dimensions of execution quality: recent evidence for u.s...
TRANSCRIPT
Dimensions of execution quality: Recent evidence for U.S. equity markets
Ekkehart Boehmer
306L Wehner Building Mays Business School Texas A&M University
College Station, TX 77845-4218 979-690-2626
First draft: March 30, 2003
This draft: October 15, 2003
Abstract This research provides a post-decimals analysis of market order execution quality in U.S. equity markets, using order-based data reported in accordance with SEC Rule 11Ac1-5. These data facilitate a comprehensive investigation of multiple dimensions of execution quality, including measures of costs and speed, for large samples of common stocks on Nasdaq and the NYSE. The evidence is consistent with highly competitive equity markets. Overall execution costs on Nasdaq exceed those on the NYSE, but orders are executed significantly faster. This relation is reversed for larger orders of 5,000 shares or more. The apparent trade-off between costs and speed persists throughout the results, which suggests that inferring execution quality from out-of-pocket costs alone may be problematical. It also illustrates the need for models of trader behavior that can accommodate more than one dimension of execution quality.
JEL Codes: G24, G23 Keywords: Securities trading, Order execution quality, SEC Rule 11Ac1-5
I thank James Angel, Paul Bennett, Hendrik Bessembinder, Beatrice Boehmer, Kee Chung, Amy Edwards, Mark Gurliacci, Charles Jones, Pamela Moulton, Gideon Saar, Lei Yu, an anonymous referee, and participants at NYSE workshops and the 2003 FMA meeting for very valuable comments and discussions. Christopher Gieckel was very helpful in assembling the data. This paper was largely completed while the author was a Director of Research at the New York Stock Exchange. The opinions expressed in this paper do not necessarily reflect those of the members or directors of the NYSE.
1 Introduction
The economic importance of execution quality in equity markets has generated substantial
attention from financial economists. Yet, data limitations have largely confined analysis to a single
dimension of execution quality, the out-of-pocket costs of completing an order, and have required
approximate algorithms to estimate these costs. While costs are clearly the single most important
component of execution quality, the recent proliferation of alternative trading systems, automated
trading algorithms, and online trading suggests that the speed of order executions is equally
important to some traders.
This study uses novel data that eliminate the need for approximations and allow a
simultaneous analysis of two dimensions of execution quality, costs and speed. I compare market-
order executions on the two dominant U.S. equity markets, the New York Stock Exchange (NYSE)
and the Nasdaq Stock Market (Nasdaq). Reports published in accordance with Securities and
Exchange Commission Rule 11Ac1-5 allow a comparison based on actual orders. Since fall 2001,
the rule has required U.S. market centers to report various standardized measures of execution
quality for orders below 10,000 shares in nearly all publicly traded securities. Compared to the
traditional approach of estimating execution costs from trade reports, order-based analysis does not
require approximate algorithms to determine trade direction and the timing of benchmark quotes.
Moreover, Rule 11Ac1-5 makes the average order execution speed, or the period between order
receipt and execution, publicly available.
If traders value speed, a negative relation between execution cost and execution speed would
suggest that the lowest-cost market is not necessarily the best market. My comparison of order
execution on the NYSE and Nasdaq provides systematic evidence of such a negative relation. This
implies that equity markets may be in a competitive equilibrium even when out-of-pocket execution
2
costs are systematically higher in one market. It also suggests caution when interpreting earlier
studies conducted before data on execution speed became available.1
Execution costs in U.S. equity markets have been a frequent academic research topic. Studies
generally fall into one of two categories: analysis of the same stocks trading in different markets, or
analysis of different stocks across markets. In the former, researchers study firms that either trade in
multiple locations (see, for example, Lee, 1993; and Easley, Kiefer and O’Hara, 1996) or have
switched trading venues (see, for example, Christie and Huang, 1994; and Barclay, 1997). In the
second approach, researchers match firms by characteristics that control for ex-ante differences in
execution costs (see, for example, Huang and Stoll, 1996; Bessembinder and Kaufman, 1997a,
1997b; Bessembinder, 1999; and SEC, 2001).
These studies were completed before substantial changes in the structure of U.S. equity
markets. The move to decimalization was completed in April 2001. SEC Rule 11Ac1-5 was
implemented between July and October 2001. Since January 2002, the NYSE has publicly displayed
all limit orders, and Nasdaq’s Supermontage quotation and execution system was launched in
October 2002. These structural changes may have had substantial effects on relative execution costs.
Evidence presented by Bessembinder (1999) and Weston (2000), for example, shows that changes in
the Nasdaq order handling rules in 1997 narrowed execution cost differences between Nasdaq and
the NYSE.2
Most researchers use publicly available trade-based data to estimate execution costs. The
drawback is that order size, order direction, and order arrival time are not observable and must be
estimated using approximation methods (see Bessembinder, 2002, for a summary). Stock or period-
specific systematic biases may affect the estimates and comparisons across markets may also be
misleading if trade report delays differ across markets (see Bessembinder, 1999). Only the SEC’s
1 Throughout the paper, the term “execution quality” refers to several components that concern a trader,
including effective spreads, realized spreads, and the speed of execution. I use the term “execution costs” to refer to a trader’s out-of-pocket costs excluding commissions (and therefore as a synonym to “effective spreads”).
2 Decimalization has received some attention in the literature. See, for example, Bacidore, Battalio, and Jennings (2003) and Bessembinder (2003b), who show that the reduction in tick size that accompanied decimalization has significantly changed order submission strategies and the relation between order size and execution costs.
3
(2001) execution quality study uses order data to compare Nasdaq to the NYSE.3 For orders of under
5,000 shares during a five-day period in June 2000, it finds that NYSE execution costs are below
Nasdaq costs (except for the smallest trades in the largest stocks, where the cost differential
vanishes), but Nasdaq orders generally execute faster. I extend this analysis in two ways. First, I
include orders between 5,000 and 9,999 shares. Second, using a large sample of 2,136 common
stocks, I examine a longer period from November 2001 through December 2002 that encompasses
recent structural changes in equity markets.
Overall, I find order execution is significantly more expensive on Nasdaq than on the NYSE
in terms of effective spreads, whether measured in dollars or relative to share price. This differential
cannot be explained by more informed order flow, because realized spreads are also significantly
higher for Nasdaq orders, and there tends to be less informed order flow on Nasdaq. I also document
that orders execute significantly faster on Nasdaq than on the NYSE, although the execution quality
differentials change with order size, and reverse for orders between 5,000 and 9,999 shares. These
larger orders execute more cheaply on Nasdaq, but faster on the NYSE. Finally, I document a market
wide decline in execution cost during the 14-month sample period that is somewhat more
pronounced for Nasdaq stocks. Nasdaq’s cost disadvantage relative to the NYSE persists, but is
diminished over the sample period. These results suggest that despite the recent changes in the
structure of U.S. equity markets, Nasdaq execution costs are still higher than NYSE execution costs.
3 Some studies use proprietary data on institutional orders to compare execution costs. Keim and Madhavan
(1997) find higher costs on Nasdaq, and Chan and Lakonishok (1995) show that, when commissions are included, Nasdaq executions are cheaper in smaller firms. In this paper, I focus on a complementary set of orders that uses the universe of market orders below 10,000 shares, rather than selected samples of large institutional orders.
Other studies based on order data compare execution quality of different order types within one market. For example, Harris and Hasbrouck (1996) compare market orders to limit orders, and Peterson and Sirri (2002) compare market orders to marketable limit orders on the NYSE.
Finally, three recent studies use Dash 5 data to explain order-routing decisions for NYSE-listed securities. Lipson (2003) examines differences in execution quality and order flow characteristics across nine market centers. He finds that market centers specialize in certain types of order flow and consequently exhibit different execution costs. Bessembinder (2003a) also documents that order flow characteristics differ across market centers. He uses a two-stage estimation procedure to investigate the effect of selection bias on execution-cost measures. Boehmer, Jennings, and Wei (2003) analyze whether execution quality statistics published in Dash 5 reports affect subsequent order routing decisions in NYSE-listed securities.
4
The important result is that high execution costs are systematically associated with fast
execution speed, and low costs are associated with slow execution speed. This relationship holds
both across markets and across order sizes. While no previous authors develop systematic evidence
or a theoretical model of the relationship between execution cost and execution speed, Battalio,
Hatch, and Jennings (2003) and Boehmer, Jennings, and Wei (2003) show that both costs and speed
are important. For a proprietary sample of small retail orders submitted in March 1999, Battalio et al.
document that Trimark Securities, Inc. executes orders at higher costs, but faster than the NYSE.
They argue that additional dimensions of execution quality, beyond out-of-pocket execution costs,
may be relevant in comparing different execution venues. Boehmer et al. analyze whether order
routing decisions in NYSE-listed securities depend on past execution quality. They find that a
market center receives more order flow when either its reported execution cost declines or its
execution speed increases. I go beyond these indicative results and provide systematic evidence,
using orders submitted to all NYSE and Nasdaq market centers, of a negative relationship between
costs and speed that is robust to different samples and methods. I provide a rationale for this
relationship that relies on differences in order handling across markets and on differences in how
market makers infer informed order flow.
The remainder of this paper is organized as follows. In section 2, I discuss data sources and
provide a detailed description of sample selection and empirical methodology. In section 3, I present
results on execution quality differentials and investigate how they evolve over time. I use section 4
to discuss the trade-off between execution costs and execution speed and identify features of current
equity market design that help explain how it arises. The final section concludes.
2 Data and methodology
The sample and several control variables are constructed using data from the NYSE’s Trade
and Quote (TAQ) data, the Center for Research in Security Prices (CRSP), and Compustat. I use
both a matched sample and a comprehensive sample with cross-sectional and time series controls to
compare execution quality. The sample selection criteria closely follow the ones used in SEC (2001).
Table 1 summarizes the procedure and shows the weight of each criterion. The initial sample
5
consists of all domestic securities on September 30, 2001 that are included in CRSP: 2,579 on the
NYSE and 3,949 on Nasdaq. I apply four categories of filters to this set: (1) basic CRSP filters that
retain only single-class common stocks with at least a two-year return history; (2) CRSP trading
filters that assure a minimum activity level during the third quarter of 2001; (3) TAQ trading filters
that eliminate low-priced and inactive securities; and (4) a filter based on Rule 11Ac1-5 data that
retains only securities that have a continuous series of monthly reports over the entire period (not
necessarily for the same order size and type). These filters result in a final sample of 1,043 NYSE-
listed and 1,093 Nasdaq-listed securities. Considering the interim market downturn that led to net
exit from the stock markets, this sample is comparable to the final samples in SEC (2001) of 1,141
NYSE and 1,441 Nasdaq securities.
2.1 Matching procedure
I closely follow the procedure in SEC (2001) to identify matches between Nasdaq stocks and
NYSE stocks. To avoid potential hindsight bias, I measure the matching criteria mostly during the
third quarter of 2001, which precedes the analysis period. I select a sample of Nasdaq stocks
stratified by dollar trading volume during the third quarter of 2001. I sort the 1,093 Nasdaq stocks in
order of diminishing volume and select every fifth stock, starting with the most active. This results in
a sample of 219 securities. Next, I produce three different rankings of the 1,093 Nasdaq stocks,
according to (1) market capitalization (as of September 30, 2001), (2) dollar trading volume, and (3)
share volume (both during the third quarter of 2001). Thirty-five different securities appear in the top
20 of at least one of these three lists (“Top20 stocks”). Out of those, the volume-stratified sample
already includes five. The remaining 30 securities are then added to the stratified sample, resulting in
249 Nasdaq securities. Including the Top20 stocks assures that the Nasdaq stocks with the lowest ex
ante execution cost are represented in the matched-pairs analysis. All results are qualitatively similar
without the additional Top20 stocks.
Using one-to-one matching without replacement, the selected Nasdaq stocks are then
matched to 249 NYSE stocks. When we match without replacement, it is difficult to determine the
optimal sequence of selections. For example, suppose that NYSE stock A is the closest match to
6
Nasdaq stock Y. Then A is removed from the set of remaining potential matches for the other Nasdaq
stocks, although the overall matching error may have been lower if we assigned A to a different
stock. To my knowledge, there is no algorithm that finds optimal matches without comparing each
possible permutation. I use a random order of matching by sorting the 249 Nasdaq stocks by symbol
(SEC (2001) sorts by name). Then the best-matching NYSE stock is selected, in that order, for each
Nasdaq firm.
The matching criteria consider four dimensions: market capitalization (MCAP) and share
price (PRC) on September 30, 2001, and average adjusted daily dollar volume (ADV) and the
average daily relative price range (RR) during the third quarter of 2001. Volume is adjusted to take
different counting procedures across markets into account; I compute it by multiplying Nasdaq
volume by 0.7 and leaving NYSE volume unchanged.4 The volatility measure, the daily relative
range, is defined as the daily range divided by the closing (or last sale) price.5 For each of the 249
Nasdaq stocks i and each of the 1,043 NYSE stocks j, I then compute average pairwise matching
errors Dij, which are weighted equally across the four dimensions, as follows:
1111 −+−+−+−=j
i
j
i
j
i
j
iij RR
RRADVADV
PRCPRC
MCAPMCAPD (1)
The NYSE stock with the lowest error is then selected as the matching firm and removed
from the sample of potential matches for the remaining Nasdaq securities.6 In sensitivity tests, I
4 See Dyl and Anderson (2002) and SEC (2001, footnote 16). I obtain qualitatively identical results when
matching without this adjustment. 5 SEC (2001) uses a different volatility measure, the weekly return volatility over 29 months preceding the
analysis period. An alternative 12-month measure produces similar results. I use a volatility measure based on daily ranges because it does not rely on stationarity assumptions over such an extended period. Alternatively, one could use some measure of intra-day volatility, but it would be highly sensitive to the way it is computed. It does not seem likely that any of my conclusions are sensitive to the choice of volatility measure.
6 SEC (2001) uses an intentionally biased weighting scheme that deems a 0.05 error in the volatility component optimal. Because there is no theoretical justification for deviations from considering zero errors optimal, I do not follow that approach. Empirically, using the 0.05 target error in the volatility component does not change any results.
7
consider a subset of 86 well-matched pairs that have a matching error below 0.7. Because all results
are qualitatively identical for this subsample, I report only results using the complete set of matches.7
The appendix provides a list of all matched pairs, and Table 2 presents descriptive statistics
for the pairs, including sample medians for both markets and the median pairwise differences.
Overall, the pairwise differences in the control variables are small compared to the respective sample
medians. In most cases, however, the differences remain statistically significant, indicating the need
for further controls in the analysis.
2.2 Execution quality data
On November 17, 2000, the Securities and Exchange Commission released new rules that
mandate standardized monthly disclosures of order flow and order execution quality. For each
symbol and month, Rule 11Ac1-5 (“Dash 5”) reports include round-trip effective spreads, round-trip
realized spreads, and measures of execution speed and order flow. Effective spreads are computed as
twice the difference between the execution price and the quote midpoint prevailing at the time an
order was received. Realized spreads for buy (sell) orders are defined as twice the (negative)
difference between the execution price and the quote midpoint five minutes later. Each measure is
reported for four different order types and four different order sizes up to 9,999 shares. These reports
have to be published by each market center (including exchange specialists, Nasdaq market makers,
alternative trading systems, and generic Nasdaq execution systems).8
For this study, I obtain Dash 5 reports on market orders and marketable limit orders for all
NYSE and Nasdaq securities from www.marketsystems.com (MSI). Because Dash 5 reports are
made by each individual market center, the reports for each stock and month have to be aggregated.
To compare NYSE and Nasdaq execution quality, it seems appropriate to include only market 7 It is not possible to directly compare the quality of matches to SEC (2001). The only information provided
there regarding the quality of matches is that 58 of 221 pairs, or 26.2%, have a matching error below 70%. In my study, 86 of 249 pairs, or 34.5%, fall into that category. If we assume the distribution of errors is similar in the two samples, this comparison indicates that the overall matching success is slightly better in my study.
8 The rule was published as SEC Release No. 34-43590. It provides specific criteria for eligible orders and lists the definitions of all measures it requires to be published by each market center. See the rule text at http://www.sec.gov/rules/final/34-43590.htm for details.
8
centers that are directly associated with one of these markets. Thus, for Nasdaq stocks, I compute
averages weighted by the number of executed shares across all reporting NASD members (Rule
11Ac1-5 participant ID starting with “T”). Averages are computed separately for each of the
different Dash 5 order type and order size categories. This process captures all reporting broker-
dealers (including ECNs) in addition to the generic Nasdaq systems. For NYSE stocks, the main
reporting entity for each stock is the responsible specialist firm. To be consistent with the treatment
of Nasdaq market centers, which often route orders away, I include all outgoing Intermarket Trading
System (ITS) orders by computing share-weighted averages for each Dash 5 category.9
The Dash 5 regulations invariably lead to double-counting of certain orders. Each market
center reports the number of shares executed on its market and shares executed elsewhere, but
provides execution-quality measures only for all orders received, regardless of where they are
executed. This implies that all orders routed from one (Dash 5 reporting) market center to another
will generally be reported twice. For NYSE securities, this primarily affects orders routed through
ITS, which is rarely used (about 2% to 3% of shares executed in my sample). For Nasdaq securities,
double-counting may be more prevalent for two reasons; there is no dominant centralized execution
service, and routing orders away is part of some market center business models.10
There is little reason to believe, however, that this double-counting will systematically bias
the measurement of execution quality. While several markets may report the same order, each will
usually report different execution quality. This is because the routing recipient would record the
order later than the sender did, and all measures are based on order receipt time. Moreover, one
9 An alternative approach would be to also include off-exchange trading. For example, orders in an NYSE-listed
security can be submitted to several regional exchanges or NasdaqIntermarket broker-dealers. Similarly, orders in Nasdaq-listed securities can currently be submitted to the American, Boston, Cincinnati, or Chicago exchanges. Including these additional market centers might be more appropriate for an issuer who needs to choose between a Nasdaq listing and an NYSE listing. I ignore off-exchange orders, because my main objective is to compare execution costs on the two major equity markets. This exclusion will tend to overstate NYSE execution costs, because competing markets often attract “easy” order flow (see Bessembinder, 2003a).
10 SuperMontage, for example, executed around 20% of Nasdaq prints during its first months of operation. Because non-market maker limit orders (as opposed to dealer quotes) have been accepted only since February 10, 2003, essentially all executions on this system are reported at least twice before this date (see Nasdaq Head Trader Alerts #2003-018 and #2003-079, available at www.nasdaqtrader.com). SuperMontage published its first Dash 5 report in December 2002, so there should be little adverse effect on the results reported in this paper.
9
could argue that both markets’ measures are relevant for traders. The Dash 5 report by the sender
(based on execution at the recipient market) represents the actual execution cost for the trader. The
Dash 5 report by the recipient market represents the cost from the perspective of the sending market,
or, alternatively, the trader’s hypothetical cost had he searched longer and then submitted to the
second market center directly.
Dash 5 reports publish effective spreads only for market orders and marketable limit orders,
but my analysis is based on market orders only. In many respects, it is easier to interpret the results
for market orders than those for marketable limits. First, because Dash 5 reports do not include
information on the opportunity cost of non-execution, ex post execution cost for marketable limits
will understate their true cost. Consequently, estimates for marketable limits would not be
comparable to those in SEC (2001), which uses an ex-post adjustment for unfilled marketable limits.
This analysis cannot be replicated using Dash 5 data, because they include only monthly aggregates.
Second, the time-to-execution for this order type is censored, because cancelled and expired orders
are not considered in the computation. Third, summary statistics on speed are dominated by orders
that happen to be submitted as the market moves away, and therefore do not execute immediately.
Finally, usage of marketable limit orders differs systematically across markets. All NYSE specialists
accept market orders, but some Nasdaq market centers do not. For example, some marketable limits
reported by Island, which does not accept market orders, are probably functionally equivalent to
market orders. For all these reasons it is difficult to interpret potential differences in execution
quality across markets for marketable limits.11
Dash 5 statistics are reported for four order-size categories: 100-499, 500-1,999, 2,000-4,999,
and 5,000-9,999 shares. Thus, the Dash 5 market order data include up to four observations for each
month-security combination. Not all market centers report in each category in each month,
especially in the larger order sizes, so there are fewer than four observations on average. Several
market centers began publishing their Dash 5 reports in June 2001, but Nasdaq participants were
11 See Peterson and Sirri (2002) for a comparison of the two order types.
10
exempt until September 2001. To leave a one-month adjustment period after implementation of the
rule, I analyze the period from November 2001 through December 2002.
The Dash 5 market order data for the final sample include 115,023 security/month/order size
combinations. Of these, 58,009 are for the 1,093 Nasdaq stocks, and 57,014 are for the 1,043 NYSE
stocks. To eliminate potential errors and non-representative Dash 5 variables, I impose two filters.
First, I drop all order-size specific observations where the reported effective or realized spreads
exceed 50% of the average share price during that month. This filter eliminates 125 Nasdaq
observations and 22 NYSE observations, including several implausible or erroneous data points.
Second, for each security and each month, I drop all order size categories based on fewer than 20
orders. This filter eliminates 17,308 Nasdaq observations and 10,400 NYSE observations, but
eliminates only a small number of shares executed (0.48% of NYSE executed market order shares
and 0.79% of Nasdaq market order shares). Results are qualitatively identical throughout the
analysis without applying this filter.
Over the 14-month period on which the analysis is based, the final data set on the 249
matched firms covers executed market orders of 33 billion shares on Nasdaq and 23 billion shares on
the NYSE. For the entire sample of 1,093 Nasdaq and 1,043 NYSE securities, the data cover 48
billion market order shares on Nasdaq and 73 billion market order shares on the NYSE. To provide
an indication of the representativeness of the sample, Table 3 shows the percentage of covered
orders for different order sizes. Panel A presents the aggregate number of shares executed during the
14-month sample period as reported to the Consolidated Tape (CT). These figures include only
trades reported by either Nasdaq or the NYSE during regular market hours. I separate out trades
between 100 and 9,999 shares because they correspond to the order sizes covered by Dash 5 reports.
Unfortunately, no publicly available data source allows a categorization of reported trades based on
the original order size, so I use trade sizes as an approximation.12 To relate the Dash 5 market order
12 Trade size may be smaller or larger than original order size. For example, partial consecutive executions of a
large buy order against several smaller sell orders imply that CT trade size is smaller than order size for the buy. The opposite occurs if the large buy executes against several small sells in one transaction.
11
volume to market-wide activity, I divide order volume by twice CT volume, which counts only one
side of each trade. These percentages are presented in Panel B.
Overall, the NYSE sample represents 24% of small trades and 13% of all trades. The
corresponding percentages for Nasdaq are 12% and 8%, respectively. The differences between the
two markets reflect the greater prevalence of marketable limits and smaller trade prints on Nasdaq.
The distribution of order sizes is similar across the two markets. These comparisons highlight the
main disadvantage of using Dash 5 data; they represent only a fraction of actual order volume. Their
advantage over trade data from TAQ, however, is that Dash 5 data are based on a homogeneous set
of actual orders, whose submission time, type, size, and direction are known to the researcher.
2.3 Methodology
I compute three alternative measures of the Nasdaq-NYSE execution quality differential. The
first set of results is based on the matched sample. It includes pairwise differences for four measures
of execution quality: effective spreads measured in dollars and in basis points (standardized by the
monthly average of daily closing prices), realized spreads in dollars, and the time between order
receipt and execution.13 Each measure is based on firm-month averages of the underlying orders,
weighted by the number of shares executed. Consequently, whenever these measures are further
aggregated across order sizes or months, I use the corresponding number of shares executed as
weights. Statistical comparisons across matched pairs or individual stocks are always based on
equally weighted averages in that cross-section. I conduct both Wilcoxon tests of the hypothesis that
the median pairwise differences are zero and t-tests of the hypothesis that the mean pairwise
differences are zero.
The second set of results also uses the matched pairs, but adds additional control variables.
The controls serve two purposes. First, they help control for residual matching errors. Second, by
incorporating monthly time-varying control variables, this approach adjusts for potentially different
13 Results using percentage realized spreads, computed analogously to percentage effective spreads, are
qualitatively identical and are therefore not reported.
12
time paths of the matching variables between Nasdaq and the NYSE. The panel regression model
estimated has one observation for each matched pair i and each month t:
itRR)ADVln()PRC/()MCAPln(IEQM ititititt
ttit εφδγβα +∆+∆+∆+∆+=∆ ∑=
114
1 (2)
where ∆ represents the difference Nasdaq – NYSE, and EQM is one of the execution quality
measures. The monthly control variables are based on equally weighted averages of daily values.
Instead of an intercept, the model uses fixed time effects, It, that equal one in month t and zero
otherwise. This approach allows the monthly difference between Nasdaq and NYSE execution
quality, αt, to vary over time. I focus on the mean coefficient, ∑ =
14114
1t tα , as a measure of execution
cost differences during the entire period, but the median yields qualitatively identical conclusions. I
use the (White) heteroskedasticity-consistent estimator for the standard errors of regression
coefficients.
The third approach uses a panel of all securities in the broad sample, and regresses the
execution quality measures on the control variables and a dummy variable indicating a Nasdaq
security:
ititititititttit RRADVPRCMCAPNasdaqIEQM εϕφδγβα ++++++=∑ )ln()/1()ln( (3)
In this regression, the coefficient β measures the difference in execution quality between Nasdaq and
the NYSE.14
14 SEC (2001) expresses the continuous independent variables as deviations from the Nasdaq mean. This
transformation affects only the estimated intercept coefficient α and its standard error. It does not change the estimated coefficient or standard error of the variable of interest, Nasdaq, or those of any other variable except the intercept. Thus, for simplicity, I use the untransformed variables in this regression.
Moreover, following SEC (2001), I also estimate a model that controls for monthly variation in the earnings/price ratio:
itititititititititit EsignusEPEPplusRRADVPRCMCAPNasdaqEQM ελκηϕφδγβα +++++++++= )minln()ln(ln()/1()ln( The additional control variables represent the earnings/price ratio if earnings are positive (EPplus), the earnings/price ratio if earnings are negative (EPminus), and a dummy variable that is one when earnings are positive (Esign). These three variables are based on the previous quarter’s earnings and current market capitalization. This model produces
13
Because execution quality may be correlated across securities, the estimated standard errors
in equation (3) may be underestimated. This is not as problematic in equation (2), because it is less
likely that the matched differences are correlated across securities. Despite the potential statistical
problems, using an extensive sample of securities may allow broader inferences. Finally, because
both models allow the control variables to vary over time, these specifications control for certain
time-specific effects. Alternative specifications of (2) and (3) using static controls, measured during
2001Q3, do not qualitatively alter the results. Similarly, results are virtually identical when equation
(3) is estimated without fixed time effects.15
3 Results
The analysis focuses on three different measures of execution quality: round-trip effective
spreads, round-trip realized spreads, and execution speed, or the time between order receipt and
execution. Effective spreads, which can be interpreted as the total price impact of a trade, measure
the non-commission out-of-pocket costs of a trader. They can be decomposed into a permanent and a
temporary component. Because the permanent component approximates the information component
of the trade, wider effective spreads may reflect more informative order flow, and not necessarily
higher execution costs for comparable orders. Therefore, I also report differences in realized spreads
(the temporary component) and price impacts (the permanent component, defined as half the
difference between effective and realized spreads). Realized spreads can be interpreted as a market’s
inherent execution cost, because they exclude the effects of the information content of order flow.
Execution speed is an important component of execution quality for some traders.16 Other
things equal, traders prefer faster executions for a variety of reasons. First, prices fluctuate over time
and a longer wait impairs a trader’s ability to react to price moves quickly. Second, because many
results that are qualitatively identical to the ones reported in the paper, but they tend to be somewhat more pronounced in magnitude.
15 The results of both models are robust to different specifications of time fixed effects. Specifically, I obtain qualitatively identical results using (1) no time effects; (2) annual time effects; and (3) period effects that divide the 14-month sample period into two, three, or four subperiods.
16 Blume (2001) cites a May 2000 survey that finds 58% of online traders rate speed as more important than a favorable price.
14
orders are executed at prices different from the quoted spreads, a longer wait makes it more difficult
for a trader to predict the execution price accurately. While the realized execution price may be
better or worse than the expected execution price, the increased uncertainty is undesirable for risk-
averse traders. Moreover, a longer wait makes it more difficult to synchronize automated trading
algorithms, where an order submission decision often depends on the outcome of a previous order.
Third, for orders that are not sent to automatic execution systems, a longer wait may make traders
more apt to perceive adverse selection to their disadvantage, whether justified or not. This would
occur, for example, if executions are fast when prices move in the trader’s favor, and take longer
when prices move against them. Finally, a longer wait may increase the risk of front running to the
traders’ disadvantage. The actual trade-off between execution speed and out-of-pocket costs depends
on individual trader characteristics and is difficult to assess. Therefore, it is not possible to
determine, say, what execution delay traders are willing to incur in exchange for a certain reduction
in effective spreads.
3.1 Execution quality across markets
Table 4 reports results for the execution quality measures aggregated over time and the four
order-size categories. Panel A presents (equally weighted) means and medians for the univariate
matched-sample comparison. It reports both the levels for the two markets and pairwise execution
quality differentials (computed throughout the paper as the difference between Nasdaq and the
NYSE). Thus, a positive cost differential implies that Nasdaq is more expensive, and a negative cost
differential implies that the NYSE is more expensive. Panels B and C show results for the matched-
pairs monthly panel regression [equation (2)] and the monthly panel regression using the broader
sample [equation (3)], respectively. Because the control variables may evolve differently in the two
markets, both regressions use observations that are pooled and not aggregated over time. This
approach controls for different time paths of ex-ante execution quality in the two markets.17
17 An alternative specification to control for potentially different effects of the control variables in the two
markets would be to interact the three control variables with the Nasdaq dummy. The estimates become more difficult to interpret, however, because in this case the total effect of the Nasdaq dummy depends on the levels of the control variables.
15
The results across the three panels support similar conclusions: Effective spreads and realized
spreads are significantly lower on the NYSE, and Nasdaq executions are significantly faster. The
extent of the differences depends on the methodology. The median pairwise difference in effective
spreads (Panel A) is 2.2 cents and the mean pairwise difference is 3.9 cents. Estimated execution
costs on Nasdaq are also higher in both regression models, where the difference is just above five
cents (matched pairs in Panel B and broad sample in Panel C). A comparison of the means and
medians of the effective spread levels suggests they are more skewed on Nasdaq than on the NYSE,
which explains why the median paired difference is lower than the mean paired difference.
Measuring effective spreads in basis points leads to very similar conclusions: Univariate
estimates of the differential are 9 basis points (median pairwise difference in Panel A) and 26 basis
points (mean pairwise differential in Panel A). For the regression models, Nasdaq executions are
between 22 and 23 basis points more expensive. Although these differences in execution cost appear
relatively small, they are economically important. My sample represents an average of 8.6 billion
executed shares every month. Therefore, a one-cent difference in effective spreads represents a
monthly execution cost differential of $86 million.
While effective spreads represent the out-of-pocket costs for order submitters, realized
spreads are a better measure of the efficiency of market making. Because realized spreads can be
interpreted as the cost of trading net of the effect of trader information, they also provide a way of
controlling for order informativeness. Table 4 shows that realized spread differentials follow a
pattern similar to effective spread differentials. For the matched sample, they are on average 5.5
cents (median 2.3 cents) higher on Nasdaq. The regressions imply a differential of 6.1 cents (paired
sample in Panel B) and 5.9 cents (broad sample in Panel C). This suggests that, for the market orders
up to 9,999 shares examined here, Nasdaq’s competing dealer trading protocol is associated with
inherently higher costs than the NYSE’s centralized specialist system. Moreover, the control
variables in the paired-sample regression, which by construction controls for security-specific
characteristics, exhibit little significance. This is consistent with the view that differences in realized
spreads across markets represent fixed costs that do not depend on trading characteristics.
16
While both effective and realized spreads are wider on Nasdaq, part of the greater effective
spreads may be due to more informed order flow. To investigate this issue, I use price impacts (the
change in the midquote from order receipt to five minutes after execution) as an approximation for
order flow information. The results in Table 4 suggest that Nasdaq spreads are wider despite lower
information content. The median pairwise difference in price impacts is -0.3 cents, and the mean is -
0.8 cents (Panel A). Although the estimates from the broad sample are not significant, the matched-
pairs regression in Panel B suggests price impacts are 0.5 cents lower on Nasdaq. Therefore,
differences in informed order flow are unlikely to explain the higher Nasdaq execution costs.18
Finally, the lower spreads on the NYSE come at the cost of more time between order receipt
and execution. An order takes about twice as long (12 seconds longer) to execute on the NYSE, and
this estimate is similar by each method.19 This is an important observation, because slower
executions reduce the benefit of lower costs for some traders. We cannot assess whether the faster
execution on Nasdaq can adequately compensate for the greater cost. Nor is there any theory of
trader behavior that explicitly models the trade-off between speed and price in the context of market
orders; the trade-off is presumably governed by market conditions and the preferences of individual
traders. If price and speed are indeed substitutes, the results in Table 4 make it difficult to assess
whether a trader is better off sending an order to Nasdaq or the NYSE.
My estimates of effective and realized spreads are close to those in Bessembinder (2003b),
the most recent broad analysis of execution costs in both markets. For a matched sample of 300
18 The result of greater price impacts for NYSE orders is consistent with several measures of information
content estimated by Stoll (2000), who uses a broad sample of NYSE and Nasdaq securities from December 1997 through February 1998. It is also consistent with studies of earlier periods, such as Bessembinder and Kaufman (1997a). It is contrary, however, to Heidle and Huang’s (2002) finding that Nasdaq securities are subject to a greater probability of informed trading. They use a sample of firms that switched from Nasdaq to the NYSE in 1996 and use Easley, Kiefer, O’Hara, and Paperman’s (1996) model to estimate the likelihood of informed trading. One reason for the different results may be their sampling period, which precedes the Nasdaq order handling rules and the accelerated development of auction markets within Nasdaq. Another potential reason is the different methodology, which is derived from an explicit sequential-trade model, but ignores much information generated in the trading process.
19 One might suspect that stopped orders take longer to execute and receive better prices. For purposes of SEC Rule 11Ac1-5, however, orders that are “stopped” (i.e., receive a firm guarantee to execute later at a specific price) are deemed executed when they are stopped. Thus, they enter the Dash 5 data with the time of the guarantee in place of the actual execution time. Consequently, stopped orders do not contribute mechanically to the negative relation between speed and cost
17
security pairs in the post-decimalization period from April through August 2001, he estimates
effective spreads of 7.3 cents for the NYSE and 10.5 cents for Nasdaq. Both figures are within 10
percent of my estimates, but they are not directly comparable. First, Bessembinder uses different
weighting schemes. He computes averages that are either equally weighted first across trades and
then across stocks, or share-weighted first across trades and then across stocks. I use share weighting
across orders, and then equal weighting across securities. Second, my estimates exclude orders of
10,000 shares or more. Because larger orders tend to have wider spreads, my averages may reflect
different order sizes.20 Finally, the overall effective spread differentials in Table 4 are also close to
those reported by Weston (2000) for 88 matched pairs during the nine-month period following
October 1996. He finds a difference of 3.4 cents between (equally weighted) Nasdaq and NYSE
effective spreads, which compares to my estimated unconditional differential of 3.9 cents.
3.2 Effect of order size on differences in execution quality
Grouping all order sizes together may be misleading, because the cost differentials between
the two markets may differ depending on order size. I thus provide separate comparisons for the four
order-size categories in the Dash 5 reports. Table 5 reports the median pairwise differences in Panel
A, the estimated intercept coefficient from equation (2) in Panel B, and the estimated coefficient of
the Nasdaq dummy variable from equation (3) in Panel C.
The results reveal that the aggregate tests mask important differences across different order
sizes. Effective spreads are wider and execution is faster for small orders sent to Nasdaq, but these
relations reverse direction for orders of 2,000 shares or more. Again, the three methods yield very
similar results. For the broad sample (Panel C), I find that effective spreads are 6.5 cents (32 bp)
higher on Nasdaq for orders below 500 shares. This differential declines almost monotonically with
order size and becomes negative for orders between 2,000 and 4,999 shares, where Nasdaq is 0.5 20 Bessembinder (2003b) reports separate statistics for trade sizes between 1,000 and 9,999 shares, which are
8.6 cents (NYSE) and 10.3 cents (Nasdaq). In fact, execution reports for trades of fewer than 10,000 shares may represent partial executions of larger orders, and therefore overestimate the cost of executing orders below 10,000 shares. Similarly, trades above 9,999 shares may also represent a pooled trade print of several smaller orders. Because the process of matching and reporting orders of different sizes may differ systematically across markets, it is difficult to interpret differences between the two studies.
18
cents cheaper than the NYSE (although relative effective spreads are 4 basis points higher). For the
largest order-size category, up to 9,999 shares, Nasdaq effective spreads are 4.0 cents (9 bp) lower
than at the NYSE. The matched-pairs analyses in Panels A and B yield very similar results.
As in Table 4, realized spreads follow a pattern similar to that of effective spreads, so that
differences in information content are unlikely to explain the differences in execution costs. Indeed, I
find that price impacts are consistently greater on the NYSE (although they are not significantly
different from Nasdaq price impacts in 3 out of 12 cases). Therefore, differential information content
of order flow cannot explain execution cost differences in any of the order-size groups. However, I
find that the execution speed differential increases with order size. Small market orders are executed
18.0 seconds faster on Nasdaq, but large market orders take 29.3 seconds longer (Panel C). This
inverse relationship between cost and speed across order sizes complements the inverse relation
across markets documented in Table 4.
While Table 4 shows that Nasdaq market orders are more expensive in the aggregate, Table 5
reveals that this result is not uniform across order sizes. Contrary both to conventional wisdom and
the findings in SEC (2001), large orders execute more cheaply on Nasdaq. An important caveat is
that this comparison takes the order-size decision as exogenous; it assumes that a trader would
optimally submit the same order size to both markets. If order size were in fact endogenous, the
basic results would not change, but they would require a somewhat different interpretation. For
example, suppose an uninformed trader, who minimizes execution costs, decides to buy 40,000
shares. Suppose also that the trader optimally submits 5 orders of 8,000 shares each for a Nasdaq
stock, but 40 orders of 1,000 shares each for a NYSE stock. In this case, the relevant comparison
should be between large Nasdaq orders and small NYSE orders, and not between orders of identical
size. Endogeneity of order size, however, would affect the interpretation of the overall cost
differentials (Table 4) much less than the interpretation of size-specific results (Table 5).
3.3 Time trends in differences across markets
U.S. equity markets experienced several structural changes during and immediately
preceding my sampling period. Events include the final phase of decimalization (April 2001); the
19
implementation of Rule 11Ac1-5 between June and October 2001; the public display of limit orders
on the NYSE (January 2002); and Nasdaq’s transition to the Supermontage order display and
execution system between October and December 2002. While I do not attempt to relate these events
causally to changes in execution quality differentials, it is conceivable they have collectively
affected execution quality directly or indirectly on both the NYSE and on Nasdaq (for example,
through changes in order-routing practices).
Panel A of Table 6 shows monthly medians of the execution quality measures for market
orders in the 249 matched firms. Effective spreads have declined dramatically between November
2001 and December 2002 on both markets, possibly reflecting the temporarily elevated execution
costs in the aftermath of the September 2001 market closure. Nasdaq spreads are 5.5 cents or 45%
lower by the end of the sample period, and NYSE spreads are 3 cents (38%) lower compared to the
beginning of the sample period. Realized spreads and price impacts show a comparable development
over time, and execution speeds appear to accelerate slightly. Given these developments, we would
like to know whether the two markets have moved together or farther apart in terms of execution
quality. To address these issues, I analyze monthly cost differentials and test for potential trends over
time.
Because the matching procedure is based on the third quarter of 2001, matching errors are
likely to increase over time. Therefore, I analyze changes over time based on unrestricted versions of
the two regression procedures. Specifically, I estimate equations (2) and (3) separately for each
month, and replace the monthly fixed effects by a constant intercept.
To test for a linear trend, I estimate 249 regressions, one for each matched pair i:
14,...,1,249,...,1, ==++=∆ tiTEQM itiiit εβα (4)
where ∆EQMi represents the 14-month time series of execution quality differentials for matched pair
i, and T is a time-trend variable with values ranging from 1 through 14, corresponding to each month
from November 2001 through December 2002. Then I compute the mean and median of βi and test
whether it is equal to zero. A rejection would indicate the presence of a linear trend, and imply that
the execution quality differentials changed monotonically between November 2001 and December
20
2002. For an alternative specification, I also use a pooled panel model corresponding to (4) that
estimates one trend coefficient using all paired observations.
Panels B and C of Table 6 present the estimated coefficients for market order execution
quality measures, aggregated across order sizes. Monthly differentials are similar in size for the
matched-pairs regressions (Panel B) and the regressions using the entire sample (Panel C). It is
important that the differences in each of the execution quality measures documented in Table 4
persist in every single month. Specifically, considering the significant monthly differences, all
spread measures are consistently higher on Nasdaq; price impacts are lower (with one exception);
and execution is faster.
The trend tests in Panel B also present some evidence that the differences between the two
markets narrow over the period. While realized spreads and price impacts exhibit no significant
linear trend, the differences in effective spreads and execution speed appear to decline (although the
coefficients are never significant for all three methods). For the matched sample, Nasdaq orders
executed in November 2001 are 6.4 cents (27 bp) more expensive than NYSE orders. By December
2002, the differential has narrowed to 3.7 cents (16 bp). The decline is comparable for the broad
sample in Panel C, where the spread differential declines from 6.4 cents (36 bp) to 3.5 cents (15 bp).
For example, the median trend coefficient of –0.066/100 implies that for every other firm the spread
differential decreased by more than 0.066 cents (0.21 bp) in every month. While these changes are
economically small, they nevertheless appear to be systematic across securities.
Figure 1 shows how execution quality differentials evolve over time in the four order-size
categories. First, the graphs show that the relation between execution quality differentials and order
size I have documented is remarkably stable over time. Second, the downward trend in effective-
spread differentials appears to be very similar for each category. While Table 6 shows no strong
trend in the overall monthly speed differential, Figure 1 suggests that Nasdaq execution speed
improves compared to NYSE execution speed for large market orders. Most importantly, the figure
21
shows that the negative relation between execution costs and speed and its dependence on order size
exists in every month of the sample period.21
4 Trade-off between execution cost and speed
The analysis of execution quality reveals substantial differences in execution quality between
Nasdaq and the NYSE. An intriguing observation is that out-of-pocket execution costs and execution
speed are inversely related across markets (see Table 4, Table 5, and Figure 1). This result is
consistent with other analysis using specific pre-decimals samples to examine different dimensions
of execution quality. Battalio, Hatch, and Jennings (2003) document a similar relationship for a
sample of retail orders that execute faster on the NYSE, but receive better prices at Trimark
Securities. SEC (2001) finds lower costs and slower executions on the NYSE for a matched sample
of NYSE and Nasdaq stocks (but the authors do not discuss this relationship). Both higher execution
costs and slower execution speed are disadvantageous to traders, so these results suggest a potential
trade-off between costs and speed. Table 5 further shows that the trade-off appears to be related to
order size; smaller NYSE orders execute cheaper but slower than similar-sized Nasdaq orders, and
larger NYSE orders are more expensive, but execute faster. Figure 1 shows these relationships
persist over the period analyzed.
These findings raise the fundamentally important question of how to rank markets in terms of
execution quality. They suggest caution in ranking based on execution costs alone, because a market
with low effective spreads may impose additional costs in the form of slow executions.
Unfortunately, it is not possible to incorporate both dimensions of execution quality objectively
without an explicit model of the trade-off between costs and speed. In addition, because order size
21 The spike in Figure 1 that characterizes realized spreads in month 9 (July 2002) is not due to obvious data
errors or ambiguous outliers. Panel A of Table 6 suggests that this month was also different regarding other measures; effective spreads, price impacts, and execution speeds exhibit a local peak, and NYSE realized spreads are the lowest over the period analyzed. During this month, the S&P 500 Index declined by 9.4%, and volume in both markets was about 50% higher than during the other months of 2002. While such market wide extremes may be related to liquidity, it is not clear why the two markets were affected so differently.
22
determines whether the NYSE or Nasdaq is better along the cost dimension, differences in market
design alone cannot easily explain the potential trade-off.22
In this section, I provide an economic and institutional rationale for the apparent trade-off
between costs and speed. I first discuss a simple conceptual framework that incorporates the cost-
speed trade-off for Dash 5-eligible orders, and then provide additional empirical observations that
are consistent with this framework.
4.1 Conceptual framework for the cost-speed trade-off
My argument to explain the apparent trade-off between execution costs and speed, and why it
changes with order size, relies on differences between the NYSE and Nasdaq in how market makers
handle orders and how traders choose between order types. I assume that traders benefit, ceteris
paribus, from both low-cost and fast executions. I do not specify a specific relation between the two,
so the trade-off may vary over time according to market conditions, trader preferences, or order
characteristics. I further assume some traders have private information, so that market makers widen
their quotes when they perceive an increased likelihood of receiving orders from a better-informed
trader. This assumption is consistent with sequential-trade models as in Glosten and Milgrom
(1985).
Because the arguments rely on differences in the treatment of small and large orders, I first
provide more information on actual order sizes during the sample period. While Dash 5 reports cover
only orders of less than 10,000 shares, these orders by far exceed average quoted depth and average
trade size on both markets. To illustrate this point, I use all trades and quotes during regular market
hours (excluding trades with irregular settlement and quotes that are not eligible for the national best
bid and offer) from TAQ to compute the time-weighted quoted depth and equally weighted average
22 Demsetz (1968) and Stoll (1978) have argued that traders who prefer fast executions will have to pay an
immediacy premium. Their models, however, treat trader choice between costs and speed as exogenous and do not attempt to model this decision. Several other authors have conceptually recognized the trade-off between costs and speed. For example, Macey and O’Hara (1997) discuss legal aspects associated with best execution, and Burdett and O’Hara (1987) analyze the trade-off in the context of block-trade decisions. The theoretical literature on limit orders has analyzed both execution price and the probability of execution in static environments, but the static risk associated with unfilled orders does not easily translate into implications for order duration in a dynamic market.
23
trade size for each stock. Using the 249 matched pairs, Panel A of Table 7 shows that the medians of
both trade size and depth are within the second-smallest Dash 5 order-size category (500-1,999
shares). The median depth of 574 shares for Nasdaq stocks compares to 1,076 shares for the matched
NYSE stocks. Median trade size is 494 shares on Nasdaq and 695 shares on the NYSE.
Given these observations, I refer to the two largest Dash 5 order-size categories (2,000-9,999
shares) as large orders, and to the 100 to 1,999 share category as small orders. One distinctive
difference is that small orders will generally execute fully at the quoted price, because they tend to
be smaller than or close to the quoted depth. Most large orders, however, can be filled only partially
against published quotes. Depending on trading protocol and market maker preferences, the portion
exceeding the quoted depth may not execute at all or may execute at prices different from the quote.
4.1.1 Differences in order handling
The first part of the argument involves the effect of different order handling protocols on
execution speed. Incoming market orders can be executed automatically or filled manually by a
market maker. Automatic execution is typically fast because it requires no human interaction, but
incoming orders generally execute at the prevailing quotes, and therefore receive no price
improvement.23 When an order exceeds quoted depth, it receives a partial execution at the quote,
and, depending on the trading protocol and order instructions, potentially further executions at or
outside the quote. Most dealers will handle the excess portion manually, but automated systems such
as ECNs or Supermontage would match it against limit orders or market maker quotes outside the
best quotes.
If automatic execution is not available, the market maker or specialist has some discretion in
how to handle the order. He can provide fast execution by emulating automatic systems in that the
order is executed against published quotes and limit orders. Alternatively, a market maker may
23 There are some exceptions to this rule. For example, B.L. Madoff Investments Securities will offer automatic
price improvement for eligible orders, although Nasdaq market officials claim that this system attracts mostly exchange-listed order flow and only few Nasdaq orders. Eligibility is determined on a client-by-client basis, subject to several constraints, so that the combination of price improvement and automatic execution is available only for some orders. See http://www.madoff.com.
24
provide price improvement in two different ways. He can attempt to match the order with opposite-
side orders that arrive contemporaneously or are solicited from other (floor) brokers, or take the
opposite side himself and fill the incoming order from inventory. Both alternatives take time,
because the market maker has to scan incoming orders or negotiate with brokers. Similarly, to trade
himself, the market maker must comply with rules that give public orders at the same price priority
over dealer trades; this requires him to survey pending public orders. Thus, providing price
improvement arguably takes longer for any given order size (Blume (2001) makes a similar
argument). Therefore, the realized cost-speed combination depends on how market makers choose to
fill an order, and I argue that this choice differs between Nasdaq and the NYSE.
Nasdaq’s market structure does not rely primarily on a centralized execution system.24
Rather, broker-dealers and ECNs operate competing systems that are accessible to other NASD
members through the Nasdaq Level II quotation terminals. Most large broker-dealers use automatic
execution systems for small market orders, but execute larger orders manually.25 ECNs, which
provide automatic executions as well, also receive a greater proportion of small orders. While
several brokers provide rebates for certain order flow from selected customers, most widely
accessible automatic execution systems do not allow for price improvement per se. Thus, one would
expect small Nasdaq orders to execute fast and close to the prevailing quotes. But because brokers
handle large Nasdaq orders manually, and because these orders likely exceed quoted size, a broker
must search the connected pools of liquidity that constitute Nasdaq to find a counterparty willing to
take the opposite side of large orders. For any given order size, this process takes longer than an
automatic execution, but may yield a better price than was displayed (at and beyond the inside
quote) at the time the order arrived.
24 Nasdaq has traditionally operated centralized systems with execution capabilities (such as SOES or
Supermontage for Nasdaq stocks, and CAES for listed stocks), but these systems do not execute orders themselves. Rather, they display the trading interests of various dealers and ECNs, and users are able to execute automatically against one of the displayed alternatives.
25 It is difficult to provide a precise characterization of the set of orders that are eligible for automatic execution. Most broker-dealers provide different criteria depending on the size of the firm, current market conditions, and client identity. Most execute orders in liquid stocks up to a multiple of quoted depth, and orders in less liquid stocks up to the quoted depth. See, for example, the description provided by Knight Trading, one of the largest Nasdaq broker-dealers, at http://www.knight-sec.com/How_the_Trade_Gets_Done/Our_Order_Handling_Protocols.
25
On the NYSE, a specialist is responsible for executing most orders. Exceptions are odd lots
(orders smaller than 100 shares or residuals below 100 shares from larger orders) and Direct+ orders,
both of which execute automatically against prevailing quotes. The automatic execution system
Direct+ accepts only limit orders, however, and is therefore not included in my empirical analysis
(Rule 11Ac1-5 does not require a separate report for Direct+). Similarly, orders smaller than 100
shares are not included in Dash 5 reports (although odd-lot portions of larger orders are). As a result,
specialists execute virtually all NYSE orders in my analysis manually, unlike Nasdaq, where
typically only large orders are handled manually.
The difference between the two markets is consistent with two findings in Table 5: the faster
execution of small Nasdaq orders (because of automatic execution systems), and a greater likelihood
of price improvement for small NYSE orders than for small Nasdaq orders (because of manual order
handling). What remains to be explained is why large NYSE orders execute faster and at higher cost
than large Nasdaq orders.
4.1.2 Differences in order choice
The second part of my argument recognizes the different ways specialists and Nasdaq dealers
can presumably identify informed traders. It is reasonable to assume that a market maker’s
perception of an order’s information content affects how he fills the order, and I argue that this
decision is made differently on Nasdaq and the NYSE. Because different execution types are
associated with different cost-speed combinations, this decision will affect the ultimate cost and
speed of the execution.
On the NYSE, traders who have either no private information or whose information is
sufficiently long-lived often use floor brokers to work large orders. This involves delegating control
over the actual trading decisions to a floor broker, who then seeks favorable (partial) executions until
the order is filled. An informed trader with short-lived information cannot afford to use this option,
because it is slow, and the trader risks others discovering the same information before the orders are
filled. Instead, one would expect informed traders to submit orders directly to the specialist. This
self-selection would help the specialist distinguish informed and uninformed orders; he knows that
26
intermediated orders are unlikely to be informed, and therefore can afford to provide better prices to
these orders. Large market orders sent electronically to the specialist post are more likely to be
information-based. The specialist will thus generally not provide price or depth improvement to
these orders but execute them at quoted prices (and possibly the book, if the order is larger than the
quote). Therefore, large NYSE market orders will tend to receive relatively fast executions, but no
favorable price because of their perceived information content.26
The situation is different on Nasdaq, because there is no standardized procedure to submit a
large patient order similar to using a NYSE floor broker. Traders can obtain fast and anonymous
executions by walking up the book displayed on ECNs or systems such as Supermontage, but this
strategy will not allow any price improvement. Alternatively, they can route orders directly to
broker-dealers, who receive the majority of market orders included in this study.27 This routing
choice is not anonymous, however, because the trader must reveal his identity to the dealer.
Therefore, the best way for an informed Nasdaq trader to protect his information is to route orders to
an anonymous execution system, and not directly to a broker-dealer.28 This suggests an important
difference between Nasdaq market makers and NYSE specialists: A large electronic market order is
more likely to be viewed as information-based on the NYSE than on Nasdaq.
26 This adverse selection process does not apply to small NYSE orders, because small orders are not typically
worked by floor brokers. To put it differently, the specialist is not able to infer information content from small orders, because patient traders cannot self-select into a more patient alternative. Moreover, it has been argued that informed traders are unlikely to use small order sizes (see Barclay and Warner, 1993).
27 For my 249 matched Nasdaq stocks and the entire sample period, 32% of Dash 5-reported market orders are routed to ECNs. Almost all of the ECN market orders (30% all reported market orders) go to Redibook and Archipelago. These markets are built on business models that rely on finding good executions on other markets, rather than providing liquidity themselves. Specifically, Redibook executes 86% of its market orders elsewhere, and Archipelago 72%. Therefore, virtually all Nasdaq executions in my sample involve non-ECN broker-dealers.
28 See Barclay, Hendershott, and McCormick (2003). Also, an indirect way to check this claim is to examine the characteristics of (anonymous) Supermontage executions. Supermontage began publishing Dash 5 reports in December 2002. I examine eight monthly reports published since then, using market orders in all reported Nasdaq stocks (762 million executed market order shares). The average effective spread (4 cents) exceeds the average quoted spread (2 cents) by a factor of two. The average execution speed is 0.1 second, which is substantially faster than any other relevant market center. These observations suggest that traders expect speed, but not better-than-quoted prices on Supermontage, and are consistent with informed users who value anonymity and speed more than price.
27
4.1.3 Summary
Taken together, these arguments support a conceptual framework that is consistent with a
systematic negative relationship between execution costs and speed, and rationalize its dependence
on order size. Large electronic market orders execute at higher costs on the NYSE, because
specialists face more risk of trading with an informed party than Nasdaq market makers. Yet, large
NYSE executions are faster because the specialist, given his perception of an informed order,
executes against published quotes and potentially the book, but does not spend time to improve the
price. Nasdaq market makers, however, only receive non-anonymous order that are less likely to be
informed. They thus attempt to shop large orders by searching at different pools of liquidity, which
yields comparatively better prices and slower execution. Small orders execute faster on Nasdaq
because of the prevalence of automatic execution systems. At the same time, automatic execution
provides no price improvement that the exchange specialist is able to provide because every market
order is intermediated. Importantly, by splitting their trading interest accordingly, traders on both
markets can determine the likely balance between out-of-pocket cost and execution speed
themselves.
4.2 Additional empirical observations
The conceptual framework involves several critical assumptions about Dash 5-eligible
market orders: that small Nasdaq orders receive automatic executions but large orders do not; that
small NYSE orders are more likely than small Nasdaq orders to receive price improvement; and that
large NYSE orders are less likely than large Nasdaq orders to receive price improvement. To judge
how reasonable these assumptions are, and to further illuminate the relationship between the cost-
speed trade-off and order size, I provide additional evidence from Dash 5 reports.
I compute levels of execution speed, net price improvement, quoted and effective spreads,
and fill rates to compare relative changes across order-size categories. One way to obtain a partial
picture of how a market maker chooses between different execution types is to examine net price
improvement (NPI), defined as the total dollar amount of price improvement, net of the total amount
of price disimprovement, both measured against the relevant side of the NBBO when the order
28
arrived. I normalize this measure by the number of shares executed. While NPI alone should not be
viewed as a measure of execution quality, it may help to gauge market maker effort relative to the
price information available at the time an order is received. Quoted spreads are not published in
Dash 5 reports, so I compute the sum of effective spreads and twice net price improvement (all in
dollars) to produce the round-trip share-weighted average of the national best bid and offer (NBBO)
spread at the time orders were received. Finally, fill rates are defined as the percentage of shares
executed and can be viewed as another measure of market maker effort.
Panel B of Table 7 presents the median levels of these variables for Nasdaq and the NYSE,
using the sample of 249 matched pairs, and p-values of a Wilcoxon test that the medians are equal
across markets. The first finding is that the level of NYSE execution speed increases gradually with
order size, from 22 seconds for the smallest category to 28 for the largest. Nasdaq speed increases
from 2 and 8 seconds for the two smaller categories to 23 and 35 seconds for the two larger ones.
While this jump could be due to a variety of reasons, it is consistent with automatic (manual)
executions of small (large) orders on Nasdaq.
Second, there is evidence that increasing order size elevates the perception of informed
trading on the NYSE, but not on Nasdaq. Two observations contribute to this conclusion: NPI is
greater on the NYSE than on Nasdaq for small orders, but lower for large orders; and effective
spreads, which naturally increase with order size, actually decline for the largest orders on Nasdaq.
Using the matched-sample regression from equation (2), Panel C shows that the changes in NPI
persist after controlling for security-specific characteristics. Results are qualitatively identical for the
broad sample and with percentage instead of dollar spreads (not reported).
The finding that the NYSE provides more price improvement for small orders and less for
large orders than Nasdaq is consistent with the conceptual framework above and the results in Table
5 and Figure 1. Unfortunately, if inter-market differentials in quoted spreads were constant across
order sizes, the declining effective spread differential would mechanically imply less NPI on the
NYSE. In this case, the differences in NPI would not be sufficient to suggest that different market-
maker behavior contributes to the change in effective spreads. Table 7, however, shows that the
quoted spread differential does change with order size. This suggests two distinct sources of the
29
wider effective spreads on the NYSE for large orders: different quoted prices, and in addition less
NPI (a poorer price relative to the quote) than on Nasdaq. Thus, NYSE specialists appear to react
differently to large market orders than Nasdaq broker-dealers.
Finally, the lower quoted depth on Nasdaq (Panel A) complicates the comparison across
order-size categories, because any given order size is more likely to exceed the Nasdaq quote than
the NYSE quote. Moreover, Panel B in Table 7 reveals that NYSE fill rates are relatively constant
across order sizes (around 99%). Nasdaq fill rates by contrast decline from 97% for the smallest
orders to 82% for the largest ones. The matched-pairs panel regression in Panel C shows that the
difference in fill rates is significant even after controlling for stock characteristics and time variation.
Declining fill rates may play a role in explaining the lower effective spreads for large Nasdaq
orders. Rule 11Ac1-5 specifies that each partial execution of a large order be reported in that size
category. As an extreme example, suppose that all 8,000 share orders always receive a 25% fill rate.
Then the reported execution quality for the largest order category would actually reflect 2,000 share
trades. To the extent that market makers can provide 2,000 share executions at lower cost, even for
originally larger orders, a declining fill rate would impose a downward bias on reported execution
cost. These conditions represent a possible alternative for part of the trade-off explanation, but the
explanation based on declining fill rates cannot easily be reconciled with the reduced execution
speed for large Nasdaq orders. If fill rates were the whole story, one would expect large Nasdaq
orders to execute faster, and not slower as shown in Panel B. Declining fill rates, therefore, are likely
only a partial explanation for the apparent trade-off between execution costs and execution speed.
5 Conclusions
I provide the first comprehensive analysis of market order execution quality in the post-
decimals environment, taking advantage of new order-based data made available through SEC Rule
11Ac1-5. The rule requires individual market centers to publish monthly standardized reports that
provide detailed statistics on various measures of execution quality for orders below 10,000 shares
on an individual-security level. The sample period, November 2001 through December 2002, covers
or follows several important changes relating to equity trading, including decimalization, the
30
implementation of Rule 11Ac1-5, the public display of all limit orders on the NYSE, and the
introduction of Nasdaq’s Supermontage quotation and execution system.
To assess differences in execution quality between Nasdaq and the NYSE, I use three
different methodologies: (1) a matched-sample analysis; (2) a regression analysis using 249 matched
pairs; and (3) a comprehensive regression analysis of 1,043 NYSE and 1,093 Nasdaq stocks. Both
regression models use a standard set of control variables to adjust for differences in ex-ante
execution quality. Throughout the analysis, the three approaches produce virtually identical results.
Overall, market order executions are significantly more expensive on Nasdaq in terms of
effective spreads, whether measured in dollars or relative to share price. This differential cannot be
explained by more informed order flow, because realized spreads are also significantly higher for
Nasdaq orders, and the information content of order flow is smaller. Orders also execute
significantly faster on Nasdaq than on the NYSE. This basic result masks important differences
across order sizes, because the cost and speed differentials reverse for larger order sizes.
Specifically, executions of orders exceeding 2,000 shares are cheaper on Nasdaq, but also slower
than on the NYSE. In contrast, executions of smaller orders are cheaper, but slower on the NYSE.
Over time, I document a weak, but significant downward trend in the effective spread
differential between Nasdaq and the NYSE. This may imply that Nasdaq has improved its trading
system in a way that reduces its cost disadvantage relative to the NYSE, but it may also be caused by
systematic variation in market wide liquidity. During the entire period, however, Nasdaq execution
costs remain significantly above NYSE execution costs, and execution speed remains faster than on
the NYSE.
The results overall suggest a trade-off between execution costs and execution speed. Costs
appear to be negatively related to speed in a systematic fashion that persists over time. This negative
relation affects the way researchers, regulators, and market professionals can measure and interpret
execution quality. While this trade-off is conceptually well understood, its mechanics remain unclear
on both a theoretical and a practical level.
31
Presumably, the trade-off depends on trader preferences, order characteristics, and market
conditions. I rationalize the observed negative relation between costs and speed based on the
different order handling procedures on Nasdaq and the NYSE and different ways market makers can
detect informed traders. The automatic execution systems on Nasdaq cause smaller market orders to
execute faster and at quoted prices. NYSE specialists operate in an auction market and can provide
price improvement, but their manual intermediation slows execution speed. For larger market orders,
NYSE specialists expect greater information content than Nasdaq market makers. The reason is that
informed Nasdaq traders prefer anonymous systems, such as ECNs or Supermontage, over non-
anonymous orders to broker-dealers. Informed NYSE traders prefer to route orders directly to the
specialist rather than to floor brokers. As a result, the specialist executes large market orders at
quoted prices, which is fast, while Nasdaq market makers spend time to provide lower-cost
executions.
In the absence of publicly available data on the speed of order execution, researchers have
traditionally suggested that lower out-of-pocket costs imply a higher-quality execution. Given the
negative cost-speed relation I have documented, and because slow execution is costly for many
traders, this inference may need to be qualified. In a highly competitive environment, one would
expect execution quality not to differ significantly across markets. When execution quality has
several dimensions in addition to out-of-pocket costs, however, a competitive equilibrium may well
mean that one market will have higher costs along one dimension, such as effective spreads, but
lower costs along another, such as speed.
My evidence is consistent with this view, and illustrates the importance of further research to
explain trader preferences and competition between markets. Better understanding the trade-offs
between execution costs and speed would allow a more precise measurement of execution quality,
and provide valuable guidance for appropriate regulation and theoretical models of market design
and trader behavior.
32
6 References
Bacidore, J., R. Battalio, and R. Jennings, 2003. Order submission strategies, liquidity supply, and
trading in pennies on the New York Stock Exchange. Journal of Financial Markets 6, 337-
362.
Barclay, M.J., 1997. Bid-ask spreads and the avoidance of odd-eighth quotes on Nasdaq: An
examination of exchange listings. Journal of Financial Economics 45, 35-60.
Barclay, M.J., T. Hendershott, and D.T. McCormick, 2003. Competition among trading venues:
Information and trading on electronic communications networks. Journal of Finance,
forthcoming.
Barclay, M.J., and J.B. Warner, 1993. Stealth and volatility: Which trades move prices? Journal of
Financial Economics 34, 281-306.
Battalio, R., B. Hatch, and R. Jennings, 2003. All Else Equal?: A Multi-dimensional Analysis of
Retail Market Order Execution Quality. Journal of Financial Markets, 143-162.
Bessembinder, H., 1999. Trade execution costs on Nasdaq and the NYSE: A post-reform
comparison. Journal of Financial and Quantitative Analysis 34, 387-407.
Bessembinder, H., 2002. Issues in assessing trade execution costs. Journal of Financial Markets 6,
233-257.
Bessembinder, H., 2003a. Selection biases and cross-market trading cost comparisons. Working
paper, University of Utah.
Bessembinder, H., 2003b. Trade execution costs and market quality after decimalization. Journal of
Financial and Quantitative Analysis, forthcoming.
Bessembinder, H., and H. Kaufman, 1997a. A comparison of trade execution costs for NYSE and
Nasdaq-listed stocks. Journal of Financial and Quantitative Analysis 32, 287-310.
Bessembinder, H., and H. Kaufman, 1997b. A cross-exchange comparison of execution costs and
information flow for NYSE-listed stocks. Journal of Financial Economics 46, 293-319.
33
Blume, M., 2001. The structure of the US equity markets. Working paper, University of
Pennsylvania.
Boehmer, E., R. Jennings, and L. Wei, 2003. Public disclosure and private decisions: The case of
equity market execution quality. Working paper, Indiana University.
Burdett, K., and M. O’Hara, 1987. Building blocks: An introduction to block trading. Journal of
Banking and Finance 11, 193-212.
Chan J., and J. Lakonishok, 1995. The behavior of stock prices around institutional trades. Journal of
Finance 50, 1147-1174.
Christie, W.G., and R. Huang, 1994. Market structures and liquidity: Transactions data study of
exchange listings. Journal of Financial Intermediation 3, 300-326.
Demsetz, H., 1968. The cost of transacting. Quarterly Journal of Economics 82, 33-53.
Dyl, E., and A. Anderson, 2002, Market structure and trading volume. Working paper, University of
Arizona.
Easley, D., N.M. Kiefer, and M. O’Hara, 1996. Cream-skimming or profit-sharing? The curious role
of purchased order flow. Journal of Finance 51, 811-834.
Easley, D., N.M. Kiefer, M. O’Hara, and J.B. Paperman, 1996. Liquidity, information, and
infrequently traded stocks. Journal of Finance 51, 1405-1436.
Glosten, L., and P. Milgrom, 1985. Bid, ask, and transaction prices in a specialist market with
heterogeneously informed traders. Journal of Financial Economics 13, 71-100.
Harris, L., and J. Hasbrouck, 1996. Market vs. limit orders: The SuperDOT evidence of order
submission strategy. Journal of Financial and Quantitative Analysis 31, 213-231.
Heidle, H.G., and R.D. Huang, 2002. Information-based trading in dealer and auction markets: An
analysis of exchange listings. Journal of Financial and Quantitative Analysis 37, 391-424.
Huang, R.D., and H. Stoll, 1996. Dealer versus auction markets: A paired comparison of execution
costs on Nasdaq and the NYSE. Journal of Financial Economics 41, 313-357.
34
Keim, D.B., and A.N. Madhavan, 1997. Transaction costs and investment style: An inter-exchange
analysis of institutional equity trades. Journal of Financial Economics 46, 371-398.
Lee, C., 1993. Market integration and price execution for NYSE-listed securities. Journal of Finance
48, 1009-1038.
Lipson, M., 2003. Competition among market centers. Working paper, University of Georgia.
Macey, J., and M. O’Hara, 1997. The law and economics of best execution. Journal of Financial
Intermediation 6, 188-223.
Peterson, M., and E. Sirri, 2002. Order submission strategy and the curious case of marketable limit
orders. Journal of Financial and Quantitative Analysis 37, 221-241.
SEC, 2001. Report on the comparison of order execution across equity market structures. U.S.
Securities and Exchange Commission, Washington.
Stoll, H.R., 1978. The supply of dealer services in securities markets. Journal of Finance 33, 1133-
1151.
Stoll, H.R., 2000. Friction. Journal of Finance 55, 1479-1514.
Weston, J., 2000. Competition on the Nasdaq and the impact of market reforms. Journal of Finance
55, 2565-2598.
Matching error
Ticker Name Top20 Size Ticker Name
AMGN AMGEN INC 1 1 MWD MORGAN STANLEY DEAN WITTER & CO 112%AMAT APPLIED MATERIALS INC 1 1 TXN TEXAS INSTRUMENTS INC 147%AMCC APPLIED MICRO CIRCUITS CORP 1 1 AMD ADVANCED MICRO DEVICES INC 125%BEAS B E A SYSTEMS INC 1 1 EMC E M C CORP MA 189%BRCD BROCADE COMMUNICATIONS SYS INC 1 1 MU MICRON TECHNOLOGY INC 218%CHIR CHIRON CORP 1 1 KG KING PHARMACEUTICALS INC 55%CSCO CISCO SYSTEMS INC 1 1 AOL A O L TIME WARNER INC 203%COST COSTCO WHOLESALE CORP 1 1 KSS KOHLS CORP 48%EBAY EBAY INC 1 1 BBY BEST BUY COMPANY INC 148%FITB FIFTH THIRD BANCORP 1 1 KMB KIMBERLY CLARK CORP 44%GENZ GENZYME CORP 1 1 APC ANADARKO PETROLEUM CORP 69%INTC INTEL CORP 1 1 GE GENERAL ELECTRIC CO 201%JDSU J D S UNIPHASE CORP 1 1 LU LUCENT TECHNOLOGIES INC 153%KLAC K L A TENCOR CORP 1 1 ADI ANALOG DEVICES INC 177%LLTC LINEAR TECHNOLOGY CORP 1 1 LOW LOWES COMPANIES INC 147%MXIM MAXIM INTEGRATED PRODUCTS INC 1 1 GDT GUIDANT CORP 157%NTRS NORTHERN TRUST CORP 1 1 WY WEYERHAEUSER CO 30%NVLS NOVELLUS SYSTEMS INC 1 1 PVN PROVIDIAN FINANCIAL CORP 211%NVDA NVIDIA CORP 1 1 PCS SPRINT CORP 211%ORCL ORACLE CORP 1 1 HD HOME DEPOT INC 238%PAYX PAYCHEX INC 1 1 HDI HARLEY DAVIDSON INC 47%PSFT PEOPLESOFT INC 1 1 AES A E S CORP 281%QLGC QLOGIC CORP 1 1 SFA SCIENTIFIC ATLANTA INC 322%QCOM QUALCOMM INC 1 1 PFE PFIZER INC 246%RFMD R F MICRO DEVICES INC 1 1 WFT WEATHERFORD INTL INC NEW 269%SANM SANMINA HOLDINGS INC 1 1 LSI L S I LOGIC CORP 132%SOTR SOUTHTRUST CORP 1 1 MAY MAY DEPARTMENT STORES CO 26%SUNW SUN MICROSYSTEMS INC 1 1 BA BOEING CO 350%VRSN VERISIGN INC 1 1 JPM J P MORGAN CHASE & CO 297%VRTS VERITAS SOFTWARE CORP 1 1 T A T & T CORP 365%BRCM BROADCOM CORP 0 1 Q QWEST COMMUNICATIONS INTL INC 231%DELL DELL COMPUTER CORP 0 1 MER MERRILL LYNCH & CO INC 218%MSFT MICROSOFT CORP 0 1 C CITIGROUP INC 323%PMCS P M C SIERRA INC 0 1 MOT MOTOROLA INC 259%SEBL SIEBEL SYSTEMS INC 0 1 CD CENDANT CORP 352%COMS 3COM CORP 0 2 VRC VARCO INTERNATIONAL INC DEL 105%ALTR ALTERA CORP 0 2 CPN CALPINE CORP 86%AEOS AMERICAN EAGLE OUTFITTERS INC NE 0 2 TER TERADYNE INC 76%ANAT AMERICAN NATIONAL INS CO 0 2 CBH COMMERCE BANCORP INC NJ 134%ANDW ANDREW CORP 0 2 CRA APPLERA CORP 55%APPB APPLEBEES INTERNATIONAL INC 0 2 YRK YORK INTL CORP NEW 33%ADSK AUTODESK INC 0 2 NBL NOBLE AFFILIATES INC 37%CELG CELGENE CORP 0 2 IRF INTERNATIONAL RECTIFIER CORP 93%CEPH CEPHALON INC 0 2 LEN LENNAR CORP 106%CMVT COMVERSE TECHNOLOGY INC 0 2 HAL HALLIBURTON COMPANY 174%CYTC CYTYC CORP 0 2 DO DIAMOND OFFSHORE DRILLING INC 53%DLTR DOLLAR TREE STORES INC 0 2 CY CYPRESS SEMICONDUCTOR CORP 67%ERTS ELECTRONIC ARTS INC 0 2 COF CAPITAL ONE FINANCIAL CORP 84%
Appendix: List of matching Nasdaq-NYSE pairs
The Nasdaq sample consist of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization (MCAP), dollar volume, or share volume during 2001Q3. The Top20 indicator identifies Nasdaq firms that were added following this procedure. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: MCAP, share price, adjusted dollar volume, and the daily relative price range during 2001Q3 (see equation (1) in the main text). Size is an indicator of Nasdaq market capitalization, equal to 1 for Top20 firms, 2 if MCAP > $1 billion, 3 if $0.2 billion <= MCAP <= $1billion, and 4 otherwise.
Nasdaq NYSE
35
JKHY HENRY JACK & ASSOC INC 0 2 OCR OMNICARE INC 39%HGSI HUMAN GENOME SCIENCES INC 0 2 NSM NATIONAL SEMICONDUCTOR CORP 120%IFIN INVESTORS FINANCIAL SERVS CORP 0 2 JEC JACOBS ENGINEERING GROUP INC 49%LNCR LINCARE HOLDINGS INC 0 2 HCR MANOR CARE INC NEW 47%MVSN MACROVISION CORPORATION 0 2 CAM COOPER CAMERON CORP 104%MEDI MEDIMMUNE INC 0 2 BHI BAKER HUGHES INC 88%NATI NATIONAL INSTRUMENTS CORP 0 2 CNX CONSOL ENERGY INC 52%NTAP NETWORK APPLIANCE INC 0 2 BJS B J SERVICES CO 155%ORLY O REILLY AUTOMOTIVE INC 0 2 HP HELMERICH & PAYNE INC 57%PDCO PATTERSON DENTAL CO 0 2 TRI TRIAD HOSPITALS INC 64%PIXR PIXAR 0 2 NDE INDYMAC BANCORP INC 84%BPOP POPULAR INC 0 2 TSS TOTAL SYSTEM SERVICES INC 76%QSFT QUEST SOFTWARE INC 0 2 CCI CROWN CASTLE INTERNATIONAL CORP 182%QTRN QUINTILES TRANSNATIONAL CORP 0 2 ESV E N S C O INTERNATIONAL INC 65%RLRN RENAISSANCE LEARNING INC 0 2 NFX NEWFIELD EXPLORATION CO 82%RESP RESPIRONICS INC 0 2 HAR HARMAN INTERNATIONL INDS INC NEW 73%SYMC SYMANTEC CORP 0 2 SII SMITH INTERNATIONAL INC 68%TLAB TELLABS INC 0 2 JBL JABIL CIRCUIT INC 100%USAI U S A NETWORKS INC 0 2 JCP PENNEY J C INC 69%TTEN 3 T E C ENERGY CORP 0 3 NCS N C I BUILDING SYSTEMS INC 94%ACDO ACCREDO HEALTH INC 0 3 SKE SPINNAKER EXPLORATION CO 56%AFFX AFFYMETRIX INC 0 3 NOI NATIONAL OILWELL INC 80%PCSA AIRGATE P C S INC 0 3 RYL RYLAND GROUP INC A 56%ALSC ALLIANCE SEMICONDUCTOR CORP 0 3 CHB CHAMPION ENTERPRISES INC 26%ALOY ALLOY INC 0 3 LSS LONE STAR TECHNOLOGIES INC 25%AMSY AMERICAN MANAGEMENT SYSTEMS INC 0 3 DY DYCOM INDUSTRIES IN 52%AMWD AMERICAN WOODMARK CORP 0 3 TRR T R C COMPANIES INC 71%ABCW ANCHOR BANCORP WISCONSIN INC 0 3 JLL JONES LANG LASALLE INC 36%APOG APOGEE ENTERPRISES INC 0 3 TFS THREE FIVE SYSTEMS INC 105%AGII ARGONAUT GROUP INC 0 3 CDI C D I CORP 60%ASTE ASTEC INDUSTRIES INC 0 3 DRQ DRIL QUIP INC 50%APWR ASTROPOWER INC 0 3 PRX PHARMACEUTICAL RESOURCES INC 133%BEBE BEBE STORES INC 0 3 SAH SONIC AUTOMOTIVE INC 45%BBOX BLACK BOX CORP DEL 0 3 SGR SHAW GROUP INC 109%CSAR CARAUSTAR INDUSTRIES INC 0 3 GRB GERBER SCIENTIFIC INC 37%CEGE CELL GENESYS INC 0 3 ENZ ENZO BIOCHEM INC 40%CHDN CHURCHILL DOWNS INC 0 3 TRC TEJON RANCH CO 57%CTBK CITYBANK LYNNWOOD WASHINGTON 0 3 BDG BANDAG INC 81%COLM COLUMBIA SPORTSWEAR COMPANY 0 3 BBI BLOCKBUSTER INC 51%CRXA CORIXA CORP 0 3 CKP CHECKPOINT SYSTEMS INC 94%CSGP COSTAR GROUP INC 0 3 MKT ADVANCED MARKETING SERVICES INC 46%CRGN CURAGEN CORP 0 3 CVD COVANCE INC 90%CYBX CYBERONICS INC 0 3 SOL SOLA INTERNATIONAL INC 32%DLIA DELIAS CORP 0 3 RRC RANGE RESOURCES CORP 74%DIGL DIGITAL LIGHTWAVE INC 0 3 PDE PRIDE INTERNATIONAL INC DEL 199%DCTM DOCUMENTUM INC 0 3 TWR TOWER AUTOMOTIVE INC 94%DCLK DOUBLECLICK INC 0 3 ETS ENTERASYS NETWORK INC 103%DYII DYNACQ INTERNATIONAL INC 0 3 STW STANDARD COMMERCIAL CORP 69%EXAR EXAR CORP 0 3 PWR QUANTA SERVICES INC 96%FFIV F 5 NETWORKS INC 0 3 KMX CIRCUIT CITY STORES INC 125%FBAN F N B CORP PA 0 3 MTW MANITOWOC INC 33%FINL FINISH LINE INC 0 3 PVH PHILLIPS VAN HEUSEN CORP 54%FTFC FIRST FEDERAL CAPITAL CORP 0 3 PRA PROASSURANCE CORP 29%FFBC FIRST FINANCIAL BANCORP OHIO 0 3 RDK RUDDICK CORP 58%FPFC FIRST PLACE FINANCIAL CORP NM 0 3 PNN PENN ENGINEERING & MFG CORP 29%GLDB GOLD BANC CORP INC 0 3 GES GUESS INC 30%GTRC GUITAR CENTER INC 0 3 HDL HANDLEMAN CO 90%HBHC HANCOCK HOLDING CO 0 3 CW CURTISS WRIGHT CORP 79%HDWR HEADWATERS INC 0 3 SEI SEITEL INC 107%HOTT HOT TOPIC INC 0 3 CHS CHICOS FAS INC 37%
36
IDXX I D E X X LABORATORIES INC 0 3 WGR WESTERN GAS RESOURCES INC 32%IMGN IMMUNOGEN INC 0 3 UNT UNIT CORP 94%IMDC INAMED CORP 0 3 LNY LANDRYS RESTAURANTS INC 31%ISSX INTERNET SECURITY SYSTEMS INC 0 3 RDC ROWAN COMPANIES INC 218%ISLE ISLE OF CAPRI CASINOS INC 0 3 PWN CASH AMERICA INTERNATIONAL INC 55%JDAS J D A SOFTWARE GROUP INC 0 3 BHE BENCHMARK ELECTRONICS INC 64%KNSY KENSEY NASH CORP 0 3 TTI TETRA TECHNOLOGIES INC 74%KEYS KEYSTONE AUTOMOTIVE INDS INC 0 3 TWP TREX INC 32%NITE KNIGHT TRADING GROUP INC 0 3 OO OAKLEY INC 73%LGTO LEGATO SYSTEMS INC 0 3 KEG KEY ENERGY SERVICES INC 96%LTBG LIGHTBRIDGE INC 0 3 BGC GENERAL CABLE CORP DEL NEW 90%LFIN LOCAL FINANCIAL CORP 0 3 IDT I D T CORP 28%MSBK MAIN STREET BANKS INC 0 3 ESL ESTERLINE TECHNOLOGIES CORP 114%MANH MANHATTAN ASSOCIATES INC 0 3 CHP C & D TECHNOLOGIES INC 109%MANU MANUGISTICS GROUP INC 0 3 GTW GATEWAY INC 217%MCSI MCSI INC 0 3 MNS M S C SOFTWARE CORP 32%MMSI MERIT MEDICAL SYSTEMS INC 0 3 SFY SWIFT ENERGY CO 153%MOVI MOVIE GALLERY INC 0 3 ASF ADMINISTAFF INC 146%NBTB N B T BANCORP INC 0 3 TCC TRAMMELL CROW CO 63%NAUT NAUTICA ENTERPRISES INC 0 3 MWY MIDWAY GAMES INC 52%OLOG OFFSHORE LOGISTICS INC 0 3 KWD KELLWOOD COMPANY 14%OCAS OHIO CASUALTY CORP 0 3 EYE V I S X INC 24%OSUR ORASURE TECHNOLOGIES INC 0 3 VTA VESTA INSURANCE GROUP INC 81%PRXL PAREXEL INTERNATIONAL CORP 0 3 WLM WELLMAN INC 71%PRKR PARKERVISION INC 0 3 NEV NUEVO ENERGY CO 112%PEGS PEGASUS SOLUTIONS INC 0 3 ITN INTERTAN INC 64%PIOS PIONEER STANDARD ELECTRONICS INC 0 3 TGX THERAGENICS CORP 30%PLXS PLEXUS CORP 0 3 PCP PRECISION CASTPARTS CORP 88%POWI POWER INTEGRATIONS INC 0 3 WMS W M S INDUSTRIES INC 118%RADS RADIANT SYSTEMS INC 0 3 ALN ALLEN TELECOM INC 59%RARE RARE HOSPITALITY INTL INC 0 3 PPD PRE PAID LEGAL SERVICES INC 37%RBNC REPUBLIC BANCORP 0 3 UCI UICI 52%SBAC S B A COMMUNICATIONS CORP 0 3 AXL AMERICAN AXLE & MFG HLGDS INC 63%POOL S C P POOL CORP 0 3 SPF STANDARD PACIFIC CORP NEW 30%SNDK SANDISK CORP 0 3 ANN ANNTAYLOR STORES CORP 162%SASR SANDY SPRING BANCORP INC 0 3 AWR AMERICAN STATES WATER CO 71%SCIO SCIOS INC 0 3 ACI ARCH COAL INC 74%SECD SECOND BANCORP INCORPORATED 0 3 CGX CONSOLIDATED GRAPHICS INC 145%SHFL SHUFFLE MASTER INC 0 3 ZQK QUIKSILVER INC 103%TSFG SOUTH FINL GROUP INC 0 3 OLN OLIN CORP 43%SLNK SPECTRALINK CORP 0 3 PBY PEP BOYS MANNY MOE & JACK 149%SRCL STERICYCLE INC 0 3 CRY CRYOLIFE INC 50%STEI STEWART ENTERPRISES INC 0 3 OI OWENS ILL INC 62%SYKE SYKES ENTERPRISES INC 0 3 OMM O M I CORP NEW 133%SCTC SYSTEMS & COMPUTER TECHNOLOGY 0 3 NR NEWPARK RESOURCES INC 148%TGIC TRIAD GUARANTY INC 0 3 CKH SEACOR HOLDINGS INC 75%TRMB TRIMBLE NAVIGATION LTD 0 3 CTS C T S CORP 93%TRMS TRIMERIS INC 0 3 EVG EVERGREEN RESOURCES INC 72%TRYF TROY FINANCIAL CORP 0 3 CV CENTRAL VERMONT PUB SVC CORP 54%TUES TUESDAY MORNING CORP 0 3 DNR DENBURY RESOURCES INC 47%UNBJ UNITED NATIONAL BANCORP NJ 0 3 THO THOR INDUSTRIES INC 42%UNFI UNITED NATURAL FOODS INC 0 3 GPI GROUP 1 AUTOMOTIVE INC 87%UEIC UNIVERSAL ELECTRONICS INC 0 3 MPH CHAMPIONSHIP AUTO RACING TM INC 107%USFC USFREIGHTWAYS CORP 0 3 SUP SUPERIOR INDUSTRIES INTL INC 34%VARI VARIAN INC 0 3 FDS FACTSET RESEARCH SYSTEMS INC 40%WFSI W F S FINANCIAL INC 0 3 TG TREDEGAR CORP 121%WDFC WD-40 CO 0 3 STC STEWART INFORMATION SVCS CORP 55%WCBO WEST COAST BANCORP ORE NEW 0 3 RNT AARON RENTS INC 62%ZOLL ZOLL MEDICAL CORP 0 3 ATW ATWOOD OCEANICS INC 89%ANSI ADVANCED NEUROMODULATION SYS INC 0 4 CGC CASCADE NATURAL GAS CORP 48%
37
ALCO ALICO INC 0 4 SKY SKYLINE CORP 69%ALLE ALLEGIANT BANCORP INC 0 4 RSC REX STORES CORP 50%ASGR AMERICA SERVICE GROUP INC 0 4 CSV CARRIAGE SERVICES INC 154%AINN APPLIED INNOVATION INC 0 4 KTO K 2 INC 74%ARTI ARTISAN COMPONENTS INC 0 4 HKF HANCOCK FABRICS INC 96%AVTR AVATAR HOLDINGS INC 0 4 NTK NORTEK INC 107%AVID AVID TECHNOLOGY INC 0 4 SIE SIERRA HEALTH SERVICES INC 57%OZRK BANK OF THE OZARKS INC 0 4 CRN CORNELL COMPANIES INC 148%BSET BASSETT FURNITURE INDUSTRIES INC 0 4 OSM OSMONICS INC 78%BELM BELL MICROPRODUCTS INC 0 4 NSS N S GROUP INC 44%BJCT BIOJECT MEDICAL TECHNOLOGIES INC 0 4 CPE CALLON PETROLEUM CO DEL 62%BCGI BOSTON COMMUNICATION GROUP INC 0 4 KEI KEITHLEY INSTRUMENTS INC 48%BUCA BUCA INC 0 4 MVK MAVERICK TUBE CORP 68%CCCG C C C INFORMATION SVCS GROUP INC 0 4 CAO C S K AUTO CORP 116%CLZR CANDELA CORP 0 4 ENC ENESCO GROUP INC 64%CLRS CLARUS CORP DEL 0 4 USG U S G CORP 161%CMTL COMTECH TELECOMMUNICATIONS CORP 0 4 TT TRANSTECHNOLOGY CORP 68%CVAS CORVAS INTERNATIONAL INC 0 4 FWC FOSTER WHEELER LTD 97%CCEL CRYO CELL INTERNATIONAL INC 0 4 CKR C K E RESTAURANTS INC 158%CYGN CYGNUS INC 0 4 MTZ MASTEC INC 85%DSSI DATA SYSTEMS & SOFTWARE INC 0 4 WZR WISER OIL CO 112%DIOD DIODES INC 0 4 LMS LAMSON & SESSIONS CO 67%EMBX EMBREX INC 0 4 HEI HEICO CORP NEW 61%ENMD ENTREMED INC 0 4 RTI R T I INTERNATIONAL METALS INC 71%FESX FIRST ESSEX BANCORP INC 0 4 MHO M I SCHOTTENSTEIN HOMES INC NEW 78%FMAR FIRST MARINER BANCORP 0 4 GI GIANT INDUSTRIES INC 166%FFBK FLORIDAFIRST BANCORP INC NEW 0 4 LAD LITHIA MOTORS INC 72%FLOW FLOW INTERNATIONAL CORP 0 4 SRI STONERIDGE INC 87%GMCR GREEN MOUNTAIN COFFEE INC 0 4 KDE 4 KIDS ENTERTAINMENT INC 108%GRKA GREKA ENERGY CORP 0 4 FOB BOYDS COLLECTION LTD 130%GSOF GROUP 1 SOFTWARE INC NEW 0 4 IMR I M C O RECYCLING INC 75%HEPH HOLLIS EDEN PHARMACEUTICALS INC 0 4 DFS DEPARTMENT 56 INC 166%ICTG I C T GROUP INC 0 4 LDL LYDALL INC 133%IIVI II VI INC 0 4 OFG ORIENTAL FINANCIAL GROUP INC 110%IMCO IMPCO TECHNOLOGIES INC 0 4 VTS VERITAS D G C INC 165%JJSF J & J SNACK FOODS CORP 0 4 SRT STARTEK INC 118%JACO JACO ELECTRONICS INC 0 4 FJC FEDDERS CORP 127%JOSB JOS A BANK CLOTHIERS INC 0 4 CBZ COBALT CORP 146%KVHI K V H INDUSTRIES INC 0 4 RWY RENT WAY INC 173%LJPC LA JOLLA PHARMACEUTICAL CO 0 4 HXL HEXCEL CORP NEW 133%LOJN LO JACK CORP 0 4 DAB DAVE & BUSTERS INC 75%LNET LODGENET ENTERTAINMENT CORP 0 4 MEH MIDWEST EXPRESS HOLDINGS INC 98%MIPS M I P S TECHNOLOGIES INC 0 4 SFP SALTON INC 104%SHOO MADDEN STEVEN LTD 0 4 FLE FLEETWOOD ENTERPRISES INC 130%MTSN MATTSON TECHNOLOGY INC 0 4 SKS SAKS INC 191%MESA MESA AIR GROUP INC NEV 0 4 PME PENTON MEDIA INC 198%MSSN MISSION RESOURCES CORP 0 4 MMR MCMORAN EXPLORATION CO 58%NUCO N U C O 2 INC 0 4 RES R P C INC 118%NARA NARA BANCORP INC 0 4 CPY C P I CORP 72%NEOG NEOGEN CORP 0 4 UAG UNITED AUTO GROUP INC 181%OGLE OGLEBAY NORTON CO 0 4 AZZ A Z Z INC 140%OSBC OLD SECOND BANCORP INC 0 4 OXM OXFORD INDUSTRIES INC 124%ONXX ONYX PHARMACEUTICALS INC 0 4 UNA UNOVA INC 181%PBIX PATRIOT BANK CORP NEW 0 4 HUF HUFFY CORP 109%PTIX PERFORMANCE TECHNOLOGIES INC 0 4 INT WORLD FUEL SERVICES CORP 137%PHAR PHARMANETICS INC 0 4 OS OREGON STEEL MILLS INC 195%QRSI Q R S CORP 0 4 APN APPLICA INC 124%RNBO RAINBOW TECHNOLOGIES INC 0 4 MWL MAIL WELL INC 98%RCOT RECOTON CORP 0 4 CDT CABLE DESIGN TECHNOLOGIES CORP 110%REFR RESEARCH FRONTIERS INC 0 4 CHH CHOICE HOTELS INTERNATIONAL INC 177%
38
RELL RICHARDSON ELECTRONICS LTD 0 4 CAE CASCADE CORP 82%RGLD ROYAL GOLD INC 0 4 AVL AVIALL INC NEW 140%SCAI SANCHEZ COMPUTER ASSOC INC 0 4 CRK COMSTOCK RESOURCES INC 109%SAFM SANDERSON FARMS INC 0 4 POP POPE & TALBOT INC 99%STCO SIGNAL TECHNOLOGY CORP 0 4 CKC COLLINS & AIKMAN CORP NEW 166%SPEX SPHERIX INC 0 4 LUB LUBYS INC 147%SSNC SS & C TECHNOLOGIES INC 0 4 ACO AMCOL INTERNATIONAL CORP 101%STLY STANLEY FURNITURE CO NEW 0 4 SWM SCHWEITZER MAUDUIT INTL INC 92%STTX STEEL TECHNOLOGIES INC 0 4 SHS SAUER DANFOSS INC 134%STSA STERLING FINANCIAL CORP WASH 0 4 MFI MICROFINANCIAL INC 38%SNBC SUN BANCORP INC 0 4 PCU SOUTHERN PERU COPPER CORP 92%SUPC SUPERIOR CONSULTANT HLDNG CORP 0 4 AOR AURORA FOODS INC 170%SYNM SYNTROLEUM CORP 0 4 HYC HYPERCOM CORP 194%WRLS TELULAR CORP 0 4 FMT FREMONT GENERAL CORP 198%TRFX TRAFFIX INC 0 4 BYD BOYD GAMING CORP 196%PANL UNIVERSAL DISPLAY CORP 0 4 WNC WABASH NATIONAL CORP 87%URBN URBAN OUTFITTERS INC 0 4 BWS BROWN SHOE INC NEW 77%VLNC VALENCE TECHNOLOGY INC 0 4 MPS MODIS PROFESSIONAL SERVICES INC 174%VXGN VAXGEN INC 0 4 GFF GRIFFON CORP 146%XICO XICOR INC 0 4 IKN IKON OFFICE SOLUTIONS INC 151%ZIGO ZYGO CORP 0 4 ABF AIRBORNE INC 210%
39
Criterion NYSE Nasdaq
General CRSP filters
All U.S. domestic securities on 9/30/2001 2579 3949Dual-class stock -215 -147Non-common-stock securities -967 -215No price on 9/30/2001 -6 -38No SIC code on 9/30/2001 -1 -1No link to Compustat on CCM 9/30-12/31/2001 -79 -259No daily return data 10/1/1999-12/31/2001 -65 -495
1246 2794
CRSP trading filters 7/1-9/30/2001
Switched trading venue -6 -5Mean daily trading volume < $20,000 -25 -598Missing price, any day -4 -7Missing volume, any day 0 0Change in share class or type 0 -2
1211 2182
TAQ trading filters, 7/1-9/30/2001
Lowest price < $3.00 -78 -601Average daily number of trades < 20 -64 -405
1069 1176
Rule 11Ac1-5 filters Nov 2001-Dec 2002
No continuous data for at least one category -26 -83
Final sample 1043 1093
Table 1: Sample selectionThe table describes the selection of the final sample from the universe of all securities included in the CRSP database. CCM refers to the CRSP-Compustat link file. The filters are not mutually exclusive, so their weight (how many securities they remove) depends on their ordering.
40
N Nasdaq median NYSE medianMedian pairwise
differenceWilcoxon p-value N
Median pairwise difference
Wilcoxon p-value
MCAP 249 $333,301,919 $390,976,000 -$36,784,148 0.00 86 -$11,216,638 0.10Share price 249 $14.55 $14.77 $0.34 0.13 86 $0.47 0.47Daily price range 249 6.0% 4.4% 1.3% 0.00 86 0.4% 0.00Daily volume 249 $2,037,009 $2,373,191 -$48,570 0.29 86 -$60,059 0.00
All matched pairs
The table presents pairwise differences in the matching variables of 249 Nasdaq-NYSE matched pairs. The Nasdaq sample consists of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization (MCAP), dollar volume, or share volume during 2001Q3. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: MCAP, share price, the daily relative price range during 2001Q3, and adjusted dollar volume during 2001Q3. The best matches include only pairs that have a matching error below 0.7.
Table 2: Descriptive statistics on matched pairs of Nasdaq and NYSE securities
Best matches only
41
Panel A: CT trading volume 100 shares < trade size < 10,000 shares All trades
CT volume NYSE sample 151,810,665,300 278,718,503,260CT volume Nasdaq sample 202,376,157,200 291,495,039,500Total volume both markets 354,186,822,500 570,213,542,760
NYSE market orders 73,269,557,281Nasdaq market orders 47,595,973,131
NYSE market orders (% of twice NYSE CT volume) 24% 13% … 100-499 share orders 6% 3% … 500-1999 share orders 10% 6% … 2000-4999 share orders 5% 3% … 5000-9999 share orders 3% 2%
Nasdaq market orders (% of twice Nasdaq CT volume) 12% 8% … 100-499 share orders 2% 2% … 500-1999 share orders 5% 4% … 2000-4999 share orders 3% 2% … 5000-9999 share orders 2% 1%
Panel B: Executed market orders in the sample and their composition by order size
The table is based on monthly SEC Rule 11Ac1-5 execution-quality reports between November 2001 and December 2002. The sample consists of 1043 NYSE common stocks and 1093 Nasdaq common stocks. CT volume refers to the trading volume for the sample reported on the consolidated tape during regular trading hours between November 2001 and December 2002.
Table 3: Order volume represented in the sample
42
Nasdaq median NYSE medianMedian pairwise
difference Nasdaq mean NYSE meanMean pairwise
difference
Effective spread (in $) $0.092 $0.069 $0.022 (0.00) *** $0.113 $0.074 $0.039 (0.00) ***Effective spread / price 0.0068 0.0045 0.0009 (0.00) *** 0.0085 0.0059 0.0026 (0.00) ***Realized spread (in $) $0.035 $0.010 $0.023 (0.00) *** $0.059 $0.004 $0.055 (0.00) ***Price impact (in $) $0.024 $0.028 -$0.003 (0.00) *** $0.027 $0.035 -$0.008 (0.05) **Time to execution (seconds) 10.6 23.8 -12.2 (0.00) *** 12.2 24.0 -11.9 (0.00) ***
Dependent variable
Average monthly intercept ∆ln(MCAP) ∆(1/PRC) ∆ln(ADV) ∆RR
p-valuefor F-test
∆Effective spread (in $) 0.052 0.039 -0.096 -0.039 0.398 (0.00) (p-value) (0.00) (0.00) (0.00) (0.00) (0.00)∆Effective spread / price 0.0023 0.0004 0.0083 -0.0031 0.0810 (0.00) (p-value) (0.00) (0.19) (0.01) (0.00) (0.00)∆Realized spread (in $) 0.061 0.032 -0.047 -0.040 0.153 (0.00) (p-value) (0.00) (0.03) (0.12) (0.00) (0.58)∆price impact (in $) -0.005 0.004 -0.024 0.001 0.123 (0.00) (p-value) (0.00) (0.59) (0.09) (0.91) (0.35)∆Time to execution (seconds) -12.053 -0.547 2.092 -0.089 -24.304 (0.00) (p-value) (0.00) (0.30) (0.15) (0.81) (0.15)
Table 4: Average market-order execution quality on Nasdaq and the NYSE
Panel B: Monthly panel regression using pairwise differences (249 Nasdaq stocks matched to 249 NYSE stocks, 3423 observations)
Wilcoxon p-value
The table is based on monthly SEC rule 11Ac1-5 execution-quality reports between November 2001 and December 2002. Panels A and B use 249 Nasdaq-NYSE matched pairs. The Nasdaq sample consist of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization, dollar volume, or share volume during 2001Q3. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: market capitalization (MCAP), share price (PRC), adjusted daily dollar volume (ADV), and the daily relative price range (RR) during 2001Q3. Panel A uses market-order execution-quality data that are based on averages across order sizes and months, weighted by shares executed. It presents statistics on pairwise differences of Nasdaq execution quality – NYSE execution quality.
Panel A: Matched-sample pairwise comparison (249 Nasdaq stocks matched to 249 NYSE stocks)p-value of t-
statistic
Panels B and C are based on equations (2) and (3) and do not aggregate over time. Instead, the control variables vary over time with observations corresponding to the month of the 11Ac1-5 report. The results in Panel B are based on a panel regression of pairwise matched differences in execution quality on pairwise differences (∆) in four control variables (ln(MCAP), ln(1/PRC, ln(ADV), and RR). Instead of an intercept, the model uses monthly time fixed effects. The average of these monthly coefficients measures the Nasdaq-NYSE difference in execution quality. The associated test statistic refers to the null hypothesis that all monthly fixed-effect coefficients are jointly equal to zero.The results in Panel C use all securities in the final sample and are based on a panel regression of execution quality on a Nasdaq dummy and levels of the four control variables in Panel B. The regression includes monthly time fixed effects in place of an intercept, but their estimated coefficients are omitted from the table. All regression p-values refer to robust t-statistics, and p-values are in parentheses.
43
Dependent variable
Average monthly intercept Nasdaq dummy ln(MCAP) 1/PRC ln(ADV) RR
p-valuefor F-test
Effective spread (in $) 0.181 0.053 0.014 -0.112 -0.026 -0.224 (0.00) (p-value) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)Effective spread / price 0.0125 0.0022 0.0019 0.0120 -0.0035 0.1098 (0.00) (p-value) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)Realized spread (in $) -0.059 0.059 0.021 -0.029 -0.024 0.015 (0.00) (p-value) (0.00) (0.00) (0.00) (0.00) (0.00) (0.86)Price impact (in $) 0.120 -0.003 -0.003 -0.042 -0.001 -0.120 (0.00) (p-value) (0.00) (0.15) (0.04) (0.00) (0.49) (0.01)Time to execution (seconds) 46.007 -12.583 -0.809 -0.392 -0.320 -15.326 (0.00) (p-value) (0.00) (0.00) (0.00) (0.69) (0.00) (0.01)
Panel C: Monthly panel regression of all securities (1093 Nasdaq stocks and 1043 NYSE stocks, 29611 observations)
44
Order sizeMedian
pairwise difference p-value
Median pairwise
difference p-value
Median pairwise
difference p-value
Median pairwise
difference p-value
Number of pairs 249 248 188 110Effective spread (in $) 0.031 (0.00) 0.019 (0.00) -0.008 (0.03) -0.026 (0.00)Effective spread / price 0.0016 (0.00) 0.0008 (0.00) -0.0005 (0.08) -0.0011 (0.00)Realized spread (in $) 0.038 (0.00) 0.023 (0.00) 0.000 (0.76) -0.019 (0.00)Price impact (in $) -0.004 (0.00) -0.003 (0.01) -0.004 (0.12) -0.005 (0.12)Time to execution (seconds) -18.1 (0.00) -13.9 (0.00) -2.6 (0.12) 4.7 (0.00)
Number of observations 3395 3227 1828 879Effective spread (in $) 0.063 (0.00) 0.046 (0.00) -0.005 (0.00) -0.054 (0.00)Effective spread / price 0.0033 (0.00) 0.0024 (0.00) 0.0003 (0.00) -0.0013 (0.00)Realized spread (in $) 0.086 (0.00) 0.035 (0.00) -0.006 (0.00) -0.033 (0.00)Price impact (in $) -0.011 (0.03) 0.006 (0.01) 0.000 (0.00) -0.010 (0.00)Time to execution (seconds) -17.7 (0.00) -11.0 (0.00) 4.9 (0.00) 19.5 (0.00)
Number of observations 29492 28578 19537 9525Effective spread (in $) 0.065 (0.00) 0.046 (0.00) -0.005 (0.00) -0.040 (0.00)Effective spread / price 0.0032 (0.00) 0.0023 (0.00) 0.0004 (0.00) -0.0009 (0.00)Realized spread (in $) 0.078 (0.00) 0.041 (0.00) -0.001 (0.20) -0.027 (0.00)Price impact (in $) -0.007 (0.04) 0.003 (0.25) -0.002 (0.01) -0.006 (0.00)Time to execution (seconds) -18.0 (0.00) -11.5 (0.00) 5.0 (0.00) 29.3 (0.00)
Table 5: Differences in market-order execution quality on Nasdaq and the NYSE by order size and order typeThe table is based on monthly SEC rule 11Ac1-5 execution-quality reports between November 2001 and December 2002. Panels A and B use 249 Nasdaq-NYSE matched pairs. The Nasdaq sample consist of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization, dollar volume, or share volume during 2001Q3. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: market capitalization (MCAP), share price (PRC), adjusted daily dollar volume (ADV), and the daily relative price range (RR) during 2001Q3. Panel A uses market-order execution-quality data that are based on averages across months, weighted by shares executed. It presents statistics on pairwise differences of Nasdaq execution quality – NYSE execution quality.Panels B and C are based on equations (2) and (3) and do not aggregate over time. Instead, the control variables vary over time with observations corresponding to the month of the 11Ac1-5 report. The results in Panel B are based on a panel regression of pairwise matched differences in execution quality on pairwise differences (∆) in four control variables (ln(MCAP), ln(1/PRC, ln(ADV), and RR). Instead of an intercept, the model uses monthly time fixed effects. Panel B reports only the average of these monthly coefficients that measures the Nasdaq-NYSE difference in execution quality. The associated test statistic refers to the null hypothesis that all monthly fixed-effect coefficients are jointly equal to zero.The results in Panel C use all securities in the final sample and are based on a panel regression of execution quality on a Nasdaq dummy and levels of the four control variables in Panel B. The regression includes monthly time fixed effects in place of an intercept. Panel C reports only the estimated coefficient on the Nasdaq dummy. All regression p-values refer to robust t-statistics, and p-values are in parentheses.
Panel A: Matched-sample pairwise comparison (249 Nasdaq stocks matched to 249 NYSE stocks)
Panel B: Monthly panel regression using pairwise differences (249 Nasdaq stocks matched to 249 NYSE stocks)
Panel C: Monthly panel regression using all securities (1093 Nasdaq stocks and 1043 NYSE stocks)
100-499 shares 500-1999 shares 2000-4999 shares 5000-9999 shares
45
NNasdaq NYSE Nasdaq NYSE Nasdaq NYSE Nasdaq NYSE Nasdaq NYSE
Nov-01 243 $0.121 $0.080 0.78% 0.52% $0.055 $0.016 $0.023 $0.029 9.4 22.1Dec-01 242 $0.106 $0.069 0.70% 0.43% $0.046 $0.017 $0.026 $0.025 8.9 23.1Jan-02 246 $0.102 $0.069 0.59% 0.42% $0.045 $0.015 $0.024 $0.026 9.6 22.1Feb-02 246 $0.092 $0.072 0.61% 0.44% $0.041 $0.015 $0.021 $0.026 6.5 21.5Mar-02 248 $0.084 $0.069 0.50% 0.38% $0.035 $0.013 $0.020 $0.026 8.0 22.3Apr-02 248 $0.087 $0.063 0.56% 0.33% $0.035 $0.012 $0.021 $0.025 7.5 21.9May-02 248 $0.087 $0.064 0.55% 0.34% $0.036 $0.011 $0.021 $0.024 7.5 22.2Jun-02 248 $0.097 $0.066 0.71% 0.38% $0.038 $0.010 $0.022 $0.027 8.4 23.9Jul-02 249 $0.101 $0.071 0.79% 0.48% $0.036 $0.004 $0.025 $0.035 9.0 26.6Aug-02 247 $0.084 $0.065 0.70% 0.48% $0.036 $0.011 $0.019 $0.026 7.0 21.6Sep-02 246 $0.076 $0.062 0.69% 0.44% $0.032 $0.011 $0.018 $0.024 5.4 20.4Oct-02 232 $0.082 $0.060 0.69% 0.47% $0.021 $0.007 $0.024 $0.025 5.7 22.3Nov-02 244 $0.072 $0.054 0.52% 0.40% $0.032 $0.010 $0.015 $0.022 7.9 20.5Dec-02 247 $0.066 $0.050 0.50% 0.36% $0.031 $0.011 $0.011 $0.018 7.1 18.8
Panel A: Median execution quality for 249 matched Nasdaq and NYSE securities
Effective spread (in $)Effective spread /
price Realized spread (in $) Price impact (in $)Time to execution
(seconds)
Table 6: Time trends in market order execution quality on Nasdaq and the NYSEThe table is based on monthly SEC rule 11Ac1-5 execution-quality reports between November 2001 and December 2002. Both panels use market-order execution-quality data that are based on averages across order sizes, weighted by shares executed. Panels A and B use 249 Nasdaq-NYSE matched pairs. The Nasdaq sample consist of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization, dollar volume, or share volume during 2001Q3. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: market capitalization (MCAP), share price (PRC), adjusted daily dollar volume (ADV), and the daily relative price range (RR) during 2001Q3. Panels B and C are based on equations (2) and (3), respectively, and are estimated separately for each month. The results in Panel B are from a regression of pairwise matched differences in execution quality on pairwise differences (∆) in four control variables (ln(MCAP), ln(1/PRC, ln(ADV), and RR). Panel B reports the intercept coefficient, which measures the Nasdaq-NYSE difference in execution quality, and the p-value of the associated robust t-statistic. To test for a linear trend, monthly differences in execution quality are regressed at the firm level on an intercept and a linear time-trend variable (with values ranging from one for Nov 2001 to 14 for Dec 2002). Panel B reports the mean and median coefficient of the trend variable and trend coefficient from a pooled model. The results in Panel C use all securities in the final sample and are based on a regression of execution quality on a Nasdaq dummy and levels of the four control variables in Panel B. Panel C reports the coefficient on the Nasdaq dummy variable, which measures the Nasdaq-NYSE difference in execution quality, and the p-value of the associated robust t-statistic, and p-values are in parentheses.
46
N
Effective spread (in $)
p-value for t-statistic
Effective spread /
pricep-value for t-statistic
Realized spread (in $)
p-value for t-statistic
Price impact (in $)
p-value for t-statistic
Time to execution (seconds)
p-value for t-
statistic
Nov-01 243 0.064 (0.00) 0.0027 (0.00) 0.083 (0.00) -0.009 (0.00) -11.3 (0.00)Dec-01 242 0.050 (0.00) 0.0027 (0.00) 0.055 (0.00) -0.002 (0.24) -14.2 (0.00)Jan-02 246 0.049 (0.00) 0.0025 (0.00) 0.063 (0.00) -0.007 (0.00) -3.9 (0.18)Feb-02 246 0.044 (0.00) 0.0023 (0.00) 0.061 (0.00) -0.009 (0.00) -10.2 (0.00)Mar-02 248 0.032 (0.00) 0.0013 (0.00) 0.051 (0.00) -0.010 (0.00) -10.8 (0.00)Apr-02 247 0.044 (0.00) 0.0019 (0.00) 0.023 (0.36) 0.010 (0.44) -10.0 (0.00)May-02 247 0.044 (0.00) 0.0016 (0.00) -0.040 (0.59) 0.042 (0.10) -11.5 (0.00)Jun-02 247 0.062 (0.00) 0.0026 (0.00) 0.055 (0.00) 0.004 (0.29) -14.3 (0.00)Jul-02 248 0.068 (0.00) 0.0034 (0.00) 0.216 (0.01) -0.074 (0.02) -15.9 (0.00)Aug-02 246 0.054 (0.00) 0.0026 (0.00) 0.086 (0.00) -0.016 (0.11) -14.2 (0.00)Sep-02 245 0.047 (0.00) 0.0025 (0.00) 0.052 (0.00) -0.002 (0.18) -14.3 (0.00)Oct-02 231 0.046 (0.00) 0.0028 (0.00) 0.044 (0.00) 0.001 (0.62) -15.4 (0.00)Nov-02 242 0.038 (0.00) 0.0013 (0.00) 0.060 (0.00) -0.011 (0.08) -10.3 (0.00)Dec-02 245 0.037 (0.00) 0.0016 (0.00) 0.019 (0.33) 0.009 (0.34) -11.6 (0.00)Mean linear trend in 1/100 (t-statistic p-value)
249 -0.070 (0.21) -0.01003 (0.07) 0.058 (0.92) -0.0639 (0.82) -8.2 (0.67)
Median linear trend in 1/100 (Wilcoxon p-value)
249 -0.066 (0.00) -0.00248 (0.00) -0.025 (0.11) 0.0075 (0.81) -10.6 (0.45)
Trend coefficient from pooled model in 1/100 (robust t-statistic p-value)
3423 -0.160 (0.00) -0.00375 (0.12) -0.084 (0.22) -0.0386 (0.56) -19.3 (0.01)
Panel B: Nasdaq-NYSE differentials estimated from monthly regressions using 249 pairwise differences
47
N
Effective spread (in $)
p-value for t-statistic
Effective spread /
pricep-value for t-statistic
Realized spread (in $)
p-value for t-statistic
Price impact (in $)
p-value for t-statistic
Time to execution (seconds)
p-value for t-
statistic
Nov-01 2104 0.064 (0.00) 0.0036 (0.00) 0.083 (0.00) -0.010 (0.00) -13.0 (0.00)Dec-01 2115 0.054 (0.00) 0.0030 (0.00) 0.068 (0.00) -0.007 (0.00) -12.4 (0.00)Jan-02 2127 0.048 (0.00) 0.0029 (0.00) 0.063 (0.00) -0.008 (0.00) -7.8 (0.00)Feb-02 2114 0.040 (0.00) 0.0023 (0.00) 0.056 (0.00) -0.008 (0.00) -13.4 (0.00)Mar-02 2124 0.031 (0.00) 0.0016 (0.00) 0.052 (0.00) -0.010 (0.00) -10.2 (0.00)Apr-02 2127 0.043 (0.00) 0.0020 (0.00) 0.059 (0.00) -0.008 (0.12) -11.6 (0.00)May-02 2124 0.045 (0.00) 0.0020 (0.00) 0.064 (0.02) -0.010 (0.40) -12.5 (0.00)Jun-02 2127 0.058 (0.00) 0.0026 (0.00) 0.064 (0.00) -0.003 (0.05) -14.2 (0.00)Jul-02 2132 0.070 (0.00) 0.0031 (0.00) 0.182 (0.00) -0.056 (0.00) -18.3 (0.00)Aug-02 2128 0.055 (0.00) 0.0019 (0.00) 0.059 (0.00) -0.002 (0.62) -13.6 (0.00)Sep-02 2122 0.049 (0.00) 0.0021 (0.00) 0.055 (0.00) -0.003 (0.02) -14.3 (0.00)Oct-02 2038 0.049 (0.00) 0.0023 (0.00) 0.049 (0.00) 0.000 (0.99) -15.0 (0.00)Nov-02 2110 0.032 (0.00) 0.0015 (0.00) -0.006 (0.86) 0.019 (0.12) -11.0 (0.00)Dec-02 2119 0.035 (0.00) 0.0015 (0.00) -0.010 (0.64) 0.023 (0.00) -10.0 (0.00)
Panel C: Nasdaq-NYSE differentials estimated from monthly regressions using all securities (1093 Nasdaq stocks, 1043 NYSE stocks)
48
Nasdaq median
NYSE median
Median pairwise
differenceWilcoxon p-
value
Time-weighted quoted depth 574 1,076 -417 (0.00)Average trade size 494 695 -196 (0.00)
Order size
Nasdaq median
NYSE median
Wilcoxon p-value
Nasdaq median
NYSE median
Wilcoxon p-value
Nasdaq median
NYSE median
Wilcoxon p-value
Nasdaq median
NYSE median
Wilcoxon p-value
Number of pairs 249 248 188 110
Time to execution (seconds) 2.3 21.9 (0.00) 7.7 23.2 (0.00) 22.6 25.1 (0.21) 34.7 27.5 (0.00)Net price improvement (in $) -0.0002 0.0056 (0.00) -0.0075 -0.0055 (0.00) -0.0190 -0.0251 (0.00) -0.0217 -0.0366 (0.00)Effective spread (in $) 0.072 0.039 (0.00) 0.091 0.066 (0.00) 0.096 0.100 (0.03) 0.080 0.117 (0.00)Quoted spread (in $) 0.069 0.051 (0.00) 0.075 0.055 (0.00) 0.057 0.050 (0.00) 0.035 0.042 (0.07)Fill rate (in %) 97.1% 99.4% (0.00) 92.7% 98.7% (0.00) 86.5% 98.8% (0.00) 81.9% 98.7% (0.00)
Number of observations 3395 3227 1828 879Fill rate (in %) -2.9% (0.00) -7.1% (0.00) -13.7% (0.00) -18.8% (0.00) (0.00)Net price improvement (in $) -0.007 (0.00) -0.003 (0.00) 0.007 (0.00) 0.022 (0.00) (0.00)
Panel A: Matched-sample pairwise comparison of trade sizes and quoted depth (249 Nasdaq stocks matched to 249 NYSE stocks)
5000-9999 sharesPanel B: Matched-sample pairwise comparison (249 Nasdaq stocks matched to 249 NYSE stocks)
Panel C: Monthly panel regression using pairwise differences (249 Nasdaq stocks matched to 249 NYSE stocks)
100-499 shares 500-1999 shares 2000-4999 shares
Table 7: The relation between order size and the characteristics of order executionThe table reports results for 249 Nasdaq-NYSE matched pairs. The Nasdaq sample consist of a dollar-volume stratified sample of 219 securities, plus all 30 securities that were in the top 20 of either market capitalization, dollar volume, or share volume during 2001Q3. The NYSE firms are matched by minimizing the absolute matching error across four dimensions: market capitalization (MCAP), share price (PRC), adjusted daily dollar volume (ADV), and the daily relative price range (RR) during 2001Q3. Panel A is based on trades and quotes during regular market hours that are reported for this sample between November 2001 and December 2002 in the TAQ database. Panels B and C are based on monthly SEC rule 11Ac1-5 execution quality reports between November 2001 and December 2002. Panel B uses market order execution-quality data that are based on averages over months, weighted by shares executed. It presents statistics on pairwise differences of Nasdaq execution quality – NYSE execution quality. Net price improvement is the total dollar amount of price improvement (relative to the relevant side of the quote) minus the total dollar amount of price disimprovement. This measure is normalized by the number of shares executed. Quoted spread is the national best bid and offer spread, computed as the sum of effective spread and twice net price improvement (not normalized). Fill rate is computes as the ratio of shares executed and shares placed. Panel C is based on equation (2) and does not aggregate over time. Instead, the control variables vary over time with observations corresponding to the month of the 11Ac1-5 report. I estimate a panel regression of pairwise matched differences in execution quality on pairwise differences (∆) in four control variables (ln(MCAP), ln(1/PRC, ln(ADV), and RR). Instead of an intercept, the model uses monthly time fixed effects. The average of these monthly coefficients measures the Nasdaq-NYSE difference in execution quality. The associated (robust) test statistic refers to the null hypothesis that all monthly fixed-effect coefficients are jointly equal to zero, and p-values are in parentheses.
49
Figure 1: Time trends in market order execution quality on Nasdaq and the NYSE by order sizeThe sample consists of 1093 domestic common stocks on Nasdaq and 1043 domestic common stocks on the NYSE. The figures graph the Nasdaq-NYSE difference in execution quality, estimated as the coefficient on a Nasdaq dummy variable coefficient from a regression of execution quality on an intercept, a Nasdaq dummy, and four control variables (ln(market capitalization), ln(1/share price), ln(adjusted daily dollar volume), and the daily relative price range). Both panels use execution quality data from 11Ac1-5 reports between November 2001 and December 2002.
-$0.08
-$0.04
$0.00
$0.04
$0.08
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Month (Nov 2001 = 1)
Eff
ectiv
e sp
read
diff
eren
tial
100-499 shares 500-1999 shares2000-4999 shares 5000-9999 shares
-0.3%
-0.2%
0.0%
0.2%
0.3%
0.5%
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Month (Nov 2001 = 1)
Rel
ativ
e ef
fect
ive
spre
ad
diff
eren
tial
100-499 shares 500-1999 shares2000-4999 shares 5000-9999 shares
-$0.05
$0.00
$0.05
$0.10
$0.15
$0.20
$0.25
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Month (Nov 2001 = 1)
Rea
lized
spre
ad d
iffer
entia
l
100-499 shares 500-1999 shares2000-4999 shares 5000-9999 shares
-24-12
01224364860
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Month (Nov 2001 = 1)
Exe
cutio
n sp
eed
diff
eren
tial
(in se
cond
s)
100-499 shares 500-1999 shares2000-4999 shares 5000-9999 shares
50