Period of Staggered Anonymity Regime Change This table presents the midpoints of the information share over the sample period. “Transparent” is sub-period (1) from 22 April 2004 to 27 November 2005, during which broker identifications were disclosed in the both markets; “transition” is sub-period (2) from 28 November 2005 to 5 July 2007 during which anonymous trading took place only in the Australian market; “anonymous” is sub-period (3) from 6 July 2007 to 10 February 2009, during which both markets became anonymous. The difference in information share between the three sub-periods is also calculated. The one-tail sign test is conducted to examine whether these changes are significant. The null is: H0: P (–) <= P (+) v.s. H1: P (– ) > P (+), where P denotes probability, “ – ” represents the decreased information share on NZX, and “ + ” represents the increased information share on NZX during the transition period. The testing statistic for the sign test is M=T–n/2, where T is the number of decreased information share, and n is the total number of information share.

NZX (%) ASX (%) Change on NZX (%) Change on ASX (%) Transparent (1) Transition (2) Anonymous (3) Transparent (1) Transition (2) Anonymous (3) (2) – (1) (3) – (2) (2) – (1) (3) – (2) Panel A: New Zealand Companies AIA 52.55 67.85 72.40 47.45 32.15 27.60 15.30 4.55 -15.30 -4.55 AIR 50.45 22.45 69.40 49.55 77.55 30.60 -28.00 46.95 28.00 -46.95 FBU 83.30 79.70 76.90 16.70 20.30 23.10 -3.60 -2.80 3.60 2.80 FPA 82.95 62.85 84.70 17.05 37.15 15.30 -20.10 21.85 20.10 -21.85 FPH 54.45 74.05 79.45 45.55 25.95 20.55 19.60 5.40 -19.60 -5.40 SKC 88.50 71.35 75.95 11.50 28.65 24.05 -17.15 4.60 17.15 -4.60 TEL 57.65 43.10 46.65 42.35 56.90 53.35 -14.55 3.55 14.55 -3.55 TWR 87.20 42.75 56.65 12.80 57.25 43.35 -44.45 13.90 44.45 -13.90 AVERAGE 69.63 58.01 70.26 30.37 41.99 29.74 -11.62 12.25 11.62 -12.25

Panel B: Australian Companies AMP 15.50 8.60 33.15 84.50 91.40 66.85 -6.90 24.55 6.90 -24. 55 TLS 14.60 13.70 18.95 85.40 86.30 81.05 -0.90 5.25 0.90 -5.25 WBC 24.35 13.30 21.35 75.65 86.70 78.65 -11.05 8.05 11.05 -8.05 AVERAGE 18.15 11.87 24.48 81.85 88.13 75.52 -6.28 12.62 6.28 -12.62

Panel C: Sign test Average change -10.16 12.35 10.16 -12.35 Testing statistics (-3.5)** (4.5)*** (3.5)** (-4.5)*** p-value 0.0327 0.0059 0.0327 0.0059

145

The sign test reported in panel C indicates that this change is statistically significant at the 1% significance level. These results show again the positive impact of anonymity on the price discovery process. Consistent with

Hypothesis4.2, NZX’s introduction of an anonymity regime leads to improved information share on NZX and deteriorated information share on ASX.

These trends can be better observed in Figure 4-2, which plots the average midpoint of information share on NZX and ASX every five months for New

Zealand and Australian domiciled stocks respectively. Clearly, these downward/upward trends demonstrate the contribution of anonymity to the price discovery process.

146

Figure 4-2: Hasbrouck Information Share over Time

% Panel A: NZX domiciled stocks 100

55 NZX ASX 40

% Panel B: ASX domiciled stocks 100 85 70 55 NZX ASX 40 25 10

4.6.2 Change in Information Share from ECM

Table 4-8 presents the main results regarding the estimates of 훼 and 훼 of the

ECM equation (4.1). As the model predicted, the error coefficients 훼 and 훼 carry opposite signs, indicating that error-correcting adjustments occur in either one market or both markets to maintain the long-run equilibrium between the two price series following a price shock.

147

Table 4-8: ECM Estimation Results

This table reports the estimates of nzx and asx from the error correction model (4.1) over the sample period. “Transparent” is sub-period (1) from 22 April 2004 to 27 November 2005, during which broker identifications were disclosed in the both markets; “transition” is sub-period (2) from 28 November 2005 to 5 July 2007 during which anonymous trading took place only in the Australian market; “anonymous” is sub-period (3) from 6 July 2007 to 10 February 2009, during which both markets became anonymous. The interpretation for nzx is the average adjustment of NZX price towards ASX price; the interpretation for asx is the average adjustment of the ASX to NZX price. *, ** and *** denote significance levels of 10%, 5% and 1% respectively.

NZX ASX Transparent (1) Transition (2) Anonymous (3) Transparent (1) Transition (2) Anonymous (3) Panel A: New Zealand Companies AIA a -0.02 -0.06 -0.03 0.09 0.24 0.22 t-value (-0.4) (-3.9)*** (-2.3)** (3.6)*** (12.3)*** (12.8)*** AIR a -0.08 -0.40 -0.06 0.17 0.19 0.15 t-value (-1.6) (-15.7)*** (-5.0) (3.5)*** (11.7)*** (9.6)*** FBU a -0.02 -0.02 -0.01 0.19 0.10 0.06 t-value (-2.9)*** (-4.1)*** (-6.6)*** (16.3)*** (16.3)*** (23.2)*** FPA a -0.03 -0.07 0.00 0.23 0.18 0.19 t-value (-2.5)** (-4.3)*** (-0.1) (15.3)*** (8.7)*** (13.2)*** FPH a -0.07 -0.03 -0.02 0.54 0.19 0.14 t-value (-0.7) (-2.1)** (-2.8)*** (5.7)*** (9.5)*** (14.9)*** SKC a -0.03 -0.03 -0.02 0.36 0.09 0.09 t-value (-2.8)*** (-5.3)*** (-3.6)*** (18.5)*** (11.6)*** (12.9)*** TEL a -0.01 -0.01 -0.02 0.01 0.01 0.02 t-value (-8.2)*** (-11.9)*** (-15.1)*** (11.3)*** (9.9)*** (13.5)*** TWR a -0.06 -0.14 -0.06 0.31 0.09 0.09 t-value (-13.7)*** (-15.7)*** (-9.6)*** (35.9)*** (10.3)*** (12.3)*** AVERAGE a -0.04 -0.10 -0.03 0.24 0.14 0.12

Panel B: Australian Companies AMP a -0.12 -0.11 -0.17 0.00 0.01 0.05 t-value (-12.2)*** (-11.9)*** (-8.7)*** (0.3) (1.8)* (2.4)** TLS a -0.11 -0.22 -0.52 0.01 0.02 0.03 t-value (-12.1)*** (-13.5)*** (-15.3)*** (1.1) (1.5) (0.9) WBC a -0.19 -0.19 -0.27 0.01 0.00 0.04 t-value (-10.8)*** (-14.1)*** (-14.2)*** (0.4) (0.3) (2.5)** AVERAGE a -0.14 -0.17 -0.32 0.01 0.01 0.04

148

Consistent with Table 4-6, price discovery occurs mainly in the home market, as greater error-correction occurs in the foreign market. Panel A shows that on average, the New Zealand domiciled stocks have greater ASX adjustment

coefficients 훼 , which are approximately three times larger than the NZX adjustment , On the other hand, Panel B shows that Australian domiciled stocks have high values of NZX adjustment , which are approximately 10 times greater than the ASX adjustment.

When considering the impact of ASX’s anonymous trading on the price discovery process, Table 4-8 further shows that the information role of ASX became more significant after its adoption of trader anonymity, in particular for the stocks AIA,

AIR and FPH. The NZX adjustments of these stocks are almost zero and statistically insignificant in the transparent period, indicating the lack of price influence of ASX on NZX. However, they become statistically significant at least at the 5% level during the transition period. ASX contributes its information role to the price discovery process in the trading of these three stocks after the adoption of an anonymous market. Interestingly, when considering the impact of

NZX’s anonymous trading, the information role of NZX became significant after its adoption of anonymity in the trading of stock AMP and WBC. The ASX adjustments 훼 of these stocks became statistically significant at the 5% significance level in the anonymous period, indicating enhanced information role by NZX in the price discovery process due to the anonymity regime.

149

After estimating the coefficients 훼 and 훼 above in the error correction model, Table 4-9 computes the explicit measure for the relative contribution to price discovery of each market using equations (4.7) and (4.8). Consistent with

Table 4-7, it shows that on average the share of price discovery is higher in the home market. After the introduction of anonymous trading by ASX, price discovery improves on ASX, and decreases correspondingly on NZX. In the trading of New Zealand companies, the average information share of New

Zealand companies on ASX is only 17.75% during the transparent period, indicating a marginal information role of ASX prior to anonymity regime change.

After commencing anonymous trading, all New Zealand stocks record improved price discovery in ASX. The average rises to 35.98%, while the average

reduces to 64.02% (which was 82.25% prior to the change).

Panel B shows that the trading of TLS and WBC on ASX has also improved information share during the transition period. On average, ASX anonymity results in the NZX information share decreasing and ASX information share increasing by 12.76% correspondingly. The one-tail sign test shows that these changes are significant at the 1% significance level. These findings are consistent with the previous results, providing support to Hypothesis4.1. The share of price discovery increases in ASX, and decreases correspondingly in NZX, after ASX introduced anonymous trading.

150

Table 4-9: Information Share from ECM This table presents the share of price discovery of the ASX and NZX respectively. For each stock, it calculates the share of price discovery in ASX based on equation (4.7), and in NZX based on equation (4.8), where the error correction coefficients are obtained from the error correction model (4.1). “Transparent” is sub-period (1) from 22 April 2004 to 27 November 2005, during which broker identifications were disclosed in the both markets; “transition” is sub-period (2) from 28 November 2005 to 5 July 2007 during which anonymous trading took place only in the Australian market; “anonymous” is sub-period (3) from 6 July 2007 to 10 February 2009, during which both markets became anonymous. The difference in the share of price discovery between the three sub-periods is also calculated. The one-tail sign test is conducted to examine whether these changes are significant. The null is: H0: P (–) <= P (+) v.s. H1: P (–) > P (+), where P denote probability, “ – ” represents the decreased information share on NZX, and “ + ” represents the increased information share on NZX during the transition period. The testing statistic for the sign test is M=T–n/2, where T is the number of decreased information share, and n is the total number of information share.

NZX (%) ASX (%) Change on NZX (%) Change on ASX (%) Transparent (1) Transition (2) Anonymous (3) Transparent (1) Transition (2) Anonymous (3) (2) – (1) (3) – (2) (2) – (1) (3) – (2) Panel A: New Zealand Companies AIA 80.81 78.86 87.38 19.1 9 21.14 12.62 -1.95 8.52 1.95 -8.52 AIR 68.31 32.30 71.37 31.69 67.70 28.63 -36.01 39.07 36.01 -39.07 FBU 90.45 85.17 84.29 9.55 14.83 15.71 -5.28 -0.88 5.28 0.88 FPA 90.09 71.38 99.26 9.91 28.62 0.74 -18.71 27.88 18.71 -27.88 FPH 89.08 85.01 88.83 10.92 14.99 11.17 -4.07 3.81 4.07 -3.81 SKC 93.30 76.45 84.63 6.70 23.55 15.37 -16.85 8.18 16.85 -8.18 TEL 62.01 43.81 50.10 37.99 56.19 49.90 -18.21 6.29 18.21 -6.29 TWR 83.93 39.16 59.47 16.07 60.84 40.53 -44.76 20.30 44.76 -20.30 AVERAGE 82.25 64.02 78.16 17.75 35.98 21.84 -18.23 14.15 18.23 -14.15

Panel B: Australian Companies AMP 2.66 10.39 21.19 97.34 89.61 78.81 7.72 10.80 -7.72 -10.80 TLS 8.03 7.89 4.85 91.97 92.11 95.15 -0.14 -3.04 0.14 3.04 WBC 3.88 1.79 13.73 96.12 98.21 86.27 -2.09 11.94 2.09 -11.94 AVERAGE 4.86 6.69 13.26 95.14 93.31 86.74 1.83 6.57 -1.83 -6.57

Panel C: Sign test Average change -12.76 12.08 12.76 -12.08 Testing statistics (-4.5)*** (3.5)** (4.5)*** (-3.5)** p-value 0.0059 0.0327 0.0059 0.0327

151

Table 4-9 also computes the change in information share between the transition and anonymous periods. The results show that after the introduction of anonymity on NZX, nine out of 11 stocks record improved price discovery on NZX. The average of New Zealand-incorporated stocks bounces up to 78.16%, while

drops to 21.84% in the anonymous period. NZX also increases its share of price discovery in the trading of ASX domiciled stocks. The average information share in the anonymous period (13.26%) is about three times greater than in the transition period (4.86%). Though NZX’s share of price discovery had decreased

12.76% after ASX introduced anonymity, it recovered 12.08% after NZX’s adoption of anonymity. These results provide consistent evidence that anonymity improves price discovery, supporting Hypothesis4.2.

4.7 Summary

The impact of an anonymous trading structure on the limit order book is the subject of considerable research. A main focus of this literature is the investigation of liquidity impact, while implications for price discovery have not been previously examined. This chapter bridges this gap in the literature by examining anonymity effects on price discovery for cross-listed stocks on ASX and NZX.

The estimation of Hasbrouck information share and the error correction mechanism shows two important trends regarding the change in price discovery.

The share of price discovery increases on ASX and decreases correspondingly on

152

NZX, after ASX switched to anonymous trading. On the other hand, the share of price discovery bounces up on NZX with a corresponding decrease on ASX, after

NZX switched to anonymous trading. These downward/upward trends in information share demonstrate clearly that anonymous trading enhances the price discovery process. These results are consistent with the findings in Chapter 3, which shows that an anonymous market attracts the trading of cross-listed stocks from the foreign counterpart.

These results also provide support to the prior literature on the choice of anonymous markets by informed traders (Grammig et al. 2001; Heidle and Huang,

2002; Foucault et al., 2007), while more information-based trading in turn enhances price informativeness (e.g., Easley and O’Hara, 2004; Inci et al., 2010;

Eun and Sabherwal, 2003).

153

References

Baillie, R., Booth, G., Tse, Y., and Zabotina, T., 2002. Price discovery and common factor models. Journal of Futures Markets, 5(3), pp.309-321.

Barclay, M.J., and Hendershott, T., 2003. Price Discovery and Trading after

Hours. The Review of Financial Studies, 16(4), pp.1041-1073.

Barclay, M.J., Hendershott, T., and McCormick, D.T., 2003. Competition among

Trading Venues: Information and Trading on Electronic Communications

Networks. The Journal of Finance, 58(6), pp.2637-2665.

Barclay, M.J., and Warner, J.B., 1993. Stealth trading and volatility : Which trades move prices? Journal of Financial Economics, 34(3), pp.281-305.

Beveridge, S., and Nelson, C.R., 1981. A new approach to decomposition of economic time series into permanent and transitory components with particular attention to measurement of the “business cycle”. Journal of Monetary

Economics, 7(2), pp.151-174.

Booth, G.G., Lin, J.C., and Tse, Y., 2002. Trading and Pricing in Upstairs and

Downstairs Stock Markets. The Review of Financial Studies, 15(4), pp.1111-1135.

Comerton-Forde, C., Frino, A., and Mollica, V., 2005. The impact of limit order anonymity on liquidity: Evidence from Paris, Tokyo and Korea. Journal of

Economics and Business, 57(6), pp.528-540.

154

Dickey, D.A., and Fuller, W.A., 1981. Likelihood Ratio Statistics for

Autoregressive Time Series with a Unit Root. Econometrica, 49(4), pp.1057-

1072.

Ding, D.K., Harris, F.H.deB., and Lau, S.T., 1999. An investigation of price discovery in informationally-linked markets: equity trading in Malaysia and

Singapore. Journal of Multinational Financial Management, 9(3-4), pp.317-329.

Easley, D., and O’Hara, M., 2004. Information and the Cost of Capital. The

Journal of Finance, 59(4), pp.1553-1583.

Enders, W., 1995. Applied Econometric Time Series, Wiley, New York.

Engle, R.F., and Granger, C.W.J., 1987. Co-Integration and Error Correction:

Representation, Estimation, and Testing. Econometrica, 55(2), pp.251-276.

Eun, C.S., and Sabherwal, S., 2003. Cross-Border Listings and Price Discovery:

Evidence from U.S.-Listed Canadian Stocks. The Journal of Finance, 58(2), pp.549-575.

Foucault, T., Moinas, S., and Theissen, E., 2007. Does Anonymity Matter in

Electronic Limit Order Markets? Review of Financial Studies, 20(5), pp.1707-

1747.

Frijns, B., Gilbert, A., and Tourani-Rad, A., 2010. The dynamics of price discovery for cross-listed shares: Evidence from Australia and New Zealand.

Journal of Banking & Finance, 34(3), pp.498-508.

155

Forster, M.M., and George, T.J., 1992. Anonymity in securities markets. Journal of Financial Intermediation, 2(2), pp.168-206.

Grammig, J., Schiereck, D., and Theissen, E., 2001. Knowing me, knowing you:

Trader anonymity and informed trading in parallel markets. Journal of Financial

Markets, 4(4), pp.385-412.

Gonzalo, J., and Granger, C., 1995. Estimation of common long-memory components in cointegrated. Journal of Business & Economic Statistics, 13(1), pp.27-35.

Grossman, S., 1976. On the Efficiency of Competitive Stock Markets Where

Trades Have Diverse Information. The Journal of Finance, 31(2), pp.573-585.

Gujarati, D.N., 2007. Basic Econometrics, McGraw, Australia.

Harris, F.H., McInish, T.H., Shoesmith, G.L., and Wood, R.A., 1995.

Cointegration, Error Correction, and Price Discovery on Informationally Linked

Security Markets. The Journal of Financial and Quantitative Analysis, 30(4), pp.563-579.

Harris, L. 2003. Trading and exchanges: market microstructure for practitioners,

Oxford, New York.

Hasbrouck, J., 1995. One Security, Many Markets: Determining the Contributions to Price Discovery. The Journal of Finance, 50(4), pp.1175-1199.

156

Heidle, H.G., and Huang, R.D., 2002. Information-Based Trading in Dealer and

Auction Markets: An Analysis of Exchange Listings. The Journal of Financial and Quantitative Analysis, 37(3), pp.391-424.

Huang, R.D., 2002. The Quality of ECN and Nasdaq Market Maker Quotes. The

Journal of Finance, 57(3), pp.1285-1319.

Inci, A.C., Lu, B., and Seyhun, H.N., 2010. Intraday Behaviour of Stock Prices and Trades around Insider Trading. Financial Management, 39(1), pp.323-363.

Johansen, S., 1988. Statistical analysis of cointegration vectors. Journal of

Economic Dynamics and Control, 12(2-3), pp.231-254.

Lamont, O.A., and Thaler, R.H., 2003. Anomalies: The Law of One Price in

Financial Markets. The Journal of Economic Perspectives, 17(4), pp.191-202.

Lok, E., and Kalev, P.S., 2006. The intraday price behaviour of Australian and

New Zealand cross-listed stocks. International Review of Financial Analysis,

15(4-5), pp.377-397.

O’Hara, M., 1995. Market Microstructure Theory, Blackwell Publishing,

Cambridge.

O’Hara, M., 2003. Presidential Address: Liquidity and Price Discovery. The

Journal of Finance, 58(4), pp.1335-1354.

157

Schreiber, P.S., and Schwartz, R.A. 1986. Price discovery in securities markets.

The Journal of Portfolio Management, 12(4), pp.43-48.

Schwarz, T.V., and Szakmary, A.C., 1994. Price discovery in petroleum markets:

Arbitrage, cointegration, and the time interval analysis. Journal of Futures

Markets, 14(2), pp.147-167.

Theissen, E., 2002. Price discovery in floor and screen trading systems. Journal of

Empirical Finance, 9(4), pp.455-474.

Theissen, E., 2003. Trader Anonymity, Price Formation and Liquidity. European

Finance Review, 7(1), pp.1-26.

158

Appendices

Appendix 4-1: Unit Root Test for New Zealand Stocks

An augmented Dickey-Fuller (1981) test for unit roots is performed over the full study period9. Three equations are used: ∑𝑖 𝑖 𝑖 , 0 ∑𝑖 𝑖 𝑖 ,

0 ∑𝑖 𝑖 𝑖 where is each price series, is the differenced price variable, t is the trend term, and k is the lag length determined by Akaike and Schwarz Bayes Information Criteria. NC, regression without constant and time trend; C, regression containing a constant; CT, regression containing a constant and a liner time trend; * indicates that the hypothesis of a unit root is rejected at significance levels at or greater than 10%.

Stock Code NC t-test C t-test CT t-test         Panel A: Unit root test in levels

AIA NZX 0.00 0.3 0.00 0.6 0.00 0.6

ASX 0.00 -1.8* 0.00 -1.9 0.00 -2.0

AIR NZX 0.00 -0.3 0.00 -0.4 0.00 0.0

ASX 0.00 -1.1 0.00 -0.6 0.00 -0.6

FBU NZX 0.00 -0.1 0.00 0.5 0.00 0.5

ASX 0.00 0.0 0.00 -1.0 0.00 -1.0

FPA NZX 0.00 -0.6 0.00 3.6 * 0.00 3.6 *

ASX 0.00 -1.2 0.00 -0.3 0.00 -0.2

FPH NZX 0.00 -2.6* 0.00 -0.2 0.00 -0.1

ASX 0.00 -0.4 0.00 0.7 0.00 0.7

SKC NZX 0.00 1.3 0.00 -0.5 0.00 -0.5

ASX 0.00 -1.3 0.00 0.5 0.00 0.5

TEL NZX 0.00 -0.7 0.00 1.4 0.00 -0.4

ASX 0.00 0.0 0.00 0.3 0.00 0.0

TWR NZX 0.00 1.3 0.00 -0.3 0.00 -0.3

ASX 0.00 0.2 0.00 -0.2 0.00 -0.2

Panel B: Unit root test in differences

AIA NZX -1.33 -192.2* -1.33 -192.2* -1.33 -192.2 *

ASX -0.99 -200.2* -0.99 -200.2* -0.99 -200.2*

AIR NZX -1.17 -182.3* -1.17 -182.3* -1.17 -182.3 *

ASX -1.01 -339.8* -1.01 -339.7* -1.01 -339.7*

FBU NZX -1.43 -184.8* -1.43 -184.8* -1.43 -184.8 *

ASX -1.14 -191.6* -1.14 -191.6* -1.14 -191.6*

FPA NZX -1.25 -183.4* -1.25 -183.4* -1.25 -183.4 *

ASX -1.02 -179.9* -1.02 -179.9* -1.02 -179.9*

FPH NZX -1.27 -186.3* -1.27 -186.3* -1.27 -186.3 *

ASX -1.00 -173.1* -1.00 -173.1* -1.00 -173.1*

SKC NZX -1.33 -198.4* -1.33 -198.4* -1.33 -198.4 *

ASX -1.08 -200.5* -1.08 -200.5* -1.08 -200.6*

TEL NZX -1.67 -211.9* -1.67 -211.9* -1.67 -211.9 *

ASX -1.53 -234.6* -1.53 -234.6* -1.53 -234.6*

TWR NZX -1.15 -183.9* -1.15 -183.9* -1.15 -183.9 *

ASX -1.20 -228.5* -1.20 -228.5* -1.20 -228.5*

Critical value -1.6 -2.6 -3.2

9 The same conclusion holds for the all sub-periods. Results are available on request.

159

Appendix 4-2: Unit Root Test for Australian Stocks

An augmented Dickey-Fuller (1981) test for unit roots is performed over the full study period10. Three equations are used: ∑𝑖 𝑖 𝑖 , 0 ∑𝑖 𝑖 𝑖 ,

Stock Code NC t-test C t-test CT t-test        Panel A: Unit root test in levels

AMP NZX 0.00 -0.6 0.00 -0.3 0.00 -0.3 ASX 0.00 -0.7 0.00 0.5 0.00 0.5

TLS NZX 0.00 -1.4 0.00 -2.3 0.00 -2.3 ASX 0.00 -1.4 0.00 -1.6 0.00 -1.6

WBC NZX 0.00 0.2 0.00 -1.6 0.00 -1.6 ASX 0.00 -1.7* 0.00 -1.3 0.00 -1.7

Panel B: Unit root test in differences

AMP NZX -1.52 -214.7 * -1.07 -197.9 * -1.07 * -197.9 ASX -1.07 -197.9* -1.52 -214.7* -1.52* -214.7

TLS NZX -1.04 -341.3* -1.04 -341.3* -1.04* -341.4 ASX -2.22 -231.1* -2.22 -231.1* -2.22* -231.1

WBC NZX -1.02 -261.1* -1.03 -261.1* -1.03* -261.1 ASX -1.14 -271.4* -1.13 -304.2* -1.14* -271.4

Critical value -1.6 -2.6 -3.2

10 The same conclusion holds for the all sub-periods. Results are available on request.

160

Appendix 4-3: Hasbrouck Information Share over Time

This table calculates the upper and lower bound of information for cross-listed stocks on ASX and NZX over “transparent”, “transition” and “anonymous” periods.

Transparent (%) Transition (%) Anonymous (%) Upper Lower Upper Lower Upper Lower Midpoint Midpoint Midpoint Bound Bound Bound Bound Bound Bound Panel A: New Zealand Companies AIA NZX 99.90 5.20 52.55 96.40 39.30 67.85 98.80 46.00 72.40 ASX 94.80 0.10 47.45 60.70 3.60 32.15 54.00 1.20 27.60

AIR NZX 98.20 2.70 50.45 36.90 8.00 22.45 92.70 46.10 69.40 ASX 97.30 1.80 49.55 92.00 63.10 77.55 53.90 7.30 30.60

FBU NZX 97.80 68.80 83.30 97.20 62.20 79.70 95.90 57.90 76.90 ASX 31.20 2.20 16.70 37.80 2.80 20.30 42.10 4.10 23.10

FPA NZX 98.30 67.60 82.95 92.40 33.30 62.85 99.90 69.50 84.70 ASX 32.40 1.70 17.05 66.70 7.60 37.15 30.50 0.10 15.30

FPH NZX 99.90 9.00 54.45 98.90 49.20 74.05 98.50 60.40 79.45 ASX 91.00 0.10 45.55 50.80 1.10 25.95 39.60 1.50 20.55

SKC NZX 98.70 78.30 88.50 92.40 50.30 71.35 97.40 54.50 75.95 ASX 21.70 1.30 11.50 49.70 7.60 28.65 45.50 2.60 24.05

TEL NZX 73.50 41.80 57.65 63.30 22.90 43.10 64.40 28.90 46.65 ASX 58.20 26.50 42.35 77.10 36.70 56.90 71.10 35.60 53.35

TWR NZX 95.30 79.10 87.20 77.10 8.40 42.75 80.30 33.00 56.65 ASX 20.90 4.70 12.80 91.60 22.90 57.25 67.00 19.70 43.35

Panel B: Australian Companies

AMP NZX 31.00 0.00 15.50 15.70 1.50 8.60 64.80 1.50 33.15 ASX 100.00 69.00 84.50 98.50 84.30 91.40 98.50 35.20 66.85

TLS NZX 28.70 0.50 14.60 26.40 1.00 13.70 37.60 0.30 18.95 ASX 99.50 71.30 85.40 99.00 73.60 86.30 99.70 62.40 81.05

WBC NZX 48.10 0.60 24.35 26.60 0.00 13.30 41.00 1.70 21.35 ASX 99.40 51.90 75.65 100.00 73.40 86.70 98.30 59.00 78.65

161

CHAPTER 5

QUOTE-BASED PRICE MANIPULATION: THE HONG KONG

EVIDENCE

5.1 Abstract

This chapter identifies trading structure effects on price manipulation. It presents initial evidence for quote-based manipulation, a form of costless closing price manipulation facilitated by the closing mechanism of HKEx. Closing price can be manipulated solely through quotes without trading. The manipulator is able to inflate (deflate) closing prices through placing orders to buy (sell) small quantities of shares at prices higher (lower) than the prevailing market prices near the market close. Using 123 prosecuted cases of closing price manipulation, it is found that quote-based price manipulation results in increased closing prices, bid returns, and bid frequencies immediately prior to the close of the market. In contrast to previous studies, the stocks that are most susceptible to this kind of manipulation are found to be those with low trading volume and depth. This chapter also investigates the potential for closing price manipulation in the current market climate and finds evidence that the HKEx still suffers from numerous potential manipulations. The continued existence of costless manipulation suggests that illiquid stocks would benefit from the introduction of non-continuous call auctions and/or market-makers as practised in Euronext Paris and the London Stock

Exchange.

162

5.2 Introduction

To quote the great Austrian economist Friedrich von Hayek (Hayek, 1945, pp.526-7):

"We must look at the price system as such a mechanism for communicating information if we want to understand its real function ... The most significant fact about this system is the economy of knowledge with which it operates, or how little the individual participants need to know in order to be able to take the right action. In abbreviated form, by a kind of symbol, only the most essential information is passed on and passed on only to those concerned. It is more than a metaphor to describe the price system as a ... system of telecommunications which enables individual producers to watch merely the movement of a few pointers”.

Hayek intends to assert the information role of the price mechanism, and consequently, it is very important to understand how the market mechanism works.

O’Hara (1995) stresses that since the decisions of market participants are not independent of the specific trading mechanism used, then understanding how various characteristics of the trading mechanism affect the behaviour of security prices is clearly important. It is true that even manipulating traders study the price formation process in order to know the best way to move prices to their own benefit.

Price manipulation can be carried out in many different ways and take many forms. It can occur when the manipulator short-sells a stock, pressing the price downwards by inducing others to sell, and covers his position at a depressed price

163

(Gerard and Nanda, 1993). The most common form of price manipulation is

“pump-and-dump”, through artificially pushing up the price in order to sell larger volumes at a higher price. This chapter examines the impact of the price mechanism on price manipulation in the Hong Kong Stock Exchange (HKEx), using a sample of 123 prosecuted cases from 1999 to 2006, all of which involve manipulators’ attempts to inflate security prices. The key contribution of this chapter is to provide initial evidence for quote-based manipulation, a costless form of manipulation facilitated by the unique closing price procedure of HKEx.

The manipulator is able to inflate (or deflate) the closing price by placing small orders to buy (or sell) at prices higher (or lower) than the prevailing best bid near the market close. As the closing price is constructed based on the best bid and ask quotes in the absence of a trade, closing price manipulation can be achieved solely through quotes, allowing prices to be manipulated without trading, and hence without explicit cost.

The existing literature focuses mainly on manipulation schemes which involve the actual trading of stocks, and have not specifically considered the impact of differing closing mechanisms. Many studies try to generalise multiple forms of manipulation in one model, while different modelling frameworks result in different predictions. Aggarwal and Wu’s (2006) model demonstrates that the presence of “information seekers” (momentum traders) is necessary for successful manipulation, otherwise the “unravelling problem”11 rules out the possibility of profitable manipulation. Hanson and Oprea (2009) model market manipulation in

11 Purchasing shares will push up prices, while selling shares will depress the stock price (Aggarwal and Wu, 2006, p1916).

164

a prediction market. They theorise that manipulation induces more traders to be better informed, resulting in improved price accuracy.

Based on a sample of 123 prosecuted cases from 1999 to 2006, this chapter finds compelling evidence that the closing price calculation method used in Hong Kong encourages manipulation techniques. Due to the costless nature of manipulation, the “unravelling problem” documented by Aggarwal and Wu (2006) is not applicable. The presence of “information seekers” is not necessary for the closing price manipulation to occur; however, their presence allows for the profitable sale of shares by the manipulator after the manipulation. Quote-based manipulation is associated with inflated closing prices, low trading activity and depth, narrower spreads, and much less mean reversion the following day than has been documented in previous studies. The conduct of manipulators in Hong Kong differs significantly from the typical manipulation documented in studies such as

Hillion and Suominen (2004) and Comerton-Forde and Putniņš (2011), with manipulators not seeking to actively purchase shares at the end of the day, nor is the typical reversion in price on the following day observed.

These findings have significant implications for the design of market structures generally. The closing mechanism of HKEx is particularly susceptible to price manipulation for illiquid stocks. The potential scale of the problem is exacerbated by the fact that the Hong Kong market trades numerous ‘‘penny’’ stocks. The introduction of call auctions and/or market-makers may be necessary to minimise this kind of manipulative conduct.

165

The remainder of this chapter is organised as follows. The next section describes

HKEx market structure and its closing procedure. Section 5.4 presents the data and Section 5.5 reviews the literature and outlines several hypotheses. Section 5.6 describes the research methodology and Section 5.7 reports the results. Section

5.8 discusses implications for the current market structure in HKEx and Section

5.9 concludes.

5.3 Hong Kong Market and the Closing Procedure

In 2011, the Hong Kong Stock Exchange was the sixth largest equity market in the world in terms of market capitalisation, and the third largest in Asia-Pacific after the Tokyo Stock Exchange and Shanghai Stock Exchange. The Hang Seng

Index is the main indicator of overall market performance. It consists of the 48 largest stocks, representing about 60% of capitalisation of HKEx. A striking phenomenon in HKEx is that the market has an unusually large number of penny stocks. In December 2011, 713 of the 1462 listed stocks that recorded transactions had share prices at or below HK$1.00, with 501 stocks (34%) exhibiting prices below HK$0.50. Many of these penny stocks receive significant retail investment, with retail participation in HKEx securities by approximately 33.8% of the Hong

Kong adult population in 201112.

12 For further information on the composition of the Hong Kong Market see the HKEx Fact Book 2011.

166

5.3.1 Market Structure

The Exchange operates an order-driven trading system. Buyers and sellers trade with each other through an electronic trading platform. It does not have market- makers except in Exchange Traded Funds, and securities listed on Nasdaq or the

American Stock Exchange under the Nasdaq-Amex Pilot Programme.

HKEx uses a call auction to open trading, then a continuous trading auction throughout the trading day. Current trading is conducted in two continuous trading sessions: a morning session from 9:30 to 12:00 and an afternoon session from

13:00 to 16:00 (effective from 5 March 2012). Trading during this study occurred in the morning from 10:00 to 12:30; and in the afternoon from 14:30 to 16:00.

Traders place their orders directly into an order book in which orders continuously interact during trading hours. There are two main types of orders. A limit order is an instruction to trade at the best price available (i.e., a minimum ask price to sell securities or a maximum bid price to buy securities). A limit order allows matching only at the specified price. An order with no specified price is called a market order, which is an instruction to trade at the best price currently available in the market. Market orders usually fill quickly, but sometimes at inferior prices.

In HKEx, both limit orders and market orders are accepted during the opening auction session, while only limit orders are accepted during the continuous trading session for automatic matching under AMS/3 (Third Generation Automatic Order

Matching and Execution System) in price and time priority. However, investors can submit market orders to their brokers who will place them in the form of limit

167

orders that match the best price on the other side of the book13. The information regarding the limit-order book is immediately available to all market participants through the electronic screen. The AMS/3 displays the best five bid and ask prices and aggregate order volume, along with the broker identifiers. Orders which are not executed by the end of the trading day will automatically lapse or expire.

Investors will need to re-enter the order if they want to place it for the next trading day.

HKEx has a minimum trading unit of shares, called single-board-lot. Each stock has its own individual board-lot size, ranging from 500 shares to 20000. The tick size for these stocks is dependent on their prices, with stocks priced between

HK$0.01–0.25 trading with a tick size of 0.1 cent, stocks between HK$0.25–0.5 trading with a tick size of 0.5 cent, stocks with prices between HK$0.5–10.0 have a tick size of 1 cent, stocks between HK$10–20 have a tick size of 2 cents, between HK$20–100 have a tick size of 5 cents and stocks with prices greater than HK$100 have a tick size of 10 cents.

Order submission rules for trading in the continuous trading are also applied during the closing procedure, because it is part of the continuous trading session.

In general, a sell order input price cannot be made at a price below the best bid price, whereas the buy order input price cannot be made at a price above the best ask price, if available. The “24-tick” (or “24-spread”) rule applies to orders submitted during the continuous, and hence closing portion of trading. Limit orders to be input should be at a price within +/‒ 24 ticks from the prevailing

13 Ahn et al. (2001) have a good discussion of Hong Kong market operation.

168

market price (i.e., the current best bid-ask or previous closing price at the open).

This rule places maximum bounds on quote-based manipulation in the case where no ask price exists.

For a HK$20 stock, 24 ticks correspond to HK$0.48 or a 2.4% price increment

(tick size of 2 cents); however, for a $0.20 stock, 24 ticks correspond to $0.024 or a 12% price increment (tick size of 0.1 cent). This reveals that small stocks have the potential to trigger noticeable spikes in prices; even a movement of a few cents can generate handsome percentage returns.

5.3.2 The Closing Procedure and Quote-based Manipulation

The closing price of a stock is determined by taking the median of five nominal prices at 15-second intervals in the last minute of the continuous trading session.

More specifically, if a stock trades frequently on a given day, the closing price for that day is determined by observing the last traded price as at 15:59:00, 15:59:15,

15:59:30, 15:59:45 and 16:00:00. The median of these five prices becomes the closing price for the day.

If there is only one trade after 15:59:00 (or for the day), the trade price will form the closing price as long as it is between the current best bid and ask quotes. If the last traded price is below the best bid, the best bid will become the closing price.

Similarly if the last traded price is above the best ask price, the ask price will become the closing price. Numeric examples of these situations are provided in

Panel A of Table 5-1.

169

Table 5-1: Examples of HKEx Closing Price Determination

Panel A: when a stock has been traded up to the market close Best Bid > Last Price Best Ask < Last Price Last Price between Bid and Ask

Last traded $0.220 Last traded $0.220 Last traded $0.220

Best bid $0.230 Best bid $0.200 Best bid $0.215

Best ask $0.250 Best ask $0.210 Best ask $0.225

Closing Price $0.230 Closing Price $0.210 Closing Price $0.220

Panel B: when a stock has not been traded up to the market close Ask Missing Both Bid and Ask Exist Both Bid and Ask Missing

Last traded Nil Last traded Nil Last traded Nil

Best bid $0.244 Best bid $0.224 Best bid Nil

Best ask Nil Best ask $0.225 Best ask Nil

Previous Close $0.220 Previous Close $0.220 Previous Close $0.244

Closing Price $0.244 Closing Price $0.224 Closing Price $0.244

If there are no trades for a given stock, the previous closing price will become the closing price as long as that price is between the best bid and ask prices. If the previous close is outside the best bid and ask, the closing price will be determined in the same way as when the last traded price is outside of the best bid and ask, as explained above. If one side of the market does not contain any depth, the closing price will be set to the best quote available (either bid or ask). If there are no quotes on either side of the market, the closing price is set to the previous day’s close. Numeric examples of these situations are provided in Panel B of Table 5-1.

For large and liquid stocks, choosing the median of five snapshots in the last minute of trading avoids the closing price being biased by one single trade (i.e., the last trade). For small stocks, however, this process increases the risk of stock prices being manipulated, because their closing prices can be biased by one single

170

trade. In the case of a stock with no trading on a given day, the price can be manipulated with a single bid.

In cases where the best bid is greater than the last traded price, or where no trades have occurred on a given day, closing prices can be artificially inflated by increasing the best bid immediately prior to the close. If there is a prevailing ask, a manipulator places a limit buy order at a price just below the ask price, narrowing the spread and increasing the closing price. The illiquidity of the stock and the potential to enter the manipulative bid seconds from the close of continuous trade minimises the probability of the manipulative bid being executed against.

If there is no prevailing ask, a manipulator can place a limit buy order at a price up to 24 ticks above the previous closing price or current bid price. In very low- priced stocks this can result in closing price movements in excess of 50%.

Manipulators executing this type of strategy would optimally use limit orders of the minimum board-lot size in order to minimise the risk of a counterparty executing against their standing limit order prior to the close of the market. If the stock is not traded on the next day, this inflated price will be maintained and there will be limited price reversion.

5.4 Data

This study examines prosecuted cases of closing price manipulation on HKEx from 1999 to 2006. A database of closing price manipulation cases is manually constructed by analysing HKSFC (Hong Kong Security and Future Commission)

171

Enforcement News. Cases are eliminated if the stocks they refer to: (1) are not common stock; (2) are not closing price manipulation; (3) do not have identifiable manipulation dates; (4) do not have available trade and quote data.

Trade and quote data are obtained from the Reuters DataScope Tick History

Database via the Securities Industry Research Centre of Asia Pacific (SIRCA).

The trade data is in one-minute intervals, containing fields with the security code, date, time, price and volume. In each interval, the last trade price and the total volume traded are calculated with the exception of the last one minute. The price at 16:00 is the official closing price determined by the median of five nominal prices (as explained in Section 5.3.2).

1 th The quote data is time-stamped to the nearest /100 of a second. Each quote consists of the best bid and ask prices, and the depth at the best bid or ask prices.

For consistency, a one-minute quote data set is generated from this data set. The prevailing quotes and respective depth levels are recorded at the end of each minute.

5.4.1 A Typical Example of Quote-based Manipulation Scheme

The sample of prosecuted closing price manipulations comprises 123 instances or stock-days of manipulation. They are obtained from 14 independent manipulation cases, involving 26 manipulated stocks. The specific offence date for each manipulation instance is provided by HKSFC. In all cases, the manipulator attempted to inflate the closing price, typically by using small limit buy orders

172

without an intention to trade. No attempted price deflation was prosecuted by

HKSFC. This is consistent with the sample of closing price manipulations reported in Canada and the US by Comerton-Forde and Putniņš (2011). Appendix

5-1 describes alleged quote-based manipulation schemes.

A typical example is the manipulation of shares of Chinney Alliance Group

Limited (stock code 0385.HK) by a retail investor, Lam Yat Wa. He placed repeated single-board-lot orders to buy shares at prices higher than the prevailing bid price near the market close on each of 29 June, 4 July and 10 July 2001. The following table provides snapshots of the transactions and official closing prices in the afternoon session and on each of the manipulation days.

On 29 June 2001, Lam placed two single-board-lot buy orders, one at 15:59:20 at a price of HK$0.064, and another at 15:59:35 at a price of HK$0.072. The last bid of HK$0.072 effectively set up the official closing price on the day, and resulted in a return of 30.91% compared to previous close. On 4 July, Lam used the same manipulation scheme by placing a bid 12 seconds prior to the close of the market at HK$0.065, pushing the closing price 10.17% higher than the previous day. On

10 July 2001, Lam placed three manipulative bids near the close, with one being executed, pushing the closing price 40% higher compared to the previous close.

In each of these three manipulation instances, manipulated stocks were not actively traded up to the time of manipulation. This facilitated costless manipulation, requiring only bid orders which were cancelled after the market closed.

173

Table 5-2: Three Instances of Quote-based Manipulation for Stock 0385 This table presents the transactions during the afternoon session on 29th June, 4th and 10th July 2001, respectively. A blank ask size indicates periods with no ask depth. The manipulator’s quotes are marked in bold.

Instance 1: 29/06/2001 Time Type Price Volume Bid Size Bid Price Ask Size Ask Price 14:30:18 Quote 200,000 0.059

14:31:59 Quote 120,000 0.056 200,000 0.059

15:15:55 Quote 270,000 0.055 200,000 0.059

15:15:55 Trade 0.056 120,000

15:34:45 Quote 270,000 0.055 ------

15:59:20 Quote 10,000 0.064 ------

15:59:35 Quote 10,000 0.072 ------Previous Close 0.055

Closing Price 0.072

Instance 2: 4/07/2001

14:31:08 Quote 290,000 0.054 190,000 0.06

15:49:32 Trade 0.054 100,000

15:49:32 Quote 190,000 0.054 190,000 0.06

15:52:16 Quote 100,000 0.055 190,000 0.06

15:56:19 Quote 300,000 0.055 190,000 0.06

15:58:11 Quote 300,000 0.055 ------

15:59:48 Quote 10,000 0.065 ------Previous Close 0.059

Closing Price 0.065

Instance 3: 10/07/2001

14:30:09 Quote 200,000 0.051 ------

15:58:41 Quote 10,000 0.06 ------

15:58:52 Quote 10,000 0.07 ------

15:59:57 Trade 0.07 10,000

15:59:57 Quote 10,000 0.06 90,000 0.07 Previous Close 0.05

Closing Price 0.07

Lam aimed to mislead the market into believing that the share price of these stocks was going up, and then to sell the stock at higher prices eventually. He was convicted for intentionally creating a false market by misleading the market and sentenced to six months’ imprisonment, suspended for one year, plus a fine totalling $30,000. He was also ordered to pay investigation costs of $37,006 to the

SFC. The motivations for inflating the closing price of stocks by prosecuted manipulators vary widely. Some of them attempt to sell stocks at high prices on subsequent days, while others seek to increase the collateral value of the manipulated shares and so reduced their required margin.

174

This quote-based manipulation scheme tends to be concentrated when the trading activities of stocks are particularly low. Table 5-3 provides descriptive statistics for the sample of 123 stock-days of closing price manipulation and shows that manipulators in Hong Kong try to minimise their manipulation cost by using quotes rather than trades. They overwhelmingly place bid orders of one board-lot to avoid the potential costs of execution. Only 13 out of 170 manipulative bid orders were executed against, amounting to less than 10% of all manipulative orders.

Table 5-3: Descriptive Statistics for the Sample of 123 Stock-days of Closing Price Manipulation This table summarises the number of total stock-days on which manipulation occurred, the number of manipulation days occurring at month-end, stock-days when no ask side was quoted, the number of single- board-lot bid orders used to manipulate the closing price, and the number of executions of those bid orders during the manipulation.

Executions Month-end Stock-days single-board-lot against Stock Stock-days Stock-days without Ask Bids Manipulator 0181 1 1 1 (100%) 13 0 0205 3 1 1 (33%) 9 0 0224 3 0 2 (67%) 3 0 0385 3 0 2 (67%) 6 2 0439 8 1 2 (25%) 12 0 0487 13 0 1 (8%) 15 2 0529 4 0 0 4 0 0542 9 0 4 (44%) 19 0 0544 1 0 1 (100%) 1 0 0567 5 0 0 5 2 0704 1 0 0 1 0 0725 3 1 3 (100%) 4 0 0759 1 0 0 1 1 0856 29 3 0 30 0 0938 7 1 0 7 0 0986 5 0 0 7 1 8009 4 0 0 4 0 8019 2 0 0 2 0 8037 1 0 0 1 1 8065 2 0 0 2 0 8163 1 0 1 (100%) 2 0 8171 5 1 3 (60%) 7 4 8175 4 0 0 4 0 8182 2 0 2 (100%) 2 0 8239 1 0 1 (100%) 2 0 8250 5 0 5 (100%) 7 0 Total 123 9 29 170 13

175

5.4.2 Trading Characteristics of Manipulated Stocks

All of the stocks within the sample are in the sub-HK$1 range, consistent with the idea that they are relatively illiquid stocks. Table 5-4 provides a comparison between the sample stocks and the constituents of the Hang Seng Index (HSI) for a period of 30 trading days ending 15 days prior to the first instance of manipulation.

Table 5-4: Characteristics of Manipulated Stocks Compared to HSI Stocks This table reports the daily trading summary for manipulated stocks and HSI stocks. This includes average stock price, dollar trading volume, number of trades, average percentage spread, and average market capitalisation during a period of 30 trading days ending 15 days prior to the manipulation. Manipulated HSI Difference Stocks Components Stock Price (HK$) 0.16 23.61 -23.45 Volume (HK$000) 101 123,000 -122,899 No. Trades 6 282 -277 Spread (%) 13.68 0.56 13.12 Market Cap (HK$000) 133,000 90,500,000 -90,367,000

Table 5-4 shows that the HSI components are orders of magnitude larger than the sample stocks in terms of market capitalisation, price and volume. Manipulated stocks trade few times per day, and have much wider spreads than the constituents of the HSI. Manipulators are likely to target stocks that are typically small and illiquid. This is in contrast to Comerton-Forde and Putniņš (2011), who find that manipulated stocks tend to be larger and more liquid than average. This difference can be explained by the motivations of the manipulators. Comerton-Forde and

Putniņš (2011) find that manipulators typically profit outside the manipulated market from contracts based on closing prices, where such contracts are less likely

176

on illiquid stocks. Quote-based manipulation is more akin to the pump-and-dump strategies documented in Aggarwal and Wu (2006), who find it is easier to manipulate illiquid stocks, as manipulators profit from manipulation of the underlying stock. This is consistent with the fact that under the current closing procedure in HKEx, quote-based closing price manipulation is likely to succeed for illiquid stocks.

5.5 Literature Review and Hypotheses

Stock markets are designed to facilitate the execution of trades and promote liquidity and price efficiency, while market manipulation deliberately attempts to interfere with the market operation by creating artificial market activity with respect to the price of listed securities (Harris, 2003). The closing prices of securities often serve as benchmarks for the value of derivative products and portfolios and hence are enticing targets for manipulators. For example, brokers may want to influence the closing price to alter their customers’ impression about their execution quality (Hillion and Suominen, 2004). Traders with large positions in futures markets have an incentive to manipulate underlying prices, especially when closing prices are used to determine settlement (Kumar and Seppi, 1992).

Retail investors also have an incentive to manipulate the closing price in order to profit from a desired price change as discussed in Section 5.4.

There is a large body of literature documenting the abnormal behaviour of closing prices. Harris (1989) reports an abnormally large jump in the last transaction price in the US equity markets between 1981 and 1982. More recently, Cushing and

177

Madhavan (2000) study Russell 1000 Index stocks, and find that the last five minutes of trading account for a disproportionately large fraction of the variation in daily returns.

Recent studies examine the potential of manipulation to explain this end-of-day phenomenon. Comerton-Forde and Putniņš (2011) study a sample of prosecuted cases in the US and Canada, using the difference-in-difference methodology.

They find that there is a significant increase in price, spreads and trading activity at the end of the day in the presence of manipulation. Hillion and Suominen (2004) also find that manipulation causes a significant rise in volatility, volume and bid- ask spreads in the last minute of trading on the Paris Bourse. While the activity of end-of-day manipulators has been described by these studies, this behaviour has occurred in an environment with typical closing mechanisms. Due to the unusual closing price mechanism described in Section 5.3, the end-of-day manipulation perpetrated in Hong Kong more closely resembles pump-and-dump manipulation.

A pump-and-dump strategy involves artificially pushing the stock price up, attracting interest from uninformed investors to the stock and then selling the stock at inflated prices. This manipulation strategy is traditionally categorised as engaging in the actual trading of stocks to affect prices (i.e., trade-based manipulation).

Aggarwal and Wu (2006) develop a mode to explain profitable manipulation, and the cases for their empirical analysis involve the use of rumours, wash sales, and attempts to corner the market. Their model is based on the earlier framework

178

developed by Allen and Gale (1992). The central idea is that manipulators pretend to be informed by trading large volumes at inflated prices. Other traders misidentify manipulators as informed traders, thereby exaggerating price movements and enabling manipulators to make profits. Aggarwal and Wu (2006) consider a market where a manipulator trades in the presence other traders, so called “information seekers”. They demonstrate that the presence of information seekers is necessary for profitable manipulation, due to the unravelling problem.

Their model shows that more information seekers imply a greater competition for shares, making it easier for a manipulator to exit the market. They test their model using prosecuted manipulation cases in the US dealer markets. Consistent with their theoretical prediction, they show that liquidity and return increase during the manipulation period, and subsequently decrease after the manipulation ends. They also find that manipulations happen typically in small and illiquid stocks.

Hanson and Oprea (2009) model trade-based manipulation in a prediction market.

In contrast to Aggarwal and Wu (2006), they find that manipulators are unable to mislead the market and distort prices. Instead, manipulation causes price to become more accurate. They find that manipulative trades stimulate counteracts by informed traders, who have solid financial resources and try to profit from manipulation. The central tenet of their model is the rational expectations theory, which predicts that in equilibrium asset prices will reflect all of the information held by market participants. Rational traders who are able to recognise manipulation will profitably counteract manipulation, thereby offsetting any price distortion.

179

These studies try to generalise manipulation characteristics into one model, however, with opposing predictions. This is not surprising, because these models are based on different theoretical frameworks, without reference to particular trading mechanisms and the motivation for the manipulation itself.

Unlike prior literature, this chapter shows that price mechanisms determine the techniques used to manipulate price. Hypotheses are tested regarding the effects of quote-based manipulation on trade and quote behaviour and provide evidence that the current role of closing price procedure allows successful closing price manipulation.

Though prior literature on the effects of manipulation is inconclusive, there is little doubt that manipulators are able to influence price. As identified in Section

5.3, quote-based manipulation attempts to inflate the closing price by placing bid orders with small quantities at high prices. Therefore, one would expect to observe positive abnormal return in the presence of manipulation. This is consistent with Comerton-Forde and Putniņš (2011), who find that manipulated stocks experience large abnormal day-end returns. This leads to hypothesis one

(H5.1):

Hypothesis5.1: The closing price return increases significantly in the presence of quote-based manipulation.

Aggarwal and Wu (2006) argue that manipulation succeeds because manipulators lure information seekers to trade. These information seekers mistake manipulators for informed traders and mimic their trades. This allows the manipulators to

180

establish price momentum and unwind their position. Hillion and Suominen (2004) also find that manipulation causes the rise in volume in the last minute of trading on the Paris Bourse. These studies demonstrate the ability of manipulators to generate trading interest and profits by trading large volumes. However, in the context of HKEx, the closing mechanism allows profitable manipulations without the necessity of trading large amount of shares. A manipulator can establish a high closing price without cost by submitting a small buy order at a high price just seconds before the close. In fact, most of these buy orders do not execute. Even if some of manipulative bid orders are executed, a resulting small trading volume can be expected.

Therefore, in contrast to previous studies of end-of-day manipulation, quote-based manipulation is expected to be most successful in illiquid stocks. This leads to the second hypothesis (H5.2):

Hypothesis5.2: End-of-day trading activity is low in the presence of quote-based manipulation.

Since manipulation is carried out during the last minute of the market close, leaving little time for the market to respond, the trading behaviour of information seekers might be observed on the following morning. If this were the case, information seekers would compete for shares. As a result, there would be a further increase in prices accompanied by an elevated volume on the day following the manipulation. This leads to hypothesis three (H5.3):

181

Hypothesis5.3: The presence of information seekers will result in elevated prices and volume at the open following the manipulation.

Comerton-Forde and Putniņš (2011) argue that closing price manipulation often occurs through submitting large buy orders just before the close, resulting in a significant rise in the last minute returns and trading. Again, this is based on a theoretical reasoning that manipulators pretend to be informed, whereas the closing mechanism adopted by HKEx enables a manipulator to inflate a closing price through submitting high bids but without trading. This could be considered rational behaviour from the point of view of Harris (1989), who concludes that a manipulator who wishes to bias the closing price would use as small a transaction as possible to minimise potential costs of execution. Three hypotheses (H5.4 to

H5.6) regarding the bidding behaviour of manipulators are thus developed as follows:

Hypothesis5.4: End-of-day bid price return increases significantly in the presence of manipulation.

This hypothesis is consistent with Hanson et al. (2006), who find that manipulators submit high bids to inflate prices. Due to the ability to manipulate without cost using bid orders, it is anticipated that there will be an increase in the number of bid orders during the closing procedure, leading to H5.5:

Hypothesis5.5: Manipulation increases the frequency of end-of-day bid orders.

182

On the other hand, the optimal size of such a bid order is expected to be small to minimise manipulation costs.

Hypothesis5.6: End-of-day bid depth is significantly lower in the presence of manipulation.

Comerton-Ford and Putniņš (2011) argue that the submission and execution of bid orders by the manipulator to inflate closing prices lead to the consumption of depth on the ask side of the order book as well as increases in spreads. Hillion and

Suominen (2004) argue that manipulation is the cause of an observed rise in the spread in the last minute of trading on the Paris Bourse. However, as closing prices are inflated by placing bid orders at prices higher than the current best bid, seconds before the market closes, narrower spreads in the presence of manipulation are expected. This leads to H5.7:

Hypothesis5.7: End-of-day bid-ask spreads are narrower in the presence of manipulation.

Hanson et al. (2006) argue when market price is the best predictor of the future event, market participants are able to recognise manipulation and profitably counteract it, offsetting any price distortion. In the context of quote-based manipulation, a rational trader would take advantage of these aggressive bid offers, and profit from selling shares to the manipulator at inflated prices. Given that manipulation is carried out during the last minute of the market close, leaving little time for the market to respond, this response may occur the following morning. The actions of rational traders would lead to increased activity on the

183

ask side, resulting in increased volume and price reversal the following morning.

This leads to H5.8:

Hypothesis5.8: The presence of rational sellers will result in a high level of activity on the ask side, thereby increasing trading volume and price reversal following the manipulation.

5.6 Research Methodology

Intraday analysis is conducted on quote-based closing price manipulation. The objective of this study is to examine abnormal trading and quoting behaviour during and immediately after the manipulation so as to provide evidence on trading mechanism effects. An event window is manually constructed starting at

15:00 on the day of manipulation and ending at 14:59 the next day, a period which is considered as the ith manipulation or stock-day. Manipulation impacts are examined using both the difference-in-difference estimation and benchmark approach.

5.6.1 Difference-in-difference Estimation

The Difference-in-difference (DID) is defined as the difference in average outcome in the treatment group before and after treatment minus the difference in average outcome in the untreated control group during the same period

(Ashenfelter and Card, 1985). The logic behind the DID estimator is that if the manipulated stocks are only compared before and after the manipulation, the

184

effect of market-wide shocks around the time of manipulation cannot be excluded.

By using a control group, outcomes are observed for manipulated and non- manipulated stocks for two time periods, so as to difference the potential confounding factors and isolate the effects of manipulation. Comerton-Ford and

Putniņš (2011) employ DID in their study of manipulation effects. They show that

DID is able to control for selection bias which can arise from manipulators choosing stocks that systematically differ from other stocks in observable or unobservable characteristics, e.g., liquidity, or manipulation days that differ systematically from other days.

Using this DID technique, the abnormal trading and quoting behaviour due to manipulation can be measured, first through time: before and after the manipulation; and second across subjects: between manipulated stocks and non- manipulated control stocks.

This is expressed in equation (5.1),

훼̂ { [ ] [ ]} { [ ] [ ]} (5 ) 𝑖 𝑖 𝑖 𝑖

th where, for the i manipulation, 𝑖 is the mean values of an end-of-day variable

th for the manipulated stocks; time period denotes the day of the i manipulation, and 0 is a benchmark period of 30 trading days ending 15 days prior to the date of the first manipulation.

The first term is the before-after estimator for manipulated { [ 𝑖 ] [ 𝑖 ]} stocks, indicating the difference between the value of a variable on the day of

185

manipulation and a benchmark period for the same manipulated stocks. This term removes unobservable individual stock-specific effects.

𝑖 is the mean value of a variable for the non-manipulated control stocks. Since quote-based closing price manipulation is particularly plausible for illiquid and small stocks, the most liquid stocks: Hang Send Index components are thus used as the control group. They are 44 components of the Hang Seng Index that are not subject to manipulation and for which data is available during the entire study period (1999 to 2006). The second term is the before-after { [ 𝑖 ] [ 𝑖 ]} estimator for these 44 HSI stocks as the ith manipulation. This term controls for market-wide effects.

Subtracting the second term from the first term, the DID estimator 훼̂ is obtained, which not only accounts for the effects of stock-specific characteristics, but also removes the common trends in the market, such as intraday volatility or market shocks.

The two means are tested using the following formula under the assumption of unequal variance14.

(𝑋̅ 𝑋̅0) 𝑛 𝑛 (5 )

√ ( 𝑛 𝑛0)

14 The same results are obtained when conducting the test under the assumption of equal variance. For details in this two-sample t-test, please see Selvanathan et al. (2006).

186

where (𝑋̅ 𝑋̅0) refers to the before-after estimators or DID estimators (훼̂ ), is the pooled variance estimate, defined as:

(𝑛 ) (𝑛0 ) 0 (5 ) 𝑛 𝑛

where and 0 are the variances of the two samples, and 𝑛 and 𝑛 are the size of each samples.

5.6.2 Variables Measuring Trading and Quoting Characteristics

Variables including stock price return, trading volume and trade execution are used to characterise trading behaviour. Stock price is defined as the price at which a security is being bought or sold at a given minute, except for 16:00, when the official closing price is taken. Price return is calculated as the percentage change of price every minute per stock. Since stocks are thinly traded, prices are forward filled from the previous minute if there is no trade in the current minute.

Trading volume is defined as the percentage of outstanding shares traded, which is the number of shares traded divided by the number of shares outstanding for a firm at any given time. Trade execution is calculated as the ratio between number of trades generated and number of orders posted at a given minute. This ratio measures the proportion of trades executed.

Quoting behaviour is characterised by bid/ask quote return, bid/ask depth, bid/ask frequency, and bid-ask spread. Bid (ask) quote is the best prevailing price

187

at which the shares can be bought (sold). Bid (ask) price return is calculated as the percentage change of bid (ask) per minute for each stock. Like stock prices, bid and ask prices are also forward filled from the previous minute if there is no change in the current minute. Bid (ask) depth is the quantity of shares demanded

(supplied) at the best bid (ask) price and expressed as a percentage of shares outstanding. Bid (ask) frequency is used to measure the level of quoting activity, and is calculated as the total number of bid (ask) orders placed at a given minute.

Bid-ask spread is the proportional bid-ask spread calculated by dividing the difference between bid and ask prices by the midpoint of bid-ask price.

5.6.3 Further Analysis: A Benchmark Approach

A benchmark approach is used as an additional test to provide a more descriptive analysis of abnormal performance due to manipulation. Variable values are benchmarked for manipulated stocks by computing average values of those stocks during non-manipulation trading days (or normal performance period). For each stock, an extended benchmark period is constructed consisting of 100 trading days, ending 15 days prior to the first date of manipulation. There are in total 26 benchmark periods. Variables benchmarked include abnormal price return, abnormal quote return, abnormal trading volume, abnormal bid/ask depth, and trade frequency, bid/ask frequency, and bid-ask spread.

To calculate abnormal returns for stock price and quote, a mean-adjusted model of

Brown and Warner (1980) is used. Stock-day returns are compared with expected values based on a benchmark period. The expected return, ( 𝑖 ), is the average

188

return for each stock i at each minute t estimated from 100-day benchmark periods:

( ) ∑ (5 ) 𝑖 𝑖

There are in total 123 stock-days involving 26 manipulated stocks. For mth stock- day of manipulation, the abnormal price (or quote) return ( ), is the difference between the observed return (Ri,t,m) and the expected return of corresponding manipulated stock at minute t during the event window:

𝑖 ( 𝑖 ) (5 5)

The cumulative abnormal returns (CAR) are then obtained by accumulating the average abnormal returns, which are the average at every minute across 123 stock- days.

𝐶 ∑ (5 6) where

∑ 3 (5 7)

Standardised z-scores for ARt at each minute are thus calculated, so as to examine how many standard deviations an abnormal return is above or below the mean of all values.

189

( ̅̅̅ ̅) (5 )

where ̅̅̅ ̅ is the mean of and σ is the standard deviation of from 15:00 on the day of manipulation to 14:59 the next day without including lunch break

(i.e. 240 minutes). A positive z-score indicates the variable value is above the mean of all values, while a negative z-score indicates the variable value is below the mean.

Abnormal trading volume and abnormal bid/ask depth are also calculated from a mean-adjusted model, following an approach by (Campbell and Wasley, 1996).

As previously identified, the trading volume or quoted depth ( 𝑖 ) of a stock i is as a percentage of the total number of shares (𝑛𝑖 ) of that stock outstanding

( 𝑖 ) on a given day j, given minute t:

(𝑛𝑖 ) 𝑖 (5 ) 𝑖

As suggested by Campbell and Wasley (1996), this calculation uses the natural log of the percentage of outstanding shares traded metric appearing in the above equation. Before transformation the small constant of 0.000255 is added to preclude taking the log of zero in the case of zero trading volume in a given minute.

The expected volume or depth E( 𝑖 ), is the average volume for each stock i at each minute t and is estimated from 100-day benchmark periods:

190

( ) ∑ (5 ) 𝑖 𝑖

For mth stock-day of manipulation, the mean-adjusted abnormal volume (or depth)

( ), is the difference between the observed volume/depth ( 𝑖 ) and the expected volume/depth of a manipulated stock i at minute t:

𝑖 ( 𝑖 ) (5 )

The average abnormal volume/depth ( ) at each minute across 123 stock-days, with standardised z-scores are calculated as:

∑ 3 (5 )

( ̅̅̅ ̅) (5 )

where ̅̅̅ ̅ is the mean of and σ is the standard deviation of from 15:00 to

14:59 the next day. The exponent of the abnormal volume/depth measure is the ratio of 0.000255 plus the actual percentage volume to 0.000255 plus the expected percentage volume on mth manipulation15.

15 This can be presented in the algebra:

55 ( ) [ ( 55 ) ( 55 ( ))] 55 ( )

191

Trade frequency, bid/ask frequency and bid-ask spread are ratio variables, which are computed as a ratio between the value of a variable on stock-day and the average value of a variable during the 100-day benchmark. The benchmark average is the pooled average across 100 benchmark days and 26 stocks at each minute t.

∑𝑖 ∑ 𝑖 𝑛 (5 ) 6 ( )

For each mth stock-day at each minute t, a ratio for each variable with its z-score is calculated as:

(5 5) 𝑛

̅̅̅ ̅̅ ̅ ̅ ̅ (5 6) 𝑖

where

∑ 3 (5 7)

and ̅̅̅ ̅̅ ̅ ̅ ̅ is the mean of , and σ is the standard deviation of from

15:00 to 14:59 the next day.

192

5.7 Results

This section analyses manipulation effects. The first part tests Hypothesis5.1 to

Hypothesis5.3 for the impact of quote-based manipulation on trading behaviour.

The second part tests Hypothesis5.4 to Hypothesis5.8 for the impact on quoting behaviour, and the final part provides further daily evidence on quote-based manipulation.

5.7.1 Manipulation Effects on Trading Characteristics

Table 5-5 reports the mean difference-in-difference estimates for the last three minutes of trading and the first three minutes of the market open following the manipulation. The before-after estimates of stock price return reported in Panel C show a significant increase in end-of-day stock prices for manipulated stocks on the day of manipulation when compared to the 30-day benchmark period. The before-after estimates for HSI stocks are near zero, suggesting there are no strong market-wide factors on the manipulation days that can explain the significant increases in end-of-day stock prices.

After controlling for stock-specific and market-wide effects, the DID estimator in

Panel E shows an excess return of 13.73% at the last minute of trading. This abnormal return caused by manipulation is approximately 29 times larger than its usual level (0.47%), and is significantly greater than zero at the 1% level.

The robustness of these results is examined using the benchmark method.

Abnormal stock price returns with corresponding z-scores are calculated at each

193

minute during manipulated days. This benchmark approach confirms the findings.

Due to reason of space, results are presented in Appendix 5-216. These findings support Hypothesis5.1, and are also consistent with Comerton-Forde and Putniņš

(2011), who find that manipulated stocks experience large abnormal end-of-day returns.

16 Due to reasons of space, Appendix 5-2 only presents z-scores for the last three minutes of trading and first three minutes of the market open. Results are available on request.

194

Table 5-5: DID Estimation of the Impact of Quote-based Manipulation on Trading Behaviour This table reports the manipulation effects on trading characteristics using the DID estimation. n is the number of observation used in the calculation. In Panels C, D and E. *, ** and *** represent significance at the 10%, 5% and 1% levels, respectively.

Stock Price Trade Execution Trading Volume Stock Price Trade Execution Trading Volume Panel Time Time Return (%) (per 100 orders) (% per hour) Return (%) (per 100 orders) (% per hour) 15:58 0.00 0.00 0.00 10:01 -0.01 0.27 0.00 A: Manipulated stocks 15:59 0.21 3.36 0.00 10:02 0.00 0.00 0.00 on stock-days (n=123) 16:00 14.20 6.22 0.01 10:03 -0.01 1.64 0.02

15:58 0.04 38.43 0.05 10:01 0.00 3.27 0.05 B: Manipulated stocks 15:59 0.05 37.70 0.05 10:02 0.00 5.38 0.05 during 30-day benchmark (n=3690) 16:00 0.47 34.46 0.05 10:03 0.01 15.31 0.04

C: Before-after 15:58 -0.04 -38.43*** -0.05*** 10:01 -0.01 -2.99*** -0.05*** estimator for 15:59 0.16 -34.34*** -0.05*** 10:02 0.00 -5.38*** -0.05*** manipulated stocks 16:00 13.73*** -28.24*** -0.03*** 10:03 -0.02 -13.67*** -0.02 (n=3813)

D: Before-after 15:58 0.01 -1.11 -0.01 10:01 0.01 1.40 0.01 estimator for non- 15:59 0.00 0.40 0.00 10:02 0.01 0.90 -0.01 manipulated HSI stocks 16:00 0.00 0.40 0.00 10:03 0.00 0.80 -0.01 (n=167,772)

15:58 -0.05* -37.32*** -0.04*** 10:01 -0.02 -4.39*** -0.06*** E: DID estimator 15:59 0.16 -34.74*** -0.05*** 10:02 -0.01 -6.28*** -0.04** (171,585) 16:00 13.73*** -28.64*** -0.03** 10:03 -0.02 -14.47*** -0.01

195

Table 5-5 also presents evidence that end-of-day trading activity is lower in the presence of quote-based manipulation, supporting Hypothesis5.2. Trade execution ratio and trading volume are used to measure the level of trading activity. Panel A indicates that for every 100 orders placed, there are only about 6 trades executed on manipulated days, compared to an average of 34 trades executed in the control period. Trading volume on manipulation days is also significantly lower than the benchmark average. These differences are statistically significant at the 1% level.

The relative illiquidity of the manipulation sample is consistent with manipulators preferring illiquid stocks.

There is no evidence to support Hypothesis5.3. The data shows neither an increase in prices nor volume at the open following the manipulation. A small and insignificant price reversal is observed at the open. The mean price change at

10:01 is negative 0.01% (Panel A), and negative 0.02% (Panel E) after controlling for stock-specific and market-wide effects. Price reversion is also calculated at

10:30 and 11:00 the following morning to ensure that there is sufficient time allowed from the open for price discovery to take place (Comerton-Forde and

Putniņš, 2011). It is found that price reversal at 10:30 is negative 0.53%, while at

11:00 is 0.81%, all of which are still statistically insignificant.

Figure 5-1 provides details on stock price behaviour using the benchmark method.

The graph plots cumulative average abnormal returns from 15:00 on the day of manipulation to 14:59 the following day.

196

Figure 5-1: Cumulative Abnormal Return of Stock Price

19% 16% 13% 10% 7% 4% 1%

-2%

15:11 15:21 15:31 15:41 15:51 10:09 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:19 12:29 14:39 14:49 14:59

-5% 15:01 Market close Market

Stock prices increase an average of 12.7% just before the market close across the sample of manipulated stocks. There is a minor reversion on the following day, with the average CAR 10.1% on the following day by 14:59. Aggarwal and Wu

(2006) predict that stock price will increase further due to interaction between information seekers and manipulators. There is no evidence indicating that this process is occurring. These results also stand in contrast to the findings of

Comerton-Forde and Putniņš (2011), who indicate that an inflated stock price is followed by price reversals in the following open. The lack of price reversion may indicate that the Hong Kong market for illiquid stocks is not efficient.

Figure 5-2 depicts the behaviour of trading volume around the event window.

Consistent with Table 5-4, trading volume at the end-of-day manipulation is much lower than the benchmark period.

197

Figure 5-2: Abnormal Trading Volume

0.1

0.05

-0.05

12:09 12:29 14:49 15:01 15:11 15:21 15:31 15:41 15:51 10:09 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:19 14:39 14:59 -0.1

-0.15 close Market

NaturalLogarithm -0.2

-0.25

-0.3

The abnormal trading volume measure is between negative 0.27 and negative 0.14 during the last three minutes of trading. The actual percentage volume is only about 76% to 86% of the predicted percentage trading volume on the manipulation day (Exp(-0.27)=0.76 and Exp(-0.14)=0.86). This difference is statistically significant at the 1% level (see Appendix 5-2). This suggests that manipulators target stocks which are not actively traded. In the open following the manipulation, no unusual trading patterns are observed, consistent with a lack of information seekers.

5.7.2 Manipulation Effects on Quoting Characteristics

Table 5-6 reports the results of difference-in-difference estimation of the quoting behaviour of manipulated stocks. The results indicate that the bid return and bid frequency are significantly higher in the presence of manipulation. Manipulation is also characterised by significantly lower bid depth and narrower spreads.

198

Panel A of Table 5-6 shows that bid prices surge towards the market close on manipulated days. The average return is 7.18% at 16:00 (Panel A) on manipulation days, as compared to 0.36% (Panel B) during the benchmark period.

This results in a 6.82% (Panel C) excess return, which is statistically significant at the 1% level. The before-after estimates for HSI Index stocks are all insignificant, suggesting that such unusual bid price behaviour around the manipulation is not related to market-wide phenomena. Moreover, a 2.1% bid price reversion (Panel

C and Panel E) seems evident at the following open, consistent with the removal of the manipulators’ bids; however, this only accounts for about one-fifth of the abnormal returns.

199

Table 5-6: DID Estimation of the Impact of Quote-based Manipulation on Bid Behaviour This table reports the manipulation effects on quoting characteristics using the DID estimation. n is the number of observations used in the calculation. In Panels C, D and E, *, ** and *** represent significance at the 10%, 5% and 1% levels respectively.

Bid Price Bid Frequency Bid Depth Bid Price Bid Frequency Bid Depth Panel Time Spread (%) Time Spread (%) Return (%) (per hour) (%) Return (%) (per hour) (%)

15:58 2.96 10.73 0.021 11.77 10:01 -3.08 9.76 0.010 7.68 A: Manipulated 15:59 5.30 24.88 0.011 8.45 10:02 -0.61 6.34 0.013 8.63 stocks on stock-days (n=123) 16:00 7.18 38.05 0.004 2.80 10:03 -0.58 2.44 0.016 9.41

15:58 0.27 4.39 0.017 12.51 10:01 -0.99 8.41 0.005 13.44 B: Manipulated 15:59 0.76 8.49 0.016 11.77 10:02 -0.12 3.25 0.008 13.62 stocks during 30-day benchmark (n=3690) 16:00 0.36 6.33 0.016 11.58 10:03 -0.11 3.11 0.008 13.79

C: Before-after 15:58 2.69*** 6.34** 0.005 -0.74 10:01 -2.10** 1.35 0.005* -5.77** estimator for 15:59 4.53*** 16.39*** -0.005* -3.32** 10:02 -0.49* 3.09* 0.006* -4.99*** manipulated stocks 16:00 6.82*** 31.72*** -0.012*** -8.78*** 10:03 -0.47 -0.67 0.007** -4.38*** (n=3813)

D: Before-after 15:58 0.00 5.34 -0.001 -0.01 10:01 0.00 -4.74 0.000 0.03 estimator for non- 15:59 0.00 3.48 -0.001 0.00 10:02 0.00 -2.60 -0.001 0.02 manipulated HSI 16:00 0.00 2.10 -0.001 0.00 10:03 0.00 -5.49 -0.001 0.02 stocks (n=167,772)

15:58 2.69*** 1.00 0.006 -0.73 10:01 -2.10*** 6.09* 0.006* -5.80*** E: DID estimator 15:59 4.53*** 12.91*** -0.004 -3.32*** 10:02 -0.49 5.69*** 0.006* -5.01*** (171,585) 16:00 6.81*** 29.62*** -0.011*** -8.78*** 10:03 -0.48 4.82 0.008** -4.40***

200

Figure 5-3 plots the cumulative abnormal returns of bid price from 15:00 on the day of manipulation to 14:59 the following day. The corresponding z-scores are calculated in Appendix 5-317, pointing to the same conclusion as in Table 5-6.

Figure 5-3: Cumulative Abnormal Return of Bid Price

19.0%

16.0%

13.0%

10.0%

7.0%

4.0%

1.0%

-2.0%

15:41 10:29 11:19 12:09 15:01 15:11 15:21 15:31 15:51 10:09 10:19 10:39 10:49 10:59 11:09 11:29 11:39 11:49 11:59 12:19 12:29 14:39 14:49 14:59

Market close Market

There is a significant increase in bid prices around the market close. The cumulative abnormal bid returns of 18.56% are recorded at the last minute of trading. A bid price reversion of 2.51% is observed at 10:01 (see Appendix 5-3), followed by further small reversals. The cumulative abnormal bid returns remain at 11.02% by 14:59. These results support Hypothesis5.4, providing evidence that manipulators submit high bids to inflate stock prices, and are consistent with

Hanson et al. (2006).

Given manipulators use bid orders to conduct their manipulation, Hypothesis5.5 anticipated more frequent bid orders at the end of the day. The DID estimators for bid order frequency reported in Table 5-6 show that the end-of-day bid order

17 Due to reasons of space, z-scores are presented only for the last three minutes of trading and the first three minutes of the market opening. The full results are available on request.

201

frequency is significantly higher in the presence of manipulation. An average of

38 bid orders per hour (Panel A) is placed in the final minute of trading on manipulation days compared to only 6 orders (Panel B) in the benchmark period.

The DID estimators (Panel E) confirm that manipulation causes a significant increase in bid order frequency near the market close, which is significant at the 1% level.

Figure 5-4 plots the ratio of bid order frequency on manipulation days against the benchmark average.

Figure 5-4: Bid Order Frequency Ratio

6.0

5.0

4.0

3.0 Ratio 2.0

1.0

0.0

14:59 15:01 15:11 15:21 15:31 15:41 15:51 10:09 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:19 12:29 14:39 14:49 Market close Market

The graph shows that bid order frequency on manipulation days is around 6 times greater than the benchmark average. These results are consistent with manipulators repeatedly submitting high bids in order to inflate stock prices on

HKEx.

202

There is also evidence for Hypothesis5.6, providing support to Harris’s (1989) conclusion, that manipulators, who have no desire to actually transact at the quotes posted, will minimise the cost of potential execution by minimising the size of the order they place in the market. Table 5-6 indicates that the size of bid orders drops suddenly at the market close on manipulation days. Bid depth is only about a quarter of its benchmark level. This is consistent with manipulators submitting small bid orders to minimise manipulation costs. Figure 5-5 also shows a substantial decrease in bid depth at the market close graphically. The actual percentage depth at the end-of-day manipulation is about 24% of the predicted depth (Exp(-1.44)=0.24).

Figure 5-5: Abnormal Bid Depth

0.5

15:01 10:29 15:11 15:21 15:31 15:41 15:51 10:09 10:19 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:19 12:29 14:39 14:49 14:59

-0.5

NaturalLogarithm Market close Market -1

-1.5

During the first 20 minutes of trading on the following day, the average abnormal bid depth measure is about 0.5, while actual percentage depth is about 1.65 times the predicted bid depth (Exp(0.5)=1.65). The z-scores between 10:07 to 10:20 are statistically significant at the 10% level18. Although bid depth bounces up at the

18 Full results are available on request.

203

market open, it gradually decreases thereafter, indicating insufficient evidence to demonstrate that there is a strong demand for the stock.

Quote-based manipulation is only constrained by the ask side of the market.

Manipulators will not enter a bid price equal to the current best ask, lest their orders be immediately filled, incurring the cost of execution. This is one of the unique features of costless manipulation, as Comerton-Ford and Putniņš (2011) and Hillion and Suominen (2004) show that closing price manipulation causes wider spreads. Panel A of Table 5-6 shows that the bid-ask spreads are considerably narrower towards the market close. At 16:00, the average bid-ask spread is only 2.8% compared to 11.58% during the benchmark periods. This difference is statistically significant at the 1% level.

Figure 5-6 shows that the bid-ask spread is significantly reduced near the market close. At 16:00, it is approximately one quarter of the benchmark average. It increases the day after the manipulation, and remains narrower on average than it had been prior to manipulation. These results support Hypothesis5.7.

Figure 5-6: Spread Ratio

1.4 1.2 1

0.8

Ratio 0.6 0.4 0.2

10:09 14:49 15:01 15:11 15:21 15:31 15:41 15:51 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:19 12:29 14:39 14:59

Market close Market

204

The data shows weak evidence for Hypothesis5.8, which tests whether there are counteractions by rational traders (who recognise manipulations) leading to increased volume and price reversal on the following morning. According to

Hanson et al. (2006), if market participants are able to identify manipulators, they will take advantage of a manipulator’s aggressive buying by selling shares at inflated prices. This process will increase the level of activity on the ask side and the counteractions by rational traders will result in an increase in volume and price reversal after the manipulation. Table 5-7 reports that the end-of-day ask returns during manipulation (Panel A) are slightly higher than, but insignificantly different from, the returns during the benchmark period. The DID estimator of

Panel E shows that ask prices increase 3% the morning after manipulation, which is statistically significant at the 5% level. This seems to imply that the rational traders described by Hanson et al. (2006) are attempting to take advantage of the perceived change in the fundamental value of the security by increasing the price to sell the manipulated security.

205

Table 5-7: DID Estimation of the Impact of Quote-based Manipulation on Ask Behaviour This table reports the manipulation effects on selling behaviour using the DID estimation. n is the number of observations used in the calculation. In Panels C, D and E, *, ** and *** represent significance at the 10%, 5% and 1% levels respectively.

Ask Price Ask Frequency Ask Price Ask Frequency Panel Time Ask Depth (%) Time Ask Depth (%) Return (%) (per hour) Return (%) (per hour) 15:58 0.26 1.95 0.013 10:01 3.68 19.51 0.008 A: Manipulated 15:59 0.20 4.39 0.013 10:02 0.39 4.39 0.010 stocks on stock-days (n=123) 16:00 0.31 5.85 0.016 10:03 -0.15 6.83 0.013

B: Manipulated 15:58 0.00 2.28 0.038 10:01 0.71 13.37 0.007 stocks during 30-day 15:59 0.07 4.72 0.038 10:02 0.09 5.11 0.010 benchmark (n=3690) 16:00 0.03 2.63 0.038 10:03 0.14 4.47 0.011

C: Before-after 15:58 0.26 -0.33 -0.025*** 10:01 2.97** 6.15* 0.001 estimator for 15:59 0.13 -0.33 -0.025*** 10:02 0.30 -0.72 0.000 manipulated stocks 16:00 0.27 3.22 -0.022*** 10:03 -0.29 2.36 0.001 (n=3813)

D: Before-after 15:58 0.00 6.84 0.001 10:01 -0.03 -5.46 0.000 estimator for non- manipulated HSI 15:59 -0.01 5.76 0.001 10:02 0.01 -1.97 0.001 stocks (n=167,772) 16:00 0.00 -3.26 0.001 10:03 -0.01 -2.74 0.000

15:58 0.26 -7.17** -0.026*** 10:01 3.00** 11.61*** 0.001 E: DID estimator 15:59 0.14 -6.09** -0.026*** 10:02 0.29 1.25 0.000 (171,585) 16:00 0.28 6.48 -0.023*** 10:03 -0.28 5.09 0.001

206

Figure 5-7 further plots the cumulative average abnormal returns for the ask price during the event window. The graph employs the same vertical scale as used in

Figure 5-3 to enable easy comparisons. The corresponding z-scores for the last three minutes of trading and the first three minutes of market opening are calculated in Appendix 5-4. This shows that the end-of-day ask price does not experience a significant change; however, its overnight return has seen a significant increase at 10:01 followed by some small reversals, and then ask price maintains for the rest of the day.

Figure 5-7: Cumulative Abnormal Return of Ask Price

19.0%

16.0%

13.0%

10.0%

7.0%

4.0%

1.0%

-2.0%

15:21 12:19 15:01 15:11 15:31 15:41 15:51 10:09 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:29 14:39 14:49 14:59 Market close Market

Given the ability of the manipulator to enter the order almost immediately prior to the close, rational traders may arrive in the market the following day. There is also a significant increase in the frequency of ask orders in the first minute of the day following the manipulation, as shown in Panel E of Table 5-7. Figure 5-8 also plots the ask frequency ratio against the benchmark for the whole event window.

207

Figure 5-8: Ask Order Frequency

6.0 5.0

4.0

3.0 Ratio 2.0 1.0

0.0

10:09 15:01 15:11 15:21 15:31 15:41 15:51 10:19 10:29 10:39 10:49 10:59 11:09 11:19 11:29 11:39 11:49 11:59 12:09 12:19 12:29 14:39 14:49 14:59 Market close Market

The graph shows that the ask order frequency near the market close during the manipulation days is below the benchmark average, but increases during the first hour of trading on the following day. In conjunction with ask price behaviour, these results indicate that there is some evidence of an increase in selling interests at the following open.

Figure 5-9 shows the abnormal ask depth around the manipulation date. The ask depth on the date of manipulation is less than half the benchmark period. This is consistent with manipulation occurring in stocks that have low liquidity and are potentially missing any ask side demand. After the manipulation occurs, the depth on the ask side increases to levels slightly higher than those observed during the benchmark period. The ask depth then gradually decreases, indicating the decline in stock supply.

208

Figure 5-9: Abnormal Ask Depth

0.5

11:19 11:59 15:01 15:11 15:21 15:31 15:41 15:51 10:09 10:19 10:29 10:39 10:49 10:59 11:09 11:29 11:39 11:49 12:09 12:19 12:29 14:39 14:49 14:59

-0.5

Market close Market Natural Natural Logarithm

-1

-1.5

These results may provide some evidence of increased selling interests, in the form of increased ask price, ask frequency and depth at the open following the manipulation. However, since no significant increase in volume and price reversal is observed at the open, the evidence of counteraction by rational traders is not clear. It is also suspected that the increased selling interest may be due to manipulators’ desire to sell stocks at inflated prices. If this is the case, the short- term manipulation profits may be quick but not large, due to illiquidity and the low price of manipulated stocks.

5.7.3 Long-run Effects of Quote-based Manipulation

The intraday results in the previous sections show that the manipulator performs a pump-and-dump manipulation scenario through manipulating the bid side at the market close. The manipulator buys at higher prices, giving a misleading appearance of a high demand for the manipulated stock. Although this false signal does not create enough credibility to mislead other market participants (such as

209

information seekers), the manipulation is successful due to the market structure of closing price procedure on HKEx.

Hanson and Oprea (2009) argue that since manipulative traders are in essence

‘noise’ traders without private information, their manipulation attack should not have a discernible effect on prices, except during the manipulation. They expect that the market will correct such temporary price distortions and the price will finally revert to the underlying values in the long-run. Aggarwal and Wu (2006) show that the stock price declines after the manipulation ends, though manipulated prices do not fully revert to their pre-manipulation levels even one year post-manipulation.

To examine whether a manipulation attack by an uninformed trader will only result in a temporary distortion in prices, the daily behaviour of price and volume is further examined for manipulated stocks over 120 trading days (about six months) following the manipulation. Daily abnormal price return and abnormal trading volume are further calculated based on the mean-adjusted model (see equations 5.5 and 5.11). The expected value is estimated over a 100-day estimation period which is from 115 through 15 days prior to the manipulation.

Figure 5-10 plots average cumulative abnormal stock price returns from day 5 to day120.

210

Figure 5-10: Daily Cumulative Abnormal Return of Stock Price

85%

70%

55%

40%

25%

10%

-5% 5

-5

10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

105 110 115 120 Event Days 100

Manipulation

The graph shows that the stock prices on average do not revert back below their pre-manipulation level; instead they increase gradually through time. The cumulative return of the stock price is more than 80% six months post- manipulation. To examine whether this price increase is a market-wide phenomenon, Figure 5-11 compares the average return of the manipulated stocks with the constituents of the HSI from five days before the manipulation to 120 days post-manipulation.

Figure 5-11: Daily Price Returns of Manipulate Stocks Compared to HSI Stocks

13%

-5

95 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90

105 110 115 120 -2% 100 Event Days Stock Return Market Return Manipulation

211

It appears that the abnormal returns recorded by the manipulated stocks are not a result of market-wide influences, with the returns of manipulated stocks significantly in excess of the HSI constituents for a majority of the six months following the manipulation.

Figure 5-12 plots abnormal daily trading volume to further examine whether an increased price is accompanied by an increased trading volume. Apparently, the graph shows that the average trading volume of manipulated stocks increases gradually 10 days after the manipulation. By day 120, the abnormal trading volume measure is 1.69. This shows that the actual trading volume is about 5.4 times the expected trading volume (Exp(1.69)=5.4).

Figure 5-12: Daily Abnormal Trading Volume

2.5

1.5 1 0.5

-5

55 10 15 20 25 30 35 40 45 50 60 65 70 75 80 85 90 95

100 105 110 115 120

Natural Natural Logarithms -0.5

-1 Event Days Manipulation

This daily evidence shows the long-run manipulation effects, which may enable manipulators to make a handsome profit from manipulating small and illiquid stocks.

212

5.8 Implications

Given the large number of illiquid stocks on HKEx and the findings of Comerton-

Forde and Putniņš (2011) that a large amount of manipulation goes unprosecuted, it is suspected that the practice of quote-based manipulation may still be prevalent in Hong Kong. This section analyses the existence of potential manipulations in the current market climate of HKEx, and discusses the potential solutions for the trading of illiquid securities.

5.8.1 The Persistent Risk of Quote-based Manipulation in HKEx

This chapter analyses all 1464 stocks listed in 2011, a period for which no manipulation cases were pursued by the HKSFC. Since quote-based manipulation is carried out through placing high bids at the market close without intention to trade, the bid return and trade execution are thus calculated at 16:00, the last minute of trading. The number of stocks, stock-days and trade execution are counted when the end-of-day last minute bid return is greater than 5%, 10% and

20%, respectively. There is a total of 392 stocks involving 916 stock-days, which exhibit end-of-day bid returns in excess of 5%. These cases of potential manipulation are then categorised into the eight price levels.

Table 5-8 shows that the unusual bid returns are concentrated in lower-priced stocks and trade execution in the last minute for those stocks is quite low. Similar patterns are observed when bid returns are greater than 10% and 20%.

213

Table 5-8: Suspected Instances of Quoted-based Manipulation This table analyses all 1464 stocks in 2011. It counts the number of stocks, stock-days, trade execution, when the day-end bid return at 16:00 is greater than 5%, 10% and 20%, respectively. There are 392 stocks with day-end returns greater than 5%.

Bid Return (at 16:00) >=5% Bid Return (at 16:00) >=10% Bid Return (at 16:00) >=20% Price Level Minimum No. of No. of Proportion Stock- Trade Execution No. of Stock- Trade Execution No. of Stock- Trade Execution (HK$) Tick (HK$) Stocks (A) Stocks (B) (B/A) days (per 100 orders) Stocks days (per 100 orders) Stocks days (per 100 orders)

0.01 - 0.25 0.001 294 124 42% 367 21 60 116 11 20 26 8

0.25 - 0.50 0.005 211 86 41% 180 20 31 47 20 3 5 15

0.50 - 1.00 0.010 208 66 32% 184 23 33 75 18 7 25 8

1.00 - 2.00 0.010 253 68 27% 122 25 23 36 18 6 11 23

2.00 - 5.00 0.010 253 33 13% 46 33 6 7 18 1 1 39

5.00 - 10.00 0.010 102 8 8% 9 50 1 1 100 0 0 0

10.00 - 20.00 0.020 76 3 4% 4 21 1 1 33 0 0 0

20.00 -100.00 0.050 67 4 6% 4 7 1 1 0 0 0 0

Total 1464 392 27% 916 156 284 37 68

214

These stocks represent approximately 39% of the total number of stocks within the sub-HK$1 range. Among stocks trading above HK$5, few have last minute bid returns greater than 5%. This is consistent with active markets for these less liquid stocks, preventing quote-based manipulation from being as effective as traditional trade-based closing price manipulation.

The continued and widespread existence of potential closing price manipulation during 2011 demonstrates that the problem of quote-based manipulation is ongoing for HKEx-listed securities. The problem is exacerbated by the fact that

49% of the stocks listed on HKEx have prices below HK$1.00 and are relatively illiquid.

5.8.2 Potential Solutions for Illiquid Securities

Easley et al. (1996) indicate that the risk of information-based trading is higher for small and inactively traded stocks, leading to a lack of uninformed traders.

Continuous trading relies on limit orders to provide liquidity; a trader placing a limit order is equivalent to writing a free option to informed traders, and he may end up trading at an unfavorable price. This adverse selection problem combined with wide bid-ask spreads discourages uninformed traders from participating in small illiquid stocks. On the other hand stocks with high volume tend to have a high probability of information events and higher arrival rates of informed traders; however, these are more than offset by the higher arrival rate of uninformed traders. Easley et al. (1996) refer to this as “the almost universal failure of screen trading for inactive stocks”.

215

To overcome this problem, Easley et al. (1996) suggest employing different trading mechanisms for active and inactive stocks. While active stocks trade in continuous auctions, inactive stocks trade periodically through non-continuous call auctions. Indeed, numerous studies on call auctions have shown that they improve price discovery and reduce price manipulation. The use of call auctions increases the number of participants and concentrates liquidity at a single point in time, achieving an asset equilibrium price. Madhavan (1992) indicates that call auctions aggregate information efficiently and are more robust to problems of information asymmetry in that they can operate where continuous markets fail. An empirical study by Comerton-Forde et al. (2007) on the Singapore Stock

Exchange shows that the introduction of closing call auctions has reduced the incidence of closing price manipulation.

Many European exchanges have started using non-continuous call auctions for thinly traded stocks, while maintaining continuous trade for actively traded stocks.

In England Tradepoint offers periodic auctions of less active stocks, and so does the Paris Bourse (now Euronext Paris). Lauterbach (2001) reports that in Germany some small-firm stocks have a single daily auction; while on the Tel Aviv Stock

Exchange (TASE), stocks with high trading activity are traded on a continuous trading system (V-method), and more illiquid stocks are traded in a call-auction phase (C-method). Illiquid stocks are updated once every three months. Stocks that demonstrate high trading activity on the auction system will enter into the continuous trading system, while stocks that exhibit persistent low trading activity on the continuous trading system will move back to the auction method.

Lauterbach (2001) finds that illiquid stocks show improved liquidity after moving

216

to a daily call auction in TASE.

Exchanges operating separate trading methods for liquid and illiquid securities include the London Stock Exchange (LSE) and NYSE Euronext. LSE has several electronic platforms on which the different types of security trade. For example,

SETS (Stock Exchange Electronic Trading Service) trades only liquid securities, including indexed securities, Exchange Traded Funds, Exchange Trading Products and London Standard Listed securities. SETSqx (Stock Exchange Electronic

Trading Services—quotes and crosses) is a trading platform for securities less liquid than those traded on SETS. NYSE Euronext employs Universal Trading

Platform, where the most liquid equities are traded continuously throughout the day, while less liquid equities are traded at call auctions at 11:30 and 16:30 each day.

The use of market-makers and specialists has also been introduced to assist the trading of thinly traded stocks. For example, the London Stock Exchange SEAQ trading mechanism for small and mid-cap stocks in AIM securities (not traded on

SETS or SETSqx) is also employing market-maker services. NYSE Euronext use market-makers to support liquidity for small and mid-cap securities traded continuously or by auction. More recently, the Australian Stock Exchange has made proposals to introduce equity market-makers for small to mid-caps. The contribution of market-makers to the liquidity and efficiency of stocks has been well-documented in many studies (e.g., Kehr et al., 1998; Benveniste et al., 1992).

Lauterbach (2001) argues that call auctions and market-maker assisted trade for

217

thinly traded stocks provide a partial remedy for the asymmetric information problem.

The European experience shows that, for illiquid stocks, continuous price trading would probably not provide the same benefit as it does for liquid stocks. The less liquid securities are in most need of additional assistance – from daily auctions to the presence of market-makers. Such systems, if applied in HKEx, could help to increase liquidity and reduce the ability of manipulators to commit costless closing price manipulation.

5.9 Summary

This chapter examines trading mechanism effects on price manipulation. It complements the prior literature by investigating quote-based manipulation on

HKEx, a new form of manipulation using prevailing quotes in the absence of trade.

Manipulators are able to influence closing prices by placing these bids immediately prior to the close of the market, increasing prices without incurring any actual costs. Based on a sample of 123 prosecuted cases of closing price manipulation on HKEx from 1999 to 2006, this chapter shows that the closing price mechanism of HKEx encourages this unique manipulation technique. The manipulated stocks are typically small, and not actively traded at the time of manipulation. Manipulators places high bids, creating the appearance of increased demand for the manipulated stocks.

218

The intraday analysis shows that quote-based manipulation can result in high end- of-day price returns, bid returns, and bid order frequency. This is accompanied by low liquidity, as well as limited price reversion at the open following the manipulation. Although the presence of information seekers does not seem necessary for manipulators to inflate prices, they are necessary if shares are to be sold at a profit after the manipulation, closely resembling a typical pump-and- dump strategy. The existence of information seekers is evident when examining the six-month performance of daily stock prices and trading volume. Results show that the manipulated prices do not revert to their initial values, instead increasing gradually, enabling manipulators to make a handsome profit in the long term.

Given that HKEx has a high proportion of penny stocks, the end-of-day bid returns and trade execution for all 1464 stocks listed during 2001 are analysed. It is found that there is a significant amount of potential quote-based manipulation occurring on HKEx, despite the complete lack of prosecution by the regulator.

The anomalous last minute returns are concentrated in low-priced and illiquid stocks. The existence of such potential manipulations is supported by the work of

Comerton-Forde and Putniņš (2011) who find that the majority of closing price manipulations go unprosecuted.

Lastly, this chapter documents non-continuous call auctions/market-makers solution, which have been used in many European exchanges including LSE and

Euronext, and which Lauterbach (2001) has shown to be successful at increasing liquidity and reducing volatility. Comerton-Forde et al. (2007) have additionally shown that the introduction of call auctions can reduce the incidence of closing

219

price manipulation. The introduction of either or both of these mechanisms on

HKEx would significantly increase the cost of such manipulation, and in so doing likely reduce its occurrence.

The results reported in this chapter have important regulatory implications for designing market structure. A thorough understanding of the link between the trading mechanism and price formation process is extremely important for securities exchanges and regulators in designing a fair and efficient marketplace.

Clearly, one model is not able to characterise all forms of manipulation, while one trading structure does not fit all types of stocks.

220

References

Aggarwal, R.K., and Wu, G., 2006. Stock Market Manipulations. The Journal of

Business, 79(4), pp.1915-1953.

Ahn, H.-J., Bae, K.-H., and Chan, K., 2001. Limit Orders, Depth, and Volatility:

Evidence from the Stock Exchange of Hong Kong. The Journal of Finance, 56(2), pp.767-788.

Allen, F., and Gale, D., 1992. Stock-Price Manipulation. The Review of Financial

Studies, 5(3), pp.503-529.

Ashenfelter, O., and Card, D., 1985. Using the Longitudinal Structure of Earnings to Estimate the Effect of Training Programs. The Review of Economics and

Statistics, 67(4), pp.648-660.

Benveniste, L.M., Marcus, A.J., and Wilhelm, W.J., 1992. What’s special about the specialist? Journal of Financial Economics, 32(1), pp.61-86.

Brown, S.J., and Warner, J.B., 1980. Measuring security price performance.

Journal of Financial Economics, 8(3), pp.205-258.

Campbell, C.J., and Wasley, C.E., 1996. Measuring Abnormal Daily Trading

Volume for Samples of NYSE/ASE and NASDAQ Securities Using Parametric and Nonparametric Test Statistics. Review of Quantitative Finance & Accounting,

6(3), pp.309-326.

221

Cushing, D., and Madhavan, A., 2000. Stock returns and trading at the close.

Journal of Financial Markets, 3(1), pp.45-67.

Comerton-Forde, C., Ting Lau, S., and McInish, T., 2007. Opening and closing behavior following the introduction of call auctions in Singapore. Pacific-Basin

Finance Journal, 15(1), pp.18-35.

Comerton-Forde, C., and Putniņš, T.J., 2011. Measuring closing price manipulation, Journal of Financial Intermediation, 20(2), pp.135-158.

Easley, D., Kiefer, N.M., O’Hara, M., and Paperman, J.B., 1996. Liquidity,

Information, and Infrequently Traded Stocks. The Journal of Finance, 51(4), pp.1405-1436.

Gerard, B., and Nanda, V., 1993. Trading and Manipulation Around Seasoned

Equity Offerings. The Journal of Finance, 48(1), pp.213-245.

Hanson, R., Oprea, R., and Porter, D., 2006. Information aggregation and manipulation in an experimental market. Journal of Economic Behaviour &

Organization, 60(4), pp.449-459.

Hanson, R., and Oprea, R., 2009. A Manipulator Can Aid Prediction Market

Accuracy. Economica, 76(302), pp.304-314.

Harris, L., 1989. A Day-End Transaction Price Anomaly. The Journal of

Financial and Quantitative Analysis, 24(1), pp.29-45.

222

Harris, L., 2003. Trading and Exchanges: Market Microstructure for Practitioners,

Oxford University Press, New York.

Hayek, F.A., 1945. The Use of Knowledge in Society. The American Economic

Review, 35(4), pp.519-530.

Hillion, P., and Suominen, M., 2004. The manipulation of closing prices. Journal of Financial Markets, 7(4), pp.351-375.

Kehr, C., Krahnen, J., and Theissen, E., 1998. The anatomy of a call market:

Evidence from Germany. Working paper, Johann Wolfgang Goethe University,

Frankfurt.

Kumar, P., and Seppi, D.J., 1992. Futures Manipulation with “Cash Settlement”.

The Journal of Finance, 47(4), pp.1485-1502.

Lauterbach, B., 2001. A note on trading mechanism and securities’ value: The analysis of rejects from continuous trade. Journal of Banking & Finance, 25(2), pp.419-430.

Madhavan, A., 1992. Trading Mechanisms in Securities Markets. The Journal of

Finance, 47(2), pp.607-641.

O’Hara, M., 1995. Market Microstructure Theory, Blackwell Publishing,

Cambridge.

223

Selvanathan, A., Selvanathan, S., Keller, G., and Warrack, B., 2006. Australian

Business Statistics, Thomson, Australia.

224

Appendices

Appendix 5-1: Summary of Quote-based Closing Price Manipulation Cases

Cases News release Stock Single- Manipulator Position Quote-based Manipulation Scheme date Code board-lot 1 5-Dec-00 0181 2000 Wang Fang Investor Placing 13 consecutive single-board-lot bid orders, which were not executed, pushing up the closing price from $0.140 to $0.192. 2 20-May-03 0487 2000 N/A Floor trader Placing single-board-lot bid orders at prices 2% to 23% higher than the prevailing market price, in order to sell the stocks for his sister-in-law at higher prices on the following days. 0224 2000 Poon Lak To Retail investor Placing single-board-lot bid orders at prices 9% to 18% higher than the prevailing market price in order to reduce margin deposit. 3 10-Jun-03 0439 2000 Choi Kam Tui Investor Placing single-board-lot bid orders at prices 4% to 150% higher than the prevailing market price, in order to sell the shares at higher prices. 4 29-Jul-03 0205 2000 N/A N/A Placing seven single-board-lot orders at prices six to eight spreads (ticks) higher than the prevailing market price, in order to sell the shares at higher prices on the next day. 5 5-Sep-03 0544 20000 Lam Yat Wa Retail investor Placing single-board-lot orders on a number of days in June and July 2001 to buy shares at high 0725 2000 prices , pushing the closing price up in order to sell these stocks at higher prices eventually. 0385 10000 6 9-Jun-04 0542 2000 N/A N/A Placing a series of single-board-lot bid orders at high prices, pushing the closing price 8% to 60% higher than the previous close. 7 15-Nov-04 8163 4000 N/A Dealer Placing a series of purchase orders at a high price, causing the closing price to rise by 30%.

8 12-Jan-06 8065 10000 Cheung Wan Investor Placing a series of single-board-lot bid orders at high prices, causing the closing price of Innovis Chiu shares to rise by 16%, and KanHan shares by 26%. 8175 10000 9 26-Apr-06 0529 2000 Wong Wei Yin Former broker Placing single-board-lot bid orders at high prices, pushing up the share price by 8% to 10%, in order to dispose of the shares at higher prices on the following days. 10 9-Mar-07 856 8000 Chaw Chi Wai Retail investor Placing buy orders at a price higher than the prevailing market price near the market close, resulting in 938 2000 Chaw’s orders becoming the closing price. 11 26-Jul-07 986 2000 Cheng Ngai Retail investor Placing a number of single-board-lot orders near the market close to establish a higher closing price, in 567 2000 order to sell the shares at a higher price. 12 3-Jan-08 759 2000 Wong Win Hing Retail investor Placing single-board-lot orders near the market close, in order to establish higher closing prices. 8037 5000 704 10000 13 12-Jun-08 8019 5000 Leung Kam Lai Retail investor Placing single-board-lot buy orders in the last minutes of trading for the five stocks to inflate their 8171 4000 closing prices, so as to dispose of the shares at higher prices on the following trading days. 8182 6000 8239 5000 8250 8000 14 7-Jul-08 8009 2000 Ng Yu Hon Licensed Placing orders to buy small quantities of I Merchants shares at prices higher than the prevailing market representative prices, in order to inflate the closing price of the stock.

225

Appendix 5-2: Impact of Quote-based Manipulation on Trading Behaviour Using Benchmark Analysis This table reports results for the last three minutes of trading and the first three minutes of market opening based on the benchmark approach. The corresponding z-scores are calculated in parentheses. Trade frequency and trading volume in manipulated days are calculated as ratios of the 100-day benchmark average. *** indicates significance at 1% level, ** indicates significance level at 5%, and * indicates significance level at 10%.

Manipulation Following Opening

Variable 15:58 15:59 16:00 10:01 10:02 10:03 Mean Value Standard Deviation

Abnormal Stock Price Return -0.03% 0.13% 13.72% -0.05% -0.03% -0.03% 0.04% 0.009 (-0.1) (0.1) (15.3)*** (-0.1) (-0.1) (-0.1) Abnormal Trading Volume -0.25 -0.27 -0.14 -0.08 -0.07 0.00 -0.04 0.042 (natural log) (-5.0)*** (-5.3)*** (-2.3)** (-1.0) (-0.6) (0.9) Trade Frequency Ratio 0.00 0.39 0.73 0.24 0.00 0.91 0.36 0.360 (-1.0) (0.1) (1.0) (-0.3) (-1.0) (1.5)

226

Appendix 5-3: Impact of Quote-based Manipulation on Bid Behaviour Using Benchmark Analysis This table reports results for the last three minutes of trading and the first three minutes of market opening based on the benchmark approach. The corresponding z-scores are calculated in parentheses. All ratio variables are calculated as ratios of the 100-day benchmark average. *** indicates significance at 1% level, ** indicates significance level at 5%, and * indicates significance level at 10%.

Manipulation Following Opening Variable 15:58 15:59 16:00 10:01 10:02 10:03 Mean Value Standard Deviation

Abnormal Bid Price Return 2.81% 4.89% 6.79% -2.51% -0.57% -0.53% 0.05% 0.006 (4.5)*** (7.8)*** (10.9)*** (-4.1)*** (-1.0) (-0.9) Bid Frequency Ratio 2.50 3.76 5.75 1.02 1.92 0.71 0.71 0.629 (2.9)*** (4.8)*** (8.0)*** (0.5) (1.9)* (0.0) Abnormal Bid Depth -0.14 -0.74 -1.44 0.27 0.40 0.45 0.27 0.216 (natural log) (-1.9)* (-4.6)*** (-7.9)*** (0.0) (0.6) (0.9) Spread Ratio 0.98 0.71 0.24 0.59 0.66 0.71 0.98 0.148 (-0.7) (-1.9)* (-3.5)*** (-1.4) (-1.1) (-1.0)

227

Appendix 5-4: Impact of Quote-based Manipulation on Ask Behaviour Using Benchmark Analysis This table reports results for the last three minutes of trading and the first three minutes of market opening based on the benchmark approach. The corresponding z-scores are calculated in parentheses. All ratio variables are calculated as ratios of the 100-day benchmark average. *** indicates significance at 1% level, ** indicates significance level at 5%, and * indicates significance level at 10%.

Manipulation Following Opening

Variable 15:58 15:59 16:00 10:01 10:02 10:03 Mean Value Standard Deviation

Abnormal Ask Price 0.25% 0.12% 0.28% 3.20% 0.33% -0.19% 0.02% 0.002 Return (1.0) (0.4) (1.1) (13.7)*** (1.4) (-0.9) Ask Frequency Ratio 0.47 0.78 0.91 1.27 0.74 1.57 0.61 0.517 (-0.3) (0.3) (0.6) (1.3) (0.2) (1.9)* Abnormal Ask Depth -0.88 -0.85 -0.75 0.55 0.53 0.56 -0.07 0.467 (natural log) (-1.8)* (-1.7)* (-1.5) (1.3) (1.3) (1.3)

228

CHAPTER 6

CONCLUSION

This thesis consists of three studies and is motivated by a desire to offer an original way of examining three main themes in market microstructure: liquidity, price discovery and price manipulation. It aims to present a comprehensive analysis of these issues of significance under various market structures and to provide empirical evidence to exchange regulators in designing trading mechanisms. This chapter summarises the main points that can be drawn from this thesis about: (1) how anonymous trading affects liquidity migration and (2) the price discovery process in the trading of cross-listed stocks; (3) how trading mechanism affects price manipulation; and (4) suggestions for future research.

This chapter ends with the implications for economic growth and policy.

6.1 How Anonymous Trading Affects Liquidity Migration of Cross-listed

Stocks

Chapter 3 examines liquidity migration of cross-listed stocks before and after the change to the broker anonymity regime. While numerous studies explore the liquidity impact of anonymous trading with mixed and anecdotal evidence, the liquidity migration effect of the rule change has not previously been considered.

The staggered removal of trader identifiers on the Australian Stock Exchange

(ASX) and New Zealand Stock Exchange (NZX) provides an ideal natural setting

229

to examine liquidity change in the trading of cross-listed stocks on both markets simultaneously. Chapter 3 also takes into account endogeneity, using the instrumental variables two-stage least squares (2SLS) technique, controlling for both stock-specific and market-wide factors. It is found that ASX’s introduction of anonymous trading attracts the trading of cross-listed stocks from NZX to ASX, in the form of a significant improvement of bid-ask spreads and quoted depth on

ASX, and a significant deterioration of liquidity on NZX. On the other hand, when considering NZX’s adoption of anonymous trading, liquidity migrates from

ASX to NZX in the trading of cross-listed stocks. These results suggest that trader anonymity attracts liquidity migration of cross-listed stocks from the foreign counterpart and yields significant benefits to both exchanges. Chapter 3 adds to the limited literature on the effects of broker anonymity by providing additional evidence using a natural experiment on two exchanges. It also contributes to the ongoing debate as to whether or not trader anonymity can benefit exchanges in terms of market liquidity.

6.2 How Anonymous Trading Affects the Price Discovery Process

Chapter 4 further examines anonymity effects from the price discovery aspect by utilising the natural experiment on ASX and NZX. Although there have been extensive studies on anonymity and price discovery separately, the link between the two concepts is not clear. Chapter 4 bridges a research gap between previous work on anonymity and price discovery, by directly observing the conduct of market traders, given the choice between transparent and anonymous markets.

230

Using Minspan synchronous transactions data on New Zealand and Australian cross-listed stocks, estimations with both the Hasbrouck (1995) information share and the error correction model of Harris et al. (1995) demonstrate clearly that trader anonymity improves the price discovery process. Information share improves on ASX, but deteriorates on NZX, after ASX switched to anonymous trading. On the other hand, information share increases on NZX, but decreases on

ASX, after NZX adopted anonymous trading. Consistent with the findings presented in the previous chapter, Chapter 4 concludes that the adoption of an anonymity regime has yielded significant benefits for both ASX and NZX in the trading of cross-listed stocks.

6.3 How Trading Mechanism Affects Price Manipulation

Chapter 5 of this thesis focuses on the impact of trading mechanism on price manipulation in the Hong Kong Stock Exchange (HKEx). Microstructure research suggests the importance of understanding how the market mechanism works and how it affects the price creation process; however, the existing studies on price manipulation have not specifically considered the role of the trading mechanism.

Chapter 5 therefore complements market manipulation literature by providing initial evidence of quote-based manipulation, a costless form of manipulation facilitated by the unique closing price procedure of HKEx. This involves the placing of orders to buy small quantities of shares at prices higher than the prevailing best bid near the market close. Quote-based manipulation differs significantly from the typical manipulation schemes documented in existing

231

studies, such as Hillion and Suominen (2004) and Comerton-Forde and Putniņš

(2011). Results show that quote-based price manipulation is associated with increased closing prices, bid returns, bid order frequency, and decreased bid-ask spreads immediately prior to the close of the market. The stocks that are most susceptible to this kind of manipulation are those with low trading volume and depth. This manipulation scheme is consistent with the fact that under the current closing procedure in HKEx, it is likely to succeed for illiquid stocks. These results have important regulatory implications for the design of market structure.

Implementing non-continuous call auctions and/or market-makers, as practised in

Euronext and London Stock Exchange, may be necessary for the trading of illiquid securities in order to minimise this kind of manipulative conduct and thereby enhance the accuracy of the price creation process.

6.4 Avenues for Future Research

Several potential further research directions can be drawn from this thesis. First, in terms of the impact on liquidity of anonymous trading, as examined in Chapter

3, further evidence from other exchanges is called for to validate and extend these findings, considering that the results in Chapter 3 may be potentially affected by the global financial crisis, starting in August 2007. Second, the examination of anonymity effects on price discovery in Chapter 4 is a new area, raising two issues for future research. One area of concern with this analysis is the classification of cross-listed stocks into two categories: stocks with more active trading, and stocks with less active trading in the foreign market compared to the

232

home market. It would also be interesting to see whether the results for the limit order book systems of the Australian and New Zealand stock exchanges can be extended to other markets. Finally, the results in Chapter 5 provide original evidence regarding the trading procedure and quote-based price manipulation in the Hong Kong market. Future research on how markets can be structured to improve the integrity of trading in small stocks and reduce instances of manipulation would be valuable.

6.5 Implications for Economic Growth and Policy

An important theme in this thesis, and microstructure literature in general, is how specific trading arrangements affect market liquidity and thus the price formation process. Market structure is fundamental to all trading. As indicated by O’Hara

(2001), the behaviour of prices and even the capability of markets depend on the ability of the trading structures to match the trading desires of sellers and buyers.

On the micro level, the structure of trading matters to investors. To trade effectively, investors need to know the structure of every market in which they trade, because having different trading strategies for different market conditions is a crucial factor of trading success. Market structure also matters to practitioners and market microstructure researchers when translate their analysis into applicable methods. From a policy perspective, regulators must understand how market design change affects transaction cost, price discovery, market manipulation and insider trading, in order to promote market quality and foster competition for the supply of liquidity.

233

On the macro level, there are wider implications for the economy as a whole. An influential study by King and Levine (1993) presents evidence on the relationship between the financial system and economic development. They identify a strong and statistically significant link between various financial liquidity and depth measures and economic growth. Levine (1997) also shows that the structure of financial markets can promote economic growth and argues that less-developed countries can accelerate their growth rates by changing the structure of their financial systems. Levine also reports that Germany’s more rapid economic growth rate than the United Kingdom during the latter half of the 19th century and the first decade of the 20th century is simply because of a well-established financial structure, which allowed for more efficient capital allocation and more effective corporate operation. Allen and Gale (2000, 2007) develop models for understanding the characteristics of financial market structure that may result in financial crises. They indicate that a well-developed, smoothly-operating financial market play an important role in contributing to financial stability, and hence the health and efficiency of an economy.

Given the importance of market structure from both the micro and macro point of view, as well as the findings presented in this thesis, four ways to improve trading structure can be suggested. First, it is recommended to provide a legal system which protects investors and encourages information sharing, and communications technology which enables public investors to be equally informed. Rajan and Zingales (2001) indicate that the informational structure is important because it enables economies to become more transparent and thus attract more capital from investors. Given the severe information asymmetry, in

234

particular for illiquid stocks, the second suggestion is to develop an appropriate trading mechanism for the trading of small-cap and illiquid securities so as to allow traders to capture liquidity in low-volume situations. As practised in many

European exchanges (e.g., Euonext) and the London Stock Exchange, separate trading methods can be developed for the trading of liquid and illiquid securities.

Moreover, Lauterbach (2001) suggests the use of market-makers and specialists, which is another effective way to support liquidity for the trading of thinly traded stocks. As discussed in Chapter 5, assistance to the trading of illiquid stocks can improve the provision of liquidity, and also reduce the risk of stock prices being manipulated. The third suggestion is to reduce the prevalence of price manipulation by developing an appropriate surveillance system that is able to improve the accuracy of detection methods and aid market participants to identify manipulation. It is also recommend that exchange regulators need to consider the time, market condition, types of securities, and types of orders permitted in a trading mechanism when formulating market manipulation rules. Finally, microstructure research requires not only an understanding of price behaviour, but also of how different trading protocols affect price formation. In Hayek’s (1945) words, “we must look at the price system as such a mechanism for communicating information if we want to understand its real function ... The most significant fact about this system is the economy of knowledge with which it operates, or how little the individual participants need to know in order to be able to take the right action”.

235

References

Allen, F., and Gale, D., 2000. Comparing Financial Systems, MIT press, London.

Allen, F., and Gale, D., 2007. Understanding Financial Crises. Clarendon

Lecture Series in Finance, Oxford University Press, UK.

Comerton-Forde, C., and Putniņš, T.J., 2011. Measuring closing price manipulation. Journal of Financial Intermediation, 20(2), pp.135-158.

Hayek, F.A., 1945. The Use of Knowledge in Society. The American Economic

Review, 35(4), pp.519-530.

Hillion, P., and Suominen, M., 2004. The manipulation of closing prices. Journal of Financial Markets, 7(4), pp.351-375.

Harris, F.H., McInish, T.H., Shoesmith, G.L., and Wood, R.A., 1995.

Cointegration, Error Correction, and Price Discovery on Informationally Linked

Security Markets. The Journal of Financial and Quantitative Analysis, 30(4), pp.563-579.

Hasbrouck, J., 1995. One Security, Many Markets: Determining the Contributions to Price Discovery. The Journal of Finance, 50(4), pp.1175-1199.

King, R.G., and Levine, R., 1993. Finance and Growth: Schumpeter Might be

Right. The Quarterly Journal of Economics, 108(3), pp.717-737.

236

Levine, R., 1997. Financial Development and Economic Growth: Views and

Agenda. Journal of Economic Literature, 35(2), pp.688-726.

Lauterbach, B., 2001. A note on trading mechanism and securities’ value: The analysis of rejects from continuous trade. Journal of Banking & Finance, 25(2), pp.419-430.

O’Hara, M., 2001, Designing Markets for Developing Countries. International

Review of Finance, 2(4), pp.205-215.

Rajan, R.G., and Zingales, L., 2001. Financial Systems, Industrial Structure, and

Growth. Oxford Review Economic Policy, 17(4), pp.467-482.

237