Quote Stuffing
Jared Egginton
egginton@latech.edu
Louisiana Tech University
Bonnie F. Van Ness
bvanness@bus.olemiss.edu
University of Mississippi
Robert A. Van Ness
rvanness@bus.olemiss.edu
University of Mississippi
1
Quote Stuffing
Abstract
This study examines the impact of 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 experience at least one episode during 2010.
We find that stocks experience decreased liquidity, higher trading costs, and increased shortterm volatility during periods of intense quoting activity. We find that the most quote stuffing
events occur on the NYSE, ARCA, NASDAQ and BATS. We find that during these quote stuffing
events that the number of new orders and cancelled orders increases substantially while the
order size and order duration decrease.
2
1. Introduction
Quote stuffing is a 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 cancelations have come under scrutiny from the media and regulators. 1
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, page A. 1). Others have likened the practice to an auctioneer
placing “plants” and “shills” in the audience in an attempt to manipulate prices thru 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 paper seeks to address these questions.
The practice of quote stuffing is often linked to high frequency trading (hereafter, HFT).
HFT 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 trading strategy
where securities are rapidly purchased and sold through the use of computer algorithms.
Holding periods for securities bought and sold by high frequency traders are typically very
short, lasting just seconds or milliseconds. Further, high frequency traders may move in and
out of positions thousands of times per day. The Securities and Exchange Commission (SEC)
calls high frequency trading “One of the most significant market structure developments in
1
See for example Lauricella and Strasburg (2010).
1
recent years” SEC (2010). SEC chairwoman Mary Schapiro describes the regulatory scheme
that applies computer based low-latency trading as “[an] area that warrants close review”
(Schapiro, 2010). Today HFT makes up a significant portion of U.S. equities market volume.2
Despite the criticism of HFT 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, and 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 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 high frequency 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 come from manipulative
purposes. Large episodic spikes in quoting activity may be generated for technological reasons
2
Brogaard(2010) estimates that HFT makes up 77% of dollar trading volume in U.S. equities.
See Gomber, Arndt, Lutat, and Uhle (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.
3
2
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. While the intent of the large burst of message flow may not be part
of a nefarious plan to manipulate the market, these quoting episodes may still impact market
conditions.
In this study, we identify and analyze the impact that intense episodic spikes in quoting
activity have on market conditions, including liquidity and volatility. We find that episodes of
large bursts of quote updates are pervasive with events occurring each trading day and
impacting 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 related to the broader research on market manipulation (see Allen and
Gorton [1992], Jarrow [1992], Kumar and Seppi [1992], Mei, Wu, and Zhou [2004], and
Goldstein and Guembel [2008]). Aggarwal and Wu (2006) present theory 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 model
suggests a strong role for regulation to discourage manipulation.
In 2010 FINRA fined Trillium Brokerage Services, LLC and nine of its traders $2.26 million
for illicit high frequency trading strategies. FINRA accused Trillium Brokerage Services of
3
creating a “false appearance of buy- or sell-side pressure.” 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
would then immediately cancel orders that had been designed only to create the false
appearance of market activity. Trillium traders’ improper high frequency 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 remainder of the paper is structured as follows: Section 2 summarizes related
literature. Section 3 describes the data as well as the procedure we employ to identify quote
stuffing events. Section 4 outlines the methodology and results of our study of the impact of
quote stuffing intervals on traditional measures of market quality. Section 5 provides
discussion on the implications of the study’s results and concludes.
2. Background
This paper is most closely related to a small, but growing, body of literature that
addresses issues concerning high frequency and algorithmic trading (hereafter, AT).5
Hendershott, Jones, and Menkveld (2011) explain that declining technology costs, as well as
trading becoming increasing 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 U.S.
equities market volume (Hendershott, Jones, and Menkveld).
5
AT is broadly defined as the use of a computer algorithm to automatically submit, cancel, and otherwise manage
orders. HFT is a subset of AT.
4
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. 6 Algorithmic traders also contribute more to price discovery
than 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, Jones, and Menkveld (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.
Others project the impact of HFT on financial markets. Theoretical models of HFT
trading show that it is possible for HFT to 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
6
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
non-marketable 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 non-marketable orders add liquidity.
5
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 to have a moderate to significantly positive impact
on traditional market quality measures (see Jones, 2013, for a survey of current research on
high-frequency trading). Brogaard (2010) examines the impact of HFT on the US equities
market using a unique HFT dataset for 120 stocks listed on NASDAQ. Brogaard finds that HFT
improves market conditions. HFT 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 non-high frequency traders. Broggard’s
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
6
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 non-high 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 uses 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 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, Kyle, Samadi, and Tuzun (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. They conclude
that, while 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 prior to 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. Menkveld documents how one large high frequency trader
7
who acts as a market maker is critical to the operation of a new market, Chi-X. He provides a
detailed analysis on the trading behavior of the high frequency trader. The high frequency
trader provides liquidity and the entrance of the high frequency trader corresponds with a
decrease in spreads.
Ye, Yao, and Gai (2012) study an exogenous trading shock that increased 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 stating that it creates a false sense of the true supply
and demand for a stock and may adversely impact market quality. Also, unlike previous
empirical studies of HFT in U.S. equities markets, which use data from a single market center,
we examine quote stuffing behavior across all U.S. exchanges.7 Considering the fragmentation
of order flow in U.S. markets, we believe that we will glean a more complete picture of the
impact of quote stuffing on overall market conditions by using data from all US exchanges.
3. Data and Identification of Quote Stuffing
3.1 Data
7
Gai, Yao, and Ye (2012) examine orders on only NASDAQ (using the NASDAQ TotalView-ITCH data, which does 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.
8
The primary data source for this paper is the NYSE Trade and Quote (TAQ) data. 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 (9:30am to 4:00 pm). We
follow Bessembinder (2003) when merging trades and quotes and do not lag quote time
stamps. CRSP data is used to compute daily trading statistics and to determine listing
exchange. We also use the NASDAQ TotalView ITCH data to examine orders, executions, and
cancellations of trades during identified quote stuffing events. The NASDAQ TotalView ITCH
data, which is used for our final analysis, provides information on transactions that execute on
NASDAQ only.
3.2 Use of TAQ Data to Identify Quote Stuffing
It is not typically possible to identify orders that are generated by computer algorithms
in US equity markets in most data sources. 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, cancelations, and executions handled
by an individual exchange. For example, Hendershott, Jones, and Menkveld (2011) use the
number of electronic messages handled by NYSE’s SuperDOT system and captured in the NYSE’s
9
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
submission, cancelations, and trade executions for orders received by NASDAQ. Using this 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, page
19).
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
does not include information on individual order submissions and cancelations, but contains
consolidated quotes from all exchanges in the national market system. Despite not containing
information on individual orders, submissions and cancelations 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.8
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,
8
We check a number of instances where Hasbrouck and Saar (2010) identify an elevated number of “strategic
runs”, all instances are marked with a substantial increase in quoting activity reported in TAQ.
10
examining HFT behavior across market centers should provide a more complete picture of the
impact of quote stuffing on overall market conditions.
3.3 Identification of Quote Stuffing
Gai, Yao, and Ye (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, however, rather than attempting to identify all quote stuffing
events, we locate extreme episodic spikes in quoting activity. To locate these events we first
divide the trading day (9:30-4:00) 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 twenty-day windows. We
identify intense quoting episodes as segments where the level of quoting activity exceeds the
previous twenty-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 twenty-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 one
to ten minutes.9 As our goal is to identify information-free intense episodes of quoting activity,
9
Events with duration longer than 10 minutes are excluded from the sample.
11
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] window
surrounding an earnings or dividend announcement as identified in Compustat and CRSP.
Finally, we eliminate events if there is an influx in trading in the ten minutes prior to 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 prior to 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 will examine in more
detail in this study. 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.
Panel A of Table 1 reports summary statistics for sample firms that undergo at least one
quote stuffing event during the year. Mean daily volume of shares traded ranges from 720 to
211 million, with a mean of 877,000 thousand shares. Sample firm size also spans a large range
from $520,000 to $273 billion. Median closing price and daily returns are $16.16 and 0.05%,
respectively. Daily statistics are computed as the average over the entire trading year.
The magnitude 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
12
30 standard deviations and an additional 42% of events occur between 30 and 40 standard
deviations (see Table 1 Panel B). Panel C of table 1 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 summary statistics not tabulated in table 1 are noteworthy. 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 impact a large number of firms; 5292 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 7010 quote updates per minute.
4. Impact of Quote Stuffing on Market Quality
4.1 Measures of Market Quality
We employ an event study methodology to gauge the impact of quote stuffing on
market quality. We use TAQ data to compute several measures of market quality for each
minute in the ten-minute window immediately prior to 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. We calculate HighLow
as 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 Hasbrouck 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,
, then averaged over the one-minute interval. Effsprd is a measure of
13
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, while table 2 reports, mean market quality statistics
for the quote stuffing interval (time 0), the ten minutes prior (time -10 thru -1), and the ten
minutes immediately following (+1 thru +10) the events. All three measures of liquidity (Qsprd,
Pqsprd, and Effsprd) remain relativity constant in the minutes prior to 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 4. In
the pre-event window Effsprd follows a similar pattern to Qsprd and Pqsprd, remaining
relatively constant before increasing sharply to a level of $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 thru +10.
Volatility measures also follow patterns similar to that 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.
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 expected, quoting activity peaks at time 0 at a level of more than four
14
times the pre and post 30 one-minute averages. Trading activity peaks at time +1 and remains
elevated through minute +10.
4.2 Regression Results
To further explore the impact of quote stuffing on market quality, we run a series of
panel regressions, which control for factors that may impact market quality. Each regression
uses data from the event period as well as the ten one-minute periods immediately preceding
(pre-periods) and following (post-periods) the event. We estimate the following equation to
test for a relation between quote stuffing and effective spread:
.
Where
(1)
is the average effective half-spread for stock i in minute t;
a dummy variable equal to 1 for event segments and 0 otherwise;
equal to 1 for the period following the event;
quote midpoint for stock i in minute t;
is
, a dummy variable, is
is the standard deviation of the
, a measure of activity, is 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 subsequent regressions.
We estimate similar models to examine the impact of quote stuffing on quoted and
percentage quoted spreads:
(2)
15
(3)
Where
and
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
:
(4)
Table 3 presents the estimated coefficients for our market quality regressions. The
coefficient of primary interest is
, which measures the impact of identified quote stuffing
events on market quality. The coefficient of During,
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 shortterm volatility.
The coefficient of the
dummy variable is positive in the Effsprd and Volitil
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 post-event
window, both effective spreads and short-term volatility remain slightly elevated compared to
their pre-event levels.
Given that the identified quote stuffing episodes are of varying durations, it is feasible
that the impact of quote stuffing on market quality depends on the duration of the quote
16
stuffing event. To determine if event duration matters, we estimate panel regressions
separately for events of varying durations. Table 4 reports regression results for four
subsamples, consisting of four subdivided event period duration lengths (the dependent
variables are Effsprd (panel A), Qsprd (panel B), and Volatil (panel C)). (0,1] refers to event
periods that last one minute or less, (1,4] includes event periods with a duration longer than
one minute and up to four minutes, (4,7] includes events that have a duration longer that four
minutes and up to seven minutes, and (7,10] are events that have a duration of longer than
seven minutes and up to 10 minutes. The coefficient of
lengths for Effsprd, Qsprd, and
is positive for all duration
regressions, although it is not significant in the Qsprd and
regressions for the Qsprd (4,7] length and
(7,10] length regressions. These
predominantly positive coefficients for During indicate that effective spreads, quoted spreads
and volatility are increasing during quote stuffing events. The coefficient of
is 0.004 in
the Effsprd regression for events with one minute duration and 0.002 for the (4-7] events. The
coefficient of
in the
regression displays a declining pattern. The lower
coefficients for During in the higher duration estimations indicate that the longer the quote
stuffing event lasts, the less impact there is on volatility.
follows a slightly altered pattern in the
regression as event period
duration increases. The coefficient increases from a level of 0.00859 for (0,1] to 0.00973 for
(1,4] before declining to 0.00439 for events lasting between seven and ten minutes. The
coefficients of
imply that periods of quote stuffing experience average spreads of 0.4¢
to 1¢ higher than in pre-event periods.
17
Quote stuffing may impact the market quality of large market capitalization stocks
differently than small capitalization stocks. We report market quality regression estimates for
four quartiles of firm size in table 5. Size quartiles are based on firms’ average market
capitalizations computed over the 2010 trading year. Q1 represents the smallest market
capitalization quartile. Consistent with our previous analysis, the coefficient of During is
positive for all size quartiles and measures of market quality. Additionally, the coefficient of
During declines from the small market capitalization stocks to the large capitalization stocks,
indicating that smaller firms tend to have larger trading costs and volatility impacts due to the
quote stuffing events.
To test if quote stuffing impacts the market quality of NYSE/ARCA- and NASDAQ-listed
stocks differently, we run our analysis separately for stocks listed on the NYSE/ARCA and
NASDAQ exchanges and report our results in table 6. Consistent with our previous analysis, the
coefficient of During is positive for both NYSE/ARCA- and NASDAQ-listed stocks for all measures
of market quality. Intense quoting activities appear to influence the spread measures more for
NASDAQ stocks than NYSE/ARCA ones. While multiple reasons are possible, we posit that the
higher spread measures may be due to the lower market capitalization of the NASDAQ stocks.
Overall, our analyses imply that quote stuffing can adversely impact 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.
4.3 Causes of Quote Stuffing Events
18
Quote stuffing occurs for many potential reasons. In this section we attempt to pinpoint
several reasons for quote stuffing by classifying quote stuffing events identified in section 3 into
four different strategies. The first strategy, Type 1: Same-Stock Cross-Venue, involves quote
stuffing with the purpose of slowing 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 non-stuffing traders are required to process the barage 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.
Type 2: Multi-Stock Same-Venue are events 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.
Type 3: Liquidity Consuming events are occasions when 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
19
on one venue will not be immediately followed by cancellations of outstanding limit orders on
other venues. We classify events in which 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 as Type 3 events.
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. We classify an event as Type 4: ETF
strategy if 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 7. We are able to definitively
categorize 12632 of the 24623 identified quote stuffing episodes. Of the quote stuffing events
we categorize, 8295 are Type 1, 3688 are Type 2, 554 are Type 3, and 95 are Type 4. The most
common quote stuffing strategy involves slowing down other traders in the same stock across
exchanges.
Next, we re-run our market quality regressions separately for each identified quote
stuffing strategy. These regression results are reported in table 8. The coefficient of During in
the Effsprd, Qsprd, and Volitil regressions is positive for all event types, indicating that effective
spread, quoted spread, and volatility are increasing during all quote stuffing events types.
4.4 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
20
determine if quotes that occur during quote stuffing events are a 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. Table 9 (panel A) 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 in a
row 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, quotes update runs tend to be either
short, in runs of 10 or less, or long, in runs of 300 or more.
Panel B of table 9 shows Percentage of Bid (Ask) update runs prior to, during, and after
the quote stuffing events. Pre is the 10-minute period prior to a quote stuffing event, and post
is the 10 minutes after a quote stuffing event. Short runs of 1-10 quote updates occur more
frequently prior to the event for both bid and ask side updates. Long runs of 301+ updates are
more common during the event than prior to the event. These long run updates are also more
common in the post-period than in the pre-period.
21
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 if the events are isolated on an exchange (our first look at
whether or not a trade might be “walking the book”). Table 10 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 10 reports 345
quote stuffing events occur largely on AMEX, while 4748 occur on the NYSE, 7144 on ARCA, and
6928 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 the most quote stuffing events occur.
85.8% of the quotes of an average quote stuffing event on the NYSE come from the NYSE, but
6.2% and 3.7% of the quotes are reported by ARCA and NASDAQ as well. 64.9% of the quotes
of the average NASDAQ quote stuffing event are reported by NASDAQ, but 16.3% of the quotes
are reported by ARCA and 7.6% by BATS. When there is a quote stuffing event on BATS, 62.9%
of the quotes are reported by BATS, while 17.2% are reported by ARCA, and 10.3% by
22
NASDAQ.10 While all venues in Table 10 report quotes during the events, ARCA, NASDAQ, and
BATS show the most quoting activity in events occurring on other exchanges.
Panel B of table 10 displays the exchange with the most quotes during a quote stuffing
event for each event type. We know, from Table 7, that Type 1 events are the most prevalent
of the identified quote stuffing events. Panel B shows that, with the exception of the PSX, Type
1 events are the most prevalent events for all venues. Type 1 events involve quote stuffing on
one exchange while trading occurs on other venues. Of the identified events, Type 1, 3, and 4
events occur most frequently on NASDAQ. A Type 2 event is one in which multiple stocks
experience quote stuffing on the same exchange simultaneously. Type 2 events occur most
frequently on ARCA (36.0%), followed by NASDAQ (23.9%) and NYSE (25.0%).
4.5 Trading Behavior during a Quote Stuffing Event
We look at trading on the various exchanges during quote stuffing events.11 We know,
from Table 7, that Type 1 quote stuffing events—events where more than 50% of quote
updates occur on one venue and more trading occurs on other venues—are the most prevalent
of the categorized events. Panel A of table 11 reinforces this finding, showing that trading
occurs on various exchanges during the quote stuffing events. 30.5% of trades that occur
during an NYSE quote stuffing event are reported by the NYSE, 26.2% of the trades execute on
10
We recognize that not all rows sum to 100%. There are two reasons. First, we report averages of averages (we
average first by firm, then by exchange). So, 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 one percent 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 if trading is occurring on or away from the exchange where the quote
stuffing event occurs.
23
the NASD, 13.9% on NASDAQ, and 13.1% on ARCA. 29.1% of trades occurring during a NASDAQ
quote stuffing event are reported by the NASDAQ and 29.2%, 8.8%, and 17.7% execute on the
NASD, NYSE, and ARCA, respectively. While 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 high frequency trader submitting a
large volume of orders, thereby requiring other high frequency traders to process these orders,
yielding the submitting trader an advantage (Brogaard, 2010). If Brogaard’s 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 table11. However, panel A also reports
trading on exchanges other than the one 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 prior to the event to
determine if an abnormal amount of trading is taking place on any one venue (Panel B, table
11).
We concentrate our discussion to the exchanges with the most quote stuffing events:
NYSE, ARCA, NASDAQ, and BATS. There is 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% of the trades execute on the NASD, 13.9% on NASDAQ,
and 13.1% on ARCA. 29.1% of trades occurring during a NASDAQ quote stuffing event are
24
reported by the NASDAQ and 29.2%, 8.8%, and 17.7% execute on the NASD, NYSE, and ARCA,
respectively.
We note that the percentage change in volume on the NYSE during an NYSE quote
stuffing event versus the 10 minutes prior to 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).
4.6 Orders during a Quote Stuffing Event
We use the NASDAQ TotalView ITCH data to get more detailed information during these
quote stuffing events. The NASDAQ TotalView ITCH data contains order-level data for all stocks
that execute on the NASDAQ exchange. Table 12 Panel A 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, the number of new orders, cancelled orders,
and order cancelation rate. New orders increase from 2.03 per second to 23.92 during the
event, while cancelled orders increase from 1.4 per second to 13.93 per second. Panel 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 cancelled orders.
One concern regarding the number of same-side quote updates (table 9) is that large
orders could 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 12, alleviates this
concern as the order size significantly decreases during extreme quoting events. Average order
25
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.
5. 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 (non-quote-stuffing stocks). We
obtain our matched sample by identifying quote stuffing-stock and non-quote-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, 1997b), and
Chung, Van Ness, and Van Ness (2001). Specifically, we form quote stuffing stock and nonquote-stuffing stock pairs by calculating the following score:
 X kiS  X kjC 
k  X S  X C / 2
 ki kj 
2
,
S
Where
X ki
is firm characteristic k for firm i that experiences a quote stuffing event and
C
X 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
non-quote-stuffing stock with the smallest score to match each quote stuffing stock.
Table 13 reports summary statistic for quote stuffing-stocks and control non-quotestuffing stocks. Mean market capitalization, daily trading volume, returns, and price are similar
26
for quote stuffing and non-quote-stuffing-stocks. T-tests for differences in means between
quote stuffing and non-quote-stuffing stocks do not yield statistically significant results,
suggesting that our matching procedure is able to identify firms with similar characteristics.
We report market quality statistics for quote stuffing-stocks and control stocks in table
14 and test the differences in market quality statistics for quote stuffing stocks and non-quotestuffing stock pairs for each period surrounding the quote stuffing event (similar to table 2). As
expected, the difference in mean number of quotes per minute (NQS) increases substantially
for quote stuffing-stocks relative to non-quote-stuffing stocks during the event period, rising
from a mean difference of 59 quotes in minute -10 to a difference of 6587 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 non-quotestuffing stocks also increases during the event period. The difference between Qsprd for quote
stuffing and non-quote-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 preevent level. The difference in Pqsprd follows a similar pattern to Qsprd. The difference in quote
midpoint volatility between quote stuffing and non-quote-stuffing stocks, as measured by
Highlow, also increases during event periods. The difference between Highlow for quote
stuffing and non-quote-stuffing stocks is over seven times higher during the event window
compared to the difference in Highlow during the pre- and post-event periods.12
12
We do not compare Effsprd and Voltil for matched pairs due to the high number of lost observations for nonquote-stuffing stocks, resulting from non-quote-stuffing stocks not having a trade every event minute.
27
To further shed light on differences in market quality between quote stuffing and nonquote-stuffing stocks, we use a difference-in-difference regression approach, similar to Chung,
Van Ness, and Van Ness (2001). The dependent variable is the difference in market quality
statistics for quote stuffing-stock i in minute t and market quality for the matched non-quotestuffing stock j in minute t. We difference the independent variables,
similar fashion. As in previous regressions we include dummy variables
and
, in a
and
which equal 1 for event minutes and minutes after the event, respectively.
Table 15 reports results from difference-in-difference regressions. Consistent with our
univariate tests, the indicator variable During is positive for all regressions, suggesting that
market quality for quote stuffing-stocks diminishes relative to control stocks in quote stuffing
event periods. The coefficient of the dummy variable Post is not significantly different from
zero in any of the regression specifications, suggesting that differences in market quality
statistics for quote stuffing and non-quote-stuffing stocks return to a pre-event level in the
period subsequent to the quote stuffing episode.
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 a result of events and not
other exogenous factors.
6. Conclusion
In this study, we analyze the impact of intense episodic spikes in quoting activity on
market quality, including liquidity and volatility. We find that quote stuffing is pervasive,
28
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, cancelled at higher rates, are
for shorter durations, are executed at lower rates, and are smaller in size during quote stuffing
events. Our study suggests that the HFT strategy of quote stuffing may exhibit some of the
adverse features that are criticized in the media.
29
References
Aggarwal, Rajesh K., and Guoun Wu, 2006, Stock Market Manipulations, Journal of Business 79,
1915-1953.
Allen, Franklin, and Gary Gorton, 1992, Stock Price Manipulation, Market Microstructure and
Asymmetric Information. European Economic Review 36:624-30.
Bessembinder, Hendrik, 1999, Trade Execution Costs on NASDAQ and the NYSE: A Post-Reform
Comparison, Journal of Financial and Quantitative Analysis 34, 387-407.
Bessembinder, Hendrik, 2003, Quote-based Competition and Trade Execution Cost in NYSElisted Stocks, Journal of Financial Economics 70, 385-422.
Bessembinder, Hendrik and Herbert Kaufman, 1997a, A Cross-Exchange Comparison of
Execution Costs and Information Flow for NYSE-listed Stocks, Journal of Financial
Economics 46, 293-319.
Bessembinder, Hendrik and Herbert Kaufman, 1997b, A Comparison of Trade Execution Costs
for NYSE and NASDAQ-Listed Stocks, Journal of Financial and Quantitative Analysis 32,
287-310.
Biais, Bruno, and Paul Woolley, 2011, High Frequency trading. Working paper, Toulouse
University.
Brogaard, Jonathan A., 2010, High frequency trading and its impact on market quality, Working
paper, Northwestern University.
Cartea, Àlvaro, and Josè Penalva, 2011, Where is the Value in High Frequency Trading?,
Working paper, Universidad Carlos III de Madrid.
Cvitanic, Jaksa, and Andrei Kirilenko, 2010, High Frequency Traders and Asset Prices, Working
paper, Cal Tech.
Chung, Kee, Bonnie Van Ness and Robert Van Ness, 1999, Limit Orders and the Bid-Ask Spread,
Journal of Financial Economics 53, 255-287.
Davies, Ryan, and Sang Kim, 2009, Using Matched Samples to Test for Differences in Trade
Execution Costs, Journal of Financial Markets 12, 173-202.
Elder, James, 2010, Quote Stuffing is One More Reason Not to Invest in Stocks, The Wall Street
Journal, September 8, A18.
30
Financial Industry Regulatory Authority, September 13,2010. FINRA Sanctions Trillium
Brokerage Services, LLC, Director of Trading, Chief Compliance Officer, and Nine Traders
$2.26 Million for Illicit Equities Trading Strategy. Rel No P121951.
Gai, Jiading, Chen Yao, and Mao Ye, 2012, The Externality of High Frequency Trading, working
paper, University of Illinois.
Goldstein, Itay, and Alexander Guembel, 2008, Manipulation and the Allocational Role of Prices,
Review of Economic Studies 75, 133-164.
Gomber, Peter, Björn Arndt, Marco Lutat, and Tim Uhle, 2011, High Frequency Trading,
Working paper, Goethe University.
Hasbrouck, Joel, and Gideon Sarr, 2013, Low-Latency Trading, Working paper, Journal of
Financial Markets 16, 646-679.
Hendershott, Terrence, Charles M. Jones, and Albert J. Menkveld, 2011, Does Algorithmic
Trading Improve Liquidity?, The Journal of Finance 66, 1-33.
Hendershott, Terrence, and Pamela V. Moulton, 2009, Speed and stock market quality: The
NYSE Hybrid, Working paper, University of California at Berkeley.
Hendershott, Terrence, and Ryan Riordan, 2009, Algorithmic Trading and Information, Working
paper, NET Institute.
Hirschey, Nicholas, 2013, Do High-Frequency Traders Anticipate Buying and Selling Pressure?
Working paper, London School of Business.
Huang, Roger, Hans 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, Robert A., 1992, Market Manipulation, Bubbles, Corners, and Short Squeezes. Journal of
Financial and Quantitative Analysis 27, 311-36.
Jones, Charles, 2013, What do we know about high-frequency trading? Working paper,
Columbia University.
Kirilenko Andrei, Albert Kyle, Mehrdad Samadi, and Tugkan Tuzun, 2010, The Flash Crask: The
Impact of High Frequency Trading on an Electronic Market, Working paper, CFTC.
Kumar, Praveen, and Duane J. Seppi, 1992, Futures Manipulation with “Cash Settlement.”
Journal of Finance 47, 1485-1502.
31
Lauricella, Tom, and Jenny Strasburg, 2010, SEC Probes Canceled Trades, The Wall Street
Journal, September 2, A1.
Madhavan, Ananth, 2012, Exchange-Traded Funds, Market Structure and the Flash Crash,
Working paper, Blackrock.
Mei, J., G. Wu and C. Zhou, 2004, Behavior Based Manipulation, NYU Working paper
Menkveld, Albert J., 2011, High Frequency Trading and the New-Market Makers, Working
Paper, VU University of Amsterdam.
O’Hara, Maureen, and Mao Ye, 2011, Is Market Fragmentation Harming Market Quality?
Journal of Financial Economics 100, 459-474.
Remark before Economic Club of New York, SEC Chairwoman Mary L. Schapiro, September 7,
2010.
Securities and Exchange Commission, January 13,2010. Concept Release on Equity Market
Structure. Rel No 34-61358.
Zhang, X. Frank, 2010, High-Frequency Trading, Stock Volatility, and Price Discovery, Working
paper, Yale University.
32
Table 1: 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 twenty-day mean number of quotes-perminute by at least 20 standard deviations. The number of events that are between 20-30, 3050, 50-100, 100-250, and >250 standard deviations of their previous twenty day mean number
of quotes-per-minute are reported. Panel C lists the number of events by duration and their
cumulative distribution.
Panel A: Firm Characteristics
Mean
Median
Std
Min
Max
MktCap($Million)
2378
3626
10349
0.52
273560
Daily Volume(1000s)
877
173
4747
.72
211465
Return(%)
0.047
0.051
0.14
-2.166
5.683
Closing Price ($)
25.47
16.16
39.82
0.09
1891.79
Panel B: Quote Stuffing Events
#Of Standard Deviations above Mean
Number 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
Number of Events
Length in Minutes
Cumulative Percent
17893
<1
72.3%
3074
1-2
84.7%
1171
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%
33
Table 2: Market Quality Stats by period
This table reports mean market quality statistics for the quote stuffing interval (time 0) and the
ten minutes prior (time -10 thru -1) and the ten minutes immediately following (+1 thru +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
and then averaged over the one-minute interval,
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.
Qsprd
Pqsprd
Effsprd
Voltil
HighLow
-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
34
Table 3: Regression Results
This table reports the results of regression analyses analyzing the impact of quote stuffing on
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
and then averaged over the one-minute interval, 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 oneminute interval. The following model is then estimated:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, Midpvolit, is the standard deviation of
the midpoint,
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 parenthesis and are based on event cluster corrected robust standard errors.
Effsprd
Qsprd
Pqsprd
Voltil
Post
0.00100***
-0.00133***
-0.00012***
0.00231***
(5.048)
(-2.780)
(-2.992)
(3.462)
During
0.00326***
0.00908***
0.00062***
0.00535***
(10.108)
(13.500)
(11.367)
(4.966)
Midpvolit
0.03095
0.25797**
0.01178**
(1.463)
(2.277)
(2.305)
0.00002***
-0.00002***
-0.00000***
(3.105)
(-2.938)
(-2.959)
Constant
0.01556***
0.08104***
0.00606***
0.01261***
(64.459)
(85.904)
(131.768)
(37.243)
Observations
260,798
475,674
475,674
R-squared
0.69
0.89
0.89
F test
30.28
92.51
70.75
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
35
222,128
0.28
16.75
Table 4: Regressions by Length of Quote Stuffing Event
This table reports the results of regression analyses, separately for events of different
durations. (0,1] refers to event periods that last one minute or less, (1,4] includes event
periods with a duration longer than one minute and up to four minutes, (4,7] and (7,10] are
similarly defined. 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
and then averaged over the one-minute interval, 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 oneminute interval. The following model is then estimated:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, Midpvolit, is the standard deviation of
the midpoint,
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 parenthesis and are based on event cluster corrected robust standard errors.
Panel A: Market Quality Effsprd
Duration
(0,1]
(1-4]
(4-7]
(7-10]
Post
0.00112***
0.00072
-0.00004
-0.00014
(5.146)
(1.399)
(-0.044)
(-0.078)
During
0.00360***
0.00295***
0.00175**
0.00318*
(8.809)
(4.299)
(2.238)
(1.886)
Midpvolit
0.02145
0.11927
0.09884
0.40232
(1.298)
(1.187)
(0.722)
(1.191)
Nts
0.00002**
0.00001**
0.00001
0.00002
(2.476)
(1.988)
(0.786)
(0.420)
Constant
0.01467***
0.01703***
0.01820***
0.01705***
(66.474)
(16.910)
(14.382)
(5.388)
Observations
R-squared
F test
192,597
0.68
23.37
49,749
0.72
14.73
Table 4 Continued:
36
11,313
0.65
2.288
7,139
0.68
1.772
Panel B: Market Quality Qsprd
Duration
Post
During
Midpvolit
Nts
Constant
(0,1]
-0.00025
(-0.465)
0.00859***
(11.819)
0.16732**
(2.047)
-0.00002***
(-2.613)
0.07549***
(104.723)
Observations
330,235
R-squared
0.89
F test
60.41
Panel C: Market Quality Volatil
(1-4]
-0.00273**
(-2.232)
0.00973***
(7.136)
0.95720***
(6.522)
-0.00006***
(-5.309)
0.08601***
(64.452)
(4-7]
-0.00830***
(-3.669)
0.00316
(1.556)
1.05939***
(5.673)
-0.00003***
(-3.134)
0.09713***
(47.855)
(7-10]
-0.00914***
(-2.969)
0.00439*
(1.690)
1.27127***
(5.013)
-0.00011***
(-2.975)
0.10032***
(35.235)
99,762
0.88
44.69
28,399
0.89
13.77
17,278
0.90
14.53
(4-7]
0.00041
(0.548)
0.00198*
(1.834)
0.01370***
(28.699)
(7-10]
-0.00868
(-1.143)
0.01059
(1.014)
0.02489***
(8.476)
Duration
Post
During
Constant
(0,1]
0.00204***
(2.949)
0.00534***
(6.160)
0.01208***
(33.754)
(1-4]
0.00510**
(2.412)
0.00426***
(3.846)
0.01289***
(12.121)
Observations
166,909
40,584
8,902
R-squared
0.22
0.36
0.55
F test
19.26
7.398
1.737
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
37
5,733
0.19
0.772
Table 5: Regressions by firm size
This table reports the results of regression analyses performed separately for firm-size
quartiles. Q1 (Q4) is comprised of the smallest (largest) firms as measured by market
capitalization during 2010. 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
and then averaged over the one-minute interval,
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:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, Midpvolit, is the standard deviation of
the midpoint,
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 parenthesis and are based on event cluster corrected robust standard errors.
Panel A: Market Quality Effsprd
Size
Q1
Q2
Q3
Q4
Post
0.00455***
0.00127**
0.00051
0.00053**
(3.355)
(2.393)
(1.525)
(2.342)
During
0.00949***
0.00392***
0.00280***
0.00194***
(4.569)
(7.667)
(6.468)
(3.487)
Midpvolit
0.00979
0.05229
0.39541***
-0.03801
(0.705)
(1.111)
(3.929)
(-0.625)
Nts
0.00008*
0.00005***
-0.00000
0.00002***
(1.726)
(2.777)
(-0.275)
(2.700)
Constant
0.04128***
0.01662***
0.00950***
0.01206***
(46.654)
(36.192)
(12.476)
(20.661)
R-squared
F test
0.74
7.624
0.65
17.99
Table 5 Continued:
38
0.53
18.82
0.71
6.868
Panel B: Market Quality Qsprd
Size
Post
During
Midpvolit
Nts
Constant
Q1
-0.00375**
(-2.471)
0.00714***
(4.179)
0.09553*
(1.709)
-0.00013***
(-2.754)
0.16109***
(163.521)
R-squared
0.87
F test
18.88
Panel C: Market Quality Volatil
Q2
0.00059
(0.581)
0.01737***
(12.707)
0.54967
(1.588)
-0.00009
(-1.450)
0.08043***
(35.009)
Q3
-0.00142***
(-3.403)
0.00723***
(10.788)
1.23208***
(10.295)
-0.00012***
(-4.634)
0.03942***
(47.109)
Q4
-0.00177***
(-2.810)
0.00581***
(5.070)
0.68662***
(4.061)
-0.00003***
(-3.640)
0.03571***
(23.789)
0.88
78.53
0.89
58.46
0.90
10.35
Q3
0.00225**
(2.322)
0.00303***
(5.689)
0.01031***
(21.019)
Q4
0.00109***
(2.929)
0.00425***
(3.844)
0.01324***
(60.779)
Size
Post
During
Constant
Q1
0.01323
(1.383)
0.01388***
(3.093)
0.01668***
(3.299)
Q2
0.00258
(1.141)
0.00952*
(1.736)
0.01218***
(10.373)
R-squared
0.13
0.33
0.31
F test
6.178
2.388
17.16
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
39
0.48
8.637
Table 6: Regressions by listing exchange
This table reports the results of regression analyses for stocks listed on NYSE/ARCA and
NASDAQ stock exchanges. 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
and then averaged over the one-minute interval,
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:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, Midpvolit, is the standard deviation of
the midpoint,
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 parenthesis and are based on event cluster corrected robust standard errors.
Panel A:NYSE/ARCA
Effsprd
Qsprd
Pqsprd
Voltil
Post
0.00102***
-0.00203***
-0.00009***
0.00295***
(4.129)
(-4.394)
(-4.733)
(3.110)
During
0.00280***
0.00527***
0.00021***
0.00424***
(7.765)
(8.282)
(7.949)
(2.920)
Midpvolit
0.07631
0.43252**
0.01028**
(1.350)
(2.202)
(2.219)
Nts
0.00001
-0.00003**
-0.00000**
(1.518)
(-2.361)
(-2.491)
Constant
0.01161***
0.04958***
0.00212***
0.01180***
(24.350)
(38.209)
(66.152)
(24.974)
Observations
R-squared
F test
172,502
0.57
19.24
283,693
0.88
38.74
Table 6 Continued:
40
283,693
0.81
46.50
150,192
0.26
8.569
Panel B:NASDAQ
Post
During
Midpvolit
Nts
Constant
Effsprd
0.00095***
(2.944)
0.00416***
(6.024)
0.04751
(1.328)
0.00002**
(2.397)
0.02174***
(43.586)
Qsprd
0.00192*
(1.866)
0.01778***
(12.154)
0.44917***
(3.243)
-0.00003***
(-3.043)
0.12434***
(81.276)
Pqsprd
0.00002
(0.155)
0.00149***
(10.581)
0.01983**
(2.262)
-0.00000***
(-2.697)
0.01203***
(114.605)
Observations
78,122
166,799
166,799
R-squared
0.76
0.89
0.89
F test
11.94
71.05
51.18
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
41
Voltil
0.00066***
(2.591)
0.00502***
(8.446)
0.01391***
(94.841)
64,315
0.64
35.67
Table 7: Quote Stuffing Strategies
A quote stuffing event is categorized as a Type 1: Same-Stock Cross-Venue event when more than 50% of
quote updates occur on one venue and that venue does not have 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 quote updates. Type 3: Liquidity Consuming
Strategy events are where 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
events labeled Type 4: ETF Strategy 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: Multi-Stock Same Venue
3,688
Type 3: Liquidity Consuming Strategy
554
Type 4: ETF Strategy
95
42
Table 8: Regressions by Type of Quote Stuffing Strategy
This table reports the results of regression analyses, separately for events quote stuffing (QS)
strategies. Same-Stock Cross-Venue (Type 1) are QS events in which greater than 50% of
quote updates occur on one venue which does not have highest number of trades. MultiStock Same Venue (Type 2) are events in which two or more QS events occur simultaneously
on one exchange with greater than 50% of quote updates. Liquidity Consuming Strategy
(Type 3) are event in which the distribution of quote updates and trades are relatively evenly
dispersed across multiple exchanges and no single exchange has greater than 33% of quote
updates or trades. ETF Strategy (Type 4) are event where in which an ETF and one or more of
the ETF’s constituents have simultaneous QS events. We use TAQ data to compute several
measures of market quality: Voltil is the one-minute standard deviation of trade prices,
Qsprd is the average spread (ask price minus bid price) of the one minute interval, Pqsprd is
the spread scaled by the midpoint
and then averaged over the one-minute interval,
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:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, Midpvolit, is the standard deviation of
the midpoint,
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 parenthesis and are based on event cluster corrected robust standard errors.
Panel A: Market Quality Effsprd
Strategy
Type 1
Type2
Type3
Type 4
Post
0.00043**
0.00071**
0.00103
-0.00046
(2.403)
(2.065)
(1.378)
(-0.206)
During
0.00156***
0.00225***
0.00540**
0.00864**
(4.071)
(3.569)
(2.478)
(2.123)
Midpvolit
0.07979
0.18655***
-0.00905
0.08935
(1.295)
(3.136)
(-0.645)
(0.398)
Nts
0.00002***
0.00001
0.00000
0.00002
(2.812)
(1.059)
(1.337)
(0.967)
Constant
0.01187***
0.01160***
0.01363***
0.02170***
(22.756)
(23.706)
(29.491)
(7.561)
Observations
R-squared
F test
127,344
0.73
13.51
45,837
0.75
11.79
43
4,656
0.63
2.484
1,307
0.59
2.419
Panel B: Market Quality Qsprd
Strategy
Post
During
Midpvolit
Nts
Constant
Type 1
0.00048
(0.937)
0.00321***
(3.890)
0.35915**
(2.274)
-0.00006***
(-3.276)
0.04376***
(37.519)
Observations
162,085
R-squared
0.87
F test
8.572
Panel C: Market Quality Volatil
Type2
-0.00069
(-0.867)
0.00631***
(4.611)
0.82518***
(6.276)
-0.00014***
(-5.197)
0.06059***
(58.030)
Type3
-0.00321
(-1.384)
0.00574**
(2.265)
0.01863
(0.312)
-0.00000
(-1.224)
0.06833***
(51.221)
Type 4
-0.00390
(-1.041)
0.03711**
(2.075)
1.12771***
(3.371)
-0.00007***
(-2.694)
0.03900***
(8.270)
72,623
0.90
15.50
10,546
0.89
4.074
1,964
0.69
3.746
Type3
0.00408
(1.627)
0.00598***
(3.716)
0.00833***
(6.573)
Type 4
0.00473
(1.458)
0.01155***
(3.311)
0.01780***
(11.163)
Strategy
Post
During
Constant
Type 1
0.00089***
(2.707)
0.00237***
(5.541)
0.01127***
(63.347)
Type2
0.00110***
(4.806)
0.00241***
(6.366)
0.01128***
(94.250)
Observations
111,751
39,047
3,632
R-squared
0.48
0.63
0.30
F test
15.46
26.70
8.473
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
44
1,165
0.44
8.658
Table 9: Runs of Quote Updates During Quote Stuffing Events
The table reports the number and percentage of quote updates that occur on the bid and ask side
of the quote during quote stuffing events. Panel A reports the number of bid and ask side update
runs. A bid (ask) run is a series of sequential quotes from the same exchange that that are 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. # of Bid (Ask) Side Updates in a row 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 B 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: Runs of Quote updates
# of Bid Side
Percentage of Bid # of Ask Side
Percentage of
Runs
Updates in a row updates
Updates in a row Ask 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 B: Percentage Runs of Quote updates
Runs
1 – 10
11– 50
51 – 100
101 – 150
151 – 200
201 – 250
251 – 300
301+
Percentage of Bid updates
Pre
During
Post
48.4%
32.2%
35.2%
15.4%
9.3%
11.1%
4.9%
3.9%
3.5%
2.8%
2.5%
2.1%
2.0%
2.0%
1.6%
1.6%
1.6%
1.3%
1.3%
1.4%
1.1%
23.5%
47.2%
44.0%
45
Percentage of Ask updates
Pre
During
Post
48.2%
33.6%
36.4%
14.9%
8.6%
11.0%
4.9%
3.9%
3.6%
2.8%
2.5%
2.1%
2.1%
1.9%
1.6%
1.6%
1.5%
1.3%
1.4%
1.4%
1.1%
24.2%
46.2%
42.8%
Table 10: 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 the number 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 types
(See table 7).
Panel A: Percentage of Quotes during a Quote Stuffing Event by Exchange
Exchange
N
AMEX
Boston
National
ISE
NYSE
ARCA
NASDAQ
CBOE
PXS
BATS
AMEX
345
54.7%
0.0%
0.2%
1.3%
0.0%
21.5%
11.0%
4.3%
0.0%
4.4%
Boston
1180
0.1%
70.3%
0.3%
0.4%
1.2%
6.0%
17.3%
0.1%
0.0%
1.9%
National
273
0.2%
2.1%
68.0%
0.6%
1.0%
4.7%
8.7%
1.3%
0.0%
5.8%
ISE
314
0.8%
0.8%
0.3%
57.4%
2.4%
16.8%
8.6%
2.1%
0.0%
10.8%
NYSE
4748
0.0%
0.4%
0.1%
0.4%
85.8%
6.2%
3.7%
0.3%
0.0%
1.6%
ARCA
7144
0.4%
0.5%
0.3%
1.1%
2.3%
72.4%
12.4%
1.3%
0.0%
6.9%
NASDAQ
6928
0.7%
2.8%
0.6%
1.3%
2.1%
16.3%
64.9%
1.3%
0.0%
7.6%
CBOE
485
1.6%
0.0%
0.3%
1.4%
1.0%
9.2%
11.8%
61.3%
0.0%
11.4%
PSX
12
0.0%
4.6%
0.2%
0.0%
3.9%
3.9%
0.6%
0.0%
32.9%
1.1%
BATS
3194
0.2%
1.0%
0.4%
1.6%
1.5%
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
Type 3
Type 4
N
%
N
%
N
%
N
%
N
%
AMEX
345
1.4%
27
0.3%
2
0.1%
8
1.4%
0
0.0%
Boston
1180
4.8%
639
7.7%
135
3.7%
17
3.1%
9
9.5%
National
273
1.1%
85
1.0%
14
0.4%
26
4.7%
2
2.1%
ISE
314
1.3%
59
0.7%
3
0.1%
10
1.8%
1
1.1%
NYSE
4748
19.3%
1791
21.6%
924
25.0%
59
10.6%
6
6.3%
ARCA
7144
29.0%
1980
23.9%
1331
36.0%
132
23.8%
18
18.9%
NASDAQ
6928
28.1%
2303
27.8%
977
26.5%
160
28.9%
38
40.0%
CBOE
485
2.0%
135
1.6%
18
0.5%
10
1.8%
0
0.0%
PSX
12
0.0%
2
0.0%
5
0.1%
5
0.9%
0
0.0%
BATS
3194
13.0%
1274
15.4%
284
7.7%
127
22.9%
21
22.1%
46
Table 11: Location of Trading during a Quote Stuffing
This table reports the location of trading during a quote stuffing event. N = is the number of events when the exchange on far left 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 to the 10 minute time period
prior to the quote stuffing event.
Panel A: Percentage of Volume
Exchange
N
AMEX
Boston National
NASD
ISE
Chicago
NYSE
ARCA
NASDAQ
CBOE
PSX
BATS
AMEX
345
48.0%
0.1%
0.6%
25.5%
1.6%
0.0%
0.0%
11.2%
9.9%
0.5%
0.0%
2.6%
Boston
1180
1.2%
3.8%
1.2%
28.4%
0.8%
0.1%
7.1%
17.6%
27.9%
0.0%
0.0%
11.8%
National
273
0.9%
3.6%
14.3%
42.7%
0.6%
0.0%
7.5%
9.6%
13.9%
0.3%
0.0%
6.6%
ISE
314
3.5%
0.9%
0.7%
53.1%
4.6%
0.0%
6.5%
12.6%
13.9%
1.1%
0.0%
3.1%
NYSE
4748
0.0%
1.0%
4.5%
26.2%
2.0%
1.2%
30.5%
13.1%
13.9%
0.1%
0.1%
7.5%
ARCA
7144
0.9%
1.0%
1.0%
31.9%
1.6%
0.1%
15.8%
19.7%
20.3%
0.3%
0.0%
7.5%
NASDAQ
6928
3.3%
1.2%
1.1%
29.2%
1.2%
0.1%
8.8%
17.7%
29.1%
0.2%
0.0%
8.0%
CBOE
485
9.3%
0.8%
3.2%
24.4%
2.0%
0.4%
3.7%
14.1%
35.0%
2.8%
0.0%
4.3%
PSX
12
0.0%
0.9%
2.8%
40.3%
0.0%
0.0%
21.7%
19.0%
9.1%
0.0%
0.2%
6.1%
BATS
3194
1.8%
1.5%
0.7%
29.9%
1.4%
0.1%
16.3%
17.8%
20.4%
0.1%
0.0%
9.8%
Panel B: Change in Percentage of Volume
Exchange
N
AMEX
Boston National
NASD
ISE
Chicago
NYSE
ARCA
NASDAQ
CBOE
PSX
BATS
AMEX
345
108.0%
85.8%
-100%
-18.0%
-81.4%
-100.0%
0.0%
-16.3%
-15.6%
-98.1%
0.0%
-26.2%
Boston
1180
142.5%
74.9%
62.3%
5.9%
122.1%
-15.9%
16.8%
87.2%
44.7%
12.1%
3.2%
62.6%
National
273
-57.2%
42.0%
559.4%
23.1%
111.3% -100.0%
23.3%
3.3%
63.4%
-100%
-100%
-13.6%
ISE
314
-32.7%
49.2%
-87.3%
25.5%
133.7% -100.0%
48.1%
247.3%
2.5%
-100%
0.0%
10.1%
NYSE
4748
0.0%
-6.7%
222.4%
36.1%
124.6%
155.4%
14.7%
30.8%
31.5%
-33.7%
-9.9%
44.4%
ARCA
7144
44.9%
26.5%
38.0%
19.4%
28.9%
82.4%
40.7%
396.8%
31.7%
106.3%
81.0%
45.9%
NASDAQ
6928
92.5%
75.1%
9.5%
18.9%
26.7%
-29.4%
24.6%
45.4%
48.9%
-36.5%
-12.5%
56.1%
CBOE
485
42.3%
212.2%
106.7%
-14.2%
102.1%
-9.1%
-27.9%
-42.2%
-17.3%
80.2%
0.0%
184.2%
PSX
12
-100%
-65.4% 1429.6% -28.3%
0.0%
0.0%
57.2%
208.6%
-1.3%
-100%
4.0%
4.1%
BATS
3194
90.9%
21.3%
47.4%
15.9%
15.7%
-45.0%
19.9%
49.3%
32.7%
-50.8%
-47.3%
37.8%
47
Table 12. Order Statistics
This table reports information about order submission, cancelation and execution activity on the NASDAQ exchange for sample firms
surrounding quote stuffing events. Data for order activity is 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
episode, and Post is the 10 minute window following the quote stuffing event. Messages Per Sec is the mean number of messages
per seconds reported in ITCH for sample firms; New Orders Per Sec, Canceled Orders Per Sec, and Executed Orders Per Sec is the
mean number of new orders submitted canceled, and executed per second respectively; New Buy (Sell) Orders Per Sec is the average
number of new buy (sell) orders submitted per second; Order Cancelation (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; Cancelation (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 less than 100
shares; 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 statistic for all sample stocks, Panel B reports
statistics for NYSE/ARCA listed stocks, and Panel C reports statistics for NASDAQ listed stocks.
Panel A: All Stocks
Pre
During
Post
During-Pre
During-Post
Post-Pre
Messages Per Sec
3.39
37.59
3.54
34.20***
34.04***
0.16**
New Orders Per Sec
2.03
23.92
2.10
21.89***
21.82***
0.07
Canceled Orders Per Sec
1.40
13.93
1.48
12.53***
12.45***
0.08**
Executed Orders Per Sec
0.10
0.15
0.11
0.05***
0.03***
0.02***
New Buy Orders Per Sec
1.14
12.84
1.14
11.70***
11.70***
0.00
New Sell Orders Per Sec
1.10
13.24
1.16
12.14***
12.08***
0.06*
Order Cancelation Rate (%)
57.0
72.0
56.8
15.4%***
15.2***
0.2***
Order Execution Rate (%)
6.8
4.7
7.0
-2.1***
-2.3***
0.2*
Order Duration (Sec)
8.55
4.15
8.38
-4.40***
-4.23***
-0.17***
Cancelation Duration (Sec)
8.63
4.19
8.45
-4.44***
-4.27***
-0.18***
Execution Duration (Sec)
6.86
5.49
6.72
-1.36***
-1.22***
-0.14***
Odd-Lot Order (%)
27.0%
20.9
26.6
-6.1***
-5.7***
-0.4%***
Order Size
421.65
362
412.49
-59.51***
-50.35***
-9.16***
48
Panel B: NYSE/ARCA Stocks
Post
3.47
1.90
1.60
0.10
1.01
1.04
55.6
5.7
9.05
9.12
6.96
27.4
566.73
During-Pre
29.08***
15.98***
13.37***
0.01***
8.21***
9.14***
15.2***
-1.1***
-3.89***
-3.91***
-1.13***
-4.1***
-62.88***
During-Post
28.82***
15.87***
13.22***
0.00
8.20***
9.03***
15.1***
-1.4***
-3.75***
-3.77***
-0.92***
-3.6***
-54.54***
Post-Pre
0.27***
0.11**
0.15***
0.01***
0.01
0.11**
0.1%
0.3***
-0.14***
-0.14***
-0.20***
-0.5***
-8.35***
Pre
During
Post
Messages Per Sec
3.71
48.62
3.88
New Orders Per Sec
2.49
35.20
2.56
Canceled Orders Per Sec
1.29
13.61
1.36
Executed Orders Per Sec
0.12
0.22
0.14
New Buy Orders Per Sec
1.43
19.52
1.45
New Sell Orders Per Sec
1.36
18.75
1.41
Order Cancelation Rate (%)
57.6
72.4
58.0
Order Execution Rate (%)
8.5
4.9
8.5
Order Duration (Sec)
7.63
2.69
7.41
Cancelation Duration (Sec)
7.74
2.72
7.49
Execution Duration (Sec)
6.41
4.92
6.33
Odd-Lot Order (%)
25.0%
16.5%
24.6%
Order Size
191.58
180
189.56
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively.
During-Pre
44.91***
32.71***
12.33***
0.10***
18.10***
17.39***
14.8***
-3.6***
-4.94***
-5.02***
-1.49***
-8.5***
-11.62***
During-Post
44.75***
32.64***
12.25***
0.08***
18.08***
17.34***
14.3***
-3.6***
-4.72***
-4.78***
-1.41***
-8.1***
-9.59***
Post-Pre
0.17**
0.07
0.07***
0.03***
0.02
0.05
0.4***
0.0
-0.22***
-0.24***
-0.08
-0.4**
-2.02
Messages Per Sec
New Orders Per Sec
Canceled Orders Per Sec
Executed Orders Per Sec
New Buy Orders Per Sec
New Sell Orders Per Sec
Order Cancelation Rate (%)
Order Execution Rate (%)
Order Duration (Sec)
Cancelation Duration (Sec)
Execution Duration (Sec)
Odd-Lot Order (%)
Order Size
Panel C: NASDAQ listed Stocks
Pre
3.20
1.78
1.46
0.09
1.00
0.93
55.5
5.4
9.19
9.25
7.17
27.8
575.08
During
32.29
17.76
14.82
0.10
9.21
10.07
70.7
4.3
5.30
5.35
6.04
23.7
512
49
Table 13: Match Summary Statistics
This table presents summary statistics for quote stuffing-stocks (QS) and matched Non quote
stuffing-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.
Panel A: Firm Characteristics
QS-Stocks
Non-QS Stocks
Difference(QSNonQS)
MktCap($Million)
2378
2398
-20
Daily Volume(1000s)
877
952
-75
Return(%)
0.047
0.048
-0.001
Closing Price ($)
25.47
25.73
-0.27
*,**,*** Indicate statistically significant difference at the 10%, 5%, and 1% level respectively
50
Table 14: Differences in QS-Stocks and Non QS -Stock Market Quality Stats by period
This table reports difference in mean market quality statistics for QS-stocks and non QS-stocks.
NQS is the mean number of quotes 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
and then averaged over the one-minute interval, HighLow is the highest
quoted midpoint in the one-minute interval minus the lowest quoted midpoint in the interval.
NQSQS-NQSnon-QS QsprdQS-Qsprdnon-QS
PqsprdQS-Pqsprdnon-QS HighlowQS-HighlownonQS
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+10
*,**,***
59***
0.024***
0.002***
0.005***
55***
0.023***
0.002***
0.005***
57***
0.023***
0.002***
0.005***
74***
0.025***
0.002***
0.005***
87***
0.027***
0.002***
0.006***
103***
0.025***
0.002***
0.006***
134***
0.027***
0.002***
0.007***
177***
0.028***
0.002***
0.010***
256***
0.031***
0.002***
0.010***
815***
0.036***
0.003***
0.019***
6587***
0.051***
0.004***
0.039***
844***
0.042***
0.003***
0.019***
287***
0.032***
0.003***
0.012***
196***
0.030***
0.002***
0.011***
144***
0.030***
0.002***
0.010***
110***
0.028***
0.002***
0.009***
99***
0.025***
0.002***
0.008***
70***
0.025***
0.002***
0.007***
71***
0.024***
0.002***
0.007***
71***
0.024***
0.002***
0.006***
66***
0.025***
0.002***
0.007***
Indicate statistically significant difference at the 10%, 5%, and 1% level respectively
51
Table 15: Difference-In-Difference Regression
This table reports difference-in-difference regression analyses analyzing the impact of quote
stuffing (QS) on market quality for stocks that experience a quote stuffing event (QS-stock)
and matched stocks that do not experience a quote stuffing event (non QS-stocks). Qsprd is
the average spread (ask price minus bid price) of the one minute interval, Pqsprd is the
spread scaled by the midpoint
and then averaged over the one-minute interval,
HighLow is the highest quoted midpoint in the one-minute interval minus the lowest quoted
midpoint in the interval. Qsprddiff is the Qsprd for QS sotck i in one minute interval t for QS
stocks less the Qsprd for non QS-stock j in minute t , PqsprdDiff and HMLowdiff are
calculated similarly. The following model is then estimated:
During is a dummy variable equal to 1 for event segments and 0 otherwise; Post is a dummy
variable equal to 1 for the period following the event, MidpvolitDiff, is the standard deviation
of the quote midpoint during minute t for the QS-stock less the standard deviation of the
quote midpoint for the matched non QS-stock during minute t,
is the number of
trades executed in minute t for the QS-stock less the number of trades executed in minute t
for the non QS-stock. We include event window fixed effects in each regression which
uniquely identifies each event. T-Stats are reported in parenthesis and are based on event
cluster corrected robust standard errors.
QsprdDiff
PqsprdDiff
HMLdiff
Post
-0.00053
-0.00006
-0.00049
(-0.938)
(-1.484)
(-1.054)
During
0.00959***
0.00068***
0.01339***
(13.073)
(12.668)
(10.276)
MidpvolitDiff
0.60918***
0.02843***
4.27990***
(4.192)
(4.330)
(5.297)
-0.00004***
-0.00000***
-0.00003
(-4.547)
(-4.543)
(-0.768)
Constant
0.02452***
0.00186***
0.00031
(61.593)
(72.174)
(0.202)
Observations
361,507
361,507
R-squared
0.88
0.88
F test
78.96
83.47
*,**,*** Statistically significant at the 10%, 5%, and 1% level respectively
52
361,507
0.93
823.4
9:30 AM
9:43 AM
9:56 AM
10:09 AM
10:22 AM
10:35 AM
10:48 AM
11:01 AM
11:14 AM
11:27 AM
11:40 AM
11:53 AM
12:06 PM
12:19 PM
12:32 PM
12:45 PM
12:58 PM
1:11 PM
1:24 PM
1:37 PM
1:50 PM
2:03 PM
2:16 PM
2:29 PM
2:42 PM
2:55 PM
3:08 PM
3:21 PM
3:34 PM
3:47 PM
# of Quotes
9:30 AM
9:43 AM
9:56 AM
10:09 AM
10:22 AM
10:35 AM
10:48 AM
11:01 AM
11:14 AM
11:27 AM
11:40 AM
11:53 AM
12:06 PM
12:19 PM
12:32 PM
12:45 PM
12:58 PM
1:11 PM
1:24 PM
1:37 PM
1:50 PM
2:03 PM
2:16 PM
2:29 PM
2:42 PM
2:55 PM
3:08 PM
3:21 PM
3:34 PM
3:47 PM
# of Quotes
Figure 1: Examples of Quote Stuffing Events
WHR - January 15, 2010
50000
45000
40000
35000
30000
25000
20000
15000
10000
5000
0
Time of Day
WMGI - June 14, 2010
40000
35000
30000
25000
20000
15000
10000
5000
0
Time of Day
53
9:30 AM
9:43 AM
9:56 AM
10:09 AM
10:22 AM
10:35 AM
10:48 AM
11:01 AM
11:14 AM
11:27 AM
11:40 AM
11:53 AM
12:06 PM
12:19 PM
12:32 PM
12:45 PM
12:58 PM
1:11 PM
1:24 PM
1:37 PM
1:50 PM
2:03 PM
2:16 PM
2:29 PM
2:42 PM
2:55 PM
3:08 PM
3:21 PM
3:34 PM
3:47 PM
# of Quotes
Figure 1 Continued:
HSC - November 11, 2010
30000
25000
20000
15000
10000
5000
0
Time of Day
54
-25
-24
-23
-22
-21
-20
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Pqsprd
0.005
Qsprd
Figure 2
Spread by Period
0.01
0.14
0.009
0.12
0.008
0.007
0.1
0.006
0.08
0.004
0.06
0.003
0.04
0.002
0.001
0.02
0
0
Pqsprd
Qsprd
55
-25
-24
-23
-22
-21
-20
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Voltil, Effsprd
Figure 3
Effective Spread and Volatility by Period
0.045
0.04
0.035
0.03
0.025
0.02
0.015
0.01
0.005
0
Voltil
EffSprd
56
-30
-29
-28
-27
-26
-25
-24
-23
-22
-21
-20
-19
-18
-17
-16
-15
-14
-13
-12
-11
-10
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
Standardized # of Quotes
4.5
4
3.5
3
0.1
2.5
2
0.05
1.5
0
1
-0.05
0.5
-0.1
0
-0.15
-0.5
-0.2
Quotes
Trades
57
Standardized # of Trades
Figure 4
Standardized Quotes and Trades
0.25
0.2
0.15