Security Market Manipulations and the Assurance of Market Integrity

Shan Ji

Supervisor: Professor Mike Aitken

Co-supervisor: Professor Frederick H. deB. Harris

This thesis is presented for the degree of Doctor of Philosophy in Finance At The University of New South Wales

Banking and Finance Australian School of Business 2009

Abstract

This dissertation is motivated by two major factors. First, there have been no direct studies conducted for the relationship between market integrity and market efficiency and the driving forces behind the cross-sectional variations in market quality. Second, a better understanding the relationships among market integrity, market efficiency and other mechanism design factors for securities exchanges will facilitate securities exchanges achieve a satisfactory level of market quality.

This dissertation consists of three chapters. In Chapter 1, a review of literature on market manipulation will be given. A series of common securities market manipulation strategies and corresponding market surveillance alerts will be explained and defined.

In Chapter 2, we develop a testable hypothesis that market manipulation as proxied by the incidence of ramping alerts would raise transaction cost for completing larger trades. We find ramping alert incidence positively related to effective spreads in 8 of 10 turnover deciles from most liquid to thinnest-trading securities. The magnitude of the increase in effective spreads when ramping manipulation incidence doubles is economically significant, 30 to 40 basis points in many moderate liquidity deciles. This compares with an average effective spread of 72 basis points for index-listed securities in the most efficient electronic markets worldwide.

In Chapter 3, In Chapter 3 of this thesis, we test the correlation between the levels of market integrity as proxied by the incidence of ramping alerts and a combination of proxies for factors from the following four potential drivers deciding the market quality across securities exchanges: • Securities Markets Trading Regulations • Securities Markets Technologies

• Securities Market Infrastructure • Securities Market Participants

The model we developed to test the correlation between the proxies for level of market integrity and seven proxies for the four potential drivers were estimated with Ordinary Least Square (OLS) and Two-stage Least Square (2SLS) error structures assumed, respectively to learn the most about the possible endogeneity of spreads and volatility. By performing Hausman-Wu specification tests, we concluded that simultaneity bias in the thickly-traded deciles is not material for the AI-Volatility and AI-Spread equation pairs. Subsequently, we used the PROBIT model to analyse the probability of adopting RTS across the 240 securities exchange deciles and the likelihood proves to be systematically related to four determinants in our sample. Finally we estimate the structural equations to investigate possible cross-equation correlation of the disturbances with either seemingly unrelated regression (SURL) estimation.

Our findings are three-fold. Firstly, in the moderately-traded deciles, we find that the presence of a closing auction (CloseAucDum) reduces the incidence of ramping alerts. Trade-based manipulation proves more difficult when a manipulator’s counterparties can use closing auctions to unwind their intraday exposures. The RTS dummy variable is significantly positively related to alert incidence. In the absence of any panel data on the dynamic effects of adopting RTS, what we are observing in cross section is the perceived vulnerability of certain exchanges to manipulation and their consequent adoption of RTS plus the regulatory regimes required to have a salutary effect on market integrity. Second, in the moderately-traded deciles, we find that the closing auctions and more regulations in pursuit of market integrity lower quoted spreads. RTS and a regulation specifically prohibiting ramping indicate in cross- section the perceived likelihood of more ramping. Thirdly, in terms of the probability of the deployment of a real-time surveillance system, the estimations again differ by liquidity decile grouping. In the moderately-traded deciles, higher alert incidence, the presence of DMA, and higher FDI again increase the likelihood of adopting a real- time surveillance system.

iii

Our findings have a couple of policy implications for many securities exchanges in terms of market design and market surveillance. First, the exhibited relationship between alert incidence and effective spreads indicates trade-based manipulation has a significant impact on execution costs. Therefore, the prevention of securities market manipulation not only serves the indirect purpose of improving an exchange’s reputation for market integrity but also contributes directly to achieving a more efficient marketplace. Second, our results indicate that some market design changes can enhance the regulatory efforts to prevent securities market manipulations. For example, to prevent manipulators from marking the closing price, some exchanges could choose to adopt a closing auction or a random closing time, which would make manipulation more costly. Nevertheless, no securities exchange can be designed perfectly. Consequently, exchange and broker-level surveillance backed by effective regulatory enforcement is a necessary and pivotal complement to good design choices.

Table of Contents

ABSTRACT ...... II

TABLE OF CONTENTS ...... V

CERTIFICATION ...... VIII

ACKNOWLEDGEMENTS ...... IX

LIST OF DIAGRAMS ...... XI

LIST OF TABLES ...... XII

CHAPTER 1 ...... 1

MANIPULATION: A SURVEY AND LITERATURE REVIEW ...... 1 1.1 INTRODUCTION ...... 1 1.2 LITERATURE REVIEW ...... 4 1.2.1 THE EXISTENCE OF MANIPULATION IN SECURITIES MARKET ...... 5 1.2.2 THE FORMS OF SECURITIES MARKET MANIPULATION ...... 7 1.2.3 VOLATILITY, SPREAD AND SECURITIES MARKET MANIPULATIONS ...... 9 1.3 SECURITIES MARKET MANIPULATIONS ...... 12 1.3.1 RAMPING ...... 12 1.3.2 UNUSUAL EXPIRATION DAY ACTIVITY ...... 13 1.3.3 MISLEADING ORDER AND TRADING STRATEGIES ...... 13 1.3.4 WASH TRADES ...... 15 1.3.5 DERIVATIVE-UNDERLYING PRICE MANIPULATION ...... 15 1.3.6 LAYERING THE ORDERBOOK ...... 16 1.3.7 CHURNING ...... 17 1.3.8 CORNERING THE MARKET ...... 18 1.3.9 SQUEEZING THE MARKET ...... 19 1.3.10 FRONT RUNNING ...... 19 1.4 ALERTS FOR SECURITIES MARKET MANIPULATIONS ...... 21 1.4.1 INTRODUCTION ...... 21

1.4.2 THE MARKING THE CLOSE ALERT ...... 22 1.4.3 REVERSAL THE NEXT TRADING DAY ...... 23 1.4.4 THE BAIT AND SWITCH ALERT ...... 24 1.4.5 WASH TRADE ...... 24 1.4.6 DERIVATIVES/UNDERLYING PRICE MANIPULATION ...... 25 1.4.7 CORNERING THE MARKET ...... 26 1.4.8 SQUEEZING THE MARKET ...... 26 1.4.9 LAYERING THE ORDER BOOK ...... 27 1.4.10 CHURNING ...... 28 1.4.11 FRONT RUNNING ...... 29

CHAPTER 2 ...... 30

MARKET INTEGRITY AND MARKET EFFICIENCY: A CROSS-MARKET COMPARISON ...... 30 2.1 INTRODUCTION ...... 30 2.2 RESEARCH METHODOLOGY ...... 32 2.2.1 MARKET INTEGRITY DEFINED FOR THIS RESEARCH ...... 32 2.2.2 MARKET EFFICIENCY DEFINED FOR THE THIS RESEARCH ...... 34 2.2.3 RANDOM EFFECTS MODEL ...... 36 2.2.4 MODEL SPECIFICATIONS ...... 40 2.3 DATA AND MEASUREMENT ...... 42 2.3.1 DATA ...... 42 2.3.2 RAMPING ALERT INCIDENCE ...... 44 2.3.3 TIME-WEIGHTED QUOTED SPREAD AND VOLUME-WEIGHTED EFFECTIVE SPREAD ...... 47 2.3.4 DESCRIPTIVE STATISTICS ...... 49 2.3.5 LIMITATIONS OF RESEARCH DESIGN ...... 55 2.4 EMPIRICAL SPECIFICATION ...... 57 2.5 EMPIRICAL RESULTS ...... 59 2.5.1 ANNUAL AVERAGE DAILY QUOTED SPREAD FOR ALL DECILES ...... 59 2.5.2 ANNUAL AVERAGE DAILY QUOTED SPREADS (THICKLY-TRADED DECILES) ...... 62 2.5.3 ANNUAL AVERAGE DAILY QUOTED SPREADS (THINLY-TRADED DECILES) ...... 72 2.5.4 SUMMARY OF QUOTED SPREADS RESULTS ...... 82 2.5.5 ANNUAL AVERAGE DAILY EFFECTIVE SPREADS (ALL DECILES)...... 83 2.5.6 ANNUAL AVERAGE DAILY EFFECTIVE SPREADS (THICKLY-TRADED DECILES) ...... 85 2.5.7 ANNUAL AVERAGE DAILY EFFECTIVE SPREADS (THINLY-TRADED DECILES) ...... 95 2.5.8 SUMMARY OF RESULTS FOR EFFECTIVE SPREADS ...... 105 2.6 DISCUSSION OF RESULTS ...... 106 2.7 SUMMARY AND CONCLUSION ...... 108

CHAPTER 3 ...... 110

AN EMPIRICAL MODEL OF THE INCIDENCE OF SECURITIES MARKET MANIPULATION ...... 110 3.1 INTRODUCTION ...... 110 3.2 PRIOR LITERATURE ...... 113 vi

3.2.1 REGULATIONS ...... 113 3.2.2 INFORMATION ...... 114 3.2.3 TECHNOLOGY ...... 114 3.2.4 MARKET INFRASTRUCTURE ...... 116 3.2.5 PARTICIPANTS ...... 116 3.3 RESEARCH METHODOLOGY ...... 118 3.3.1 PROXY FOR MARKET INTEGRITY ...... 118 3.3.2 PROXIES FOR SECURITIES MARKET REGULATIONS ...... 118 3.3.3 PROXY FOR TECHNOLOGY ...... 119 3.3.4 PROXIES FOR SECURITIES MARKET INFRASTRUCTURE ...... 119 3.3.5 PROXIES FOR SECURITIES MARKET PARTICIPANTS ...... 120 3.3.6 MODEL SPECIFICATION ...... 121 3.4 DATA AND MEASUREMENT ...... 123 3.4.1 DATA ...... 123 3.4.2 RAMPING ALERT INCIDENCE ...... 125 3.4.3 TIME-WEIGHTED QUOTED SPREAD ...... 127 3.4.4 AVERAGE VOLATILITY ...... 128 3.4.5 AVERAGE LIQUIDITY ...... 128 3.4.6 DUMMY VARIABLES ...... 129 3.4.7 DESCRIPTIVE STATISTICS ...... 129 3.4.8 LIMITATIONS OF RESEARCH DESIGN ...... 131 3.5 EMPIRICAL SPECIFICATION ...... 132 3.6 POSSIBLE ENDOGENEITY OF ALERT INCIDENCE, VOLATILITY, AND SPREADS ...... 135 3.6.1 OLS, 2SLS AND FIML RESULTS FOR ALL DECILES ...... 135 3.6.2 OLS, 2SLS AND FIML RESULTS FOR THICKLY-TRADED DECILES ...... 140 3.6.3 OLS, 2SLS AND FIML RESULTS FOR MODERATELY-TRADED DECILES ...... 143 3.6.4 OLS, 2SLS AND FIML RESULTS FOR THINLY-TRADED DECILES ...... 146 3.7 PROBABILITY OF REAL-TIME SURVEILLANCE ...... 149 3.7.1 MODEL SPECIFICATION ...... 149 3.7.2 EMPIRICAL RESULTS ...... 150 3.8 STRUCTURAL EQUATION ESTIMATES ...... 155 3.8.1 THE SIMPLE MODEL ...... 155 3.8.2 THE FULL INTERACTION EFFECTS MODEL ...... 161 3.9 SUMMARY AND CONCLUSION ...... 165

BIBLIOGRAPHY ...... 168

APPENDIX 1 QUOTES FROM VARIOUS LEAD EXCHANGE WEBSITES ...... 172

APPENDIX 2 ANNOTATED BIBLIOGRAPHY ...... 175

vii

Certification

I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantially proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product my own work, except to the extent that assistance from others in the project’s design and conception or in style, presentation and linguistic expression is acknowledged

Shan Ji 30th March 2009

viii

Acknowledgements

If it wasn’t clear to me before, it’s certainly clear to me now that a doctorate is an enormous personal undertaking. While it is my name appears on this dissertation, this has nonetheless been a team effort, and there are a number of people and institutions who have provided me with considerably guidance and assistance and who I wish to acknowledge here.

First and foremost I express my deep appreciation to my beloved parents and wife, and Dawn and Bryce Butterwort for all they have done for me through the years. For their sustained support of my, my education, and of this dissertation.

Professor Rick Harris’ supervision of this research has been magnificent. I am profoundly grateful to Professor Harris for the incredible efforts he made guiding, educating and mentoring me through the entire Ph.D. programme.

A number of organizations have supported my Ph.D. dissertation both financially and materially. SMARTS Group International generously sponsored my scholarship in partnership with Capital Markets Cooperative Research Centre (CMCRC). I thank Mr. Lorne Chambers for his guidance and support to the Ph.D. program. I spent the entire duration of my dissertation working out of the offices of SMRATS Group, firstly as a surveillance alerts developer, then as a senior business analyst. Overwhelmingly my colleagues have been a great source of stimulation and support.

Finally, I’d like to express my deepest acknowledgements to my principal supervisor Professor Michael Aitken, for creating and leading an organisation as innovative as the CMCRC with which I could achieve so much. Michael, thank you for offering me with this precious opportunity.

List of diagrams Figure 1 Ramping ...... 13 Figure 2 Order book for the Misleading Order Strategies Scenario ...... 14 Figure 3 Illustration of a Successful Misleading Order Strategy ...... 15 Figure 4 Illustration of Derivative-Underlying Price Manipulation...... 16 Figure 5 Layering the Order book ...... 17 Figure 6 Two Scenarios of Churning ...... 18 Figure 7 Histograms for Spread Measure before and after log transform ...... 51

List of tables

Table 1 Common Alerts Data Requirements ...... 33 Table 2 List of Securities Exchanges Covered by this Research ...... 43 Table 3 Descriptive Statistics for average Quoted Spread and Effective Spread per security per year across 34 Markets for the period 2000-2005 ...... 49 Table 4 Sample Mean of Spreads by Deciles ...... 52 Table 5 Descriptive Statistics for Ramping Alerts Incidence across 34 Markets for the period 2000-2005 ...... 53 Table 6 Sample Mean of Alerts Incidence by Deciles ...... 55 Table 7 Random Effects Model Results (Quoted Spread for All Deciles) ...... 59 Table 8 Random Effects Model Results (Quoted Spread for Decile 1) ...... 62 Table 9 Random Effects Model Results (Quoted Spread for Decile 2) ...... 64 Table 10 Random Effects Model Results (Quoted Spread for Decile 3) ...... 66 Table 11 Random Effects Model Results (Quoted Spread for Decile 4) ...... 68 Table 12 Random Effects Model Results (Quoted Spread for Decile 5) ...... 70 Table 13 Random Effects Model Results (Quoted Spread for Decile 6) ...... 72 Table 14 Random Effects Model Results (Quoted Spread for Decile 7) ...... 74 Table 15 Random Effects Model Results (Quoted Spread for Decile 8) ...... 76 Table 16 Random Effects Model Results (Quoted Spread for Decile 9) ...... 78 Table 17 Random Effects Model Results (Quoted Spread for Decile 10) ...... 80 Table 18 Random Effects Model Results (Effective Spread for All Deciles) ...... 83 Table 19 Random Effects Model Results (Effective Spread for Decile 1) ...... 85 Table 20 Random Effects Model Results (Effective Spread for Decile 2) ...... 87 Table 21 Random Effects Model Results (Effective Spread for Decile 3) ...... 89 Table 22 Random Effects Model Results (Effective Spread for Decile 5) ...... 91

xii

Table 23 Effects Model Results (Effective Spread for Decile 5) ...... 93 Table 24 Effects Model Results (Effective Spread for Decile 6) ...... 95 Table 25 Effects Model Results (Effective Spread for Decile 7) ...... 97 Table 26 Effects Model Results (Effective Spread for Decile 8) ...... 99 Table 27 Effects Model Results (Effective Spread for Decile 9) ...... 101 Table 28 Random Effects Model Results (Effective Spread for Decile 10) ...... 103 Table 29 List of Securities Exchanges Covered by this Research ...... 124 Table 30 Descriptive Statistics for 24 Markets in year 2005 ...... 129 Table 31 Correlation between Dummy Variables ...... 131 Table 32 Regression Results for Potential Endogenous Equations for all deciles across 24 Securities Exchanges in 2005 ...... 136 Table 33 Regression Results for Potential Endogenous Equations for thickly-traded deciles across 24 Securities Exchanges in 2005 ...... 140 Table 34 Regression Results for Potential Endogenous Equations for moderately-traded deciles across 24 Securities Exchanges in 2005 ...... 143 Table 35 Regression Results for Potential Endogenous Equations for thinly-traded deciles across 24 Securities Exchanges in 2005 ...... 146 Table 36 PROBIT Analysis of Real-time Surveillance System Deployment ...... 151 Table 37 Structural Equation Estimates for the Simple Model ...... 156 Table 38 The Full Interaction Effects Model Estimates ...... 164

xiii

xiv

Chapter 1

Manipulation: A Survey and Literature Review

1.1 Introduction Financial markets have deeply penetrated to almost every corner of the world, from contentious policy making to peoples’ daily lives. Wall Street, NASDAQ, bear market or bull market, those terms appear with high frequency from vast media and people’s conversations. The recent Sub-prime Melt Down in the U.S. has not only severely stricken investors’ confidence but also hit hard the world economy. Multi-billions of dollars have been pumped into the banking system and problematic financial institutions for the purpose of maintaining sufficient liquidity and avoiding more collapses such as Lehman Brothers. It’s perhaps a little bit late to call for tougher regulations to save the sub-prime mortgage industry. However, a vast of academic and empirical literature has shown that well-functioning markets are directly connected to the perception of fairness of the market (La Porta, Lopez-de-Silanes, Shleifer and Vishny 1997; Bhattacharya and Daouk 2002).

When the sub-prime crisis is cured and the order of the financial markets is restored and the prosperity can be expected again. With the pain still fresh, it can be expected that investors will insist on a “fair market” which has a high level of integrity and efficiency. One crucial measure to evaluate the integrity of a securities market is the level of market misconduct. For example, insider trading and securities market manipulations are two common types of market misconduct that can be observed. Aggarwal and Wu (2006) argue that the possibility that stock markets (both developed 1

and emerging) can be manipulated is an important issue for the regulation of trade and the efficiency of the financial market. As a result, more and more securities exchanges have demonstrated their commitment to the twin goals of market efficiency and market integrity.

Securities market manipulation can be defined in many different ways. The Centre for Futures Education defines market manipulation as illegal acts of creating a false impression of trading volume or price for a security. The New York Stock Exchange expounds that securities market manipulation is illegal action of buying or selling a security for the purpose of creating false or misleading appearance of active trading or for the purpose of raising or depressing the price to induce purchase or sale by others. The focus of this research paper will be on securities market manipulation which always relates to transactions as distinguished from insider trading which is driven by information.

Securities market manipulations can occur in a variety of ways. For example, by purchasing a large amount of stock, a broker can drive the price up. If that trader can subsequently sell those shares and if the price does not revert before completing his sales, then the broker can profit from such a trading strategy.

The rest of this chapter is organized as follows. In section 1.2, the review of literature on market manipulation will be given. In section 1.3, a series of common securities market manipulation strategies will be explained: • Ramping • Bait and Switch • Wash Trades • Derivative/Underlying Price Manipulation • Cornering the Market • Squeezing the Market • Unusual Expiration Day Activity • Layering the Order Book • Churning • Front Running

Section 1.4 will define the surveillance alerts that have been designed for detecting the incidence of these common market manipulations described in section 1.3.

1.2 Literature Review Securities market manipulation, no matter the forms of its existence, temporarily distorts a security’s price when manipulation strategies are successfully deployed. Perhaps a sensible question to ask is whether securities market manipulation is worth the effort of the academia to study and the energy of government (regulators) to act against it.

Although there perhaps has been no formal research conducted on the laws and regulations against securities market manipulation, Bhattacharya & Daouk (2002) report that up to 1998, there are 103 countries with stock exchanges and 87 of them having insider trading laws or regulations. Carlton and Fischel (1983) depict the argument in academia about the necessity of those insider trading laws or regulations as some believe that equilibrium price provides a reliable and efficient source of power to aggregate information effectively, and thus reduce or eliminate information asymmetries in the economy. Bhattacharya and Spiegel (1991) provides solid ground in theory to prove the severity of insider trading and the importance of laws and regulations against it. Their research, based on a two-date exchange economy, shows that the risk premium on a security can be completely characterized by the information and stock held by the market’s outsiders; so if insider trading laws and regulations do not exist, the market may fail completely as an aggregator of equilibrium price. A good example would be when the majority of the non-informed traders in the market believe they are severely disadvantaged by information asymmetry, they will not trade with the informed traders, which directly cause a liquidity collapse.

The same objections apply to securities market manipulations. When a securities market is full of manipulators, no investors will have confidence to invest their money in the market. Without sufficient investors, companies would be reluctant to get listed on the market raising the cost of capital and damaging overall social welfare (e.g., superfund’s performance). Consequently, the understanding of securities market manipulation has wide and critical economical, legal and social impact. In the next section, theoretical and empirical literature will be reviewed in terms of the existence

of manipulations in securities market, the forms of securities market manipulation and the relationship between volatility, spread and securities market manipulations.

1.2.1 The Existence of Manipulation in Securities Market Knowing the critical role manipulations can play in securities market, let us firstly have a glance at the likelihood of their existence. Perhaps a more precise terminology we should use here is profitable securities market manipulation as Jarrow (1992) defines it as large traders, those with market power, manipulate prices to their advantage and generate profits at no risk. Allen and Gale (1992) defines securities market manipulations and insider trading by defining trade-based manipulation as a trader attempting to manipulate a stocks imply by buying and then selling, without taking any publicly observable actions to alter the value of the firm or releasing false information to change the price. In contrast, Van Bommel (2003) looks at non- manipulation situations in which traders gain abnormal returns by spreading rumours in the market about their trades.

Hart (1977) perhaps is the very first study trying to prove the existence of profitable manipulation strategies in securities market. This paper builds up a time homogenous price process model in an infinite horizon, deterministic economy and shows that opportunities always exist for profitable manipulation if the economy is dynamically unstable and under certain cases even when the economy is stable. The stability of an economy in this study refers to the stability of market equilibrium.

In 1992, a series of studies were published with a common theme of providing a general model for the existence of securities market manipulations in stochastic economy with time dependent price processes. Jarrow (1992) investigates whether large traders, whose trades change securities’ prices, can gain profits through manipulation strategies. Again, one of the underlying assumptions of this study is that those large traders have no private information. This paper concludes that the existence of profitable market manipulation strategies is related to the time asymmetry in the sensitivity of price changes to the speculator's trades. That is, as the price changes due as a result of large traders’ manipulation strategies, noise traders follow those manipulators to trade with a lag with the belief of price momentum in

which increase in price caused by the speculator’s trade at one date tends to increase prices at future dates.

Allen and Gorton (1992) employ another theory to demonstrate the general existence of securities market manipulations. Their model is based on ‘Asymmetry of Price Elasticities’ which depicts the natural asymmetry between liquidity purchases and liquidity sales. When the likelihood of liquidity sales becomes more than liquidity purchases, the liquidity sales turns out to be less informative because it is less likely that the trader is informed. The bid price then moves less in response to a sale than the ask price does in response to a purchase. This asymmetry of price elasticities can create an opportunity for profitable manipulation strategy in which a manipulator can repeatedly buy stocks, causing a relatively large effect on prices, and then sell with relatively little effect.

On the other hand, Allen and Gale (1992) assert that when all agents in a market have rational expectations and maximize expected utility, profitable securities market manipulations still exist due to asymmetric information. The information asymmetry Allen and Gale refer here is the belief of the existence of informed traders rather than the existence of private information. Traders are uncertain whether a large trader who buys the stock does so because he knows it is undervalued based on his private information or because it is part of his manipulation strategy. In other words, it is this pooling equilibrium allowing profitable securities market manipulation to exist under general terms. The definition of pooling equilibrium given by the Dictionary of the Social Sciences (Calhoun, 2002) is equilibria that arise in economic interactions in the presence of incomplete information, where one side of the market has more information than the other side.

Aggarwal and Wu (2006) extend the Allen and Gale (1992) by incorporating information seekers and arbitrageurs into the pooling equilibrium model. This paper provides evidence to show why manipulation strategy can be successful when the manipulator trade in the presence of other trades who seek out information about the stock’s true value. In a market without manipulators, information seekers improve market efficiency by pushing prices up or down to the level private information points

to. In a market with manipulators, those information seekers are the ones being manipulated as they can’t distinguish between a market manipulator and an informed trader. This worsens the market efficiency from the perspective of price transparency.

In conclusion, the existence of profitable securities market manipulations has been firmly approved. This suggests a strong role for government regulations to discourage manipulation.

1.2.2 The Forms of Securities Market Manipulation Securities market manipulations exist in a wide range of forms though the overall aim is to drive the security price to the direction beneficial to the manipulator who then liquidates his holdings of the security at a price better than the portfolio establishment price for a profit.

Two traditional forms of securities market manipulations are Corner the Market and Squeeze the Market. They will be explained in details in Section 2 but in general, those two manipulation strategies are often conducted together and the essence is to forcefully shift the equilibrium price by gaining dominant control over the supply of a security. Allen, Titov and Mei (2006) analyse the expected utilities of the uninformed, the arbitrageurs and the manipulator and show that cornering the market can occur when everybody is behaving rationally, which is consistent with Aggarwal and Wu (2006) extend the Allen and Gale (1992). By examining a hand-collected data set of stock market and commodity corners which occurred between 1863 and 1980 in the U.S., this paper finds strong evidence that large investors and corporate insiders possess market power that allows them to manipulate prices. In addition, this paper also suggests that corners normally occur when there is bad news.

Merrick, Naik and Yadav (2005) investigate and present a specific example of market manipulation, Squeeze the Market, detailing the price effects alongside trading positions of participants. With a detailed analysis on the physical delivery squeeze in the March 1998 Long-Term U.K. Government Bond Futures contract traded on LIFFE, this paper depicts a simple squeeze strategy which involves

• Buying up the cheapest to deliver bond issue and taking a substantial long position in the bond future • By restricting supply this increases the price of the cheapest-to-deliver bond and also forces participants with shorter term futures contracts to deliver higher value bonds instead.

Jarrow (1992) sheds some lights on the possible reasons behind successful squeeze the market strategy. First of all, the short traders may not realize the market is cornered, because he cannot observe the speculator's trades. Secondly, it may be that the speculator has special information about a technical corner, rather than an actual corner, which the other traders do not share. A technical comer occurs when the speculator's holdings exceed the floating supply, those shares available for sale, and the floating supply is less than the actual supply of shares outstanding. An example is provided by Cornell and Shapiro (1989) as shares may sit in trusts or escrow accounts that cannot (or will not) be sold.

Kumar and Seppi (1992) extend the Kyle (1985) Model and discover that uninformed investors can make profits by taking positions in futures contracts with cash settlement followed by manipulations of prices of the underlying securities. If the manipulator’s futures position is larger than his/her spot position, the net expected gain from this manipulation strategy (i.e., profit on futures less loss on the spot) is positive. The major reason behind is the imperfect informational linkage of futures and the underlying markets which leads to “price pressure” in the futures market. To make this manipulation strategy work, noise traders are necessary in the futures market, which is again consistent with the pooling equilibrium theory.

Drudi and Massa (2005) conduct a unique case study of price manipulation in parallel markets with different transparency by using trading data from the Primary and Secondary Treasury Bond Market in Italy. The design of the Treasury Bond Market in Italy is very exceptive as it consists of a Primary Market which is based on uniform- type auctions, with a uniform cut-off price paid by all winning bidders; and a Secondary Market in which treasury bonds are traded continuously. It is not obscure that the primary market is less transparent than the secondary market since the auction

price, overall demand and bonds allotment won’t be known when the primary market closes. This study discovers that the informed dealers with positive news can profit by simultaneously placing bids in the primary market and sell in the secondary market, repurchasing when the primary market closes.

Manipulation of Closing Price becomes the theme of a number of studies. Harris (1989) reports the existence of a large mean day-end transaction price change in the U.S. market between December 1981 and January 1983. Hillion and Souminen (2004) find that broker manipulates the closing price of a stock precede large customer trades in order to improve customer’s impression of his execution quality. An earlier study by Flexison and Pelli (1998) provides empirical evidence found in the Finnish securities market supporting Hillion and Souminen’s theory. Carhart, Kaniel, Kusto and Reed (2002), and Bernhardt and Davies (2005) assert mutual funds manipulate shares’ closing prices at the end of evaluation period to improve fund performance as closing price is a common performance benchmark. Such manipulation strategy is also known as “Painting the Tape”.

A couple of empirical researches also study the expiration-day effect. Stoll (1987), and Chamberlain, Chueng and Kuan (1989) find empirical evidence in the North American markets that on the expiration day of index futures/options, the price mean- reversals are significantly higher than month or quarter end without index futures/options expiration. Stoll and Whaley (1991) suggests that the change of settlement procedure to use next day’s opening price in New York Futures Exchange and New York Stock Exchange may only shift the position of expiration-day volatilities. This provides another reason for the manipulation of closing price. Although, Corredor, Lechon and Santamaria (2001) don’t find any significant expiration effect on returns in the Spanish market which, as a small and less liquid market, is expected to have even stronger evidence.

1.2.3 Volatility, Spread and Securities Market Manipulations Having seen the various forms in which securities markets manipulations can exist, now let us lay our sight on the relationship between securities market manipulations and various market attributes.

Hart and Kreps (1986) laid the foundation for the connection between securities market manipulations and volatility. Traditional theory believes speculation stabilizes prices because speculators buy when the prices are low and sell when the prices are high. But other researches show that speculators will buy when the chances of price appreciation are high, which may or may not be when prices are low. This paper discovers that even though non-speculators and speculators alike behave rationally and speculators are competitive, speculation can destabilize prices. Applying this theory to a world with the existence of securities market manipulations, two propositions can be drawn. i) Price manipulators, as a kind of speculators, conduct market manipulation when they believe that profit can be made by doing so. This would destabilize security prices or increase the volatility of security prices. ii) Noise traders, another kind of speculators, follow manipulators to trade since they cannot differentiate informed traders and manipulators. This would also increase the volatility of security prices. Stoll (1987, 1991), Chamberlain, Chueng and Kuan (1989), Chiou, et al (2007) all find empirical evidence suggesting the volatility of security prices is higher during the period of manipulation.

Foucault (1999) develops a Theory of Order Placement which connects volatility, order placement strategies and spread. In Foucault’s theory, order placement strategy consists of two components, Order Type and Order Aggressiveness, which will be defined first. There are two basic order types available for traders to choose from. Market order is submitted without a price which will be executed against the prevailing best price and thus, known as the source of liquidity supply. On the contrary, limit order, which is submitted with a price and stored in the order book waiting for future execution with market orders, consumes liquidity supply. When the non-execution risk is high, traders will use market orders to gain immediate execution; when the pick-off risk is high, limit orders turn out to be a better choice. Order aggressiveness refers to how close the order price is to the prevailing best price when the order is entered or amended. Foucault’s theory predicts that when the volatility of security price increases, traders will tend to hold limit orders rather than

market orders to reduce the pick-off risk at the cost of non-execution risk being increased. When execution risk is high, liquidity traders are under the pressure to trade immediately upon arrival because the probability of being executed with a limit order is small. For this reason, traders are willing to place market orders at more unfavourable pries, which result in limit order traders to be less aggressive by posting larger spread in order to take advantage of liquidity traders.

Aitken, Almeida, Harris and McInish (2007) provide evidence supporting Foucault’s Theory of Order Placement. They conclude that hedge funds with short-lived information face high cost of non-execution, and thus their Order Placement Strategy is very aggressive. Insurance companies and mutual funds with the emphasis of cost control tend to be less aggressive.

Comerton-Forde and Putnins (2009) studies the effects of closing price manipulation in an experimental market to evaluate the social harm caused by manipulation. They find that manipulators, given incentives similar to many actual manipulation cases, decrease price accuracy and liquidity, thereby undermining economic efficiency. They conclude that the mere possibility of manipulation alters market participants’ behaviour, leading to reduced liquidity.

1.3 Securities Market Manipulations In this section, common securities market manipulations will be described.

1.3.1 Ramping Consistent with the overarching goal of maintaining market integrity, a key goal of a securities market must be to ensure that no one investor can manipulate prices for their benefit, that is, deliberately cause a short term supply/demand imbalance. The ability to manipulate a market would be difficult if individual investors were to invest primarily on their own account. However, given that investors now congregate in funds, the effective size of these new types of investors means that manipulation is feasible. Most funds base their quoted prices and value off of the closing prices of securities in their portfolio. Unusual price movements which revert the following day – especially at quarter ends, may indicate market manipulation to artificially inflate the price of securities. This type of market manipulations is referred as Ramping or Painting the Tape specifically referring to fund managers manipulating security’s closing price at the end of the evaluation period.

A successful ramping case normally evolves in two stages – Marking the Close and Reversal at the Start of the next Trading Day. Marking the Close is a form of price manipulation describing the practice of executing purchase or sale orders at or near the close of the trading session to raise or lower the closing price or to raise or lower the bid or offer artificially for the purpose of reducing margin or net capital requirements for enhancing profit and loss, or to influence the mark-to-market for credit or reporting purposes if holding a large position in the derivatives contract.

Significant price changes at the end of the day may also be explained by reasons other than a participant trying to drive the price for manipulative reasons, for example, information announcements. If the price changes substantially at the end of a day and then reverses quickly at the start of the next trading day due to manipulator’s liquidation, the possibility of such case becoming ramping manipulation will be notable. Figure 1 represents a successful ramping case.

Figure 1 Ramping

1.3.2 Unusual Expiration Day Activity Unusual Expiration-Day Activity in its essence is very similar to Ramping but occurs during the "triple-witching" hour - the last hour of trading on days on which index futures, index options and options on index futures expire simultaneously. The purpose of manipulating closing prices of index stocks is not obscure. For example, long position holders of index derivatives will be settled at better price if they can successfully manipulate prices of index stocks upward.

1.3.3 Misleading Order and Trading Strategies Order depth, a widely-used measure of market liquidity, becomes a popular target for market manipulations via Misleading Order and Trading Strategies. Participants carry out misleading order and trading strategies both in the auction period and in continuous trading which are designed to give false perception of order depth on either the bid or ask side of the market. When those strategies are successfully deployed, extra demand or supply for a security can be created, while the participant trades on the other side of the market at more favorable prices.

One form of misleading order and trading strategy is Bait and Switch which is described through the following example. Assume broker B1 intends to clear his holding of 10,000 shares of S1. The prevailing order book can be illustrated by Figure 1 below.

Figure 2 Order book for the Misleading Order Strategies Scenario

It can be seen that prevailing best bid/ask price is $11.25 and $11.35, which creates a bid-ask spread of $0.10. Further assume the price tick of security S1 is $0.05. Rather than trading with bids at $11.25, broker B1 enters a bid with price $11.30 and volume 10,000. The behaviour of broker B1 draws attention from other investors and some of them subsequently enter bids at $11.30 as well. Once the total depth of bids at $11.30 from other investors approaches to 10,000, broker B1 withdraws his bid at $11.30 and simultaneously enters a market order (ask) with volume 10,000 which gets traded with remaining bids at $11.30. Figure 2 below illustrates this Misleading Order Strategy. The succesful deployment of this strategy helps broker B1 increase the potential execution price from $11.25 to $11.30.

B1 enters a market order at B1 deletes this $11.30 to trade order when with those bids there is enough order depth created

Figure 3 Illustration of a Successful Misleading Order Strategy

1.3.4 Wash Trades Manipulators conduct Wash Trades by actively trading on the market for a security between two accounts that they control, typically buying through one broker, and simultaneously selling through another. Wash trades involves no change in the beneficial ownership of the securities. One obvious intention of Wash Trades is to give false impression of active trading of the security being manipulated, which may attract day-traders to trade. Wash Trades can also be utilized as an avenue for brokers to increase their market share in an instrument leading up to a secondary float of shares. For example, in Australia, lead brokers for the secondary float of Telstra shares (Telstra is the biggest telecommunication company in Australia) were chosen on the basis of existing market share, which may potentially encourages brokers to conduct Wash Trades to compete for the lead broker position.

1.3.5 Derivative-Underlying Price Manipulation In today’s uncertain financial environment, derivative instruments have been widely deployed for hedging and speculating. According to the Bank for International Settlements, the total market value of global derivatives has exceeded $516 trillion in 2007. This makes derivative instruments as heavily manipulated as equities.

Rather than directly manipulating the price of derivative instruments, manipulators often choose to detour by conducting Derivative-Underlying Price Manipulation. 15

Manipulators will take a highly leveraged position1 in a derivative security and then manipulate the price of the underlying security to improve the value of their derivative position (e.g., increase the price of call options or decrease the price of put options). This distorts the true value of the underlying security and creates unnecessary intraday volatility. According to the Strait Time (8 March 2008), recently a Derivative-Underlying Manipulation case was successfully prosecuted in Singapore where two proprietary traders manipulated the price of the Straight Times Index by trading in 12 of the constituent stocks, which made their derivative positions on those constituent stocks profitable. The manipulators then clear both of their positions on the derivative securities and underlying securities to realize the profits. The two manipulators were each fined SG$100,000 (approximately AU$80,000).

3. Clear the position on both the underlying security and the derivative security at a certain stage to realize the profit

2. Then manipulate the 1. Establish the position price of the underlying in the derivative security security in order to first improve the price of the derivative security

Figure 4 Illustration of Derivative-Underlying Price Manipulation

1.3.6 Layering the Orderbook Layering the Orderbook is a form of Misleading Order and Trading Strategy in which a manipulator places a large number of differently priced orders below the best bid price or above the best ask price to create fake appearance of multiple demand or supply sources for a security. In anonymous markets, other participants just see many orders and don’t realise they are all for the same investor. A successful case of

1 Extensive futures positions can reduce the flexibility of manipulators to meet margin requirements, to cancel and replace orders, and to execute on very short-term arbitrage opportunities, etc. 16

Layering the Orderbook may provoke other investors to trade, which could help to either raise or lower the price of the security in the manipulators favour. For example, a manipulator places many bids at different price level to bring the market up, to a level that the manipulator wants to sell at. The manipulator then submits offers that get traded immediately at the favourable price, and withdraws those fake bids after the trades.

Figure 5 Layering the Order book

1.3.7 Churning Churning is normally conducted by a group of brokers/traders who pass share parcels of a security within the group in order to create a high turnover for the security being manipulated but very low net turnover for the group on that security. Figure 3 below illustrates two Churning scenarios. The first one is conducted by Broker 1 (B1) and Broker 2 (B2). B1 sells 10,000 shares of security S1 to B2 then B2 sells 10,000 shares of security S1 back to B1. By doing this, a total turnover of 20,000 volume unit is created for security S1 while the net turnover for B1 and B2 on security S1 is zero. The second scenario is conducted by B1, B2 and Broker 3 (B3) on security S1. B1 sells 10,000 shares of S1 to B2; B2 then pass that parcel of shares to B3 who finally finishes the churning cycle by selling that parcel of shares back to B1. Similar to the

first scenario, a total turnover of 30,000 volume unit is created for S1 while none of the three brokers holds a position on S1 at the end of the churning cycle.

Figure 6 Two Scenarios of Churning

Several motivations could initiate churning. Some manipulators may try to increase their commissions by excessively trading between each other (in this case, a group of client accounts will be used for settlement); some manipulators pre-arrange churning to create an impression of active market on illiquid securities so that more investors can be attracted to improve the execution price in manipulators’ favor; in some markets, warrant’s turnover has become one of the most important benchmarks for non-institutional investors to determine which warrant to trade. For example, when 10 underwriters issue a total of 10 warrants for Google with similar product structure, non-institutional investors would often rank those warrants by turnover and invest in the ones with high turnover. As a result, a warrant underwriter may collaborate with a group brokers to make use of Churning to increase the turnover of a warrant issued by him in order to compete against warrants issued by other underwriters for the same underlying.

1.3.8 Cornering the Market Cornering the Market is normally conducted on derivatives contracts settled by physical deliveries. To successfully corner the market, a manipulator needs to acquire a large quantity of a physical good that is the deliverables for some derivatives

contracts in order to shift the market equilibrium by reducing the supply of that physical good so that strong price pressure can be exerted. For example, when a derivatives contract is cornered, investors who hold short positions on that contract will be forced to close out their positions by buying back the physical good from manipulators cornering the market at a much higher price.

1.3.9 Squeezing the Market Manipulators can bid up the price of a security if they suspect the cumulative short position on a security is high. At some point, investors may be forced to buy back to close their position, which will result in more demand for that security and further driving the price of the security higher. The manipulator can then sell the securities to the market at a higher price. This form of manipulation is normally known as Squeezing the Market that is most effective where momentum traders are present which creates enough demand for the manipulator to sell back to. The Corning the Market strategy is often accompanied by the Squeezing the Market strategy.

1.3.10 Front running Brokers trade not only for their customers but also for themselves. Front running is a kind of stock market manipulations in which a stock broker executes orders on a security for their own account (as principal) before executing orders previously submitted by their customers (as agency). After the broker has made their original transactions, they can expect to close out their position at a profit based on the new price level. Front running may involve either • Brokers buying for their own accounts, driving up the price before executing customer buy orders; or • Brokers selling for their own account, driving down the price before executing customer sell orders.

For example, a broker buys 20,000 shares of a stock for $10 per share just before buying a large block of 100,000 shares for a customer. The broker’s principal trade may drive the price up to $10.5 per share. If the broker is able to sell their newly purchased shares at $10.3, it will have made $6,000 in a few minutes, which is likely

to be part of the additional cost to the customer's purchase caused by the broker's principal dealing.

1.4 Alerts for Securities Market Manipulations

1.4.1 Introduction In collaboration with the Capital Markets Cooperative Research Centre Limited (CMCRC) and Smarts Group International Pty. Ltd., the author of this dissertation have gained broad knowledge in the area of securities markets manipulation detection by implementing market surveillance alerts for a number of securities exchanges. Some of those exchanges are: • The Australian Securities Exchange (ASX)

• The Saudi Arabian Stock Exchanges (TADAWUL)

• The Dubai Financial Market (DFM)

• The Euronext Stock Exchange (NYSE Euronext)

• Tokyo Commodity Exchange (TOCOM)

• The Singapore Stock Exchange (SGX)

• The Hong Kong Stock Exchange (HKEX)

• The Securities and Futures Commission of Hong Kong (HKSFC)

• The PLUS Market Group (PLUS)

• The Stock Exchange of Thailand (SET)

• The Securities and Exchange Board of India (SEBI)

• Malaysia Securities Commission (MSC)

In Section 1.4, market surveillance alerts that are developed by the author of this dissertation for securities exchanges listed above to detect market manipulation cases will be explained in terms of purpose, data requirements and algorithm design.

1.4.2 The Marking the Close Alert 1.4.2.1 Purpose The Marking the Close Alert is designed to detect attempts to manipulate the close price of a security. Marking the Close is normally referred as the first stage of a Ramping case. This alert can also identify Unusual Expiration Day Activity.

1.4.2.2 Data Requirements The Marking the Close Alert is executed on trade level data including trade time and trade price. If trading participants data is provided, this alert can become more informative by pointing out the participant who is responsible for majority of the price movements.

1.4.2.3 Alert Algorithm A Marking the Close Alert will be issued if the following conditions are satisfied: 1) When the security is closed for trading; and 2) The absolute percentage difference between the close price of the security and its price X minutes before closing is greater than the threshold.

The calculation of threshold is described as below. 1) For each trade of a security over the past T days, calculate the absolute percentage difference between the trade price and the price X minutes ago; 2) Assign the absolute percentage difference calculated in 1) to the price change distribution for that security; 3) At last, calculate the mean and standard deviation for the price change distribution for each security. If there are more than 50 observations in the price change distribution for a security, set the threshold to the 3 standard deviations away from of the price change distribution as the threshold for that security; otherwise determine the percentage threshold from the following Price Band-Threshold Table. The rationale behind the threshold calculation is to capture unrepresentative price change. This algorithm has been implemented at more than 30 national exchange and regulators by SMARTS.

Last Trade Price of the Security Percentage Threshold Price From ($) Price To ($) 0.00 0.10 20% 0.11 0.25 15% 0.26 0.50 12% 0.51 1.00 10% 1.01 5.00 8% 5.01 10.00 5% 10.01 10000000.00 3%

1.4.3 Reversal the Next Trading Day 1.4.3.1 Purpose The Reversal the Next Trading Day Alert is designed to detect security, which was issued a Marking the Close Alert the previous day, has a price reversal towards the VWAP price X minutes before the previous day’s closing during the first Y minutes of the current day’s trading. Reversal the Next Trading Day is normally referred as the second stage of a Ramping case.

1.4.3.2 Data Requirements The Marking the Close Alert is executed on trade level data including trade time, trade price and trading participants.

1.4.3.3 Alert Algorithm A Reversal the Next Trading Day Alert will be triggered for a security when the following conditions are satisfied: 1) If a Marking the Close Alert was triggered for a security the previous day; and 2) A trade occurs for that security during the first Y minutes of trading on the current day; and 3) The absolute percentage difference between the trade price and the security’s VWAP price X minutes before the previous day’s closing is smaller than a pre-defined threshold (e.g., 50%). 23

1.4.4 The Bait and Switch Alert 1.4.4.1 Purpose The Bait and Switch Alert is designed to identify order strategies both in the auction period and in continuous trading which are designed to give the perception of depth on either the bid or ask side of the market, while the participant submitting the orders trades on the other side.

1.4.4.2 Data Requirements The Bait and Switch Alert requires both order level and trade level data including best bid/offer price, order entry time and price, trade time and price and buying/selling participants.

1.4.4.3 Alert Algorithm A Bait and Switch Alert will be triggered when the following conditions are satisfied: 1) A participant (the suspect participant) enters a bid/offer which improves the prevailing best bid/offer price; and 2) Other participants follow by entering bid/offer at the same price or even better price; and 3) When the total order depth created by other participants at the new best bid/offer exceeds the intended volume the suspect participant wants to trade, the suspect participant withdraws his initial bid/offer; and 4) The suspect participant enters order at the opposite side which gets traded subsequently.

1.4.5 Wash Trade 1.4.5.1 Purpose The Wash Trade Alert identifies trades where there is no change in beneficial ownership of the parcel of shares.

1.4.5.2 Data Requirements The Wash Trade Alert requires trade level data, and client data or participant dealing capacity data including trade time, volume and trading client/broker dealing capacity (e.g., Principal or Agency)

1.4.5.3 Alert Algorithm A Wash Sale Alert will be triggered when the following conditions are satisfied: (1) The buy client and the sell client of a trade are the same entity or the dealing capacity of the buy side and the sell side are the same (e.g., both the buy broker and the sell broker are trading for the same and the broker is trading for himself) (2) The number of wash sales conducted by a client/broker on one security exceeds the threshold.

1.4.6 Derivatives/Underlying Price Manipulation 1.4.6.1 Purpose The Derivatives/Underlying Price Manipulation Alert is designed to identify participant who gains profits on his/her derivatives position by manipulating the underlying’s price.

1.4.6.2 Data Requirements The Derivatives/Underlying Price Manipulation Alert requires trade level data for the derivatives instruments and their underlying, including trade time, price, volume and buying/selling participant.

1.4.6.3 Alert Algorithm A Derivatives/Underlying Price Manipulation Alert will be triggered when the following conditions are satisfied: (1) A participant (e.g., client or principal broker) establishes a long (short) position in call (put) option(s) or warrant(s) for an underlying (2) After the long (short) position is established for the call (put) option(s) or warrant(s), the same participant starts continuously buying (selling) the underlying security in order to drive the underlying price up (down) (3) When the price of the call (put) option(s) or warrants(s) increases by a certain percentage, the participant closes his position in the option(s) or warrant(s) to realize the profit

(4) After the derivatives position is closed, the participant starts to close his position in the underlying as well (5) At the end of the trading day, the participant holds a zero (or near-zero) position in the underlying security with trivial profit or loss from the day trading in the underlying security

1.4.7 Cornering the Market 1.4.7.1 Purpose The Cornering the Market Alert is designed to identify participant who tries to manipulate the price of futures contracts by dominating the holdings of the underlying security or commodity. 1.4.7.2 Data Requirements The Cornering the Market Alert requires derivatives trading data and underlying holdings data including trade time, price, volume, participant and the participant’s holding quantity of the underlying security of commodity.

1.4.7.3 Algorithm The Cornering the Market Alert will be triggered when the following conditions are satisfied: (1) At the start of a trading day, check each participant’s holding position in securities or commodities that are underlyings for derivatives contracts; (2) If one participant holds the majority of the underlying securities or commodities In the (e.g., more than 25%), issue an alert for Excessive Holdings of Underlyings; (3) During continuous trading, if the trading price of a derivatives contract significantly varies from the reference price (e.g., close price from 3 trading days ago), then check if the underlying of that derivatives contract has been issued an Excessive Holdings of Underlying Alert before; if yes, issue a Cornering the Market Alert for the underlying security and the participant dominating the holding of the underlying security or commodity.

1.4.8 Squeezing the Market 1.4.8.1 Purpose

The Squeezing the Market Alert is designed to identify participant who continuously drives the price of a derivatives contract up when the cumulative short position is relatively high in the derivatives contract.

1.4.8.2 Data Requirements The Squeezing the Market Alert requires derivatives trading data and derivatives holding data including trade time, price, volume, participant and participant’s holding quantities of the derivatives contract.

1.4.8.3 Algorithm The Squeezing the Market Alert will be triggered when the following conditions are satisfied: (1) At the start of a trading day, look over the derivatives holding database and calculate the cumulative short position of each derivative contract; (2) If the cumulative short position of a derivative contract exceeds the benchmark (e.g., 30% of the total contracts issued), flag that derivative contract as highly sensitive; (3) During continuous trading, if a participant continuously drives up the price of a derivatives contract which has been flagged as highly sensitive before, issue a Squeezing the Market Alert.

1.4.9 Layering the Order Book 1.4.9.1 Purpose The Layering the Order Book Alert is designed to identify participant who creates a false impression of an active market for a security by placing multiple buying (selling) orders at different price level with the intension of trading at the opposite side with a better price.

1.4.9.2 Data Requirements The Layering the Order Book Alert requires the order level data and trade level data including transaction time, price, volume and participant for orders and trades.

1.4.9.3 Algorithm

The Layering the Order Book Alert will be triggered when the following conditions are satisfied: (1) A participant (broker, trader or client) places multiple orders with different price at the bid (ask) side of a security; (2) The same participant subsequently enters orders at the opposite side and those orders get traded; (3) After the trade, the participant quickly withdraws those fake bids (asks) entered before at various price levels.

1.4.10 Churning 1.4.10.1 Purpose The Churning Alert is designed to identify participants-pair that creates a false impression of an active market for a security by trading between each other without holding a substantial quantity of the security overnight.

1.4.10.2 Data Requirements The Churning Alert requires trade level data including trade time, price, volume and participants.

1.4.10.3 Alert Algorithm The Churning Alert will be triggered when the following conditions are satisfied: (1) On every trade, check whether the buy participant-sell participant pair has traded the security before; (2) If yes, increase total turnover from that participant pair for that security by the volume of the current trade and adjust the net turnover in between the participant pair accordingly. For example, assume Broker A is the buy broker while Broker B is the sell broker for the current trade with volume of 5,000 shares. Suppose Broker A was the sell broker while Broker B was the buy broker from a previous trade for the same security and the trade volume was 5,000 shares. Consequently, the total turnover, for that security, from the Broker A- Broker B pair after the current trade is 10,000 shares while the latest net turnover in between the broker pair is zero.

(3) At the end of the day, if the total turnover from a participant pair on a single security exceeds the total turnover threshold and the net turnover as a percentage ratio to the total turnover from that participant pair is below the net turnover threshold, issue a Churning Alert.

1.4.11 Front Running 1.4.11.1 Purpose The Front Running Alert is designed to identify participant who executes orders for himself before executing clients’ orders in order to enjoy better execution price. 1.4.11.2 Data Requirements The Front Running Alert requires order and trade level data including order/trade transaction time, price, volume, participant and dealing capacity (e.g., trading as principal or agency).

1.4.11.3 Alert Algorithm The Front Running Alert will be triggered when the following conditions are satisfied: (1) A participant executes a bid (offer) order for a client and the execution moves the security price up (down); (2) Before the execution of the client’s bid (offer) order, if the participant executes bid (offer) order(s) as a principal at a price better than the execution price of the client’s order, flag the execution(s) as potential front running trade(s) and increase the potential front running volume executed by that participant on that security by the volume of the potential front running trade(s); (3) At the end of the day, if the total front running volume for a participant- security pair exceeds X% (e.g., 30%) of the total volume that participant executed on that security during the day, issue a front running alert.

Chapter 2

Market Integrity and Market Efficiency: A Cross-Market Comparison

2.1 Introduction After the financial crisis of 2008, more than ever investors prefer to put their money into a “fair market” which has a high level of integrity and efficiency. Accordingly, more and more securities exchanges have declared and begun to demonstrate their commitment to the twin goals of market efficiency and market integrity. NASDAQ states on its website that

“NASDAQ is among the world’s most regulated stock markets, employing sophisticated surveillance systems…to protect investors and provide a fair and competitive trading environment.”

“Offering growth and liquidity, fostering innovative technologies…NASDAQ continues to build the most efficient trading environment…to the benefit of all market participants and investors.”

Appendix 1 contains a sample of relevant statements from the websites of other major world securities exchanges worldwide.

Although significant resources have been invested in the improvement of market integrity and market efficiency by the those major stock exchanges, little is known about the direct relationship between market integrity and market efficiency and whether an improvement of market integrity helps in achieving better market efficiency. The purpose of this Chapter is to address these questions directly.

In Chapter 1, a review of theoretical and empirical literature was given for the existence of securities market manipulations and the relationship between manipulations and some market attributes. Hart and Kreps (1986) laid the foundation for the connection between securities market manipulations and volatility; Foucault (1999) and Aitken, Almeida, Harris and McInish (2007) find a negative correlation between volatility and aggressiveness of order placement strategy. When volatility is high, traders submit orders less aggressively because non-execution risk decreases and picking off risk increases, which are eventually reflected in wider spreads. There have been a number of empirical studies conducted to test the relationship between securities market manipulations and volatility, and we would expect a follow-on correlation between volatility and spreads.

However, there has been no direct test of the relationship between securities market manipulations and spreads. One of the fundamental reasons perhaps is the extreme difficulty of collecting data on market manipulations. In this chapter, a random effects correlation analysis will be conducted for the relationship between securities market manipulations and market efficiency. We believe this work to be the first of its kind.

The rest of this chapter is organized as follows. In section 2.2, the research methodology will be specified. Section 2.3 defines the data and measurement used by this research while section 2.4 gives the empirical specification of the statistical regression model. Section 2.5 and 2.6 analyses and discusses the empirical results.

2.2 Research Methodology

2.2.1 Market Integrity Defined for this Research Securities market manipulations exist in various forms. Section 1.3 provides an illustration of ten common securities market manipulation strategies. The nature of each manipulation strategy decides the surveillance approach to detection. Though thousands of ways could exist to detect one kind of market manipulation strategy versus another, a series of detection rules (a.k.a., Alerts), which the author has designed and implemented for 12 securities exchanges and regulators around the world are presented in Section 1.4. For the purpose of this research, the Ramping Manipulation Strategy will be the focus, and the combination of the Marking the Close Alert and the Reversal the Next Morning Alert, which is normally used to detect Ramping will become an important part of the research methodology. The reason behind this selection is given below.

Table 1 lists the minimum level of data required by each of the 10 types of market manipulations presented in Section 1.4.

Table 1 Common Alerts Data Requirements

Alert Name Data Required (minimum) Marking the Close Alert Trade Level Data Reversal the Next Trading Day Alert Trade Level Data Bait and Switch Alert Order Book Data Trade Level Data Wash Trade Trade Level Data Participant Data Derivatives/Underlying Price Trade Level Data Manipulation Participant Data Cornering the Market Trade Level Data Participant Holdings Data Squeezing the Market Trade Level Data Participant Holdings Data Layering the Order Book Order Book Data Trade Level Data Churning Trade Level Data Participant Data Front Running Trade Level Data Participant Data Deal Capacity Data

From Table 1, it can be seen that Marking the Close Alert and Reversal the Next Trading Day Alert are the only two alerts that can be executed by using the Trade Level Data only (Trade level data includes security, trade time, trade price and volume). All the other alerts require Order Level Data, Participant Data, Participant Holdings Data or Deal Capacity Data. Hart and Kreps (1986) point out that data collection for market manipulations is extremely hard. The only order book data database that can be publicly accessed is NYSE trades, orders, reports, and quotes (TORQ) Database which contains time-ordered transactions for 144 selected stocks for the short three month period from November 1990 to January 1991 for the purpose of an order book audit. Even more so, the participant data required for the surveillance and enforcement of rules violated by other manipulations (and their associated alerts) is strictly confidential to stock exchanges; no publicly accessible database can provide the participant data.

Therefore, this research will use Ramping Market Manipulation (which consists of Marking the Close Alert and Reversal the Next Morning Alert) as a proxy for securities market manipulation for the cross-markets analysis.

A recent case detected by the Australian Securities Exchange (ASX) illustrates the behaviour underlying Ramping. On Friday, 29th June 2001 between 4 and 4.15pm the Standard & Poor’s ASX 200 Index (SPI 200) increased 45.5 points following the closing single price auction (CSPA) on the ASX. By market open on the following Monday, this unusual increase was reversed. The last trading day of the financial year always pushes share prices a little higher, but on 29 June the All Ordinaries Index rose by 67 points, or two per cent, and the ASX is concerned market manipulation may have been involved. On 2 July, the index fell by 54 points, as the "ramping" buyers, believed to be fund managers and derivative players, withdrew.

2.2.2 Market Efficiency Defined for the this Research Aitken and Berry (1993) state that academia typically define market efficiency based on information as the market reacts in a speedy and unbiased fashion to information. Once Kendall (1957) argued that stock prices displayed ‘random walk’ behaviour, economists came to realize that random price movements indicated a well-functioning or efficient market. The three standard forms of efficiency come from broad consensus where evidence has shown that markets are at least weak form and often semi-strong form efficient under the Efficient Market Hypothesis (e.g., Fama (1970, 1991 and 1998), Jensen (1978) and Malkiel (1996)). Grossman and Stiglitz (1980) claim that investors will have an incentive to spend time and resources to analyse and uncover new information only if such activity is likely to generate higher investment returns.

Gilson and Krakman (2004) came up with the following general conclusion. The level of market efficiency with respect to a particular fact is dependent on a number of mechanisms • Informed trading, e.g., professionally-informed trading, derivatively informed trading, etc; and

• Uninformed trading which is operated to cause that fact to be reflected in market price. Which mechanism is operative depends on the breadth of the fact’s distribution, which in turn depends on the cost structure of the market for information. They further argue that the lower the cost of information, the wider its distribution, the more effective the operative efficiency mechanism and, finally, the more efficient the market.

In order for a market to become efficient, investors must perceive that a market is inefficient and possible to beat. Investment strategies intended to manipulate inefficiencies are actually the fuel that keep a market efficient. However it is often these investment strategies that not only affect the level of efficiency but the integrity of the market place. This is consistent with the pooling equilibrium theory for the existence of profitable securities market manipulations suggested by Allen and Gale (1992) and Aggarwal and Wu (2006).

For the purpose of this research, we shy away from traditional academic notions of market efficiency that tend to focus on information efficiency as previously discussed to a more all encompassing definition which concerns itself with the ability to instantaneously convert cash into securities and back again. The more efficient the market the cheaper is the conversion process; or more conventionally, the lower are transaction costs.

Key components of transaction costs include brokerage costs, market impact costs, and opportunity costs. Unfortunately none of these is directly observable in the data available to us. We therefore proxy transaction costs by measuring (1) The absolute cost of a round trip transaction (the quoted spread) as a percentage of the spread midpoint; and (2) Average cost beyond the midpoint to complete all trades (the effective spread) as a percentage of the quote midpoint; and (3) Average cost for immediate versus more patient trades (the realized spread), again as a percentage of the quote midpoint.

These relative bid/ask spreads are widely used and accepted measures of the relevant transaction costs.

2.2.3 Random Effects Model By maintained hypothesis, market integrity affects transaction costs. A lower level of integrity and higher level of market manipulations is hypothesized to raise spreads. Specifically, the null hypothesis of our research is

: Higher ramping manipulation (as proxied by ramping alert incidence AI) is associated with higher long-term quoted and effective spreads.

where AI is reported by the surveillance system created by the leading surveillance technology and consulting firm, Smarts Group International.

The alternative hypothesis is that ramping alerts represents information arrivals, not market manipulations. Consequently, under the alternative hypothesis, information both positive and negative will temporarily raise spreads, but spreads will then mean revert, and ramping alert incidence (AI) will be unrelated to long-term average spreads.

Since spreads in round trip transactions, and the spread to complete large trades, and the price impact of acquiring large positions are (along with expected legal consequences) substantive costs of engaging in market manipulations, we also expect alert incidence to be determined by spreads. Hence, our prior is that relative spreads and AI are endogenously determined. In addition, both information arrivals and market manipulations are inherently unobservable variables. For these reasons, this research utilizes a Random Effects Model to test the correlation between Ramping Alert Incidence data and relative bid/ask spread data collected for 34 securities markets over 6 years (2000-2005). This Random Effects Model is expressed as follows:

, , , (1)

where

, = the relative spread measure of market i at time t

= the constant that would be estimated

= the correlation coefficient between relative spread and Ramping

Alert incidence

, = the average number of Ramping Alerts of markets i at time t

= Observational error on information arrivals at time t

= Observational error on Ramping Alert detection of market i

̃, = The residual error term

It is essential to understand that the Random Effects Model specified above will be used to test the correlation rather than the causality between relative spread and Ramping Alert incidence across securities markets over time. The Random Effects Model is based on Error Component Modelling. That is, we expect spreads to reflect both – the observational error on information arrivals at time t as well as – the

observational error on Ramping Alert detection in market i. If , is independent of

and (i.e., if Cov(,,) = 0 and Cov(,,) = 0), then , is not jointly

depent with , on random observational errors in detecting informational arrivals over time or market manipulations across markets. In that case, we will estimate the correlation between relative spreads and alert incidence with moving average time- series models (like Da Silva’s method).

On the other hand, if Cov(,, ) 0 and/or Cov(,, ) 0, then we will

proceed to fixed effects modelling. Initially, the correlation between , and , will be estimated by controlling for exchange-specific dummy variables that capture idiosyncratic surveillance, detection, prosecution, or enforcement reasons why manipulation may be observed only with error. Then only we will estimate all those cross-sectional fixed effects as well as time-series dummy variables that control for idiosyncratic reasons in each time period why information may arrive but be observed 37

only with error. In both estimation models, the null hypothesis is that despite all these fixed effects control variables, the positive association between , and , will continue to be present and statistically reliable.

Examples for Cov(,, ) = 0:

The following examples illustrate various possible findings for the null hypothesis below.

: The covariance between Ramping Alert incidence and information arrivals over time is zero.

(1) As per Section 1.4, the Ramping Alert will not be issued unless a Reversal the Next Trading Day Alert is triggered after a Marking the Close Alert for a security. The Reversal the Next Trading Day Alert is triggered when the suspect security has a price reversal. The price reversal detected the next morning could be a true mean reversion following information arrivals and the associated liquidity available in resilient limit order books or it could result the day following a Marking the Close Alert triggered by true closing price manipulation2. In either case, the Ramping Alert is triggered. As a result, the null hypothesis of zero systematic relationship (i.e., zero covariance) between information arrivals and Ramping Alert incidence over time would be accepted; (2) In opposition to (1), if positive (negative) news is announced overnight when the Marking the Close Alert is triggered for upward (downward) closing price manipulation, there will be no price reversal the next morning as the market continues to trend, which in turn stops the Ramping Alert from triggering. In such case, the null hypothesis of zero covariance between information arrivals and Ramping Alert incidence over time is rejected; (3) The information event could also have pure white noise. For example, an announcement by a small listed company could be out of traders’ sight, which

2 Some may also suspect that price reversal is purely caused by order re-balancing. But according to the results presented in Section 2.3.4 of this thesis, the average ramping alert incidence per security-day across the 34 securities markets over the period of 2000-2005 is only 0.001, which is too infrequent to be order rebalancing. 38

doesn’t impact on the issue of the Ramping Alert. In such case, the null hypothesis of zero covariance between information arrivals and Ramping Alert incidence over time is also accepted.

Examples for Cov(,,) = 0:

The following examples illustrate various possible findings for the null hypothesis below:

: The covariance between Ramping Alert incidence and Ramping Alert detection error across securities market is zero. (1) When closing price manipulation truly present for a security from a market and a Ramping Alert is triggered for that security, the null hypothesis of zero covariance between Ramping Alert incidence and manipulation detection error across securities market is accepted; (2) When closing price manipulation is truly present for a security, but due to insufficient monitoring by momentum traders or arbitrage traders (e.g., for highly liquid securities), no one emulates the manipulators’ trading activities, the manipulation strategy eventually therefore fails and no Ramping alert is triggered. In such a scenario, the null hypothesis of zero covariance between Ramping Alert incidence and manipulation detection error across securities market is rejected; (3) The alert detection process could also generate pure white noise. For example, an arbitrary time period before closing (e.g., 15 minutes) is used in all markets to detect closing price manipulation (AI). But the manipulation could occur 30 to 15 minutes before closing or at other randomly chosen distances from closing. In such case, the null hypothesis of zero covariance between manipulation detection error and Ramping Alert incidence over time is also accepted.

2.2.4 Model Specifications Based on Section 2.2.3, five model specifications will be tested for the Random Effects Model. They are (1) Random One Model:

, , ,̃ (2)

which only exams the random observational error on Ramping Alert detection of market i;

(2) Random Two Model:

, , ,̃ (3)

which exams both the random observational error on Ramping Alert detection of market I and the observational error on information arrivals at time t;

(3) Da Silva Model:

, , , , ,̃ (4)

which assumes a mixed variance-component moving average model for the error structure;

(4) Fix One Model:

, , ,̃ (5)

which controls for the systematic effects from observational errors on manipulation presence across i securities markets with dummy variables for each of the 34 securities exchanges.

(5) Fix Two Model:

, , ,̃ (6)

which controls for the systematic effects from observational error on information arrival at time t as well as manipulation presence in market i with two sets of dummy variables over time and across securities.

2.3 Data and Measurement

2.3.1 Data The data for this research is obtained from the Reuters database maintained by the Securities Industry Research Centre of Asia-Pacific (SIRCA). This database contains intra-day trade and quote data for seven years for more than 200 world markets including most of the equity markets. The period of analysis for our study extends from January 2000 to December 2005. The period of December 1999 is used to generate initial benchmark for January 2000.

The analysis will be conducted on both the entire sample of listed securities in all ten liquidity deciles from each securities market. Liquidity deciles are determined by dividing the total number of securities in each market into 10 groups, based on their monthly trading turnover. Table 2 below lists the 34 securities exchanges studied by this research.

Table 2 List of Securities Exchanges Covered by this Research

Index Securities Exchange Home Country 1 American Stock Exchange U.S. 2 Australian Securities Exchange Australia 3 Bombay Stock Exchange India 4 Cairo Stock Exchange Egypt 5 Colombo Stock Exchange Sri Lanka 6 Copenhagen Stock Exchange Denmark 7 Deutsche Boerse-Xetra Germany 8 Euronext Amsterdam Netherlands 9 Euronext Brussels Belgium 10 Euronext Paris France 11 Euronext Portugal Portugal 12 Hong Kong Stock Exchange China 13 Istanbul Stock Exchange Turkey 14 Jakarta Stock Exchange Indonesia 15 Johannesburg Stock Exchange South Africa 16 Korea Stock Exchange Korea 17 Kuala Lumpur Stock Exchange Malaysia 18 London Stock Exchange U.K. 19 Madrid Stock Exchange Spain 20 Milan Stock Exchange Italy 21 NASDAQ U.S. 22 National Stock Exchange, India India 23 New York Stock Exchange U.S. 24 New Zealand Stock Exchange New Zealand 25 Oslo Bors Norway 26 Shanghai Stock Exchange China 27 Shenzhen Stock Exchange China 28 Singapore Stock Exchange Singapore 29 Stock Exchange of Thailand Thailand 30 Stockholm Borsen Sweden 31 Swiss Stock Exchange Switzerland 32 Taiwan Stock Exchange` China 33 Tokyo Stock Exchange Japan 34 Toronto Stock Exchange Canada

2.3.2 Ramping Alert Incidence The Ramping Alert applied in this study is programmed in the ALICE Language, which is the proprietary language from the SMARTS Real-time Securities Market Surveillance Platform. The algorithm of the Ramping Alert described below has been implemented at more than 30 national securities markets and regulators.

Benchmark Period and Threshold For date T, a historical price change distribution for the past month (the benchmarking period) is created for each security. The observations in this distribution are sampled wherever on market trades occur throughout the benchmarking period. Fifteen minutes after the market opens, we calculate the percentage change between the trade price and the true price 15 minutes earlier. True price is defined as (1) the previous trade price; or (2) the best bid (offer) price at time t-15 minutes if the previous trade price is below (above) the best (offer) price at t-15 minutes.

Finally, we take the absolute value of the calculated percentage price change and add it to the historical distribution.

At the end of the benchmarking period, we check the number of observations from each security’s historical price change distribution. If there are more than 50 observations, then we set the ramping price change threshold for that security as the 99% histogram distribution cut-off. If there are 50 or less observations from the distribution, then we determine the ramping price change threshold for that security as per the following Preset Price Change Threshold Table:

Last Trade Price of the Security Percentage Threshold Price From ($) Price To ($) 0.00 0.10 20% 0.11 0.25 15% 0.26 0.50 12% 0.51 1.00 10% 1.01 5.00 8% 5.01 10.00 5% 10.01 10000000.00 3%

The purpose of the benchmark process is to identify the top 1 percent least frequent or unrepresentative price changes for a security during the benchmark period. Assuming that there are approximately 20 trading days in a month and 100 trades in each trading day (assuming 6 trading hours per day), there are approximately 2000 price change observations each month. If these observations are sorted, the largest 20 price change (or 1 percent of the distribution) can be identified. The value of the 20th price change is where the threshold for ramping for that security is set. For example, if the 20th highest price change for BHP Billiton is 0.5% during the September, then the security is deemed to have been subject to ramping if the return in the last 15 minutes of 31 October was greater than 0.5%.

Conditions for Marking the Close Alert After market i closes on date t, for each security, we trigger a Marking the Close Alert if the absolute percentage difference between the closing price and the true price 15 minutes prior is greater than the ramping price change threshold for that security3.

Conditions for Reversal the Next Morning Alert On date T+1, for each security, we trigger a ramping alert if the following conditions are satisfied:

3 It is also true that for some illiquid securities, the time when the closing price is determined could be mid day or even earlier 45

(1) if there was a Marking the Close Alert triggered for that security on date T; and (2) During the first 15 minutes of trading on date T+1 or among the first ten trades on date T+1 (i.e., considering illiquid securities may not be traded during the first 15 minutes), if the Marking the close Alert triggered on date T is for driving up (down) closing price by X%, at least one trade occurs at price P that is below (above) the closing price on date T by more X% or more (a.k.a., Reversal the Next Morning).

The above algorithms are run daily across the 34 securities exchanges over 6 years (2000-2006) to derive the annual alert incidence of daily ramping manipulation. The annual alert incidence of daily ramping manipulation per security for market i in year Y is calculated as the number of ramping alerts triggered for all securities from market i in year Y normalized by the average number of listed securities per month for the same market over the same period. The annual alert incidences of daily ramping manipulation per security per decile for market i in year Y is also calculated in the same manner.

2.3.3 Time-Weighted Quoted Spread and Volume-Weighted Effective Spread 2.3.3.1 Annual Average Daily Quoted Spread (Time-Weighted) To calculate the Quoted Spread for each security, the following formula is used:

0.5

The time weight was calculated by taking the time that each spread existed during a trading day. A summation of the changes in spreads multiplied by the time it was available is created for each security for each trading day using the following:

where

To obtain the final estimate of Annual Average Daily Time-Weighted Quoted Spread for market i in year Y, the time weighted quoted spreads are averaged across all securities over all trading days for market i in year Y. Negative spreads and instances where one side of the spread was absent were removed from the sample. The Annual Average Daily Time-Weighted Quoted Spread per decile is determined in the same manner.

2.3.3.2 Annual Average Daily Effective Spread (Volume-Weighted) To calculate the Effective Spread for each security, the following formula is used:

, , 2 , ,

where

Di,t = the trade direction (D = 1 for buyer initiated trades with trade price above the midpoint price and D = -1 for seller initiated trades with trade price below the midpoint price)

The volume weight was calculated by taking the volume of each trade as a proportion of the total daily traded volume for each security. A summation of the changes in spreads multiplied by the volume weights is created for each security for each trading day using the following formula:

where

To obtain the final estimate of Annual Average Daily Effective Spread for market i in year Y, the volume weighted effective spreads are averaged across all securities over all trading days for market i in year Y. Negative spreads and instances where the midpoint was absent were removed from the sample. The Annual Average Daily Effective Volume-Weighted Effective Spread per decile is determined in the same manner.

2.3.4 Descriptive Statistics 2.3.4.1 Spreads The descriptive statistics for the average Quoted Spread and Effective Spread per security per year across the 34 securities exchanges over the 6 years period (2000- 2005) are presented in Table 2 below.

Table 3 Descriptive Statistics for average Quoted Spread and Effective Spread per security per year across 34 Markets for the period 2000-2005

Panel A: Moments for Raw Spreads Quoted Spread Effective Spread Mean 6.99% 6.64% Std. Dev 0.4994 0.1383 Skewness 1.2689 5.1899 Kurtosis 1.7947 28.7423 No. of Observations 204 204

Panel B: Moments for Natural Log of Spreads Quoted Spread Effective Spread Mean -3.1555 -3.4645 Std. Dev 1.1513 1.2172 Skewness -0.6831 -0.1480 Kurtosis -0.3900 0.9826 No. of Observations 204 204

It can be seen that the average Quoted Spread and Effective Spread per security per year across 34 markets for the period 2000-2005 are demonstrably non-normal. For example, the effective spread (“ES”) has mean ( "" ) of 6.64% and standard deviation ("") of 0.1383 with skewness of 5.1899 and kurtosis of 28.7423. After a natural log transform, we observe the distribution of ln to be approximately normal ( = -3.4645 and = 1.2172) with skewness of -0.1480 and kurtosis of 0.9826. The same is also observed for quoted spread. Figure 7 below represents the histogram for the two spread measures before and after the natural log 49

transform. Using the properties of the lognormal distribution and assuming exact log-

. normality for our observations, an estimator of would be . Our data ... yields such an estimate of = 6.56%. This figure differs from of 6.64% because our sample differs slightly from a pure lognormal distribution.

Figure 7 Histograms for Spread Measure before and after log transform

2.3.4.2 Sample Mean of Annual Average Daily Spreads by Deciles Table 4 below represents the sample mean of annual average daily quoted and effective spreads for each of the ten deciles.

Table 4 Sample Mean of Spreads by Deciles

Quoted Spread Effective Spread Decile 1 0.73% 2.19% Decile 2 1.28% 2.88% Decile 3 1.85% 3.26% Decile 4 2.56% 7.90% Decile 5 3.72% 5.42% Decile 6 5.36% 5.23% Decile 7 7.52% 7.21% Decile 8 10.89% 7.70% Decile 9 15.02% 10.91% Decile 10 20.98% 13.56%

Average Annual Quoted spread ranges from 0.73% for the most liquid decile to 20.98% for the most illiquid decile. Effective Spread varies from 2.19% to 13.56% for the 10 liquidity deciles. Overall, less liquid deciles exhibit an order of magnitude higher average annual daily spreads.

2.3.4.3 Ramping Alerts Incidence The descriptive statistics for the average Ramping Alerts Incidence per security across the 34 securities exchanges and 6 year period (2000-2005) are presented in Table 5 below.

Table 5 Descriptive Statistics for Ramping Alerts Incidence across 34 Markets for the period 2000-2005

Panel A: Moments for Raw Ramping Alerts Incidence Mean 0.21

Std. Dev 0.1841 Number of Observations 204

Panel B: Average Ramping Alerts Incidence per security per Market Market Mean American Stock Exchange 0.19 Australian Securities Exchange 0.25 Bombay Stock Exchange 0.44 Cairo Stock Exchange 0.14 Colombo Stock Exchange 0.16 Copenhagen Stock Exchange 0.17 Deutsche Boerse-Xetra 0.10 Euronext Amsterdam 0.32 Euronext Brussels 0.35 Euronext Paris 0.30 Euronext Portugal 0.37 Hong Kong Stock Exchange 0.07 Istanbul Stock Exchange 0.03 Jakarta Stock Exchange 0.21 Johannesburg Stock Exchange 0.07 Korea Stock Exchange 0.02 Kuala Lumpur Stock Exchange 0.40 London Stock Exchange 0.32 Madrid Stock Exchange 0.15 Milan Stock Exchange 0.23 NASDAQ 0.08 National Stock Exchange, India 0.37 53

Panel B: Average Ramping Alerts Incidence per security per Market Market Mean American Stock Exchange 0.19 New York Stock Exchange 0.26 New Zealand Stock Exchange 0.28 Oslo Bors 0.17 Shanghai Stock Exchange 0.03 Shenzhen Stock Exchange 0.06 Singapore Stock Exchange 0.36 Stock Exchange of Thailand 0.13 Stockholm Borsen 0.23 Swiss Stock Exchange 0.15 Taiwan Stock Exchange` 0.45 Tokyo Stock Exchange 0.16 Toronto Stock Exchange 0.22

It can be seen that the average ramping alerts incidence per security for the period 2000-2005 ranges from 0.02 to 0.45 across the 34 markets, which leads to an overall average ramping alerts incidence per security of 0.214 with a standard deviation of 18.41%.

4 0.21 is the grand mean of ramping incidents per security across market and across 6 years. At the security-day level (assuming 250 trading days per year), this is an alert incidence of only 0.001. Hence, security-day analysis is impractical. 54

2.3.5 Limitations of Research Design Table 6 below represents the annual average alert incidences of daily ramping manipulation by security per decile across 34 securities exchanges over 6 years.

Table 6 Sample Mean of Alerts Incidence by Deciles

Average Alert Incidence

Decile 1 0.31 Decile 2 0.27 Decile 3 0.22 Decile 4 0.23 Decile 5 0.23 Decile 6 0.22 Decile 7 0.19 Decile 8 0.16 Decile 9 0.12 Decile 10 0.13

It can be seen that the annual average alerts incidence of daily ramping manipulation by security per decile is gradually decreasing from 0.31 in decile 1 to 0.13 in decile 10. We would expect manipulations to be highest in the moderate liquidity deciles. In the most liquid stocks, the capital required to ramp a security is too extensive. In the least liquid securities, detection of manipulation activities is too likely.

Our highest alert incidence results could be caused by alert detection errors in deciles 1 and 2. Ramping alerts are not equivalent to genuine ramping manipulation cases. As a matter of fact, much surveillance workflow applied by those securities exchanges the author has designed alerts for is designed to validate alerts triggered based on a variety of other information. It is quite common that a big proportion of alerts triggered each day are false positives that can be explained by a legitimate reason. And it has become common practice to adjust various attributes of alerts (e.g., alerting

conditions, thresholds, etc.) to have alerts issued more or less often based on the client exchanges’ needs.

So the problem of false positives in ramping alerts is well known and likely to be highly correlated with trading volume. The securities in higher liquidity-deciles are traded a lot more frequently than those from lower liquidity-deciles. Due to the nature of design of surveillance alerts, it would be expected there is greater opportunity of false alerts to be triggered for securities from higher liquidity-deciles than for securities from lower liquidity-deciles. This probably explains the declining monotonic mean AI statistics from decile 1 to decile 10.

We should like to point out however that this false positive AI data collection problem biases our findings against (not in favour of) our research hypothesis of a positive spread-AI relationship. When mean reversion occurs due to an equilibrating response of resilient limit order books to random shocks in the absence of informational arrivals rather than the existence of a manipulator, liquidity providers would tend to tighten the spreads. Hence false positives in our AI data collection would falsify our hypothesis of a positive Spread-AI relationship, not the reverse.

An explanation for the absolute quantity of ramping alerts triggered for each securities exchange included in the data sample is outside the scope of the present research. Chapter 2 is intended to explain the impact of cross-sectional and time-series variations of alerts incidence on relative spreads across those 34 exchanges over 6 years. Chapter 3 will examine the drivers of alert incidence.

2.4 Empirical Specification We hypothesize the theoretical relation between relative spread and ramping alert incidence as

, , ,

and transform for estimation to the regression relation as

, , ,̃

Equation includes a normally distributed error term, ̃, . Recall that , is also subject to observational errors of two types, and these are the observational error on informational arrivals at time t and observational errors on Ramping Alert detection in market i.

To investigate the role of observational errors, we analyse the 204 time-series cross- section observations with a Hausman specification test in the Random Effects Model.

, , ,̃ where

, = the relative spread measure of market i at time t

= the constant that would be estimated

= the correlation coefficient between relative spread and Ramping

Alert incidence

, = the average number of Ramping Alert incidence of market i at time t

= Observational error on information arrivals at time t

= Observational error on Ramping Alert detection in market i

̃, = The residual error term

When Cov(,,) = 0 and Cov(,,) = 0, the Ramping Alert incidence regressor is orthogonal to observational errors. In this circumstance, autoregressive OLS estimation is appropriate. We employ the Da Silva model to analyse those cases. On

the other hand, when Hausman specification test rejects ( : Cov(,,) = 0 and

: Cov,, 0), then we conclude that autoregressive OLS would be a misspecification because Ramping Alert incidence is observed with error (i.e., is itself stochastic). In this circumstance, fixed effect dummy variables are used to control for observation errors in measuring the arrival of information over time and the presence of a manipulator across securities exchanges. Hausman (1978) test statistic is distributed with 1 d.o.f.

We introduce fixed effects in stages to observe first the effect of cross-section dummy variables (Cross-Sectional Fixed Effects) and then both time series and cross-section control variables (Full Fixed Effects).

2.5 Empirical Results

2.5.1 Annual Average Daily Quoted Spread for All Deciles The random effects model results for annual average daily quoted spread for all deciles are presented in Table 7 below.

Table 7 Random Effects Model Results (Quoted Spread for All Deciles)

Cross Sectional Random Effects N 204 α -3.0600 ***5 (t= -15.96) β 0.0456 ** (t= 2.26) Hausman Test m= 4.08 (Pr > m: 0.0435) R-square 0.0246

Full Random Effects N 204 α -3.0651 *** (t= -15.23) β 0.0432 ** (t= 2.32) Hausman Test m= 2.96 (Pr > m: 0.0841) R-square 0.0260

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -4.2314 *** (t= -28.91) β 0.0411 ** (t= 2.02) Cross Sectional Dummies 31 of 34 are significant at 95% R-square 0.9259

Full Fixed Effects N 204 α -4.5003 *** (t= -32.44) β 0.0392 ** (t= 2.09) Cross Sectional Dummies 31 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9426

Pooling all ten deciles of securities across 34 markets over six years, a doubling of the alert incidence is responsible for a 4.41 percent (e.0432 -1) change of quoted raw spreads that average 6.99%. This aggregate relationship reflects ten deciles that

5 *** stands for 99% confidence level; ** stands for 95% confidence level; * stands for 90% confidence level 59

combine very distinctly different thickly versus thinly-traded securities. Below we perform a Chow test to discern how to separate the deciles for more precise analysis.

The pooling of thickly-traded deciles in liquidity deciles 1-5 and thinly traded stocks in deciles 6-10 is rejected by a Chow test (F 10.09 with p-value less than 0.01)6. Consequently, we report and discuss the regressions for several subsets of the ten liquidity deciles.

A second econometric issue is that the Hausman specification test indicates these estimates are potentially subject to simultaneity bias. In particular, in estimating all ten deciles of securities, the Hausman specification test rejects H0: Cov(AI, w) = 0 in the cross-sectional random effects model at p-value 0.04 and also rejects Cov(AI, v) = 0 in the cross-sectional time-series random effects model at p-value 0.08. This means that AI is not orthogonal to either set of observational errors affecting spreads, and therefore AI can not be considered exogenous as a determinant of quoted spreads.

Because alert incidence is not orthogonal to cross-sectional or time-series observational errors, we introduce 34 cross-sectional and 6 time-series dummy variables. The inclusion of cross-sectional fixed effects alone raises R-squared from .026 to .926. Thirty-one of 34 exchanges have statistically significant fixed effects. Assuming exact log normality, the mean spread is e-3.15+0.5*1.15^2 = 0.068 while the sample spread is 0.070. Even after controlling for these fixed effects, alert incidence (AI) is statistically significantly positively related to quoted spreads. A doubling of the alert incidence is responsible for a 4.2 percent (e.0411 -1) change of quoted raw spreads.

6 The Chow (1960) test is used to ascertain whether a number of sets of observations can be pooled into one larger set for regression purposes. In our case we have two sets: thickly traded securities from liquidity deciles 1 through 5 and thinly-traded securities from liquidity deciles 6 through 10. To compute the test statistic (“C”) we need the sum of squared residuals from running a regression using k terms on the pooled set () of n, together with the sum of squared residuals from running a regression using the same model on the subsets ( ).

/ = 0.69/0.07 = 10.09 /

The Chow test statistic C has an Fk, n1+n2-2k distribution. For further discussion of the Chow test see Gujarati (2003), pp.275-279. 60

Finally, we re-estimate the model with both cross-sectional and time-series fixed effects. Again, 31 of 34 cross-section dummy variables as well as 4 of 5 time-series dummy variables are statistically significant. With the additional fixed effects, R- squared increases from 0.926 to 0.942 and = 9.103 with p-value less than 0.017 for inclusion of these time-series dummy variables as regressors. With all the fixed effects included, AI is still significantly positively related to quotes spreads. A doubling of the alert incidence is responsible for 4.0 percent (e.0392 -1) change of quoted raw spreads.

7 The F test here is used to ascertain whether adding the time-series fixed effects will significantly change the model fitness. To compute the F statistics, we need the R-squared for the full fixed effects model, the R-squared for the Cross-sectional Fixed Effects, for J linear (β=0) restrictions, and n-k degrees of freedom.

/ = 0.003/0.0004 = 9.103 / 61

2.5.2 Annual Average Daily Quoted Spreads (Thickly-traded Deciles) 2.5.2.1 Decile 1 The random effects model results for annual average daily quoted spread for decile 1 are presented in Table 8 below.

Table 8 Random Effects Model Results (Quoted Spread for Decile 1)

Cross Sectional Random Effects N 204 α -5.2730 *** (t= -39.81) β 0.0016 (t= 0.14) Hausman test m= 1.41 (Pr > m: 0.2345) R-square 0.0001

Full Random Effects N 204 α -5.2730 *** (t= -31.20) β 0.0014 (t= 0.18) Hausman test m= 1.26 (Pr > m: 0.2623) R-square 0.0002

Da Silva Model N 204 Order of MA 4 α -5.2746*** (t= -30.04) β 0.001** (t= 2.04) RMSE 7.6335

Using all securities from decile 1 across 34 markets over six years, a doubling of the alert incidence is responsible for an insignificant 0.14 percent (e.0014 -1) change of quoted raw spreads that average less than 1 percent (0.73%) in stocks this liquid.

In estimating all securities from decile 1, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI,v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of quoted spreads in decile 1.

Since neither the cross-sectional random effects model nor the full cross-sectional time-series random effects model yields to a significant relationship with AI, we now look at the results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a small but significant 0.10% percent (e.001 -1) change of quoted raw spreads in decile 1.

2.5.2.2 Decile 2 The random effects model results for annual average daily quoted spread for decile 2 securities are presented in Table 9 below.

Table 9 Random Effects Model Results (Quoted Spread for Decile 2)

Cross Sectional Random Effects N 204 α -4.6814 *** (t= -34.68) β 0.0007 (t= 0.06) Hausman test m= 1.47 (Pr > m: 0.2123) R-square 0.0001

Full Random Effects N 204 α -4.6964 *** (t= -27.87) β -0.0048 (t= -0.50) Hausman test m= 1.31 (Pr > m: 0.2517) R-square 0.0002

Da Silva Model N 204 Order of MA 3 α -0.4665*** (t= -28.07) β 0.006** (t= 5.74) RMSE 7.2007

Using all securities from decile 2 across 34 markets over six years, the cross-sectional time-series random effects model indicates an insignificant correlation coefficient of - 0.0048 with AI, as in decile 1.

In estimating all securities from decile 2, the Hausman specification test does not

reject H0: Cov (AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI,v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can again be considered exogenous as a determinant of quoted spreads, as in decile 1.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we again look at the results of the Da Silva Autoregressive Moving Average Model. At the moving

average order of 3 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a small but significant 0.60 percent (e.006 -1) change of quoted raw spreads in decile 2. The first two most liquid deciles are equivalent. Spreads are tiny and yet, decreased ramping incidence leads to lower spreads even in the presence of full fixed effects.

2.5.2.3 Decile 3 The random effects model results for annual average daily quoted spread for decile 3 securities are presented in Table 10 below.

Table 10 Random Effects Model Results (Quoted Spread for Decile 3)

Cross Sectional Random Effects N 204 α -4.3351 *** (t= -30.16) β -0.0041 (t= -0.31) Hausman test m= 6.88 (Pr > m: 0.0087) R-square 0.0005

Full Random Effects N 204 α -4.3468 *** (t= -25.40) β -0.0087 (t= -0.82) Hausman test m= 8.36 (Pr > m: 0.0038) R-square 0.0034

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -5.1596 *** (t= -32.50) β -0.0104 (t= -0.79) Cross Sectional Dummies 30 of 34 are significant at 95% R-square 0.8567

Full Fixed Effects N 204 α -5.4876 *** (t= -41.93) β -0.0131 (t= -1.23) Cross Sectional Dummies 30 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9147

Using all securities from decile 3 across 34 markets over six years, the cross-sectional time-series random effects model indicates an insignificant correlation coefficient of - 0.0087 with AI.

Here, unlike decile 1 and 2, the Hausman specification test indicates these estimates are potentially subject to simultaneity bias. In particular, in estimating with all securities in decile 3, the Hausman specification test rejects H0: Cov (AI, w) = 0 in the cross-sectional random effects model and also rejects Cov(AI, v) = 0 in the cross- sectional time-series random effects model. This means that AI is not orthogonal to 66

either set of observational errors affecting spreads, and therefore AI cannot be considered exogenous as a determinant of quoted spreads.

Finally, we re-estimate the model with both cross-sectional and time-series fixed effects. Again, it indicates an insignificant correlation coefficient of -0.013 with AI for decile 3.

2.5.2.4 Decile 4 The random effects model results for annual average daily quoted spread for decile 4 securities are presented in Table 11 below.

Table 11 Random Effects Model Results (Quoted Spread for Decile 4)

Cross Sectional Random Effects N 204 α -4.0537 *** (t= -25.98) β -0.0094 (t= -0.64) Hausman test m= 4.60 (Pr > m: 0.0320) R-square 0.0020

Full Random Effects N 204 α -4.0532 *** (t= -22.30) β -0.0092 (t= -0.74) Hausman test m= 8.36 (Pr > m: 0.0038) R-square 0.0027

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -4.9232 *** (t= -29.35) β -0.0171 (t= -1.12) Cross Sectional Dummies 29 of 34 are significant at 95% R-square 0.8568

Full Fixed Effects N 204 α -5.2521 *** (t= -36.95) β -0.0145 (t= -1.13) Cross Sectional Dummies 30 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9094

Using all securities from decile 4 across 34 markets over six years, the cross-sectional time-series random effects model indicates again insignificant correlation coefficient of -0.0092 with AI.

The Hausman specification tests again indicate these estimates are potentially subject to simultaneity bias. In particular, in estimating with all securities in decile 4, the

Hausman specification test rejects H0: Cov(AI, w) = 0 in the cross-sectional random effects model and also rejects Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is not orthogonal to either set of observational 68

errors affecting spreads, and therefore AI cannot be considered exogenous as a determinant of quoted spreads.

Because alert incidence is not orthogonal to cross-sectional or time-series observational errors, we introduce 34 cross-sectional and 6 time-series dummy variables. Like decile 3, the inclusion of cross-sectional fixed effects alone indicates an insignificant correlation coefficient of -0.017 with AI and when we re-estimate the model with both cross-sectional and time-series fixed effects, again, it suggests an insignificant correlation coefficient of -0.014 with AI for decile 4.

2.5.2.5 Decile 5 The random effects model results for annual average daily quoted spread for decile 5 securities are presented in Table 12 below.

Table 12 Random Effects Model Results (Quoted Spread for Decile 5)

Cross Sectional Random Effects N 204 α -3.7629 *** (t= -21.92) β -0.0115 (t= -0.86) Hausman test m= 3.70 (Pr > m: 0.0583) R-square 0.0036

Full Random Effects N 204 α -3.7800 *** (t= -19.43) β -0.0177 (t= -1.51) Hausman test m= 4.688 (Pr > m: 0.0350) R-square 0.0112

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -4.7191 *** (t= -27.06) β -0.0159 (t= -1.15) Cross Sectional Dummies 29 of 34 are significant at 95% R-square 0.8667

Full Fixed Effects N 204 α -5.0725 *** (t= -32.58) β -0.0216 (t= -1.16) Cross Sectional Dummies 30 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9086

Using all securities from decile 5 across 34 markets over six years, the cross-sectional time-series random effects model again indicates an insignificant correlation coefficient of -0.0177 with AI.

As with deciles 3 and 4, the Hausman specification tests indicate these estimates are potentially subject to simultaneity bias. In particular, in estimating with all securities in decile 5, the Hausman specification test rejects H0: Cov(AI, w) = 0 in the cross- sectional random effects model and also rejects Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is not orthogonal to either set 70

of observational errors affecting spreads, and therefore AI cannot be considered exogenous as a determinant of quoted spreads in decile 5.

Because alert incidence is not orthogonal to cross-sectional or time-series observational errors, we introduce 34 cross-sectional and 6 time-series dummy variables. The inclusion of cross-sectional fixed effects alone indicates an insignificant correlation coefficient of -0.016 with AI while re-estimating the model with both cross-sectional and time-series fixed effects indicates an insignificant correlation coefficient of -0.022 with AI for decile 5.

2.5.3 Annual Average Daily Quoted Spreads (Thinly-traded Deciles) 2.5.3.1 Decile 6 The random effects model results for annual average daily quoted spread for decile 6 securities are presented in Table 13 below.

Table 13 Random Effects Model Results (Quoted Spread for Decile 6)

Cross Sectional Random Effects N 204 α -3.5268 *** (t= -18.45) β -0.0258* (t= -1.69) Hausman test m= 1.44 (Pr > m: 0.2308) R-square 0.0140

Full Random Effects N 204 α -3.5052 *** (t= -17.06) β -0.0182 (t= -1.30) Hausman test m= 0.42 (Pr > m: 0.5169) R-square 0.0083

Da Silva Model N 204 Order of MA 4 α -3.3621*** (t= -16.15) β 0.0319*** (t= 25.39) RMSE 6.6252

Using all securities from decile 6 across 34 markets over six years, the cross-sectional time-series random effects model indicates an insignificant correlation coefficient of - 0.0182 with AI.

In estimating with all securities from decile 6, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of quoted spreads in decile 6.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we look at the

results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a significant 3.25 percent (e.032 -1) change of quoted raw spreads in decile 6. Doubling AI from the mean in decile 6 of 0.21 to 0.42 increases quoted spreads by 3.25% ceteris paribus.

2.5.3.2 Decile 7 The random effects model results for annual average daily quoted spread for decile 7 securities are presented in Table 14 below.

Table 14 Random Effects Model Results (Quoted Spread for Decile 7)

Cross Sectional Random Effects N 204 α -3.2288 *** (t= -15.67) β -0.0130 (t= -1.00) Hausman test m= 0.25 (Pr > m: 0.6187) R-square 0.0050

Full Random Effects N 204 α -3.2106 *** (t= -14.50) β -0.0074 (t= -0.64) Hausman test m= 0.01 (Pr > m: 0.9123) R-square 0.0020

Da Silva Model N 204 Order of MA 3 α -3.1244*** (t= -13.96) β 0.0201*** (t= 16.90) RMSE 6.6559

Using all securities from decile 7 across 34 markets over six years, the cross-sectional time-series random effects model again indicates an insignificant correlation coefficient of -0.0074 with AI.

In estimating with all securities from decile 7, the Hausman specification test does not reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of quoted spreads in decile 7 as well.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we look at the results of the Da Silva Autoregressive Moving Average Model. At the moving

average order of 3 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a significant 2.02 percent (e.020 -1) change of quoted raw spreads in decile 7. Doubling the AI in decile 7 from the mean AI of 0.19 to 0.38 increases quoted spreads by 2%, ceteris paribus.

2.5.3.3 Decile 8 The random effects model results for annual average daily quoted spread for decile 8 securities are presented in Table 15 below.

Table 15 Random Effects Model Results (Quoted Spread for Decile 8)

Cross Sectional Random Effects N 204 α -2.8532 *** (t= -12.98) β 0.0054 (t= 0.46) Hausman test m= 1.01 (Pr > m: 0.3161) R-square 0.0011

Full Random Effects N 204 α -2.8421 *** (t= -12.21) β -0.0085 (t= 0.82) Hausman test m= 1.24 (Pr > m: 0.2652) R-square 0.0033

Da Silva Model N 204 Order of MA 4 α -2.8353*** (t= -12.24) β 0.0104*** (t= 18.71) RMSE 6.0701

Using all securities from decile 8 across 34 markets over six years, the cross-sectional time-series random effects model indicates an insignificant correlation coefficient of - 0.0085 with AI.

In estimating with all securities from decile 8, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of quoted spreads in decile 8.

average order of 4 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a significant 1.01 percent (e.010 -1) change of quoted raw spreads in decile 8. Doubling AI from 0.16 of its mean in decile 8 to 0.32 raises quoted spreads by 1%.

2.5.3.4 Decile 9 The random effects model results for annual average daily quoted spread for decile 9 securities are presented in Table 16 below.

Table 16 Random Effects Model Results (Quoted Spread for Decile 9)

Cross Sectional Random Effects N 204 α -2.5765 *** (t= -11.26) β -0.0038 (t= -0.33) Hausman test m= 2.58 (Pr > m: 0.1080) R-square 0.0005

Full Random Effects N 204 α -2.5505 *** (t= -10.75) β 0.0021 (t= 0.20) Hausman test m= 4.42 (Pr > m: 0.0350) R-square 0.0002

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -3.7753 *** (t= -22.81) β -0.001 (t= -0.09) Cross Sectional Dummies 32 of 34 are significant at 95% R-square 0.9254

Full Fixed Effects N 204 α -4.0258 *** (t= -24.64) β 0.0057 (t= 0.52) Cross Sectional Dummies 31 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9381

Using all securities from decile 9 across 34 markets over six years, the cross-sectional time-series random effects model still indicates no significant correlation coefficient of 0.0021 with AI.

The Hausman specification test indicates these estimates are potentially subject to simultaneity bias. In particular, in estimating with all securities in decile 9, the

errors affecting spreads, and therefore AI cannot be considered exogenous as a determinant of quoted spreads in decile 9.

Because alert incidence is not orthogonal to cross-sectional or time-series observational errors, we introduce 34 cross-sectional and 6 time-series dummy variables. Both the inclusion of cross-sectional fixed effects alone and the inclusion of cross-sectional and time-series fixed effects indicates insignificant correlation coefficient of -0.001 as well as 0.057 with AI for decile 9.

2.5.3.5 Decile 10 The random effects model results for annual average daily quoted spread for decile 10 securities are presented in Table 17 below.

Table 17 Random Effects Model Results (Quoted Spread for Decile 10)

Cross Sectional Random Effects N 204 α -2.1036 *** (t= -9.96) β 0.001 (t= 0.11) Hausman test m= 5.83 (Pr > m: 0.0158) R-square 0.0003

Full Random Effects N 204 α -2.0494 *** (t= -9.43) β 0.0108 (t= 1.00) Hausman test m= 8.63 (Pr > m: 0.0033) R-square 0.0049

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -3.2824 *** (t= -18.01) β 0.0044 (t= 0.43) Cross Sectional Dummies 30 of 34 are significant at 95% R-square 0.8856

Full Fixed Effects N 204 α -3.5052 *** (t= -18.36) β 0.0317 ** (t= 2.10) Cross Sectional Dummies 30 of 34 are significant at 95% Time Dummies 4 of 5 are significant at 95% R-square 0.9038

Using all securities from decile 10 across 34 markets over six years, the cross- sectional time-series random effects model indicates an insignificant correlation coefficient of 0.0021 with AI.

The Hausman specification test indicates these estimates are potentially subject to simultaneity bias. In particular, in estimating with securities from decile 10, the

errors affecting spreads, and therefore AI cannot be considered exogenous as a determinant of quoted spreads in decile 10.

Finally, we re-estimate the model with both cross-sectional and time-series fixed effects. The model suggests that alert incidence is responsible for a significant 3.25 percent (e.032 -1) change of quoted raw spreads in decile 10.

2.5.4 Summary of Quoted Spreads Results Decile 1 and 2 exhibit statistically significant but extremely small (1/10 of the 1 percent and 6/10 of the 1 percent) changes in quoted spreads as alert incidence doubles, in the presence of full fixed effects. The Quoted Spreads-Alert Incidence relationship is statistically insignificant in decile 3, 4 and 5.

In decile 6, 7 and 8, we have statistically orthogonal alert incidence driving, 3.25 percent, 2 percent and 1 percent change of spreads in the Da Silva Autoregressive Moving Average Model. Finally, in decile 10 we get a 3.25 percent in quoted spreads from a doubling of alert incidence, even in the presence of full fixed effects.

2.5.5 Annual Average Daily Effective Spreads (All Deciles) The random effects model results for annual average daily effective spread for all deciles are presented in Table 18 below.

Table 18 Random Effects Model Results (Effective Spread for All Deciles)

Cross Sectional Random Effects N 204 α -3.2457 *** (t= -16.53) β 0.1045 *** (t= 2.98) Hausman Test m= 1.08 (Pr > m: 0.2990) R-square 0.0422

Full Random Effects N 204 α -3.2289 *** (t= -15.62) β 0.1126 *** (t= 3.26) Hausman Test m= 0.23 (Pr > m: 0.6351) R-square 0.0500

Da Silva Model N 204 Order of MA 3 α -3.3142*** (t= -16.47) β 0.0718*** (t= 4.00) RMSE 1.7641

Pooling all ten deciles of securities across 34 markets over six years, a doubling of the alert incidence is responsible for an 11.92 percent (e.113 -1) change of effective spreads that average 6.64%. This aggregate relationship reflects ten deciles that combine very distinctly different thickly versus thinly-traded securities. The pooling of thickly- traded deciles in liquidity deciles 1-5 and thinly traded stocks in deciles 6-10 is rejected by a Chow test (F 12.36 with p-value less than 0.01). Consequently, we report and discuss the regressions for several subsets of the ten liquidity deciles.

In estimating all securities across 34 markets over six years, the Hausman specification test does not reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model for effective spreads. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads. 83

In both of the cross-sectional random effects model and the full cross-sectional time- series random effects model we find a significantly positively relationship with AI. In addition, in the Da Silva Autoregressive Moving Average Model, at the moving average order of 3 (which minimizes RMSE), the Da Silva Model suggests that alert incidence is responsible for a significant 7.43 percent (e.072 -1) change of effective spreads across 34 markets over 6 years.

2.5.6 Annual Average Daily Effective Spreads (Thickly-traded Deciles) 2.5.5.1 Decile 1 The random effects model results for annual average daily effective spread for decile 1 are presented in Table 19 below.

Table 19 Random Effects Model Results (Effective Spread for Decile 1)

Cross Sectional Random Effects N 204 α -4.7881 *** (t= -22.09) β 0.0297 (t= 1.01) Hausman test m= 0.82 (Pr > m: 0.3662) R-square 0.0051

Full Random Effects N 204 α -4.7756 *** (t= -21.55) β 0.0342 (t= 1.17) Hausman test m= 0.01 (Pr > m: 0.9179) R-square 0.0067

Da Silva Model N 204 Order of MA 4 α -4.8213*** (t= -20.96) β 0.0174 (t= 0.85) RMSE 1.1658

Using all securities from decile 1 across 34 markets over six years, an insignificantly positive correlation coefficient of 3.42 percent is found with AI and full random effects.

In estimating all securities from decile 1, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we now look 85

at the results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model also suggests an insignificantly positive correlation coefficient of 1.74 percent with AI in decile 1. Decile 1 (most liquid) securities do not exhibit the efficiency-integrity relationship. Below we will discuss this negative finding later in this chapter.

2.5.5.2 Decile 2 The random effects model results for annual average daily effective spread for decile 2 are presented in Table 20 below.

Table 20 Random Effects Model Results (Effective Spread for Decile 2)

Cross Sectional Random Effects N 204 α -4.4626 *** (t= -24.84) β 0.0481** (t= 2.33) Hausman test m= 1.27(Pr > m: 0.2600) R-square 0.0262

Full Random Effects N 204 α -4.4717 *** (t= -22.45) β 0.0447 ** (t= 2.23) Hausman test m= 1.46 (Pr > m: 0.2267) R-square 0.0241

Da Silva Model N 204 Order of MA 2 α -4.5228 *** (t= -22.85) β 0.0258 (t= 1.35) RMSE 1.0671

Using all securities from decile 2 across 34 markets over six years, a doubling of the alert incidence is responsible for a 4.57 percent (e.0447 -1) change of effective spreads that average 2.88%.

In estimating all securities from decile 2, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 2.

Though both of the cross-sectional random effects model and the full cross-sectional time-series random effects model suggest a significantly positively relationship with 87

AI, we will still have look at the results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 2 (which minimizes RMSE), the Da Silva Model indicates an insignificantly positive correlation coefficient of 2.58 percent with AI across 34 markets over 6 years in decile 2.

2.5.5.3 Decile 3 The random effects model results for annual average daily effective spread for decile 3 are presented in Table 21 below.

Table 21 Random Effects Model Results (Effective Spread for Decile 3)

Cross Sectional Random Effects N 204 α -4.2541 *** (t= -23.90) β 0.0448 (t= 1.58) Hausman test m= 1.23 (Pr > m: 0.2669) R-square 0.0122

Full Random Effects N 204 α -4.2460 *** (t= -20.24) β 0.0480 (t= 1.61) Hausman test m= 0.83 (Pr > m: 0.3631) R-square 0.0146

Da Silva Model N 204 Order of MA 2 α -4.3282 *** (t= -20.92) β 0.0154 (t= 1.06) RMSE 1.0671

Using all securities from decile 3 across 34 markets over six years, the full random effects model suggests an insignificantly positive correlation coefficient of 4.80 percent with AI.

In estimating all securities from decile 3, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 3.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we now look 89

at the results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 2 (which minimizes RMSE), the Da Silva Model also suggests an insignificantly positive correlation coefficient of 1.54 percent with AI in decile 3.

2.5.5.4 Decile 4 The random effects model results for annual average daily effective spread for decile 4 are presented in Table 22 below.

Table 22 Random Effects Model Results (Effective Spread for Decile 5)

Cross Sectional Random Effects N 204 α -4.0677 *** (t= -22.84) β 0.0353 (t= 1.57) Hausman test m= 0.91 (Pr > m: 0.3411) R-square 0.0120

Full Random Effects N 204 α -4.0524 *** (t= -20.68) β 0.0411* (t= 1.86) Hausman test m= 0.34 (Pr > m: 0.5575) R-square 0.0168

Da Silva Model N 204 Order of MA 4 α -4.0594 *** (t= -20.01) β 0.0384** (t= 2.26) RMSE 1.2863

Using all securities from decile 4 across 34 markets over six years, the full random effects model suggests a weakly significant (at 90% level) correlation coefficient of 4.11 percent with AI, which means alert incidence is responsible for a 4.19 percent (e.041 -1) change of effective spreads that average 7.90% in decile 4.

In estimating all securities from decile 4, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 4.

Now we will have look at the results of the Da Silva Autoregressive Moving. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests 91

alert incidence is responsible for a 3.87 percent (e.038 -1) change of effective spreads in decile 4. Doubling AI from its in decile 4 of 0.25 increases effective spreads by 3.87%.

2.5.5.5 Decile 5 The random effects model results for annual average daily effective spread for decile 5 are presented in Table 23 below.

Table 23 Effects Model Results (Effective Spread for Decile 5)

Cross Sectional Random Effects N 204 α -3.8324 *** (t= -22.71) β 0.0343** (t= 2.00) Hausman test m= 2.52 (Pr > m: 0.1127) R-square 0.0193

Full Random Effects N 204 α -3.8328 *** (t= -20.69) β 0.0342** (t= 2.06) Hausman test m= 1.99 (Pr > m: 0.1585) R-square 0.0206

Da Silva Model N 204 Order of MA 2 α -3.9197 *** (t= -21.81) β 0.027*** (t= 3.21) RMSE 1.8727

Using all securities from decile 5 across 34 markets over six years, the full random effects model suggests that alert incidence is responsible for a 3.45 percent (e.034 -1) change of effective spreads that average 4.90% in decile 5.

In estimating all securities from decile 5, the Hausman specification test does not reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 5.

Now we will have look at the results of the Da Silva Autoregressive Moving. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests

alert incidence is responsible for 3.04 percent (e.03 -1) change of effective spreads in decile 5.

2.5.7 Annual Average Daily Effective Spreads (Thinly-traded Deciles) 2.5.6.1 Decile 6 The random effects model results for annual average daily effective spread for decile 6 are presented in Table 24 below.

Table 24 Effects Model Results (Effective Spread for Decile 6)

Cross Sectional Random Effects N 204 α -3.5403 *** (t= -18.71) β 0.0647*** (t= 2.88) Hausman test m= 3.02 (Pr > m: 0.0821) R-square 0.0395

Full Random Effects N 204 α -3.5233 *** (t= -18.19) β 0.0706*** (t= 3.15) Hausman test m= 0.37 (Pr > m: 0.5434) R-square 0.0468

Da Silva Model N 204 Order of MA 4 α -3.5193 *** (t= -17.20) β 0.0720 *** (t= 4.43) RMSE 1.8727

Using all securities from decile 6 across 34 markets over six years, the full random effects model suggests that alert incidence is responsible for a 7.32 percent (e.071 -1) change of effective spreads that average 5.92% in decile 6.

In estimating all securities from decile 6, the Hausman specification test does not reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 6.

Now we will have look at the results of the Da Silva Autoregressive Moving. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests

alert incidence is responsible for a significant 7.47 percent (e.072 -1) change of effective spreads in decile 6.

2.5.6.2 Decile 7 The random effects model results for annual average daily effective spread for decile 7 are presented in Table 25 below.

Table 25 Effects Model Results (Effective Spread for Decile 7)

Cross Sectional Random Effects N 204 α -3.3235 *** (t= -17.09) β 0.0411** (t= 2.35) Hausman test m= 0.99 (Pr > m: 0.3205) R-square 0.0265

Full Random Effects N 204 α -3.2957 *** (t= -16.19) β 0.0504*** (t= 2.85) Hausman test m= 0.00 (Pr > m: 0.9662) R-square 0.0387

Da Silva Model N 204 Order of MA 4 α -3.2021 *** (t= -15.63) β 0.0786 *** (t= 57.99) RMSE 1.5682

Using all securities from decile 7 across 34 markets over six years, the full random effects model suggests that alert incidence is responsible for a 5.12 percent (e.050 -1) change of effective spreads that average 7.21% in decile 7.

In estimating all securities from decile 7, the Hausman specification test does not reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 7.

Now we will have look at the results of the Da Silva Autoregressive Moving. At the moving average order of 4 (which minimizes RMSE), the Da Silva Model suggests

alert incidence is responsible for a significant 8.18 percent (e.0786 -1) change of effective spreads in decile 7.

2.5.6.3 Decile 8 The random effects model results for annual average daily effective spread for decile 8 are presented in Table 26 below.

Table 26 Effects Model Results (Effective Spread for Decile 8)

Cross Sectional Random Effects N 204 α -3.0862 *** (t= -14.44) β 0.0217 (t= 1.53) Hausman test m= 0.09 (Pr > m: 0.7677) R-square 0.0115

Full Random Effects N 204 α -3.0731 *** (t= -13.96) β 0.0251 (t= 1.61) Hausman test m= 0.12 (Pr > m: 0.7339) R-square 0.0163

Da Silva Model N 204 Order of MA 4 α -2.9887 *** (t= -13.17) β 0.0495 *** (t= 63.96) RMSE 6.2391

Using all securities from decile 8 across 34 markets over six years, the full random effects model suggests that an insignificantly positive correlation coefficient of 2.51 percent in decile 8.

In estimating all securities from decile 8, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 8.

Since neither cross-sectional random effects model nor the full cross-sectional time- series random effects model yields to a significant relationship with AI, we now look at the results of the Da Silva Autoregressive Moving Average Model. At the moving

average order of 4 (which minimizes RMSE), the Da Silva Model also suggests alert incidence is responsible for a significant 5.13 percent (e.050 -1) change of effective spreads that average 7.70% in decile 8.

100

2.5.6.4 Decile 9 The random effects model results for annual average daily effective spread for decile 9 are presented in Table 27 below.

Table 27 Effects Model Results (Effective Spread for Decile 9)

Cross Sectional Random Effects N 204 α -0.4588 ** (t= -2.04) β 0.0227* (t= 1.91) Hausman test m= 0.65 (Pr > m: 0.4193) R-square 0.0178

Full Random Effects N 204 α -0.4352 * (t= -1.88) β 0.0281*** (t= 2.46) Hausman test m= 1.87 (Pr > m: 0.1713) R-square 0.0291

Da Silva Model N 204 Order of MA 4 α -0.3289 *** (t= -2.40) β 0.05216 *** (t= 71.46) RMSE 8.3060

Using all securities from decile 8 across 34 markets over six years, the full random effects model suggests that an insignificantly positive correlation coefficient of 2.81 percent in decile 9.

In estimating all securities from decile 9, the Hausman specification test does not

reject H0: Cov(AI, w) = 0 in the cross-sectional random effects model and does not reject Cov(AI, v) = 0 in the cross-sectional time-series random effects model. This means that AI is orthogonal to both sets of observational errors affecting spreads, and therefore AI can be considered exogenous as a determinant of effective spreads in decile 9.

In addition, we also look at the results of the Da Silva Autoregressive Moving Average Model. At the moving average order of 4 (which minimizes RMSE), the Da

101

Silva Model also suggests alert incidence is responsible for a significant 5.33 percent (e.052 -1) change of effective spreads that average 10.91% in decile 9.

102

2.5.6.5 Decile 10 The random effects model results for annual average daily effective spread for decile 10 securities are presented in Table 28 below.

Table 28 Random Effects Model Results (Effective Spread for Decile 10)

Cross Sectional Random Effects N 204 α -2.2066 *** (t= -8.74) β 0.0059 (t= 0.37) Hausman test m= 6.74 (Pr > m: 0.0094) R-square 0.0007

Full Random Effects N 204 α -2.1582 *** (t= -8.26) β 0.0146 (t= 0.87) Hausman test m= 8.63 (Pr > m: 0.0033) R-square 0.0037

Da Silva Model N/A N/A Cross Sectional Fixed Effects N 204 α -3.5639 *** (t= -12.34) β 0.0140 (t= 0.86) Cross Sectional Dummies 26 of 34 are significant at 95% R-square 0.8217

Full Fixed Effects N 204 α -3.8307 *** (t= -12.57) β 0.0333 ** (t= 1.93) Cross Sectional Dummies 27 of 34 are significant at 95% Time Dummies 3 of 5 are significant at 95% R-square 0.8333

Using all securities from decile 10 across 34 markets over six years, the cross- sectional time-series random effects model suggests an insignificantly positive correlation coefficient of 1.46 percent with AI.

The Hausman specification tests here indicate these estimates are potentially subject to simultaneity bias. In particular, in estimating all ten deciles of securities, the

errors affecting spreads, and therefore AI can not be considered exogenous as a determinant of quoted spreads in decile 10.

Finally, we re-estimate the model with both cross-sectional and time-series fixed effects and it suggests that alert incidence is responsible for a significant 3.36 percent (e.033 -1) change of effective spreads that average 13.56% in decile 10.

104

2.5.8 Summary of Results for Effective Spreads In the first three (most liquid) deciles only decile 2 exhibits a statistically significant effect of alert incidence (4.5% or 287 basis points effective spreads from a doubling of the alert incidence). In moderate liquidity deciles 4, 5, 6 and 7, alert incidence is statistically significantly positively related to effective spreads. The elasticities vary from 3.45 percent to 8.18 percent change in effective spreads averaging from 523 to 790 basis points for a doubling of the alert incidence.

In the least liquid deciles (8, 9 and 10), the effect of alert incidence is statistically insignificant until we examine moving average Da Silva models. For those three deciles, the elasticities of 5.13%, 5.33% and 3.36% corresponding to a doubling of the alert incidence increase effective spreads of 770, 1019 and 1356 basis points.

Section 2.6 will further interpret and discuss the above results.

105

2.6 Discussion of Results We find in examining the surveillance data across 34 exchanges that the efficiency- integrity relationship is much more sensitive when we account for actual trading prices (with effective spreads) rather than the quoted spreads on offer for doing tiny size. In 8 of 10 deciles, doubling alert incidence increases effective spreads from 3.3% to 7.3% where as in quoted prices, only two deciles (decile 6 and 10) show an effect on quoted spreads of this magnitude.

The price of immediacy, what liquidity demanders must pay to get substantial size done, in the face of a doubled incidence of ramping alerts increases substantially even though the quotes for trivial size are largely unchanged. So it is institutional clients seeking larger trades who would be expected to pressure exchanges to detect potential manipulators and exclude them from the marketplace.

As to the magnitudes, a doubling of ramping alert incidence from its mean in decile 4 of 0.23 to 0.46, raises the 790 basis points mean effective spread by approximately 4.00% (4.11% in the full random effects model and 3.87% in the Da Silva Autoregressive Moving Average Model). That means twice the average ramping incidence would be associated with 31.6 basis points of extra transaction costs. And the results in other moderately liquid deciles are similar. For example, in decile 6, 7.32% full random effects model and 7.47% Da Silva Model elasticities raise 523 basis points spreads -- i.e., 38.7 basis points.

In decile 10, a doubling of ramping alert incidence from its mean of 0.12 to 0.24, increases the mean effective spread of 1356 basis points by approximately 3.33 percent (from full fixed effects model) or 44.7 basis points. Other illiquid deciles exhibit similar results as well. For example, in decile 9, 5.33% elasticity in the Da Silva Autoregressive Moving Average Model raises the effective spread by 58.2 basis points. In decile 8, 5.13% elasticity in the Da Silva model raises effective spreads by 39.5 basis points. The elasticity of spreads with respect to alert incidence in decile 8, 9 and 10 are not as big as for moderately-liquid deciles. Decile 6, for example, exhibits an elasticity of 7.47% and decile 7 of 8.18% with more comparable basis points impact on effective spreads of 39.1 b.p and 58.9 b.p. The lower elasticities in deciles

106

8, 9 and 10 can be explained by the fact that detection of manipulation activities is too likely in the least liquid deciles, which drives manipulators away.

In decile 1, no significant correlation is found with alert incidence which is consistent with the expectation that manipulations are more costly and difficult to implement in highly liquid securities.

107

2.7 Summary and Conclusion We developed a testable hypothesis that market manipulation as proxied by the incidence of ramping alerts would raise transaction cost for completing larger trades. The alternative hypothesis was that ramping alerts represent information arrivals that are quickly reversed, unmasked as rumours, or false leads. If mean reversion of prices the next morning following a marking at the close represents manipulation, then we would expect market makers and limit order placers to reduce their order aggressiveness as protection against picking-off risk. Quoted spreads for thin volumes at the BBO might remain unchanged but limit orders for larger volume would be spread farther away from the BBO to avoid being triggered by a manipulator’s walking the order book. This rational response to possible manipulation presence would show up as higher effective spreads – i.e., higher volume-weighted (trade price – mid-point quote)/mid-point quote averaged over long periods. Information arrivals that are quickly reversed do not lead to wider spreads when averaged over long periods.

This second chapter has been a correlation analysis between two potentially endogenous variables, relative spreads and ramping alert incidence, not a factor analysis of the spread. The right-hand-side regressor is a proxy for the inherently unobservable presence of a manipulator. The event of information arrival is also inherently unobservable. Consequently, we have modelled the error components across exchanges and over time as observational errors, and we have tested whether Smarts’ measure of alert incidence is orthogonal to these observational errors on information arrival and manipulator presence.

When our alert incidence data is orthogonal to observational errors, we proceed to autoregressive modelling of the spreads-alert incidence relationship. When alert incidence is not orthogonal to the observational errors, we specify a fixed effects model and control for all the exchange-specific and time-specific idiosyncratic factors that influence both alert incidence and relative spreads.

Our findings are two-fold. First, variation in market integrity across securities exchanges and over time does affect the transaction costs of doing larger trades. We

108

find ramping alert incidence positively related to effective spreads in 8 of 10 deciles from most liquid to thinnest-trading securities. In contrast, quoted spreads are substantially affected by alert incidence in only the thinnest-trading tenth decile. Secondly, the magnitude of the increase in effective spreads when ramping manipulation incidence doubles is economically significant, 30 to 40 basis points in many moderate liquidity deciles. This compares with an average effective spread of 72 basis points for index-listed securities in the most efficient electronic markets worldwide.

Our findings have implications in terms of market design and market surveillance for many securities exchanges. Firstly, the exhibited relationship between alert incidence and relative spreads indicates trade-based securities market manipulations have significant impact on transaction costs. Therefore, the prevention of securities market manipulations not only serves the indirect purpose of improving the exchanges reputation for market integrity but also contributes directly to achieving more efficient marketplaces. Secondly, our results strongly suggest that market design should be continuously enhanced to help prevent securities market manipulations. For example, to prevent manipulators from marking the closing price, some exchanges could choose to adopt a closing auction or a random closing time, which would make manipulation more costly and unpredictable. Thirdly, notwithstanding the fact that market design enhancements can improve market integrity and eventually market efficiency, no securities exchange can be designed perfectly. Consequently, market surveillance is a necessary and important complement to a market design that enhances market integrity. In addition, brokers should be encouraged to monitor possible facilitation of market manipulation by conducting their own surveillance.

109

Chapter 3

An Empirical Model of the Incidence of Securities Market Manipulation

3.1 Introduction The concept of securities markets competing with each other has until recently been a rather foreign concept. Except for NASDAQ and NYSE which have been long-time competitors, national exchanges have had almost no competition other than that which is implied in cross-listings. As a consequence there has been limited interest in comparing and contrasting markets outside of NASDAQ and NYSE. 8 However, in the wake of the successful introduction of new markets like BATS and Direct Edge into the US and Chi-X, BATS and Turquoise into the UK and European marketplaces much more interest is now being generated in how to judge the quality of one market versus another, especially where they traverse national boundaries.

In our view, market quality is reflected by market efficiency and market integrity (often referred to as fairness) and the key task of all securities exchanges is to get the right combination of mechanism design factors: regulations, information, technology, market infrastructure (i.e., instruments) and participants in order to build a satisfactory level of market efficiency and integrity. Various researches have studied the impact of

8 One recent example of cross-national market comparisons is “Market Design and Execution Cost for Matched Securities Worldwide”, by Aitken, M., Cook R., Harris, F.H. and McInish, T.H, in Institutional Investor Inc., A Guide to Global Liquidity, Liquidity II, Bloomberg Tradebook, Winter 2009, 38-76. See also Harris, F.H., “Panel Discussion: Execution Costs and Market Design Worldwide,” Journal of Trading, 2007 (Winter), Vol 3, Issue 1. 110

those five factors on securities exchanges. For example, in terms of regulations, Bhattacharya & Daouk (2002) find that the cost of equity in a country, after controlling for a number of variables, does not change after the introduction of insider trading laws, but decreases significantly after the first prosecution. Under asymmetric information, Allen and Gale (1992) show that when all agents in a market have rational expectations and maximize expected utility, profitable securities market manipulations still exist.

However no prior research has directly studied the relationship between market integrity and those five factors, which could be the drivers of cross-sectional variation in market integrity. One of the fundamental reasons perhaps is the extreme difficulty of collecting data on market integrity (i.e., market manipulations). However, understanding the relationship between market integrity and those potential drivers can not only help investors compare various markets, but also assist securities exchanges and regulators themselves in enhancing market efficiency and integrity.

In Chapter 2, we test the relationship between securities exchange integrity (proxied by ramping alerts incidence) and efficiency (proxied by quoted spreads and effective spreads). Using observational error components to represent the presence of a manipulator and the arrival of information in a random effects model, we show that effective spreads are positively related to the incidence of ramping alerts across 8 of 10 liquidity deciles. The magnitude is economically significant. A doubling of this form of securities market manipulation raises the effective spread 31 to 59 basis points which is approximately 10% in the middle liquidity deciles across all the listed securities on the 34 exchanges we study.

Notwithstanding the fact that the findings from Chapter 2 are compelling, the cross- sectional variations in ramping alerts incidence across the 34 markets are left unexplained. In particular, we make no attempt in the research design of Chapter 2 to estimate the structural equations relationship between integrity and the potential drivers of market manipulations. As defined before, trade-based manipulations are implemented purely by buying and then selling securities, without taking any publicly observable actions to alter the value of the firm or releasing false information to

111

change the price. Therefore, the following categories of potential drivers will be the focus of this study: • Regulations • Technology • Security Market Infrastructure • Participants

The rest of this chapter is organized as follows. In section 3.2, the research methodology will be specified. Section 3.3 and 3.4 defines the data and measurement used by this research. Section 3.5 and 3.6 present analysis and discussion of the empirical results.

112

3.2 Prior Literature

3.2.1 Regulations Free market fundamentalism has dominated the development of global capital markets for more than three decades. But the discovery of more and more securities markets manipulations and financial markets crisis (e.g., Asia Financial Crisis, the burst of the high-tech bubble and the recent Sub-prime Meltdown, etc) have aroused the awareness of the great importance of external factors, such as regulations and technologies, to a healthy financial market.

Rajan and Zingales (1995) look at determinants of the size and extent of capital markets in most developed countries. Modigliani and Perotti (1996) and La Porta et al (1997) focus on the legal solutions to agency problems in financial markets. On the other hand, Levine (1997) and Rajan and Zingales (1996) examine the interactions between financial markets development and economic growth.

La Porta et al (1997) is the first study that empirically tests the correlation between legal environment and the financial markets. They argue that the root cause of the cross-sectional variation in the size and extent of different capital markets is legal environment because a well-established legal environment protects investors’ benefits, which raises their willingness and confidence to invest and hold securities.

Bhattacharya and Daouk (2002) study the relationship between cost of equity and the existence together with the enforcement of insider trading laws across 108 countries. They find that that the cost of equity in a country, after controlling for a number of variables, does not change after the introduction of insider trading laws, but decreases significantly after the first prosecution. Unlike Bhattacharya and Daouk (2002) that focus on insider trading law only, Cumming and Sofia (2008) manually collect data on trading regulations against trade-based manipulations and insider trading adopted across 25 securities exchanges in order to examine the hypothesis that securities exchanges having comprehensive rules generate higher turnover and larger market size, etc.

113

Our research examines a similar hypothesis to the one applied in La Porta et al (1997), Bhattacharya and Daouk (2002) and Cumming and Johan (2008) but rather than examining the consequences for financial market size and cost of equity, we focus on market integrity (proxied by trade-based market manipulations) which is more directly correlated with the legal environment.

3.2.2 Information Allen and Gale (1992) assert that when all agents in a market have rational expectations and maximize expected utility, profitable securities market manipulations still exist due to asymmetric information. The information asymmetry Allen and Gale refer to is that traders are uncertain whether a large trader who buys the stock does so because he knows it is undervalued based on his private information or because it is part of his manipulation strategy. It is this pooling equilibrium that allows profitable securities market manipulation to exist under general conditions.

Many research have provided evidence of the influence that information disclosure and transparency have on financial markets (Pagano and Roell 1996，Lang and Lundholm 1996 and Oved 2002 etc). Since the focus of this study is on trade-based manipulation, we will not elaborate the information factor further.

3.2.3 Technology As stated by the world largest trading engine company NASDAQ OMX on their website, technology has the power to drive capital formation, transform business and fuel economic growth around the world.

Garbade and Silber (1978) asserts that one of the most important causes of inter- market price differentials is the existence of time delays in communicating price information between market centres. From the domestic telegraph system on the New York Market and Philadelphia Stock Market during the 1840s to the consolidated stock market ticker tape on the various regional stock exchanges in 1975, there is a

114

close connection between advancement of technology and the prosperity of financial markets during the pre-computer and pre-Internet age.

The recent revolution of trading technology has been even faster-paced. In terms of trading engine (or trading platform) of a securities exchange, the aim has been providing investors with a real-time and centralized order book with a expedite channel for order submission. With the help of computer and network technology, centralized order books became standard in securities exchanges decades ago. Internet trading has unbound investors from their dependence on brokers and telephone lines. The latest development on some exchanges (e.g., the London Stock Exchange, Direct Edge etc) is the effort to launch Enhanced Liquidity Provider Program (ELP), which provides subscribed traders an integrated view of both displayed and dark pool order books. Historically, traders had to seek executions in either the displayed market or a single "dark pool". ELP offers a comprehensive solution for traders looking to aggregate liquidity of all types. According to Reuters (2008), Direct Edge's ELP Program broke the 100 million shares traded/per day mark on 15th July 2008 while the overall trading volume was 1.23 billion shares on the same day.

Technology has also transformed the institutional traders’ trading behaviour. Program trading perhaps is the most convincing example. The most common definition of program trading is the use of computers in stock markets to engage in arbitrage and portfolio insurance strategies. Stoll and Whaley (1987) states that although program trading is used for a number of reasons unrelated to index futures (by index funds, for example), much of it is associated with the index futures arbitrage process. Hence, program trading is one of the motivations behind the expiration day effect on index futures and options. Harris, Sofianos and Shapiro (1994) also conclude that program trading and intraday volatilities in the SP 500 Index are correlated.

On the other hand, trading surveillance technology has also been gradually catching up. Real-time market surveillance technology has been replacing traditional T+N market surveillance or transaction log books on paper. For example, the SMARTS Real-time Securities Market Surveillance Platform has been deployed in more than 50 national exchanges and regulators. But there has been no prior research studying the

115

relationship between surveillance technology and the quality of securities market. This study will try to fill this gap.

3.2.4 Market Infrastructure Hart and Kreps (1986) laid the foundation for the connection between securities market manipulations and volatility by discovering that even though non-speculators and speculators alike behave rationally and speculators are competitive, speculation can destabilize prices.

Foucault (1999) develops a Theory of Order Placement which connects volatility, order placement strategies and spread. When the non-execution risk is high, traders will use market orders to gain immediate execution; when the pick-off risk is high, limit orders turn out to be a better choice. Foucault’s theory predicts that when the volatility of security price increases, traders will tend to hold limit orders rather than market orders to reduce the pick-off risk at the cost of non-execution risk being increased, which are eventually reflected in wider spreads.

Empirically, Stoll (1987, 1991), Chamberlain, Chueng and Kuan (1989), Chiou, et al (2007) all find evidence that price volatility is higher during the period of manipulation. Consistent with this result, in Chapter 2, we also discovered that daily effective spreads are positively related to the daily incidence of ramping alerts across 34 securities exchanges for the period 2000-2005.

3.2.5 Participants Flexison and Pelli (1998) provides empirical evidence found in the Finnish securities market supporting the theory that traders have to manipulate securities’ closing price since research has shown that closing prices are one of the most used benchmarks against which trader performance is evaluated. Drudi and Massa (2005), Kumar and Seppi (1992) and Merrick, Naik and Yadav (2005) etc study manipulations performed on specific trading instruments. For example, in the case of squeezing the long-Term U.K. Government Bond Futures contract in 1998, uninformed investors can make

116

profits by taking positions in futures contracts with cash settlement followed by manipulations of prices of the underlying securities.

A lot of emerging markets have opened up to foreign investors with the rationale that informed foreign investors will boost liquidity and tend to stabilize the financial market. But financial crisis in emerging markets have also brought up concerns regarding the excessive volatilities or bubbles introduced by those foreign investors. Chiou, et al (2007) study the behaviour of the Taiwan Stock Index from 1996 to 2005 and conclude that informed foreign investors have motivations to manipulate the stock market when (1) enough time is available for them to push up the price and get out completely before the private information is fully disclosed to the public, and (2) the profit from manipulating the market prices will be higher than the cost of doing so.

117

3.3 Research Methodology

3.3.1 Proxy for Market Integrity As stated in Chapter 1, one crucial measure to evaluate the integrity of a securities market is the level of market misconduct. In Chapter 2, the Ramping Market Manipulation (which consists of Marking the Close Alert and Reversal the Next Morning Alert) is applied as a proxy for securities market manipulation for the cross- markets analysis of the relationship between spreads and alerts. From Table 1, it can be seen that Marking the Close Alert and Reversal the Next Trading Day Alert are the only two alerts that can be studied using the Trade Level Data only (Trade level data includes security, trade time, trade price and volume). All the other alerts require Order Level Data, Participant Data, Participant Holdings Data or Deal Capacity Data.

Hart and Kreps (1986) point out that data collection for market manipulations is extremely hard. The only order book data database that can be publicly accessed is NYSE trades, orders, reports, and quotes (TORQ) Database which contains time- ordered transactions for 144 selected stocks for the short three month period from November 1990 to January 1991 for the purpose of an order book audit. Even more so, the participant data required for the surveillance and enforcement of rules violated by other manipulations (and their associated alerts) is strictly confidential to stock exchanges; no publicly accessible database can provide the participant data.

3.3.2 Proxies for Securities Market Regulations Cumming and Sofia (2008) manually collect data on the trading regulations against trade-based manipulations and insider trading adopted across 25 securities exchanges in order to examine the hypothesis that securities exchanges having comprehensive rules are better in terms of turnover, market size, etc. For the purpose of this research, we will utilize the following data items provided by Cumming and Sofia as proxies for Securities Market Regulations: • The number of trading regulations that are monitored (i.e., having corresponding surveillance alerts designed by SMARTS Group International) • The existence of a trading regulation for ramping

118

Though Bhattacharya and Daouk (2002) state that the enforcement of insider trading laws has far more significant impact than the introduction of the insider trading laws, we will be focusing on the existence of trading regulation against market manipulations due to the extreme difficulty of obtaining enforcement data for manipulation cases from various securities exchanges. We will use the Standard Surveillance Alerts Portfolio provided by SMARTS to its clients around the world as a benchmark to measure the number of trading regulations that are monitored in daily market supervision across those securities exchanges in the sample. Securities exchanges with higher number of trading regulations that have corresponding surveillance alerts are expected to have fewer ramping alerts.

3.3.3 Proxy for Technology The technology we focus on here is the Market Surveillance Technology. As the leading real-time market surveillance platform, SMARTS has been deployed by more than 50 national securities exchanges and regulators around the world. We will use the deployment of SMARTS as a proxy for the Experience with Real Time Market Surveillance Technology of a securities exchange.

3.3.4 Proxies for Securities Market Infrastructure The following will be used as proxies for the impact of Financial Market Infrastructure on market integrity: (1) Volatility (defined as standard deviations of daily returns) (2) Market Liquidity (defined as market turnover) (3) Quoted Spread (4) The existence of a closing call auction

According to the Theory of Order Placement developed by Foucault (1999), higher volatility leads to wider spreads, which would increase the cost of manipulation. In other words, volatility should be negatively correlated with the number of alerts. But on the other hand, manipulations carried out on securities with higher volatilities should have lower probability of detection since price dynamics would less often trigger surveillance.

119

Quoted spreads, as a measure for the absolute cost of a round trip transaction, is a good proxy for the potential transactional costs involved in manipulations. It is expected that higher quoted spreads will cause lower alerts incidence.

Securities that are highly liquid normally have big market capital and are difficult to manipulate due to the higher costs involved. Hence, it could be expected that higher market liquidity is associated with lower number of alerts. But as explained in Chapter 2, ramping alerts are not equivalent to genuine ramping manipulation cases. As a matter of fact, much surveillance workflow applied by those securities exchanges the author has designed alerts for is designed to validate alerts triggered based on a variety of other information. It is quite common that a big proportion of alerts triggered each day are false positives that can be explained by a legitimate reason. So the problem of false positives of ramping alerts is well known in surveillance research and likely to be positively correlated with market volatility.

Many securities exchanges have introduced closing call auctions with the purpose of improving market quality (Pagano and Schwartz 2003, Comerton-Forde and Rydges 2006, etc). The findings from previous studies provide mixed evidence on the effectiveness of the closing call auction.

3.3.5 Proxies for Securities Market Participants Chiou, et al (2007) has two major findings. The first is that informed foreign portfolio investments have motivations to manipulate the market. The second is that in financial markets, extremely low transaction costs and high speed of adjustment of trading creates greater chances for the informed to manipulate the market and take advantage of the uninformed. Based on those findings, we will use the following as proxy for the impact of financial market participants on market integrity: • The existence of direct market access which is expected to be positively correlated with the ramping alerts incidence.

120

Direct Market Access (DMA) is generally defined as electronic facilities which allow brokers to offer clients direct access to the exchange trading system through the broker’s infrastructure without manual intervention by the broker. Some of the well- known advantages of DMA are listed below: (1) direct control of clients over orders; (2) faster execution of client orders; (3) lower impact costs for large orders; (4) better use of hedging and arbitrage opportunities through the use of algorithm trading

3.3.6 Model Specification Assume the number of ramping alerts triggered in a securities exchange proxies the overall level of market manipulation in that exchange. The following regression relation is designed to test the relationship between the level of market manipulation or conversely the level of integrity of an exchange and a number of proxies for Securities Market Regulations, Surveillance Technology, Securities Market Infrastructure and Market Participants.

̃ where

= the average numbers of Ramping Alerts of market i

= the constant that would be estimated

= the correlation coefficient between Alerts Incidence and the corresponding proxy

= The dummy variable for the deployment of Real Time Surveillance System (i.e., SMARTS) in market i

= The average standard deviation of daily returns of market i at (endogenous variable)

= The average market turnover of market i

121

= The relative spread measure of market i

= The dummy variable for the existence of Closing Call Auction in market i

= The dummy variable for the existence of direct market access of market i

= The dummy variable for the existence of ramping regulation of market i

̃ = The residual error term

The model will be estimated with Nonlinear Ordinary Least Square (OLS) and Nonlinear Two-stage Least Square (2SLS) error structures assumed, respectively to learn the most about the possible endogeneity of spreads and volatility. The Hausman- Wu test will be conducted to test whether the endogeneity of effective spreads and volatility are significantly biased the Nonlinear OLS model. The rest of the independent variables are specified as instrument variables. We predict Volatility by using the following regression relation:

̃ Where = The number of trading regulations that have corresponding surveillance alerts in market i

We predict Spreads by using the following regression relation:

122

3.4 Data and Measurement

3.4.1 Data We utilize the Trading Regulations Database from Cumming and Sofia (2008) to create the proxy for Regulations for our study. Cumming and Sofia (2008) study the correlation between trading regulations against market manipulations and the market turnover across 25 national securities exchanges for the period 2005-2008.

We obtain the primary database for our study from the Reuters database maintained by the Securities Industry Research Centre of Asia-Pacific (SIRCA). This database contains intra-day trade and quote data for seven years for more than 200 world markets including most of the equity markets. Initially, the period of analysis for our study extends from January to December 2005. But due to the fact that the Trading Regulations Database from Cumming and Sofia (2008) only covers years between 2005 and 2008, the regression analysis in this Chapter will be conducted for year 2005.

We checked the official website of each national securities exchange from our sample and the Reuters Financial New Database to confirm • Whether the exchange had implemented Close Call Auction in 2005; and • Whether the exchange had allowed Direct Market Access in 2005.

Direct Market Access facilitates algorithmic trading which makes market manipulation more difficult. The reason is that market manipulators must exit their positions faster than the execution of hedge funds and proprietary trading desks who often adopt algorithmic trading with computer “bots”.

The analysis will be conducted on the entire sample of listed securities grouped into ten liquidity deciles of aggregate data from each of the 24 securities exchanges. Liquidity deciles are determined by dividing the total number of securities in each market into 10 groups, based on their monthly trading turnover. Table 29 below lists the 24 securities exchanges studied by this Chapter.

123

Table 29 List of Securities Exchanges Covered by this Research

Index Securities Exchange Home Country 1 Australian Securities Exchange Australia 2 Bombay Stock Exchange India 3 Cairo Stock Exchange Egypt 4 Deutsche Boerse-Xetra Germany 5 Hong Kong Stock Exchange China 6 Istanbul Stock Exchange Turkey 7 Jakarta Stock Exchange Indonesia 8 Johannesburg Stock Exchange South Africa 9 Korea Stock Exchange Korea 10 Kuala Lumpur Stock Exchange Malaysia 11 London Stock Exchange U.K. 12 Madrid Stock Exchange Spain 13 NASDAQ U.S. 14 National Stock Exchange, India India 15 New York Stock Exchange U.S. 16 New Zealand Stock Exchange New Zealand 17 Shanghai Stock Exchange China 18 Shenzhen Stock Exchange China 19 Singapore Stock Exchange Singapore 20 Stock Exchange of Thailand Thailand 21 Stockholm Borsen Sweden 22 Swiss Stock Exchange Switzerland 23 Taiwan Stock Exchange` China 24 Tokyo Stock Exchange Japan

124

3.4.2 Ramping Alert Incidence The Ramping Alert applied in this study is programmed in the ALICE Language, which is the proprietary language from the SMARTS Real-time Securities Market Surveillance Platform. The algorithm of the Ramping Alert is described as below.

Benchmark Period and Threshold For date T, a historical price change distribution for the past month (the benchmarking period) is created for each security. The observations in this distribution are sampled wherever on market trades occur throughout the benchmarking period. Fifteen minutes after the market opens, we calculate the percentage change between the trade price and the true price 15 minutes earlier. True price is defined as (3) the previous trade price; or (4) the best bid (offer) price at time t-15 minutes if the previous trade price is below (above) the best (offer) price at t-15 minutes.

Finally, we take the absolute value of the calculated percentage price change and add it to the historical distribution.

125

Last Trade Price of the Security Percentage Threshold Price From ($) Price To ($) 0.00 0.10 20% 0.11 0.25 15% 0.26 0.50 12% 0.51 1.00 10% 1.01 5.00 8% 5.01 10.00 5% 10.01 10000000.00 3%

The purpose of the benchmark process is to identify the top 1 percent of least frequent or unrepresentative price changes for a security during the benchmark period. Assuming that there are approximately 20 trading days in a month and 100 trades in each trading day (assuming 6 trading hours per day), there are approximately 2000 price change observations each month. If these observations are sorted, the largest 20 price change (or 1 percent of the distribution) can be identified. The value of the 20th price change is where the threshold for ramping for that security is set. For example, if the 20th highest price change for BHP Billiton is 0.5% during the September, then the security is deemed to have been subject to ramping if the return in the last 15 minutes of 31 October was greater than 0.5%.

Conditions for Reversal the Next Morning Alert On date T+1, for each security, we trigger a ramping alert if the following conditions are satisfied: (3) if there was a Marking the Close Alert triggered for that security on date T; and

9 It is also true that for some illiquid securities, the time when the closing price is determined could be mid day or even earlier 126

(4) During the first 15 minutes of trading on date T+1 or among the first ten trades on date T+1 (i.e., considering illiquid securities may not be traded during the first 15 minutes), if the Marking the close Alert triggered on date T is for driving up (down) closing price by X%, at least one trade occurs at price P that is below (above) the closing price on date T by more X% or more (a.k.a., Reversal the Next Morning).

The above algorithms are run daily across the 24 securities exchanges for 2005 to derive the annual alert incidence of daily ramping manipulation. The annual alert incidence of daily ramping manipulation per security per decile for market i in year 2005 is calculated as the number of ramping alerts triggered for all securities from market i in year 2005 normalized by the average number of listed securities per month for the same market over the same period.

3.4.3 Time-Weighted Quoted Spread To calculate the Quoted Spread for each security, the following formula is used:

0.5

where

127

To obtain the final estimate of Annual Average Daily Time-Weighted Quoted Spread per decile for market i in year 2005, the time weighted quoted spreads are averaged across all securities per decile over all trading days for market i in year 2005. Negative spreads and instances where one side of the spread was absent were removed from the sample.

3.4.4 Average Volatility We measure of volatility by using the standard deviation of logarithmic daily return. To calculate the standard deviation of logarithmic daily return for each security, the following formula is used:

∑ . . 1

where

n = Number of trading days in year 2005

To obtain the final estimate of Annual Average Standard Deviation of Logarithmic Daily Return per security per decile for market i in year 2005, the standard deviation of logarithmic daily returns are averaged across all securities per decile over all trading days for market i in year 2005.

3.4.5 Average Liquidity We use trading turnover to proxy liquidity. To calculate the Annual Average Turnover per decile for market i in year 2005, the trading volume are aggregated across all securities per decile overall trading day for market i in year 2005.

128

3.4.6 Dummy Variables We create the following three dummy variables. (1) Closing Auction Dummy Variable which equals to 1 if a securities exchange has implemented Closing Auction before 2005 and equals to 0 otherwise; (2) Direct Market Access Dummy Variable which equals to 1 if a security exchange has allowed Direct Market Access before 2005 and equals to 0 otherwise; (3) SMARTS Deployment Dummy Variable which equals to 1 if a security exchange has deployed the SMARTS Real-time Surveillance Platform before 2005 and equal to zero otherwise.

3.4.7 Descriptive Statistics The descriptive statistics for the following variables are presented in Table 30 below. • Annual Alert incidence of daily ramping manipulation per security per decile across the 24 securities exchanges in year 2005 (AI); • Annual Average Quoted Spread per security per decile across the 24 securities exchanges in year 2005 (SPR); • Annual Average Standard Deviation of Logarithmic Daily Return per security per decile across the 24 securities exchanges in year 2005 (Vol); • Annual Average Turnover per decile across the 24 securities exchanges in year 2005 (Turnover); • Number of Trading Regulations that are monitored across the 24 securities exchanges in year 2005 (Regs)

Table 30 Descriptive Statistics for 24 Markets in year 2005

Panel A: Moments for Raw Observations AI SPR Vol Turnover Regs Mean 0.1775 5.36% 0.0409 97120486 6.7500 Std. Dev 0.1876 0.0956 0.0446 365031297 3.9053 Skewness 1.7931 3.4166 3.9966 7.3444 0.6062 Kurtosis 3.2889 13.1504 17.1601 63.4832 -0.2879

129

Panel A: Moments for Raw Observations AI SPR Vol Turnover Regs Mean 0.1775 5.36% 0.0409 97120486 6.7500 Std. Dev 0.1876 0.0956 0.0446 365031297 3.9053 No. of Observations 240 240 240 240 240

Panel B: Moments for Natural Log of Observations AI SPR Vol Turnover Regs Mean -2.5002 -3.9911 -3.4882 15.1705 1.3483 Std. Dev 0.8547 1.4776 0.7877 3.3841 2.2723 Skewness 0.1977 0.2425 -1.5412 -0.7449 -4.1936 Kurtosis 3.4063 -0.6729 3.3485 0.2987 16.9851 No. of Observations 240 240 240 240 240

It can be seen that except for the Number of Trading Regulations that are monitored (“Regs”) across the 24 securities exchanges in year 2005, all the other 4 variables are

demonstrably non-normal. For example, the quoted spread (“SPR”) has mean ("")

of 5.36% and standard deviation ("") of 0.0956 with skewness of 3.4166 and kurtosis of 13.1504. After a natural log transform, we observe the distribution of

ln to be approximately normal ( = -3.9911 and = 1.4776) with skewness of 0.2425 and kurtosis of -0.6729. The same is also observed for AI, Vol and Turnover. Using the properties of the lognormal distribution and assuming

. exact log-normality for our observations, an estimator of would be .

Our data yields such an estimate of ... = 5.51%. This figure differs

from of 5.36% because our sample differs slightly from a pure lognormal distribution.

130

3.4.8 Limitations of Research Design Table 31 below display the correlation matrix for the three dummy variables specified in Section 3.4.6.

Table 31 Correlation between Dummy Variables

Closing Auction Direct Market Access SMARTS Dummy Dummy Dummy

Closing Auction 1 0.4584 0.0084 Dummy (p=0.0243) (p=0.9689)

Direct Market 0.4584 1 0.4584 Access Dummy (p=0.0243) (p=0.0243)

SMARTS Dummy 0.0084 0.4584 1 (p=0.9689) (p=0.0243)

It can be seen that the correlation between the Closing Auction Dummy variable and the Direct Market Access Dummy variable is as same as the one between the Closing Auction Dummy variable and the SMARTS Dummy variable. We count the full-

correlation value pairs (e.g., Closing Auction Dummyi = 1 and Direct Market Access

Dummyi = 1) among those three dummy variables and discover that the number of full-correlation value pairs between the Closing Auction Dummy variable and the SMARTS Dummy variable is 17. This coincidence results in high collinearity which may impair the explanatory power of those dummy variables although the joint F-test rejects the null hypothesis that the closing auction dummy variable and the SMARTS dummy variable are jointly zero (F= 3.57, 22 d.f.).

131

3.5 Empirical Specification Based on the fact that the following four variables are near log-normally distributed, • Annual Alert incidence of daily ramping manipulation per security per decile across the 24 securities exchanges in year 2005 (AI), • Annual Average Quoted Spread per security per decile across the 24 securities exchanges in year 2005 (SPR), • Annual Average Standard Deviation of Logarithmic Daily Return per security per decile across the 24 securities exchanges in year 2005 (Vol), • Annual Average Turnover per decile across the 24 securities exchanges in year 2005 (Turnover), we transform for estimation the regression relation as

, , ,

, ,+,

, ,̃ where

, = the average numbers of Ramping Alerts of market i at time t

= the constant that would be estimated

= the correlation coefficient between Alerts Incidence and the corresponding proxy

, = The dummy variable for the deployment of Real Time Surveillance System (i.e., SMARTS) in market i at time t

, = The average standard deviation of daily returns of market i at time t (endogenous variable)

, = The average market turnover of market i at time t

, = The relative spread measure of market i at time t

, = The dummy variable for the existence of Closing Call Auction in market i at time t

, = The dummy variable for the existence of direct market access of market i at time t 132

, = The dummy variable for the existence of ramping regulation of market i at time t

̃, = The residual error term

We predict endogenous variable, ,, by using the following regression relation:

, , ,

, ,̃ Where

, = The number of trading regulations that have corresponding surveillance alerts in market i at time t

Similarly, we predict Spreads by using the following regression relation:

, , ,

, ,̃

Since the data we analyse aggregates all listed securities at the national exchange level to the liquidity decile at the national market level, we introduced fixed effects for the liquidity decile, omitting decile seven. The results were qualitatively identical because a continuous measure of liquidity itself is a right-hand-side variable in all of our models.

In Chapter 2, we discovered that securities exchanges’ fixed effects have significant explanatory power on the relationship between relative spreads and ramping alerts incidence in many deciles. Therefore, we will also add exchange dummy variables into the regression analysis as:

133

, , ,

, ,+ ,

…

, ,̃ Where

= The dummy variable for securities exchange. The number of exchange dummy variables is 23.

Similarly, we predict endogenous variable, , , , by using the following regression relation:

, , ,

, ,

… ,

, ,̃

, , , ,

, ,

… +,

, ,

134

3.6 Possible Endogeneity of Alert Incidence, Volatility, and Spreads

In this section, we investigate four econometric issues that arise in estimating the drivers of market integrity measured in this thesis as Ramping Alert Incidence (AI). First, we explore the possible endogeneity of effective spreads and volatility, two theoretically indicated determinants of AI. By performing Hausman-Wu specification tests, we will assess the simultaneity bias, if any, on ordinary least squares parameters for these variables and the need for instrumental variables with two-stage least squares (TSLS) estimation. Depending on the result of these endogeneity tests, in a subsequent section estimating the structural equations we will also investigate possible cross-equation correlation of the disturbances with either seemingly unrelated regression (SUR) or 3SLS estimation. Second, we will estimate the possible idiosyncratic effect on the parametric estimates of fixed effects pertaining to the 24 security exchanges. Third, we will test using Chow and Wald tests the validity of pooling the ten liquidity deciles rather than estimating separate subgroups of thickly-traded, thinly-traded, and those in-between that we refer to as moderately liquid securities. Finally, we will relax our log linear functional form specification of the AI and Spreads models by employing maximum likelihood estimation.

3.6.1 OLS, 2SLS and FIML results for All Deciles The ordinary least square regression (OLS), two-stage least square regression (2SLS) and Full Information Maximum Likelihood (FIML) results for all deciles are presented in Table 32. Here we focus on the overall model fit and the issues of 1) Pooling versus grouped (in liquidity deciles) regressions; and 2) The inclusion of fixed effects at the national securities exchange level of observation. In subsequent Tables we focus on parameter estimates and interpretation.

135

Table 32 Regression Results for Potential Endogenous Equations for all deciles across 24 Securities Exchanges in 2005

Panel A: OLS

Without Fixed Effects AI Model Volatility Model Spread Model 10 F-test 2.30** 27.67*** 57.47*** R-square 0.0623 0.4161 0.5968 0.0159 (t= 0.13) 0.2545***(t= 6.92) N/A βspr -0.6576** (t= -1.99) 0.4639***(t= 4.39) -0.8848***(t=-5.70) βcall -0.8331** (t= -2.53) N/A N/A βDMA 0.9937*** (t= 2.88) -0.3951***(t= -3.55) 1.5994***(t=10.92) βRTS -0.2107( t=-0.74) -2.7026***(t= -10.04) 1.0050***( t=4.77) βRampReg 0.1891 (t= 0.96) N/A 0.6478***(t=6.97) βvol N/A -0.0513***( t= -4.42) -0.0285 (t=-1.20) βregs N/A 0.2026***(t= 12.56) -0.3704***(t=-16.21) βliq α -2.5283** ( t= -2.07) -5.4091***(t=-25.98) 3.9478***(t=6.73) With Fixed Effects AI Model Volatility Model Spread Model F-test 11.87*** 20.63*** 82.87*** R-square 0.4428 0.4988 0.9064 -0.2016*** (t= -2.69) 0.3019*** (t=7.29) N/A βspr -1.4939***(t= -3.76) 0.7911***( t=6.88) -3.1688***(t=-10.22) βcall 0.3964 (t= 0.1892) N/A N/A βDMA -0.3710 (t=-1.24) -0.7065*** (t=-5.97) 9.0239***(t=11.35) βRTS 0.2286(t= 0.0.84) -0.4595*** (t=-3.22) 19.2534***(t=12.12) βRampReg 0.0872 (t= 0.69) N/A 0.3941***(t=5.78) βvol N/A -0.0552***(t=-3.63) -1.3167***(t=-11.26) βregs N/A 0.2189*** (t= 12.78) -0.3943***(t=-26.36) βliq 8 out of 23 5 out of 23 19 out of 23 βexch α -1.7210*** ( t= -2.89) -5.3339***(t =-24.38) 5.6862***(t=14.47)

10 *** stands for 99% confidence level; ** stands for 95% confidence level; * stands for 90% confidence level

136

Panel B: 2SLS

Without Instrument: Volatility Instrument: Spread Fixed Effects

Hausman- 3.88 18.47 Wu Test AI Model Volatility Model AI Model Spread Model F-test 2.94*** 0.25 1.01 23.03*** R-square 0.1137 0.0211 0.0253 0.3723 -0.3875 (t=-1.47) 2.9281(t= 0.80) -0.1419( t=-1.46) N/A βspr 0.1253 (t=0.16) 3.3214(t= 0.80) 1.1812*(t=1.83) -1.4392***(t=-5.04) βcall -3.0131(t= -1.62) N/A -2.3260( t=-1.52) N/A βDMA 5.9743(t= 1.63) -9.9692(t= -0.74) 4.1051(t=1.41) 2.9115***(t=6.28) βRTS 0.9070(t= 1.10) -7.2539(t= -0.75) 0.5470(t=0.90) 1.4528***(t=3.64) βRampReg -0.7629 (t= -0.79) N/A 0.4906(t=-0.60) 2.1063***(t=3.24) βvol N/A 0.4825( t= 0.64) N/A 0.0155 (t=0.27) βregs N/A 1.1169(t= 0.64) N/A -0.5870***(t=-6.69) βliq α -6.9418 ( t=-1.45) -9.1317(t=-1.64) -5.8087***(t=-1.46) 11.9030***(t=3.48)

With Fixed Instrument: Volatility Instrument: Spread Effects

Hausman- 2.09 1.15 Wu Stats AI Model Volatility Model AI Model Spread Model F-test 11.61*** 16.10*** 11.16*** 2.19*** R-square 0.4277 0.4371 0.4278 0.1739 -0.3436** (t= -2.10) 0.2355*** (t=3.78) -0.3436**( t=-2.10) N/A βspr -1.4771***(t= -3.59) 0.8026*** (t=6.36) -1.4771*** (t=-3.59) -2.0140(t=-1.52) βcall 0.4609 (t= 1.44) N/A 0.4609(t=1.44) N/A βDMA -0.5700* (t=-1.83) -0.4284*** ( t=-3.73) -0.5700*(t=-1.83) 1.2078(t=1.20) βRTS 0.2074(t=0.74) -0.3191** ( t=-2.14) 0.2074(t=0.74) 1.9571( t=1.01) βRampReg -0.1911 (t= 0.49) N/A -0.1911(t=-0.49) 4.7182(t=1.24) βvol N/A -0.0767***(t=-5.01) N/A 0.1460(t=0.43) βregs N/A 0.1961*** (t=9.50) N/A -0.9172*(t=-1.91) βliq 7 out of 23 5 out of 23 7 out of 23 3 out of 23 βexch α -3.2706* ( t= -1.75) -5.2026***( t =-23.10) -3.2706*(t=1.75) 25.7755(t=1.32)

137

Panel C: FIML

Without Instrument: Volatility Instrument: Spread Fixed Effects AI Model Volatility Model AI Model Spread Model R-square 0.0604 0.3986 0.06113 0.5943 -0.0146 (t=-0.15) 0.2837***(t= 5.20) -0.0590( t=-0.56) N/A βspr 0.6610 (t=1.27) 0.4590***(t= 3.88) 0.6554(t=1.27) -0.9000***(t=-3.83) βcall -0.7779 (t= -1.57) N/A -0.7781( t=-1.48) N/A βDMA 1.0267***(t= 2.98) -0.6095***( t= -4.18) 1.0683***(t=2.70) 1.5631***(t=8.79) βRTS 0.0222( t=0.05) -0.4608***( t= -3.75) 0.0219(t=0.05) 0.8379***(t=5.11) βRampReg 0.3122(t= 0.79) N/A 0.2061(t=0.77) 0.6733***(t=8.54) βvol N/A 0.2068***(t= 13.15) N/A -0.3776***(t=-15.72) βliq α -1.8551( t=-1.15) -5.5063(t=-1.15) -2.4101**(t=-2.07) 4.0229***(t=7.01)

With Fixed Instrument: Volatility Instrument: Spread Effects AI Model Volatility Model AI Model Spread Model R-square 0.2662 0.4609 0.3167 0.8493 -0.2629** (t= -1.96) 0.3282*** (t=5.21) -0.3615***( t=-2.78) N/A βspr -1.3177*(t=-1.87) 0.7345*** (t=6.36) -1.4254*(t=-1.92) -1.0803***(t=-2.96) βcall 0.1938 (t=-0.41) N/A -0.3363(t=0.72) N/A βDMA 0.2604 (t=0.55) -0.7899*** (t=-4.92) 0.1479(t=0.33) 0.4574(t=0.80) βRTS 0.3725(t=0.87) -0.7523***( t=-5.84) 0.4641(t=1.10) 1.3537***(t=4.89) βRampReg 0.6261* (t=1.89) N/A 0.1310(t=-0.49) 0.6909***(t=14.30) βvol N/A 0.2132*** (t=12.89) N/A -0.4151***(t=-23.88) βliq 3 out of 23 3 out of 23 6 out of 23 7 out of 23 βexch α -0.2059 (t=0.14) -5.3834***( t =-26.34) -2.2191**(t=-2.39) 4.3960***(t=11.84)

It can be seen that from Table 32 Panel A, B and C that fixed effects alter the sign and significance of most regressor in the AI Model whilst 1 of 18 in the Volatility and Spread Model. The R-square and the F-statistics for the various AI and Spread Models with fixed effects are much higher than those without fixed effects. As a result, in subsequent tables, we will focus on the regression results from models with fixed effects only.

Chow testing of pooled samples in Chapter 2, Chow-test proves that the pooling of the liquidity ten deciles combine very distinctly different thickly versus thinly-traded 138

securities. The same phenomenon holds for the current analysis as well. Below we perform a Chow test to confirm that the liquidity deciles subgroups analysis is appropriate.

The pooling of thickly-traded deciles in liquidity deciles 1-3, the moderately-traded stocks in deciles 4-7 and the thinly-traded stocks in deciles 8-10 are rejected by a Chow test (F 4.04 with p-value less than 0.01)11. Consequently, we report and discuss the regressions for 3 subsets of the ten liquidity deciles in our subsequent analysis.

11 The Chow (1960) test is used to ascertain whether a number of sets of observations can be pooled into one larger set for regression purposes. In our case we have three sets: thickly traded securities from liquidity deciles 1 through 3, moderately-traded securities from liquidity decile 4 through 7 and thinly- traded securities from liquidity deciles 8 through 10. To compute the test statistic (“C”) we need the sum of squared residuals from running a regression using k terms on the pooled set () of n, together with the sum of squared residuals from running a regression using the same model on the subsets (, ).

/ = 7.45/1.84 = 4.04 /

The Chow test statistic C has an Fk, n1+n2-2k distribution. For further discussion of the Chow test see Gujarati (2003), pp.275-279. 139

3.6.2 OLS, 2SLS and FIML results for Thickly-traded Deciles The ordinary least square regression (OLS), two-stage least square regression (2SLS) and Full Information Maximum Likelihood (FIML) results for thickly-traded deciles (decile 1, 2 and 3) are presented in Table 33.

Table 33 Regression Results for Potential Endogenous Equations for thickly-traded deciles across 24 Securities Exchanges in 2005

Panel A: OLS

With Fixed Effects AI Model Volatility Model Spread Model F-test 16.69*** 26.39*** 55.97*** R-square 0.7079 0.9309 0.9682 0.2550*(t= 1.99) -0.1596*** (t=-2.97) N/A βspr 0.3995(t=1.06) 1.7444***(t=9.76) -2.3986***(t=-7.87) βcall -0.7370** (t=-2.00) N/A N/A βDMA -0.1391(t=-0.38) -2.0008*** (t=-7.97) 7.8736***(t=9.10) βRTS 0.0043(t= 0.01) -1.2003*** (t=-3.50) 18.5677***(t=8.92) βRampReg -0.3466(t= -1.39) N/A -0.1900(t=-1.01) βvol N/A -0.1138***(t=-3.79) -1.4223***(t=-8.51) βregs N/A 0.0194(t=0.46) -0.3184***(t=-11.25) βliq 3 out of 23 12 out of 23 18 out of 23 βexch α -1.7210*** ( t= -2.89) -3.8249***(t =-17.38) 3.2526***(t=4.36)

140

Panel B: 2SLS

With Fixed Instrument: Volatility Instrument: Spread Effects

Hausman- 11.53 0.40 Wu Stats AI Model Volatility Model AI Model Spread Model F-test 8.53*** 0.61 15.58*** 29.30*** R-square 0.5531 0.2370 0.6934 0.9374 0.2940(t=1.61) -0.0752(t=-0.11) 0.1488( t=0.68) N/A βspr 0.9707(t=1.12) 0.7978(t=0.91) 0.4894(t=1.20) -2.4694***(t=-4.80) βcall -2.2920(t=-1.15) N/A -0.9907**(t=-2.08) N/A βDMA 1.9729(t=0.72) 2.5008 (t=0.46) 0.2303(t=0.60) 7.6212***(t=10.50) βRTS 0.6088(t=0.68) 2.6388(t=0.78) 0.0541(t=0.14) 17.9909***( t=5.27) βRampReg -1.3295(t=-1.10) N/A -0.5944(t=-1.27) -0.4863(t=-0.22) βvol N/A -0.3716**(t=-1.96) N/A -1.3883***(t=-3.58) βregs N/A 0.0934*** (t=2.64) N/A -0.2845(t=-1.49) βliq 3 out of 23 2 out of 23 3 out of 23 17 out of 23 βexch α -5.3393( t= -1.24) -3.3627(t =-1.34) -3.2541**(t=-1.97) 1.6922(t=0.18)

Panel C: FIML

With Fixed Instrument: Volatility Instrument: Spread Effects AI Model Volatility Model AI Model Spread Model R-square 0.6792 0.5565 0.7076 0.9177 0.2260(t= 0.86) 0.2289*(t=1.85) 0.2861(t=1.09) N/A βspr 0.5078(t=0.90) 0.5998(t=0.94) 0.4041(t=0.78) -2.0735***(t=-3.41) βcall -0.8471(t=-1.35) N/A -0.7395(t=-1.33) N/A βDMA 0.1387(t=0.23) -0.9939*** (t=-4.22) -0.1624(t=-0.31) 1.1614***(t=2.55) βRTS -0.0882(t=-0.18) -0.4786*( t=-1.93) 0.0168(t=0.04) 1.0172***(t=2.52) βRampReg 0.1954(t=0.29) N/A -0.3522(t=-0.74) 1.0886***(t=9.15) βvol N/A 0.2452*** (t=4.86) N/A -0.3822***(t=-8.12) βliq 4 out of 23 2 out of 23 7 out of 23 8 out of 23 βexch α -0.2363 (t=-0.09) -6.4161***( t =-7.67) -1.6954(t=-0.89) 5.7912***(t=5.19)

141

In the OLS estimation for thickly-traded securities (Table 33 Panel A), fast electronic trading access (DMA) reduce alert incidence. This result suggests that counterparties can unwind their intra-day exposures before potential manipulators can execute ramping manipulation strategy. Lower standard errors may result from subsequent instrumental variable and systems estimation. Closing Call Auctions lower spreads while increasing volatility. Real time surveillance in contrast is associated with more asymmetrical information environments (in the cross-sectional) and therefore higher spreads. Importantly, RTS does lower volatility. Similarly, the presence of regulations specifically prohibiting ramping is suggestive of perceived vulnerability to manipulators and is therefore associated with higher spreads but ramping regulations also lower volatility, ceteris paribus. Other regulations about market integrity lower both spreads and volatility.

The 2SLS results in Table 33 Panel B are consistent with the OLS estimates in Panel A for the AI-Spread Equation Pair. As a result, the Hausman-Wu specification test rejects the need for IV TSLS estimation. Similarly, where significant, 2SLS variables match the sign of their OLS counterparts in the AI-Volatility equation pair. Again, the Hausman-Wu test rejects the need for IV 2SLS estimation. We conclude that simultaneity bias in the thickly-traded deciles is not material for the AI-Volatility and AI-Spread equation pairs. Subsequent system estimation will address, however, the remaining possibility of cross-equation correlation of the error terms, using Seemingly Unrelated Regressions (SURL).

142

3.6.3 OLS, 2SLS and FIML results for Moderately-traded Deciles The ordinary least square regression (OLS), two-stage least square regression (2SLS) and Full Information Maximum Likelihood (FIML) results for moderately-traded deciles (decile 4, 5, 6 and 7) are presented in Table 34.

Table 34 Regression Results for Potential Endogenous Equations for moderately-traded deciles across 24 Securities Exchanges in 2005

Panel A: OLS

With Fixed Effects AI Model Volatility Model Spread Model F-test 6.86*** 10.09*** 161.97*** R-square 0.7102 0.7023 0.9830 -0.3522(t= -1.21) 0.1995*** (t=3.55) N/A βspr -1.3882***(t=2.66) 2.1948***(t=9.23) -3.9521***(t=-17.99) βcall -0.8916 (t=-0.76) N/A N/A βDMA 1.1498***(t=3.18) -2.6347*** (t=-10.07) 10.5837***(t=20.16) βRTS 0.6178(t=1.45) -2.7780** (t=-8.16) 22.1870***(t=19.87) βRampReg 0.0803(t= 0.82) N/A 0.4125***(t=4.68) βvol N/A 0.0473**(t=2.13) -1.4952***(t=-17.38) βregs N/A 0.2327(t=4.74) -0.3485***(t=-15.03) βliq 5 out of 23 13 out of 23 20 out of 23 βexch α -1.4953( t=-1.55) -2.7532***(t =-12.51) 5.5170***(t=13.43)

143

Panel B: 2SLS

With Fixed Instrument: Volatility Instrument: Spread Effects

Hausman- 2.92 0.03 Wu Stats AI Model Volatility Model AI Model Spread Model F-test 5.12*** 4.61*** 5.12*** 16.07*** R-square 0.4015 0.5352 0.4015 0.8446 -0.2482(t=-1.19) 0.4970**(t=2.51) 0.2482( t=1.19) N/A βspr -1.3473***(t=-2.61) 1.9304***(t=5.36) -1.3473**(t=-2.61) -3.7108***(t=-5.35) βcall -0.3400(t=-0.78) N/A -0.3400(t=-0.78) N/A βDMA 0.9515**(t=2.53) -2.3076***(t=-5.90) 0.9515**(t=2.53) 11.8457***(t=7.17) βRTS 0.5707*(t=1.69) -2.3800***(t=-4.69) 0.5707*(t=1.69) 26.4773***( t=8.00) βRampReg 0.1409(t=0.27) N/A 0.1409(t=0.27) 0.9962***(t=4.09) βvol N/A 0.2137(t=0.67) N/A -1.9181***(t=-7.88) βregs N/A 0.1671*** (t=3.47) N/A -0.2325(t=-4.27) βliq 4 out of 23 8 out of 23 4 out of 23 20 out of 23 βexch α -1.6026( t=-0.71) -3.9844***(t =-8.82) 1.6026(t=-0.71) 2.3468***(t=2.77)

Panel C: FIML

With Fixed Instrument: Volatility Instrument: Spread Effects AI Model Volatility Model AI Model Spread Model R-square 0.3986 0.7950 0.4025 0.9075 -0.1457(t= -0.58) 0.3227***(t=4.25) -0.2492(t=-1.16) N/A βspr -1.2827(t=-1.35) 0.9743***(t=.53) -1.4949*(t=-1.66) -2.3351***(t=-3.52) βcall -0.5574(t=-0.74) N/A -0.3674(t=-0.52) N/A βDMA 1.2169***(t=2.74) -1.5618*** (t=-12.56) 1.1738***(t=2.78) 2.0440***(t=2.71) βRTS 0.5158(t=0.98) -1.6987***( t=-11.19) 0.6469(t=0.89) 2.8160***(t=4.28) βRampReg 0.0430(t=0.11) N/A 0.0916(t=0.25) 1.2853***(t=15.45) βvol N/A 0.1876*** (t=12.78) N/A -0.3866***(t=-4.28) βliq 9 out of 23 8 out of 23 7 out of 23 6 out of 23 βexch α -1.5516 (t=-0.98) -4.3848***( t =-25.42) -1.6998(t=-1.13) 5.9362***(t=6.60)

144

Many more significant drivers of alert incidence emerge here in the moderately-traded deciles. As hypothesized, the costs of manipulation are not like the thickly-traded securities yet excessive and detection proves much more difficult than in the think- stocks. Call auctions at the close reduce alert incidence in OLS and 2SLS estimates. Real time surveillance and ramping regulations lower volatility and are associated in cross-section with vulnerability to trade-based manipulation and therefore both higher alert incidence and higher spreads in OLS and 2SLS. Volatility raises spreads in all models.

Here market integrity regulations not specifically about ramping prohibited conduct lower spreads in all three models. Not surprisingly, we again conclude based on the Hausman-Wu test that IV estimation is not required. Simultaneity bias in the AI- Spread (market integrity – market efficiency pair) is minimal.

145

3.6.4 OLS, 2SLS and FIML results for Thinly-traded Deciles The ordinary least square regression (OLS), two-stage least square regression (2SLS) and Full Information Maximum Likelihood (FIML) results for thinly-traded deciles (decile 8, 9 and 10) are presented in Table 35.

Table 35 Regression Results for Potential Endogenous Equations for thinly-traded deciles across 24 Securities Exchanges in 2005

Panel A: OLS

With Fixed Effects AI Model Volatility Model Spread Model F-test 3.48*** 19.90*** 32.47*** R-square 0.6543 0.9154 0.9464 -0.0009(t= -0.02) 0.4132***(t=3.02) N/A βspr 2.1942(t=1.08) 0.8322(t=1.09) -3.3923***(t=-5.88) βcall 3.6283*(t=1.82) N/A N/A βDMA -6.6846**(t=-2.07) -1.5364(t=-0.80) 9.8075***(t=7.84) βRTS -5.7615**(t=-2.09) -2.5914(t=-0.67) 19.9007***(t=8.02) βRampReg -0.8529(t= -0.63) N/A 0.3999***(t=5.39) βvol N/A 0.0434(t=0.16) -1.3158***(t=-7.20) βregs N/A 0.3760***(t=6.86) -0.3351***(t=-5.71) βliq 6 out of 23 5 out of 23 20 out of 23 βexch α -5.6848**( t=-2.46) -6.3500***(t =-9.63) 4.7542***(t=5.39)

146

Panel B: 2SLS

With Fixed Instrument: Volatility Instrument: Spread Effects

Hausman- 5.45 0.47 Wu Stats AI Model Volatility Model AI Model Spread Model F-test 4.76*** 12.73*** 5.96*** 0.02 R-square 0.4087 0.6488 0.4640 0.0015 -0.5534(t=-1.02) 0.4971**(t=2.49) 0.1943( t=0.62) N/A βspr 0.2628(t=0.36) 0.4125(t=1.53) 0.7056(t=1.17) -4.2293(t=-0.05) βcall 0.7475(t=0.87) N/A -0.0439(t=-0.07) N/A βDMA -1.1742(t=-1.62) -0.7679***(t=-2.99) -1.2001*(t=-1.81) 61.7170(t=0.03) βRTS -1.3938*(t=-1.98) -1.1679***(t=-3.28) -1.3280**(t=-2.14) 117.3303( t=0.03) βRampReg -0.7525(t=-1.27) N/A -0.0096(t=-0.03) -42.2121***(t=-0.03) βvol N/A -0.0418(t=-1.19) N/A -11.1745***(t=-0.03) βregs N/A 0.4442*** (t=8.10) N/A 9.1848(t=0.03) βliq 2 out of 23 3 out of 23 2 out of 23 0 out of 23 βexch α 7.0739*( t=-1.92) -7.0548***(t =-17.03) -2.0224(t=-0.94) -248.988(t=-0.03)

Panel C: FIML

With Fixed Instrument: Volatility Instrument: Spread Effects AI Model Volatility Model AI Model Spread Model R-square 0.4728 0.7231 0.4737 0.8834 -0.0725(t= -0.37) 0.3023**(t=2.26) -0.0626(t=0.7738) N/A βspr 0.5224(t=0.61) 0.3380(t=1.18) 0.5538(t=0.57) -1.9801***(t=-3.53) βcall 0.1725(t=0.22) N/A 0.1128(t=0.12) N/A βDMA -1.1316*(t=-1.83) -0.8771*** (t=-3.56) -0.8943(t=-1.57) 1.7965***(t=3.22) βRTS -1.2396(t=-1.59) -1.2414***( t=-5.76) -1.0741(t=-1.20) 2.4100***(t=4.88) βRampReg -0.0088(t=-0.03) N/A -0.0976(t=-0.32) 0.8189***(t=9.14) βvol N/A 0.3715*** (t=11.66) N/A -0.3615***(t=-5.61) βliq 2 out of 23 2 out of 23 7 out of 23 8 out of 23 βexch α -2.8525* (t=-1.87) 6.9477***( t =-20.26) -3.2102**(t=-2.42) 4.2593***(t=4.57)

147

It can be seen from Table 35 that in the thinly-traded securities deciles, higher spreads are associated with greater volatility in cross-sectional OLS, 2SLS or FIML. Call auctions at the close to unwind intra-day exposures lowers spreads. Direct Market Access in these thinly-traded securities appear to advantage the manipulators and raise the incidence of alerts, perhaps because with little liquidity available counterparties at the end of a ramping scenario cannot find buyers even though execution speed is very fast.

Unlike in thickly-traded and moderately-traded deciles, real time surveillance and ramping regulations lower alert incidence and volatility in OLS and FIML estimates while continuing to be associated with higher spreads. Volatility, as usual, raise spreads suggesting more asymmetric information. Market quality assurance in the form of regulations promoting fair and orderly markets lowers the spreads. Liquidity (such as it is) in these thinly-traded stocks is sufficient to lower the spread despite higher volatility. Again, Hausman-Wu tests reject the hypothesis that SSE is reduced by IV estimation; single equation results are unbiased.

148

3.7 Probability of Real-time Surveillance Surveillance of financial markets has a long history. The reasons why are both obvious and subtle. The assurance of market integrity typically requires an aggressive surveillance regime in tandem with regulatory enforcement against those who conduct prohibited practices. In addition, however, as a self-regulatory organization (SRO), many exchanges have more extensive obligations to monitor trading, detect manipulative behaviour, and punish violators. With these SRO obligations comes a greater quality assurance need of the market participants and higher consequent standards for scrutinizing market conduct than might exist in an industry like insurance that operates under detailed and continuous regulatory review and approval.

The adoption of real-time surveillance (RTS) has grown in the last decade concurrent with the growth of electronic trading. Lower latency and an explosion of trade executions barely imaginable a few years ago, necessitate real-time mechanisms for capturing and processing surveillance data. Across the 240 exchange deciles we analyse the probability of adopting RTS proves to be systematically related to four determinants.

In addition to those factors from the aspects of Regulations, Technology, Market Infrastructure and Participants which have been used in previous analysis, we also collected the Foreign Direct Investment as percentage of Gross Fixed Capital (FDI) in 2005 across the 24 nations in our sample from the United Nations Conference on Trade and Development Database. We will include FDI in our RTS PROBIT analysis which hopefully will find some evidence about the impact of foreign direct investment (e.g., foreign investment banks) on securities exchanges integrity.

3.7.1 Model Specification

̃ where

149

Prob = The probability of the deployment of Real Time Surveillance System (i.e., SMARTS) in market i at time t

= the constant that would be estimated

= the correlation coefficient between Alerts Incidence and the corresponding proxy

= the average numbers of Ramping Alerts of market i at time t

= The average standard deviation of daily returns of market i at time t

= The average market turnover of market i at time t

= The dummy variable for the existence of direct market access of market i at time t

= Foreign direct investment as % of Gross Fixed Capital to the country of market i at time t

̃ = The residual error term

3.7.2 Empirical Results Table 36 reports a PROBIT analysis of real-time surveillance system development. Panel A pools the results from all liquidity deciles which in total exhibit 33% frequency of RTS, meaning 8 of the 24 exchanges. Subsequent panels B, C and D, report the sub-groupings of liquidity deciles found to be relevant in our foregoing study of endogeneity.

150

Table 36 PROBIT Analysis of Real-time Surveillance System Deployment

Panel A: All Deciles Discrete Response Frequency 0 (66%) 1 (33%)

Without Fixed Effects With Fixed Effects N 240 240 Log Likelihood -97.76 -3.4718E-8 94.22 289.75 Likelihood Ratio 0.28 0.55 Aldrich-Nelson 0. 1327** (t=2.13) 1.3573( t=0.25) Βai

1.5636***( t =6.48) 142.39***(t=59.88) βDMA -0.0988***( t= -3.29) -3.2514(t=-0.09) βregs -0.3184**( t=-1.97) 7.0394 (t=0.70) βvol 0.0968*** ( t= 2.67) -2.4852(t=-0.08) Βliq 1.7875*** (t= 3.46) 83.6882***(t=87.90) βFDI --3.3926*** ( t= -3.36) -37.7427***(t=-15.87) α β N/A 16 out of 23 exch

Panel B: Decile 1, 2, 3 Without Fixed Effects With Fixed Effects 72 72 N Log Likelihood -21.50 -3.2837E-8

Likelihood Ratio 43.91 86.92

Aldrich-Nelson 0.38 0.55 0. 4221* (t=1.65) -382.0626***( t=-7.26) Βai 1.5285**( t =2.49) 4224.9291***(t=1283.78) βDMA -0.1133*( t=-1.72) -9.5370(t=-0.02) βregs -2.0903**( t=-2.52) -1143.8055*** (t=-105.47) βvol 0.4128*** ( t= 2.97) 49.9484(t=0.57) Βliq β 2.5194* (t= 1.69) 3155.9683***(t=1191.19) FDI α -14.8848*** ( t= -3.20) -10306***(t=-3130.69) N/A 3 out of 23 βexch

151

Panel C: Decile 4, 5, 6, 7

Without Fixed Effects With Fixed Effects N 96 96 Log Likelihood -28.19 -6.4724E-8 Likelihood Ratio 59.52 115.9 0.38 0.55 Aldrich-Nelson 1.1149*** (t=3.19) 1118.7557**( t=2.17) Βai 2.8118***( t =3.91) 2315.6456***(t=7.68) βDMA

-0.2222***( t=-2.94) -94.0829(t=-0.62) βregs -0.1999( t=-0.45) 207.2831 (t=0.98) βvol 0.3120** ( t= 2.53) 102.5849(t=0.71) Βliq 0.6523 (t= 0.77) 704.7527***(t=39.28) βFDI α -4.3470* ( t= -1.65) -378.0217***( t =9.92) N/A 11 out of 23 βexch

Panel D: Decile 8, 9, 10

Without Fixed Effects With Fixed Effects N 72 72 Log Likelihood -31.09 -5.944E-8 Likelihood Ratio 24.73 86.92 Aldrich-Nelson 0.26 0.55 0.0372 (t=0.51) 2.1517( t=0.19) Βai β 1.5268***( t =3.56) 56.5862***(t=15.74) DMA -0.0910*( t=-1.64) -3.3795(t=-0.13) βregs -0.1113( t=-0.47) -1.1287 (t=-0.08) βvol Β 0.0606( t= 0.78) -1.6398(t=-0.04) liq 1.8022* (t=1.96) 37.6652***(t=141.53) βFDI -2.2170 ( t=-1.31) -9.3265***( t=-2.59) α N/A 16 out of 23 βexch

152

In the All Deciles results in Panel A, alert incidence is positively related to the adoption of RTS, indicating a perceived vulnerability to manipulation that RTS can help mitigate. Secondly, direct market access (DMA) facilitates quick responses by both manipulators and counterparties, requiring an expanded capability by the market surveillance officials to monitor the situations as they evolve. More extensive regulations (REG) serve as something of a substitute for RTS, reducing the likelihood of surveillance. Higher liquidity (LIQ) increases RTS, perhaps because higher turnover accentuates the problem of false positives in scrutinizing potentially manipulative trades. RTS assists in separating the true and false positives in a surveillance program. Finally, a higher percentage of foreign direct investment (FDI) raises the vulnerability of an exchange to ramp-and-dump manipulative schemes, so RTS increases to combat it. Overall 16 out of 23 exchanges have significant idiosyncratic effects, and the LR rises from 94.22 to 289.75 when these exchange fixed effects are incorporated into the specification.

In Panel B for the most liquid deciles, many of the aforementioned results are quite comparable: DMA, higher liquidity, and foreign direct investment increase RTS, whilst more extensive regulations unrelated to ramping reduce the adoption of RTS. In addition, higher volatility reduces RTS. Here in the most liquid stocks not only is trade-based manipulation more difficult to detect whatever the means of surveillance, but in addition, higher volatility offers safe harbor legal defences to alleged manipulators. This makes successful prosecution and enforcement actions much more difficult. Finally, one anomalous result is that with fixed effects, alert incidence appears to reduce the adoption of RTS in these most liquid stocks. Overall 3 out of 23 exchanges have significant idiosyncratic effects, and the LR rises from 43.91 to 86.92 when the exchange fixed effects are incorporated into the specification.

In Panel C (deciles 4, 5, 6 and7), alert incidence is at its peak because costs of trade- based manipulation are moderate (unlike in highly liquid securities), yet detection is still quite difficult (unlike in thinly-traded stocks). For these moderately liquid stocks, higher alert incidence, DMA, higher liquidity, and foreign direct investment increase RTS. More extensive regulations unrelated to ramping and greater volatility reduce the adoption of RTS. In addition, higher regulatory burdens separable from ramping

153

regulations reduce the adoption of RTS. Overall 11 out of 23 exchanges have significant idiosyncratic effects, and the LR rises from 59.52 to 115.9 when the exchange fixed effects are incorporated into the specification.

Finally, in Panel D for the least liquid stocks, only DMA and FDI consistently increase the adoption of RTS. The startling finding in this grouping is that FDI decreases RTS. That is, the determinants of Prob(RTS) are indeed radically different for very illiquid stocks in small exchanges. Alert incidence probably does not matter because no exchanges develop RTS systems to monitor their trivial most illiquid stocks. Variation in volatility probably does not matter because manipulation in these thinnest trading stocks is very easy to detect. Liquidity probably does not matter because so little successful manipulation is attempted in these 8.9.10 decile stocks that false positives seldom arise. Overall 16 out of 23 exchanges have significant idiosyncratic effects in these thinnest stocks, and the LR rises from 24.73 to 86.92 when the exchange fixed effects are incorporated into the specification

154

3.8 Structural Equation Estimates

Having established that the likelihood of adopting RTS is itself related to alert incidence, volatility, and spreads, we estimate a simultaneous system of equations characterizing integrity (AI), efficiency (SPR), and Prob(RTS). Two possible model structures about the role of RTS are pursued: one in which the dichotomous decision to adopt RTS has simple intercept-shifting effects on the other dependent variables (the Simple Model), and another in which Prob(RTS) is specified as affecting a full set of interaction terms involving all the slope coefficients throughout the three structural equations (the Full Interaction Effects Model). Because Hausman-Wu tests indicate possible cross-equation covariance of the error structure, the estimation method for the Simple Model is Seemingly Unrelated Regression (SUR). The (AI, SPR, PROBIT) system is estimated with maximum likelihood using the SAS QLIM procedure.

3.8.1 The Simple Model As in the earlier error components modelling of alerts and spreads, Wald tests show that the liquidity deciles cannot be validly pooled.12 We therefore again pursue three separate estimations of alert incidence and spreads in the highly liquid deciles (1, 2, 3), the moderately liquid deciles (4, 5, 6, 7), and the illiquid deciles (8, 9, 10). In addition, fixed effects of the twenty-four exchanges are introduced to adjust the parameters for potentially idiosyncratic features of several exchanges. 13 Table 37 displays the results for the Simple Model.

12 Possible heteroskedastic error variance across the liquidity decile groupings necessitates a Wald test. 13 We retain all exchange fixed effects found to be significant in OLS regressions at the 5% level. 155

Table 37 Structural Equation Estimates for the Simple Model

Panel A: Integrity (AI) Equation

Deciles Group βspr βcall βDMA βRTS βRampReg βvol βexch α

1, 2,3 OLS F: 16.69*** 0.2550* 0.3995 -0.7370** -0.1391 0.0043 -0.3466 3/23 -1.8322* 2 R : 0.6655 (t=1.99) (t=1.06) (t=-2.00) (t=-0.38) (t=0.01) (t=-1.39) (t=-1.80) SURL 0.3147** 0.3660 -0.7148** -0.2136 0.0540 -0.3581 3/23 -1.5561 (t=2.46) (t=0.97) (t=-1.94) (t=-0.59) (t=0.16) (t=-1.44) (t=-1.53)

4,5,6,7 OLS F: 5.20*** -0.1733 -1.3882*** -0.4209 1.1498*** 0.6178* 0.0803 5/23 -1.4953 2 R : 0.3274 (t=-1.55) (t=-2.66) (t=-1.06) (t=3.18) (t=1.82) (t=0.34) (t=-1.55) SURL -0.1538 -1.2624** -0.5052 1.2577*** 0.5927* 0.0697 4/23 -1.5476 (t=-1.25) (t=-2.42) (t=-1.27) (t=3.49) (t=1.75) (t=0.29) (t=-1.61)

8, 9,10 OLS F: 6.29*** -0.0737 0.5507 0.1341 -1.0455 -1.2259** -0.1219 2/23 -3.2921*** 2 R : 0.4012 (t=-0.45) (t=0.96) (t=0.24) (t=-1.60) (t=-2.02) (t=-0.63) (t=-3.42) SURL -0.0792 0.8079 -0.1618 -0.4121 -1.2589** -0.1504 2/23 -3.5921*** (t=-0.50) (t=1.45) (t=-0.30) (t=-0.64) (t=-2.10) (t=-0.77) (t=-3.75)

Panel B: Efficiency (Spread) Equation

Deciles Group βcall βRTS βRampReg βvol βregs βexch α

1, 2,3 OLS F: 15.00*** -2.0506*** 9.9511*** 23.9762*** -1.1643*** -1.9436*** 20/23 -4.2743*** 2 R : 0.8194 (t=-5.70) (t=6.65) (t=6.23) (t=-4.03) (t=-6.31) (t=-7.48) SURL -2.0977*** 9.8739*** 23.8535*** -1.1541*** -1.9315*** 20/23 -4.2743*** (t=-5.85) (t=6.62) (t=6.23) (t=-4.01) (t=-6.30) (t=-7.48)

4,5,6,7 OLS F: 38.22*** -5.2551*** 11.9170*** 22.6508*** 0.2043 -1.5534*** 20/23 0.6857 2 R : 0.9039 (t=-12.76) (t=11.28) (t=9.94) (t=1.15) (t=-8.85) (t=1.31) SURL -5.5283*** 11.7625*** 22.5318*** 0.1955 -1.5432*** 20/23 0.6680 (t=-12.78) (t=11.16) (t=9.91) (t=1.10) (t=-8.81) (t=1.28)

8, 9,10 OLS F: 19.41*** -5.1905*** 11.4847*** 21.8973*** -0.0863 -1.4977*** 20/23 0.5972 2 R : 0.8617 (t=-8.31) (t=7.30) (t=6.89) (t=-0.66) (t=-6.43) (t=0.93) SURL -5.1548*** 11.4143 22.0818*** -0.0966 -1.5131*** 20/23 0.6136 (t=-8.28) (t=7.29) (t=6.98) (t=-0.74) (t=-6.53) (t=0.96)

156

Panel C: RTS Equation

Deciles Group βAI βDMA βregs βvol βliq βFDI βexch α

1, 2,3 OLS F: 295.20*** -0.0006 0.1138*** 0.1632*** 0.0199 -0.0001 0.9440*** 9/23 -0.1256 2 R : 0.9841 (t=-0.15) (t=5.01) (t=4.53) (t=1.14) (t=-0.01) (t=21.70) (t=-1.28) SURL -0.0018 0.1159*** 0.0164*** 0.0211 -0.0001 0.9440*** 9/23 -0.1247 (t=-0.42) (t=5.11) (t=4.57) (t=1.22) (t=-0.02) (t=21.75) (t=-1.28)

4,5,6,7 OLS F: 19.92*** 0.0639*** 0.3640*** -0.0054 -0.1278 0.0123 0.7184*** 4/23 -0.5507 2 R : 0.7214 (t=3.76) (t=5.64) (t=-0.62) (t=-1.43) (t=1.10) (t=4.72) (t=0.1789) SURL 0.0718*** 0.3679*** -0.0063 -0.1354 0.0155 0.7045*** 4/23 -0.5985 (t=4.23) (t=5.70) (t=-0.72) (t=-1.52) (t=1.38)) (t=4.63) (t=-1.48)

8, 9,10 OLS F: 16.69*** 0.0548*** 0.7257*** -0.0591*** -0.0730** 0.0352** 0.1508*** 5/23 -0.2597 2 R : 0.7418 (t=3.87) (t=8.97) (t=-5.88) (t=-2.03) (t=2.58) (t=2.95) (t=0.93) SURL 0.0751*** 0.7414*** -0.0591*** -0.718** 0.0293** 0.1689 5/23 -0.1258 (t=5.47) (t=9.29) (t=-6.00) (t=-2.03) (t=2.22) (t=1.10) (t=-0.46)

Integrity Equation (AI) In Panel A, for the most liquid deciles 1, 2 and 3, high speed execution on DMA results in fewer ramping alerts and increased market integrity.

Our interpretation is that counterparties are able to unwind exposed positions fast enough to prevent profitable ramp-and-dump manipulation. In addition, we find that higher spreads (which are associated with more asymmetric information in these most liquid deciles) increase the incidence of ramping alerts. Manipulation always necessitates substantial trading volume but especially so in these highly liquid securities, and the more asymmetric the information, the more likely momentum traders will become involved in reaching that requisite volume. Three exchanges (Shenzhen, Korea and New Zealand) have significant dummy variable effects at α = 0.05 (all negative, suggestive of greater market integrity), and the alert incidence model exhibits adjusted R-square of 0.665 and a 99% F(16.69, 9 d.f.) test.

In the moderate liquidity deciles 4,5,6,7 in Panel B, we find that the presence of a closing auction (CloseAucDum) reduces the incidence of ramping alerts. Trade-based manipulation proves more difficult when a manipulator’s counterparties can use closing auctions to unwind their intraday exposures. The RTS dummy variable is

157

significantly positively related to alert incidence. In the absence of any panel data on the dynamic effects of adopting RTS, what we are observing in cross section is the perceived vulnerability of certain exchanges to manipulation and their consequent adoption of RTS plus the regulatory regimes required to have a salutary effect on market integrity. Four exchanges (Shanghai, Shenzhen, Bombay and NASDAQ) have significant dummy variable effects at α = 0.05 (again all negative, suggestive of greater market integrity), and the alert incidence model for moderate liquidity deciles exhibits an adjusted R-square of 0.327 and a 99% F(5.20, 11 d.f.) test.

In the thin liquidity deciles 8,9,10, displayed in Panel C, we find that the presence of a regulation specifically prohibiting ramp-and-dump manipulation reduces the incidence of ramping alerts. None of the other determinants from the more liquid deciles are significant. Two exchanges (Xetra and New Zealand) have significant dummy variable effects at α = 0.05 (the first negative, but the second positive, suggestive of lower market integrity). The alert incidence model for the least liquid deciles exhibits an adjusted R-square of 0.401 and a 99% F(6.29, 9 d.f.) test.

Efficiency Equation (Spr) Table 37 Panel B displays results for the determinants of volume-weighted effected spreads. For the most liquid deciles, the absence of closing auctions, fewer regulations in pursuit of market integrity, and less volatility increases the time-weighted quoted spread. The latter findings are perhaps anomalous, but recall that reduced volatility increases the ability of market makers to detect the presence of trade-based manipulators in the marketplace. The market makers rationally raise spreads, ceteris paribus, when manipulators are detected to increase the cost of ramp-and-dump strategies. Across these 72 exchanges in 2005, spreads are higher in the presence of regulations against ramping and the RTS necessary to detect it. Numerous exchanges have significant dummy variable fixed effects at α = 0.05 (both negative and positive). The Spreads model for these most liquid deciles exhibits an adjusted R-square of 0.819 and a 99% F(15.0, 23 d.f.) test.

As in very liquid stocks, closing auctions and more regulations in pursuit of market integrity lower quoted spreads in the moderate liquidity deciles 4,5,6,7. Also

158

consistent with the earlier subsample, RTS and a regulation specifically prohibiting ramping indicate in cross-section the perceived likelihood of more ramping which again raises quoted spreads. Nineteen exchanges have significant fixed effects at α = 0.05 (both negative and positive). The Spreads model for these moderately liquid deciles exhibits an adjusted R-square of 0.904 and a 99% F(38.2, 24 d.f.) test.

Even in the thin liquidity deciles 8, 9 and 10, displayed in Panel C, closing auctions and more regulations in pursuit of market integrity lower quoted spreads. Consistent with both earlier subsamples, RTS and a regulation specifically prohibiting ramping again raise quoted spreads. Numerous exchanges have significant fixed effects at α = 0.05 (both negative and positive). The Spreads model for these illiquid deciles exhibits an adjusted R-square of 0.862 and a 99% F(19.4, 24 d.f.) test.

Real-time Surveillance Equation (RTS) In modelling the decision to adopt real-time surveillance systems as a dichotomous dummy variable, the estimations again differ by liquidity decile grouping. In the most liquid deciles 1, 2 and 3, DMA, integrity regulations (REGS), and foreign direct investment (FDI) all increase the likelihood of developing an RTS system. Faster execution channels imply a predisposition towards more sophisticated electronic trading and thereby more sophisticated surveillance to assure market integrity. More integrity regulations convey a commitment to quality assurance for market integrity as well as well as market efficiency. Higher FDI is thought to increase threats to market integrity from hedge funds and proprietary trading desks offshore.

In exchanges with a preponderance of moderately liquid deciles, the presence of DMA and higher FDI again raise the likelihood of adopting RTS systems. In addition, higher alert incidence signals increased vulnerability and also increases the likelihood of adopting RTS. Integrity regulations unrelated to ramping, here have no significant effect on RTS.

In the exchanges with a preponderance of thin trading, higher alert incidence, the presence of DMA, and higher FDI again increase the likelihood of adopting RTS. In addition, within these thinnest trading deciles, more liquidity increases the adoption of

159

RTS. Two variables in deciles 8, 9 and 10 retard the likelihood of RTS system development: REGS and Volatility. Recall that more volatility offers legal safe harbours to alleged manipulators seeking to avoid prosecution, so RTS is less valuable as a tool for assuring market integrity in those cases. An emphasis on insider trading or other integrity regulations unrelated to ramping appears to decrease the willingness to invest in RTS in exchanges characterized by thin trading.

160

3.8.2 The Full Interaction Effects Model Integrity Equation (AI) Table 38 Panel A displays results for the most liquid deciles 1, 2 and 3. High speed execution on DMA lines again results in fewer ramping alerts and increased market integrity. Higher spreads are associated with higher alert incidence. Our interpretation remains that with more asymmetric the information, momentum traders can be enticed to become involved in reaching the requisite volume to ramp a thickly-traded stock. As in the Simple Model, three exchanges (Shenzhen, Korea and New Zealand) have significant dummy variable effects at α = 0.01. Here, the first two fixed effects are negative, suggestive of greater market integrity, and the third is positive suggestive of the reverse.

In the moderate liquidity deciles 4, 5, 6 and 7 in Panel B, we again find that access to DMA lines reduces alert incidence. In addition, RTS is significantly positively related to alert incidence. As we have emphasized before, what we are observing in cross section is RTS adoption by those exchanges perceived to have greater vulnerability to manipulation. Two exchanges (Shanghai and Shenzhen) have significant dummy variable effects at α = 0.05 (both negative, suggestive of greater market integrity).

In the thinnest-trading deciles 8, 9 and 10, the only significant determinant of alert incidence here in the full interaction effects model is RampReg. In the presence of a regulation specifically prohibiting ramping, potential manipulators appear convinced that the exchange will pursue violators, and fewer ramping alerts are observed. One exchange (Xetra) has a significant negative dummy variable effect at α = 0.05 (suggestive of greater market integrity), and one exchange (New Zealand) has the reverse.

Efficiency Equation (Spread) Table 38 Panel B displays results for the determinants of spreads. For the most liquid 1,2 and 3 and moderately liquid deciles 4, 5, 6 and 7, closing auctions and regulations in pursuit of market integrity as well as greater liquidity result in lower time-weighted quoted spreads. Across these 72 and 96 exchange deciles in 2005, spreads are higher in the presence of 1) RTS systems that respond to the perceived threat to market

161

integrity and 2) the regulations against ramping necessary to punish it. Numerous exchanges have significant dummy variable fixed effects at α = 0.05 (both negative and positive).

Even in the 72 exchange deciles with the thinnest trading (deciles 8, 9 and 10), displayed in Panel C, closing auctions and more regulations in pursuit of market integrity lower quoted spreads. Consistent with both earlier subsamples, RTS and a regulation specifically prohibiting ramping again raise quoted spreads. As in the earlier decile sub-groupings, numerous exchanges again have significant fixed effects at α = 0.05 (both negative and positive).

Prob(RTS) Equation In modelling the probability of adoption of RTS in a maximum likelihood system of equations for Ai, Spreads, and Prob(RTS), the estimations again differ by liquidity decile grouping. In the most liquid deciles 1, 2 and 3, higher spreads, DMA lines and foreign direct investment (FDI) again increase the likelihood of developing an RTS system. In addition, however, higher alert incidence (AI) and higher liquidity in these already relatively liquid stocks increase Prob(RTS). A focus on other types of market integrity regulations, other than ramping regulations, reduces the Prob(RTS) perhaps because client-broker conflicts compete for regulatory attention. Finally, because high volatility leads to less successful ramping indictments attributed to the legal safe harbours afforded by volatile price patterns, greater volatility reduces the Prob(RTS).

In exchange deciles with moderately liquid stocks, the presence of DMA lines and higher FDI again raise the likelihood of adopting RTS systems. In addition, higher alert incidence and increased trading volume (liquidity) increases the likelihood of adopting RTS. In this sub-grouping of liquidity deciles, integrity regulations unrelated to ramping and volatility have no effect on Prob(RTS).

In the thin trading exchange deciles, higher alert incidence, the presence of DMA lines, and higher FDI again increase the likelihood of adopting RTS. Here, liquidity which is already quite low and the number of market integrity regulations are unrelated to Prob(RTS). In illiquid stocks and exchanges, greater volatility again

162

discourages the adoption of RTS presumably because of the difficulty of securing convictions.

163

Table 38 The Full Interaction Effects Model Estimates

Panel A: Integrity (AI) Equation

Deciles Group βspr βcall βDMA ΒProb(RTS) βRampReg βvol βexch α

1, 2,3 ML N: 72 0.2297* 0.3458 -0.8107** 0.1364 -0.0005 -0.3078 3/23 -1.8397* Log LH: -89.33 (t=1.74) (t=0.98) (t=-2.35) (t=0.39) (t=-0.01) (t=-1.34) (t=-1.91) 4,5,6,7 ML N: 96 -0.0269 -0.2287*** -1.3116*** 1.6521*** 0.0147 -0.4314 2/23 -3.3295*** Log LH: -209.37 (t=-0.15) (t=-0.60) (t=-3.66) (t=4.73) (t=0.04) (t=-1.42) (t=-3.44) 8, 9,10 ML N: 72 -0.1803 0.3619 0.0074 -0.6112 -1.3196** -0.0826 2/23 -3.3267*** Log LH: 10.64 (t=-1.11) (t=0.64) (t=0.01) (t=-0.93) (t=-2.18) (t=-0.43) (t=-3.49)

Panel B: Efficiency (Spread) Equation

Deciles Group βcall βProb(RTS) βRampReg βvol βregs βexch α

1, 2,3 ML N: 72 -2.4053*** 7.9109*** 18.5797*** -0.3078 -1.9436*** 20/23 3.2416*** Log LH: -89.33 (t=-9.83) (t=11.36) (t=11.11) (t=-1.34) (t=-6.31) (t=5.41) 4,5,6,7 ML N: 96 -0.9427* 2.6807*** 7.7208*** 0.1463 -0.5519** 13/23 2.4904*** Log LH: -209.37 (t=-1.61) (t=2.63) (t=2.77) (t=0.52) (t=-2.24) (t=2.80) 8, 9,10 ML N: 72 -5.1504*** 11.5069*** 21.8500*** -0.0660 -1.4930*** 20/23 0.6328 Log LH: 10.64 (t=-9.97) (t=8.83) (t=8.31) (t=-0.61) (t=-7.76) (t=1.19)

Panel C: Prob(RTS) Equation

Deciles Group βAI βDMA βregs βvol βliq βFDI βexch Α

1, 2,3 ML N: 72 0.6544*** 1.5227*** -0.0718*** -0.1033*** 0.1109*** 0.6186*** 1/23 -1.7417*** Log LH: -89.33 (t=55.64) (t=406.53) (t=-5.46) (t=-7.48) (t=3.18) (t=2463.8) (t=-11.44) 4,5,6,7 ML N: 96 0.9138*** 1.5227*** -0.0600 0.0969 0.1094*** 1.0354*** 7/23 -0.7022*** Log LH: -209.37 (t=6.27) (t=43.79) (t=-0.27) (t=0.42) (t=8.30) (t=39.19) (t=-20.19) 8, 9,10 ML N: 72 0.5512*** 2.3054*** -0.0883 -0.1247** 0.0058 0.9240*** 6/23 -0.3590*** Log LH: 10.64 (t=8.77) (t=12.15) (t=-0.07) (t=-1.94) (t=0.26) (t=184.19) (t=8.31)

164

3.9 Summary and Conclusion In Chapter 3 of this thesis, we tested the correlation between the levels of market integrity and a combination of factors from the following four potential drivers deciding the market quality across securities exchanges: • Securities Markets Trading Regulations • Securities Markets Technologies • Securities Market Infrastructure • Securities Market Participants

The sample we analysed in this Chapter consisted of 24 national securities exchanges and the period of our analysis is for year 2005. Among those 24 national securities from our sample, 14 are from the Asia-Pacific Region, 6 are from European countries, 2 are from the United States the rest are from Africa.

We proxied the level of market integrity by the incidence of ramping alerts that are also applied in Chapter 2 for the cross-markets analysis of the relationship between spreads and alerts. The reason for choosing ramping alert is because the order book data is extremely difficult to be acquired while the ramping alert is the only alert that is dependent on just trade level data from the SMARTS Market Surveillance Alerts Portfolios.

Based on the Securities Markets Trading Regulations from Cumming and Cumming and Sofia (2008), we built up a proxy for securities markets trading regulations as the number of trading rules that are monitored (e.g., by the SMARTS Real-time Surveillance Platform) from each securities exchange our example. In terms of Securities Markets Technology, we used the deployment of SMARTS to create a dummy variable as a proxy for the Experience with Sophisticated Surveillance Technology of a securities exchange. The proxies we created for Securities Market Infrastructure were Volatility (defined as standard deviations of daily returns), Market Liquidity (defined as market turnover), Quoted Spread and a dummy variable for the existence of a closing call auction. For securities Market Participants, we surveyed the website of each securities exchange from our sample to check if that securities

165

exchange had allowed direct market access (DMA) before year 2005 and a dummy variable was created for that.

166

167

Bibliography

Aggarwal, R., Wu, G. (2006). "Stock Market Manipulation." Journal of Business 79(4): 1915-1953.

Aitken, M., Almeida, N., Harris, F.H., McInish, T.H. (2007). "Liquidity Supply in Electronic Markets." Journal of Financial Markets 10: 144-168.

Aitken, M., Berry, J.H (1991). Market Surveillance at the Australian Stock Exchange : An Overview. ASX Working Paper.

Aitken, M., Cook, R., Harris, F.H., McInish, T.H. (2008). "Market Design and Execution Cost for Matched Securities Worldwide." A Guide to Liquidity 2(Winter): 1-39.

Allen, F., Gale, D. (1992). "Stock-Price Manipulation." The Review of financial Studies 5(3): 503-529.

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

Allen, F., Titov, L., Mei, J.P. (2006). "Large Investors, Price Manipulation, and Limits to Arbitrage: An Anatomy of Market Corners." Review of Finance 10: 645- 693.

Bernhardt, D., Davies, R.J. (2005). "Painting the tape: Aggregate evidence." Economic Letters 89: 306-311.

Bhattacharya, U., Daouk, H. (2002). "The World Price of Insider Trading." The Journal of Finance 7(1): 75-108.

Bhattacharya, U., Spiegel, M. (1991). "Insiders, Outsiders and Market Breakdowns." The Review of Financial Studies 4(2): 255-282.

Bommel, V. (2003). "Rumors." Journal of Finance 58: 1499-1521.

Calhoun, C. (2002). Dictionary of the Social Sciences, Oxford University Press: 584.

168

Calrlton, D., Fischel, D. (1983). "The Regulation of Insider Trading." Stanford Law Review 35: 857-896.

Carhart, M. M., Kaniel, R., Kusto, D.K., Reed, A.V. (2002). "Leaning for the Tape: Evidence of Gaming Behavior in Equity Mutual Funds." Journal of Finance 57(2): 661-693.

Chamberlain, T. W., Chueng, C.S., Kuan, C.C.Y. (1989). "Expiration-Day Effects on Index Futures and Options: Some Canadian Evidence." Financial Analysts Journal 45: 67-71.

Chiou, J. S., Wu, P.S., Wang, A.W., Huang, B.Y. (2007). "The asymmetric information and price manipulation in stock market." Applied Economics 39: 883- 891.

Comerton-Forde, C., Putnins, T.J. (2009). Pricing accuracy, liquidity and trader behavior with closing price manipulation. 22nd Australasian Finance and Banking Conference 2009

Comerton-Forde, C., Rydge, J. (2006). "The Current State of Asia-Pacific Stock Exchanges: A Critical Review of Market Design." Pacific-Basin Finance Journal 14: 1-32.

Cornell, B., Shapiro, A. (1989). "The Mispricing of U.S. Treasury Bonds: A Case Study." The Review of Financial Studies 2(3): 297-311.

Corredor, C., Lechon, P., Santamaria, R. (2001). "Option-Expiration Effects in Small Markets: The Spanish Stock Exchange " The Journal of Futures Markets 21(10): 905- 928.

Cumming, D., Johan, S. (2008). "Global Market Surveillance." American Law and Economics Review Forthcoming.

Drudi, F., Massa, M. (2005). "Price Manipulation in Parallel Markets with Different Transparency." Journal of Business 78(5): 1625-1658.

Fama, E. F. (1970). "Efficient Capital Markets: A Review of Theory and Empirical Work." Journal of Finance 25(2): 383-417.

Fama, E. F. (1991). "Efficient Markets: II." Journal of Finance 46(December): 1575- 1617.

Fama, E. F. (1998). "Market Efficiency, Long-Term Returns, and Behavioural Finance." Journal of Financial Economics 49(September): 283-306.

Felixson, K., Pelli, A. (1999). "Day End Returns - Stock Price Manipulation." Journal of Multinational Financial Management 9: 95-127.

169

Foucault, T. (1999). "Order flow composition and trading costs in a dynamic limit order market." Journal of Financial Markets 2: 99-134.

Garbade, K. D., Silber, W.L. (1978). "Technology, Communication and The Perfromance of Financial Markets: 1840-1975." The Journal of Finance 33(3). Gilson, R. J., Krakman, R. (2004). The Mechanisms of Market Efficiency Twenty Years Later : The Hindsight Bias. SSRN Law Series.

Grossman, S. J., Stiglitz, J.E. (1980). "On the Impossibility of Informationally Efficient Markets." American Economic Review 70(June): 220-240.

Harris, F. H. (2008). "Execution costs and market design worldwide: A panel discussion." Journal of Trading 3(1): 9-25.

Harris, L. (1989). "A Day-End Transaction Price Anomaly." Journal of Financial Quantitaive Analysis 24(1): 29-45.

Harris, L., Sofianos G., Shapiro, J.E. (1994). "Program Trading and Intra-day Volatility." The Review of Financial Studies 7(4): 653-686.

Hart, O. D. (1977). "On the Profitability of Speculation." Quarterly Journal of Economics 91(4): 579-597.

Hart, O. D., Kreps, D.M. (1986). "Price Destablizing Speculation." The Journal of Political Economy 94(6): 927-952.

Hillion, P., Souminen, M. (2004). "The Manipulation of Closing Prices." The Journal of Financial Markets 7: 351-375.

Jarrow, R. A. (1992). "Market Manipulation, Bubbles, Corners, and Short Squeezes." Journal of Financial Quantitaive Analysis 27(3): 311-336.

Jensen, M. (1978). "Some Anomalous Evidence Regarding Market Efficiency." Journal of Financial Economics 6(2): 95-101.

Kendall, M. G. (1957). "The Analysis of Economic Time-Series, Part I. Prices." Journal of the Royal Statistical Society 96: 11-25.

Kumar, P., Seppi, D.J. (1992). "Futures Manipulation with "Cash Settlement"." The Journal of Finance 18(4): 1485-1502.

La Porta, R., Lopez-De-Silanes, F., Shleifer, A., Vishny, R.W. (1997). "Legal Determinants of External Finance." Journal of Finance 52(3): 1131-1151.

Lang, M. H., Lundholm, R.J. (1996). "Corporate Disclosure Policy and Analysis Behavior." The Accounting Review 71(4): 467-493.

Levine, R. (1997). "Financial Development and Economic Growth: Views and Agenda." Journal of Economic Literature XXXV: 688-727.

170

Malkiel, B. G. (1996). A Random Walk Down Wall Street, W. W. Norton.

Merrick, J. J., Naik, N.Y., Yadav, R.K. (2005). "Strategic Trading Behavior and Price Distortion in a Manipulated Market: Anatomy of a Squeeze." Journal of Financial Economics 77: 171-218.

Modigliani, F., Perotti, E. (1996). "Protection of Minority Interest and Development of Security Markets." Mimeo, MIT.

Oved, Y. (2002). "Information Disclosure Costs and the Choice of Financing Source." Journal of Finance 4(1): 3-21.

Pagano, M. S., Schwartz, R.A. (2003). "A Closing call's impact on market quality at Euronext Paris." Journal of Financial Economics 68: 439-484.

Pagano, M., Roell, A. (Journal of Finance). "Transparency and Liquidity: A comparison of Auction and Dealers Markets with Inofrmed Trading." Journal of Finance L1(2): 579-611.

Rajan, R. G., Zingales L. (1995). "The Tyranny of Inequality." NBER Working Papers 5396.

Rajan, R. G., Zingales L. (1996). "Financial Dependence and Growth." NBER Working Papers 5758.

Stoll, H. R., Whaley, R.E. (1987). "Program trading and expiration-day effects." Financial Analysts Journal 43(2): 16-28.

Stoll, H. R., Whaley, R.E. (1991). "Expiration Day Effects: What has changed?" Financial Analysts Journal: 58-72.

171

Appendix 1 Quotes from Various Lead Exchange Websites

• New York Stock Exchange – www.nyse.com “To help reassure investors and support customers, the Exchange further reduced trading costs and increased operating efficiencies, strengthened regulatory and governance standards, and introduced new ways for customers to access the market.” “Providing the highest possible market quality was our top priority, along with ensuring the liquidity and transparency that market participants have come to expect.”

• NASDAQ Market – www.nasdaq.com “NASDAQ is among the world’s most regulated stock markets, employing sophisticated surveillance systems…to protect investors and provide a fair and competitive trading environment.” “Offering growth and liquidity, fostering innovative technologies…NASDAQ continuous to build the most efficient trading environment…to the benefit of all market participants and investors.“

• London Stock Exchange- www.londonstockexchange.com “The FSA summarises its job as “To maintain efficient, orderly and clean financial markets and help retail investors achieve a fair deal…”

172

• Euronext- www.euronext.com A business corporation that supervises listings on the exchange, ensures efficient trading, provides a guarantee of final settlement of transactions, disseminates market data in real time, and promotes securities markets in general.” “Euronext aims to provide a fair and orderly market with built-in safeguards for the quality of price formation. Euronext is of the opinion that market participants should have a level playing field.”

• Toronto Stock Exchange- http://www.tse.com/en/aboutUs/tse/index.html “Toronto Stock Exchange provides an efficient, liquid market for senior equities”. “Market Regulation Services Inc. is a national, not-for-profit, self-regulatory organization. It seeks to foster investor confidence in the Canadian securities market and to safeguard investors by maintaining fair and orderly marketplaces. It is jointly owned by the TSX and the Investment Dealers Association of Canada.”

• Australian Stock Exchange- http://www.asx.com.au “…the growing market capitalisation of the market have combined to increase the depth and liquidity of the market - two of the most crucial elements, along with integrity of a successful market;” “The reputation of ASX's markets for fairness and integrity is very important to ASX. Maintaining this reputation involves constant and vigilant supervision.”

• Tokyo Stock Exchange- http://www.tse.or.jp “The management aims are stipulated in the Tokyo Stock Exchange's constitution as, in order to contribute towards the protection of the public interest and investors, the trading of securities must be carried out in a fair and efficient manner.”

173

• Hong Kong Stock Exchange- http://www.hkex.com.hk “HKEX is committed to performing its public duty to ensure orderly and fair markets and that risks are managed prudently, consistent with the public interest and in particular, the interests of the investing public.” “The powerful resources of its new integrated market structure will ensure that Hong Kong remains one of the most important centres for providing critical hedging and risk management facilities and for financing the development of China. At the same time, Hong Kong has the upward momentum to develop as a leading market with maximum liquidity and minimum transaction costs.”

• Taiwan Stock Exchange- http://www.tse.com.tw/docs/eng_home.htm “Mission Statement: To provide innovative, efficient and superior services.” “To maintain a fair open and safe trading market.”

174

Appendix 2 Annotated Bibliography

Price Destabilizing Speculation by Hart and Kreps 1986

Research Questions (1) Can speculation destabilize prices when speculators and consumers are both rational?

Research Motivations (1) Traditional theory believes speculation stabilizes prices because speculators buy when they prices are low and sell when the prices are high. But other researches show that speculators will buy when the chances of price appreciation are high, which may or may not be when prices are low.

(2) Those researches are not conclusive since they rely either on there being a small number of imperfectly competitive speculators or on non-speculators having irrational expectations. The purpose of this paper is to show that speculation can destabilize prices even when speculators are competitive and both speculators and non-speculators have rational expectations.

Data Not applicable since this is a theory paper.

Research Methodology

175

(1) The research is based on a simple two-participant model which only has speculators and non-speculators. (2) Markov Chain Analysis

Research Findings (1) In spite of the fact that non-speculators and speculators alike behave rationally and speculators are competitive, speculation can destabilize prices. Moreover, this is not because a lack of foresight by speculators. In fact, making speculators more foresighted may actually worsen the problem because the more the speculators know about the future, the less impact surprises in the future will have on next period’s prices. For example, when speculators know the probability of future prices rising to a level that can bring them profit is low, they will dump their holdings in the next period, which may further depress prices.

(2) But it doesn’t mean speculation cannot stabilize prices. Speculation can stabilize prices in a very weak form under two extreme conditions:

a. Consumption demand are is independently and identically distributed over time; and b. Speculators have no foresight about future demand at all

(3) Speculators will buy or sell according to their expectation of large-scale changes. They may withdraw supplies from the market when they believe supply-demand conditions will change, and when the danger recedes, they will dump those supplies back on the market. While such activity may well smooth the major transitions when they happen, it may also trigger more small-scale fluctuations for which a tangible cause will rarely be identifiable.

Future Research Directions This research paper sheds some light on the relationship between stock price volatility and speculation. When manipulations exist in the market, speculators may be attracted

176

to trade with speculative expectations built upon the false impression of the market created by manipulations. When the speculative expectation fails, speculators leave the market by introducing more volatility to the market and eventually increase the level of spread in the market. Future research can test this hypothesis.

177

Program Trading and Expiration-Day Effect by Stoll and Whaley 1987

Research Questions 1. Does the expiration of futures and options on index have significant impact on price, volatility and volume behaviour in the U.S. market? 2. Does the transaction cost eliminate the arbitrage opportunities from the expiration day effects? 3. Who are affected by expiration-day effects?

Research Motivations 1. The introduction of futures contract and options on index brought significant impact on price and volume on expiration days. No study has analysed this behaviour before Stoll and Whaley (1987)

2. The expiration-day effect suggests arbitrage opportunities. It would be interesting to know how profitable can those arbitrage opportunities be. Data 1. Open interest in S&P 500 Index Futures Contracts on the day before, the day of and the day after contract expiration between 1984 and 1985. 2. Volume of block trades classified type of expiration (i.e., non-expiration days, expiration days only for options, expiration days for both futures and options) 3. Price data of index during the last half hour on expiration day and first hour on the day after expiration between 30/12/1983 and 27/12/1985.

Research Methodology (1) Descriptive statistics: a. Relative volume mean for Thursday and Friday close and Monday open (2) Test of portfolios and stock prices reversals at close and at open (3) Test of price reversals from last half hour on expiration day to the first hour on the day after expiration

178

‐ Reversal Type 1: keep the sign of the first hour return on day after expiration if return from the last hour on expiration day is negative; otherwise take the opposite sign ‐ Reversal Type 2: If sign of the returns from the two periods are different, take the absolute value of returns from the first hour on the day after expiration; otherwise set that return to zero. This overstates the price effects because reversals due to information are fully reflected, but failure of reversals due to information is ignored. ‐ Reversal Type 3: Another version of reversal 2 by using the return from the last half hour on the expiration day

Research Findings 1. Open interests on the expiration day amounted 40% of the average month-end open interests. 2. Volume during the last half hour was significantly higher on quarterly expiration Fridays for index securities 3. Volatility and returns are significantly higher for the same period for S&P 500 stocks on Futures Expiration days. Option Expiration doesn’t have this phenomenon. 4. Price reversals are significantly higher for the same period 5. The average expiration day price effect 0.4% of the transaction value. 0.25% out of that are transaction costs. The 0.15% in excess is the additional cost for liquidity.

Future Research Directions Conduct the same analysis in different types of markets. For example, big markets vs. small markets, markets with different settlement procedures. This will shed some light on how to improve settlement procedures.

179

Expiration-Day Effects on Index Futures and Options: Some Canadian Evidence By Chamberlain, T.W., Chueng, C.S., Kuan, C.C.Y.

Research Question: Chamberlain, Cheung and Kuan (1989) examines whether the volume and price behaviour of Toronto Stock Exchange 300 is affected to the expiration-day effect. The study tested two hypotheses: (1) The volume from non-expiration days is the same to volume from expiration days (2) The price change from non-expiration days is the same to price change from expiration days

Research Motivation: The stock market crash on October 1987 inspired immense effort to study whether and what extent the price and volume behaviour of the underlying stock markets are affected by trading in stock index derivatives. Study from the U.S. (Stoll and Whaley) has shown that in the U.S. market, there is an expirary-day effect especially during so- called triple-witching hour - the period immediately preceding the simultaneous expiration of stock index futures, index options and options on index futures. During the triple-witching hour in the U.S. market, it has been found that trading volume and returns from the underlying stock market both are significantly higher than non- trading days. Before Chamberlain, Cheung and Kuan (1989), the empirical study on the expiration-day effects of index futures an doptions on the stock on the stock markets appear to be only focusing on the U.S. market despite the global prevalence of index-related products. This study expands the literature on expiration-day effects by studying the leading Canadian stock index, the Toronto Stock Exchange (TSE) 300.

Research Methods and Data: (a) Data: (1) Scope: November 1985 to May 1987 (2) Index Value of TSE 300 at 15:30 and 16:00 on Fridays (3) Index Value of TSE 300 at 10:00 on Mondays (4) Daily Volume of Index Constituents on Fridays

180

(b) Methodology: 1) Hypo 1: The volumes from expiration days are the same to volumes from non-expiration days - t-test and F-test are used to test whether the mean of volumes and standard deviations on expiration days are the same to those on non- expiration days.

(2) Hypo 2: The price changes from expiration days are the same to price changes from non-expiration days - t-test and F-test are used to test whether the mean of returns and standard deviations for TSE during the last half hour on expiration Fridays are the same to those on non-expiration Fridays.

- Mean Reversals and Rate of Return correlations for TSE 300 DURING the last half-hour of trading on Fridays and the first half- hour of trading on Mondays are also examined. Three types of mean reversals are calculated

- The following regression is tested to examine whether price change reversals associated with expiration and non-expiration Fridays are the same:

rt+1 = a0 + a1rt + a2D + et

where rt and rt+1 are returns from the last half-hour on Fridays and returns from the first half-hour on Mondays correspondingly. D is a dummy variable that equals 1 when the Friday is a expiration day and 0 otherwise.

(3) This study also examined the tendency of price behaviour from the underlying stock markets on expiration days by comparing the futures price/index price relative and the net cost of carrying on expiration Fridays.

Findings and Direction for Future Research (a) Findings

181

(1) The volume from expiration days is the not significantly different to the volume from non-expiration days

(2) The price changes from expiration days are significantly higher than price changes from non-expiration days. And the triple-witching hour from the U.S. market does affect such behaviour from the Canadian market.

(3) The mean reversals on expiration Fridays are much more higher than those from non-expiration Fridays. The mean reversals on expiration Fridays are negatively auto-correlated.

(4) Regression test on reversals shows that the dummy variable for expiration Fridays is a significant explanatory variable for price reversals. This is consistent with the interpretation of abnormal returns on expiration Fridays vis-a-vis non-expiration Fridays as being attributable to the unwinding of short arbitrage positions.

(5) By comparing the futures price/index price relative and the net cost of carrying on expiration Fridays, it's also found there is a strong tendency that futures prices are lower than the spot index value before expirations especially when trading costs are taken into considerations. Hence, for practical purposes, only short arbitrage opportunities were available at the daily close on Friday. This again is consistent with the tendency for prices to rise temporarily in the final half-hour of trading before futures contracts expire.

(b) Directions for Future Research Similar study can be carried on other markets especially small/illiquid markets to examine whether expiration day effects exist in those markets as well.

182

Insiders, Outsiders and Market Breakdowns By Bhattacharya and Spiegel (1991)

Research Question: (1) What conditions lead to a liquidity collapse in a financial market? (2) Is insider trading laws essential?

Research Motivation: (1) Before Bhattacharya and Spiegel (1991), no models included all of strategic behaviour by the informed trader, a Walrasian price formation mechanism (i.e., auctions in which traders submit demand functions) an endogenous motive for trade due to the random endowment of the insider and rational expectations by the uninformed traders. (2) Academic law-and-economics commentary has used the argument of equilibrium prices aggregate information effectively, and thus eliminate information asymmetries in the economy to disapprove the regulation of insider trading. But why so many countries adopt insider trading laws? Hence, there is need to develop a model to prove the effectiveness of insider trading laws.

Research Methodologies: This research is based on a model for two-date exchange economy.

Research Data: Not applicable as this is a theory paper. Findings: (1) When traders are too uninformed, they will not trade with the more informed. Hence, liquidity collapse doesn’t necessarily mean participants are fully informed. (2) Insiders are not always trading for information motivations. They often trade for hedging purposes. Otherwise, no counter parties will be willing to trade.

183

(3) If Insider Trading Laws do not exist, the market may fail completely as an aggregator of equilibrium price. (4) The risk premium on a security is completely characterized by the information and stock held by the market’s outsiders. (5) The insider’s utility is maximized when the linear demand schedule is employed. The market-maker, therefore, has another positive role to play: to choose and coordinate the equilibrium demand schedule.

The essence of the game of Asymmetric Information is pooling equilibrium vs. separating equilibrium.

184

Expiration Day Effects: What has Changed by Stoll and Whaley 1991

Research Questions 1. Has settlement at the open reduced the expiration-day price volatility of stocks that comprise the indexes on which futures and options are traded? 2. Do the futures and options on index that continue to settle at the close experience unusual price volatility around the contracts’ expiration

Research Motivations 3. In June 1987, New York Stock Exchange and New York Futures Exchange adopted the suggestion to settle their index futures and index options on the S&P500 Index, the NYSE Composite Index and NYFE Composite Index at the opening price. While other futures and options on index will continue to be settled at the Friday close. This provides an opportunity to study whether settlement procedure can reduce the expiration-day effects

4. No analysis had been carried out on volatility at the open before Stoll and Whaley 1991.

Data 1. A time-series record of stocks in each of the indexes 2. Transaction prices and volumes for individual stocks between January 1985 and December 1986 (before the settlement price change) 3. Transaction prices and volumes for individual stocks between January 1987 and June 1989. Especially at quarterly expiration Fridays and Monthly expiration days. Research Methodology 1. Descriptive statistics: a. Relative volume mean for Thursday and Friday close and Monday open 2. Test of portfolios and stock prices reversals at close and at open

185

Research Findings 1. At quarterly expirations, both trading activity and price volatility in S&P 500 and NYSE contracts were smaller than at the close after the adoption of settlement at open 2. At the open, trading volume and price reversals increased significantly between the pre-June 1987 period and the post-June 1987 period. The price effect observed at the open on quarterly expirations since June 1987 has been somewhat smaller than the price effect observed at the close in the period before June 1987. This may reflect the impact brought by the new settlement procedures or the fact that expiration-day trading is now split between open and close. 3. At monthly (non-quarterly) expirations, index stocks behave like non-index stocks on expiration days and like index stocks on non-expiration days. Trading activity and price reversals do not appear to have changed, in a statistical sense, since June 1987, nor are they large in an absolute sense. 4. Switching the settlement may only shift the position of expiration-day volatilities. 5. Settlement procedures matter to where and when manipulations may occur.

Future Research Directions Gaining evidence from different types of market design changes will help the creation of more efficient exchanges.

186

Futures Manipulation with “Cash Settlement” by Kumar and Seppi 1992

Research Questions 1. Can uninformed investors make profits by taking positions in the futures followed by manipulations of prices of the underlying securities?

Research Motivations 2. Classical Economics Theory states that in order to gain profits by taking positions in the futures followed by manipulations of prices of the underlying securities (equilibrium), the underlying’s price will be divergent from its true fundamental values. But such divergence can be captured by other participants and thereby offsetting any manipulation. But this theory doesn’t seem to be true. 3. Previous research has been conducted on futures with physical delivery. But no one has studies the impact of cash settlement.

Data Not applicable since this is a theory paper.

Research Methodology 1. This research is based on the extended Kyle Model.

Research Findings 1. Exogenous spot “noise” trading is unnecessary for the spot market to open, although futures “noise” trading is needed.

2. Profits from manipulation disappear in the limit as manipulations are added, but welfare and price liquidity effects persist. 3. Imperfect informational linkage of futures and cash markets leads to “price pressure” (i.e., subsequently reversed price changes) in the futures market without market maker risk aversion or inventory control problems.

187

4. Existence of wide variety of manipulation per se is shown to be robust to a strikingly wide range of distributional, timing and preference assumptions. 5. Manipulation succeeds because manipulators’ trades are confused with those of the informed trader (pooling equilibrium) Thus one way to view manipulation is as a mechanism transferring liquidity from futures to cash markets (which provides the theory for setting the existence of futures in stock as an independent factor in regression testing)

188

Market Manipulation, Bubbles, Corners, and Short Squeezes by Jarrow 1992 Research Questions: (1) Can market manipulations exist in a stochastic economy, either finite or infinite horizon, with time dependent price processes?

Data: (1) A unique data set of stock market manipulation cases by analysing SEC litigation releases from 1990 to 2001. There are 142 cases of stock market manipulation that the authors are able to identify. (2) Daily stock prices, trading volume, and capitalization from January 1989 to December 2001 from the online service FACTSET.

Methodology: (1) Descriptive statistics for the manipulated stocks. Sample mean, standard deviation, skewness, and kurtosis coefficients for daily returns and turnover are computed. (2) Regression Analysis is conducted to analyse the relationship between number of manipulation cases (prosecuted) and liquidity, volatility, returns. Only considering prosecuted cases perhaps will bring selection bias.

Findings: (1) Manipulations in the stock market can be successful under pooling equilibrium. Under pooling equilibrium, there are noise traders and large number of professional momentum traders which make manipulations profitable. (2) Stock prices rise throughout the manipulation period and then fall in the post- manipulation period. (3) Prices are higher when the manipulator sells than when the manipulator buys, suggesting that the unravelling problem does not apply in practice. After the manipulation ends, prices fall (because the existence of noise traders and momentum traders under pooling equilibrium)

189

(4) Liquidity is higher when the manipulator sells than when the manipulator buys. (5) At the time the manipulator sells, prices are higher when liquidity is greater. This result is consistent with returns to manipulation being higher when there are more information seekers in the market. (6) At the time the manipulator sells, prices are higher when volatility is greater. This result is consistent with returns to manipulation being higher when there is greater dispersion in the market’s estimate of the value of the stock. All these results are consistent with the model.

Future Research Direction: (1) Future research can test other implications from the model from Aggarwul and Wu (2006)

Notes: From Allen and Gorton’s (1992) the asymmetry of price elasticities can create an opportunity for profitable price manipulation. As a result, a manipulator can repeatedly buy stocks, causing a relatively large effect on prices, and then sell with relatively little effect.

190

Stock-Price Manipulation By Allen and Gale 1992

Research Questions:

(1) Can profitable, trade-based manipulation is theoretically possible in a model where all agents are rational (i.e., maximize their utility)?

Research Motivations: (1) This paper doesn’t rely on either of the price momentum theory or corner the market theory. In contrast, Allen and Gale develop a model with asymmetric information where all agents have rational expectations and maximize expected utility.

Research Data: Not applicable since this is a theory paper

Research Methodologies: This research paper is based a simple three-date model. In the model, there exist uninformed investors, uninformed manipulators and noise traders. Research Findings: (1) As long as the investors are sufficiently risk averse and attach a positive probability to the manipulator being an informed trader, there exists a pooling equilibrium at date 1 in which the manipulator achieves strictly positive profits.

(2) There must be an informed trader for the manipulator to imitate (which is a pooling equilibrium) in order for manipulation to be profitable. If it is common knowledge that the large trader in the market is an uninformed manipulator, there is no way the manipulator can make profits simply from trading with a given set of investors.

191

(3) The importance of trade-based manipulation schemes is, of course, an empirical question. However, casual observation suggests they might be important. Large traders frequently buy and then sell substantial blocks of stock, even though they are apparently not interested in taking over the firms. Some part of the profits from this trade may be the result of manipulation of the type discussed in this article.

Rational Invstors Momentum Trading Cornering the Market

•Buying by blcok trades in •Momentum tradres •Investors are forced to response to manipulators constantly leapfrogging trade at a higher activities the manipulator's attemp equilibrium price to acquire a position

192

Order Flow Composition and Trading Costs in a Dynamic Limit Order Market By Thierry Foucault (1999)

Research Question: (1) What is the optimal choice between market order and limit orders (i.e., order placement strategy)? (2) What is the optimal price for limit orders (order aggressiveness strategy)?

Research Motivation: (1) In order driven securities markets, market order supplies liquidity while limit order consumes liquidity. Though limit order may improve execution price, limit trader traders face a winner’s curse problem since they are more likely to be picked off when their orders become mispriced due to new information arrives. (2) The behaviour of the mix between market and limit order hasn’t been addressed in empirical literature before Foucault 1999. The objective of this article is to develop a simple model in which the mix between market and limit orders can characterized, in equilibrium.

Research Data Not applicable since this is a theory paper

Research Methodology: (1) This research paper is based on a model for sequential trading process in which traders can choose between market orders and limit orders. (2) There are two major risks traders are facing: a. Non-execution risk b. Picking-off risk

Research Findings (1) Volatility of the asset is a main determinant of the mix between market and limit orders. When the asset volatility increases, the probability of being picked off are larger. For this reason, limit order traders ask for a larger compensation for the

193

risk of being picked off in markets with high volatility. Thus, more traders will choose to hold limit orders. (2) When volatility is high, since more traders choose to use limit orders, the liquidity supply is reduced and hence, the execution probabilities for limit orders are reduced. (3) When volatility is high, posted spread will also increase (4) Based on 1, 2 and 3, it can be suggested that small firms should have a larger proportion of limit orders, lower fill rates and larger spreads than large firms, in limit order markets. (5) Hasbrouck (1991) shows that volatility is negatively related to equity capitalization. Thus, the proportion of limit orders for stocks with small capitalization must larger than for stocks with large capitalization. Limit order traders react to an increase in execution risk by posting larger spreads, because when execution risk is high, traders are under pressure to trade immediately upon arrival because the probability of being executed with a limit order is small. For this reason, traders are willing to place market orders at more unfavourable prices. This explains why spread tends to increase at the end of the trading period.

194

Option-Expiration Effects in Small Markets: The Spanish Stock Exchange By Corredor, C., Lechon, P. and Santamaria, R. (2001)

Research Questions: Corredor, Lechon and Santamaria (2001) examines the expiration-day effect in the Spanish market. This study aims to answer the following two questions: (1) Is the expiration effect on prices, volatility and trading volume in smaller markets greater than the major markets? (2) Is the existence of futures markets reduces the effect of the option expiration date on the underlying market?

Research Motivations: (1) Most previous studies on expiration effect have been focuses on major international markets. Study from the U.S. (Stoll and Whaley) has shown that in the U.S. market, there is an expiry-day effect especially during so-called triple- witching hour - the period immediately preceding the simultaneous expiration of stock index futures, index options and options on index futures. Chamberlain, Cheung and Kuan (1989) also provides some evidence on higher prices and volatility in the underlying market on expiration days in Canada. (2) However, there is scarce evidence for the expiration effect on small markets. Small markets are less liquid and deep, it would be interesting to test whether arbitrage can result in a greater impact on prices, volatility and trading volumes. (3) Small markets may be especially susceptible to market manipulation. Hence, arbitrage can produce a greater impact on prices, volatility and volume. On the other hand, transaction costs on options are normally higher in smaller markets. (4) Finally, the evidence from the expiration effect on small markets may complete the international evidence about expiration-day effect.

Research Data (1) Daily prices and volumes for the period between January 1992 and December 1995 from the Spain Stock Market.

195

Research Methodologies (1) Regression Test for price:

rt = a0 + a1rt-1 + a2D + et (2) The ARCH, GARCH, GJR and EGARCH models is used to test the above regression

(3) Robustness Test for price: Mean Reversal test

(4) Regression Test for the Conditional Variance of the underlying assets

Rt = a+b*Rt-1 +ut

(5) Regression Test for trading volume

Research Findings (1) No significant expiration day effects on the returns at expiration in the Spain Market. (2) No significant expiration day effects on volatility of the underlying markets (3) Significant higher volume on expiration days. The existence of futures contracts seems to prevent arbitrage opportunities. Because individual stocks in the Spain market don’t have futures contracts issued on them and the options for those stocks have to be settled by delivery. This forced arbitrageurs and investors to unwind their position by trading in the underlying market hence increase the volume of the underlying. The index options are settled by cash.

196

Learning for the Tape: Evidence of Gaming Behavior in Equity Mutual Funds By Carhart, Kaniel, Musto and Reed (2002) Research Question: (1) Is there a pattern of abnormal high equity returns at quarter-end? (2) Do fund managers inflate quarter-end equity prices?

Research Motivations: (1) In general, open-end domestic equity mutual funds calculate their net asset values per share NAVs from the closing transaction prices of their holdings (2) Quarter-end and especially year-end equity mutual fund prices are abnormally high. Finding out the reasons behind that is important to both academics and the stock exchanges and regulators. (3) This research paper explores the influence of last-minute trading on last-trade prices, which allows fund managers to move performance between periods with last-minute trading in stocks they already hold, a practice alternately known as “painting the tape” (Market regulators regard this practice as illegal. See Sugawara 2000).

Research Data: (1) Annual returns from the CRSP mutual fund database (2) SBBI yearbook for S&P total returns

Research Methodology: (1) Regression Analysis for Index Fund abnormal returns a. Tests using Lipper Mutual Fund Indices b. Tests using Daily Individual Mutual Fund Returns

(2) Daily Cross-sectional regression on Lagged Mutual Fund Returns

Findings: (1) Equity fund returns, net of the S&P500, are abnormally high on the last day of the quarter, especially the fourth, and abnormally low the next day.

197

Magnitudes range from around 50 basis points per year for large-cap funds to well over 200 basis points for small-cap funds. (2) There is little or no effect at month-ends that are not quarter-ends. (3) The link between the quarter-end rise and the next-day decline is confirmed by showing that larger increases precede larger decreases in the cross section, which is not the case for fund returns on other days. (4) The year’s best-performing funds have the largest abnormal year-end return reversals, and the quarter’s best-performing funds have the largest abnormal quarter-end return reversals. Intra-day data isolate much of the pattern in a small window of trading time around the quarter-end day’s close.

The stocks in the disclosed portfolios of the best-performing funds, controlling for capitalization and recent return, show significantly more price inflation at year-end than do other stocks.

198

The Manipulation of Closing Prices By Hillion and Souminen 2004

Research Questions: (1) Can closing-price manipulation (ramping) be explained by agency theory? (2) Can Closing Auction Call reduce manipulation (i.e., ramping)?

Research Motivations: (1) Hillion and Souminen (2004) is motivated by recent changes in the closing price mechanism in many stock exchanges around the world due to concerns about close price manipulation. (2) No existing models tried to explain close price manipulations from the perspective of broker agency theory

Research Data Not applicable since this is a theory paper

Research Methodologies This research paper is based on a model of equity markets, with an explicit role for a broker to reduce the price impact of his customer’s trades. The model characterizes an equilibrium in which the broker exerts an effort when executing his customer’s order, and manipulates the closing price in order to influence his customer’s perception of his performance.

Research Findings: (1) Broker manipulates the closing price of a stock in order to improve customer’s impression of his execution quality. (2) Closing price manipulations should precede large customer trades the next day.

(3) Broker’s effort to improve the execution quality and amount of manipulation is positively correlated to the size of customer’s current orders. But when the size of 199

the order is too big, customer can tell broker’s ability by using the transaction price. Therefore, the needs for manipulation decreases. (4) The bigger the uncertainty of broker’s ability, the higher amount of manipulation (5) The more volatile stocks are, the higher amount of manipulation. (6) Broker’s effort and amount of manipulation is positively correlated with the risk customer’s aversion level (7) Hence, manipulations of close prices are often seen on volatile and illiquid shares. (8) Brokers may need to continuously prove their abilities to customers due to changes in employment and market conditions. Hence, close-price manipulation exists in a long term. (9) Even in a competing brokers scenario, though brokers can manipulate closing price to different directions, manipulations will still be observable in equilibrium, because if no one manipulates, then any broker would have an incentive to do so. (10) Closing Call Auction can reduce the level of broker manipulations (11) For highly liquid market, the interval for trade interruption before auction should be smaller (12) Investor stability, such as capital flow, market openness, also affects the interval of trade interruption before Closing Call Auction (13) If other price benchmarks are used to evaluate broker’s ability, brokers might manipulate the prices near the time of the customer’s transaction.

200

Painting the Tape: Aggregate Evidence By Bernhadt and Davies 2005

Research Questions: (1) Are mutual funds manipulating share prices at the end of evaluation period to improve fund performance?

Research Motivations (1) Mutual funds have become big player in the market. If their manipulation behaviour is systematic to the market at the end of evaluation period, then Regulators and Exchanges will need to design corresponding rules and regulations to mitigate such behaviour (2) Individual investors need to be more cautious when trading at the end of evaluation period.

Data: (1) This paper uses the Centre for Research in Security Prices (CRSP) Equally and Value-Weighted Index Returns to calculate (i) the difference in returns on the last trading day of the period (quarter or month) (ii) and the return on the first trading day of the following period, (iii) the difference in returns on the last day of the period versus the average return on the other trading days.

(2) Mutual funds holdings of corporate equity (Reference FL653064000) and market value of domestic corporations were collected from the Federal Reserve US Flow of Funds accounts for the first quarter 1970 to third quarter 2001.

Research Methodology: (1) Descriptive Statistics (2) Regression Analysis for abnormal returns and holding weights of mutual funds

201

Findings: (1) The paper finds that the equally weighted index return on the last trading day of a quarter is significantly higher than both the return on the first trading day of a quarter and the average return on other trading days; (2) These return differences rise with the percentage of total equity that is held by mutual funds. This strong empirical evidence indicates that the incentives of fund managers to distort investments are so high at the end of a quarter that their behaviour significantly alters aggregate market outcomes.

202

Price Manipulation in Parallel Markets with Different Transparency By Drudi and Massa 2005

Research Questions (1) Does the existence of a less transparent market (e.g., the primary bond market) affect dealer’s trading behaviour in the more transparent market (e.g., the secondary bond market)?

(2) Can dealers manipulate bonds’ price by exploiting the information asymmetry between the more transparent market and the less transparent market?

Research Motivations (1) This is the first empirical analysis of investors simultaneously operating in two different markets characterized by strikingly different levels of transparency but trading the same asset (Italian T-bond).

(2) Moreover, given that the authors focus on T-bonds, this is also the first paper that studies bond trading behaviour on both the primary and the secondary market, with data disaggregated at the dealer level.

Data (1) This paper uses a unique dataset of the Italian bond market, which at a disaggregated level, records the behaviour of each single bond dealer in both the primary and the secondary markets.

(2) The secondary market data set contains all the transactions from September 29, 1994, to February 28, 1996, for all the listed bonds (a total of 37). In all, they total 1,393,437 transactions.

Research Methodology (1) Descriptive Statistics by type of brokers in the secondary market

203

‐ Volume ‐ trade counts, ‐ bids, winning bids ‐ mean profit (2) Regression analysis for the profit earned by each dealer on trading bonds and quantity bought/sold in the two parallel markets.

(3) Kyle-model for the relationship between market depth and price change on auction days

Research Findings (1) In the Italian T-bond market, when informed dealers have positive news which may affect the bond prices, they place sell orders in the secondary market before the auction on the auction day when they have a higher informational advantage. The purpose of this strategy is to affect other uninformed dealers’ expectations on the primary bond market.

(2) Simultaneously, those informed dealers aggressively place bids in the primary market and buy back when the primary market closes.

(3) This strategy generates losses in the more transparent market (secondary market) for the period during which the less transparent market (primary market) is open, because the selling time perhaps is just before the bond prices start rising. But since other uninformed dealers may be affected by the gloomy expectation of the bond prices which are created by the manipulations, the auction prices from the primary market could be lower than expected and the allotment of bonds to those informed dealers could be higher. When the auction closes and only the continuous market is open, bond price will increase rapidly and informed dealers can gain substantial profit.

(4) The existence of a less transparent market may actually increase the liquidity of the more transparent one.

Future Research Directions 204

Similar study may be difficult to carry out for equity markets. Can similar approaches be applied to securities that are cross-listed in different markets?

205

Strategic trading behavior and price distortion in a

manipulated market: anatomy of a squeeze By John J. Merrick Jr, Narayan Y. Naik, PradeepK. Yadav (2005)

Research Questions (1) This paper investigates and presents a specific example of market manipulation, detailing the price effects alongside trading positions of participants. Specifically they look at a bond delivery squeeze.

Findings of Note

• Detailed analysis of manipulation in bond and futures. • Identify two types of market participants: o Squeezers initiate and reinforce the squeeze o Contrarians identify the squeeze attempt and bet aggressively that the squeeze will fail • Showed a case of participant learning, whereby a participant learnt of the manipulation by intermediating trades for the manipulator and then heavily joined the manipulation attempt. This was prior to the squeeze becoming market knowledge. Once the manipulation became market knowledge many participants either joined the manipulation attempt or bet heavily against it. • Found that once known, the manipulation in the bond market reduced market depth. Implies that hedgers stayed away from the market due to mis-pricing and uncertainty, representing a real cost as hedgers are denied this facility. In other words manipulation also presents a cost to non-speculative investors who did not directly lose money in the manipulation scheme. • Found that once the manipulation was know, 89% of trading occurred between squeezers and contrarians.

Behaviour being examined

206

• Attempted delivery squeeze in the March 1998 long-term U.K. government bond futures contract traded on LIFFE. • A simple bond delivery squeeze involves: o Buying up the cheapest to deliver bond issue and taking a substantial long position in the bond future o By restricting supply this increases the price of the cheapest-to-deliver bond and also forces participants with shorter term futures contracts to deliver higher value bonds instead. • Cheapest to Deliver Issue o Bond futures often require the delivery of a basket of orders, so as to avoid manipulation. However not all bonds have the same value, so the exchange provides a conversion factor to make the different bonds equivalent. However this conversion factor assumes that spot yield is the same as the notional coupon in the contract and the slope of the yield curve is flat. This is not always true and one bond will be the “cheapest to deliver”, obviously this will be the bond that everyone will want to deliver.

In the case being examined the squeeze attempt failed due to intervention by the Bank of England and the manipulators received a small loss whilst the contrarians achieved a large gain. The biggest losers were those which had joined the manipulation attempt late.

207

Stock Market Manipulations

By Aggarwal and Wu 2006

Research Questions:

(1) Can market manipulations be successful under market equilibrium?

(2) What characteristics do stock market manipulations have?

Research Motivations:

(1) Aggarwal and Wu (2006) is the first study to test models of stock market

manipulation using a comprehensive sample of cases.

(2) Understanding the relationships between manipulations and market trading

attributes will be helpful on improving market surveillance and enforcement

Data:

(1) A unique data set of stock market manipulation cases by analysing SEC

litigation releases from 1990 to 2001. There are 142 cases of stock market

manipulation that the authors are able to identify.

(2) Daily stock prices, trading volume, and capitalization from January 1989 to

December 2001 from the online service FACTSET.

Methodology:

(1) Descriptive statistics for the manipulated stocks. Sample mean, standard

deviation, skewness, and kurtosis coefficients for daily returns and turnover

are computed.

(2) Regression Analysis is conducted to analyse the relationship between number

of manipulation cases (prosecuted) and liquidity, volatility, returns. Only

considering prosecuted cases perhaps will bring selection bias.

208

Findings:

(1) Manipulations in the stock market can be successful under pooling

equilibrium. Under pooling equilibrium, there are noise traders and large

number of professional momentum traders which make manipulations

profitable.

(2) Stock prices rise throughout the manipulation period and then fall in the post-

manipulation period.

(3) Prices are higher when the manipulator sells than when the manipulator buys,

suggesting that the unravelling problem does not apply in practice. After the

manipulation ends, prices fall (because the existence of noise traders and

momentum traders under pooling equilibrium)

(4) Liquidity is higher when the manipulator sells than when the manipulator

buys.

(5) At the time the manipulator sells, prices are higher when liquidity is greater.

This result is consistent with returns to manipulation being higher when there

are more information seekers in the market.

(6) At the time the manipulator sells, prices are higher when volatility is greater.

This result is consistent with returns to manipulation being higher when there

is greater dispersion in the market’s estimate of the value of the stock. All

these results are consistent with the model.

Future Research Direction:

(1) Future research can test other implications from the model from Aggarwul and

Wu (2006)

209

Notes:

From Allen and Gorton’s (1992) the asymmetry of price elasticities can create an opportunity for profitable price manipulation. As a result, a manipulator can repeatedly buy stocks, causing a relatively large effect on prices, and then sell with relatively little effect.

210

Large Investors, Price Manipulation, and Limits to Arbitrage: An Anatomy of Market Corners By Allen, Titov and Mei 2006

Research Questions:

(1) Can corner the market arise as a result of rational behavior? (2) What are the characteristics of cornered stocks and market around corner date?

Research Motivations: (1) There is growing literature on the theory of stock market manipulation, but limited empirical research on manipulation due to the difficulties of getting data (2) This paper fills gap of both theory and empirical literature on cornering the market.

Data: (1) Hand-collected data set of stock market and commodity corners which occurred between 1863 and 1980 in the US. (2) The authors double check all the corners reported by Wyckoff (1972) using reports by Brooks (1969), Clews (1888), Sobel (1965), Stedman (1905), and Thomas (1989) and eliminate those that cannot be verified independently and we restrict our cases to those that happened between 1863 and 1928, because trading data were not available before 1863. (3) Hand-collected data set of price and trading volume from the New York Times (4) Using Wall Street Journal to search for information that is missing due to the poor publication quality of historical newspapers.

Research Methodology: (1) Descriptive statistics for daily returns, volatility, autocorrelation, price dispersion, and trading volume and for the pre-corner period, as well as in two corner sub-periods: corner period one – ten days before the corner to the

211

corner date (included), and corner period two – the day after the corner date to ten days following it. - Significant Higher returns in corner period 1 - Returns significantly declining in period 2 which provides mean reversion evidence of manipulation. - Higher volatility in both periods which implies the market has a lot more different opinions in both periods. - Higher dispersion consistent around corner date with the manipulators owns private info. - Abnormal trading Volume increasing in period 1 and decrease in period 2. - Returns significant inversely correlated in period 1 (positive) and 2 (negative) - Market Returns increasing in period 1 and decreasing in period 2 due to the fact that short sale investors pressed for liquidity sell off their portfolio

(2) Regression analysis from Llorente et al. (200) for return-trading volume: Volume * Returns

Research Findings: (1) This analysis of the expected utilities of the uninformed, the arbitrageurs and the manipulator shows that corners can occur when everybody is behaving rationally (2) This paper finds strong evidence that large investors and corporate insiders possess market power that allows them to manipulate prices. Manipulation leading to a market corner tends to increase market volatility and has an adverse price impact on other assets. (3) The presence of large investors makes it risky for would-be short sellers to trade against the mispricing. Therefore, regulators and exchanges need to be concerned about ensuring that corners do not take place since they are accompanied by severe price distortions. (4) In the case where the corner fails because it is not possible to buy up the existing supply the volume traded with successful corners will be higher than the volume traded with unsuccessful corners. On the other hand, when the

212

failure is due to new supply becoming available the volume will be higher for unsuccessful corners. The model suggests that corners only occur when there is bad news.

213

The asymmetric information and price manipulation in stock market by Chiou, Wu, Chang and Huang 2007

Research Questions (1) Does the existence of informed foreign portfolio investments increase the volatility of a stock market? (2) Do informed foreign portfolio investments have motivations to manipulate the stock market?

Research Motivations (3) A lot of emerging markets have opened up to foreign investors with the rationale that informed foreign investors will help the microstructure of the stock market and tend to stabilize the financial market. But financial crisis occurred to those emerging markets have also brought up concerns regarding the excessive volatilities or bubbles introduced by those foreign investors. This paper will examine whether foreign investors could gain profit in those emerging markets via stock market manipulations and how emerging markets should react in the sense of market design policy marking.

(4) The interaction between asymmetrically informed parties has been mostly investigated in theoretical frameworks but there are relatively few empirical studies. And those limited empirical study mostly focus on cross-sectional analysis and use very short samples. The purpose of this study is to examine the extent of volatility of financial markets relative to their tranquil periods by using time series data for Taiwan Stock Exchange.

Data (1) Daily stock index for Taiwan from 1996 to 2005 are used in this study.

Research Methodology (3) Markov Regime Switching Model and Switching GARCH Model for returns on the index (4) GARCH test for returns on the index as robustness test 214

Research Findings (4) Examining the potential gain and loss, it is found that the Foreign Portfolio Investments’ potential gain is greater than that of individual investors in the state of high volatility in the Taiwan market for the period between 1996 and 2005.

(5) Informed Foreign Investors have motivations to manipulate the stock market when

a. There will be enough time for them to push up the price and get out completely before the private information is fully disclosed to the public b. The profit from manipulating the market prices will be higher than the cost of doing so

(6) In financial markets, extremely low transaction costs and high speed of adjustment or trading create great chances for the informed to manipulate the market and take advantage of the uninformed. This may create unnecessary high volatility and inefficiency in the stock markets.

(7) Based on (2) and (3), one way to increase the efficiency of those emerging markets after opening up to foreign investors is to i) increase transaction costs; or ii) to limit trading frequency; or iii) order size

Those measures seemingly make the market more inefficient, but as a matter of fact, they are similar to circuit breaker which slows down the speed of trading and gives more time for information flow and, thus, adding transparency to the markets and improving the efficiency of the market.

215

Future Research Directions

Once a liberalization policy is adopted in emerging markets, those markets must adopt other appropriate measures to prepare for the impact of implementing such a policy. For example, increase transaction costs, to limit trading frequency, etc. Future research can conduct survey on market design and policy making in emerging markets after they are open up to foreign investors and examine the analysis of new measures those emerging markets have taken in response to the impact foreign investors have brought in to the markets.

216

Market Design and Execution Cost for Matched Securities Worldwide by Aitken, M., Cook R., Harris, F.H. and McInish, T.H, in Institutional Investor (2009)

RESEARCH QUESTION: • Investigate the impact of market design changes on quoted effective spread, realised spread and price impact for 2,330 matched stocks, across 13 markets.

Background Theory: Market design studies generally utilise one of the three methodologies: 1. Compare execution costs in cross-listed stocks. 2. Event studies on market design changes. 3. Analyse spreads and price impact for matched pairs from different exchanges.

RESEARCH MOTIVATION: • Minimal agreement exists on what factors constitute the optimal market design. This study doesn't attempt to find the optimal market design; it does however work towards it. • Liquidity deciles are used: allows the investigation of market design features for thinly traded stocks (there is little prior research that analyse market design for thinly traded stocks). • Extend on prior research by analysing numerous market design changes (where prior studies frequently collapsed market design differences, so that individual changes were unobservable. To achieve this and to over come 'selection bias' that was witnessed in prior research (Jain 2002 & Swan and Westerholm 2004) [selection bias - firm specific characteristics may have confounded the design changes]. This study matches on market cap, daily traded value, share price and industry. Unbias is assured after regressing a matrix of design features and control variables on execution costs. • Matched pair research is dominated by US studies. • Conflicting evidence has arisen in full transparency research.

217

Research Methodology: • Hypotheses: o The presence of a crowd will lower spreads because of real-time assessment of the state of the market. [Only NYSE has this design feature. Trading floors exist for both NYSE and Xetra]. o Highlight the common role of providers of facilitation services designed to increase liquidity (market makers). [Markets LSE, NASDAQ, Paris, NYSE, Xetra]. o Any pooling of liquidity to reduce execution costs [All markets except NASDAQ] o Presence of circuit breaker will reduce execution costs as market participants will have a greater confidence of market stability [only NYSE].

Research Data: • Data for every stock from the top 7 markets, (that data could be obtained for) was gathered. Markets LSE, NYSE, Paris, Tokyo, Frankfurt and ASX. They were then matched against stocks on NASDAQ. The period analysed October 2002-September 2003. • Intraday trade and quote data for each stock was gathered from SIRCA (Reuters data). [The best bid and ask]. • Industry classification, market price and market cap from Thomson's DataStream Advance. All prices are converted to USD. • For firm's to be included they must trade at least 62 days in the year. Only trades in normal trading hours are included. Treatment & Control Groups:

NASDAQ is used as the matched group. NASDAQ is superior to NYSE as the average price of each decile is considerably higher in NYSE than for other exchanges, making matching challenging.

218

RESEARCH FINDINGS: • Execution Costs – o ASX, NYSE and TSE the average quoted and effective spreads are less than NASDAQ. o Average quoted and effective spread is greater for LSE than NASDAQ. o Paris is indistinguishable. o NYSE and TSE, average quoted spread is less than NASDAQ. o ASX, NYSE, TSE effective spread is less than NASDAQ. o LSE quoted spread is greater than NASDAQ. o TSE and PAR effective spread is greater than NASDAQ.

• Reduction in execution costs in thickly and thinly traded stocks in markets with no trading floors and no market makers (NYSE) and avoidance of hybrid execution systems (LSE). • Individual depth steps present (full transparency) is neutral to execution costs; however price impact falls by 15%. • Full order book transparency to brokers increases especially for thickly traded stocks; or they advise clients to trade away from the best bid offer.

219