THE ORDER BOOK, ORDER FLOW, AND THE IMPACT OF ORDER CANCELLATIONS ON EQUITY INDEX FUTURES

Transcription

1 THE ORDER BOOK, ORDER FLOW, AND THE IMPACT OF ORDER CANCELLATIONS ON EQUITY INDEX FUTURES A dissertation submitted to the Kent State University Graduate School of Management in partial fulfillment of the requirements for the degree of Doctor of Philosophy by Sara E. Bennett August, 2012

2 Dissertation written by Sara E. Bennett B.B.A., Augusta State University, 2003 M.B.A., Augusta State University, 2005 Ph.D., Kent State University, 2012 Approved by Chair, Doctoral Dissertation Committee Member, Doctoral Dissertation Committee Member, Doctoral Dissertation Committee Accepted by Doctoral Director, Graduate School of Management Dean, Graduate School of Management ii

3 Acknowledgments There are so many people who have offered me support and advice as I have worked on this dissertation. I am incredibly grateful for their support and efforts. First, I want to thank my committee members, Drs. Mark Holder, Jay Muthuswamy, and Eric Johnson. Their comments, suggestions, and support made this dissertation possible. Particular thanks must be giving to my chair, Dr. Mark Holder. Without his guidance and support, including access to an amazing dataset, this dissertation would not be here today. I also want to thank Shengxiong Wu for his efforts in converting the raw data into a usable format and also for just being there to bounce ideas around. Additionally, I owe a tremendous debt of gratitude to Dr. Mike Tomas. Casual conversations with him helped me solve two of the biggest obstacles I encountered in my analysis. Above all else, I want to thank my parents, Helen and Chris. Their unwavering support has been the rock that I could cling to in the most difficult moments in this journey. They had faith in me even when I had lost faith in myself and their love and support means more than I can ever express. I would also like to thank my grandparents and the rest of my family for the love, prayers, and support. Without this wonderful support system I would not be where I am today. Special thanks go to Bob Antenucci and Randy Kimmel for helping to keep me sane during my graduate student years at Kent. iii

4 Table of Contents Acknowledgments... iii List of Tables... vi List of Figures... vii Introduction... 1 Chapter 1: Survey of Literature... 3 Order Flow and Price Discovery... 4 Information Content of Limit Order Book... 7 Futures Markets Versus Cash Markets Order Cancellations and Spoofing Impact of Macroeconomic News Announcements The Contribution of this Dissertation to the Literature Chapter 2: Methodology The Existence of Spoofing Orders Identifying Possible Spoofing Orders Intraday Patterns in Spoofing Orders Impact of Macroeconomic News Announcements Chapter 3: Data Description Equity Index Futures Trading at Eurex Eurex Historical Order Book Data Description Chapter 4: Empirical Results Is there evidence of potential spoofing in front-month FDAX and FESX contracts?.. 45 Spoofing at a Five-Deep Imbalance of the 95 th Percentile for the FDAX contract Ask Side Detailed Analysis Bid Side Detailed Analysis Summary iv

5 Spoofing at a Five-Deep Imbalance of the 95 th Percentile for the FESX contract Ask Side Detailed Analysis Bid Side Detailed Analysis Summary Spoofing at Ten-Deep 95th Percentile Imbalances for FDAX and FESX contracts FDAX FESX Impact of Scheduled Macroeconomic News Announcements on Spoofing Intraday patterns Chapter 5: Conclusions Works Cited Appendix A Appendix B v

6 List of Tables Table 1: Contract Specifications Table 2: Key Variables Table 3: Daily Trading Statistics Table 4: Summary Statistics on FDAX Table 5: Summary Statistics on FESX Table 6: Statistics on FDAX Depth Table 7: Statistics on FESX Depth Table 8: Number of Imbalance5 Intervals by Product and Length Table 9: Number of Imbalance10 Intervals by Product and Length Table 10: Imbalance5 FDAX Time Between Large Imbalances Table 11: Statistics on Imbalance5 FDAX Intervals of One Second Duration Table 12: Statistics on Ask Imbalance5 FDAX Intervals Table 13: Statistics on Bid Imbalance5 FDAX Intervals Table 14: FDAX Imbalance5 Time Between Large Imbalances Table 15: Statistics on Ask Imbalance5 by Price Change Direction Table 16: Kruskal-Wallis ANOVAs for Ask Imbalance5 FDAX Table 17: Mann-Whitney U Tests for Ask Imbalance5 FDAX Table 18: Statistics on Bid Imbalance5 by Price Change Direction Table 19: Kruskal-Wallis ANOVAs for Bid Imbalance5 FDAX Table 20: Mann-Whitney U Tests for Bid Imbalance5 FDAX Table 21: Imbalance5 FESX Time Between Large Imbalances Table 22: Statistics on Imbalance5 FESX Intervals of One Second Table 23: Statistics on Ask Imblance5 FESX Intervals Table 24: Statistics on Bid Imbalance5 FESX Intervals Table 25: Imbalance5 Time Between Large Imbalances Table 26: Statistics on Ask Imbalance5 by Price Change Direction Table 27: Kruskal-Wallis ANOVAs for Ask Imbalance5 FESX Table 28: Mann-Whitney U Tests for Ask Imbalance5 FESX Table 29: Statistics on Bid Imbalance5 by Price Change Direction Table 30: Kruskal-Wallis ANOVAs for Bid Imbalance5 FESX Table 31: Mann-Whitney U Tests for Bid Imbalance5 FESX Table 32: Statistics on Ask Imbalance10 FDAX Intervals Table 33: Statistics on Bid Imbalance10 FDAX Intervals Table 34: Statistics on Ask Imbalance10 FESX Intervals Table 35: Statistics on Bid Imbalance10 FESX Intervals Table 36: Percentages of Large Imbalances Occurring Around Economic Reports Table 37: FDAX Ask5 Economic Announcements - Imbalance Statistics Table 38: FDAX Bid5 Economic Announcements - Imbalance Statistics Table 39: FESX Ask5 Economic Announcements - Imbalance Statistics Table 40: FESX Bid5 Economic Announcement - Imbalance Statistics vi

7 List of Figures Figure 1: Order Book Reconstruction at a Specific Time Point Figure 2: FDAX Daily Settlement Prices Figure 3: FESX Daily Settlement Prices Figure 4: FDAX Ask Imbalance5 Number of Large Intervals by Hour Figure 5: FDAX Ask Imbalance5 Percentage of Intervals by Hour and Price Change Figure 6: FESX Ask Imbalance5 Number of Large Intervals by Hour Figure 7: FESX Ask Imbalance5 Percentage of Intervals by Hour and Price Change Figure 8: FDAX Bid Imbalance5 Number of Large Intervals by Hour Figure 9: FDAX Bid Imbalance5 Percentage of Intervals by Hour and Price Change Figure 10: FESX Bid Imbalance5 Number of Large Intervals by Hour Figure 11: FESX Bid Imbalance5 Percentage of Intervals by Hour and Price Change Figure 12: FDAX Daily Trading Information Plots Figure 13: FESX Daily Trading Information Plots Figure 14: FDAX Ask Imbalance5 Boxplots with Standard Errors Figure 15: FDAX Bid Imbalance5 Boxplots with Standard Errors Figure 16: FESX Ask Imbalance5 Boxplots with Standard Errors Figure 17: FESX Bid Imbalance5 Boxplots with Standard Errors vii

8 Introduction Ever since financial markets have been open, there have been market participants looking to beat the market through with a variety of trading techniques. One possible technique to beat the market is spoofing, a microstructure based trading strategy. Section 747, Antidisruptive Practices Authority, of the Dodd-Frank Wall Street Reform and Consumer Protection Act amends section 4c(a)(5) of the Commodity Exchange Act as follows: It shall be unlawful for any person to engage in any trading, practice, or conduct on or subject to the rules of a registered entity that (c) is, is of the character of, or is commonly known to the trade as spoofing (bidding or offering with the intent to cancel before execution). In essence, a spoof order is an order that is submitted to the book by a trader who has no intention of filling the order. The trader s main objective is to mislead other traders regarding the supply and demand of a financial instrument, ultimately creating an illusion of either excess supply or excess demand. The trader will place a large order on one side of the book, say the ask, to create the illusion of a large order imbalance of excess supply. The trader is hoping that this illusion of imbalance will entice other traders to sell ahead of him and thereby push the price down. When this occurs, the trader will cancel the spoofing sell order and actually buy contracts at a lower price. His intention was always to buy contracts at a price lower than the prevailing market price at the time the spoof order was submitted. However, this trading strategy is not without risk as it does provide liquidity to the market and a subsequent submission of a large market buy order could consume the spoofer s large limit sell order before the price decreases and/or before the spoofer has a chance to cancel the sell spoof order. If this occurs, the 1

9 2 spoofer is obligated to sell the contracts even though this was never his intention or desire. In their study of activity on the Island ECN, Hasbrouck and Saar (2002) discover an interesting trend regarding limit orders. They find that approximately 27% of limit order submissions are fleeting. Fleeting is defined by the authors as orders that are cancelled within two seconds of submission. One possible explanation for some of these observations is the submission of spoof orders. Only two studies of spoofing, Eom, Lee, and Park (2009) and Kong and Wang (2011), were identified during the literature review. Eom, Lee, and Park (2009) examine spoofing in Korean equities in a unique institutional setting while Kong and Wang (2011) study specific spoofing trades by trader Jianming Zhou who manipulated fifteen stocks on the Shanghai Stock Exchange. The goal of this dissertation is to contribute to the body of knowledge by providing insight into order placement strategies for equity index futures contracts as well as identifying the potential use and frequency of a trading strategy that could have a significant impact on trading and that is now prohibited under Section 747 of the Dodd-Frank Act. This dissertation will use limit order book data for two equity index futures contracts (DAX futures and DJ Euro STOXX 50 futures) to determine if spoofing might occur in these markets over the period from January 2, 2007 through April 30, 2007.

10 Chapter 1: Survey of Literature Over the past several decades, there has been an increased interest in understanding market microstructure. Hasbrouck (2007) defines market microstructure simply as the study of trading mechanisms used for financial securities. Garman (1976), in his paper titled Market Microstructure, defines microstructure as moment-tomoment aggregate exchange behavior. In his survey of microstructure, Madhavan (2000) writes market microstructure is concerned with how various frictions and departures from symmetric information affect the trading process. Essentially microstructure is concerned with the dynamic process by which prices come to reflect new information. Microstructure theory recognizes that asset prices do not necessarily reflect the value from full-information expectations because of possible sources of friction such as market regulations and trading mechanisms. For example, order flow has been shown to play a critical role in price discovery. The impact of order flow on prices has been modeled and also documented in many studies of equity markets. Two key models, inventory models and information-based models, have different predications concerning how and why order flow impacts prices of equity securities. The Literature review will detail: order flow and price discovery models; information content of the limit order book; how futures markets differ from cash markets; order cancellations and spoofing; and the contribution of this dissertation to the Literature. 3

11 4 Order Flow and Price Discovery In the inventory model, Garman (1976) models the relationship between dealer quotes and inventory levels. In this model, a dealer is needed because buyers and sellers do not necessarily arrive at the market at the same time. When the intensity of buying and selling is not relatively equal, a dealer has to maintain a safety margin of both cash and the security in question. Garman s model posits that the dealer has to adjust prices relative to his inventory, and thus change the price he quotes (see also Stoll (1978); Amihud and Mendelson (1980); Ho and Stoll (1981); Ho and Stoll (1983)). For instance, when order flow is such that dealers are buying, their inventories will, as a consequence, increase and cause a move away from an optimal or ideal inventory position which would lead to losses for dealers due to financing the cost of carrying excess inventory. This increase creates an incentive to begin selling back some of this inventory. To accomplish this goal, dealers will lower their quote to entice other market participants to buy and bring the inventories of dealers back to optimal levels. Therefore, implicit in these models is the existence of a bid-ask spread to compensate dealers 1. Inventory models tell us that dealers will make a temporary adjustment to their prices in response to incoming orders. In contrast to inventory models, information based models predict that some traders are informed and therefore have private information regarding the future value of a security. A trade by an informed investor will have the effect of a permanent price adjustment for the security as new information (information that was previously private) becomes incorporated through trading as the price of the security converges with the 1 A dealer earns a profit on the spread by quoting an ask price to buyers and a bid quote to sellers where the ask price is higher than the bid price. The difference between the ask and bid prices is the spread.

12 5 intrinsic value. Glosten and Milgrom (1985) develop a model of uninformed (liquidity) and informed traders. In their model, quotes of the provider of liquidity will be conditional on the direction of the trade since these traders are uninformed. For instance, if a trader submits an order to buy, an uninformed dealer will revise his expectation of future value upward knowing that the order might be made by an informed trader. Kyle (1985) develops a model where there is a single informed trader who behaves strategically and only market orders are allowed. This trader will set his trade size while considering the adverse price associated with large quantity traders. Kyle s model allows for the trader to re-enter the market, thereby allowing the informed trader to spread out his trades over time if desired. This model shows that market prices eventually incorporate all available information, albeit gradually. While Stoll (1978) and Amihud and Mendelson (1980) model the bid/ask spread based on the premise of dealers adjusting for inventory cost and Glosten and Milgrom (1985) and Kyle (1985) model the bid/ask spread on the idea of adverse selection (based on the presence of informed traders), an important question is whether or not there are information effects in pricing or if changes in the bid/ask spread are transitory as with the inventory cost models. Several empirical studies of dealer inventory models analyze NYSE specialist inventories. Ho and Macris (1984) examine dealer trading book information for AMEX options. They find that the specialist s quotes are impacted by the inventory position of the specialist and that the dealer s bid-ask spread serves a role as an explanatory variable in returns. Specifically, the dealer will lower (raise) both the buy and sell quotes when he has a positive (negative) inventory accumulation. Madhavan and Smidt (1991) and Madhavan and Smidt (1993) develop and test a model of intraday

13 6 security price movement that incorporates both inventory and information, using transaction data from a NYSE specialist. They find that the inventory effect is weak with inventories of the specialist showing slow mean reversion. Glosten and Harris (1988) present a technique for modeling the bid/ask spread, with two components. One component arises from asymmetric information and the other is due to inventory costs, specialist monopoly power, and clearing costs. Results indicate that a significant amount of the spread is due to the presence of private information rather than inventory effects. Holden and Subrahmanyam (1992) extend this model by including multiple risk-averse insiders who have private information. Through aggressive competition, their private information is revealed almost immediately. Studies have found that the impact of inventory control on prices is temporary while the impact of information effects on prices is permanent. For instance, Hasbrouck (1988), in an empirical analysis on the relationship between trades and quote revisions for stock listed on the NYSE, finds evidence that the information content of trades is significant. However, there seems to be little evidence of inventory control. Madhavan and Sofianos (1998) find that dealers control their inventory positions by selectively timing the size and direction of trade, as opposed to adjusting their quotes. They also find that specialists tend to participate more in smaller trades and when the bid-ask spread is wide. These empirical results indicate that there is an information component in market orders. However, many markets heavily rely on the submission of limit orders rather than market orders. Therefore, it is appropriate to ascertain whether or not there is an information component in limit order books.

14 7 Information Content of Limit Order Book Models by Glosten (1994), Rock (1996), and Seppi (1997) include informed traders who are assumed to prefer market orders, suggesting that there may be little or no information in the limit order book in the absence of designated market makers. It is assumed that informed traders would prefer to place a market order to avoid the uncertainty over whether or not a limit order will be filled. To avoid this uncertainty (non-execution risk) an informed trader will submit a market order and be guaranteed to trade on his private information. While many models assume that informed traders prefer market orders, many studies suggest that there is additional information contained in the order book. Several studies attempt to understand the impact of limit orders on price formation and liquidity. Many exchanges, at least partially, rely on limit orders to provide liquidity to the market 2. Seppi (1997) develops a model of liquidity provision and concludes that a hybrid specialist/limit order market, such as NYSE, provides better liquidity to small retail and institutional traders. This is contrasted with the conclusion that a pure limit order market, such as the Paris Bourse, provides better liquidity on midsize offers. In their study of NYSE specialist trading decisions, Harris and Panchapagesan (2005) study the dynamic between specialist who are competing with limit order traders to provide liquidity. Since specialist see all limit orders and limit order traders only know their individual trades, the author investigate whether or not specialist have a unique advantage. They conclude that the limit order book actually provides 2 Ahn, Bae, and Chan (2001) find evidence of limit order traders entering the market and placing orders when liquidity is needed. Hamao and Hasbrouck (1952) find strong liquidity on the Tokyo Stock Exchange where all liquidity is supplied by limit order traders as there are no designated dealers or market makers.

15 8 information about future price movements and that specialists use this information in ways that are advantageous to them. This result is more pronounced in more actively traded securities where there is typically increased competition between liquidity traders and specialists. Niederhoffer and Osborne (1966) examine ticker tape transaction records for Dow Jones Industrial Average stocks over a twenty-two day trading period. They find strong evidence of price clustering in stock prices which is inconsistent with efficient markets and could indicate that NYSE specialists possibly take advantage of their unique information while trading. Niederhoffer and Osborne find that there are generally price reversals between trades and that the reversals concentrate near barriers where slower traders offer to buy and sell. As limit order systems have become more advanced, these trends do continue. For instance, Hasbrouck and Ho (1987) find evidence of autocorrelation in both intra-day stock returns as well is in the arrival of buy and sell orders (order flow) in stocks traded on the NYSE. Harris (1991) finds evidence of price clustering in stocks traded on the NYSE. These studies raise questions as to market efficiency and the impact of the specialist monopoly power to the bid/ask spread, at least in a pre-decimal pricing environment and before more sophisticated, electronic trading platforms. A question of interest is whether the order book provides market participants with additional information that can be used by traders for order submission strategies. Biais, Hillion, and Spatt (1995) examine the limit order book and order flow on the Paris Bourse. Rather than focusing on the profitability of a trade, the authors are interested in the interaction between the order book and order flow. They are interested in certain

16 9 aspects of order placement, such as liquidity, priority, and information effects. They find that the conditional probability of an investor placing a limit order (market order) is larger when the bid-ask spread is large (small) or the order book is thin. This finding is consistent with the notion that investors provide liquidity when it is needed by the market and consume liquidity when it is plentiful in the marketplace. This is due to the tradeoff between non-execution risk and price risk. For instance, if depth is already large, a new limit order at a given price would have low time priority and is less likely to be executed. In this case, a market order is preferred because it ensures execution even if at a less favorable price. They also find that, although a great deal of order flow occurs at the best market quote, the depth of the order book is larger away from the quote. The findings of their study also indicate that there are information effects in the order process. For instance, after a large sale that consumes liquidity at the quote and decreases the bid, there is often a new sell order placed in the quotes that will decrease the best ask. This is indicative of the market adjusting to the information content found in that trade. They also find that large purchases or sales often occur in succession and quite rapidly. Foucault (1999) develops a model of price formation and order placement strategies in a limit order market. His model concludes that limit orders result in better execution prices but face additional risks in terms of non-execution risk and a winner s curse quandary. Kaniel and Liu (2006) find that informed traders prefer to submit limit orders and that, in equilibrium, limit orders could convey more information than market orders. Additionally, the greater the expected time horizon of the private information, the more likely it is that a trader will submit a limit order rather than a market order. Earlier theoretical models often assumed that an informed trader would prefer market orders over

17 10 limit orders because of the non-execution risk associated with limit orders (i.e. a limit order might not be filled/executed while a market order is guaranteed to be filled). While limit orders have execution risk, market orders are exposed to price risk (i.e. if the market order size is greater than the prevailing depth at the quote, the trader will have price risk due to the fact that the order is filled at subsequently worse prices). Private information that would have a longer time horizon would allow a trader to limit execution risk while placing limit orders, as it is likely that other participants will trade on this information and the order will be filled, while eliminating price risk that could arise from placing a market order. The authors findings suggest that limit orders are more informative than market orders. A significant amount of research has suggested that traders use information contained in the top of the order book to determine whether to submit a market order or a limit order. These studies generally indicate that a large bid-ask spread discourages the use of market orders while a large number of limit orders at the top of the book encourages market order submissions. However, relatively few studies look beyond the top of the order book to analyze order placement strategies. Over the past decade, markets have opened the limit order book to provide greater transparency to market participants. Cao, Hansch, and Wang (2008) determine that, while the top of the order book always affects order submissions, amendments, and cancellations, the remainder of the books also affects both order amendments and cancellations. Cao, Hansch, and Wang (2009), using data from the Australian Stock Exchange, examine the information content of the limit-order book with an interest in the marginal information that might be

18 11 contained in limit orders away from the best bid and ask. They find that 78 percent of the information comes from the best bid and ask prices as well as the last transaction and that the limit order book contributes 22 percent to price discovery. Futures Markets Versus Cash Markets While there are several studies examining order flow in equity markets, there are significantly fewer studies analyzing order books in futures markets. Subrahmanyam (1991) notes that there are two motives for trading in equity index futures. One motive is that informed traders will trade on the basis of their private information. Another motive is that liquidity traders will trade for reasons not directly associated with their payoff such as the desire for immediate consumption or to manage some risk exposure. He demonstrates that equity index futures are actually preferred over the cash market by uninformed liquidity traders who seek to trade portfolios. Another area where futures and equity markets could differ is in the relative time it takes for information to be processed and reflected. Stoll and Whaley (1990) analyze the intraday returns of stock index and stock index futures contracts and find that index futures returns lead stock market returns by approximately five minutes, and in some cases by as long as ten minutes. The authors note that this result could reflect the greater speed with which investors views or information regarding the general direction of the market are reflected in futures markets. For instance, a speculator may prefer to trade on his view using stock index futures due to lower transaction costs compared with trading on this view using individual stocks as well the higher liquidity in futures markets compared with individual stocks. Similar results are found by Herbst, McCormack, and West (1987), Kawaller, Koch, and Koch (1987), and Chan (1992) in studies of the

19 12 S&P500 index and index futures. A common factor in these earlier studies is that all focus on U.S. equities and equity index futures. Booth, So, and Tse (1999) investigate whether German equity index derivatives lead the German stock index. Specifically, they use the spot index, DAX futures, and DAX options and find that DAX futures as well as the index react to new information faster than DAX options. In fact, DAX futures contracts provide the greatest contribution to price discovery. Nieh, Chang, Wong, and Shi-jie (2008) analyze liquidity provisions for TAIEX (TX) stock index futures and conclude that there are differences from predictions of equity markets. They find that the buy-side volume for institutional investors is higher than sell-side volume. This result is surprising if one expects institutional investors to be using stock index futures to hedge their long positions in the spot market. The authors also find that the percentage of market orders for both individual and institutional investors is significantly higher in the last interval of the trading day which is inconsistent with the findings of Bloomfield, O Hara and Saar (2005), who find that informed traders actually play a dual and sometimes complex role in the market. Using experimental asset markets, they find that informed traders will typically prefer market orders initially (and earlier in the day) to trade on their private information before prices adjust. Therefore, they will typically take liquidity earlier in the trading day. Once private information begins to be incorporated in the share price, informed traders will shift to limit orders and provide liquidity later in the day.

20 13 Order Cancellations and Spoofing While a significant number of order book studies examine order submission strategies in equity markets, while very few take order cancellations into consideration. Cao, Hansch, and Wang (2008) study order books from the Australian Stock Exchange to determine whether the order book affects order placement strategies. The authors note that there is very little research examining order placement strategies beyond initial order submission. Specifically, they are interested in all order placement strategies, including order amendments and cancellations. They find that the best bid and ask always affects order submissions, amendments, and cancellations. The rest of the order book affects order amendments and cancellations. Hasbrouck and Saar (2002) study the relationship between volatility and other activity on the Island ECN, which is organized as a limit order book. They find that approximately 27.7% of all limit order submission are fleeting, meaning they are cancelled within two seconds. They suggest that these fleeting orders could be a product of several different situations. One suggestion is the use of automated order routing systems trying to realize execution at different market centers. Another possibility is that the trader submitting the orders is attempting to find possible hidden orders. The authors also comment on the fact that some of these fleeting orders could be arising from spoofing. It is also possible that the number of fleeting orders is a consequence of greater High-Frequency Trading in equity markets. For instance, Hasbrouck and Saar (2007)find that 36.7% of limit orders for a sample of 100 Nasdaq-listed stocks on INET are cancelled within two seconds (i.e. fleeting orders). The high frequency of these orders questions the idea that patient traders will submit limit orders while impatient traders

21 14 generally prefer market orders. The authors explore three possible trading hypotheses to explain the fleeting orders as. The authors find strong support for the chasing hypothesis where orders are cancelled when market prices move away from the limit order price. The idea here is that the trader will chase the order by cancelling the existing order and resubmitting closer to the current price. There is also weaker evidence to support the cost of immediacy hypothesis which finds that orders are also cancelled as the market price moves closer. This is perhaps due to the fact that the benefits of trading immediately, and a price that is slightly better than when the limit order was initially placed, outweighs the cost of waiting for the original limit order to fill. There is also evidence that some of the fleeting orders are testing for hidden liquidity. The authors note that samples from US equity markets prior to the end of the 1990s do not show similarly large numbers of fleeting orders. Better trading technologies and highfrequency trading techniques could account for the increased number of fleeting orders. A spoofing order is a visible order that is placed in order to infuse misleading information regarding the supply or demand of a security. Once a trader creates the illusion of an order imbalance, he then takes advantage of the price change his spoofing order created and cancels the spoof order. For example, assume that a spoofing trader wishes to buy at least 1 tick below the best bid. He could place a large spoofing order at least 1 tick above the best ask in the hopes that this illusion of order imbalance in terms of excess supply lures other traders to sell ahead of him. If other traders sell ahead of him, it could push the market lower, enabling him to buy at a lower price. Once the spoofing trader buys at a lower price, he will simultaneously cancel the spoofing sell order.

22 15 Eom, Lee, and Park (2009) examine how investors spoof the equity market using data from the Korea Exchange. Prior to 2002, the Korea Exchange disclosed the total quantity of orders on each side of the order book, without fully disclosing the prices at which these orders were placed, in addition to providing traders with the five best bid and ask prices and the quantities associated with those prices. This feature, which was implemented to provide greater transparency to investors, actually made it easier for investors to spoof the market since it allowed spoofing traders to create the illusion of a fairly large order imbalance. The authors find that 0.81% of all buy orders are spoofing buy orders and are observed more frequently in stocks that had higher volatility, lower capitalization, and lower prices. Beginning in 2002, the Korea Exchange changed their order-disclosure rule and no longer disclosed the total quantity of orders on each side of the book. This change led to a significant decrease in the number of spoofing buy orders on the Korea Exchange from 0.81% to 0.26%. The very small percentage of spoofing buy orders relative to buy orders is rather surprising when compared with the findings of Hasbrouck and Saar (2002) who find that over 27% of all limit order submissions on the Island ECN to be fleeting (cancelled within two seconds of submission). The findings of Eom, Lee, and Park seem to be consistent with the predictions of Allen and Gale (1992). Allen and Gale develop a model that examines whether it is possible for a speculator who makes uninformed trades to make a profit simply by buying and selling shares. They find that, as long as investors assign a positive probability to the manipulator actually being an informed trader, the manipulator can earn a profit through his actions. Jarrow (1992) and Aggarwal and Wu (2006) develop models where large uninformed traders, whose trades are able to move prices, can be profitable manipulators.

23 16 Profitable manipulation is feasible if there is price momentum where an increase in price caused by a large uninformed speculative trader at one time typically leads to an increase in prices at a later time. This can occur because traders who are seeking information trade rationally on what they observe (a possibly informed, large trader s activity). Kong and Wang (2011) use a unique dataset to study the spoofing trades of Jianming Zhou on the Shanghai Stock Exchange. They find that Zhou s trading strategies follow that of an investor pretending to be an informed trader. That is, Zhou s trades were typically in periods of greater information uncertainty such as morning opening. They also found that he preferred to submit large block spoofing buy orders. In addition to studying Zhou s particular strategies, Kong and Wang study the impact of Zhou s spoofing trades on the market. They find that stock price, turnover, volume, and volatility is higher during the manipulation period and falls after the manipulation period. The manipulation period is defined as a -2 to +2 days window and the after manipulation period is a +3 to +62 day window. The choice of a five day manipulation period is surprising given that spoofing orders are typically thought to be a high frequency trading strategy. While a successful order is capable of moving the price in a way that is advantageous for the spoofer, it is not without significant risk. If a large market order is made, it is possible that the spoofer s limit order will be executed and the spoofer is obligated to trade and could cause the spoofer to incur significant cost. Given this very real risk, it is likely that a spoofer will operate in very short window of time. Impact of Macroeconomic News Announcements The basis of this dissertation depends on large order imbalances and the corresponding price changes over large imbalance intervals. While there are several

24 17 possible reasons for large imbalances, such as possible spoofing or testing for hidden liquidity, it is also likely that there could be large imbalances related to the expectations of major macroeconomic news announcements and reactions to the released news. Literature supports the idea that macroeconomic news announcements affect trading. For instance, Ederington and Lee (1993) study interest rate futures and foreign exchange futures around scheduled macroeconomic news releases such as the unemployment report, the consumer price index, and the producer price index. They find that scheduled announcements are responsible for most of the intraday volatility patterns in these products and most occurs within one minute of the announcement. However, volatility remains significantly higher than normal for about fifteen minutes after the announcement. Buckle, Ap Gwilym, Thomas, and Woodhams (1998) examine the impact of both UK and US economic announcements on the FTSE 100 stock index futures contract. They find that UK announcements positively affect volatility and volume while US announcements positively affect volatility and negatively affect volume. For S&P 500 index futures, Erenburg, Kurov, and Lasser (2006) find that trading volume increases significantly in the first few seconds after news release. Volume gradual declines after the initial increase in volume. It is possible that spoofing would be less likely around news announcements as more individuals are trading on new information. It is also possible that there could be more spoofing as traders are looking to take advantage of uncertainties and new expectations around these announcements. In his whitepaper study for Eurex, Asplund (2011) finds that eight of the fourteen economic reports with the most impact on Euro STOXX 50 futures are actually FED

25 18 reports rather than ECB reports. For instance, the greatest price change occurring in a ten minute window after the announcement is for U.S. Payroll Employment which witnessed a price change of 0.83%. The second most important indicator was German GDP with a ten minute price change of 0.65%. Out of the top eight indicators, in terms of percentage price change, seven were U.S. indicators. The Contribution of this Dissertation to the Literature This dissertation adds to the literature by examining a little studied phenomenon. Models and empirical research indicate that it is possible for uninformed traders to be profitable manipulators. To date, only two studies examine the high-frequency trading strategy known as spoofing. This dissertation is one of a limited number of studies examining spoofing by determining whether or not likely spoofing activity occurs in equity index futures markets. The intent is that this dissertation will provide a better understanding of spoofing orders. In addition to being one of three studies examining spoofing, this is the only study to date that tests for possible spoofing in derivatives contracts, specifically European equity index futures. It had been noted earlier that futures markets do not always behave in the same manner as equity markets for a variety. Therefore, this study provides additional information on spoofing in equity index futures. Additionally, this dissertation uses limit order book data for Eurex. The use of limit order book data will provide a better understanding of limit order book characteristics of European equity index futures.

26 Chapter 2: Methodology The Existence of Spoofing Orders The purpose of this study is to determine whether or not spoofing might occur in equity index futures markets. Specifically, this dissertation will examine DAX futures and DJ Euro STOXX 50 futures for evidence of possible spoofing. If spoofing does occur with these contracts, a valid question is the relative frequency of this strategy. Eom, Lee, and Park (2009) find evidence that suggests that spoofing-buy orders account for approximately 0.81% of the total buy orders in This frequency is quite low compared with evidence of fleeting orders witnessed by Hasbrouck and Saar (2002). They note that approximately 27% of all visible orders are cancelled within 2 seconds of submission. Therefore, if spoofing does occur in FDAX and FESX, what is the frequency of occurrences? Identifying Possible Spoofing Orders To date, there are few studies that touch on spoofing orders and only two that empirically examine spoofing orders. This is the first study to explicitly examine the possible existence of spoofing orders from an order book perspective. As such, it is critical to determine what a spoofing order might look like in an order book setting. As noted earlier, for a spoofing order to work, it must create the illusion of imbalance in the order book meaning that there appears to be excess supply or excess demand after the spoofing order is placed. Therefore, this study will focus on large order book imbalances on both the bid and ask sides. Imbalance is defined as the sum of contracts being offered minus the sum of contracts being sought at a given depth; say five or ten prices deep. For 19

27 20 instance, Imbalance10 is the imbalance of supply at the ten best ask prices and demand at the ten best bid prices. A positive number would indicate that more contracts are being asked than are bid. A negative number would indicate that more contracts are being bid than are asked. Likewise, Imbalance5 is the imbalance of supply at the five best ask prices and demand at the five best bid prices. A full list of variables is provided in Table 2 in Chapter 3. Imbalances at the 95 th percentile will be studied as these are probably sufficiently large enough for traders to notice the imbalance in the book. Spoofing can occur on either side of the book. In the case of a spoofing buy order, a trader will create the illusion of imbalance on the bid side of the book through a large bid order to the extent that there should be a large negative imbalance. The intent of this spoof order is to actually sell contracts at a higher price than is currently available. If the submission of a spoofing buy order is successful, that is, other traders perceive that there is real information in this order and trade ahead of it, the midprice will increase. Once this happens, the spoofer will cancel his buy order and enjoy the benefits of selling contracts at a higher price that is at least one tick higher than the prevailing price at the time the spoof order is submitted to the book. To determine whether or not spoofing buy orders occur in either the FDAX or FESX contracts, a subset of data examining order imbalances that are less than or equal to the 5 th percentile will be used, since an imbalance on the bid side will manifest as a negative imbalance number (imbalance is calculated by taking cumulative ask orders minus cumulative bid orders). By focusing on imbalances that are less than or equal to the 5 th percentile, the largest bid imbalances will be examined. During these periods of large bid imbalance, this study first determines how long a large bid imbalance lasts (in

28 21 terms of consecutive seconds) from when it first appears in the book until it is no longer visible in the book. Anecdotal evidence would suggest that these imbalances, if they are in fact spoofing orders, will not exist for long because of the risks associated with such a strategy. As noted earlier, there is the possibility that a large market sell market order could come in and cause this spoofing order to be executed before it can be cancelled as the large sell order might be executed as multiple transactions at different prices until it is filled. Therefore, it seems that there should be strong incentive to cancel the spoof order as early as possible. In addition to examining how long large bid imbalances remain in the book, this dissertation will examine the change in midprice at the end of the interval. Midprice is the simple average of the best bid price and the best ask price. In the case of a spoofing buy order, the spoofer is anticipating that the perceived imbalance on the bid side will entice others to trade ahead of him thereby raising the midprice. Anecdotal evidence suggests that an increase of one tick would probably be sufficient for a trader. This study will examine each interval and determine whether or not the midprice increases (as would be expected with a successful spoofing buy order), decreases, or remains unchanged by the time the large bid imbalance ceases to exist compared to when the large interval first appeared. The purpose of this is to (1) determine if possible successful spoofing buy orders take place by examining whether the desired effect of an increase in midprice occurs, (2) determine the percentage of these large imbalances that are possibly successful spoofing buy orders, and (3) determine how long possible spoofing orders might last on the books.

29 22 A similar approach will be used to investigate the possibility of spoofing sell orders. In the case of a spoofing sell order, a trader will create the illusion of imbalance on the ask side of the book through large ask order(s) to the extent that there should be a large positive imbalance. The intent of this spoof order is to actually buy contracts at a lower price than is currently quoted. If the submission of a spoofing sell order is successful, that is, other traders will perceive that there is real information in this order and trade ahead of it, the midprice will decrease. Once this happens, the spoofer will cancel his sell order and enjoy the benefits of buying contracts at a lower price than was possible before the submission of the spoof sell order. To determine whether or not spoofing sell orders occur in either the FDAX or FESX contracts, a subset of data examining order imbalances that are in 95 th percentile or higher will be examined. An imbalance on the ask side will have a positive imbalance number since imbalance is calculated by subtracting the cumulative number of bid contracts from the cumulative number of ask contracts. As with the spoofing buy orders analysis, this study is interested in determining how long a large ask imbalance last at any given point (the interval of imbalance). As noted earlier, there is risk in placing a spoof order and it seems reasonable to suggest that a spoofer will not want to be exposed to this significant risk for very long. In addition to examining how long large ask imbalances remain on the book, I will examine the change in midprice at the end of the interval. For a spoofing sell order, the spoofer is anticipating the perceived imbalance on the ask side will entice others to trade ahead of him thereby lowering the midprice. I will examine each interval of imbalance and determine whether or not the midprice decreases (as would be expected

30 23 with a successful spoofing sell order), increases, or remains unchanged at the time the large ask imbalance ceases to exist compared to what the midprice was when the large imbalance first started. The purpose of this is to (1) determine if possible successful spoofing sell orders take place by examining whether the desired effect of a decrease in midprice occurs, (2) determine the percentage of these large imbalances that are possibly successful spoofing sell orders, and (3) determine how long possible spoofing sell orders might last on the books for both FDAX and FESX front-month contracts. Intraday Patterns in Spoofing Orders Analyses of order books in equity markets have yielded interesting results with respect to intraday patterns. For instance, McInish and Wood (1991) find a U-shaped pattern for volatility for stocks trade on the Toronto Stock Exchange. McInish and Wood (1992) also find volatility and spreads for NYSE stocks to be highest right after the market opening (a reverse J shape). It seems plausible that spoofing might be more likely to submit orders earlier in the day when volume is greater and spreads are larger. If spreads are larger, there could be an increased incentive to spoof as a nudge in price would be more profitable. Additionally, spoofing might be preferred later in the trading day as well when volumes are higher. This is consistent with the evidence found by Eom, Lee, and Park (2009), who find that the rate of spoofing orders is highest in the morning, decreases over the day, and experiences a slight increase near market closing. Biais, Hillion, and Spatt (1995) find a U-shaped pattern in intraday orders and trades. In the morning, there is low depth and small trades and new limit orders that are not immediately executable are preferred. This possibly indicates a greater level of price discovery in the morning. In the afternoon, there is larger depth and larger trades are

31 24 preferred. These results, combined with the predictions of Allen and Gale (1992), suggest that one would expect to see a greater proportion of spoofing orders in the earlier hours of the trading day. Therefore, this dissertation is interested in determining at what point in the day spoofing orders are most likely to occur. To determine whether or not an intraday pattern exists, the trading day will be divided into fourteen intervals. The fourteen intervals are constructed at each hour from 08:00 CET until the end of the trading day at 22:00 CET 3. For each contract, both buy and sell orders will be examined. Specifically, at each interval this study will determine the total number of large imbalance intervals as well as the percentage of possible successful buy (sell) spoof orders and the percentage of large imbalance intervals that are not successful buy (sell) spoof orders. Given strong intraday patterns in many order flow studies and the fact that these contracts trade for fourteen hours each day compared with equities which typically only trade for seven hours each day, strong intraday patterns in terms of spoofing buy and sell orders are expected. If strong intraday patterns are in fact observed, this should provide additional insights as to when spoofers prefer to act in terms of time of day and market conditions. Impact of Macroeconomic News Announcements The basis of this dissertation depends on large order imbalances and the corresponding price changes over large imbalance intervals. While there are several possible reasons for large imbalances, such as possible spoofing or testing for hidden liquidity, it is also likely that there could be large imbalances related to the expectations of major macroeconomic news announcements and reactions to the released news. Large 3 The trading day for both contracts begins at 08:00 and ends at 22:00 CET.