This action might not be possible to undo. Are you sure you want to continue?
Jonathan A. Brogaard ∗ Northwestern University Kellogg School of Management Northwestern University School of Law email@example.com August 25, 2010
I would like to thank my advisors, Tom Brennan, Robert Korajczyk, Robert McDonald, and Annette Vissing-Jorgensen, for the considerable amount of time and energy they have spent discussing this topic with me; the Zell Center for Risk Research for their ﬁnancial support; and the many other professors and Ph.D. students at Northwestern University’s Kellogg School of Management and at Northwestern’s School of Law for assistance on this paper. Please contact the author before citing this preliminary work.
Electronic copy available at: http://ssrn.com/abstract=1641387
Abstract This paper examines the impact of high frequency traders (HFTs) on equities markets. I analyze a unique data set to study the strategies utilized by HFTs, their proﬁtability, and their relationship with characteristics of the overall market, including liquidity, price efﬁciency, and volatility. I ﬁnd that in my sample HFTs participate in 77% of all trades and that they tend to engage in a price-reversal strategy. I ﬁnd no evidence suggesting HFTs withdraw from markets in bad times or that they engage in abnormal frontrunning of large non-HFTs trades. The 26 high frequency trading (HFT) ﬁrms in the sample earn approximately $3 billion in proﬁts annually. HFTs demand liquidity for 50.4% of all trades and supply liquidity for 51.4% of all trades. HFTs tend to demand liquidity in smaller amounts, and trades before and after a HFT demanded trade occur more quickly than other trades. HFTs provide the inside quotes approximately 50% of the time. In addition if HFTs were not part of the market, the average trade of 100 shares would result in a price movement of $.013 more than it currently does, while a trade of 1000 shares would cause the price to move an additional $.056. HFTs are an integral part of the price discovery process and price efﬁciency. Utilizing a variety of measures introduced by Hasbrouck (1991a, 1991b, 1995), I show that HFTs trades and quotes contribute more to price discovery than do non-HFTs activity. Finally, HFT reduces volatility. By constructing a hypothetical alternative price path that removes HFTs from the market, I show that the volatility of stocks is roughly unchanged when HFT initiated trades are eliminated and signiﬁcantly higher when all types of HFT trades are removed.
Electronic copy available at: http://ssrn.com/abstract=1641387
Financial markets continuously evolve. Whenever a change in the market composition occurs, it is important to study the impact of the new development. In the 1980’s the pertinent issues were program trading (Harris et al., 1994) and the expansion of option markets (Skinner, 1989); In the 1990’s it was the relaxation of the order book (Barclay et al., 1999); In the early 2000’s it was algorithmic trading (Hendershott et al., 2008), the decimalization of prices (Chung et al., 2004), and the introduction of electronic communication networks (Huang, 2002). Today it is high frequency trading (HFT; and I use HFTs to refer to high frequency traders). HFT has changed the composition of the market and has brought concerns with it. The fact that HFT is a new breed of trading with no trade-by-trade human interaction that can execute dozens of transactions faster than a blink of an eye is concerning and makes it important to understand the impact it is having on the market. HFT has come to make up a large portion of the U.S. equity markets, yet the academic analysis of its role in the ﬁnancial markets is non-existence. This paper aims to ﬁll the gap. Widespread interest exists in understanding the impact of HFT on market quality: HFTs argue they improve liquidity, enhance price discovery, and reduce volatility, while others express concern that HFT may exacerbate volatility, consume liquidity, and proﬁt at the expense of more traditional investors. Despite these empirically veriﬁable claims this paper appears to be the ﬁrst formal academic study of HFT, primarily because the data necessary for such a study has previously been unavailable. In the press HFT has received an increasing amount of attention with most of it emphasizing the concerns with HFT. For example, on May 6, 2010 the Dow Jones Industrial Average dropped over 1,000 points in intraday trading in what has come to be known as the “ﬂash crash”. The next day, some media blamed HFTs for driving down the market (Krudy, June 10, 2010). Others in the media blamed the temporary withdrawal of HFTs from the market as causing the precipitous fall (Lee, August 10, 2010).1 Congress and regulators have begun to take notice and vocalize concern with HFT. The SEC issued a Concept Release regarding the topic on January 14, 2010 requesting feedback on how HFTs operate and what beneﬁts and costs they bring with them (Securities and Commission, January 14, 2010). The Dodd Frank Wall
To date, the true cause of the ﬂash crash has not been determined.
Electronic copy available at: http://ssrn.com/abstract=1641387
Street Reform and Consumer Protection Act calls for an in depth study on HFT (Section 967(2)(D)). The CFTC has created a technology advisory committee to address the development of high frequency trading. Talk of regulation on HFT has already begun. Given the lack of empirical ﬁndings for such regulation, the framework for regulation is best summarized by Senator Ted Kaufman, ”Whenever you have a lot of money, a lot of change, and no regulation, bad things happen” (Kardos and Patterson, January 18, 2010). There has been a proposal (House Resolution 1068) to impose a per-trade tax of .25%. Some have suggested implementing fees when the number of canceled orders by a market participant exceeds a certain level, or limit the number of canceled orders. While others have recommended requiring quotes to have a minimum life before they can be canceled or revised. Before discussing regulation to restrict HFT it is useful to know whether HFT is harming or beneﬁting markets. In this paper I examine the empirical consequences of HFT on market functionality. I utilize a unique dataset that distinguishes HFT from non-HFT quotes and trades. This paper provides an analysis of HFTs behavior and their impact on ﬁnancial markets. Such an analysis is necessary since to ensure properly functioning ﬁnancial markets the SEC, CFTC, Congress and exchanges must set appropriate rules for traders. These rules should be based on the actual behavior and implications of market participants. It is equally important that investors understand whether or not new market developments, like the rise of HFT, beneﬁt or harm them. This paper studies HFT from a variety of viewpoints and hopes to answer two fundamental questions. First, what are the activities of HFTs? Speciﬁcally, what category of trader do HFTs relate to, what type of trading behavior do they follow, how proﬁtable are they, and what are the determinants of their market participation? Second, how does HFT impact market quality? Using research design techniques that overcome data limitations I ask whether HFT affects liquidity, contributes to the price discovery process, and generates or dampens volatility. To answer these questions, the paper ﬁrst describes the activities of HFTs, showing that HFTs make up a large percent of all trading and that they both provide and demand liquidity. Their activities tend to be stable over time and across market conditions. Second, it examines HFTs strategy and proﬁtability. HFTs generally engage in a price reversal strategy, buying after price declines and selling after price gains. They are proﬁtable, making around $3 billion each year on trading volume of $28.3 trillion dollars. Third, it considers the impact of HFT on the market, focusing on three areas - liquidity, price discovery, and volatility. HFT accounts for a signiﬁcant portion of inside quotes and their presence reduce 4
and ﬁnally volatility. 2 Literature Review HFT has received little attention to date in the academic literature. 2007). The ﬁndings suggest that an upper bound on the proﬁts HFT can earn per 5 . and Nevmyvaka (2010). now some ﬁrms’ entire trading strategy is to buy and sell stocks multiple times within a mere second. Kulesza.e. (Hendershott and Moulton. In addition. has only been around since 1999. Tradebot. The only academic paper regarding HFT is one by Kearns. price discovery. Section 4 provides descriptive statistics. Using a variety of Hasbrouck measures. Not until March 2010 did Wikipedia have an entry for HFT. July 23. 2009). This is because until recently the concept of HFT did not exist. The rest of the paper is as follows: Section 2 describes the related literature and provides deﬁnitions of relevant terms. It was brought to the general public’s attention on July 23. Second. Because the trading process is the basis by which information and risk become embedded into stock prices it is important to understand how HFT is being utilized and its place in the price formation process. the data show HFT adds to the efﬁciency of the markets.the price impact of trades by a noticeable amount. a new type of trader has evolved to take advantage of these advances: the high frequency trader. As a result. The acceleration in speed has arisen for two main reasons: First. Section 3 discusses the data. HFT is a recent phenomenon. there have been technological advances in the ability and speed to analyze information and to transport data between locations. This smaller price variation makes trading with short horizons less risky as price movements are in pennies not eighths of a dollar. increase liquidity and price efﬁciency. Given these results. and proﬁtability. Section 6 analyzes HFT impact on market quality. 2009 in a New York Times article (Duhigg. market activity. a large player in the ﬁeld who frequently makes up over 5% of all trading activity and was one of the earliest HFTs. in order of liquidity. Section 5 analyzes HFTs investment strategy. data to conduct research in this area has not been available. and reduce volatility) for all market participants. Whereas only recently an average trade on the NYSE took ten seconds to execute. the data show HFT tends to decrease volatility. HFTs appear to be a new form of market makers and they appear to make markets operate better (i. the change from stock prices trading in eighths to decimalization has allowed for more minute price variation. Section 7 concludes. and this paper shows that the maximum amount of proﬁtability that HFT can make based on TAQ data under the implausible assumption that HFT enter every transaction that is profitable. Finally.
However. Regarding volatility. This paper ﬁts in to this literature by decomposing the AT type traders into shorthorizon traders and others and focusing on the impact of the short-horizon traders on market quality. these papers that distinguish traders based on their investment horizon do not try and deﬁne the horizons. Together these papers suggest that algorithmic trading as a whole improves market liquidity and does not impact.3 billion. or may even decrease. HFT touches on a variety of related ﬁelds of research. Chaboud. Hjalmarsson. they ﬁnd a positive relationship between AT providing the best quotes for stocks and the size of the spread. The study ﬁnds that AT increases liquidity and lowers bid-ask spreads. and Chiquoine (2009) look at AT in the foreign exchange market. Gsell (2008) takes a simple algorithmic trading strategy and simulates the impact it would have on markets. Their results suggest human order ﬂow is responsible for a larger portion of the return variance. Also. The author is unaware of any theoretical work conducted that directly addresses HFT. In principal AT is similar to HFT except that the holding period can vary. price volatility. Nonetheless several papers have studied AT. Some work has been conducted to understand what the impact on market quality will be of having investors with different investment time horizons. Hendershott. Hendershott and Riordan (2009) use data from the ﬁrms listed in the Deutsche Boerse DAX index. They ﬁnd that AT increase the efﬁciency of the price process and that AT contribute more to price discovery than does human trading. It is also similar to HFT in that data to study the phenomena are difﬁcult to obtain. and Menkveld (2008) utilize a dataset of NYSE electronic message trafﬁc. He ﬁnds that the low latency of algorithmic traders reduces market volatility. Jones. They ﬁnd that AT supply 50% of the liquidity in that market. two papers directly address the scenario when there are short and long term investors in a market: “Herd on the Street: Informational Inefﬁciencies in a Market with Short-Term Speculation” 6 . and use this as a proxy for algorithmic liquidity supply. they ﬁnd no evidence of there being a causal relationship between AT and price volatility of exchange rates. the study ﬁnds little evidence between any relationship between it and AT. With that caveat. The time period of their data surrounds the start of autoquoting on NYSE for different stocks and so they use this event as an exogenous instrument for AT. the most relevant being algorithmic trading (AT). Vega.year is $21. but that the large volume of trades increases the impact on market prices. Like Hendershott and Riordan (2009).
“professional traders acting in a proprietary capacity that engages in strategies that generate a large number of trades on a daily basis” (Securities and Commission. 2010. with diffuse arrival of information. often in terms of milliseconds and seconds. short horizons enhance it” (Vives. Although the paper does not extend its model in the following direction. include “pinging” and “algorithmic trading. “an immediate-or-cancel order that can be used to search for and access all types of undisplayed liquidity. and Stein. “[T]here is an important distinction between using tools such as pinging orders as part of a normal search for liquidity 7 . Other terms of interest when discussing HFT. and “Short-Term Investment and the Informational Efﬁciency of the Market” (Vives. December 2. This paper takes the deﬁnition from the SEC: HFT refers to. short horizons reduce ﬁnal price informativeness. 1992).” The SEC deﬁnes pinging as. Even the Securities and Exchange recognizes this and says that high frequency trading “does not have a settled deﬁnition and may encompass a variety of strategies in addition to passive market making” (Securities and Commission. January 14. The SEC goes on to clarify. January 14. Vives (1995) obtains the result that the market impact of short term investors depends on how information arrives. including dark pools and undisplayed order types at exchanges and ECNs. HFT is a type of strategy that is engaged in buying and selling shares rapidly.(Froot. January 14. January 14. 1995). 2010). 2010). The informativeness of asset prices is impacted differently based on the arrival of information.1 Deﬁnitions To date there lacks a clear deﬁnition for many of the terms in rapid trading and in computer controlled trading. 1995). “with concentrated arrival of information. By some estimates HFT makes up over 50% of the total volume on equity markets daily (Securities and Commission. although not the focus of this paper. Scharfstein. The theoretical work on short horizon investors suggest that HFT may be beneﬁcial to market quality or that it may be harmful to it. Scharfstein. a decrease in the informational quality suggests a decrease in price efﬁciency and an increase in volatility. The result being a decrease in the informational quality of asset prices. Froot. Spicer. 2009). and Stein (1992) ﬁnd that short-term speculators may put too much emphasis on some (short term) information and not enough on fundamentals. 2. 2010). The trading center that receives an immediate-or-cancel order will execute the order immediately if it has available liquidity at or better than the limit price of the order and otherwise will immediately respond to the order with a cancelation” (Securities and Commission.
2008. the data show whether the liquidity was provided by HFTs or non-HFTs. 2010. whereas HFT by deﬁnition hold their position for a very short horizon and try to close the trading day in a neutral position.1 Data Standard Data The data in this paper comes from a variety of sources. 2010. 2010). 2009). or longer. Along with the standard variables for limit order data. AT and HFT are similar in that they both use automatic computer generated decision making technology. A type of trading that is similar to HFT. 3 3. It provides the 10 best price levels on each side of the market that are available on the Nasdaq book. 2009 and from February 22. The trades include a millisecond timestamp at which the trade occurred and an indicator of what type of trader (HFTs or not) is providing or taking liquidity. they differ in that AT may have holding periods that are minutes. excluding trades that occurred at the opening.26. 3. It includes the best bid and ask that is being offered by HFTs and by non-HFTs at all times throughout the day. However.2 Nasdaq High Frequency Data The unique data set used in this study has data on trades and quotes on a group of 120 stocks. The Book data is from the ﬁrst full week of the ﬁrst month of each quarter in 2008 and 2009. The Quote data is from February 22. The trade data used in this study includes those from all of 2008. 2010 to February 26. 2010. HFT must be a type of AT. but AT need not be HFT. and manage those orders after submission” (Hendershott and Riordan. and whether the liquidity was displayed or hidden. The trade data consists of all trades that occur on the Nasdaq exchange. It uses in standard fashion CRSP data when considering daily data not included in the Nasdaq dataset. Thus. closing. weeks. The Nasdaq dataset consists of 26 traders that have been identiﬁed as engaging 8 . 2010 to February 26. January 14. days. but fundamentally different is algorithmic trading (AT). September 15 .with which to trade and using such tools to detect and trade in front of large trading interest as part of an ‘order anticipation’ trading strategy” (Securities and Commission. submit orders. and during intraday crosses.19. Compustat data is used to incorporate ﬁrm characteristics in the analysis. AT is deﬁned as “the use of computer algorithms to automatically make trading decisions. and February 22 .
Firms that others may deﬁne as HFTs are not labeled as HFT ﬁrms here if they satisfy one of the following: brokerage ﬁrms who provide direct market access and other powerful trading tools to its customers. ﬁrms that engage in HFT activities but are small. considering all types of trades. This was determined based on known information regarding the different ﬁrms’ trading styles and also on the ﬁrms’ website descriptions. but otherwise a relatively close match to an average Compustat ﬁrm. The HFT ﬁrms tend to switch between long and short net positions several times throughout the day. The stocks consist of a varying degree of market capitalization. that is. HFT ﬁrms normally have a lower ratio of trades per orders placed than do non-HFT ﬁrms. This table shows that these stocks are quite average and provide a reasonable subsample 9 . the ﬁrms do not have customers but instead trade their own capital. The HFT ﬁrms engage in sponsored access providers whereby they have access to the co-location services and can obtain large-volume discounts. an independent ﬁrm that is engaged in HFT activities. integrated ﬁrm. 2010. as HFT data has not been identiﬁed before. Orders by HFT ﬁrms are of a shorter time duration than those placed by non-HFT ﬁrms. 4 Descriptive Statistics Before entering the analysis section of the paper. industries. I then compare the ﬁrm characteristics to the Compustat database and show they are on average larger ﬁrms. and listing venues. and demanding liquidity trades. market-to-book ratios. proprietary trading ﬁrms that are a desk of a larger. These statistics are taken for the ﬁve trading days from February 22 to February 26. I ﬁrst provide the basic descriptive statistics of interest. The characteristics of HFT ﬁrms that are identiﬁed are the following: They engage in proprietary trading. Table 2 describes the 120 stocks in the Nasdaq sample data set. Also. These sample stocks were selected by a group of academics. The data is for a sample of 120 Nasdaq stocks whose ticker symbols are listed in table 1. Finally. whereas non-HFT ﬁrms rarely switch from long to short net positions on any given day. I look at liquidity and trading statistics of the HFT sample and show they are typical stocks. like a large Wall Street bank with multiple trading desks.primarily in high frequency trading. supplying liquidity trades. I provide general statistics on the percent of the market trades in which HFT is involved. The HFT ﬁrms use sophisticated trading tools such as high-powered analytics and computing co-location services located near exchanges to reduce latency. but who routes its trades through a MPID of a non-HFT type ﬁrm.
List of tickers for the set of sample stocks containing HFT information.Table 1: Sample Stocks. 10 AA ARCC BZ CNQR CTSH FFIC GPS KTII MOD PNY AAPL ABD ADBE AXP AYI AZZ CB CBEY CBT COO COST CPSI DCOM DELL DIS FL FMER FPO HON HPQ IMGN LANC LECO LPNT MOS MRTN MXWL PPD PTP RIGL AGN AINV AMAT BARE BAS BHI CBZ CCO CDR CPWR CR CRI DK DOW EBAY FRED FULT GAS INTC IPAR ISIL LSTR MAKO MANT NC NSR NUS ROC ROCK ROG AMED AMGN BIIB BRCM CELG CETV CRVL CSCO EBF ERIE GE GENZ ISRG JKHY MDCO MELI NXTM PBH RVI SF AMZN ANGO APOG BRE BW BXS CHTT CKH CMCSA CSE CSL CTRN ESRX EWBC FCN GILD GLW GOOG KMB KNOL KR MFB MIG MMM PFE PG PNC SFG SJW SWN .
If a ﬁrm’s year-end is on a different date.5 107.838 Trade Size 139.617 Depth (Thousand Dollars) 71.9 billion. is used. The average trade size. Min. the average quoted half-spread of 1.161 2027.573 Daily Trading Volume (Millions) 1. Quoted half-spreads are calculated when trades occur. dividing by two and taking the average per day.983 Quoted Half Spread (cents) 1.506 Table 3 describes the 120 stocks in the HFT database compared to the Compustat database. The table shows that the average HFT database ﬁrm is larger then the average Compustat ﬁrm. making the HFT sample appear overweighted with larger ﬁrms. Max.631 37 1597 196. whereas some stock trade just 8 times on a given day while others trade as many as 59. Whereas the average Compustat ﬁrm has a market capitalization of $2.82 cents is comparable to large and liquid stocks in other markets.6 and 544.59 billion. in shares is 139.064 Daily Number of Trades per Day 5150. Dev. The data for both the Compustat and the HFT ﬁrms are for ﬁscal year end on December 31. measured by summing the depth at the bid and at the ask times their respective prices.799 times. from the very small with a market capitalization of only $80 million. The price of the stocks is on average 39. 2009.150 trades.6. 2009.628 544. The market-to-book ratio also differs between 11 .of the market. is $71.336 4. The 120 stocks are quite liquid. The Compustat database statistics include the ﬁrms that are found in the HFT database.812 8 59799 4. Table 2: Summary Statistics. This is done on average over 5.956 0.000 shares to 14 million shares.064 million shares.55 N 600 Variable Std.550. but prior to December 31. 2010. 60. Summary statistics for the HFT dataset from February 22. The daily trading volume on Nasdaq for these stocks averages 1. The Compustat ﬁrms consist of all ﬁrms in the Compustat database with data available and that have a market capitalization of greater than $10 million in 2009.421 1. the ﬁscal year-end that is most recent. Compustat includes many very small ﬁrms that reduce its mean market capitalization. Mean Price 39. the average HFT database ﬁrm has a market capitalization of $17.046 2. However the sample does span a large variation of ﬁrm sizes.57 and ranges between 4. and ranges from as small as 2. 2010 to February 26.137 0. The average depth of the inside bid and ask.6 billion.5 42. to the the very large with a market capitalization of $175.002 14.857 7591.
6% of shares trading.46 billion.9. make informational-based trades. Table 4 looks at the prevalence of HFT in the stock market. Telecommunications. They make up only 47. There is nothing inconsistent with HFTs being market makers and also demanding liquidity as Chae and Wang (2003) and Van der Wel (2008) ﬁnd market makers frequently take liquidity.HFT Involved In Any Trade splits the data based on whether HFT was involved in any way in a trade or not. and “Dollar” reports the dollar value of those shares traded. Whereas HFT have a mean market-to-book of 2. The HFT database tends to overweight Manufacturing. market capitalization. The results show that HFT make up over 77% of all trades.4% of all trades.8% of the trading volume. and underweight Energy and Other. and market-to-book values. HFTs takes liquidity in 50. and dollar-volume that have HFT involved compared to trades where no HFT participates. Panel C . HFTs tend to trade in smaller shares as per-share traded they make up just under 75%. with half of the ﬁrms listed on each exchange.Compustat and the HFT sample. Healthcare. 2 12 . and with the industries more closely matching those in the HFT dataset. and make a signiﬁcant portion of there proﬁts from their non-liquidity providing activities.HFT Involved As Liquidity Taker groups trades into the HFT category only when the HFT is on the liquidity demanding side of the transaction. suggesting they provide liquidity in stocks that are slightly higher priced. The amount of Comparing Compustat ﬁrms that are listed on NYSE or Nasdaq reduces the number of ﬁrms to 5050 with an average market capitalization of $3. The industries are determined based on the Fama-French 10 industry designation from SIC identiﬁers.65. Panel B . Finally.HFT Involved As Liquidity Supplier groups trades into HFT only when the supply of liquidity in a transaction is coming from HFT. Three different statistics are calculated for each split of the data. The HFT ﬁrms are all listed on the NYSE or Nasdaq exchange. It captures this in a variety of ways: the number of trades. The next two panels separate HFT transactions into what side of the trade it is on based on liquidity. The table provides summary statistics for the involvement of HFT in the market. Panel A . worth a dollar amount of just about the same percentage of all transactions on a dollar basis. Whereas about one-third of Compustat ﬁrms are listed on other exchanges.2 The HFT database provides a robust variety of industries. “Shares” reports the number of shares traded. the Compustat data has one of 10. shares. Based on industry. based on a dollar-volume basis of trade. the HFT sample is a relatively close match to the Compustat database. The column “Trades” reports the number of trades. they make up 73.
034 . Compustat.3 20. The industries are categorized based on the Fama-French 10 industry groups.388 8260 Compustat Dataset Std.180 .1583 Industry .5 Exchange .120 .919 .288 .080 .Wholesale . . 197012.0917 Industry .602 3.0083 Industry . 12057.Durables .502 0 Mean 2613.5 Exchange .05 Industry .56 13 Mean Market Cap.025 Industry . Max.0333 Industry .Health Care .Energy . Dev.0333 Industry .272 .049 .322 .487 .Manufacturing .NYSE .124 .894 . This table compares the HFT-identiﬁed dataset with the Compustat dataset.024 .509 .180 .156 .183 .218 .65 Industry .38 80.040 Max.502 .235 .High Tech . 37852.058 .071 .358 .467 .Nasdaq . The Compustat data consists of all ﬁrms in the Compustat database with a market capitalization of $10 million or more.2833 Exchange .330 .Table 3: HFT Sample v.34 10. (millions) 17588.374 .217 . Min.499 .001 598.779 .289 .15 Industry .452 .1667 Industry .181 .091 . Dev.Other 0 Observations 120 HFT Dataset Std.257 .Non-Durables .134 -11.Other .034 .153 .366 .1 44843.Utilities .24 Market-to-Book 2.01 10.Telecom.453 . Min.014 .126 -2489. 322334.
This is true regardless of whether they are supplying or demanding liquidity. thereafter I analyze the promising strategies in more detail. Before getting to the ordered logit. and MacKinlay (1992).1 Investment Strategy HFTs do not readily share their trading strategies. I use an ordered logit regression to show HFTs trading strategy is heavily dependent on past returns.4% of all trades having their liquidity provided by HFT. Also. spreads. it is insightful to understand more about what drives HFT activity. it drops to 45. . I begin the analysis by performing an all-inclusive ordered logit regression into the potentially important factors. . Finally. The ordered logit is such that the lowest decision is to sell. I further identify that they engage in a price reversal strategy. HFTs must be basing their decision to buy and sell from short term signals such as stock price movements. . 5. as ﬁrst discussed by Hausman. the descriptive statistics (and upcoming section 6. singular of HFTs) makes at any given moment: Does it buy. t2 . with lower volume and lower spreads and depth. Let Zk be an unobservable con- 14 . Based on number of shares this value falls to 50. . HFTs trading behavior consist of a sequence of actions Z(t1 ). the middle option is to do nothing. Z(tη ) ∗ observed at regular time intervals t0 . HFT tends to occur in larger. . Z(t2 ). t1 . I summarize the theoretical reason for why an ordered logit is appropriate in this setting. value ﬁrms. I model this setting by using a three level ordered logit. Lo. Therefore.2 results) suggest they essentially partake in market making activities by providing liquidity and a continuous market into which other investors can trade. However.liquidity supplied is only slightly more than that demanded by HFTs at 51. or does it do nothing. This decision making process occurs continuously. What is known regarding HFTs is that they tend to buy and sell in very short time periods. based on their trading activities at the aggregate level I estimate HFTs earn approximately $3 billion a year. and based on dollar-volume. . .8% of all shares traded. There are three decisions a high frequency trader (HFTr. tη . or volume. To research this. 5 HFT Strategy Before analyzing HFT impact on market quality. rather than changes in ﬁrm fundamentals. and the highest option is to buy. whereby they tend to buy stocks at short-term troughs and they tend to sell stocks at short-term peaks. . does it sell.5% of all trades.
911 Shares (%) 74.4% 49.0% 303.435 160.817.588.383 49.354 638.090.307.4% 100.427.307.138. Panel C .971.433 638.387.HFT Involved In Any Trade Type of Trader HFT Non HFT Total Trades (Sum) 2.754 26.121 73.6% 52.590 51.HFT Involved As Liquidity Taker HFT Non HFT Total 1.879.8% 49. Panel A .6% 100.241.281. Panel A .9% 25.911 47.739 3.590 50.875 100.478 334.5% $26.0% Dollar (Sum) Dollar (%) $19.347.433 3.060.221.281. Three different statistics are calculated for each split of the data.875 100.Table 4: HFT Aggregate Activity.1% $13. The column Trades reports the number of trades.424.493 50.0% $11. The table provides summary statistics for the involvement of HFT in the market. Shares reports the number of shares traded.851 702.7% 100.557 314.HFT Involved As Liquidity Supplier groups trades into HFT only when the supply of liquidity in a transaction comes from HFT.044.HFT Involved As Liquidity Taker groups trades into HFT only when the liquidity demand of a transaction comes from HFT.9% $26.5% $14.476 638.HFT Involved In Any Trade splits the data based on whether HFT was involved in any way in a trade or not.264.944.090.3% 22.6% 220.127.116.112. Panel B .0% Panel C .533.8% $6.0% Shares (Sum) 477.977.556.0% $13.829 54.590 Trades (%) 77.4% 48.090.0% Panel B .0% 15 . and Dollar reports the dollar value of those shares traded.824 3.241.281.197.157 1.959.766 1.2% $26.911 50.0% 324.307.HFT Involved As Liquidity Supplier HFT Non HFT Total 1.337.046 45.1% 100.875 100.310.2% 100.
Buy. Lo.d. but not identically.i.n. σk ) ′ (1) where ‘i. The alternative ﬁne tuned separation. the subscripts are used to denote ten second period. εk i.. Whereas Hausman. . The essence of the ordered logit model is the assumption that observed HFTs ∗ behavior Zk are related to the continuous variable Zk in the following mapping: ∗ s1 if Zk ∈ A1 . is determined by the partition boundaries calculated from the ordered logit regression. As stated in Hausman. Therefore. conditioned on the Xk ’s and the other explanatory variables. conditioned on Xk and Wk . The ordered logit speciﬁcation allows an investigator to understand the link between ζ ∗ and ζ and relate it to a set of economic variables used as explanatory variables that can be used to understand the HFT trading strategy. not transaction time. is beyond the needs of this analysis. Note. which allows for heteroscedas2 ticity in σk . In this application the sj ’s are Sell. the observable actions could also be split into size. I assume the error terms in εk ’s in equation 1 are conditionally independently.tinuous random variable where ∗ 2 Zk = Xk β + εk . etc. N (0.’ stands for the assumption that the εk ’s are independent but not identically distributed. Sell 1000 + shares. and MacKinlay (1992) deal with tick by tick stock price data. Wk . . . m ∗ where the sets Aj form a partition of the state space ζ ∗ of Zk .1000. Lo. Sell 500 . that are omitted from equation 1.n. the scenario in this paper deals with HFTs trade behavior data that is aggregated into ten second intervals. for a Gaussian εk . s ∗ if Zk ∈ Am . but I restrict the ζ partition to these three natural breaks.d. The partition ∪m ∗ will have the properties that ζ = j=1 Aj and Ai ∩ Aj = ∅ for i ̸= j. . and MacKinlay (1992). Do Nothing. distributed. k Zk = . and Xk is a q × 1 vector of predetermined variables that ∗ sets the conditional mean of Zk . . the conditional distribution is 16 . and the sj ’s are the discrete values that comprise the state space ζ of Zk . The conditional distribution of observed return changes Zk .i. for example. E[εk |Xk ] = 0. s2 if Z ∗ ∈ A2 . for instance. by subdividing the buys and selling into the number of shares exchanged.
such as 250 milliseconds. shifting the boundaries will alter the probabilities of observing each state. the empirical relationship between HFTs behavior can be analyzed with respect to the economic variables Xk and Wk . The explanatory variables then allow one to analyze the different effects of relevant economic variables to understand HFT behavior . 17 . Here. and similarly where ten lagged ten second intervals show signiﬁcance. Wk ) ′ Φ( α1 −Xk β ) if i = 1 σk ′ ′ αi −Xk β αi−1 −Xk β = Φ( σk ) − Φ( σk ) if 1 < i < m.P (Zk = si |Xk . The intuition for the ordered logit model is that the probability of the type of behavior by HFTs is determined by where the conditional mean lies relative to the ′ partition boundaries. The ten second intervals has been adopted after attempting a variety of alterations but ﬁnding this one the best for keeping the results parsimonious and still being able to uncover important results. for a given conditional mean Xk β. Do Nothing. The 3 I also tried other time intervals. ′ if i = m. Wk ) ′ = P (αi−1 < Xk β + εk ≤ αi |Xk . As the data determines where the partition boundaries the ordered logit model creates an empirical mapping between the unobservable ζ ∗ state space and the observable ζ state space. ′ α −X β if i = m. The results from these alternative suggestions are similar in signiﬁcance to the results presented in that where a ten second period shows signiﬁcance. Wk ) if 1 < i < m. The order of the outcomes could be reversed with no real consequence except for the coefﬁcients changing signs as the ordered logit only takes advantage of the fact there is some natural ordering of the events. one second and 100 second periods. 3 For each ten second interval I utilize a variety of independent variables. or Buy. I divide the time frames in to ten second intervals throughout the trading day. so does the one lagged one hundred second interval. P (αm−1 < Xk β + εk |Xk . 1 − Φ( m−1 k k ) σ ′ ′ (2) (3) where Φ(·) is the standard normal cumulative distribution function. Wk ) if i = 1 P (Xk β + εk ≤ α1 |Xk . Sell. Therefore. so does the one second interval for ten lagged period’s worth. Wk ) = P (Xk β + εk ∈ Ai |Xk .
then I would expect to see them buy after prices rise. DollarV is the dollar-volume of shares exchanged in transactions for company i in that time period. 0 or 1. It could be they base their trading decisions on the spread and so the Spread variables would have considerable power in explaining when HFTs buy or sell.t−1 . trading in high volume markets.0 = (pricei. T rades is the number of distinct trades that occurred for company i in that time period.t−1 )/pricei.0−10 +β12−22 × Depthbidlagi. For example. there is sporadic signiﬁcance in all but one place.regression I run is as follows: HF Ti. The dependent variable. and it is 1 if on net HFTs were buying shares for stock i. momentum trading.0−10 +β45−55 × T radeslagi. Firm ﬁxed effects are implemented. HF T .0−10 +β23−33 × Depthasklagi.0−10 β56−66 × Dollarvlagi. is -1. the return for time period t is deﬁned as retlagi.0−10 . The results reported in table 5 are the marginal effects at the mean for the ordered logit. it is zero if the HFTs performed no transaction or its buys and sell exactly canceled.t − pricei.0−10 +β34−44 × Spreadlagi. It takes the value -1 if during that ten second period HFTs were on net selling shares for stock i. price reversal trading. If HFTs are price reversal traders. A handful of different strategies have been suggested in which HFTs engage. Spread is the average time weighted spread for company i in that time period. Depthbid is the average time weighted best bid depth for stock i in that time period. Thus the betas represent row vectors of 1x11 and the explanatory variables column vectors of 11x1. And. Each explanatory variable has a subscript 0-10. and to sell after prices fall. Table 5 shows the results. From the ordered logit regression’s summarized results in table 5. From this ordered logit model one may expect to see a variety of potential patterns. For instance. then I would expect to observe them buying when prices fall and to sell when prices are rising.t = α +β1−11 × Retlagi. where spread is the best ask price minus the best bid price. retlag0 represents the return for the particular stock during time period t. or trading in high spread markets. If HFTs are in general momentum traders. Depthask is the average time weighted best offer depth for stock i in that time period. There is a strong relationship with higher past returns and the 18 . Subscript 0 represents the contemporaneous value for that variable. This represents the number of lagged time periods away from the event occurring in the time t dependent variable. the lagged values of company i’s stock returns.
15) (-0.00528 0.16e-13 -5.31) (-2.99) (-0.88e-13 -5.10) (-0.69) (-0.00870 -0.79e-13 1.t−1 .86) (2.99e-13 T-Stat (-1.30) Variable depthasklag1 depthasklag2 depthasklag3 depthasklag4 depthasklag5 depthasklag6 depthasklag7 depthasklag8 depthasklag9 depthasklag10 depthasklag0 tradeslag1 tradeslag2 tradeslag3 tradeslag4 tradeslag5 tradeslag6 tradeslag7 tradeslag8 tradeslag9 tradeslag10 tradeslag0 dvolumelag1 dvolumelag2 dvolumelag3 dvolumelag4 dvolumelag5 dvolumelag6 dvolumelag7 dvolumelag8 dvolumelag9 dvolumelag10 dvolumelag0 Coefﬁcient -8.97) (2.33e-14 6.82e-13 9.03e-13 -1. is -1. Depthbid is the average time weighted best bid depth for stock i in that time period.0000171 0.29) (-2.43e-13 -1.000114 0.000184 0.05) (0.30e-13 -2.33) Marginal effects.26) (-0.254∗∗∗ 2. t statistics in parentheses p < 0.31) (-0.001 19 .01.70e-13 -3. Depthask is the average time weighted best offer depth for stock i in that time period.017∗∗ 4. HF T .56) (0.Exploratory Regression.47) (0.75) (1.14) (1.79) (0.30) (1.724∗∗∗ 1.68e-12∗ 8.22e-13 -1.42) (-1.05) (1.000176∗ -0.55e-13 -2.08) (0.04) (5.80e-12 1281695 T-Stat (0.50) (-1.22) (4.000316 0.03) (0.19) (-0.01) (-1.97e-13 -5.00126 0.98) (-0.68e-12 -0.08) (2.00254 -0.15e-13 3.70) (-0.18) (1. This table includes several explanatory variables in order to uncover in which strategies HFTs are engaged.14) (5.06) (1.07e-13 8.58e-13 1.17) (-1. For example.35) (5.75e-12∗ -1.34) (0.63) (4.74) (1.744∗∗∗ 4. and it is 1 if on net HFTs were buying shares for stock i.00549 -0.06) (0.461 5. T rades is the number of distinct trades that occurred for company i in that time period. Variable retlag0 retlag1 retlag2 retlag3 retlag4 retlag5 retlag6 retlag7 retlag8 retlag9 retlag10 spreadlag1 spreadlag2 spreadlag3 spreadlag4 spreadlag5 spreadlag6 spreadlag7 spreadlag8 spreadlag9 spreadlag10 spreadlag0 depthbidlag1 depthbidlag2 depthbidlag3 depthbidlag4 depthbidlag5 depthbidlag6 depthbidlag7 depthbidlag8 depthbidlag9 depthbidlag10 depthbidlag0 N ∗ Coefﬁcient 7.000203∗∗ -0.61e-13 -1.0000886 0.t−1 )/pricei.65) (3.19e-13 2.69) (0. it is zero if the HFTs performed no transaction or its buys and sell exactly canceled. 0 or 1.Table 5: HFT Ordered Logit .00960 0.71e-13∗∗ 6.00000946 0. The dependent variable.423∗ 2.245∗∗ 1.69) (-1.07) (2. This represents the number of lagged time periods away from the event occurring in the time t dependent variable.00199 -0.49e-13 -2. ∗∗∗ p < 0. It takes the value -1 if during that ten second period HFTs were on net selling shares for stock i.176∗∗∗ 4.0 = (pricei.22) (0.88e-13 -1.19) (-1.30) (-1.16) (0.24) (0. the return for time period t is deﬁned as retlagi.05. Each explanatory variable is followed by a number between 0 and 10.t − pricei.00332 7.36) (1.216∗ 0. ∗∗ p < 0.000169∗ -0.000165∗ 0.0000884 -0.38) (0. The regression uses ﬁrm ﬁxed effects.77) (-0. Spread is the average time weighted spread for company i in that time period.21e-13 6.55) (-0.47e-13 -3.000208 -4.00456 0. DollarV is the dollar-volume of shares exchanged in transactions for company i in that time period.29e-13 2.31) (-0.405∗∗∗ 4.37e-12∗ 5.18) (-0.73e-13 -3. retlag0 represents the return for the particular stock during time period t.29) (3. And.09e-14 3.95) (1.21) (-2.49) (3.38) (1.00000749 0.65) (-1. where spread is the best ask price minus the best bid price.72e-13 -5. Subscript 0 represents the contemporaneous value for that variable.26) (0.577∗∗∗ 5.21) (-0.82) (2.81e-13 5.47) (0.50) (-1.
HFTs selling when demanding liquidity. It appears that HFTs engage in a price reversal strategy. There is some statistical signiﬁcance in other locations. HFTs buying. HFTs selling when supplying liquidity. This cannot be determined from this regression as the contemporaneous return will include within its time period HFTs transactions. This suggests that past spread size. HFTs buying when supplying liquidity. prior to HFTs executing a sale of a stock.likelihood the HFTs will be selling (and with low past returns and the likelihood the HFTs will be buying). and volume are not primary factors in HFTs trading decisions. This ﬁnding suggests HFTs in general engage in a price reversal strategy. which is consistent with a price reversal strategy. depth. To better understand HFTs trading strategy I run logit regressions on different dependent variables. all types. The third column in table 6 has as the dependent variable a one if HFTs were on net taking liquidity from the market and selling during the ten second interval and a zero otherwise. The signs are the same as before. Of the strategies discussed above. I examine HFTs buy and sell logits separately. The next column has as the dependent variable a one if HFTs were on net supplying liquidity to the market and selling during a given ten second interval and a zero otherwise. except that the magnitude and statistical signiﬁcance is not as strong. barring the ninth one. The results show the strong relationship between past returns and HFTs decision to sell. but I cannot determine whether HFTs were selling before prices fell or after they fell within this ten second increment. barring time period 8. and HFTs buying when demanding liquidity. The results found in table 6 are the marginal effects at the mean and the logit incorporates ﬁrm ﬁxed effects. There appears to be more scattered signiﬁcance of past returns. To analyze this further. I consider a total of six different regressions: HFTs selling. The results are similar to the previous results. these results are consistent with a price reversal trading strategy. with statistically signiﬁcance up to 90 seconds prior to the trade. The ﬁrst column is the results for HFT Sell. It is not clear from the logit model whether this means that HFTs initiate a sale once prices have started to fall. focusing on the lagged returns surrounding HFTs’ buying or selling stocks and decomposing the differences in demanding versus supplying liquidity activity. To further understand this potential price reversal strategy I focus on analyzing the lag returns inﬂuence on HFTs’ trading behavior. 20 . There is still strong statistical signiﬁcance from the ten past return periods. or that after they start selling prices fall. the stock tend to rise. however no where is it consistent like that of the return coefﬁcients. One large difference is the fact that the contemporaneous period return coefﬁcient is large and negative.
And. (2) HFTs on net sell and supply liquidity.13) -2.925 16.26) 1377798 (5) HFT B . retlag0 represents the return for the particular stock during time period t.498∗∗∗ (3.50) -1.145∗∗ -0.038∗ 2.881∗∗∗ 1.14) retlag10 0. (1) (2) (3) HFT S .93) -0.96) (9.395∗ (-2.34) -6.68) retlag9 2.490∗∗∗ (-3. (4) HFTs on net buy in a given ten second period.226 (-0.553∗ (-2.513∗ (-2.278∗∗∗ 2.45) retlag6 5.770∗∗ (-2.572∗∗∗ (-3.28) -1.89) -3. The reported coefﬁcients are the marginal effects at the mean.99) (0.087 (0.202∗∗∗ (-3.82) (2.Table 6: Regressions of the Sell decision.80) -2.85) -1.35∗∗∗ -16.763∗∗∗ (-4.54) retlag3 6.05. t statistics in parentheses ∗ p < 0.910∗∗∗ (-4.32) -3.0277 1.39) (-1.260∗∗ (-3.26) retlag7 3.934 7.110∗∗∗ (-6.276∗∗∗ (5.384∗∗∗ (-6.351 -0.69) (-8.257∗∗∗ (-6.194∗∗∗ 2. ∗∗ p < 0.t−1 .023∗∗∗ (-4. and (3) HFTs on net sell and demand liquidity.234 4.t − pricei.77) -2. split based on Liquidity Type.37) -6. Subscript 0 represents the contemporaneous value for that variable.274∗ (-2.521∗∗∗ 2.41) retlag5 4. This table reports the results from running a logit with dependent variable equal to 1 if (1) HFTs on net sell in a given ten second period.43) (3.291∗∗∗ (-5.38) retlag1 5.906∗∗ (-2.48∗∗∗ (0.042∗∗∗ (-4.380∗∗∗ 0.21) retlag8 0.098∗∗∗ 2.S HFT S .65) (4.87) -5.766∗∗∗ (-3.87) (1.88) 1366278 (6) HFT B .25) -4.75) -2.25) -6.270 2.594∗∗∗ (3.t−1 )/pricei.80) -2.46) (3. Firm ﬁxed effects are used.69) -1.22) N 1377798 1377798 1343177 Marginal effects.S -48.03) (6.874 (-0.44) (5.989∗∗∗ (4. (5) HFTs on net buy and supply liquidity.793 (-0. For example.01) 0.001∗∗ (-2.497 3.87) (4.521∗ 0.48∗∗∗ (18.91) (3. the return for time period t is deﬁned as retlagi.21) -4.28) -2.60) (5.38) -4.854 1.408 (-1.923 0.A -2.80) retlag2 5.049∗ (-2.13) (1.209∗∗∗ (4. Each explanatory variable is followed by a number between 0 and 10.230∗∗∗ 1.439∗∗∗ (3.01.04) (1.13) (-0.445 (0.43) -2.08) -1.03) retlag4 4.10∗∗∗ (-14.84) -3.908∗∗ (-2.83) (4. and 0 otherwise.022 (-1.610∗∗ (2.D 53.684∗∗∗ 4.67) -1.29) -6.179 4.533∗∗∗ (-4.460∗∗∗ (-6.14) -3.32) -2.603∗ (0. ∗∗∗ p < 0.44) -7. and (6) HFTs on net buy and demand liquidity.66) (3. This represents the number of lagged time periods away from the event occurring in the time t dependent variable.84) (0.802∗∗∗ (-3.0 = (pricei.D retlag0 4.67) 1377798 21 .001 (4) HFT B .28) (2.619∗∗∗ (3.A HFT S .53) (0.45) -0.
Next. except for the contemporaneous period return. I graph the results in ﬁgure 1. The results in table 6 show that HFT are engaged in a price reversal strategy. This is true whether they are supplying liquidity or demanding it.Demand scenario.1 Front Running A potential investing strategy of which HFTs have been claimed to be engaged in is front running. The results in the lag returns are similar to the previous results. The result of such an action by HFTs would be to drive up the cost for non-HFTs to execute the desired transaction. If they detect a large order coming through they may increase their trading activity. That is.The Buy regressions are also shown in table 6. some claim HFTs ping stock prices to detect large orders being executed.10).6. Like in the HFT Sell . The fourth column is the result for HFT Buy. and the SEC’s concern with it apply. the stock tend to fall. The x-axis is the 20 different non-HFTr initiated 22 . The ﬁfth column has as the dependent variable a one if HFTs were on net supplying liquidity to the market and buying during a given ten second interval and a zero otherwise. The last column in table 6 has as the dependent variable a one if HFTs were on net taking liquidity from the market and buying during the ten second interval and a zero otherwise. Some charge HFTs with detecting when other market participants hope to move a large number of shares in a company and that the HFTs enters into the same position just before the other market participant. as deﬁned in section 2. I look at the average percent of trades that were initiated by HFTs for different number of trades prior to a non-HFTr initiated trades (for prior trades 1 . It is in this context where pinging. all types. 5. except that the magnitude of the coefﬁcients are smaller.1. To see whether or not this is occurring on a systematic basis I perform the following exercise: For each stock over the database time series I create twenty bins based on trade size for trades initiated by non-HFTs. Each bin has roughly the same number of observations.1. which is large and positive. The signs for the lag returns are negative as expected. with statistically signiﬁcance up to 100 seconds prior to the trade. There is still some statistical signiﬁcance from the ten past return periods. Prior to HFTs executing a purchase of a stock. it is not clear from this logit model how to interpret this. The results show the strong relationship between past returns and HFTs decision to buy. but only in time periods 0 and 3 . There is an especially large relationship with the contemporaneous period return and the HFTs decision to supply liquidity and buy in a trade.
trades initiated by one type of market participant have a greater probability of being preceded by the same type of market participant. The graph shows the percent of trades initiated by HFTs for different prior time periods that precede different size non-HFTr initiated trades. the z-axis is the different prior trade periods. it appears that larger trades. The x-axis is the 20 different non-HFTr initiated trade size bins. The ﬁgure suggests front running by HFTs before large orders is not systematically occurring. Figure 1: HFT Front Running. the y-axis is the fraction of trades for different non-HFTr trade size bins for different prior trade periods that were initiated by HFTs. The non-HFTr trades that are preceded by the highest number of HFTr initiated trades are those that are small and those that are of moderate size.2 HFT Market Activity In addition to understanding the trading behavior of HFTs at the trade by trade level. In fact. relative to each stock. which may explain why the graph’s pattern occurs.trade size bins. the y-axis is the fraction of trades for different non-HFTr trade size bins for different prior trade periods that were initiated by a HFTr. the z-axis is the different prior trade periods. tend to be preceded by fewer HFTr initiated trades. it is informative to understand what drives HFTs to trade in certain stocks 23 . Also. it is interesting that the immediately preceding trades tend to have fewer HFTr initiated trades than those further out. 5.1. As will be shown later.
and those in Pan24 .5% of trades in the average stock per day.9%.2. Panel A shows that HFT share of the market varies a great deal depending on the stock and the day.8% to 93. HFT demand liquidity in 39.1 HFT Quote Revisions and Cancelations This section provides summary statistics on HFTs quote revision and cancelation behavior.2. Panel C is the percent of trading variation of HFT and non-HFT in demanding liquidity for a particular stock on a given day. I look at the frequency of quote changes at the inside bid and ask for HFT and non-HFT quotes. but less than when they supply liquidity. Panel B looks at HFT supplying liquidity. Quote cancelations and revisions have been found to have net economically signiﬁcant beneﬁts by reducing the non-execution cost that would otherwise occur (Fong and Liu.8% of all trades. notice the wide variation in the supply of liquidity.1. as they make up 77% of all trades in the entire market. in the Non-HFT row the numerator is the number of those changes which where for non-HFT quotes. 2010). Analysis of liquidity will be provided in detail in section 6. the HFT demand for liquidity varies substantially ranging from 3. while in others they supply 74%. Table 7 shows the variation in HFT market makeup in different stocks on different days. HFT must supply liquidity in stocks that trade more frequently. compared to the numbers seen in the descriptive statistics from table 4. With this as the denominator.6% to 79. which. This number is substantially smaller than the 50% they were found to supply in the market as a whole in table 4. suggests they trade more in stocks that trade frequently. HFT supply liquidity in 35. I examine the data in three ways. So HFT must be taking liquidity in stocks that trade more frequently. They average being involved in 61. Also. The results are in table 8. based on number of trades. Panel B is the percent of trading variation of HFT and non-HFT in supplying liquidity for a particular stock on a given day. These results.on certain days. in the HFT row the numerator is the number of those changes which where for HFT quotes. the next step is to consider which determinants result in an increase or decrease in HFT activity. In Panel A I sum up the total number of quote changes for each stock on each day. 5. Its percent of all trades varies from 10. Thus. Before doing so. however.6%. Panel C looks at HFT demanding liquidity. The results in table 7 show there is a large variation in the degree of HFT in different stocks over time. Panel A is the percent of trading variation of non-HFT and HFT in a certain stock on a given day. in some stocks they provide no liquidity. Also.6% of trades in the average stock per day. I brieﬂy show the summary statistics of HFT quote changes.
0%100.0%100.HFT Involved In A Stock As Liquidity Taker HFT Non HFT Total 39.4% 33.5% 15.0% 100.0%100.2% 66.4% 92.0%67.Table 7: HFT Market Participation Summary Statistics.0% 18.HFT Involved In A Stock Trades Mean Median Std.7% 41.3% 16. Shares Mean Median Std.24 6.4% 64.8% 37.6% 58.8% 90.4% 100.7% 16.49 2.8% 35.7% 14. Panel C is the percent of trading variation of HFT and non-HFT in demanding liquidity for a particular stock on a given day.1% 19.0% 25 .3% 37.8% 93.9% 37. Panel B is the percent of trading variation of HFT and non-HFT in supplying liquidity for a particular stock on a given day.8% 62.0% Panel C .63 25.83 9.4% 32.99 7.1% 60.0% 100. Min Dev.8% 64.1% 93.4% 59.6% 18.0%100.3% 16.4% 17.1% 65. This table shows the variation in HFT market makeup.9% 15. Panel A .6% 79.0%100.7% 16.25 10.6% 78.13 33.9% 39. Panel A is the percent of trading variation of non-HFT and HFT in a certain stock on a given day.5% 67.0%100.73 21.54 0. Type of Trader HFT Non HFT Total Max Max 61.2% 42.0% Panel B .3% 16.6% 40.6% 100.1% 96.9% 61.1% 97.4% 62.0% 100.6% 100. Min Dev.1% 100.99 0% 74.HFT Involved In A Stock As Liquidity Supplier HFT Non HFT Total 36.43 3.70 20.0% 100.
093 595.093 595.els B and C show the minimum. and maximum quote changes.445 p75 0. These results.830 1.000 595. in the HFT row the numerator is the number of those changes which where for HFT quotes.016 0. median. 25th.926 0. Panel C uses the same HFT / Non-HFT variable.005 0.324 p50 0.559 0. Panel C uses the same N on−HF T variable.901 0. and those in Panels B and C show the minimum.322 3.559 0.000 26 . With this as the denominator.676 0.000 1. 0. median. 25th.258 297.683 0.073 1.969 1. Table 8: Quote Change Frequency.005 0.451 0.247 2.005 0. Observing the data this way emphasizes the higher frequency at which HFT change their quotes.454 0. Panel A Type HFT Non-HFT Panel B HFT / Non-HFT Panel C Low Market Cap.770 2. 75th. but splits it into two groups based on the median market capitalization of the data. 0.247 2.000 1. the HFT quote changes occur about 50% more often HF T than do non-HFT quote changes.009 0.000 1.538 9.005 1. Total Mean Min. and maximum changes. In large market cap ﬁrms. Panel B does this. those in which HFT tend to trade more in.995 N 595. An alternative approach which directly compares HFT and non-HFT quote change HF T behavior is to look at the ratio of changes.089 9. High Market Cap. HFT make quote changes almost twice as frequent as do non-HFT. This table examines the frequency of quote changes by HFT and non-HFT.555 0.099 p25 0. Panel B does this.000 1.830 1. in the Non-HFT row the numerator is the number of those changes which where for non-HFT quotes. where N on−HF T = the fraction of HFT quote changes divided by non-HFT quote changes for each ﬁrm day.549 0. An alternative approach which directly compares HFT and non-HFT quote change behavior HF T is to look at the ratio of changes. but splits it into two groups based on the median market capitalization of the data. 75th.089 9.546 Max.093 298. In Panel A I sum up the total number of quote changes for each stock on each day. where N on−HF T = the fraction of HFT quote changes divided by non-HFT quote changes for each ﬁrm day.
t = α + M Ci ∗ β1 + M Bi ∗ β2 + N Ti. Also statistically signiﬁcant and with moderate economically signiﬁcant is the dollar volume of non HFT. the more important its role in impacting the dependent variable. 2009. which is winsorized at the 99th percentile. 2009. the larger the coefﬁcient. instead of running the typical OLS regression on the regressors.t ∗ β4 + Depi.t . N V is the volume of non HFT dollars that were exchanged.1. The results in the full regression. where i is the subscript representing the ﬁrm. columns (1) and (2) show that market capitalization is very important and has a positive relationship with HFT market activity.t ∗ β6 + ACi. all else being equal. M B is the market to book ratio as of December 31.t ∗ β7 + ϵi. scaled by market capitalization. both dependent and independent. M C is the log market capitalization as of December 31. The coefﬁcients reported can be understood as signaling that when there is a one standard deviation change in an independent variable.5. I run the following regression: Hi. I perform an OLS regression with the dependent variable being the percent of share volume in which HFT were involved in for a given company on a given day. H is the percent of share volume in which HFT are involved out of all trades. the variables. t is the subscript for each day. V ol is the ten second realized volatility summed up over the day.t ∗ β3 + N Vi. are de-meaned. This makes the regressors underlying scale of units irrelevant to interpreting the coefﬁcients. Thus. equally weighted. and are divided by their respective standard deviations so as to standardize all variables. but with a very small negative coefﬁcient. AC is the absolute value of the Durbin-Watson score minus two from a regression of returns over the current and previous ten second period. That is. Column two reports the normal coefﬁcients. suggesting HFT tends to occur slightly more often in value ﬁrms. which is interpreted as HFTs preferring to trade when there is less volume.3 HFT Market Activity Determinants Table 9 examines which determinants drive HFT trading.t ∗ β5 + V oli. Table 9 reports the standardized regression coefﬁcients in column (1). The spread and depth variables are 27 . the coefﬁcient is the expected change in standard deviations that will occur in the dependent variable. The market to book ratio is slightly statistically signiﬁcant. N T is the number of non HFT trades that occurred. scaled by market capitalization. Dep is the average depth of the bid and of the ask.
all else being equal.5 A Closer Look at Volatility What has been shown so far has dealt mostly with means. HFTs reduce their trading activity. 28 . equity markets were especially volatile. In each graph there are three lines. The line labeled “All HFT” represents the fraction of exchanges in which HFT was involved either as providing liquidity or as taking liquidity. for example when volatility increases. the line “HFT Liquidity Demanded” represents the fraction of trades in which HFT was demanding liquidity. autocorrelation. 5. stopped trading. or increase. HFTs prefer to trade when there is less depth and lower spreads between bids and asks. Many of the explanatory variables may be endogenously determined and as a result the OLS estimator may be a biased. Thus. The results are also in table 9. I look at each trading day and count the fraction of activity in which HFT was involved.1. Especially of note. market experienced large losses. such as HFTs. Although the database I have does not include the May 6. but a major concern is that HFTs are around during normal times. but during extreme market conditions.1. There are three graphs. it does span 2008 and 2009.S. and the number of non HFT trades are not statistically signiﬁcant. 2010 data.S. Volatility. The ﬁrst is a time series of the fraction of trades were HFT was involved in during 2008 and 2009. The second graph looks at the fraction of shares in which HFT was involved during this period. when the U. there is no abnormally large drop. with column (3) reporting the standardized beta coefﬁcients and column (4) reporting the standard OLS coefﬁcient results. This restricted regression increases the magnitude of both explanatory variables to a degree. The results are shown in ﬁgure 2.4 HFT Market Activity Time Series A concern surrounding the May 6 “ﬂash crash” was that the regular market participants. The ﬁnal graph looks at the fraction of dollar volume in which HFT was involved during this period. and especially around time periods when the U. I run the same regression but only keeping Market Capitalization and Market to Book. which were volatile times in U. the line labeled “HFT Liquidity Supplied” represents the fraction of transactions in which HFT was providing liquidity.S. All three graphs have minimal volatility among the three measures. in HFT participation in the sample data as a whole occurring in September of 2009. To see whether HFT percent of market trades varies signiﬁcantly from day to day. but otherwise the economic impact is roughly the same. equity markets. 5.statistically signiﬁcant and both have medium economically signiﬁcance.
M C is the log market capitalization as of December 31.0000239∗∗∗ (-3.16) 0.017 (-0.51) -0. Column one shows the standardized beta coefﬁcients.533 29 .000932 (-0.62) # of Non HFT Trades 0.0725∗∗∗ (25.t .t + Depi. Dep is the average depth of the bid and of the ask. 2009. N V is the volume of non HFT dollars that were exchanged.138∗∗∗ (-3.042 (0. scaled by market capitalization.82) -0.54) Observations 590 590 590 Adjusted R2 0.0000426∗∗∗ (-3.t ∗ βi.41) 590 0. 2009.575 0. M B is the market to book ratio as of December 31. 0.0823∗ (2.16) $ of Non HFT Volume -0.t ∗ βi. ∗∗∗ p < 0.13) -0. AC is the absolute value of the Durbin-Watson score minus two from a regression of returns over the current and previous ten second period.063∗ (-2.111∗∗∗ (-3.01. t statistics in parentheses ∗ p < 0.t ∗ βi. where i is the subscript representing the ﬁrm. column three shows the standardized beta coefﬁcients for the regression excluding explanatory variables that may be endogenously determined.00195 (-1. ∗∗ p < 0.07) Autocorrelation -0.0194 (0.79) Volatility -0.62) 0.85) Market / Book -0.575 0. 0.00609∗ -0.51) (25.t ∗ βi.98) (4) OLS Coef.88) Average Depth -2.722∗∗∗ (19. Economic Impact Market Cap. which is winsorized at the 99th percentile. column two shows the regular coefﬁcients.85) -0.t + V oli. (2) (3) OLS Coef. equally weighted.132∗∗∗ (-4.79) -0.t ∗ βi.98) Constant 0.0682∗∗∗ 0.Table 9: Determinants of HFT Percent of the Market This table shows the result of the following OLS regression: Hi.533 Standardized beta coefﬁcients in (1) and (3). V ol is the ten second realized volatility summed up over the day. and column four shows the regular coefﬁcients.05.13) (-4. N T is the number of non HFT trades that occurred.88) -0.001 (1) Economic Impact 0.07) -0.031 (-1. scaled by market capitalization.757∗∗∗ (19. H is the percent of share volume in which HFT are involved out of all trades.t = α + M Ci ∗ βi + M Bi ∗ βi + N Ti.122∗∗∗ (-2.95e-11∗∗∗ (-4. t is the subscript for each day.t + N Vi.0119∗∗∗ (-4.t + ACi.0309 (1.82) Average Spread -0.
80 30 01 Jan 08 40 50 60 70 01 Jul 08 01 Jan 09 sas_date 01 Jul 09 01 Jan 10 HFT Liquidity Demanded Trades HFT Liquidity Supplied Trades All HFT Trades 20 01 Jan 08 40 60 80 01 Jul 08 01 Jan 09 sas_date 01 Jul 09 01 Jan 10 HFT Liquidity Demanded Shares HFT Liquidity Supplied Shares All HFT Shares 30 01 Jan 08 40 50 60 70 80 01 Jul 08 01 Jan 09 sas_date 01 Jul 09 01 Jan 10 HFT Liquidity Demanded DVolume HFT Liquidity Supplied DVolume All HFT DVolume 30 . The ﬁnal graph looks at the fraction of dollar volume in which HFT was involved. another line represents transactions in which HFT was providing liquidity. In each graph three lines appear.Figure 2: Time Series of HFT Market Participation The ﬁrst graph is a time series of the fraction of trades in which HFT was involved in during 2008 and 2009. The second graph looks at the fraction of shares in which HFT was involved. One line represents whether HFT was involved as either a liquidity provider or a liquidity taker. the ﬁnal line represents when HFT was demanding liquidity.
but the constant also rises substantially. such as the median. and this would not be picked up at a day level analysis. I can separate the analysis into positive and negative return episodes.t ∗ β1 + ϵi.HFT . Firm ﬁxed effects are used to control for HFT variability due to ﬁrm characteristics. in Panel A it includes trades with HFT in any capacity. To study how HFTs behaves in different levels of volatility I implement a quantile regression. I implement an OLS regression with dummy variables and interaction terms to capture the variation in the HFT . 80% and 95% (high volatility). HFT percent of trades increases. As volatility increases the coefﬁcients on HFT Demand Percent increase. HFT activity is not statistically signiﬁcant.t .] Table 10 looks at day level volatility. I perform the regression: V olai. The results of Panel C. in Panel B it is the percent where HFT demand liquidity. I am interested in understanding the relationship between HFT and volatility as volatility levels change. Given this. where V olai. Note though that the constant. In addition. except in the 95th percentile. is the opposite: as volatility increases the HFT Supply Percent coefﬁcient tends to decrease. Panel B and C break down HFT activity by liquidity. The results are in table 10.ALL shows that as volatility increases from the 5% group to the 80% group. This regression is done at the 5% (low volatility). Again though the constant is increasing in volatility. I look at the data in 15 minute intervals. instead of looking at volatility I examine returns in the 15 minute period. HFT supplying liquidity activity .An alternative concern is that HFT induces heightened levels of volatility. those ﬁrm-days in the 95th percentile. The coefﬁcient doubles between the 80% and 95%. But higher frequencies are also of interest given that prices can ﬂuctuate dramatically throughout the day but end relatively unchanged. [Note: statistical analysis available soon. Panel B looks at HFT demanding liquidity activity. 60%. 40%.3. The OLS regression determines coefﬁcients based on the conditional mean. and in Panel C it is the percent where HFT supply liquidity. and HF Ti. but for the most extreme groups.t is the percent of shares for ﬁrm i on day t involving HFT. Instead of using the quintile regression approach as I did above. which is positive but less than the 80 % regression. Therefore. The regression I run is: 31 . Panel A . or baseline HFT participation increases with volatility. the quantile regression determines coefﬁcients based on a conditional observation level.t is the 15 minute realized return volatility for ﬁrm i on day t.volatility relationship across different market conditions. which is speciﬁcally addressed in Section 6. Next.t = α + HF Ti. 20%. Therefore.
80% in column (5) and 95% (high volatility) in column (6) levels.0940∗∗∗ (0.00429) 0.00176) (0.00135) 61014 (3) 40 % 0.0292∗∗∗ (0. and in Panel C it is the percent where HFT is supplying liquidity.00176) Constant 0.00191) 61014 -0.0193∗∗∗ 0.0136∗∗∗ -0.227∗∗∗ (0.t is the 15 minute realized return volatility for ﬁrm i on day t.0685∗∗∗ (0.101∗∗∗ (0.00164) Constant 0.00167) 0.05.Table 10: HFT .00367) 61014 (6) 95 % -0.0411∗∗∗ (0.00169) Observations 61014 0.0106∗∗∗ (0.00258) 61014 (5) 80 % 0.0226∗∗∗ (0.t = α + HF Ti. ∗∗∗ p < 0.00804 (0.00780) 0.0489∗∗∗ (0.0560∗∗∗ (0.227∗∗∗ (0.129∗∗∗ (0.0111∗∗∗ (0.01.0163∗∗∗ (0.00224) 0.0488∗∗∗ (0.0677∗∗∗ (0. ∗∗ p < 0.00876) 61014 32 .0313∗∗∗ (0.00297) 0.00150) 61014 0.0741∗∗∗ (0. Firm ﬁxed effects are used. 20% in column (2).00265) 0. The regression is performed at the 5% (low volatility) in column (1).00802) 0.0587∗∗∗ (0.0452∗∗∗ (0.0689∗∗∗ (0. in Panel B it is the percent where HFT is demanding liquidity.HFT .00136) 0.Supply Liquidity HFT Supply Percent -0. This table reports the results from running the quantile regression.00947) 0.0569∗∗∗ (0.t ∗ β1 + ϵi.t is the percent of shares for ﬁrm i on day t involving HFT.00420) 61014 0.HFT . 40% in column (3).128∗∗∗ (0. V olai.001 -0.t where V olai.0725∗∗∗ (0.0943∗∗∗ (0.00158) 0. 60% in column (4).00146) Observations 61014 61014 Standard Errors in parentheses ∗ p < 0.Volatility Relationship.144∗∗∗ (0.ALL HFT Percent Constant Observations 0.0178∗∗∗ (0.00174) 61014 (2) 20 % 0.HFT .0103) 61014 Panel B .00355) 0.0258∗∗∗ (0.00220) 0.140∗∗∗ (0.0509∗∗∗ (0.00118) 0.235∗∗∗ (0.00216) 0.00395) 61014 -0. which in Panel A includes trades having HFT in any capacity.00188) 61014 0.00234) 61014 -0. and HF Ti.00964) 61014 Panel C .00185) 61014 (4) 60 % 0.0291∗∗∗ (0.00248) 61014 0.0174∗∗∗ (0.Demand Liquidity HFT Demand Percent 0.00170) 0. (1) 5% Panel A .00236) (0.
t is the percent change in price for stock i during time t.40% of all observations and zero otherwise. similar to column (6). the interactive slope coefﬁcient tends simply to cancel out the baseline slope coefﬁcient in each column.t ∗ β9 + 40%Di.t variable. Column (8) Sell-Supply is high in the low and high return period and near zero in the middle events.t ∗ β6 + Reti.t ∗ Reti. There is no clear pattern in any of the columns. These results suggest that during large price declines HFTs do not make unusually large sell demands.HF Ti. 20%D is a dummy variable equal to one if the price decline (incline) for ﬁrm i in time t was in the smallest 20% .t ∗ β11 +80%Di.t ∗ β5 +95%Di.t . Reti.t ∗ β1 + 20%Di.t . As for the interactive terms.t ∗ β2 + 40%Di.t ∗ β7 + 5%Di. and 95%Di.t . Column (6) Buy-Supply is high in the low and high return periods. The table 11 shows the results.t ∗ β13 + ϵi. Like the price decline results though during large price inclines.t ∗ β4 + 80%Di.t ∗ Reti. 80%Di. The baseline dummy excluded from the regression is for return periods that are in the lowest 5%.t ∗ Reti.t ∗ β3 + 60%Di. The remaining explanatory variables are interactions between the different dummy ranges and the Reti. and Supply/Demand refers to HFTs role in the transaction. or Sell-Supply where each deﬁnes HF T as the percent of all trades that occur in the market that satisfy the criteria implied in the name. where the Buy/Sell refers to HFTs activity.t ∗ β10 + 60%Di. and near zero in the middle events.20% of all observations and zero otherwise.t ∗ Reti.t .t ∗ Reti. The positive return dummy results are also noisy. Column (5) Buy-Demand tends to decrease as the price incline increases. Sell-Demand.t dummy variables. Buy-Supply. column (1) Buy-Demand and column (2) Buy-Supply tend to increase with larger sized declines. Column (3) Sell-Demand and column (4) Sell-Supply do not show a noticeable pattern. but across the four columns as the price drop increases the coefﬁcient tends to cancel out the Ret coefﬁcient value.t ∗ Reti. The interactive terms also lack a clear pattern. HF T takes on one of four deﬁnitions: Buy-Demand. Focusing on the dummy variables.they appear noisy. and they do not stop providing liquidity to those who 33 .t is a dummy variable equal to one if the price decline (incline) for ﬁrm i in time t was in the smallest 5% . but this increase is non-monotonic. 5%Di.t = α + 5%Di. and the negative returns ﬁrst.t ∗ β12 + 95%Di. Column (7) Sell-Demand is large until the 95% Dummy.t ∗ β8 +20%Di. 60%Di.Similarly deﬁned are the 40%Di.
I regress the percent of shares in which HFT were involved on a dummy variable. in column (2) it is the percent of shares in stock i in which HFT was involved and was demanding liquidity. it is at elevated levels. 2008 . and zero otherwise. not only is it that the volatility is likely exogenous.] The above results still don’t overcome the likely endogeneity between HFT and volatility. and there was a high level of information uncertainty. 2008 and zero otherwise.. This was the week in which Lehman Brothers collapsed. Another time in which there was an identiﬁable exogenous shock to volatility was the week of September 15 . 2008 and the week of November 3. in column (2) it is the percent of shares in stock i in which HFT was involved and was demanding liquidity. the increase arises from HFT supplying liquidity in a larger fraction of shares. and they do not stop providing liquidity to those who are buying. a natural time to expect exogenous shocks to volatility is during quarterly ﬁrm earnings announcements. volatility spiked. The dependent variable in column (1) is the percent of shares in stock i in which HFT was involved. The difference is small. coming from news and not traders’ churning. Using OLS regression.are selling. For the quarterly earnings announcements. which is one for ﬁrm i if the observation is on the day of or the day after ﬁrm i reported its quarterly earnings. LehmanW eekDummy. Similarly with price inclines HFTs do not make unusually large buy demands. I regress the percent of shares in which HFT were involved on a dummy variable.September 19. 2008 (this week is chosen as it is sufﬁciently far away to reduce the autocorrelation impact of volatility. I therefore test whether there is a difference in HFT activity during the week of September 15. Like the quarterly earnings announcements. the week in September when Lehman failed and a randomly chosen week in November also shows that ﬁrms in the September week have statistically signiﬁcantly higher volatility. in column (3) it is the percent of shares in stock i in which HFT was involved and was supplying liquidity. which is one for all ﬁrms for observations on the dates September 15. The results in table 12 Panel A show that HFT activity increases with a shock to volatility. but not too far away as for there to have been a signiﬁcant change in HFTs strategies. but statistically significant. In the HFT sample dataset. To overcome this one must ﬁnd situations in which there are exogenous shocks to volatility. QuarterlyEADummy. [Note: statistical analysis available soon. Using OLS regression. 2008.September 19. Exogenous shocks to volatility typically come from new information entering the public domain. Thus. Thus. The dependent variable in column (1) is the percent of shares in stock i in which HFT was involved. in 34 . days on which ﬁrms announce their quarterly earnings have higher volatility than the average non-announcement day for that stock.
t ∗ β11 + 80%Di.4∗∗∗ 237.1∗∗∗ (35.00862) 0.β1 + 20%Di.166∗∗∗ 0.t ∗ β10 + 60%Di. where the Buy/Sell refers to HFTs activity.197∗∗∗ 0.00816) 0.01.202∗∗∗ (0. Sell-Demand.38) (39.0734∗∗∗ (0.001 (4) S-S 0.44) -362.00770) 40% Dummy 0.4∗∗∗ (35.51) (29.92) (29. 80%Di.85) 156.0805∗∗∗ (0.86) -228.9∗∗∗ -808.5∗∗∗ 400.00673) -0.36) (39.00799) 367468 0.84) -766.00745) 0.0150∗ (0.00812 -0.00888) (0.7∗∗∗ -581.t ∗ β9 + 40%Di.00709) Observations 464727 464727 464727 Adjusted R2 0.39) -36.4∗∗∗ 434.00901) (0.88) (30.107∗∗∗ 0. Reti.6∗∗∗ (33.00637) (0.t ∗ β12 + 95%Di.92 (46.34) -1.51) (29.388∗∗∗ (0.00853) 0.36) (39.38) Constant 0.68) (29.4∗∗∗ 287.0∗∗∗ 232.2∗∗∗ (34.82) 0.t ∗ β3 + 60%Di.00870) (0.t ∗ 35 Negative Returns (1) (2) (3) B-D B-S S-D 5 % Dummy 0.38) 5% Dummy*Ret 519.0646∗∗∗ (0.90) (44.00834) (0.139∗∗∗ (0. D or S.0395∗∗∗ 0.00793) 407.90) (44.0573∗∗∗ (0.00663) 464727 0.00631) (0.95) (44.3∗∗∗ (33.0146 (0.00470 (0.20% of all observations and zero otherwise.00585) (0.00968 (0.0946∗∗∗ 0.92) (44.228∗∗∗ 0.0576∗∗∗ 0.0838∗∗∗ (0.780 (37.2∗∗∗ (45.044 0.129∗∗∗ (0.t = α+5%Di.00870) 0.7∗∗∗ (47.00823) 9.238∗∗∗ (0.4∗∗∗ (37.4∗∗∗ (44.00720) 0.190∗∗∗ (0.33) -8. or Sell-Supply where each deﬁnes HF T as the percent of all trades that occur in the market that satisfy the criteria implied in the name.8∗∗∗ -228.t ∗ Reti.1∗∗∗ (34.5∗∗∗ -482.00822) -0.86) -518.36) (39.40) 80% Dummy*Ret 450.00734) 0.25) -301.t ∗ β4 + 80%Di.83) 118.7∗∗∗ (33.38) 95% Dummy*Ret 451. Similarly deﬁned are the 40%Di.2∗∗∗ (34.t .50 (44.00594) (0.6∗∗∗ -538.6∗∗∗ (36.087 (8) S-S 0.00550) (0.00620) (0.65) -702.t ∗ Reti. The ﬁrst letter. Firm ﬁxed effects are used.54) (29.797 (46.96) 181.0748∗∗∗ 0.59) 20% Dummy*Ret 557.084 (5) B-D 0.70 (46.8∗∗∗ (44.00831) 0.t ∗ β13 + ϵi.00750 0.218∗∗∗ 0.106 (46.00738) Ret -446.7∗∗∗ (36.00602) (0.00601) 367468 0.84) 0.t ∗ β5 + 95%Di.89) -514.00566) (0.95) -8.210∗∗∗ (0.47) (39.91) (44.00690) -125.171∗∗∗ (0.t variable.136∗∗∗ 0.t is a dummy variable equal to one if the price decline (incline) for ﬁrm i in time t was in the smallest 20% .66) -406.00756) 95% Dummy 0.7∗∗∗ (33.00801) (0.00895) 0. The remaining explanatory variables are interactions between the different dummy ranges and the Reti.t ∗ β6 + Reti.5∗∗∗ (46.00797) 20% Dummy 0. Table 11: HFT Behavior Around Different Size Price Changes This table shows the results to the regression: HF Ti.152∗∗∗ (0.17) -297.4∗∗∗ -506. stand for Demand or Supply liquidity.082 .00585) (0.152∗∗∗ (0.00847) 0.0235∗∗ (0.170∗∗∗ 0.00785) 60% Dummy 0. stand for Buy or Sell.136∗∗∗ 0.32) 0.00574) (0.30) (45.0956∗∗∗ (0. 60%Di.6∗∗∗ (36.65) 0.t ∗ Reti.00564) (0.9∗∗∗ 391.t .055 Positive Returns (6) (7) B-S S-D 0.t ∗ Reti.74) 40% Dummy*Ret 458. B or S.48) (45.242∗∗∗ (0.67) -403.7∗∗∗ (34. 5%Di.136∗∗∗ (0. Columns (1) to (4) are for negative returns .74∗∗ 54.t dummy variables.6∗∗ (36.t .084 0.00637) 0.t ∗ Reti.40% of all observations and zero otherwise.210∗∗∗ 0.26) (40.9∗∗∗ (34.00619) 491.139∗∗∗ (0.4∗∗∗ (34.00771) 367468 367468 0.00707) 0.00862) -0. ∗∗ p < 0.8∗∗∗ 507.71) -319.137∗∗∗ (0. and columns (5) to (8) are for positive returns.202∗∗∗ 0.7∗∗∗ 509.00791 (0.00839) 0.194∗∗∗ (0.1∗∗∗ 367.00642) 0.1∗∗∗ (34.114∗∗∗ 0.024 Standard errors in parentheses ∗ p < 0.t is the percent change in price for stock i during time t. The baseline dummy excluded from the regression is for return periods that are in the lowest 5%.05.1∗∗∗ (33.110∗∗∗ (0.t .00759) 80% Dummy 0.131∗∗∗ 0.2∗∗∗ 336.6∗∗∗ (36.53) 60% Dummy*Ret 437.049 0.00858) (0.t ∗ β8 + 20%Di.32) -357.t ∗ β2 + 40%Di.00648) 0.337∗∗∗ (0. and 95%Di.216∗∗∗ (0.0480∗∗∗ 0.191∗∗∗ (0.112∗∗∗ (0. and Supply/Demand refers to HFTs role in the transaction. Buy-Supply.0∗∗∗ (44.5∗∗∗ 95.00528) (0.8∗∗∗ (44.00655) (0.00710) 0.00855) (0.00655) 0.149∗∗∗ (0.82) 354. ∗∗∗ p < 0.942 (46.00624) (0.16) 196.63) (39.t ∗ Reti.85) -490.4∗∗∗ -609. the second letter. HF T takes on one of four deﬁnitions: Buy-Demand.5∗∗∗ (44.52) (29. 20%D is a dummy variable equal to one if the price decline (incline) for ﬁrm i in time t was in the smallest 5% .11) -14.t ∗ β7 + 5%Di.217∗∗∗ 0.4∗∗∗ -403.0885∗∗∗ (0.0817∗∗∗ 0.8∗∗∗ 239.41) -490.224∗∗∗ (0.
Nasdaq makes up 20% . For the Lehman Week increase in HFT. Thus. Given their trading amount a question of interest is how proﬁtable is their behavior. HFTs have been portrayed as making tens of billions of dollars from other investors. there will be many other trades that occur that the dataset does not include. I take all HFT buys and sells at their respective prices and calculate how much money was spent on purchases and received from sales. I can estimate the total proﬁtability of these 26 ﬁrms. but I cannot distinguish which HFT ﬁrm is buying and selling at a given time. I consider all HFT actions to come from one trader. Summing up the P rof it for each stock on a given day 36 . The HFT labeled trades come from many ﬁrms. HFTs regularly switch between being net long and net short throughout the day. the daily proﬁtability for each stock is calculated as: T ∑ t=1 P rof it = [1Sell ∗ P ricet ∗ Sharest − 1Buy ∗ P ricet ∗ Sharest ] + 1 Sharest T ∑ t=1 ∑T t=1 [P ricet ∗ Sharest ] . I can only provide an estimate of the proﬁtability of HFTs.2 Proﬁtability HFTs engage in a price reversal strategy and they make up a large portion of the market. With these considerations in mind. Also. 5. recall that the dataset only contains Nasdaq trades. Sharest is the number of shares exchanged in transaction t. 1Buy is similarly deﬁned for HFTs buying. Therefore. Due to the limitations of the data. the increase arises from HFT supplying liquidity and demanding liquidity in a larger fraction of shares. As many HFTs do not end the day with an exact net zero position in each stock I take any excess shares and assume they were traded at the mean price of that stock for that day. but at the end of the day they tend to hold very few shares.30% of all trades and so two out of every three trades are unobserved. P ricet is the price at which transaction t occurred.column (3) it is the percent of shares in stock i in which HFT was involved and was supplying liquidity. I circumvent these limitations by making estimates using the market behavior results from above to arrive at an overall annual proﬁtability of HFT. The results show that HFT activity increases with a shock to volatility. where 1Sell is a dummy indicator that equals one if HFTs sold a stock in transaction t and zero otherwise.
606 0.001 Panel B .Exogenous Volatility.545∗∗∗ (0.05. The dependent variable in column (1) is the percent of shares in stock i in which HFT was involved.ALL HFT . Lehman Failure (1) HFT . ∗∗∗ p < 0.743 0.402∗∗∗ 0.HFT .ALL 0. and zero otherwise. This table shows the results from two different approaches of trying to understand the impact volatility has on HFT activity.HFT .Table 12: HFT .0242) 1200 0.0103∗ (0. 2008 and zero otherwise.883 Lehman Week Dummy Constant Observations Adjusted R2 Standard errors in parentheses ∗ p < 0.0257) 1200 0.Demand HFT .00322) 0.00231) Constant 0. ∗∗ p < 0.730 (3) HFT . ∗∗∗ p < 0. In Panel A the explanatory variable. Quarterly Earnings (1) (2) (3) HFT .Exogenous Volatility Relationship.756∗∗∗ 0.00684∗ -0. LehmanW eekDummy is one for all ﬁrms for observations on the dates September 15.542∗∗∗ (0.01. In Panel B the explanatory variable. in column (2) it is the percent of shares in stock i in which HFT was involved and was demanding liquidity. QuarterlyEADummy which is one for ﬁrm i if the observation is on the day and next day on which ﬁrm i reported its quarterly earnings.00464) 0.390∗∗∗ (0.001 37 .01. Firm ﬁxed effects are used.00116 0.Demand 0.0125) Observations 4806 4806 4806 Adjusted R2 0.00311) (0. Panel A .783 Standard errors in parentheses ∗ p < 0.05.00435) 0.0155) (0.Supply 0.Supply Quarterly EA Dummy 0. in column (3) it is the percent of shares in stock i in which HFT was involved and was supplying liquidity.753∗∗∗ (0. ∗∗ p < 0.0142∗∗∗ (0.00288) (0.Exogenous Volatility. 2008 .0167) (0.0124∗∗∗ (0.September 19.840 (2) HFT .0130∗∗ (0.0179) 1200 0.
per day. HF T is calculated ˆ for each stock. I arrive at the annual estimated proﬁts of HFTs by: N T ] 1 ∑∑[ ˆ HF T i.0119239. where M arketCap is the log of the daily shares outstanding of company i multiplied by the closing price of company i.3 (million).0725419−M arket/Book ∗.t ∗ DV olumei.739. as a HFTr exchange with a HFTr will have a net zero proﬁt when considering HFTs proﬁt in the aggregate. Second. There is no adjustment made for transaction costs yet.156.6835[.non-HFT dollar volume traded ($151. Nasdaq is one of several venues where trades occur and on average makes up between 20 . a fraction of Compustat ﬁrms’ combined market capitalization is $17. and I multiply this value by the proﬁt per dollar traded by HFTs found above.3 (million).995 billion annually.000.110. winsorized at the 99th percentile.917. the 120 stocks have a combined market capitalization of $2.30% of trading activity.089. Thus.000106 2 i=1 t=1 ˆ HF T AnnualP rof it = (4) The .113. To account for these limitations in the data I carry out the following exercise: I use the non-endogenous regression coefﬁcient estimates in the HFT percent determinant estimation found in table 9 to estimate the percent of trades involving HFT and multiply it by the fraction of shares where HFTs are not trading with each other.346.000).1 from the 120 stocks in my sample on trades that occur on Nasdaq. the reason being that when HFT provide liquidity they re- 38 .results in the total HFT proﬁtability for that day. I multiply HF T by the dollar volume traded for each stock on each day in 2008 and 2009.574/$2. HFTs make $298. The above number substantially underestimates the actual proﬁtability of HFTs. and M arket/Book is the ratio of the M arketCap divided by the Compustat book value based on the the most recent ˆ preceding quarterly report. The result of this calculation is that HFTs gross proﬁt is approximately $ 2. ˆ HF T = . First. such costs will be relatively small. The result of this exercise is that on average.0308866+M arketCap∗. However.t ∗ .]. It is determined by taking the total proﬁt of HFTs from the 120 sample ﬁrms over the sample time period and divide it by the HFT .000106 value represents the proﬁt per dollar volume HFT traded with a non-HFT.589.
even after smoothing out the day to day ﬂuctuations. Using the data from Coughenour and Harris. and still less than a fourth as expensive as traditional market makers post-decimalization. On the other hand.00036) per dollar traded in medium stocks. From the above results.548. A rough estimate of the cost of trading is calculated by assuming that there is an equal number of shares demanded and supplied.00052) per dollar traded in small stocks.05 per trade. To try to understand what drives the changes in proﬁtability per day I look at the determinants for what stocks on different days are the most proﬁtable.0025 ($. I regress the proﬁtability on several potentially important variables. and $.t ∗ β5 + Depi. for example Nasdaq offers $. What is important isn’t the level of proﬁtability of HFT.00059) per dollar traded in large stocks. 39 .t ∗ β8.000106) per dollar traded. I compare how proﬁtable HFTs trades are per dollar traded as compared to other market makers. Thus. specialists before HFT and before decimalization (and after decimalization made $. these two values practically cancel themselves out. I run the following standardized regression: P rof iti. but only half of trades are demanding liquidity.469. $. HFTs are less than an eighth as expensive as pre-decimalization market makers.025 (. Figure 3 displays the time series of HFT proﬁtability per day.ceive a rebate from the exchange.25 per 100 shares for which trades take liquidity. Proﬁtability varies substantially from day to day.) If I assume the average stock price is $30. As the amount of liquidity demanded is slightly less than the liquidity supplied by HFT. the annual transaction cost for HFTs is $344.00292 ($. thus I can estimate each trade of 100 shares costs . but this is only for large volume traders like HFTs. the same ones used in the regression to determine HFT percent of the market.000894 ($. then using the total number of implied shares traded (including HFTr-to-HFTr transactions).t ∗ β6 + V oli. speciﬁcally specialists of NYSE stocks in 2000.20 per 100 shares for which traders provided liquidity. Nasdaq charges $. but what it is relative to the alternative.t ∗ β4 + N Vi. The graph is a ﬁve day-moving average of proﬁtability of HFT per day for the 120 ﬁrms in the dataset.t ∗ β7 + ACi. HFT make on average 1/100th of a penny ($. Hasbrouck and Soﬁanos (1993) and Coughenour and Harris study the trading activity and proﬁtability of the NYSE specialists.t ∗ β1 + M Ci ∗ β2 + M Bi ∗ β3 + N Ti.t = α + Hi. From this perspective. a market consisting of non-HFT market makers.
For any end-ofday imbalance the required number of shares are assumed traded at the average share price for the day in order to end the day with a net zero position in each stock. Proﬁtability is calculated by aggregating all HFT for a given stock on a given day and comparing the cost of shares bought and the revenue from shares sold.Figure 3: Time Series of HFT Proﬁtability Per Day. 3000000 −1000000 0 01 Jan 08 $ Profit Per Day 1000000 2000000 01 Jul 08 01 Jan 09 sas_date 01 Jul 09 01 Jan 10 40 . The ﬁgure shows the 5-day moving average proﬁtability for all trading days in 2008 and 2009 for trades in the HFT data set.
Again. in the third column it is the number of HFT shares traded for stock i on day t. 41 .where all variables are deﬁned as before. The Proﬁt per HFT Share Traded regression has no statistically or economically signiﬁcant variables and has a negative r-squared. The third regression. shows the largest coefﬁcient magnitude of 0. M arketCap. Each analysis uses different techniques to study the relationship between HFT and each type of market quality. Next. has two coefﬁcients that are statistically signiﬁcant. The second and third regression decompose the parts of the ﬁrst regression’s dependent variable. price discovery. HFTs are proﬁtable.622. I investigate the role HFT plays in the demand and supply of liquidity.098. 6 Market Quality The following section analyzes HFT impact on market quality. In addition. Also. that HFT tends to occur more in large stocks with relatively low volume with narrow spreads and depth. and that the proﬁtability is related to volatility. which also has a positive relationship with the dependent variable. and volatility. The V olatility measure has a large positive economic impact and is highly statistically signiﬁcant. making approximately $3 billion a year. and the V olatility coefﬁcient. the reported coefﬁcients have been standardized so that the coefﬁcient value represents a one standard deviation movement in a particular variable’s impact on P rof it. but on a dollar traded basis they are signiﬁcantly less expensive than traditional market makers. In the ﬁrst column P rof it is deﬁned as the proﬁt per HFT share traded averaged over stock i on day t. in the second column it is the amount of money HFT made for stock i on day t. HFT shares traded. is positive with a coefﬁcient of 0. Autocorrelation has a smaller coefﬁcient and is negative. there is little change in HFT activity during extreme market conditions and HFT slightly increases with exogenous shocks to volatility. The results are displayed in table 13 and are standardized coefﬁcients. has three statistically signiﬁcant and economically signiﬁcant variables. implying the less predictable price movements in a stock the more proﬁtable is that stock for HFT. Market quality refers to liquidity. The second regression. This section has shown that HFTs engage in a price reversal trading strategy. Autocorrelation and V olatility. the AverageDepth coefﬁcient is positive and has a coefﬁcient of 0. and the dependent variable P rof it takes on three different deﬁnitions.21. with the dependent variable as proﬁts.
24) 0. Proﬁt per HFT Share Traded -0.90) 0.025 (-0.36) -0.43) -0.024 (-0. in the second column it is the amount of money HFT made for each stock on each day.16) 0.036 (-1.67) -0.16) 0.42) 0.014 (0.001 (-3.033 (-0.078∗ (-1.Table 13: Determinants of HFT Proﬁts Per Stock Per Day The dependent variable for the ﬁrst column is deﬁned as the proﬁt per HFT share traded averaged over each stock i on day t.359∗∗∗ (8. in the third column it is the number of HFT shares traded.70) -0.532 42 . ∗∗ p < 0.41) (1.004 (-0.038 (0.69) ∗∗∗ HFT Percent Market Cap.087 (-1.20) 0.05.76) -0.01) 0.96) Observations 360 Adjusted R2 -0.66) 590 0.04) -0.058 (-0.97) -0.009 (-0.027 (0.111 HFT Shares Traded 0.29) Proﬁts -0.019 (0.210∗∗∗ (4.014 Standardized beta coefﬁcients.06) -0.33) 0.015 (-0.67) -0.62) 0. ∗∗∗ p < 0.622∗∗∗ (20.40) -0.015 (-0. t statistics in parentheses ∗ p < 0.07) 0.000 (0.23) -0.003 (-0.013 (0.01.098∗∗∗ (3.45) 590 0. Market / Book $ of Non HFT Volume Average Spread Average Depth Volatility Autocorrelation # of Non HFT Trades Constant (0.19) 0.008 (-0.031 (0.062 (1.35) -0.073 (-1.86) -0.059 (-0.
in dollar terms.4% of all trades. and this clustering is stronger for HFTs than for non-HFTs. HFT in two of the three categories are the most engaged in these transactions. HFTs are not as involved as nonHFTs. This section will examine how HFT initiated trades tend to behave compared to non-HFT trades. but tend to demand the most. It is this size trade that Chakravarty (2001) and Barclay and Warner (1993) ﬁnd to have the largest impact on stock prices and thus are the size of trades informed investors tend to use. This suggest that HFT are liquidity takers in large trades and liquidity providers 43 . but they make up a signiﬁcantly smaller portion of trades at the opening and close of the trading day. There appears to be clustering in trades. In the $30.6. then I examine the role of HFT in supplying liquidity.000 to $14. Trades that either proceed a HFTr trade or follow a HFTr trade tend to occur more quickly than those proceeding or following a non-HFTr.000 to $4. As trade size increases. In this section I look at the descriptive statistics of HFT demanding liquidity. and the same is true for sales.1 HFT Liquidity Liquidity supply and demand in the microstructure literature refers to which side of a trade had a limit order in place that was executed and which side of the transaction entered the marketable order. The ﬁrst column of Table 14 reports the fraction of trading volume for different combinations of HFTs and non-HFTs as liquidity providers and takers.1. it is much more likely the next trade will also be a buy. this is consistent with the previous results that show HFTs tend to trade more in stocks with large market caps. except for when they are demanding liquidity. Most trades occur in the value range of $1. HFTs liquidity demand is quite consistent across the day. For small trades. and this is true regardless of the size of the ﬁrm. which typically have stock prices in the double digits. 6. HFTs tend to demand liquidity in similar dollar size trades as do non HFTs. Finally. and with different HFTs and non-HFTs liquidity providers and demanders.999. whereby if a previous trade is a buy. those worth less than $1.999 category.000 plus category of trades. Table 14 looks at the percent of all transactions for different size trades. the time between trade decreases. HFTs share of trades engaged in falls in the $5. The side with the limit order is the liquidity supplier.000.1 HFT Liquidity Demand The results in table 4 show that liquidity is demanded by HFTs in 50. HFTs provide the least amount of liquidity. ﬁnally I analyze how much liquidity is provided in the quotes and the order book by HFT. and the marketable order side is the liquidity taker.
793 N 245. where the ﬁrst letter represents the liquidity demander and the second the liquidity supplier.108.473.7% 23.793 in small shares. Let Hs be the number of HFT liquidity suppliers.047 940.0% 100.609 3.8% 833.0% 25.5% 27. This table reports dollar-volume participation by HFTs and non-HFTs in ﬁve dollar-trade size categories.0% 100.2% 25.6% 26. H represents a HFTr.9% 32.5% 23.634 308. A concern among many is that if HFTs use similar trading strategies.0% 100. NH. The second letter in the column labels represents the passive party of a trade.883 27.0% 100. Hd be the number of HFT liquidity demanders.3% 26.5% 25. The ﬁrst letter in the column labels represents the liquidity seeking side of a trade.0% 3. N represents a non-HFTr.3% 28.864 25. To determine whether HFTs strategies are more correlated than those of non-HFTs I examine the frequency at which HFTs trade with each other and compare it to a benchmark model used in Chaboud et al.supplier) under the assumption that traders’ activities are random and independent.4% 24.999 15. NN. HH. Nd be the number of non-HFT liquidity demanders. As above. and the probability the +H 44 .7% 25.Table 14: HFT Volume by Trade-size Category.8% 23.8% 30.4% 757.108. Ns be the number of non-HFT liquidity suppliers. HN.999 5.000 1. Dollar Size Categories 0-1.401 1.0% 713. there are four types of trades. Then I can compare the actual occurrence of different trades to the predicted amount.869 100.7% 24.000 + Total N Type of Liquidity Taker and Liquidity Supplier HH HN NH NN Total 21.000-29. (2009) that produces theoretical probabilities of different types of trades (demander .2% 19.000-14.000-4. The probability that a HFTr will provide liquidity is then αs = P rob(HF T − supply) = NsHs s .8% 803.0% 22. which is consistent with the theory that HFTs are concerned with informed traders in big trades.999 30.1% 17. they will exacerbate market movements.0% 100.102 141.3% 26.5% 21.177 25.
20 are in the 50%’s. Starting with %DaysR < 1. The ﬁrst letter in the rows for Panel A and B is who is demanding liquidity at Time t-1. Seeing a HFTr demand liquidity in time t − 1 followed by a HFTr demanding liquidity P rob(N N ) P rob(N H) 45 . all but two stocks have the ratio R < 1 over 50% of the time. The interpretation of this result is that HFTs engage in a more diverse variety of strategies than non-HFTs. provides evidence on the clustering of HFT in trade sequences. Panel A reports the unconditional frequency of observing HFTr and non-HFTr trades. and 6 always have R < 1. As a result. The column Std. 17 are in the 80%’s. Of the 120 ﬁrms. ˆ ˆ The proxy used for R in the data is RN = V ol(N N ) and RH = V ol(HN ) . Overall 79% of days have R < 1. whereby the diverse strategies result in one HFTr deciding to buy and another HFTr deciding to sell simultaneously. or that HFT trade with non-HFT more than expected. The column R shows the results for each stock of the average RH per day. similar to that in Biais. 12 are in the 70%’s. regardless of the number of non-HFTs or HFTs. P rob(HN ) = (αd )(1 − αs ). 88 have an average R less than 1. The probability that a HFTr will demand liquidity is αd = P rob(HF T − demand) = NdHd d . In the table. 37 are in the 90%’s. the follow fraction holds: P rob(N N ) ≡ P rob(HN ) . R ≡ RH will equal one as RN non-HFT will take liquidity from other non-HFT in the same proportion as HFTs take liquidity from other HFTs. the ratio of ratios. if R > 1 then it is the case that HFTs trade with each other less than expected. [Note: statistical analysis available soon. Table 16. 26 are in the 60%’s. is the standard deviation of the R RN ratio for that stock over time. if R = 1. it must be that HFTs and non-HFTs trade with each other as much as expected when there trading strategies are equally correlated.Dev. and the probability +H the liquidity is demanded by a non-HFTr is 1 − αd . The probabilities of a speciﬁc demander and supplier can be calculated: P rob(HH) = (αd )(αs ). the next table examines the conditional frequency and occurrence of different types of trades. This suggests that HFTs trade with each other more than expected or that HFTs trade with non-HFTs less than expected. When non-HFTs are greater than HFTs then P rob(N N ) > P rob(N H) and P rob(HN ) > P rob(HH).] Table 14 analyzed the frequency of different types of trades. and Spatt (1995) and Hendershott and Riordan (2009). However. H stands for HFTr and N stands for non-HFTr. Hillion. The second letter in these two panels is who is demanding liquidity at time t. V ol(N H) V ol(HH) Table 15 shows the results. P rob(N N ) = (1 − αd )(1 − αs ). The column %DaysR < 1 is the fraction of days in which R < 1.liquidity is supplied by a non-HFTr is 1 − αs . Let RN ≡ P rob(N H) P rob(HH) be the non-HFTr demanding liquidity ratio and RH ≡ P rob(HN ) be the P rob(HH) HFTr demanding liquidity ratio. Therefore. P rob(N H) = (1 − αd )(αs ).
71 Std.48 0.63 0.94 1.68 0.82 0.87 0.77 0.15 40.79 1.14 1.79 1.71 0.51 0.58 0.66 0.62 5.15 0.96 0.00 0.29 1.53 0.99 0. 46 Symbol AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI Total R 0.53 0.93 0.47 0.23 10.94 0.93 0.03 0.41 0. 0.64 0.66 0.96 0.11 0.81 1.47 1.63 0.76 1.66 0.68 1.01 0. This table reports the mean per stock of the daily ratio R = RH/RN where ˆ ˆ RN = V ol(N N ) and RH = V ol(HN ) .59 0.72 0.81 0.07 % Days R < 1 1.77 0.20 1.98 0.98 0.15 1.22 0.23 0.73 1.29 0.99 0.57 0.98 0.12 0.86 0.82 0.59 0.14 0.63 0.29 0.15 0.90 0.93 0.84 0.13 0.82 0.13 0.68 1.77 20.54 0.69 0.63 0.57 0.15 0.12 1.66 0.46 0.95 0.85 0.96 0.91 0.62 0.37 0.91 2.13 0.55 0.83 0.58 0.12 0.98 3.66 0.61 0.92 0.53 77.94 0.19 0.80 0.62 0.68 0.13 0.69 0.74 0.35 4.77 0.90 0.71 0.08 0.13 0.76 1.77 0.99 0.69 0.83 1.19 0.54 0.40 0.60 0.32 0.67 0.72 1.Table 15: Diversity of HFTs Strategies.74 0.78 0.60 1.69 0.57 0.02 1.63 0.64 0.17 0.95 0.65 0.14 0. The results are for the full sample period.22 0.13 0.12 0.48 0.76 5.89 5.97 Std.24 0.58 0.57 0.98 0.36 0.13 0.64 1.95 0.64 1.85 0.99 0.60 0.38 1.44 0.53 0.72 0.59 0.72 8.05 1.79 Symbol CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE ESRX EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG R 0.13 0.56 0.77 0.00 0.94 0.21 0.65 0.77 0.35 0.36 1.69 0.01 0.31 2.17 0.85 0.65 0.49 0.67 0.60 0.14 % Days R < 1 0.70 0.53 0.63 0.99 1.56 0.62 0.98 0. is the standard deviation of the V ol(N H) V ol(HH) daily ratio R for that particular stock over time. 0.33 0.73 0.56 0. Dev.60 19.83 0.74 0.91 0.13 0.12 30.88 0.64 Std.22 0.44 0.32 1.38 2.29 0.67 0.08 0.60 0.92 3.36 0.85 0.52 0.05 0.35 6.81 0.88 0.71 0.12 0.75 0.84 0.12 0.96 0.94 0.57 0.13 0.62 0.88 0.71 0.Dev.66 0.60 0.60 0.72 0.17 0.74 0.99 0.99 0.61 .98 0.18 0.96 0.74 0.95 0.06 0.48 0.00 0.91 0.70 0.14 0.81 0.85 0.55 1.13 0.26 0.35 0.87 0.88 0.15 0.54 0.01 0.81 0.60 0.88 0.67 0.59 1.97 0.87 0.13 0.58 0.55 1.68 0.26 0.81 0.06 0.54 0.98 0.16 0.97 0.96 0.17 2.93 0.99 0.12 2.81 0.64 0.99 0.08 0.70 0.12 0.60 1.26 0.58 0.58 1.93 0.83 0.98 0.69 0.96 0.34 1.00 0.74 1.88 0.17 1.53 0.45 0. The column %DaysR < 1 is the fraction of days in which R < 1 for that stock.19 Symbol JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NC NSR NUS NXTM PBH PFE PG PNC PNY PPD PTP RIGL ROC ROCK ROG RVI SF SFG SJW SWN R 0.56 0.66 0.93 1.17 0.68 0.63 0.87 0.14 1.99 0.67 0.27 0. 0. Dev.87 0.82 0.86 0.16 0.17 0.13 0.71 1.95 0.00 0.24 3.76 0.00 0.46 0. Dev.52 0.80 0.36 0. Std.71 0.80 0.52 2.66 0.74 0.27 0.48 1.92 0.13 0.82 0.04 2.73 0.72 1.64 5.80 1.15 0.57 0.99 0.68 0.94 0.97 0.82 0.14 0.97 0.60 0.59 0.98 0.85 0.88 % Days R < 1 0.
in time t is as common as seeing any other time t − 1. the columns are whether the liquidity taker is buying (B) or selling (S). and the larger the dollar size of a trade the higher the likelihood the next trade will be large. In addition. The ﬁrst 47 . and when a HFTr is the liquidity demander at t − 1.000 . All times reported are in seconds. Panel B reports the conditional frequency of observing HFTr and non-HFTr initiated trades after observing trades of other participants. The results show that trades of size and type of liquidity demander are highly dependent on the previous trade type. HFTs tend to trade with HFTs. regardless of what type of trader was demanding the liquidity.000. or a nonHFTr. in both buying and selling. and the same with sales. 1 represents a trade of size $0 -$999. either H for HFTr or N for non-HFTr. In Panel B.000 . two HFTr liquidity demanding trades.999. regardless of who demands liquidity at time t. Panel A reports the average amount of time between two trades. If all HFTs were market makers and non-HFTs were not.999. not time.$4. The rows represent the type of trader taking liquidity at time t − 1. whereby market makers trade with each other to eliminate inventory imbalances. The second letter represents what the liquidity taker is doing in the time t trade. are more rapidly executed.000 . In column and row headings t indexes trades.$29. for HFT demanders than it is for non-HFT demanders. The ﬁrst letter represents what the liquidity taker is doing in the time t−1 trade. the rows are further partitioned based on the size of the trade. The clustering affect is stronger. The next set of results regarding type of trader initiated trading looks at the time between trades. Panel C provides conditional probabilities based on the previous trade’s size and type of trader. 3 represents a trade of size $5. and two non-HFTr liquidity demanding trades. and 5 represents a trade of size greater than $30. The columns identify who was the liquidity demander at time t (H or N) and is further partitioned along the size categories discussed above. 2 represents a trade of size $1. Table 17 reports the average time between trades dependent on different trade characteristics. and between a trade where the t − 1 trade was initiated by a trader who was a HFTr. Panel B provides the average amount of time between two different trade orderings and total dollar-volume and per trade dollar-volume categories. measured by the dollar size of shares exchanged in the t-1 trade. 4 represents a trade of size $15. The results suggest that one tends to see liquidity demanders purchase shares follow a previous trade of a liquidity demander purchasing shares.999.$14. Both trades when the liquidity demander is a HFTr at both t − 1 and t. t sequence. then it appears that about a quarter of trades are ”hot potato” rebalancing as discussed by Lyons (1997).
8% 11. The second letter in these two panels is who is demanding liquidity at time t.6% 6.7% 16.6% 0.1% 42.3% 8.0% 100.7% 0.3% 0.0% 100.7% 23.5% 1.5% 4.0% 14.0% Panel C lag LD Size H1 H2 H3 H4 H5 N1 N2 N3 N4 N5 Total H1 % 25.5% 43.3% 0.2% 1.7% 3.9% 5.7% 6.1% 18. Panel A reports the unconditional frequency of observing HFTr and non-HFTr initiated trades.Table 16: Trade Frequency Conditional on Previous Trade.1% 48 2.0% 15.0% 4.5% 17.7% 7.7% 1.7% 8.7% 0.9% 0.0% 12.2% Total % 100.3% 35.5% 2.8% 12.4% 3.9% 2.3% 23.8% 23.1% 3.8% 1.4% 3.9% 43.0% 3.4% 0.1% 41.8% 2.6% 5.9% 4.2% 3.0% 100.3% 7.0% 100.5% 1. Panel C provides conditional probabilities based on the previous trade’s size and participant.3% 2.5% 0. Panel B reports the conditional frequency of observing HFTr and non-HFTr trades after observing trades of other participants.4% 11.0% 100.7% 6.8% 1.0% 2.2% 100.9% 1.3% 2.6% 3.6% 6.9% 41.6% 5. In column and row headings t index trades.6% 18.9% 9.7% 3.5% 57.0% 100.5% 7.9% 6.5% 3.9% H3 % 13.3% 44.5% 7.0% 100.2% 18.2% 17.8% 46.9% 5.6% 9.0% 5.0% 100.6% 0.0% 7.8% 1.0% 7.8% 7.9% 2.4% H4 % H5 % LD Size N1 % 7.9% 43.2% N2 % 10.0% 100.9% 23. Panel A T-1 Type and T Type HH HN NH NN Total % 24.9% 6.1% 23.3% 0.0% 34.6% 14.4% 25.6% 17.3% 14.3% .3% 27.7% 1.2% 23. The ﬁrst letter in the rows for Panel A and B represents who is demanding liquidity at Time t − 1.0% 100.1% 17.8% 26.9% 43.5% 1.9% 5.3% 0.0% 100.0% 100.7% H2 % 32.0% 2.1% 7.5% 44.7% 1.0% 100.0% 100.8% 6.5% 2.8% 0.8% 18.2% 58.2% 1.6% N3 % 5.0% 100.7% 29.5% 4.9% 42.3% 4.4% 3.4% 4.8% N5 % 0.8% 0.0% Panel B T-1 Type and T Type HH HN NH NN Total T-1 Buy or Sell and T Buy or Sell BB BS SB SS Total % % % % % 44.9% N4 % 0.8% 18.5% 39.5% 4.6% 0.1% 2.3% 1.
This result is evidence that HFTs actively monitor the market for liquidity. usually the shorter the time before another trade occurs. 2010 and put them in to ten second bins based on the time of day they occurred. regardless of the day. if its between $1. Then. Finally.000 and $4.the time between trades decreases. The last two columns are similar except that the columns are distinguished based on the time it takes when the time t − 1 liquidity taker is a certain type (H or N). NH.999 its a 2. During the day the trading ratios are quite stable. This is the opposite of what Hendershott and Riordan (2009) ﬁnd. If HFTs do try and end the day with a near net zero position in stocks then they should wind down their trading before the end of the trading day in order to prevent getting stuck with assets they do not want to hold overnight. where the columns are separated based on the time t liquidity taker. at the beginning of the trading day HFTs will have few positions in which they are trying to maintain a near net zero position in and so trading should be less prevalent. I take all trades that occur on February 22. except at the beginning and end of the day.999 its a 3.M. Rows S1 through L5 represent different types of stocks.February 26. The ﬁrst character. they see that small orders for AT tend to execute faster than large orders. The four different patterns. If the trade was less than $1000 then it is a 1. the larger the trade. Finally. Figure 4 shows the make up of different types of trades throughout the day.000 and $29. S. HN. whether looking at time t − 1 or time t. NN refer to the type of liquidity demander (ﬁrst letter) and liquidity supplier (second letter). Also. I split them into the types of trades based on who was supplying liquidity and who was demanding liquidity and calculate the percent of each type of transaction per time period bin. or L.000 it is a 5.000 and $14. Similarly. and at time t there is a liquidity taker of H or N.2 Billion. if between $15. if between $5. HH. The ﬁgure is stacked so that each time period sums to one. represents the dollar volume traded in a given stock on a given day. within each dollar volume category. but that they focus their trading strategy around order imbalances from large trades. with S being for trades in small stocks with total dollar volume under $800 Million. and L for large stocks with dollar volume greater than $1.two columns in Panel B are for some trade type at time t − 1. 2010 . for most of the different categories. the number 1 through 5 represents the size of the particular trade. M for medium stocks with dollar volume between $800 Million and $1. I examine the intraday pattern of HFT supply and demand of liquidity. The results suggest that as greater dollar volume is traded .999 its a 4. The second character. HFTs tend to trade more rapidly.2 Billion. To analyze this I create a time series of the type of traders throughout the day. During these periods HFTs tend to trade with each other much less frequently and 49 . and if its greater than $30.
Panel B provides the average amount of time between two different trade orderings and trade-size categories (refer to the previous table for the different trade-size categories) The ﬁrst two columns in Panel B is for some trade type at t − 1 and at time t there is a liquidity taker of H or N.081 Panel B Time t Liquidity Taker Time t-1 Liquidity Taker HFT Non-HFT HFT Non-HFT S1 30.620 10.354 2.Table 17: Average Time Between Trades. Panel A reports the average amount of time between two trades.135 1.149 L4 0.098 1.576 4.136 0.851 1.760 L5 0.183 1.572 5.842 1.067 3.095 1.746 25.290 L1 0.572 M3 1.219 1.602 0.841 1. Panel A HFT Non-HFT Unconditional Time Between Trades 3.065 M1 0.668 0.513 S4 3.742 4.582 12.988 1.064 1.273 M5 1.119 0.696 9. where the columns are separated based on the time t liquidity taker.426 3.351 5.447 0. two HFTr liquidity demanding trades.989 0. and two non-HFTr liquidity demanding trades.435 1. The last two columns are similar except that the columns are distinguished based on the time it takes when the time t − 1 liquidity taker is a certain type (H or N).434 M4 1.780 0.667 Time Between Trades of Same Type of Trader 8.538 1.266 0. and between a trade where the initial trade had a HFTr.729 0.118 L2 1.134 S3 5.449 18.578 7.384 35.399 1.112 1.657 4.899 1.378 M2 1.166 0.056 S5 4.833 0.069 5.843 S2 9.746 1.065 L3 0.724 0.485 6.998 1. or a non-HFTr liquidity demander. All values are in seconds.639 50 .746 1.
HFTs also often supply the inside quotes throughout the day. 1 . I show that. Figure 4: Type of Liquidity Providers / Takers throughout the day.4% of all trades. HFTs provide the inside quotes.4 . Finally.2 HFT Liquidity Supply This section analyzes HFT and the supply of liquidity. I examine what the additional price impact would be on stocks if HFTs were not part of the order book. The ﬁgure shows the make up of traders throughout the day.6 . which shows the importance of HFT in creating liquid markets. It shows that HFTs tend to reduce their trading activity at the opening and closing of the trading day. Figure 5 also looks at trades throughout the day. I then consider the determinants that inﬂuence which stocks.1. and on what days. the ﬁrst letter is the liquidity taker. but only charts the percent of dollar volume in which HFTs are demanding liquidity (top graph) and supplying liquidity (bottom graph). Again this shows that HFTs signiﬁcantly reduce both their supply and their demand for liquidity at the start and end of trading hours. I focus on the amount of liquidity HFTs supply.8 11:6 12:30 1:53 Time of Day HH Transaction NH Transaction 3:16 HN Transaction NN Transaction 4:40 6. beyond supplying liquidity in 51. This is consistent with the scenario of HFTs trying to end the day near net-zero in their equity positions. There is a sizeable impact. 51 .HFTs tend to initiate fewer trades and to provide liquidity in fewer trades.2 9:43 Stacked Percent of Trades . the second letter is the liquidity provider.
Figure 5: HFT Liquidity Demander or Supplier throughout the day. The graph shows the make up of HFT throughout the day. The ﬁrst graphs shows the HFT demand for liquidity throughout the day. The second graph shows the HFT supply of liquidity throughout the day.
HFT Percent of Liquidity Demanded .4 .5
12:30 1:53 Time of Day
HFT Percent of Liquidity Supplied .4 .45 .5 .55
12:30 1:53 Time of Day
18.104.22.168 HFT Time at Inside Quotes To begin analyzing HFT role in providing liquidity in the stock market I look at the amount of time HFTs supply the inside bid or ask compared to non-HFTs. For each stock, on each day, I take the number of minutes HFTs are providing the inside bid or ask and, either: (a) subtract the number of minutes non-HFTs are providing the best inside bid or ask (ties are dropped), these are the “Minutes” results, or (b), divide this value by the total number of minutes were HFTs and non-HFTs did not have the same inside quotes, these are the “Percent” results. The results are shown in table 18. Table 18 looks at, for the 120 sample stocks, whether HFTs provide more liquidity than non-HFTs by considering how often they are providing the inside quote (bid or ask). The manner in which the metric is constructed results in there being a total of 780 minutes ( 2*60*6.5) that a HFT could potentially be providing the inside quote. This is twice as many minutes then what actually occur during the trading day. The table is separated into two categories - the category “HFT -” displays statistics for when HFTs provides fewer inside quotes than the non-HFT for a stock on a given day; category “HFT +” displays statistics for when HFTs provide more frequently the inside quote for a particular stock on a given day. The reason to separate out the two types is that it may be that in some stocks HFTs do not actively try and provide inside quotes, thus just taking the average across all stocks would underweight the liquidity they do provide in the stocks they are actively place competitive quotes. Table 18 has three panels, and within each panel either a category called “Minutes” or “Percent.” Panel A considers quotes for all stocks; Panel B considers quotes for stocks on days they are below their average spread; and Panel C considers quotes for stocks on days they are above their average spread. The Minutes results display the number of minutes HFTs provide the inside quotes more than non-HFTs through the following calculation: sum the number of minutes HFTs provide the best bid or ask, subtract the number of minutes non-HFTs are providing the best inside bid or ask, and drop ties. A problem with this is that since the ties will vary across days and stocks, the Minutes approach does not necessarily capture the frequency that HFTs provide better inside quotes than non-HFTs. The Percent results avoids this issue by dividing the number of minutes HFTs provide the best quotes by the total number of minutes where HFTs and non-HFTs did not have the same inside quotes. There does not appear to be a signiﬁcant difference between the stocks in which HFTs decide to place aggressive bid/ask orders and those in which it does not. The very low value for the mean of Net - by itself may imply that the HFTs do not attempt to match or out-price the quotes of non-HFTs for some stocks.
But looking at the “Percent” data, shows that the Net - and Net + results are about equally distant from .5. Thus, the “Minutes” Net - results must be biased downwards as a result of a large number of periods where HFTs and non-HFTs provide the same prices. Looking at the Panel A - Percent results, on average HFTs provide the best inside quotes 45% of the time, a signiﬁcant portion of the trading day. This suggests that HFTs act as market makers and are competitive quote suppliers. Panel B and C divide the stocks into those that are offering higher spreads than average and those offering lower spreads than average. Panel B reports the low spread stock days, Panel C the high spread stock days. The results between the two subsets do not differ much from one another. The average time HFTs offer the best quotes is slightly higher when spreads are high at 71.7% compared to 70.1% when it is low. Also, HFTs provide the best quotes more often than non-HFTs slightly more often when spreads are high, doing so 46% of the time as opposed to 45.1% when spreads are low. This is consistent with HFTs attempting to capture liquidity supply proﬁts as found in Foucault and Menkveld (2008) and Hendershott and Riordan (2009) make/take liquidity cycle, but as the difference is small it does not provide much support for it. 22.214.171.124 HFT Time at Inside Quotes Determinants Table 18 shows that HFTs produce the inside quotes frequently, but not as often as non-HFT. I perform an OLS regression similar to that found in table 9 to understand what determinants are related to which stocks and days HFT prduces the best quotes. Table 19 shows the results. It is very similar to table 9, with all variables being deﬁned exactly the same as before except the dependent variable. The regression is: Li,t = α + M Ci ∗ β1 + M Bi ∗ β2 + N Ti,t ∗ β3 + N Vi,t ∗ β4 + Depi,t ∗ β5 + V oli,t ∗ β6 + ACi,t ∗ β7 , where the variables and subscripts are deﬁned as above, and the dependent variable, Li,t is the percent of the time for which HFTs provide the best inside quotes compared to all times when HFTs and non-HFTs quotes differ. The coefﬁcients reported, like those in table 9, are standardized beta coefﬁcients which allows for an easy way to decide which determinants are more important. The results suggest there are several explanatory variables that matter, all except Autocorrelation are statistically signiﬁcant, and all except AverageDepth have coefﬁcient magnitudes greater than .16. M arketCap. and #of N onHF T T rades 54
Table 18: HFT Time at Best Quotes. This table reports the number of minutes HFTs are at the best bid or ask compared to non-HFTs. The remainder of time both HFTs and non-HFTs are both at the best quotes is not considered. Panel A is for all stocks at all times, Panel B is for days when the spread is below average for that stock, Panel C is for days when the spread is above average.
Panel A All -Minutes HFT mean min Net -255.6 -779.7 Net + 65.8 0.1 Average -124.9 -779.7 -Percent Net 0.281 Net + 0.709 Average 0.455
p25 -368.5 14.3 -223.5
p50 -190.3 55.1 -41.2
p75 -75.8 94.0 33.1
max -0.4 343.0 343.0
N 353.0 242.0 595.0
0.000 0.502 0.000
0.162 0.587 0.272
0.307 0.708 0.446
0.410 0.802 0.668
0.499 0.964 0.964
353.000 242.000 595.000
Panel B Low Spread -Minutes Net -243.1 -779.7 -367.0 Net + 67.7 0.1 17.5 Average -118.5 -779.7 -205.9 -Percent Net 0.283 Net + 0.701 Average 0.451
-183.0 58.6 -40.4
-70.9 94.0 39.8
-0.4 319.8 319.8
187.0 125.0 312.0
0.000 0.509 0.000
0.162 0.587 0.267
0.309 0.704 0.435
0.404 0.786 0.650
0.498 0.960 0.960
187.000 125.000 312.000
Panel C High Spread -Minutes Net -269.8 -779.1 -401.2 Net + 63.6 1.1 13.3 Average -131.9 -779.1 -227.1 -Percent Net 0.278 Net + 0.717 Average 0.460
-204.0 50.4 -41.2
-84.9 91.5 28.9
-1.1 343.0 343.0
166.0 117.0 283.0
0.000 0.502 0.000
0.158 0.593 0.274
0.296 0.714 0.455
0.415 0.814 0.680
0.499 0.964 0.964
166.000 117.000 283.000
with M arketCap. t statistics in parentheses ∗ p < 0. being the most important determinant of HFTs providing the best quotes. Table 19: Determinants of HFT Percent of Liquidity Supplying The dependent variable is the ratio of number of minutes HFTs provides the inside bid or ask divided by the total number of minutes of when the inside bid and ask differ between HFTs and non-HFTs.18) 0.14) -0. Market Cap.82) -0.654∗∗∗ (15.90) -0. the analysis on liquidity has been by looking at the best inside bid and ask.165∗∗∗ (-4.05.3 Price Impact Reduction from HFT Liquidity Thus far.55) 590 0.28) ∗ (-2.410 Standardized beta coefﬁcients.01.001 6.2.006 (-0. The other coefﬁcients are negative. ∗∗ p < 0.72) -0.162∗∗∗ (-3. Market / Book $ of Non HFT Volume Average Spread Average Depth Volatility Autocorrelation # of Non HFT Trades Constant Observations Adjusted R2 (1) Economic Impact 0.have positive coefﬁcients. ∗∗∗ p < 0. ﬁrms with narrower spreads.1.086∗∗ (-2. suggesting that HFTs prefer to provide the inside quotes for value ﬁrms.05) -0.65) -0. and ﬁrms with a lower book depth.163∗∗∗ (-4. Another way of looking at HFT impact on liquidity is by looking at the depth of the book supplied 56 .217∗∗∗ (4.241∗∗∗ (-7. less volatility ﬁrms.
The market capitalizations are divided so that Very Small includes ﬁrms under $ 400 million. where a 1 represents one dollar increase in the price impact if HFT were not in the book.2 HFT Price Discovery HFT makes up a signiﬁcant portion of market activity. 6. where a 1 represents a 1 basis percent increase if HFT were not in the book. if HFT were not part of the book the price impact would be . but this need not be the case if HFTs did not have many orders in the book to begin with. one can observe the book with all of the limit orders in it and then remove the liquidity provided by HFTs and see what the impact would be on the cost of executing a trade for different size trades. Thus. The number of shares varies from 100 to 1000. As the trade size increases. 2009). I consider a variety of different impacts based on the number of shares hypothetically bought. and a Basis impact. A concern with this analysis is the endogeneity of limit orders (Rosu. One might expect the very small to impacted the most and their be a downward trend in impact as one moves to the large ﬁrms.19 percent higher than it actual is because of the liquidity HFTs provide.. For an average 1000 share trade. a market participant who sees a limit order at a given price or in a certain quantity (or absence thereof) may alter his behavior as a result. The results of this exercise are presented in table 20. 2009) and the information they may contain (Harris and Panchapagesan. I present both the dollar impact. That is. In addition. both on the demand side and the supply side. In this section I utilize three of Hasbrouck’s methodologies to see whether HFTs 57 . and large are for ﬁrms valued at more than $3 billion. this concern should be even further dampened as market participants can always hide their limit orders. I analyze what difference having HFTs provide liquidity in the book provides in decreasing the price impact of a trade. First though. The price impact is substantial. the price impact increases across ﬁrms of all sizes and for all ten trade size increases. but that does not imply its activities increase price efﬁciency. place a marketable order.5 billion. it is not clear whether once the market participant observed a given limit order he would be inﬂuenced to place his own limit order entry. Medium are those between $1. Interestingly the Small category tends to be more impacted by the withdrawal of HFT liquidity than is the Very Small category.by HFT. the dynamics are not clear whether this increases or reduces the impact of the previous analysis. Small are those between $400 million and $1. That is.5 billion and $3 billion. Cao et al. Table 20 shows the price impact based on market capitalization and also for the overall sample (column All). or to withhold from entering the market. 2005.
008 3.013 1.474 0.049 0. The column label Basis is the percent basis points change in price as a result of HFTs being in the market.725 0.033 13. This table looks at the liquidity depth of HFT and non-HFT by analyzing the price impact for different size ﬁrms if a varying range of trade-sizes were to hit the book.037 25.Table 20: Liquidity Book Impact.016 1.587 0.036 0.450 0.052 Small Basis Dollars 8.048 31.013 6.739 0.749 0.539 0.051 19.686 0.038 15.176 0.330 0.176 0.328 0.032 22.734 0.026 9. Trade Size 58 100 200 300 400 500 600 700 800 900 1000 Large Basis Dollars 1.030 11.052 34.072 All Basis Dollars 5.770 0.019 8.098 33.161 0.204 0.394 0.016 7.866 0.789 0.018 2.250 0.019 6. The column label Dollars is the dollar difference as a result of HFTs being in the market.900 0.331 0.677 0.162 0.358 0.019 Medium Basis Dollars 2.014 4.056 .233 0.027 11.065 0.020 17.151 0.147 0.040 16.041 0.042 0.023 7. The two-column wide labels.042 28.934 0.016 5.472 0.283 0.004 1.035 13.065 22.011 1.088 29.022 10. Very Small to Large refer to the ﬁrm size.133 0.050 19.028 19.008 1.260 0.007 1.037 11.045 17.619 0.592 0.324 0.028 9. The different rows represent a varying number of shares traded.064 42.855 0.955 0.045 13.058 38.017 1.663 0.012 4.352 0.683 0.428 0.075 26.074 0.307 0.784 0.005 1.108 Very Small Basis Dollars 12.
2. is based on event time. as occurs with data denoted in seconds. The HFT dataset is especially well suited for this as it is in milliseconds and thus avoids problems of multiple trades occurring in one time period. I utilize the impulse response function whose results can be interpreted as the amount of private information different traders bring to prices by measuring the amount of the price adjustment from the trade that is permanent. The results show that HFTs are more important in the price discovery process than non-HFTs. The time index. a midpoint quote return equation. I use a variance decomposition technique that takes the results of the impulse response function and relates the different type of traders’ trades to the price discovery process. The 10-lag vector auto regression (VAR) is: 59 . while Barclay. I implement the information shares approach which takes the innovations in HFTs and non-HFTs quotes and decomposes the variance of the common component of the price to attribute contribution to the efﬁcient price path between the two types of traders. I estimate the model for each stock for each day. Hendershott and Riordan (2009) performed a similar calculation for trader types looking at algorithmic trading. 6. HFTs provide more private information to the market than do non-HFTs. Finally. q H is deﬁned as the signed (+1 for a buy. HFTs provide substantially more information to the price process than do nonHFTs. and McCormick (2003) used the technique to compare information from different markets. The Hasbrouck methodologies utilized in this paper are similar to those found in Hendershott and Riordan (2009) and other papers. Hendershott.provide new information to the market. and so each t is an event that is a trade or quote change. and a non-HFT trade equation. -1 for a sell) HFTs trades and q N is the similarly denoted signed non-HFTs trades. Hendershott.1 Permanent Price Impact To measure the information content of HFT and non-HFT I calculate the permanent price impact of HFTs and non-HFTs trades. a HFT equation. As in Barclay. t. I estimate three equations. not clock time. Second. First. I estimate the model on a trade-by-trade basis using 10 lags for HFT and non-HFT trades. and McCormick (2003) and Hendershott and Riordan (2009). rt is deﬁned as the quote midpoint to quote midpoint return between trade changes.
Hasbrouck (1991a) interprets ∑ the impulse response function for HFT.t e(L) f (L) ϵ2. H ρi qt−i + 10 ∑ i=0 N ζi qt−i + ϵ2. Table 21 shows the results of the HFT and non-HFT impulse response function for 10 events into the future. The expected portion of a trade should not impact prices and so should not show up in the impulse response function. Each stock is reported individually. The impulse response function estimates this impact on future trades. N qt = 10 ∑ i=1 H νi qt−i + 10 ∑ i=0 N ψi qt−i + ϵ3. I calculate the overall average HFT and non-HFT impulse response function. this calculation incorporates the 60 . There are 105 ﬁrms presented as ﬁfteen stocks do not contain enough data to calculate the VAR. After estimating the VAR model.t . h(L) i(L) ϵ3.rt = H qt = 10 ∑ i=1 10 ∑ i=1 αi rt−i + δi rt−i + πi rt−i + 10 ∑ i=0 10 ∑ i=0 10 ∑ i=0 H βi qt−i + 10 ∑ i=0 N γi qt−i + ϵ1.t (5) where the vectors a(L) .t . Also.t . Hasbrouck (1991a) and Hasbrouck (1991b) took this methodology and applied it to the microstructure literature.t . of a trade should inﬂuence the price of future trades. the unexpected portion. however. as the private information t=0 content of an innovation in HFT. The impulse response function is a technology ﬁrst used in the macro-economic literature to determine the impact of an exogenous shock to the economy as it worked its way through the economy. average (VMA) model to obtain: rt a(L) H qt = d(L) N qt g(L) I invert the VAR to get the vector moving b(L) c(L) ϵ1. The non-HFT impulse response function is ∑10 t=0 c(L) and is the private information content of an innovation in non-HFT.i(L) are lag operators. The t-test is adjusted using Newey-West standard errors to account for the time-series correlation in observations. For each stock I estimate the statistical signiﬁcance of the difference of the impulse response function for the HFT and non-HFT 5 trading days using a t-test. 10 b(L). the innovation.
2 Aggregate Amount of Information in HFT . Of the 90 in the other direction.SR Price Impact The results in table 21 show that HFT has a larger price impact than does non-HFT over the ten period interval. and Chiquoine (2009) and Hendershott and Riordan (2009).017 and Non HFT’s impulse response is 0. Thus. On average. An item of interest is whether the price impact is immediate or gradual over the ten future time periods. 6.759. I test whether the price process may cause an immediate overreaction to one type of trade and that over the next nine periods in the future the impact decreases. This suggests that HFTs trades provide more private information than do non-HFTs trades. 26 of the differences are statistically signiﬁcant. for each market participant column. HFT’s impulse response function is 1. Of the 80 ﬁrms where the LR-SR impulse response function is greater for HFTs than non-HFTs 15 are statistically signiﬁcant. 10 event forecast horizon) and short-run (SR. The overall difference is statistically signiﬁcant. None of the 15 ﬁrms where the non-HFT impulse response function is larger than HFT’s are statistically signiﬁcant. If it is the case that there is an immediate overreaction to a HFTs trade this would support the theory that HFTs increase the volatility of markets. Hjalmarsson.Newey-West correction for time series and also a correction for the cross-section correlation standard errors. a positive number implies that the LR impact of a trade is greater than the SR impact. Of the 105 The LR-SR impulse response is less for HFTs than for non-HFTs in 25 of the 105 ﬁrms. Also.1.2.2. an innovation in HFT tends to lead to a 34% greater permanent price change than does a trade by a non-HFTr. but the methodology does not directly esti61 . The results of table 22 suggest that HFTs individual innovations have more private information than non-HFTs trades and that the difference is persistent and increases beyond the immediate impact of the trade. Similar to the methodology used in Chaboud. 6. This is similar to the ﬁndings in Hendershott and Riordan (2009) with algorithmic trades. To analyze this I report the difference between the long-run (LR. Of the 105 companies represented 90 of them have the HFT impulse response function being larger than the non-HFT impulse response. Vega. immediate) impulse response functions in table 22.Variance Decomposition The permanent price impact section above shows that HFTr demanded trades add important information to the market. and a negative number implies there is a short run overreaction and that over the next nine periods the permanent price impact falls.1 LR . Of those 25 ﬁrms none are statistically signiﬁcant.
497 2.364 1.384 1.536 2.541 1.899 -0.195 3.638 4.534 2.092 0.955 1.157 3.659 -0.753 0. This table reports the average long-run (10 events in the future) impulse response function for HFT and non-HFT.783 0.592 1.719 0.589 2.540 1.539 T Test 1.015 0.604 0.405 2.548 0.313 2.982 2.688 2.320 2.003 10.726 0.045 1.150 9.565 1.485 3.394 2.193 5.793 -0.984 0.808 -0.517 0.841 1.872 5.000 2.342 0.676 2.488 2.384 0.649 0.980 0.956 3.808 2.707 4.539 0.800 1.219 -0.596 0.819 1.596 7.814 -0.752 2.129 0.768 2.826 1.041 0.670 1.091 1.041 1.857 1.535 1.482 1.604 1.759 T Test 0.050 1.237 2.546 0.003 -10.644 0.790 1.553 0.656 -0.719 0.719 0.107 1.365 2.840 0.633 3.255 0.480 0.875 2.235 0.813 0.095 0.089 0.094 1.684 -0.813 3.794 1.840 0.525 2.744 1.225 -0.840 3.713 1.202 1.973 14.431 2.802 1.301 1.671 0.945 1.579 1.700 0.960 0.495 4.786 1.047 0.781 15.488 -8.627 3.518 1.457 1.083 1.773 1.449 4.261 1.476 Stock CSCO CSE CSL CTRN CTSH DCOM DELL DIS DOW EBAY ERIE EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG JKHY KMB KNOL KR LANC HFT 2.623 2.680 0.643 3.210 0.291 1.440 1.239 10.351 2.042 1.284 -0.328 1.608 1.659 1.704 2.644 1.930 0.154 1.788 -2.349 .900 -4.647 -0.247 1.110 2.008 0.664 3.357 1.941 1.215 3.407 0.398 1.170 2.085 1.017 Non HFT 1.782 3.721 0.239 0.593 2.314 3.882 0.904 1.399 -29.979 1.428 -0.680 Non HFT -0.987 0.360 0.820 1.011 3.798 0.034 0.549 4.658 0.516 2.762 2.239 2.676 0.358 1.931 0.458 1.712 0.985 1.890 0.607 0.642 0.435 0.553 0.911 3.323 4.891 1.212 1.681 1.262 4.450 1.172 5.746 -0.730 3.393 0.782 0.846 7.19 3.782 0. 62 Stock AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN APOG ARCC AXP AYI BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CCO CDR CELG CETV CKH CMCSA CNQR COO COST CPSI CPWR CR CRI Overall HFT 1.356 1.140 6.061 0.973 1.393 0.585 0.476 -0. The last column reports the T-statistics for the HFT .814 0.887 3.191 4.621 0.968 1.136 0.287 3.418 0.845 1.691 0.840 0.575 1.009 1.181 3.145 2.431 2.877 0.130 1.080 -0.463 4.061 -1.957 1.491 1.066 1.731 -3.225 0.066 1.687 4.049 1.439 T Test 1.872 3.155 4.717 2.877 0.244 3.811 1.043 2.431 Stock LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NSR NUS NXTM PBH PFE PG PNC PNY PTP RIGL ROC ROCK SF SFG SWN HFT 0.025 2.105 2.021 3.699 0.667 0.974 5.089 2.259 0.854 5.637 3.513 0.420 2.204 1.589 3.214 -3.054 0.448 Non HFT 1.237 0.206 0.790 2.575 -0.non-HFT difference for each security.356 4.889 -18.940 0.468 1.159 0.296 1.401 -0.133 11.519 4.260 9.465 1.897 2.215 2.189 1.583 1.Table 21: HFT and non-HFT Long-Run Impulse Response Functions.867 0.511 1.864 4.398 1.077 1.375 1.483 0.159 1.177 1.675 1.647 0.087 -3.501 1.581 1.849 2.217 0.769 2.905 2.689 1.357 1.986 -0.993 0.770 1.409 0.178 2.543 2.843 0.631 13.551 3.286 2.
247 1.114 2.539 -8.383 1. The last column reports the T-statistic for the HFT .021 0.175 0.252 0.non-HFT difference for each security.036 0.493 0.55 3.781 -0.282 0.110 5.946 Non HFT 0.546 Non HFT -0.133 -5.751 0. 63 Stock AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN APOG ARCC AXP AYI BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CCO CDR CELG CETV CKH CMCSA CNQR COO COST CPSI CPWR CR CRI Overall HFT 0.292 0.513 -0.169 2.286 5.303 0.145 0.590 -0.758 0.598 -0.677 3.875 1.300 0.134 1.168 0.short run HFT and non-HFT impulse response function (IRF).294 0.Short Run Impulse Response Functions.803 0.171 0.050 4.244 0.157 0.082 -0.179 0.891 1.489 -1.051 0.327 1.465 -0.457 1.007 0.510 3.343 0.588 1.025 0.216 1.229 0.027 -0.454 -0.709 1.726 1.080 0.760 0.303 0.599 0.180 -0.784 0.418 1.292 0.955 0.266 -0.099 0.294 0.134 1.203 0.014 0.145 1.801 0.373 1.971 0.940 2.357 0.988 0.233 1.535 0.272 -0.354 0.154 -0.505 0.567 0.186 0.418 -3.405 2.527 1.295 -7.961 3.119 0.655 0.852 2.255 0.739 1.665 0.646 0.723 0.345 0.308 1.081 0.158 -0.794 2.145 1.542 0.118 2.886 1.165 0.768 3.271 7.This table reports the average long-run .364 1.660 0.839 1.467 -1.031 0.174 0.182 0.309 -0.343 0.117 11.044 3.863 1.121 3.466 0.925 T Test -0.844 1.159 0.379 0.248 2.662 0.Table 22: Long-Run .253 0.370 0.082 0.710 -0.447 -0.471 1.202 -0.382 -0.104 1.239 1.208 0.333 1.344 0.284 -0.641 0.109 1.259 0.330 1.104 -0.581 1.090 0.795 0.301 0.642 0.591 1.917 0.579 1.847 3.676 0.247 -6.318 -0.815 -0.035 0.728 -3.605 0.282 -0.849 0.772 0.972 0.195 -0.433 0.372 0.384 2.021 3.187 0.623 0.198 2.640 0.390 0.837 1.242 -0.689 -0.767 1.906 0.701 0.365 0.774 -0.392 0.318 0.318 2.751 1.392 2.481 0.778 0.084 6.497 0.260 0.436 2.225 -0.698 2.514 0.615 3.707 -4.079 7.123 0.013 -0.369 3.976 0.611 1.229 -0.150 0.170 -1.264 2.166 -2.625 0.980 3.044 0.994 1.528 0.276 1.257 0.013 0.543 0.499 -1.008 1.159 -0.158 0.870 3.483 -0.194 -11.515 0.463 1.578 1.068 0.611 1.337 0.040 0.073 0.150 0.082 0.117 Stock LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NSR NUS NXTM PBH PFE PG PNC PNY PTP RIGL ROC ROCK SF SFG SWN HFT 0.427 0.740 1.257 2.505 0.875 -0.710 -0.611 0.285 0.777 0.375 0.098 2.210 1.755 0.515 Non HFT 0.002 0.941 1.308 0.846 -3.339 0. where the long run is the 10 events in the future IRF minus the one period IRF.433 -0.757 2.504 2.231 -1.368 0.725 0.417 0.341 T Test -0.607 0.963 -0.469 0.475 -2.427 0.922 1.756 0.200 0.162 2.678 0.026 -1.416 9.486 5.036 0.459 -2.633 2.030 -0.265 2.484 6.347 T Test 1.677 0.063 1.199 -3.458 0.480 0.166 0.275 1.242 1.563 Stock CSCO CSE CSL CTRN CTSH DCOM DELL DIS DOW EBAY ERIE EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG JKHY KMB KNOL KR LANC HFT 0.580 0.023 0.158 0.684 0.328 5.740 2.298 0.848 0.527 1.706 3.129 3.154 1.494 -8.965 -0.146 2.868 .732 0.315 0.107 0.302 0.
The ﬁnal column is the t-statistic for the difference between the HFT and non-HFT contribution and is adjusted for its time-series correlation with Newey-West standard errors. The contribution to the Returns component (the public information) is the public information related to price discovery. The methodology decomposes the variance of the efﬁcient price into the portion of total price discovery that is correlated with HFT and non-HFT trades. I also report the average overall contribution. whose t-statistic is corrected for time-series correlation and for cross-sectional correlation. st is the non-persistent price component. and σϵ3 = Eϵ2 . To examine this I follow Hasbrouck (1991b) to decompose the variance of the efﬁcient price into the portion of total price discovery that is correlated with HFT and non-HFT. The HFT column is the contribution to price discovery from HFTs. σϵ2 1 2 = Eϵ2 .non-HFT contribution difference is statistically signiﬁcant. Let σϵ1 = Eϵ2 . This analysis was also in Hendershott and Riordan (2009) to determine whether algorithmic or human traders contribute more to price discovery and I follow a similar methodology. mt represents the efﬁcient price where mt = mt−1 + wt and wt is a random 2 2 walk with Ewt = 0. it is unreported here for lack of space. and the same interpretation is true with the non-HFT column. In the 50 stocks where the non-HFT contribution is 64 . c are as deﬁned in the previous section as the lag coefﬁcients ∑ 2 found in the VMA matrix. b. The results from this exercise are found in table 23. To perform the variance decomposition the return series rt (using midpoint returns to avoid the bid-ask bounce) is separated into its random walk component mt and stationary component st : rt = mt + st .mate the importance of HFT and non-HFT in the overall price formation process. but can be easily calculated by taking the difference between 1 and the sum of the HFT and non-HFT components. I report the average contribution by HFT and by non-HFT for each company over the ﬁve days. The ( i=0 ai )2 σϵ1 term is the already public information portion of price discovery. The results indicate which trades contribute more to price discovery. i=0 i=0 i=0 (6) where the a. The ( 10 bi )2 σϵ2 term represents the proportion of the i=0 ∑ 2 efﬁcient price variance attributable to HFT and the ( 10 ci )2 σϵ3 term represents i=0 ∑10 2 the non-HFT proportion of the efﬁcient price variance. I decompose the variance of the efﬁcient price mt into 2 3 trade-correlated and trade-uncorrelated changes: 2 σw 10 10 10 ∑ ∑ ∑ 2 2 2 2 2 =( ai ) σϵ1 + ( bi ) σϵ2 + ( ci )2 σϵ3 . Of the 118 ﬁrms 68 of them show HFT as having a greater contribution to price discovery. and 28 of those stocks’ HFT .
The price vector of the HFT and non-HFT price process can be put into a VMA model: ∆pt = ϵt + ψ1 ϵt−1 + ψ2 ϵt−2 . 6. and Chiquoine (2009) and Hendershott and Riordan (2009).2. and. The price process is calculated from the HFT and non-HFT midpoint. t Then the price process for HFT and non-HFT is pHF T = mt + ϵHF T and pnHF T = t t t mt + ϵnHF T respectively. The variance σu can be decomposed as: [ ] [ 2 σHF T 2 σHF T.3 Information Share This section examines the role HFTs and non-HFTs quotes play in the price discovery process. M PtHF T HF = InsideBidHF T + InsideAskt T )/2 for HFT.greater than that of the HFT.nHF T 2 σHF T. . to determine which type of market participant contributes more to the price discovery process. I use the Information Shares (IS) approach introduced by Hasbrouck (1995) and that is used in. mt = mt−1 + ut . 65 . I calculate the HFT and non-HFT price path. This approach has been used to determine which of several markets contributes more to price discovery. The market participants’ variance is considered the contribution of that participant to the information in the price discovery process.nHF T 2 σnHF T ][ 2 σu = ΨHF T ΨnHF T ΨHF T ΨnHF T ] . and done similarly for non-HFT. Vega. . The approach is as follows. whereas the previous two sections had been analyzing the role of trades. ϵnHF T ] and is the information coming from HFT and nont t 2 HFT. Next. Hjalmarsson. as will be done here. if prices follow a random walk then I can represent the change in price as a vector moving average (VMA). among others. On average HFT contributes 86% more to price discovery than do non-HFT. (7) where ϵt = [ϵHF T . Chaboud. (8) where Ψ represent the lag operator vector from above and the sigmas represent the V ar(ϵt ) from above. and the common efﬁcient price path is the random walk t process. From the VMA I gather the variance of the random walk and the coefﬁcients of the VMA innovations. . the difference is statistically signiﬁcant for 7 ﬁrms. I can decompose the VMA variance into the lag operator coefﬁcients and the variance of the different market participants’ price paths.
025 0.017 T Test 3.150 2.436 0.027 0.027 0.003 0.002 0.002 0.671 2.125 -1.003 0.009 0.024 0.059 0.020 0.003 0.638 -2.000 0.029 0.490 5.106 0.007 0.000 0.511 1.002 0.002 0.098 1.002 0.654 Stock CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG HFT % 0.111 0.020 0.008 0.021 0.477 1.002 0.701 .381 Stock JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NC NSR NUS NXTM PBH PFE PG PNC PNY PPD PTP RIGL ROC ROCK ROG RVI SF SFG SJW HFT % 0.003 0. The remainder is in the Return column (unreported) and is interpreted as the price discovery from publicly available information.564 -0.781 0.868 0.436 0.907 -1.525 -1.031 0.573 0.037 0.100 0.073 0.079 0.014 0.012 0.003 0.081 0.414 3.210 0.759 1.025 0.044 0.045 0.011 0.851 0.026 0.105 T Test 3.003 0.035 0.018 3.002 0.048 3.014 0.590 -2.738 5.102 0.662 -4.756 -3.015 0.016 0.059 .563 5.027 0.005 0.001 0.714 9.105 0.000 0.059 0.004 0.300 0.166 0.037 0.066 0.749 0.164 1.001 0.266 2.002 0.208 2.966 -0.055 0.012 0.059 0.261 1.007 0.003 0.010 0.096 0.002 0.502 1.117 0.004 0.352 3.010 0.003 0.054 1.000 0.110 0.001 0.613 5.115 -2.096 3.804 -2.428 2.001 0.139 0.359 -1.013 0.049 0.729 1.502 1.068 0.003 0.147 0.356 -0.080 0.268 0.005 0.148 -0.652 5.016 0.996 1.010 0.355 -1.022 0.177 3.112 0.218 2.023 T Test -1.002 0.001 0.933 0.002 0.009 0.107 0.024 0.027 0.484 5.non-HFT Variance Decomposition.008 0.068 0.001 0.085 0.017 0.011 -0.856 -5.040 0.019 0.494 6.365 -1.205 1.115 4.002 0.021 0.097 0.007 0.113 0.037 0.010 0.790 -1.041 0.002 0.005 1.357 -1.073 0.535 2.005 0.007 0.716 -1.799 2.817 0.124 2.012 0.846 -0.029 0.701 3.182 0.923 -1.109 0.000 0.671 -2.243 0.449 -0.030 0.047 0.007 0.899 4.267 0.207 -2.278 0.023 0.350 -0.009 0.010 0.534 2.436 4.198 0.040 0.Table 23: HFT .048 0.010 0.013 0.004 0.010 0.091 -1.016 -4.012 0.023 0.541 5.065 0.008 0.002 0.139 0.029 0.463 1.030 0.119 0.029 0.129 0.004 0.464 1.195 Non HFT % 0.326 2.141 0.682 -0.016 0.003 0.002 0.077 0.838 -0.211 0.776 -1.039 0.247 -3.009 0.085 0.043 0.002 0.197 0.897 -1.032 0.001 0.024 Non HFT % 0.929 -1.377 -0.041 0.025 0.027 0.000 0.124 2.053 0.074 -1.079 0.080 0. This table reports the percentage of the variance of the efﬁcient price correlated with HFT and non-HFT trades.205 1.032 0.002 0.013 0.015 0.020 0.109 0.366 0.152 2.026 0.078 0.020 0.859 5.973 -5.004 0.171 0.053 0.203 0. 66 Stock AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI Overall HFT % 0.030 0.001 0.035 0.001 0.003 0.250 .120 0.070 0.173 0.420 2.073 Non HFT % 0.021 0.050 0.017 0.002 0.070 0.947 -1.003 0.045 0.852 -2.102 0.039 0.031 0.034 0.039 0.001 0.008 0.097 0.013 0.002 0.774 5.370 4.189 2.037 0.212 -1.099 0.207 1.079 0.175 -1.010 0.130 -1.082 0.178 0.014 0.018 0.032 0.036 0.709 -2.802 -2.002 0.254 0.020 0.011 0.178 0.158 -1.382 -0.003 0.187 -1.043 0.018 0.319 -0.068 1.001 0.168 0.067 0.147 0.002 0.003 0.
The results suggest that HFT reduces volatility to a degree. This suggest that in quotes. The table shows the average information share (which sums to 1 for each stock) for each stock. The information share of a participant is measured as that participant’s contribution to the total variance of the common component of the price.As the quote data I have is updated every time a new inside bid or ask is posted by a HFT or a non-HFT the diagonal values of the covariance matrix should be nearly perfectly identiﬁed. The information share attributable to HFTs and non-HFTs from their quote time-series process. The results suggest that HFT and volatility are not highly related. The results are found in table 24. as the book limit order book is updated every millisecond for which an order arrives. The results in Table 24 show which quotes contribute more to price discovery. HFT or non-HFT. Of the 17 companies where the non-HFT have a larger information share than HFT. Finally I compare the price path of stocks with and without HFT being part of the data generation process.3 Volatility The ﬁnal market quality measure I analyze is the relationship between HFT and volatility. for the previous ten periods. 6. I begin this analysis by performing two simple regressions. Next. using the period surrounding the short sale ban in September. there should be no contemporaneous correlation between HFT and non HFT quote changes. I switch the regression so that the dependent variable is the HFT 67 . only two of the differences are statistically signiﬁcant. The ﬁrst is a regression with the dependent variable being volatility. 103 stocks have the HFT information share being larger than the non-HFT information share. Of those 63 of the stock have HFT being statistically signiﬁcantly providing more information in their quotes than non-HFT. Similarly. as well as the volatility. and the explanatory variables are the total shares traded during that 10 second period and the percent of trades involving a HFT during that ten second period. 2008 I evaluate the impact on volatility of an exogenous decrease in HFT. I ﬁrst do an OLS regression to observe whether there is any relationship between HFT and volatility. 2010. like in trades. calculated in terms of 10 second realized volatility for each stock over the ﬁve trading days February 22 26. especially contemporaneously. That is. The average is over the ﬁve days in the dataset. The t-statistics are based on the difference in the information share between HFT and the non-HFT and incorporates Newey West standard errors to account for time series correlation. HFT are important in the price discovery process. as well as leads and lags for these two variables.
889 11.053 67.164 0.000 0.062 0.524 0.881 0.988 36.925 0.000 0.646 0.958 0.070 0.850 0.874 -0.30e+07 31.117 0.494 5.529 0.984 1.564 12.656 4.220 51.175 1.001 0.452 0.957 0.805 0.987 nHFT 0.864 0.151 6.999 0.152 0.540 0.346 0.472 2.995 0.136 0.317 0.447 0.999 0.952 0.228 0.491 0.476 0.012 0.294 0.614 0.061 0.879 1.001 0.836 0.351 0.987 0.561 0.292 60.017 0.40e+10 0.212 3.984 0.068 0.369 0.302 0.848 0.013 Tstat 131.480 0.193 5.919 0.715 0.665 -7.616 0.601 0.999 0.635 0.482 10.220 0.613 0.553 1.574 0.009 0.089 0.992 0.455 1.001 0.038 0.653 0.004 0.511 1.715 0.864 0.994 0.358 0.983 0.666 0.053 1.932 0.473 0.824 1.326 0.999 0.426 0.649 0.473 0.459 0.788 0.025 0.998 0.356 8.665 46.000 0.016 0.258 Tstat 5.000 0.150 0.411 28.119 0.706 1.784 0.323 1.996 0.013 0.377 1.870 0.335 14.939 0.55e+10 .468 0.603 8.000 0.308 0.285 0.856 0.000 0.368 7.671 103.22e+10 37.999 0.199 3.509 0.125 0.776 4.013 0.271 1368.499 0.021 1.934 -0.988 0.430 0.501 0.136 0.087 0.002 -0.045 0.706 0.491 0.994 0.173 0.383 0.608 0.002 0.158 3.982 0.042 0.Table 24: HFT and non-HFT Information Shares: This table reports the Hasbrouck (1995) information shares for HFT and non-HFT.474 0.384 0.542 0.983 0.995 0.014 0.213 2.518 0.914 0. 68 Firm AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO MDCO MIG MOS NC NXTM PG PPD ROC RVI Overall HFT 0.875 0.253 2.334 0.142 0.332 0.294 0.941 0.047 -1.170 -1.039 0.460 0.121 0.588 0.450 0.471 0.701 1.013 8.097 19.548 1.081 0.990 1.722 0.581 0.411 0.721 20.984 0.827 0.425 56.509 0.332 0.841 1.000 0.048 0.205 32.511 -5.811 0.141 -0.048 0.911 0.482 0.288 216.345 0.178 29.913 0.550 0.412 0.520 2.526 9.074 1.489 0.134 1.883 0.094 4.489 0.772 -0.274 -1.601 -0.941 7.962 0.986 0.000 0.575 0.836 0.439 0.527 0.058 0.000 0.828 2.952 0.393 122.387 0.858 0.000 0.541 0.111 1603.397 9.545 0.191 96.616 1.354 0.017 0.346 2.953 0.642 0.000 0.059 0.006 0.654 0.942 0.532 0.043 0.185 Firm FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MFB MOD MXWL NUS PFE PNY RIGL ROG SFG HFT 0.016 0.734 -0.275 0.772 0.589 0.237 7.278 0.879 0.010 0.000 nHFT 0.757 nHFT 0.689 0.505 0.936 837.267 66.034 -1.948 1557.130 0.311 0.285 0.000 Tstat 3.001 0.658 10364.709 0.991 0.266 0.060 1.100 30.000 0.144 0.001 0.938 0.933 1588.917 199.386 0.018 0.047 -1.138 35.631 0.59e+10 0.164 0.365 0.734 0.923 Firm CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE EWBC FCN FFIC MELI MMM MRTN NSR PBH PNC PTP ROCK SF HFT 0.961 0.656 0.005 0.526 0.495 0.822 0.016 0.006 0.975 0.698 -0.242 835.291 0.624 3.236 -0.818 0.527 0.365 0.511 0.189 0.929 0.008 0.341 0.195 0.047 7.347 0.698 0.989 0.000 0.399 0.683 0.520 0.987 0.344 0.673 4.071 0.595 1.780 1.655 0.400 2.842 -1.570 0.425 0.000 0.984 2.739 429.458 0.
the Percent of HFT trading coefﬁcient is statistically signiﬁcant and negative in the two period lag period. rvlag4 = (log(pricei. and the volatility is one of the explanatory variables.0−10 β23−33 × Hperclagi.t−5 /pricei. I include them only to show a possible link between HFT and volatility. The periods lag 1.t = α + β1−11 × rvlagi.t4 ))2 . Of course. The next section attempts to avoid the econometric issues and to reduce the endogeneity problem. “HF T %” for the ﬁrst regression and “RV ” for the second regression. The betas represent row vectors of 1x11 and the explanatory variables column vectors of 11x1. many of the prior volatility coefﬁcients are statistically signiﬁcant.0−10 = α + β1−11 × rvlagi. totshares is the number of shares that were traded for a company i in that ten second time period. for example using the fourth lag.0−10 + + + + + HP erci.percent of trades in that ten second window. The two regressions are as follows: V oli. Subscript 0 represents the contemporaneous value for that variable. For both. 9. 69 . and 10 statistically signiﬁcant. with HF T P ercent as the dependent variable. along with the others previous included in the regression. Table 26 shows the results of the two regressions and only reports the variables of interest. because of the endogeneity problem not much weight should be put on these results. only the results of the variables of interest are shown. this represents the number of ten-second time periods prior to the dependent variable time t event that the variable represents. 2. In the second regression. All of the statistically signiﬁcant lag coefﬁcients are positive except for period 9.0−10 β23−33 × Hperclagi. Hperc is the percent of trades for stock i in that time period for which HFT was involved. rv is the squared price change for company i for the respective time period.0−10 β12−22 × totshareslagi. 5. when volatility is the dependent variable.t Each explanatory variable has a subscript 0-10. This suggests that after volatility has been elevated HFT tend to make up more of the market trades.0−10 β12−22 × totshareslagi. The results suggest that there is some statistically signiﬁcant relationship between the two variables. In the ﬁrst two columns are the results of the ﬁrst regression with volatility as the dependent variable. Volatility is deﬁned as. The last two columns display the result for the second regression with HFT percent of market activity as the dependent variable.
00005 0. Variable HFT % lag0 HFT % lag1 HFT % lag2 HFT % lag3 HFT % lag4 HFT % lag5 HFT % lag6 HFT % lag7 HFT % lag8 HFT % lag9 HFT % lag10 N R2 F (187.RV Relationship.02362 0.Table 25: HFT . This table tries to capture whether there is a relationship between HFT and short-term market volatility.03959∗ -0.04379 0.35190 Signiﬁcance levels : ∗ ∗ ∗ : 1% 70 .552657) Dep = RV Coefﬁcient -0.00006 0.552657) -0.00072 0.00025 -0.12203∗∗∗ 0.00025 0.05887∗∗∗ 0.00048∗ -0.00010 0.00092 -0.09531∗∗∗ -0.03216 0.00025 0.00917 0.00003 0.01746 0.00017 -0.04089∗∗ 0.02628 0.00025 0.12254 1135.01552 552845 0.02731 -0.02459 0.00025 0.00027 0.00025 0.00026 0.00025 0.01675 0.00004 -0. RV lag0 RV lag1 RV lag2 RV lag3 RV lag4 RV lag5 RV lag6 RV lag7 RV lag8 RV lag9 RV lag10 N R2 F (187.02718 ∗ : 10% ∗∗ : 5% 552845 0.00001 -0.00025 Dep = HFTPercent Variable Coefﬁcient Std.01904 0. 0.02196 0.03580 0.00022 0.01608 0. Err.89658 70461.00025 0.00018 Std.00031 0. Err.00822 0.
HF T %Changei.t ∗ β1 + ϵi.3. The results are in table 27.ban. after talking with HFT ﬁrms. which was in place until October 9. the SEC made clear that ofﬁcially designated market makers were not subject to the ban and could freely short sell the 799 stocks. I run the following OLS regression: ∆V ola = HF T %Changei. where ∆V ola is the change in volatility for ﬁrm i between the pre. Now I study whether HFT inﬂuence volatility. ﬁrst I look at a time when HFT activity was exogenously reduced and examine what happened to volatility. for the pre ban data. 2008. The exogenous shock I utilize is the September 19. 2008. As seen in Section 5. As all 13 ﬁrms are in the short sale ban. The results in column (1) shows no relationship between an exogenous shock in HFT and volatility. The ban did not directly stop HFTs from trading in those shares. However.0. and the post ban data are for September 19. One reason HFT during this period does not drop further is that a portion of ﬁrms identiﬁed as HFT are ofﬁcial market makers and so they did not experience the same trading limitations as their nondesignated counterparts. This approach produces a negative 71 . HF Tpre −HF Tpost . Column HF Tpre (1) show the results when looking at the one-day level activity. Of the 120 ﬁrms in the HFT sample dataset. and second I consider what volatility would look like if HFT activity had not participated in the market. With the 13 effected ﬁrms I use the variation in the decline in HFT activity as different levels of treatment and study the subsequent change in volatility.1. using the average per stock data of the ﬁve trading days prior to September 19.6. 13 were on the ban list. Here I use an exogenous shock on HFTs to study the reverse. I do so using two methods.1 HFT Impact on Volatility I next try to disentangle the HFT . it was in fact a defacto ban on a portion of HFTs. it is clear they avoided these stocks as their strategies require them to switch freely between being long or short a stock. In a quick-to-follow clariﬁcation.t is the change in HFT activity pre.t . the pre ban data are for September 18. and the average per stock data of the ﬁve trading days after the ban for the post ban data. their is no concern that my results are actually an implication of the ban itself. Firm ﬁxed effects are used. HFT is inﬂuenced by volatility.and post. That is.and postban period. 2008. 2008. Before I used exogenous shocks to volatility to study its inﬂuence on HFTs. Column (3) performs the same analysis but uses the week average. and observing in the data that HFT activity dropped precipitously during this period (for the 13 affected stocks). 2008 ban on short sale trading for 799 ﬁnancial ﬁrms.Volatility relationship and minimize the endogeneity problem.5.
723∗ (-2.and post. but I can supplement that with the hypothetical price series of each stock assuming that there were no HFT in the market.05.Bootstrap 0. 29.28) 0. Table 26: Exogenous HFT . using the bootstrap techniques for the Week level analysis results in HFT % Change showing statistical signiﬁcance at the 5% level. I have the actual price series for each stock. This table shows the results of an exogenous removal of HFT and its impact on volatility. ∗∗ p < 0. Given the sparse number of observations. For each stock I calculate the realized volatility. I perform this calculation for each stock on each day and do it for the actual price path. . To reduce the impact of endogeneity.t is the change in HFT HF Tpre −HF Tpost activity pre. and 30 shows the results.Bootstrap -1.t . the greater the rise in volatility. I implement a non-parametric bootstrap looping through the data 50 times (using replacement). the sum over one minute increments of the absolute value of the returns over the day.523 (0.083 (2) 1 Day . This has no impact on the statistical signiﬁcance of the Day level analysis as seen in column (2). It uses the short sale ban as the source of exogenous shock.723 (-1.t ∗ β1 + ϵi. Table 28. I run the following OLS regression: ∆V ola = HF T %Changei. where ∆V ola is the change in volatility for ﬁrm i between the pre.ban. 13 ﬁrms are impacted.091 ( 4) 1 Week .84) 13 HFT % Change Constant Observations Adjusted R2 ∗ p < 0.26) 13 0.001 I also take an alternative approach to studying the impact of HFT on volatility. column (3) shows the results using the average values from the week before and week after for pre and post data points.01. However.ban period. Columns (2) and (4) utilize a non-parametric bootstrap looping through the data 50 times (and using replacement). I take advantage of the book data I have available in one minute increments. (1) 1 Day 0. HF T %Changei.568 (1.368 (0.48) 0.98) 13 (3) 1 Week -1. That is.coefﬁcient on HF T %Change which is interpreted as the more HFT decreased.92) 13 -0.523 (0. However.41) 0.568 (1. and also for an alternative price path based on the role of HFT. the coefﬁcient is not statistically signiﬁcant above the 20th percentile.42) 0. Column (1) HF Tpre shows the results using the day before and day after data. With this data I can estimate what the price impact would have been had there been no HFTs demanding liquidity or supplying liquidity. In table 28 I remove HFTr initiated trades 72 . Firm ﬁxed effects are used.Volatility Relationship.and post.368 (0. ∗∗∗ p < 0.
I adjust the price based on the size of the trade and the price impact it would have on the book after removing the HFT book entries. Thus. the only way prices can move is further away from their previous path. Table 30. This is the mechanical portion of the HFT price reduction: when liquidity is removed. negative. unlike the previous table. Table 29 does the same calculation but only removes the HFT liquidity providing trades. were the realized volatility calculation would have used a trade by a HFTr initiated trade. The t statistics for the individual ﬁrms use Newey-West standard errors to account for the time series correlation. Table 30 again compares the realized volatility of the 120 ﬁrms. Of those stocks were HFT reduced volatility. The t statistics for the individual ﬁrms use Newey-West standard errors to account for the time series correlation. 72 of them have a higher volatility when HFTr initiated trades are present. but it compares the volatility of the actual price path with the volatility of the price path if only HFTr initiated trades are removed. Only one ﬁrm shows that volatility is increased by removing HFT from providing liquidity. the overall t-statistic also corrects for cross-sectional correlation. the overall t-statistic also corrects for cross-sectional correlation. it instead has to grab the price from the next trade that is initiated by a non-HFTr. Also. may show a positive. Table 29 looks at what happens when HFT is only removed from providing liquidity. a small majority of ﬁrms experience slightly higher volatility with HFTr initiated trades. Finally. the overall t-statistic also corrects for cross-sectional correlation. Thus. only one exhibits that volatility would not be reduced if HFT had not been in the market. Of the 48 stocks where the presence of HFTr initiated trades reduces volatility none show a statistically signiﬁcant difference in volatility. These results suggest that HFT helps to reduce the volatility in the market.and also HFTr liquidity providing trades. only one is statistically signiﬁcant. table 30 removes only the HFT initiated trades from the alternative price path. Of the 120 ﬁrms. or no direction in its impact on volatility. when the realized volatility would have had a trade where a HFTr was providing the liquidity. 73 . In table 28 of the 120 stocks. The t statistics for the individual ﬁrms use Newey-West standard errors to account for the time series correlation. Of the 119 that show HFT is reduced. However of these 72 stocks. 82 show a statistically signiﬁcant difference in volatility. These results suggest that HFT initiated trades do not result in increased volatility. 85 of them have volatility that is statistically signiﬁcantly less than what it would be if HFT had not been part of the market. thus increasing volatility.
208 0.577 -4.695 -3.646 -2.161 0.201 0.256 0.446 0.135 0.204 0.288 -2.089 0.253 0.816 -2.180 0.305 0.288 0.688 -3.158 0.202 0.094 -4.193 0.209 0.120 0.241 0.093 0.753 -0.063 0.097 0.097 0.011 0.154 0.034 -2.912 -3.149 0.206 0.215 0.140 0.851 -2.226 0.241 -3.098 0.421 -3.163 0.103 0.144 no H RV 0.188 0.092 0.140 0.305 -1.182 T-Stat -2.101 0.363 0.119 0.755 -2.889 -3.194 0.570 -3.262 no H RV 0.318 -3.644 -3. I sum the one minute realized volatility and compare its actual value with what it would be if HFTs trading and liquidity had not occurred.120 0.000 -1.093 0.391 -2.144 0.779 -3.867 -1.282 0.142 0.835 -2.948 -4.131 0.155 0.271 -5.165 0.195 0.612 -5.138 0.202 0.183 0.186 -5.130 0.225 0.150 0.235 0.213 0.503 0.131 0.085 0.069 -3.614 -2.194 -3.119 0.340 0.119 0.106 0.921 -4.151 0.145 0.142 0.056 -3.Table 27: HFT Impact on Volatility .109 0.824 0.096 0.164 0.102 0.165 0.191 0.057 0.257 0.175 0.165 0.110 -0.996 -2.139 0.534 -5.175 0.428 -3.126 0.123 0.157 0.397 -3.291 -3.073 0.167 0.199 0.206 0.231 0.506 -1.164 T-Stat -1.001 0.159 0.304 -9.538 Firm JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NC NSR NUS NXTM PBH PFE PG PNC PNY PPD PTP RIGL ROC ROCK ROG RVI SF SFG SJW SWN H RV 0.311 T-Stat -5.200 -2.144 0.196 -9.088 0.050 -5.976 -2.165 0.069 0.235 0.111 0.080 0.095 0.280 -1.293 0.265 0.176 0.173 0.448 -3.126 0.081 0.160 0.202 0.060 -3.153 0.137 0.283 0.137 0.127 0.110 0.192 0.060 0.128 0.001 0.308 0.124 0.256 0.216 0.149 0.084 0.181 0.663 -3.159 0.286 0.188 0.311 0.125 0.116 -12.124 0.044 -2.660 -3.006 -4.164 0.087 0.271 0.129 0.177 0.179 0.075 0.193 0.106 0.345 -17.214 0.140 0.134 0.135 0.003 -3.428 -9.192 -4.148 0.236 0.620 -4.114 0.840 -3.120 0.337 0.131 0.084 0.911 -10.193 0.139 0.203 0.628 -0.311 -6.287 0.305 -1.604 -3.002 0.133 0.165 0.237 0.002 0.184 0.178 0.187 0.485 0.532 -5.143 0.827 -2.391 -3.791 -7.339 -4.198 0.023 -3.116 0.102 0.751 -5.106 0.893 -4.222 0.311 0.188 0.156 0.857 -1.536 -0.802 -0.070 0.223 -2.140 0.095 -3.984 -10.169 0.291 -0.238 0.954 -3.169 0.824 -7.738 -7.106 0.562 Firm CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE ESRX EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG H RV 0.886 -3.161 0.084 -3.567 -4.433 -3. 74 Firm AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI Overall H RV 0.101 0.167 0.207 -7.192 0.103 0.128 0.187 0.033 -2. This table looks at the impact of HFT on volatility.180 0.231 0.075 -3.763 -4.206 0.221 -3.568 -6.231 0.131 0.462 -5.135 0.138 0.148 no H RV 0.079 0.No Demand or Supply of Liquidity.142 0.105 0.157 0.154 0.133 -1.097 0.114 0.812 -11.166 -3.379 -4.093 0.181 0.697 -5.335 -3.092 0.361 -2.175 0.275 -1.156 0.148 0.175 0.960 -0.086 0.377 0.210 0.071 0.136 0.289 0.093 0.461 -12.469 -2.252 0.673 -1.187 0.125 0.161 -1.253 0.516 -1.167 0.698 -11.126 0.165 0.777 -3.760 -6.218 0.079 0.152 0.208 0.166 0.197 0.152 0.123 0.190 -2.155 0.139 0.116 0.748 -1.231 0.400 -1.098 0.011 0.077 .199 -2.708 -5.612 -3.092 0.628 -5.161 0.237 0.255 0.165 0.
091 0.791 -6.142 0.237 0.180 -1.139 0.446 0.120 0.085 -1.224 0.119 0.179 0.127 0.204 0.201 0.157 -2.041 -2.459 -3.161 0.096 0.682 -3.141 no H RV 0.691 -2.262 0.487 -3.345 -6.500 -3.062 -3.293 0.984 -3.967 -4.648 -2.111 0.625 -0.267 -2.582 .182 0.108 0.341 -2.098 0.206 0.880 -7.171 0.337 0.416 -3.246 0.136 0.093 0.143 0.182 T-Stat -2.251 0.154 0.360 -7.119 0.292 0.091 0.153 0.172 0.357 -2.231 0.157 -1.442 -3.140 0.199 -3.155 0.085 0.663 -11.142 0.125 0.193 0.131 0.253 0.705 -7.214 0.169 0.186 0.210 0.222 0.139 0.155 0.572 -3.158 0.142 0.109 0.391 -1.270 -3.604 -2.166 0.225 -3.173 0.098 0.161 0.199 0.113 0.613 -2.011 -2.165 0.187 0.342 0.452 -2.187 0.165 0.712 -3.394 -4.096 0.093 0.135 -3.157 0.600 -2.464 -7.169 -2.197 0.226 0.311 0.172 0.149 0.157 0.563 -1.485 0.063 0.246 -1.702 -7.083 0.203 0.176 -4.135 0.089 0.078 0.179 0.535 -2.136 0.070 0.684 -9.190 0.208 -0.236 0.499 -2.140 0.057 0.218 0.263 0.597 -4.194 0.727 -3.257 0.092 0.806 -0.283 0.282 0.237 0.156 0.547 -2.097 0.475 -5.109 0.447 -0.092 -9.527 -4.928 -2.129 0.377 0.097 -2.208 0.144 0.173 0.148 0.823 -3.077 0.839 0.311 0.165 0.202 0.725 -5.151 0.287 -4.658 -3.208 0.135 0.122 0.079 0.144 0.960 -2.126 0.175 0.095 0.180 0.644 -2.231 0.126 0.164 0.108 0.121 0.131 0.593 -1.011 0.200 0.137 0.175 -4.138 0.178 0.171 0.216 0.074 -1.113 0.353 -3.113 -6.047 -5.189 0.092 0.148 0.714 -2.187 0.154 0.100 0.863 -6.101 0.120 0.311 0.363 -6.070 0.153 -5.237 0.125 0.165 0.160 0.471 -6.237 0.089 0.267 no H RV 0.252 0.213 0.133 0.147 0.289 0.108 0.305 0.123 0.101 0.692 -3.705 Firm CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE ESRX EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG H RV 0.141 0.213 0.002 0.558 -10.160 0.133 0.202 0.133 0.209 0.256 0.876 -0.169 0.131 0.260 -2.352 -1.073 0.127 0.235 0.593 -3.480 -1.164 0.145 -2. 75 Firm AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI Overall H RV 0.118 0.182 0.185 0.601 -2.194 0.130 0.102 0.083 0.208 0.167 0.081 0.359 0.311 T-Stat -7.949 -10.001 0.140 0.136 0.134 0.165 0.099 0.122 0.238 0.587 -3.894 -7.309 -4.001 0.423 -2.051 -9.162 0.734 -7.659 -15.257 0.167 0.158 0.No Supply of Liquidity.175 0.106 0.518 -3.665 -5.195 0.287 0.138 -3.011 0.190 -0.103 0.878 -2.162 0.452 -1.227 0.154 0.153 0.167 0.891 -6.073 0.567 -5.839 -1.190 0.202 0.099 0.088 0.503 0.128 0.630 -3.202 0.181 0. This table looks at the impact of HFT on volatility.980 -2.798 -13.428 -3.726 -3.188 0.Table 28: HFT Impact on Volatility .145 0.157 0.192 -2. I sum the one minute realized volatility and compare its actual value with what it would be if HFT trading and liquidity had not occurred.135 0.059 0.382 -1.164 T-Stat -1.230 -4.163 0.814 -1.117 0.153 0.199 0.125 0.070 0.187 0.103 0.083 0.696 -8.948 -2.249 -2.286 0.846 -2.002 0.300 -1.124 0.265 -9.102 0.273 0.092 0.148 no H RV 0.118 Firm JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NC NSR NUS NXTM PBH PFE PG PNC PNY PPD PTP RIGL ROC ROCK ROG RVI SF SFG SJW SWN H RV 0.
338 0.105 0.259 0.117 0.120 0.070 -0.181 0.144 0.214 0.068 0.122 0.194 0.263 0.023 -0.110 0.165 0.280 0.041 -0.042 -0.184 0.101 0.145 0.070 0.101 0.860 0.081 0.237 0.142 0.156 0.174 0.081 0.161 0.208 0.179 0.000 -0.143 0.003 0.239 0.390 0.401 -1.012 0.174 0.193 -0.260 0.061 -0.175 0.287 -0.141 0.187 0.No Demand of Liquidity.189 0.171 0.294 0.088 0.643 -0.134 -0.100 0.253 0.118 0.131 0.089 0.149 0.083 0.114 0.117 0.070 0.193 0.155 no H RV 0.110 0.001 0.330 -0.105 0.507 0.197 0.174 0.135 0.212 0.125 -0.543 0.402 -0.142 0.172 0.043 0.221 -0.137 0.164 0.128 0.203 0.315 0.111 0.183 0.142 0.146 0.228 0.102 0.087 -0.128 0.001 0.150 0.137 0.076 0.187 0.150 -0.183 0.092 0.068 0.163 0.188 0.153 -0.347 0.127 0.064 0.164 0.185 -0.194 0.424 -0.298 0.173 0.103 0.186 0.100 0.208 0.158 0.000 0.493 .209 0.156 -0.181 0.101 0.059 0.020 -0.147 0.161 -0.149 0.465 -0.210 0.012 0.076 0.195 0.159 0.048 0.076 0.116 0.100 0.002 -0.123 0.099 0.446 -0.105 0.105 0.117 0.205 0.257 -0.270 0.128 0.020 -0.261 0.579 -0.323 0.110 0.246 0.119 0.160 0.108 0.294 0.146 0.356 0.140 0.140 0.150 0.683 -0.155 0.167 0.171 0.99 Firm CPWR CR CRI CRVL CSCO CSE CSL CTRN CTSH DCOM DELL DIS DK DOW EBAY EBF ERIE ESRX EWBC FCN FFIC FL FMER FPO FRED FULT GAS GE GENZ GILD GLW GOOG GPS HON HPQ IMGN INTC IPAR ISIL ISRG H RV 0.416 -0.129 0.060 0.000 0.179 0.082 0.154 0.256 0.176 0.150 0.157 T-Stat -0.138 0.147 0.268 0.101 0.102 0.096 0.145 -1.185 0.108 0.343 -0.172 0.207 0.281 0.288 no H RV 0.181 0.117 0.130 0. This table looks at the impact of HFT on volatility.068 -0.144 0.193 0.464 -0.532 0.227 0.149 0.227 0.241 0.656 -0.211 0.078 0.091 0.031 0.019 -0.235 0.556 -0.152 0.304 0.188 0.212 0.269 0.163 0.116 0.148 -0.105 0.069 0.003 0.253 -0.741 0.337 -0.113 0.086 0.158 0.873 -0.197 0.052 -0.924 -0.102 0.187 0.141 0.175 0.088 -0.121 0.137 0.194 0.458 -0.225 0.000 -0.262 0.095 -0.272 -0.326 -0.128 0.739 -0.085 0.295 -0.186 0.143 0.073 0.211 0.218 -0.001 0.088 0.391 0.151 0.103 0.128 0.091 -1.025 -0.147 0.122 0.154 0.138 0.181 0.133 0.348 -1.243 0.118 0.650 Firm JKHY KMB KNOL KR KTII LANC LECO LPNT LSTR MAKO MANT MDCO MELI MFB MIG MMM MOD MOS MRTN MXWL NC NSR NUS NXTM PBH PFE PG PNC PNY PPD PTP RIGL ROC ROCK ROG RVI SF SFG SJW SWN H RV 0.209 0.198 0.097 0.059 -0.178 0.210 0.208 0.005 -0.Table 29: HFT Impact on Volatility .186 0. I sum the one minute realized volatility and compare its actual value with what it would be if HFT trading and liquidity had not occurred.372 -0.315 0.346 0.119 0.386 0.163 0.103 0.182 0.296 0.143 0.052 -0.082 -0.126 0.150 0.403 -0.125 0.200 0.147 0.189 0.140 -0.295 -0.126 0.172 0.296 0.333 0.033 0.160 T-Stat -0.145 0.093 0.321 0.165 0.560 0.100 0.181 -0.841 -0.531 -0.123 0.146 0.139 0.161 0.197 0.118 0.238 0.138 0.154 0. 76 Firm AA AAPL ABD ADBE AGN AINV AMAT AMED AMGN AMZN ANGO APOG ARCC AXP AYI AZZ BARE BAS BHI BIIB BRCM BRE BW BXS BZ CB CBEY CBT CBZ CCO CDR CELG CETV CHTT CKH CMCSA CNQR COO COST CPSI Overall H RV 0.102 0.103 0.237 0.331 0.159 no H RV 0.102 0.444 0.024 0.206 0.132 0.019 -0.090 0.282 T-Stat -0.199 0.295 0.121 0.031 -0.142 0.087 -2.210 0.141 0.253 0.117 0.037 0.075 0.098 0.096 0.
the evidence suggest HFT plays a very important role in price efﬁciency and the price discovery process. HFTs supply liquidity in about half of all trades and demand liquidity in about half as well.999. making around $3 billion annually. HFT make up a large majority of all trades. lower volatility. usually in value between $1. Finally. HFT activity either has no impact on volatility or tends to decrease it. They tend to make more money in volatile times. both in initiating trades and in providing liquidity. and companies that are considered value ﬁrms. 77 . lower spreads and depth.000 and $4. and they tend to have lower time between trades than non-HFTs. Their activities in the market. and this is stronger when they are demanding liquidity. lower spreads and depth. HFT prefer to trade in large stocks with lower volume. it provide more useful information to the price generation process than do non-HFT. and with greater number of trades.7 Conclusion This paper examines high frequency trading and its role in ﬁnancial markets. They tend to engage in a price reversal strategy. HFTs prefer to demand liquidity in small amounts. There is no evidence of abusive front running occurring. They provide the best quotes about 45% of the time. are stable over time. HFT ﬁrms are proﬁtable. with lower volume. They provide more inside quotes for larger value ﬁrms. In fact. From the different Hasbrouck measures.
Chaboud. Specialist proﬁts and the minimum price increment. J. Harris. and D.T. C Duhigg. Order preferencing and market quality on NASDAQ before and after decimalization* 1. 29(1):16–41.References M. Kandel. Working paper.html).F. Journal of Finance. (http://www. July 23. Stock Traders Find Speed Pays. M. J. K. and D. 54(1):1–34. 1995. and P. 1993. Journal of Futures Markets.H. C. and X. An empirical analysis of the limit order book and the order ﬂow in the Paris Bourse. Buffalo. Chung.B. Coughenour and L.J. Who Makes Markets? Do Dealers Provide or Take Liquidity. C. S. in Milliseconds. 2009. Cao. New York Times. Hjalmarsson. Journal of Financial Economics. 34(3):281–305. Hansch. Vega. 2009. Competition among trading venues: Information and trading on electronic communications networks. Stealth-trading: Which traders’ trades move stock prices?* 1.H. Barclay. Technical report. Rise of the machines: Algorithmic trading in the foreign exchange market. 58(6):2637–2665. Harris. Christie. Journal of Finance. Effects of market reform on the trading costs and depths of Nasdaq stocks. T. International Finance Discussion Papers. 50(5):1655–1689. and C. Wang.nytimes. Chae and A. E. P. C. Barclay. Stealth trading and volatility:: Which trades move prices? Journal of Financial Economics. Barclay and J. 78 .H. 2009. Chakravarty. Biais. Hendershott. A. McCormick. and B. 2003. 2001. 71(3):581–612.T. Wang. M. The Journal of Finance.G.com/2009/07/24/business/24trading.J. Spatt. Journal of Financial Economics. The information content of an open limit-order book. State University of New York. Hillion. W. Chuwonganant. Schultz. Chiquoine. E. 2003. B. O. McCormick. 1999. 2004. Warner.J. 61(2):289–307. J.
Program trading and intraday volatility. Journal of ﬁnancial economics. 2008. K. Journal of Banking & Finance. Stein. One security. Shapiro. 2007. 48(5):1565–1593. Review of Financial Studies. Moulton. 46(1):179–207. Herd on the street: Informational inefﬁciencies in a market with short-term speculation. Lo. Scharfstein. Hasbrouck.M. An ordered probit analysis of transaction stock prices. 31(3):319–379. Measuring the information content of stock trades. 79 . Journal of Finance. MacKinlay. Gsell. Soﬁanos. 63(1):119–158. 2009. J. Panchapagesan. Hendershott and R. Riordan. Journal of ﬁnance. 1992. SSRN. Hendershott and P.Y. Hasbrouck. Soﬁanos. In Proceedings of the 16th European Conference on Information Systems (ECIS). 47 (4):1461–1484. Harris. J. Harris and V. Assessing the impact of Algorithmic Trading on markets: A simulation approach. Liu.E. The trades of market makers: An empirical analysis of NYSE specialists. Competition for order ﬂow and smart order routing systems. 2005. 1994. Journal of Financial Markets. 4(3):571–595. UC Berkeley. J. 1993.C. and J. pages 1175–1199. Hasbrouck and G. Galway.K. T. Speed and stock market quality: The NYSEs Hybrid. Limit order revisions. M. Ireland. Review of Financial Studies. Journal of Finance. 2008.J. A.E. Foucault and A. The summary informativeness of stock trades: An econometric analysis. J. 1991a. 2010. J. and J. The information content of the limit order book: evidence from NYSE specialist trading decisions. Journal of Finance. G.S. and A.A. many markets: Determining the contributions to price discovery. The Journal of Finance. Manuscript. 8(1):25–67.C. 1995. 1991b.A. Hausman. L. T. L.C. 7(4):653–685. D.L.W. T. Hasbrouck. Menkveld. Froot. 1992. Fong and W. Algorithmic trading and information.
42(3-4):275–298. Skinner. Empirical Limitations on High Frequency Trading Proﬁtability. Journal of Finance. R. Kearns.reuters. Kardos and S. 8(1):125–160. R. 1989. Options markets and stock return volatility.html). (http://www. Van der Wel.com/story/high-speed-trading-led-toﬂash-crash-survey). 2002. 2010. 2010. (Concept Release on Equity Market Structure). Menkveld. (http://www. D. 1997. 2010.T.K. high-speed trading led to ﬂash crash: survey. D. 23(1):61–78. C. M. Working Paper. June 10.com/article/idUSN173583920091202). Securities and Exchange Commission. 1995. Is the May 6 ”ﬂash crash” in US stock markets the new normal? Reuters. Lee. Journal of Financial Economics. A dynamic model of the limit order book. 2010. X.nytimes. Krudy. Computers. E. Kulesza. 2008. Does Algorithmic Trading Improve Liquidity? Journal of Finance. I. Who’s afraid of high-frequency trading.J. January 18. Huang. Lyons. A simultaneous trade model of the foreign exchange hot potato.marketwatch. Journal of International Economics. 2009. Sarkar A. (http://www. and A. Review of Financial Studies. A. Forthcoming. 2009. August 10. Spicer. December 2. Menkveld A. Concept Release on Equity Market Structure. A. January 14. M.com/article/idUSN1057063420100610).D. Reuters. 2010.com/2009/07/24/business/24trading. Are Market Makers Liquidity Suppliers? Working paper. 57(3):1285–1319. Lash H. Hendershott. Nevmyvaka. Wall Street Journal. and Y. Jones. 22(11):4601. Rosu.J. 80 . Market Watch. The quality of ECN and Nasdaq market maker quotes.M. Review of Financial Studies. J. Vives.reuters. 2008. Senator Calls for More Regulation of High-Frequency Trade. Patterson. Short-term investment and the informational efﬁciency of the market. (http://www.