Printer Friendly

Quote stuffing.

In this study we examine intense episodic spikes in quoting activity (frequently referred to as quote stuffing) on market conditions. We find that quote stuffing is pervasive and that over 74% of US exchange-listed securities experienced at least one episode during 2010. We also find that stocks experience decreased liquidity, higher trading costs, and increased short-term volatility during periods of intense quoting activity. We find that most quote-stuffing events occur on the NYSE, ARCA, NASDAQ, and BATS and that during these quote-stuffing events, the number of new orders and canceled orders increases substantially while the order size and order duration decrease.

**********

Quote stuffing is the practice where a large number of orders to buy or sell securities are placed and then canceled almost immediately. These intense episodic spikes in order submissions and cancellations have come under scrutiny from the media and regulators (see, e.g., Lauricella and Strasburg, 2010). Market participants criticize the practice, stating that it creates a false sense of the supply and demand for a stock. Sean Hendelman, chief executive officer at T3 Capital, expressed his concern stating, "People are relying on the [stock quote data] and the data is not real" (Lauricella and Stasburg, 2010, p. A.1). Others liken the practice to an auctioneer placing "plants" and "shills" in the audience in an attempt to manipulate prices through fake bidding (Elder, 2010). Are these concerns justified? How prevalent is quote stuffing? Does quote stuffing adversely affect market conditions, and if so, to what degree? Are quote-stuffing events localized on one exchange or are quoting and trading altered on all exchanges during quote-stuffing events? This article seeks to address these questions.

Although quote stuffing is often linked to high-frequency trading (HFT), smart order routers and other algorithmic traders, who are not high-frequency traders, may also be quote stuffing. Nevertheless, HFT has garnered increased attention in the wake of the May 6, 2010 flash crash when the Dow Jones Industrial Average collapsed 998.5 points in a few minutes. HFT is a strategy where securities are rapidly purchased and sold through the use of computer algorithms. (1) Holding periods for securities bought and sold by high-frequency traders are typically very short, lasting just seconds or milliseconds. Furthermore, high-frequency traders may move in and out of positions thousands of times per day. The Securities and Exchange Commission (SEC, 2010) calls HFT "one of the most significant market structure developments in recent years." SEC Chairwoman Mary

Schapiro (2010) describes the regulatory scheme that applies computer-based low-latency trading as "[an] area that warrants close review." Today HFT makes up a significant portion of US equities market volume. (2)

Despite the criticism of HFT and algorithmic trading (AT) by the popular press and market participants, early academic work finds little evidence that the practice is detrimental to financial markets. Recent studies show that, in aggregate, HFT improves traditional measures of market quality and contributes to price discovery (Hasbrouck and Saar, 2013; Brogaard, 2010). Additionally, Menkveld (2011) examines the high-frequency trader's role as a modern market maker and finds it to be crucial to the operation of a new market.

Many AT and HFT strategies rely on the ability to trade fast and frequently. (3) Latency arbitrage is one such strategy in which high-frequency traders attempt to profit from inefficiencies in data between exchanges or other market centers. By submitting large numbers of orders that are canceled very quickly, a high-frequency trader may create exploitable latency arbitrage opportunities. Brogaard (2010) explains that latency arbitrage opportunities from quote stuffing may arise from requiring other traders to process large amounts of volume, giving the high-frequency trader submitting the orders an advantage. (4) A large number of order submissions may also cause the exchange receiving the quotes to lag other exchanges, creating arbitrage opportunities.

It is also possible that large bursts of quoting activity may not be caused for manipulative purposes. Large episodic spikes in quoting activity may be generated for technological reasons where two algorithms interact with each other and fail to converge. For example, one algorithm submits a quote that causes another algorithm to reply, causing the first algorithm respond. If this process of multiple algorithms "chasing" each other continues, a large burst of quotes will be generated. Although the large burst of message flow may not be part of a nefarious plan to manipulate the market, these quoting episodes may still be associated with degraded market conditions.

In this study, we identify and analyze intense episodic spikes in quoting activity. We examine market conditions, including liquidity and volatility around these episodic spikes. We find that episodes of large bursts of quote updates are pervasive, with events occurring every trading day and affecting over 74% of US-listed equities. Our results suggest that in periods of intense quoting activity, stocks experience decreased liquidity, higher trading costs, and increased short-term volatility. Thus, quote stuffing may exhibit some market-degrading features (and could be creating the latency arbitrage opportunities described by Brogaard, 2010 and Biais and Woolley, 2011).

This study is also related to the broader research on market manipulation (see Allen and Gorton, 1992; Jarrow, 1992; Kumar and Seppi, 1992; Mei, Wu, and Zhou, 2004; Goldstein and Guembel, 2008). Aggarwal and Wu (2006) present a theoretical model and empirical evidence of stock market manipulation. They find that in the presence of stock price manipulation, volatility increases and market efficiency worsens. Aggarwal and Wu's (2006) model suggests a strong role for regulation to discourage manipulation.

In 2010 the Financial Industry Regulatory Authority (FINRA) fined Trillium Brokerage Services LLC and nine of its traders $2.26 million for illicit market manipulation. FINRA accused Trillium of creating a "false appearance of buy- or sell-side pressure" through an illicit "layering" strategy. Layering involves traders entering multiple fake orders to create a false buying or selling interest with a goal of improperly baiting unsuspecting market participants into executing trades at unreasonably high or low prices. FINRA alleged that Trillium's trading strategy induced other market participants to enter orders to execute against limit orders previously entered by the Trillium traders. Once their orders were filled, the Trillium traders then immediately canceled the orders that had been designed only to create the false appearance of market activity. Trillium traders' improper trading strategy allowed the firm to obtain advantageous prices that otherwise would not have been available to them on over 46,000 occasions (FINRA, 2010). The layering strategy employed by Trillium differs from the quote-stuffing strategies we explore in this study. Whereas quote stuffing is frequently employed as a denial-of-service attack with an attempt to slow down the market, layering is a way of getting good execution on a trade that was planned before the layering occurred (Salmon, 2010).

I. Background

This article is closely related to a small but growing body of literature that addresses issues concerning HFT and AT. Hendershott, Jones, and Menkveld (2011) explain that declining technology costs, as well as trading that has become increasingly electronic, have made it easier and cheaper for firms to implement computer programs to make trading decisions, submit orders, and modify those orders after submission. Today, orders submitted via computer algorithms make up over two-thirds of US equities market volume (Hendershott et al., 2011).

Hendershott and Riordan (2009) use data from the 30 largest DAX stocks on the Deutche Boerse to determine the role of AT in the price-discovery process. They find AT represents a large fraction of the order flow. For sample stocks, AT demand (supply) represents 52% (50%) of trading volume. (5) Algorithmic traders also contribute more to price discovery than do their human counterparts. Algorithmic traders are more likely to be at the inside quote when spreads are high than when spreads are low, suggesting that algorithmic traders supply liquidity when it is expensive and demand liquidity when it is cheap. The authors find no evidence that AT increases volatility. Hendershott et al. (2011) examine the impact AT has on the market quality of New York Stock Exchange (NYSE)-listed stocks. Using a normalized measure of NYSE message traffic, they measure the causal effect of AT on liquidity surrounding the NYSE's implementation of automatic quote dissemination in 2003. They find that AT narrows spreads, reduces adverse selection, and increases the informativeness of quotes, especially for larger stocks. These results suggest that AT improves liquidity and market quality.

Theoretical models of HFT show that it is possible for HFT to either enhance or degrade market quality. Cvitanic and Kirilenko (2010) develop a theoretical model predicting the presence of high-frequency traders is likely to cause a change in average transaction prices, with more mass around the center and thinner tails. This price distribution arises as high-frequency traders "snipe" out human orders, which are away from the inside of the book. Volume, intertrade duration, and liquidity should all increase with changes in the speed and quantity of human order submissions. As the proportion of transactions submitted by computers grows, the ability to forecast with transaction prices should increase.

Cartea and Penalva (2011) model the impact of HFT on financial markets using a model with three types of traders: liquidity traders, market makers, and high-frequency traders. According to their model, high-frequency traders increase the price impact of liquidity trades, increasing (decreasing) the price at which liquidity traders buy (sell). These costs increase with the size of the trade, suggesting that large liquidity traders (i.e., large institutional traders making sizable changes to their portfolio) will be most affected by HFT. Market makers are compensated for losses in revenues to high-frequency traders by a higher liquidity discount. Thus, HFT does not affect the number of market makers. The authors also propose that HFT increases price volatility and doubles volume.

Most empirical studies on HFT find it has a moderate to significant positive impact on traditional market-quality measures (see Jones, 2013, for a survey of current research on HFT). Brogaard (2010) examines the impact of HFT on the US equities market using a unique HFT data set for 120 stocks listed on NASDAQ. Brogaard (2010) finds that HFT improves market conditions: it adds to the price-discovery process, provides the best bid and offer quotes for a significant portion of the trading day, and reduces volatility. However, the extent to which HFT improves liquidity is mixed as the depth high-frequency traders provide to the order book is one-fourth of that provided by nonhigh-frequency traders. Broggard's (2010) analysis also suggests that HFT is a profitable venture, generating trading profits of $2.8 billion annually. Hasbrouck and Sarr (2013) use NASDAQ order-level data to examine the impact that low-latency traders have on market characteristics, including volatility, total price impact, and book depth. They measure HFT activity by identifying "strategic runs" of submissions, cancellations, and executions. The authors find that HFT improves market quality by decreasing short-term volatility, spreads, and depth of the order book.

Contrary to the aforementioned empirical studies, Hirschey (2013) shows that HFT may increase trading costs for nonhigh-frequency traders, and Zhang (2010) finds that HFT may increase stock price volatility and impede the market's ability to incorporate firm fundamentals into asset prices. Zhang (2010) uses the Center for Research in Security Prices (CRSP) and Thomason Reuters Institutional Holdings databases to estimate HFT dollar volume. He finds a positive correlation between HFT and quarterly volatility, and this relation is strongest for larger stocks. Zhang (2010) also finds that prices of stocks with more HFT tend to overreact to firm fundamental news such as earnings surprises.

Other studies examine the role of high-frequency traders in the May 6, 2010 flash crash. Kirilenko et al. (2010) study the behavior of high-frequency traders in E-mini S&P 500 futures contracts during the events surrounding the flash crash. HFT patterns surrounding the flash crash are inconsistent with traditional market making. The authors conclude that although high-frequency traders did not cause the flash crash, their response to the high selling pressure exacerbated volatility. Madhavan (2012) analyzes the relation between market structure and the flash crash. He finds that firms with higher fragmentation before the flash crash were disproportionately susceptible to rapid price movements on the day of the crash and provides a framework with which to evaluate new market structure reforms.

HFT is also described as modern market making. Menkveld (2011) examines HFT and its role as a modern market maker. He documents how one large high-frequency trader that acts as a market maker is critical to the operation of a new market, Chi-X. Menkveld (2011) provides a detailed analysis on the trading behavior of the high-frequency trader, finding that the high-frequency trader provides liquidity and that the entrance of the high-frequency trader corresponds to a decrease in spreads.

Ye, Yao, and Gai (2012) study an exogenous trading shock that increases the speed of trading on NASDAQ from microseconds to nanoseconds. The authors find evidence that NASDAQ stocks have an abnormal amount of correlation with other stocks handled by the same channel, consistent with single-venue quote stuffing. They show that excessive message flows can slow trading for stocks within the same channel.

Our study adds to the literature by exploring quote stuffing, a strategy in which a large number of orders to buy or sell securities are placed and then canceled almost immediately. Market participants criticize this practice, arguing it creates a false sense of the true supply and demand for a stock and may adversely affect market quality. Unlike previous empirical studies of HFT in US equities markets, which use data from a single market center, we examine quote-stuffing behavior across all US exchanges. (6) Considering the fragmentation of order flow in US markets, we believe that we will glean a more complete picture of quote stuffing and the existing market conditions by using data from all US exchanges.

II. Data and Identification of Quote Stuffing

A. Data

The primary data source for this article is the NYSE Trade and Quote (TAQ) database. Our sample includes all trades and quotes for NYSE- and NASDAQ-listed stocks for all trading days in 2010. We apply conventional filters to TAQ, excluding trades and quotes that are coded as having an error or a correction, or are reported out of time sequence. In addition, we omit a quote if the bid is greater than the ask, or the bid and/or ask price is less than zero. Securities with an average trade price less than $3 are also eliminated.

We use TAQ data to both identify quote-stuffing episodes and calculate measures of market quality. Our analysis is restricted to normal trading hours (9:30 a.m. to 4:00 p.m.). We follow Bessembinder (2003) when merging trades and quotes and do not lag quote time stamps. CRSP data are used to compute daily trading statistics and to determine listing exchange. We also use NASDAQ TotalView-ITCH data to examine orders, executions, and cancellations of trades during identified quote-stuffing events. The NASDAQ TotalView-ITCH data, which are used for our final analysis, provide information on transactions that execute on NASDAQ only.

B. Use of TAQ Data to Identify Quote Stuffing

Typically, in most data sources it is not possible to identify orders that are generated by computer algorithms in US equity markets. As a result, previous studies use proxies to measure the level of AT and HFT. These proxies are typically derived using system order data, which identify electronic messages including order submissions, cancellations, and executions handled by an individual exchange. For example, Hendershott et al. (2011) use the number of electronic messages handled by NYSE's SuperDOT system and captured in the NYSE's System Order Data (SOD) database as a proxy measure of AT. Hasbrouck and Saar (2013) compute their proxy for low-latency trading using NASDAQ TotalView-ITCH, which includes submissions, cancellations, and trade executions for orders received by NASDAQ. Using these data, the authors develop a proxy for HFT by identifying "strategic runs," which the authors define as "linked submissions that are likely to be parts of a dynamic strategy" (Hasbrouck and Saar, 2013, p. 658). Unlike the proxies developed by the aforementioned studies, we use TAQ data to identify heightened periods of low-latency activity. In contrast to system order data, TAQ data do not include information on individual order submissions and cancellations, but contains consolidated quotes from all exchanges in the national market system. Despite not containing information on individual orders, submissions and cancellations of marketable orders are reflected in consolidated quote updates of TAQ. Thus, frequent quote updates in TAQ are likely to be highly correlated proxies of HFT based on system order data. (7)

An attractive feature of TAQ for our study is that it includes quote updates for all exchanges that trade US equities. Unlike the US equity market of just over a decade ago where a few venues commanded an overwhelming share of market activity, today's market is fragmented with order flow going to an increasing number of trading venues. O'Hara and Ye (2011) show that both NYSE- and NASDAQ-listed stocks exhibit substantial fragmentation. Quote stuffing is likely to involve order-submission strategies that span multiple trading venues, possibly in an attempt to exploit inefficiencies that may arise in prices across exchanges. Thus, examining quoting behavior across market centers should provide a more complete picture of quote stuffing and the existing market conditions.

C. Identification of Quote Stuffing

Ye, Yao, Gai (2012) state that quote stuffing is hard to identify and use message traffic on NASDAQ to identify quote-stuffing events in their study. Identifying all quote-stuffing events is potentially difficult; therefore, we locate extreme episodic spikes in quoting activity. To locate these events, we first divide the trading day (9:30 a.m. to 4:00 p.m.) into 390 one-minute segments. Next, we calculate the intraday variation in quoting activity by computing the average standard deviation of the number of quotes submitted in the one-minute segments for rolling 20-day windows. We identify intense quoting episodes as segments where the level of quoting activity exceeds the previous 20-day mean number of quotes per minute by at least 20 standard deviations. We also require that the average number of quotes for the entire trading day not exceed its previous 20-day rolling average by more than two standard deviations. The latter requirement is implemented to exclude trading days with an unusually high level of quoting activity.

We group multiple one-minute segments into a single quote-stuffing event when the duration between high-quoting episodes is 10 minutes or less. Grouping of one-minute segments yields a total of 58,737 unique quote-stuffing events with durations ranging from 1 to 10 minutes. (8) As our goal is to identify information-free intense episodes of quoting activity, we attempt to eliminate conflicting events by using CRSP and Compustat to identify corporate announcements. We exclude any quote-stuffing event that occurs within a [-3, +3]-day window surrounding an earnings or dividend announcement as identified in CRSP and Compustat.

Finally, we eliminate events if there is an influx in trading in the 10 minutes before the spike in quoting activity. The influx in trading restriction is implemented to eliminate large episodes of quote updating driven by increases in trading. Additionally, increases in liquidity-demanding trades may inflate market-quality measures. Filtering events near earnings and dividend announcements or with increased trading in the minutes before the influx in quoting activity yields a final sample size of 24,733 events.

Figure 1 contains examples of three quote-stuffing events that we examine in more detail. These events are for Whirlpool Corporation (WHR) on January 15, 2010, Wright Medical Group Incorporated (WMGI) on June 14, 2010, and Harsco Corporation (HSC) on November 11,2010. Figure 1 shows the number of quotes for each minute of the trading day including the quote-stuffing event for each stock. As illustrated in Figure 1, our quote-stuffing events consist of very large spikes in quote activity.

[FIGURE 1 OMITTED]

Panel A of Table I reports summary statistics for sample firms that undergo at least one quote-stuffing event during the year. Mean daily volumes of shares traded range from 720 to 211 million, with a mean of 877,000 thousand shares. Sample firm sizes also span a large range from $520,000 to $273 billion. Median closing price and daily returns are $16.16 and 0.05%, respectively.

The magnitudes of quotes during our intense quoting events range from 20 to 925 standard deviations above the previous 20-day average, 36% of the events fall between 20 and 30 standard deviations, and an additional 42% of events occur between 30 and 40 standard deviations (see Panel B of Table I). Panel C of Table I lists the number of events by duration. The majority (72%) of events last less than one minute with over 94% lasting less than six minutes.

Several additional summary statistics are tabulated in Panel D of Table I. First, large spikes in quoting activity occur relatively frequently with an average of roughly 125 such events occurring each day. These large spikes in activity also affect a large number of firms; 5,292 or roughly 74.7% of all US-listed equities experience at least one event during the 2010 trading year. During the events, there is a mean of 7,010 quote updates per minute.

III. Quote Stuffing and Market Quality

A. Measures of Market Quality

We use TAQ data to compute several measures of market quality for each minute in the 10-minute window immediately before and after the quote-stuffing event. Our measures of market quality include two measures of short-term volatility and three measures of liquidity. Voltil is the one-minute standard deviation of trade prices. HighLow is an alternative measure of short-term volatility, which is the highest quoted midpoint in the one-minute interval minus the lowest quoted midpoint in the interval (this measure is similar to the HighLow measure of Flasbrouck and Saar, 2013). We use quoted, percentage-quoted, and effective spreads (QSprd, Pqsprd, and Effsprd) to measure liquidity. Qsprd is the average spread (ask price minus bid price) of the one-minute interval. Pqsprd is the quoted spread scaled by the midpoint, [(ask - bid)/(ask + bid/2)], then averaged over the one-minute interval. Effsprd is a measure of the price impact of a trade and is computed as the average effective half-spread (absolute value of the trade price minus the prevailing midpoint) of all trades during the one-minute interval.

Figures 2 and 3 graphically depict, and Table II reports, mean market-quality statistics for the quote-stuffing interval (time 0), the 10 minutes before (time -10 to -1), and the 10 minutes immediately following (time +1 to +10) the events. All three measures of liquidity (Qsprd, Pqsprd, and Effsprd) remain relativity constant in the minutes before the influx of quoting activity then abruptly increase during the event window. In the minutes following the event, both Qsprd and Pqsprd decline gradually until reaching their pre-event average in minute four. In the pre-event window Effsprd follows a pattern similar to Qsprd and Pqsprd, remaining relatively constant before increasing sharply to $0.04. In the minutes following the event period, Effsprd declines, but unlike Qsprd and Pqsprd, it remains elevated, not dropping below $0,026 in minutes + 1 to +10.

Volatility measures also follow patterns similar to those of the liquidity measures, increasing sharply during the event period. Voltil begins increasing in minute -2 and declines to its pre-event window average by minute +5. HighLow rises from a minute -10 level of $0,025 to an event-period level of $0,061 and subsequently declines to a minute +10 level of $0,026.

[FIGURE 2 OMITTED]

The identified intense episodes of quoting activity are associated with decreased liquidity--higher trading costs and increased short-term volatility. We also standardize quotes and trades to gain more insight into trading and quoting during these quote-stuffing events. Figure 4 displays the standardized quotes and trades in the window [-30, +30] surrounding the quote-stuffing events. As we expect, quoting activity peaks at minute 0 at a level of more than four times the pre and post 30 one-minute averages. Trading activity peaks at minute +1 and remains elevated through minute +10.

B. Regression Results

To further explore quote stuffing and market quality, we run a series of panel regressions that control for factors that may be associated with market quality. Each regression uses data from the event period as well as from the 10 one-minute periods immediately preceding (preperiods) and following (postperiods) the event. We estimate the following equation to test for a relation between quote stuffing and effective spread:

[Effsprd.sub.i,t] = [[beta].sub.0] + [[beta].sub.1]Post + [[beta].sub.2]During + [[beta].sub.3][Midpvolit.sub.i,t] + [[beta].sub.4][Nts.sub.i,t] + [[epsilon].sub.i,t] (1)

where [Effsprd.sub.i,t] is the average effective half-spread for stock i in minute t; During is a dummy variable that equals one for event segments, and zero otherwise; Post is a dummy variable that equals one for the period following the event; [Midpvolit.sub.i,t] is the standard deviation of the quote midpoint for stock i in minute t, and [Nts.sub.i,t] is a measure of activity, computed as the number of trades that execute in minute t for stock i. We include event-window fixed effects, which uniquely identify each event window, for this model as well as all for subsequent regressions.

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

We estimate similar models to examine the association between quote stuffing and quoted and percentage quoted spreads:

[Qsprd.sub.i,t] = [[beta].sub.0] + [[beta].sub.1]Post + [[beta].sub.2]During + [[beta].sub.3][Midpvolit.sub.i,t] + [[beta].sub.4][Nts.sub.i,t] + [[beta].sub.4][Nts.sub.i,t] + [[epsilon].sub.i,t], (2)

[Qsprd.sub.i,t] = [[beta].sub.0] + [[beta].sub.1]Post + [[beta].sub.2]During + [[beta].sub.3][Midpvol.sub.i,t] + [[beta].sub.4][Nts.sub.i,t] + [[epsilon].sub.i,t], (3)

where [Qsprd.sub.i,t] and [Pqsprd.sub.i,t] are the average quoted and percentage quoted spreads for stock i in minute t, and all other variables are as previously described.

We also estimate the following model for one-minute [Voltil.sub.i,t]:

[Voltil.sub.i,t] = [[beta].sub.0] + [[beta].sub.1]Post + [[beta].sub.2]During + [[epsilon].sub.i,t] (4)

Table III presents the estimated coefficients for our market-quality regressions. The coefficient of primary interest is [[beta].sub.2], which measures the correlation of identified quote-stuffing events and market quality. The coefficient of During, [[beta].sub.2], is positive for all regression specifications. This positive coefficient suggests that intervals experiencing a large influx of quoting activity are associated with higher quoted and effective spreads and increased short-term volatility.

The coefficient of the Post dummy variable is positive in the Ejfsprd and Voltil regressions. However, this positive coefficient is nearly an order of magnitude smaller than the coefficient of During in both regressions. Our regressions suggest that in the postevent window, both effective spreads and short-term volatility remain slightly elevated compared to their pre-event levels.

Overall, our analyses imply that quote stuffing is adversely associated with traditional measures of market quality, regardless of the duration of the event, the market capitalization of the firm, or the listing exchange. Our results confirm that in periods of intense quoting activity, stocks experience decreased liquidity, higher trading costs, and increased short-term volatility.

C. Causes of Quote-Stuffing Events

Quote stuffing occurs for many reasons. In this section, we pinpoint the reasons for quote stuffing by classifying quote-stuffing events identified in Section II into four strategies. The first strategy, Type 1: Same-Stock Cross-Venue, is an event where quote stuffing is used to slow down other traders in the same stock across exchanges. The alleged purpose of this strategy is to create a latency arbitrage opportunity in the same stock across exchanges in which nonstuffing traders are required to process the barrage of quotes generated by the quote-stuffing trader(s). A large number of order submissions may cause the exchange receiving the quotes to lag other exchanges, creating arbitrage opportunities. Quote-stuffing events are classified as Type 1 if more than 50% of quote updates occur on a venue that does not also have the highest number of trades. Events identified using the Type 1 procedure are events in which quote stuffing is occurring on one exchange while trading is occurring on another, implying that the traders may be lagging one exchange to trade on another venue.

The second strategy, Type 2: Multistock Same-Venue, is an event where multiple stocks on the same exchange experience quote stuffing simultaneously. The purpose of this quote-stuffing strategy is to slow down the infrastructure of the stock exchange to trade other stocks on the same exchange. Quote-stuffing events are classified as Type 2 when two or more quote-stuffing events occur simultaneously on one exchange with more than 50% of quote updates in the affected stocks.

The third strategy, Type 3: Liquidity Consuming, is an event where a trader attempts to trade a relatively large quantity of one stock by consuming liquidity at several exchanges simultaneously. The trader may quote stuff to slow down multiple exchanges, such that a trade on one venue will not be followed immediately by cancellations of outstanding limit orders on other venues. Quote-stuffing events are classified as Type 3 when the distribution of quote updates and trades are relatively evenly dispersed across multiple exchanges and no single exchange has more than 33% of quote updates or trades.

Quote stuffing may also occur to create arbitrage opportunities between exchange-traded funds (ETFs) and the ETF's constituent securities. We identify all ETFs and their constituent securities using data provided by MasterDATA. The fourth strategy, Type 4: ETF, is an event where an ETF and one or more of the ETF's constituents have simultaneous quote-stuffing events.

We report the number of each type of event in Table IV. We are able to definitively categorize 12,632 of the 24,623 identified quote-stuffing episodes as follows: 8,295 as Type 1, 3,688 as Type 2, 554 as Type 3, and 95 as Type 4. Thus, the most common quote-stuffing strategy involves slowing down other traders in the same stock across exchanges.

D. Quoting Behavior during Quote-Stuffing Event

Market centers post new quotes to the consolidated data feed as a result of a change in the bid price, ask price, bid size, or ask size. To further explore periods of quote stuffing, we determine whether quotes that occur during quote-stuffing events are the result of an update to the quoted bid or ask price or to the quoted depth. We believe that identifying a pattern associated with these events (if one exists) will increase our understanding of the market's reaction to intense quoting episodes. To assess market reaction to quote-stuffing events, we count the number (and percentage) of quote updates, that is, quotes that are not the same as the previous quote. A quote update on the bid (ask) side of the quote is defined as a change in the bid price (ask price) or a change in the bid size (ask size) from the previous quote on the same exchange. We divide the number of bid (ask) updates by the total number of quote updates to obtain the proportion of quotes in which an update occurs on the bid (ask) side of the quote. Panel A of Table V reports both the number and the proportion of quote updates that occur on the bid side of the quote during quote-stuffing events. (9) The first thing to note in Panel A is the high proportion of events in which the majority of quote updates occur on only one side of the quote. In 27% of quote-stuffing events, less than 10% of the quote updates occur on the bid side. In 30% of the events, greater than 90% of the quote updates come from the bid side. This means that for the majority of the quote-stuffing events (roughly 57%), over 90% of the quote updates during the event occur on either the bid or ask side of the quote. This implies that quote stuffing is often isolated to only one side of the book.

Panel B of Table V reports the number (and the percentage) of bid and ask runs that occur during a quote-stuffing event. A bid (ask) run is a series of sequential quotes from the same exchange that are bid (ask) side updates. The size of a run is determined by the number of bid or ask updates in a series. A bid (ask) side run ends when a new quote is generated that does not update the bid (ask) side of the quote. # of Bid (Ask) Side Updates is the number of bid (ask) runs of different lengths. Percentage of Bid (Ask) Updates is the proportion of total bid (ask) updates that are part of runs of different lengths. The most frequent run length on both the bid side and ask side is 1-10. The largest percentage of both bid updates and ask updates are in runs of 300+, as both bid side and ask side have approximately 30,000 runs of 300+ updates. In summary, quote update runs tend to be either short, in runs of 10 or less, or long, in runs of 300 or more.

Panel C of Table V shows Percentage of Bid (Ask) Updates runs before, during, and after the quote-stuffing events. Pre is the 10-minute period before a quote-stuffing event, and Post is the 10-minute period after a quote-stuffing event. Short runs of 1-10 quote updates occur more frequently before the event for both bid and ask side updates. Long runs of 301 + updates are more common during the event than before the event. These long-run updates are also more common in the postperiod than in the preperiod.

Given that there are so many long bid (ask) side runs during and after quote-stuffing events, we look at the various exchanges reporting quoting activity in the security with the quote-stuffing event to determine whether the events are isolated on an exchange (our first look at whether a trade might be "walking the book"). Table VI reports the percentage of quotes on each exchange during the quote-stuffing events.

Large increases in quotes associated with an episodic quoting event tend to concentrate on a particular exchange. We deem an event as occurring on a particular exchange if the exchange has the greatest proportional increase in quoting activity during an event. We then group quote-stuffing events by the exchanges where the events occur. Table VI reports 345 quote-stuffing events occur largely on the American Stock Exchange (AMEX), 4,748 on the NYSE, 7,144 on ARCA, and 6,928 on NASDAQ. When a quote-stuffing event occurs, most of the increase in quotes comes from the exchange where the quote-stuffing event occurs. But the events are not isolated on a particular exchange; there is significant quoting activity on the other exchanges. NYSE, ARCA, NASDAQ, and BATS are the reporting venues where most of the quote-stuffing events occur. During an NYSE quote-stuffing event, 85.8% of the quotes are reported by the NYSE, but 6.2% are reported by ARCA and 3.7% by NASDAQ. During a NASDAQ quote-stuffing event, 64.9% of the quotes are reported by NASDAQ, but 16.3% are reported by ARCA and 7.6% by BATS. During a BATS quote-stuffing event, 62.9% of the quotes are reported by BATS, but 17.2% are reported by ARCA and 10.3% by NASDAQ. (10) Although all venues in Table VI report quotes during the events, ARCA, NASDAQ, and BATS show the most quoting activity in events occurring on other exchanges.

Panel B of Table VI displays the exchanges with the most quotes during a quote-stuffing event for each event type. We know from Table IV that Type 1 events, which involve quote stuffing on one exchange while trading occurs on other exchanges, are the most prevalent quote-stuffing events. Panel B shows that, except for PSX, Type 1 events are the most prevalent events for all exchanges. Type 1,3, and 4 events occur most frequently on NASDAQ, and Type 2 events occur most frequently on ARCA (36.0%), followed by NASDAQ (23.9%), and NYSE (25.0%).

E. Trading Behavior during a Quote-Stuffing Event

We look at trading on the exchanges during quote-stuffing events. (11) Panel A of Table VII reinforces the finding from Table IV that Type 1 events are the most prevalent of the categorized events. During an NYSE quote-stuffing event, 30.5% of trades are reported by the NYSE, 26.2% by the NASD, 13.9% by NASDAQ, and 13.1% by ARCA. During a NASDAQ quote-stuffing event, 29.1% of trades are reported by NASDAQ, 29.2% by the NASD, 8.8% by the NYSE, and 17.7% by ARCA. These results show that although trading occurs on the exchange where the quote-stuffing event occurs, trading occurs on other exchanges as well.

It is possible that a quote-stuffing event arises from a trader submitting a large volume of orders, thereby requiring other traders to process these orders and yielding an advantage for the submitting trader (Brogaard, 2010). If Brogaard's (2010) assertion is correct, we expect to see increases in trading on the exchange where the quote-stuffing event is identified, which is what is reported in Panel A of Table VII. However, Panel A also reports trading on exchanges other than the exchange where the quoting event is occurring. Given that trading normally takes place on multiple venues, we look at the increase in volume during the quote-stuffing event relative to the volume executing in the 10 minutes before the event to determine whether an abnormal amount of trading is taking place on any one venue (Panel B of Table VII).

We limit our discussion to the exchanges with the most quote-stuffing events: NYSE, ARCA, NASDAQ, and BATS. We find an increase in trading for the quote-stuffing exchange as well as for other exchanges during the quote-stuffing events. We identify a large number of quote-stuffing events where the highest intensity of quoting occurs on one exchange, while traders access liquidity in multiple locations (Type 1 events). However, from an aggregate point of view, we see that 30.5% of trades that occur during an NYSE quote-stuffing event are reported by the NYSE, 26.2% by the NASD, 13.9% by NASDAQ, and 13.1% by ARCA. During a NASDAQ quote-stuffing event, 29.1% of trades are reported by NASDAQ, 29.2% by the NASD, 8.8% by the NYSE, and 17.7% by ARCA.

We note that the percentage change in volume on the NYSE during an NYSE quote-stuffing event compared to the 10 minutes before the event is lower (14.7%) than that of many of the other exchanges during a quote-stuffing event on those exchanges (396.8% for ARCA, 48.9% for NASDAQ, and 37.8% for BATS).

F. Orders during a Quote-Stuffing Event

We use NASDAQ TotalView-ITCH data to get more detailed information during these quote-stuffing events. TotalView-ITCH data contain order-level data for all stocks that execute on NASDAQ. Using TotalView-ITCH data allows us to see the entire order book, unlike TAQ data where only the top of the book is visible. Panel A of Table VIII provides NASDAQ order statistics for all NYSE and NASDAQ stocks that execute on NASDAQ. Several statistics increase during our extreme quoting events: messages per second, number of new orders, canceled orders, and order cancellation rate. New orders increase from 2.03 per second to 23.92 during the event, whereas canceled orders increase from 1.4 per second to 13.93 per second. These results confirm that there is a high degree of correlation between top-of-the-book quoting activity from TAQ and full-order book activity on NASDAQ using the ITCH data. Panels B and C report the statistics for the NYSE/ARCA- and NASDAQ-listed stocks separately. The statistics are similar during the quote-stuffing events for stocks regardless of listing--large increases in messages, orders, and canceled orders.

One concern regarding the number of same-side quote updates (Table V) is that large orders can simply be walking the book, generating updated quotes as the depth at the current quote is exhausted. However, the average order size, shown in Table VIII, alleviates this concern as the order size significantly decreases during extreme quoting events. Average order size decreases for both NYSE/ARCA and NASDAQ stocks. Other statistics, such as order execution rate and order duration, decrease during quote-stuffing events as well.

Table IX reports the locations of order updates relative to the inside quote for NASDAQ. An order update is an incoming message that creates a new order, cancels an existing order, or executes an existing order. At Inside is the percentage of order updates that occur at the inside bid price for buy orders or the inside offer price for sell orders. >$0.01 and [less than or equal to] $0.02 is the percentage of order updates that occur more than one cent but less than or equal to two cents away from the inside quote; other variables (>$0.2 and [less than or equal to] $0.05, etc.) are similarly defined.

It is not surprising that most order update activity occurs near the inside price before, during, and after the quote-stuffing event. However, during the quote-stuffing event the proportion of order update activity that occurs near the inside price increases. At Inside >$0.01 and [less than or equal to] $0.02 is 2.8% (2.1%) higher during the quote-stuffing period relative to the prequote-stuffing period. During event periods the proportion of order updates that occur between $0.02 and $0.25 of the inside bid or ask prices decreases. In the postevent window the location of order updates shifts back to proportions similar to the preevent window.

IV. Matching Sample (Robustness)

For robustness, we further explore the effect of quote stuffing by adding to the sample a group of stocks that do not experience a quote-stuffing event (nonquote-stuffing stocks). We obtain our matched sample by identifying quote-stuffing and nonquote-stuffing stock pairs. To identify the control sample we use a weighting scheme similar to that employed by Huang and Stoll (1996), Bessembinder (1999), Bessembinder and Kaufman (1997a,b), and Chung, Van Ness, and Van Ness (1999). Specifically, we form quote-stuffing and nonquote-stuffing stock pairs by calculating the following score:

[summation over k] [[[X.sup.S.sub.ki] - [X.sup.C.sub.kj]/{([X.sup.S.sub.ki] + [X.sup.S.sub.kj]/2}].sup.2],

where [X.sup.S.sub.ki] is firm characteristic k for firm i that experiences a quote-stuffing event, and [X.sup.S.sub.kj] is firm characteristic k for firm j that does not experience a quote-stuffing event. As suggested by Davies and Kim (2009), we use the average stock price and market capitalization in the 30-day period immediately preceding the quote-stuffing event date for firm characteristics. Matched pairs are identified by selecting the NYSE/ARCA- and NASDAQ-listed nonquote-stuffing stock with the smallest score to match each quote-stuffing stock.

Table X reports summary statistics for quote-stuffing and nonquote-stuffing stocks. Mean market capitalization, daily trading volume, returns, and price are similar for quote-stuffing and nonquote-stuffing stocks. T-tests for differences in means between quote-stuffing and nonquote-stuffing stocks do not yield statistically significant results, suggesting that our matching procedure is able to identify firms with similar characteristics.

Table XI reports market-quality statistics for quote-stuffing and nonquote-stuffing stocks, and it tests for differences in market-quality statistics for the stock pairs for each period surrounding the quote-stuffing event (similar to Table II). As expected, the difference in the mean number of quotes per minute (NQS) increases substantially for quote-stuffing stocks relative to nonquote-stuffing stocks during the event period, rising from a mean difference of 59 quotes in minute--10 to a difference of 6,587 quotes during the event period. Following the event, NQS declines to a difference of 66 quote updates in minute +10.

The difference in other market-quality statistics for quote-stuffing and nonquote-stuffing stocks also increases during the event period. The difference between Qsprd for quote stuffing and nonquote-stuffing stocks more than doubles from minute--10 ($0.24) to the event period ($0.051). After the event, the difference in Qsprd declines to a level similar to its pre-event level. The difference in Pqsprd follows a pattern similar to Qsprd. The difference in quote midpoint volatility between quote-stuffing and nonquote-stuffing stocks, as measured by HighLow, also increases during event periods. The difference between HighLow for quote-stuffing and nonquote-stuffing stocks is over seven times higher during the event window compared to the pre- and postevent periods. (12)

Results comparing quote-stuffing stocks and control stocks are consistent with market quality diminishing during quote-stuffing events. The matched sample results also suggest that market-quality degradation observed during quote-stuffing events is the result of events and not other exogenous factors.

V. Conclusion

In this study, we analyze intense episodic spikes in quoting activity and market quality, including liquidity and volatility. We find that quote stuffing is pervasive, affecting over 74.7% of US-listed equities during our sample period. Our results show that in periods of intense quoting activity, stocks experience decreased liquidity, higher trading costs, and increased short-term volatility. We find that most quote-stuffing events occur on the NYSE, ARCA, NASDAQ, and BATS and that increased trading takes place on multiple exchanges during these intense quoting events. Orders are entered at higher rates, are canceled at higher rates, are for shorter durations, are executed at lower rates, and are smaller during quote-stuffing events. Our study suggests that intense episodes of quoting activity is associated with degraded market quality.

We thank an anonymous referee, Marc Lipson (Editor), Robert Battalio, Kathleen Fuller, Michael Goldstein, Andreas Heinen, Pawan Jain, Michael Lewis, Ananth Madhavan, Sandra Mortal, Pamela Moulton, Tom McInish, Andriy Shkilko, Jim Upson, Bob Wood, Adam Yore, and seminar participants at Auburn University, Louisiana Tech University, The University of Mississippi, Mississippi State University, Ohio University, the University of Richmond, Utah State University, the Eastern Finance Association annual meeting (2013), the Southern Finance Association annual meeting (2012), the Midwest Finance Association annual meeting (2014), the Magnolia Finance Conference (2015), and the Financial Management Association annual meeting (2012).

References

Aggarwal, R.K. and G. Wu, 2006, "Stock Market Manipulations," Journal of Business 79, 1915-1953.

Allen, F. and G. Gorton, 1992, "Stock Price Manipulation, Market Microstructure and Asymmetric Information," European Economic Review 36, 624-630.

Bessembinder, H., 1999, "Trade Execution Costs on NASDAQ and the NYSE: A Post-Reform Comparison," Journal of Financial and Quantitative Analysis 34, 387-407.

Bessembinder, H., 2003, "Quote-Based Competition and Trade Execution Cost in NYSE-Listed Stocks," Journal of Financial Economics 70, 385-422.

Bessembinder, H. and H. Kaufman, 1997a, "A Cross-Exchange Comparison of Execution Costs and Information Flow for NYSE-Listed Stocks," Journal of Financial Economics 46, 293-319.

Bessembinder, H. and H. Kaufman, 1997b, "A Comparison of Trade Execution Costs for NYSE and NASDAQ-Listed Stocks," Journal of Financial and Quantitative Analysis 32, 287-310.

Biais, B. and P. Woolley, 2011, "High Frequency Trading," Toulouse University Working paper.

Brogaard, J. A., 2010, "High Frequency Trading and Its Impact on Market Quality," Northwestern University Working paper.

Cartea, A. and J. Penalva, 2011, "Where is the Value in High Frequency Trading?" Universidad Carlos III de Madrid Working paper.

Chung, K., B. Van Ness, and R. Van Ness, 1999, "Limit Orders and the Bid-Ask Spread," Journal of Financial Economics 53, 255-287.

Cvitanic, J. and A. Kirilenko, 2010, "High Frequency Traders and Asset Prices," Cal Tech Working paper.

Davies, R. and S. Kim, 2009, "Using Matched Samples to Test for Differences in Trade Execution Costs," Journal of Financial Markets 12, 173-202.

Elder, J., 2010, "Quote Stuffing Is One More Reason Not to Invest in Stocks," Wall Street Journal, September 8, A18.

Financial Industry Regulatory Authority, 2010, "FINRA Sanctions Trillium Brokerage Services, LLC, Director of Trading, Chief Compliance Officer, and Nine Traders $2.26 Million for Illicit Equities Trading Strategy," September 13, Release No. P121951.

Goldstein, I. and A. Guembel, 2008, "Manipulation and the Allocational Role of Prices," Review of Economic Studies 75, 133-164.

Gomber, P., B. Arndt, M. Lutat, and T. Uhle, 2011, "High Frequency Trading," Goethe University Working paper.

Hasbrouck, J. and G. Sarr, 2013, "Low-Latency Trading," Journal of Financial Markets 16, 646-679.

Hendershott, T., C.M. Jones, and A.J. Menkveld, 2011, "Does Algorithmic Trading Improve Liquidity?" Journal of Finance 66, 1-33.

Hendershott, T. and R. Riordan, 2009, "Algorithmic Trading and Information," NET Institute Working paper.

Hirschey, N., 2013, "Do High-Frequency Traders Anticipate Buying and Selling Pressure?" London School of Business Working paper.

Huang, R. and H. Stoll, 1996, "Dealer versus Auction Markets: A Paired Comparison of Execution Costs on NASDAQ and the NYSE," Journal of Financial Economics 41, 313-357.

Jarrow, R.A., 1992, "Market Manipulation, Bubbles, Comers, and Short Squeezes "Journal of Financial and Quantitative Analysis 27, 311-336.

Jones, C., 2013, "What Do We Know about High-Frequency Trading?" Columbia University Working paper.

Kirilenko, A., A. Kyle, M. Samadi, and T. Tuzun, 2010, "The Flash Crask: The Impact of High Frequency Trading on an Electronic Market," US Commodity Futures Trading Commission Working paper.

Kumar, P. and D.J. Seppi, 1992, "Futures Manipulation with 'Cash Settlement,"' Journal of Finance 47, 1485-1502.

Lauricella, T. and J. Strasburg, 2010, "SEC Probes Canceled Trades," Wall Street Journal, September 2, A1.

Madhavan, A., 2012, "Exchange-Traded Funds, Market Structure and the Flash Crash," Blackrock Working paper.

Mei, J., G. Wu, and C. Zhou, 2004, "Behavior Based Manipulation," New York University Working paper.

Menkveld, A. J., 2011, "High Frequency Trading and the New-Market Makers," VU University of Amsterdam Working paper.

O'Hara, M. and M. Ye, 2011, "Is Market Fragmentation Harming Market Quality?" Journal of Financial Economics 100, 459-474.

Salmon, F., 2010, "Trillium Wasn't Quote Stuffing," Reuters, September 14.

Schapiro, M., 2010, "Remark before Economic Club of New York," September 7.

Securities and Exchange Commission, 2010, "Concept Release on Equity Market Structure," January 13, Release No. 34-61358.

Ye, M., C. Yao, and J. Gai, 2012, "The Externality of High Frequency Trading," University of Illinois Working paper.

Zhang, X.F., 2010, "High-Frequency Trading, Stock Volatility, and Price Discovery," Yale University Working paper.

Jared F. Egginton, Bonnie F. Van Ness, and Robert A. Van Ness **

* Jared F. Egginton an Assistant Professor at Louisiana Tech University in Ruston, LA. Bonnie F. Van Ness is a Professor of Finance at The University of Mississippi in University, MS. Robert A. Van Ness is a Professor of Finance at The University of Mississippi in University, MS.

(1) HFT is a subset of algorithmic trading, where algorithmic trading is broadly defined as the use of a computer algorithm to automatically submit, cancel, and otherwise manage orders.

(2) Brogaard (2010) estimates that HFT makes up 77% of dollar trading volume in US equities.

(3) See Gomber et a!. (2011) and Brogaard (2010) for detailed descriptions of HFT strategies.

(4) Biais and Woolley (2011) also discuss high-frequency traders using quote stuffing to create congestion in the market by submitting a large number of orders to the market and thus impairing the market for slow traders.

(5) Liquidity-demanding trades are trades that occur via marketable orders (i.e., market orders, limit orders to buy above the current ask, or limit orders to sell below the current bid). Liquidity-supplying trades are trades from nonmarketable orders (i.e., limit orders to sell above the current bid or limit orders to buy below the current ask). Marketable orders take liquidity from the market whereas nonmarketable orders add liquidity.

(6) Ye, Yao, Gai (2012) examine orders only on NASDAQ (using NASDAQ TotalView-ITCH data, which do not have orders that originate on other exchanges). We use trades and quotes from all exchanges for most analyses (identification and impact of quote-stuffing events, etc.), but we also use NASDAQ TotalView-ITCH in our last analysis of orders during identified quote-stuffing events.

(7) We check instances where Hasbrouck and Saar (2013) identify an elevated number of "strategic runs"; all instances are marked with a substantial increase in quoting activity reported in TAQ.

(8) Events with durations longer than 10 minutes are excluded from the sample.

(9) We report only the proportion of updates that occur on the bid side because all quote updates occur on either the bid or ask side of the quote; therefore, the proportion of bid and ask quote updates must sum to one.

(10) We recognize that not all rows in Table VI sum to 100%. There are two reasons for this. First, we report averages of averages (we average first by firm, then by exchange). Therefore, the sum of the averages of averages is not necessarily 100%. Second, we do not report the Chicago Stock Exchange in this table for brevity and because it does not have 1% of the quotes in our sample.

(11) TAQ lists the exchange where the quote or trade is reported and not the trade-reporting facility (see O'Hara and Ye, 2011, for more on differences in exchanges and trade reporting facilities). We recognize this limitation of our data but believe our analysis will indicate whether trading is occurring on or away from the exchange where the quote-stuffing event occurs.

(12) We do not compare Effsprd and Voltil for matched pairs because a large number of nonquote-stuffing stocks are lost as a result of nonquote-stuffing stocks not trading every event minute.
Table I. Summary Statistics

This table presents summary statistics for sample firms and events.
The sample period is from January 2010 to December 2010 and
includes all stocks that experience at least one period of intense
quoting activity (quote-stuffing event) during that time frame. In
Panel A, MktCap is the average market capitalization for sample
firms (in $ millions), Daily Volume is the average daily volume (in
thousands), Return is the average daily close-to-close return, and
Closing Price is the average closing price. All statistics in Panel
A are computed as daily averages for the 2010 trading year and are
computed using CRSP data. All averages are computed on an
individual stock basis and then averaged across stocks. Panel B
presents information on the distribution of the magnitude of
events. Events are defined as episodic spikes in quoting activity
in which the level of quoting activity exceeds the previous 20-day
mean number of quotes per minute by at least 20 standard
deviations. The number of events that are between 20-30, 30-50,
50-100, 100-250, and >250 SDs of their previous 20-day mean number
of quotes per minute are reported. Panel C lists the number of
events by duration and their cumulative distribution. Panel D lists
other quote-stuffing events characteristics.

Panel A. Firm Characteristics

                         Mean    Median       SD      Min.       Max.

MktCap ($ millions)     2,378    3,626    10,349      0.52    273,560
Daily Volume (1,000s)     877      173     4,747      0.72    211,465
Return (%)              0.047    0.051     0.140    -2.166      5.683
Closing Price ($)       25.47    16.16     39.82      0.09    1891.79

Panel B. Quote-Stuffing Events

No. of SD above mean               No. of events     Cumulative percent

20-29                                  8,947               36.2%
30-49                                 10,463               78.5%
50-99                                  4,132               95.2%
100-249                                1,089               99.6%
>250                                    102                100.0%

Panel C. Quote-Stuffing Events Duration

No. of events                    Length in minutes   Cumulative percent

17,893                                  <1                 72.3%
3,074                                   1-2                84.7%
1,171                                   3-4                89.5%
702                                     4-5                92.3%
503                                     5-6                94.3%
437                                     6-7                96.1%
292                                     7-8                97.3%
256                                     8-9                98.3%
219                                    9-10                99.2%
186                                    10-11              100.0%

Panel D. Other Quote-Stuffing Events Characteristics

Mean number of events per day                                125
No. of stocks with an event                                5,292
% of US equities with at an event                          74.7%
During an events mean no. of quote updates per minute      7,010

Table II. Market Quality Stats by Period (Expanded)

This table reports mean market-quality statistics for the
quote-stuffing interval (time 0) and the 10 minutes before (time
-10 to -1) and the 10 minutes immediately following (time +1 to
+10) the event. Voltil is the one-minute standard deviation of
trade prices, HighLow is the highest quoted midpoint in the
one-minute interval minus the lowest quoted midpoint in the
interval, Qsprd is the average spread (ask price minus bid price)
of the one-minute interval, Pqsprd is the spread scaled by the
midpoint [(ask - bid)/(ask + bid/2)] and then averaged over the
one-minute interval, and Effsprd measures the price impact of a
trade and is computed as the average effective half-spread
(absolute value of the trade price minus the prevailing midpoint)
of all trades during the one-minute interval. Tests for
significance are calculated using the time-series median of the
variables. We calculate the per minute time-series median for all
variables. We then use the time-series median and variance to test
for significance of the variables surrounding the event window.

            Qsprd       Pqsprd     Effsprd     Voltil     HighLow

-30 to -21  $0,099      0.008      $0,016      0.013      0.032
-20 to -11  $0,083      0.006      $0,019      0.013      0.027
-10         $0,082      0.006      $0,019      0.012      0.025
-9          $0,080      0.006      $0,020      0.011      0.025
-8          $0,080      0.006      $0,020      0.011      0.025
-7          $0,081      0.006      $0,021      0.012      0.026
-6          $0,082      0.006      $0,021      0.012      0.026
-5          $0,081      0.006      $0,021      0.012      0.026
-4          $0,082      0.006      $0,022      0.014      0.027
-3          $0,084      0.006      $0,023      0.012      0.029
-2          $0,086      0.006      $0,024      0.013      0.030
-1          $0,094      0.007      $0,027      0.015      0.040 **
0           $0.116 ***  0.009 ***  $0,039 ***  0.021 ***  0.061 ***
+1          $0,103 *    0.008 **   $0,033      0.015      0.039 *
+2          $0,092      0.007      $0,029      0.016 **   0.032
+3          $0,087      0.007      $0,028      0.016 **   0.030
+4          $0,086      0.006      $0,029      0.016 *    0.030
+5          $0,084      0.006      $0,027      0.013      0.028
+6          $0,082      0.006      $0,027      0.015      0.027
+7          $0,081      0.006      $0,027      0.015      0.027
+8          $0,079      0.006      $0,026      0.012      0.026
+9          $0,079      0.006      $0,027      0.013      0.026
+ 10        $0,080      0.006      $0,026      0.013      0.026
+11 to +20  $0,078      0.006      $0,026      0.011      0.023
+21 to +30  $0,075      0.005      $0,026      0.011      0.022

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table III. Regression Results

This table reports the results of regression analyses analyzing
quote stuffing and market quality. We use TAQ data to compute
several measures of market quality: Voltil is the one-minute
standard deviation of trade prices, HighLow is the highest quoted
midpoint in the one-minute interval minus the lowest quoted
midpoint in the interval, Qsprd is the average spread (ask price
minus bid price) of the one-minute interval, Pqsprd is the spread
scaled by the midpoint [(ask - bid)/(ask + bid/2)] and then
averaged over the one-minute interval, and Effsprd measures the
price impact of a trade and is computed as the average effective
half-spread (absolute value of the trade price minus the prevailing
midpoint) of all trades during the one-minute interval. The
following model is then estimated:

[Mkt Quality.sub.i,t] = [[beta].sub.0] + [[beta].sub.1] Post +
[[beta].sub.2] During + [[beta].sub.3] [Midpvolit.sub.i,t] +
[[beta].sub.4] [Nts.sub.i,t] + [[epsilon].sub.i,t],

where During is a dummy variable that equals one for event
segments, and 0 otherwise; Post is a dummy variable that equals one
for the period following the event; Midpvolit is the standard
deviation of the midpoint; and [Nts.sub.i,t] is the number of
trades executed in each minute. We include event window fixed
effects in each regression which uniquely identifies each event
window. T-stats are reported in parentheses and are based on
event-cluster-corrected robust standard errors.

                 Effsprd        Qsprd         Pqsprd        Voltil

Post             0.001 ***    -0.001 ***     -0.000 ***     0.002 ***
                (5.048)      (-2.780)       (-2.992)       (3.462)
During           0.003 ***     0.009 ***      0.001 ***     0.005 ***
               (10.108)      (13.500)       (11.367)       (4.966)
Midpvolit        0.031 ***     0.258 ***      0.012 ***
                (1.463)       (2.277)        (2.305)
Nts              0.000 ***    -0.000 ***     -0.000 ***
                (3.105)      (-2.938)       (-2.959)
Constant         0.016 ***     0.081 ***      0.006 ***     0.013 ***
               (64.459)      (85.904)      (131.768)      (37.243)
Observations    260,798       475,674       475,674        222,128
[R.sup.2]        0.69          0.89           0.89          0.28
F-test          30.28 ***     92.51 ***      70.75 ***     16.75 ***

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table IV. Quote-Stuffing Strategies

Type 1: Same-Stock Cross-Venue events are when more than 50% of
quote updates occur on one venue and that venue does not have the
highest number of trades. Type 2: Multistock Same-Venue events
are when two or more quote-stuffing events occur simultaneously
on one exchange and that one exchange has more than 50% of the
quote updates. Type 3: Liquidity Consuming events are when the
distribution of quote updates and trades are relatively evenly
dispersed across multiple exchanges and no single exchange has
more than 33% of quote updates or trades. Type 4: ETF are events
where an ETF and one or more of the ETF's constituent securities
have simultaneous quote-stuffing events.

Type of Event                    Number of Events

Total events                            24,733

Type 1: Same-Stock Cross-Venue           8,295

Type 2: Multistock Same Venue            3,688

Type 3: Liquidity Consuming                554

Type 4: ETF                                 95

Table V. Quote Side Updates during Quote-Stuffing Events

This table reports the proportion of quote updates that occur on
the bid and ask sides of the quote during quote-stuffing events. A
quote update on the bid (ask) side of the quote is defined as a
change in the bid price (ask price) or a change in the bid size
(ask size) from the previous quote on the same exchange. We divided
the number of bid (ask) updates by the total number of quote
updates to obtain the proportion of quotes in which the update
occurs on the bid (ask) side of the quote. Panel A reports the
proportion of quote updates that occur on the bid side of the quote
during quote-stuffing events. Note we only report the proportion of
updates that occur on the bid side because all quote updates occur
on either the bid or the ask side of the quote; therefore, the
proportion of bid and ask quote updates must cumulate to one. Panel
B reports the number of bid and ask side update runs. A bid (ask)
run is a series of sequential quotes from the same exchange as the
bid (ask) side updates. The size of a run is determined by the
number of bids or asks updates in a series. A bid (ask) side run
ends when a new quote is generated that does not update the bid
(ask) side of the quote. No. of Bid (Ask) Side Updates is the
number of bid (ask) runs of different lengths. Percentage of Bid
(Ask) Updates is the proportion of total bid (ask) updates that are
part of runs of different lengths. Panel C reports the percentage
of bid and ask side updates that are parts of runs. Pre is the
10-minute period that precedes the quote-stuffing event, During is
the period during the quote stuffing episode, and Post is the
10-minute window following the quote-stuffing event.

Panel A. Type of Quote Updates

                     Bid Side         Bid Side
Proportion of        Updates          Updates
Quotes Updates   (No. of Events)   (% of Events)

0-10%                  6,661             27%
10-20%                 1,081              4%
20-30%                   963              4%
30-40%                 1,322              5%
40-50%                 2,066              8%
50-60%                 2,039              8%
60-70%                 1,239              5%
70-80%                   972              4%
80-90%                   970              4%
90-100%                7,415             30%

Panel B. Runs of Quote Updates

                                Percentage     # of Ask     Percentage
                  No. of Bid      of Bid     Side Updates     of Ask
Runs             Side Updates    Updates       in a row      Updates

1-10              19,895,890      32.2%       20,074,373      32.6%
11-49                439,874       9.3%          409,691       8.6%
50-100                53,879       3.9%           52,314       3.9%
101-149               19,760       2.5%           19,599       2.5%
150-199               10,922       2.0%           10,642       1.9%
200-249                6,885       1.6%            6,661       1.5%
250-299                4,933       1.4%            4,800       1.4%
300+                  30,815      47.2%           29,908      46.2%

Panel C. Percentage Runs of Quote Updates

Runs      Percentage of Bid Updates   Percentage of Ask Updates

              Pre   During    Post      Pre   During    Post

1-10        48.4%    32.2%   35.2%      48%      33%     36%
11-50       15.4%     9.3%   11.1%      15%       9%     11%
51-100       4.9%     3.9%    3.5%       5%       4%      4%
101-150      2.8%     2.5%    2.1%       3%       3%      2%
151-200      2.0%     2.0%    1.6%       2%       2%      2%
201-250      1.6%     1.6%    1.3%       2%       2%      1%
251-300      1.3%     1.4%    1.1%       1%       1%      1%
301+        23.5%    47.2%   44.0%      24%      47%     43%

Table VI. Location of Quote Stuffing

This table reports the percentage of quote updates from each
exchange during a quote-stuffing event. Exchange identifies the
exchange with the most quotes during a quote-stuffing event. N is
the number of events in which the exchange (on far left of the
table) has the most quotes during a quote-stuffing event. The
table also shows the percentage of quotes each exchange has
during a quote-stuffing event. Panel B displays the exchange
with the most quotes during a quote-stuffing event by event type
(see Table VII).

Panel A. Percentage of Quotes during a Quote Stuffing Event
by Exchange

Exchange       N     AMEX    Boston   National     ISE     NYSE

AMEX         345    54.7%     0.0%       0.2%     1.3%     0.0%
Boston     1,180     0.1%    70.3%       0.3%     0.4%     1.2%
National     273     0.2%     2.1%      68.0%     0.6%     1.0%
ISE          314     0.8%     0.8%       0.3%    57.4%     2.4%
NYSE       4,748     0.0%     0.4%       0.1%     0.4%    85.8%
ARCA       7,144     0.4%     0.5%       0.3%     1.1%     2.3%
NASDAQ     6,928     0.7%     2.8%       0.6%     1.3%     2.1%
CBOE         485     1.6%     0.0%       0.3%     1.4%     1.0%
PSX           12     0.0%     4.6%       0.2%     0.0%     3.9%
BATS       3,194     0.2%     1.0%       0.4%     1.6%     1.5%

Exchange    ARCA    NASDAQ    CBOE        PXS     BATS

AMEX       21.5%    11.0%     4.3%       0.0%     4.4%
Boston      6.0%    17.3%     0.1%       0.0%     1.9%
National    4.7%     8.7%     1.3%       0.0%     5.8%
ISE        16.8%     8.6%     2.1%       0.0%    10.8%
NYSE        6.2%     3.7%     0.3%       0.0%     1.6%
ARCA       72.4%    12.4%     1.3%       0.0%     6.9%
NASDAQ     16.3%    64.9%     1.3%       0.0%     7.6%
CBOE        9.2%    11.8%    61.3%       0.0%    11.4%
PSX         3.9%     0.6%     0.0%      32.9%     1.1%
BATS       17.2%    10.3%     2.4%       0.0%    62.9%

Panel B. Location of Quote Stuffing by Event Types

               All Events           Type 1            Type 2

               N        %        N          %        N        %

AMEX         345     1.4%       27       0.3%        2     0.1%
Boston     1,180     4.8%      639       7.7%      135     3.7%
National     273     1.1%       85       1.0%       14     0.4%
ISE          314     1.3%       59       0.7%        3     0.1%
NYSE       4,748    19.3%    1,791      21.6%      924    25.0%
ARCA       7,144    29.0%    1,980      23.9%    1,331    36.0%
NASDAQ     6,928    28.1%    2,303      27.8%      977    26.5%
CBOE         485     2.0%      135       1.6%       18     0.5%
PSX           12     0.0%        2       0.0%        5     0.1%
BATS       3,194    13.0%    1,274      15.4%      284     7.7%

                 Type 3             Type 4

               N        %        N          %

AMEX           8     1.4%        0       0.0%
Boston        17     3.1%        9       9.5%
National      26     4.7%        2       2.1%
ISE           10     1.8%        1       1.1%
NYSE          59    10.6%        6       6.3%
ARCA         132    23.8%       18      18.9%
NASDAQ       160    28.9%       38      40.0%
CBOE          10     1.8%        0       0.0%
PSX            5     0.9%        0       0.0%
BATS         127    22.9%       21      22.1%


Table VII. Location of Trading during a Quote-Stuffing Event

This table reports the location of trading during a quote-
stuffing event. TV is the number of events in which the exchange
on far left side of the table had the most quotes during a quote-
stuffing event. Panel A shows the percentage of trades that occur
during a quote-stuffing event at the exchanges listed
horizontally. Panel B shows the change in the percentage of
trading volume for each exchange by comparing volume during the
quote-stuffing event with the 10-minute period before the quote-
stuffing event.

Panel A. Percentage of Volume

Exchange       N     AMEX    Boston    National     NASD       ISE

AMEX         345    48.0%      0.1%       0.6%     25.5%      1.6%
Boston     1,180     1.2%      3.8%       1.2%     28.4%      0.8%
National     273     0.9%      3.6%      14.3%     42.7%      0.6%
ISE          314     3.5%      0.9%       0.7%     53.1%      4.6%
NYSE       4,748     0.0%      1.0%       4.5%     26.2%      2.0%
ARCA       7,144     0.9%      1.0%       1.0%     31.9%      1.6%
NASDAQ     6,928     3.3%      1.2%       1.1%     29.2%      1.2%
CBOE         485     9.3%      0.8%       3.2%     24.4%      2.0%
PSX           12     0.0%      0.9%       2.8%     40.3%      0.0%
BATS       3,194     1.8%      1.5%       0.7%     29.9%      1.4%

Panel B. Change in Percentage of Volume

AMEX         345    108.0%    85.8%    -100.0%    -18.0%    -81.4%
Boston     1,180    142.5%    74.9%      62.3%      5.9%    122.1%
National     273    -57.2%    42.0%     559.4%     23.1%    111.3%
ISE          314    -32.7%    49.2%     -87.3%     25.5%    133.7%
NYSE       4,748     0.0%     -6.7%     222.4%     36.1%    124.6%
ARCA       7,144    44.9%     26.5%      38.0%     19.4%     28.9%
NASDAQ     6,928    92.5%     75.1%       9.5%     18.9%     26.7%
CBOE         485    42.3%    212.2%     106.7%    -14.2%    102.1%
PSX           12    -100%    -65.4%    1429.6%    -28.3%      0.0%
BATS       3,194    90.9%     21.3%      47.4%     15.9%     15.7%

Panel A. Percentage of Volume

Exchange   Chicago      NYSE       ARCA    NASDAQ       CBOE

AMEX          0.0%      0.0%      11.2%      9.9%       0.5%
Boston        0.1%      7.1%      17.6%     27.9%       0.0%
National      0.0%      7.5%       9.6%     13.9%       0.3%
ISE           0.0%      6.5%      12.6%     13.9%       1.1%
NYSE          1.2%     30.5%      13.1%     13.9%       0.1%
ARCA          0.1%     15.8%      19.7%     20.3%       0.3%
NASDAQ        0.1%      8.8%      17.7%     29.1%       0.2%
CBOE          0.4%      3.7%      14.1%     35.0%       2.8%
PSX           0.0%     21.7%      19.0%      9.1%       0.0%
BATS          0.1%     16.3%      17.8%     20.4%       0.1%

Panel B. Change in Percentage of Volume

AMEX       -100.0%      0.0%     -16.3%    -15.6%     -98.1%
Boston      -15.9%     16.8%      87.2%     44.7%      12.1%
National   -100.0%     23.3%       3.3%     63.4%    -100.0%
ISE        -100.0%     48.1%     247.3%      2.5%    -100.0%
NYSE        155.4%     14.7%      30.8%     31.5%     -33.7%
ARCA         82.4%     40.7%     396.8%     31.7%     106.3%
NASDAQ      -29.4%     24.6%      45.4%     48.9%     -36.5%
CBOE         -9.1%    -27.9%     -42.2%    -17.3%      80.2%
PSX           0.0%     57.2%     208.6%     -1.3%    -100.0%
BATS        -45.0%     19.9%      49.3%     32.7%     -50.8%

Panel A. Percentage of Volume

Exchange       PSX      BATS

AMEX          0.0%      2.6%
Boston        0.0%     11.8%
National      0.0%      6.6%
ISE           0.0%      3.1%
NYSE          0.1%      7.5%
ARCA          0.0%      7.5%
NASDAQ        0.0%      8.0%
CBOE          0.0%      4.3%
PSX           0.2%      6.1%
BATS          0.0%      9.8%

Panel B. Change in Percentage of Volume

AMEX          0.0%    -26.2%
Boston        3.2%     62.6%
National   -100.0%    -13.6%
ISE           0.0%     10.1%
NYSE         -9.9%     44.4%
ARCA         81.0%     45.9%
NASDAQ      -12.5%     56.1%
CBOE          0.0%    184.2%
PSX           4.0%      4.1%
BATS        -47.3%     37.8%

Table VIII. Order Statistics

This table reports information about order submission,
cancellation, and execution activity on NASDAQ for sample firms
surrounding quote-stuffing events. Data for order activity are
obtained from NASDAQ TotalView-ITCH. Order statistics are
reported for three periods: Pre is the 10-minute period that
precedes the quote-stuffing event, During is the period during
the quote-stuffing event, and Post is the 10-minute window
following the quote-stuffing event. Messages per second is the
mean number of messages per second reported in ITCH for sample
firms; New orders per second, Canceled orders per second, and
Executed orders per second are the mean number of new orders
submitted, canceled, and executed per second, respectively; New
buy (sell) orders per second is the average number of new buy
(sell) orders submitted per second; Order cancellation
(execution) rate is the average percentage of orders submitted
and subsequently canceled (executed) during a period; Order
duration is the average number of seconds an order is outstanding
before being canceled or executed; Cancellation (execution)
duration is the mean number of seconds an order is outstanding
before being canceled (executed); Odd-lot order is the percentage
of new orders submitted that are fewer than 100 shares; and Order
size is the average size in shares of newly submitted orders.
Differences in means of order statistics across periods are
reported in columns During-Pre, During-Post, and Post-Pre. Panel
A reports order statistics for all sample stocks, Panel B reports
statistics for NYSE-ARCA-listed stocks, and Panel C reports
statistics for NASDAQ-listed stocks.

                                  Pre     During    Post    During-Pre

Panel A. All Stocks

Messages per second                3.39   37.59     3.54     34.20 ***
New orders per second              2.03   23.92     2.10     21.89 ***
Canceled orders per second         1.40   13.93     1.48     12.53 ***
Executed orders per second         0.10    0.15     0.11      0.05 ***
New buy orders per second          1.14   12.84     1.14     11.70 ***
New sell orders per second         1.10   13.24     1.16     12.14 ***
Order cancellation rate (%)       57.00   72.00    56.80     15.40 ***
Order execution rate (%)           6.80    4.70     7.00     -2.10 ***
Order duration (second)            8.55    4.15     8.38     -4.40 ***
Cancellation duration (second)     8.63    4.19     8.45     -4.44 ***
Execution duration (second)        6.86    5.49     6.72     -1.36 ***
Odd-lot order (%)                 27.00   20.90    26.60     -6.10 ***
Order size                       421.65  362.00   412.49    -59.51 ***

Panel B. NYSE/ARCA-Listed Stocks

Messages per second                3.20   32.29     3.47     29.08 ***
New orders per second              1.78   17.76     1.90     15.98 ***
Canceled orders per second         1.46   14.82     1.60     13.37 ***
Executed orders per second         0.09    0.10     0.10      0.01 ***
New buy orders per second          1.00    9.21     1.01      8.21 ***
New sell orders per second         0.93   10.07     1.04      9.14 ***
Order cancellation rate (%)       55.50   70.70    55.60     15.20 ***
Order execution rate (%)           5.40    4.30     5.70     -1.10 ***
Order duration (second)            9.19    5.30     9.05     -3.89 ***
Cancellation duration (second)     9.25    5.35     9.12     -3.91 ***
Execution duration (second)        7.17    6.04     6.96     -1.13 ***
Odd-lot order (%)                 27.81   23.71    27.40     -4.10 ***
Order size                       575.08  512.00   566.73    -62.88 ***

Panel C. NASDAQ-Listed Stocks

Messages per second                3.71   48.62     3.88     44.91 ***
New orders per second              2.49   35.20     2.56     32.71 ***
Canceled orders per second         1.29   13.61     1.36     12.33 ***
Executed orders per second         0.12    0.22     0.14      0.10 ***
New buy orders per second          1.43   19.52     1.45     18.10 ***
New sell orders per second         1.36   18.75     1.41     17.39 ***
Order cancellation rate (%)       57.61   72.41    58.00     14.80 ***
Order execution rate (%)           8.51    4.91     8.50     -3.61 ***
Order duration (second)            7.63    2.69     7.41     -4.94 ***
Cancellation duration (second)     7.74    2.72     7.49     -5.02 ***
Execution duration (second)        6.41    4.92     6.33     -1.49 ***
Odd-lot order (%)                 25.01   16.51    24.61     -8.51 ***
Order size                       191.58  180.00   189.56    -11.62 ***

                                 During-Post    Post-Pre

Panel A. All Stocks

Messages per second               34.04 ***     0.16 **
New orders per second             21.82 ***     0.07
Canceled orders per second        12.45 ***     0.08 **
Executed orders per second         0.03 ***     0.02 ***
New buy orders per second         11.70 ***     0.00
New sell orders per second        12.08 ***     0.06 *
Order cancellation rate (%)       15.20 ***     0.20 ***
Order execution rate (%)          -2.30 ***     0.20 *
Order duration (second)           -4.23 ***    -0.17 ***
Cancellation duration (second)    -4.27 ***    -0.18 ***
Execution duration (second)       -1.22 ***    -0.14 ***
Odd-lot order (%)                 -5.70 ***    -0.40 ***
Order size                       -50.35 ***    -9.16 ***

Panel B. NYSE/ARCA-Listed Stocks

Messages per second               28.82 ***     0.27 ***
New orders per second             15.87 ***     0.11 **
Canceled orders per second        13.22 ***     0.15 ***
Executed orders per second         0.00         0.01 ***
New buy orders per second          8.20 ***     0.01
New sell orders per second         9.03 ***     0.11 **
Order cancellation rate (%)       15.10 ***     0.10
Order execution rate (%)          -1.40 ***     0.30 ***
Order duration (second)           -3.75 ***    -0.14 ***
Cancellation duration (second)    -3.77 ***    -0.14 ***
Execution duration (second)       -0.92 ***    -0.20 ***
Odd-lot order (%)                 -3.60 ***    -0.50 ***
Order size                       -54.54 ***    -8.35 ***

Panel C. NASDAQ-Listed Stocks

Messages per second               44.75 ***     0.17 **
New orders per second             32.64 ***     0.07
Canceled orders per second        12.25 ***     0.07 ***
Executed orders per second         0.08 ***     0.03 ***
New buy orders per second         18.08 ***     0.02
New sell orders per second        17.34 ***     0.05
Order cancellation rate (%)       14.31 ***     0.41 ***
Order execution rate (%)          -3.61 ***     0.01
Order duration (second)           -4.72 ***    -0.22 ***
Cancellation duration (second)    -4.78 ***    -0.24 ***
Execution duration (second)       -1.41 ***    -0.08
Odd-lot order (%)                 -8.11 ***    -0.41 **
Order size                        -9.59 ***    -2.02

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table IX. Order Book Updates

This table reports the locations of order updates relative to the
inside quote for NASDAQ. Data for order activity are obtained
from NASDAQ TotalView-ITCH. Order statistics are reported for
three periods: Pre is the 10-minute period that precedes the
quote-stuffing event, During is the period during the quote-
stuffing event, and Post is the 10-minute window following the
quote-stuffing event. At Inside is the percentage of order
updates that occur at the inside quote. >$0.01 and [less than
or equal to] $0.02 is the percentage of order updates that occur
more than one cent but less than or equal to two cents away from
the inside quote. Differences in means of order statistics across
periods are reported in columns During-Pre, During-Post, and
Post-Pre.

                                                          During-
                                  Pre    During   Post      Pre

At Inside (%)                    36.5     39.3    36.2     2 8 ***
>$0.01 and [less than or equal   21.1     23.2    21.1     2.1 ***
  to] $0.02 (%)
>$0.02 and [less than or equal   15.3     12.6    15.1    -2.6 ***
  to] $0.05 (%)
>$0.05 and [less than or equal   10.4      8.1    10.3    -2.3 ***
  to] $0.10 (%)
>$0.10 and [less than or equal    7.9      7.1     8.0    -0.8 ***
  to] $0.25 (%)
>$0.25 and [less than or equal    6.2      7.1     6.7     0.9 ***
  to] $1.00 (%)
>$1.00 (%)                        2.6      2.6     2.6     0.0

                                 During-
                                   Post     Post-Pre

At Inside (%)                     3.2 ***   -0.3 ***
>$0.01 and [less than or equal    2.1 ***    0.0
  to] $0.02 (%)
>$0.02 and [less than or equal   -2.4 ***   -0.2 ***
  to] $0.05 (%)
>$0.05 and [less than or equal   -2.2 ***   -0.1 ***
  to] $0.10 (%)
>$0.10 and [less than or equal   -0.9 ***    0.1 **
  to] $0.25 (%)
>$0.25 and [less than or equal    0.4 ***    0.5 ***
  to] $1.00 (%)
>$1.00 (%)                       -0.1 *      0.1 ***

*** Significant at the 0.01 level.

** Significant at the 0.05 level.

* Significant at the 0.10 level.

Table X. Match Summary Statistics

This table presents summary statistics for quote-stuffing stocks
(QS) and matched nonquote-stuffing (Non-QS) stocks. The sample
period is from January 2010 to December 2010. MktCap is the
average market capitalization for sample firms (in $ millions),
Daily Volume is the average daily volume (in thousands), Return
is the average daily close-to-close return, and Closing Price is
the average closing price.

                                                   Difference (QS -
                       QS-Stocks   Non-QS Stocks   Non-QS Stocks)

MktCap ($ millions)      2,378         2,398              -20
Daily Volume (1000s)       877           952              -75
Return (%)               0.047         0.048           -0.001
Closing Price ($)        25.47         25.73            -0.27

Table XI. Differences in QS Stock and Non-QS Stock Market-
Quality Stats by Period

This table reports difference-in-mean market-quality statistics
for quote-stuffing (QS) stocks and nonquote-stuffing (non-QS)
stocks. NQS is the mean number of quote updates per minute. Qsprd
is the average spread (ask price minus bid price) of the one-
minute interval, Pqsprd is the spread scaled by the midpoint
[(ask - bid)/(ask + bid/2)] and then averaged over the one-
minute interval, and HighLow is the highest quoted midpoint in
the one-minute interval minus the lowest quoted midpoint in the
interval.

                                                           [HighLow
      [NQS.sub.QS] -   [Qsprd.sub.QS] -   [Pqsprd.sub.    .sub.QS] -
      [NQS.sub.non-    [Qsprd.sub.non-    QS] -[Pqsprd   [HighLow.sub.
           QS]               QS]          .sub.non-QS]      non-QS]

-10      59 ***           0.024 ***        0.002 ***       0.005 ***
-9       55 ***           0.023 ***        0.002 ***       0.005 ***
-8       57 ***           0.023 ***        0.002 ***       0.005 ***
-7       74 ***           0.025 ***        0.002 ***       0.005 ***
-6       87 ***           0.027 ***        0.002 ***       0.006 ***
-5      103 ***           0.025 ***        0.002 ***       0.006 ***
-4      134 ***           0.027 ***        0.002 ***       0.007 ***
-3      177 ***           0.028 ***        0.002 ***       0.010 ***
-2      256 ***           0.031 ***        0.002 ***       0.010 ***
-1      815 ***           0.036 ***        0.003 ***       0.019 ***
0     6,587 ***           0.051 ***        0.004 ***       0.039 ***
+1      844 ***           0.042 ***        0.003 ***       0.019 ***
+2      287 ***           0.032 ***        0.003 ***       0.012 ***
+3      196 ***           0.030 ***        0.002 ***       0.011 ***
+4      144 ***           0.030 ***        0.002 ***       0.010 ***
+5      110 ***           0.028 ***        0.002 ***       0.009 ***
+6       99 ***           0.025 ***        0.002 ***       0.008 ***
+7       70 ***           0.025 ***        0.002 ***       0.007 ***
+8       71 ***           0.024 ***        0.002 ***       0.007 ***
+9       71 ***           0.024 ***        0.002 ***       0.006 ***
+10      66 ***           0.025 ***        0.002 ***       0.007 ***

*** Significant at the 0.01 level.
COPYRIGHT 2016 Financial Management Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Egginton, Jared F.; Van Ness, Bonnie F.; Van Ness, Robert A.
Publication:Financial Management
Article Type:Report
Date:Sep 22, 2016
Words:13899
Previous Article:Independent directors and favoritism: when multiple board affiliations prevail in mutual fund families.
Next Article:Momentum, reversals, and fund manager overconfidence.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters