ARTICLE IN PRESS

Journal of Financial Economics 87 (2008) 249–268

www.elsevier.com/locate/jfec

Liquidity and market efficiency$

Tarun Chordiaa, Richard Rollb,, Avanidhar Subrahmanyamb
a
Goizueta Business School, Emory University, Atlanta GA, 30322, USA
b
Anderson School of Management, University of California, Los Angeles, CA 90095, USA
Received 23 May 2006; received in revised form 28 February 2007; accepted 22 March 2007
Available online 9 October 2007

Abstract

Short-horizon return predictability from order flows is an inverse indicator of market efficiency. We find that such
predictability is diminished when bid-ask spreads are narrower, and has declined over time with the minimum tick size.
Variance ratio tests suggest that prices were closer to random walk benchmarks in the more liquid decimal regime than in
other ones. These findings indicate that liquidity stimulates arbitrage activity, which, in turn, enhances market efficiency.
Further, as the tick size decreased, open-close/close-open return variance ratios increased, while return autocorrelations
decreased. This suggests an increased incorporation of private information into prices during more liquid regimes.
r 2007 Elsevier B.V. All rights reserved.

JEL classification: G12; G14

Keywords: Liquidity; Market efficiency; Order flow

1. Introduction

As recognized at least since Hillmer and Yu (1979), Epps (1979), and Patell and Wolfson (1984), evidence
that a market is quite efficient (Fama, 1970) over a daily horizon does not preclude inefficiencies at shorter
horizons. This is because investors need time to absorb and act on new information. Indeed, recent papers,
e.g., Cushing and Madhavan (2000), and Chordia, Roll, and Subrahmanyam (2005) document that short-
horizon returns are predictable from past order flows. The determinants of this short-horizon predictability
deserve a thorough investigation by finance scholars.

$
We are grateful to two anonymous referees and Bill Schwert (the editor) for their constructive insights and encouragement. We also
thank Doron Avramov, Suman Banerjee, Ekkehart Boehmer, Michael Brandt, Guy Charest, Nai-fu Chen, Gustavo Grullon, Soeren
Hvidkjaer, Vojislav Maksimovic, Barbara Ostdiek, Lemma Senbet, Haluk Unal, James Weston, and seminar participants at Faculdade de
Economia, Universidade do Porto, ISCTE, Rice University, Tulane University, University of California at Irvine, University of Georgia,
University of Maryland, University of Montreal, University of Quebec at Montreal, Universidad de Santiago de Chile, the Norwegian
School of Economics, and the conference on Microstructure of International Financial Markets at the Indian School of Business, for
valuable comments. We also thank Morgan Stanley for the market microstructure grant that funded this project.
Corresponding author. Tel.: +1 310 825 5355; fax: +1 310 206 5455.
E-mail address: [email protected] (R. Roll).

0304-405X/$ - see front matter r 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jfineco.2007.03.005
 ARTICLE IN PRESS
250 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

Short-horizon return predictability should be diminished by arbitrage trading, which should be more
extensive and effective during times in which the market is more liquid. Recent evidence indicates that liquidity
does vary over time and that market conditions can occasionally be so turbulent that effective trading costs
may jump dramatically and liquidity can decline or even disappear.1 An important but hitherto unaddressed
empirical question is whether such fluctuations in liquidity are related to variations in the degree of market
efficiency.2 Further, successive reductions in the minimum tick size have led to exogenous decreases in bid-ask
spreads (see, for example, Bessembinder, 2003; or Chordia, Sarkar, and Subrahmanyam, 2005), and a related
issue is whether such secular increases in liquidity have enhanced market efficiency. In this paper we
address these questions by using return, order flow, and liquidity data for a large sample of NYSE stocks over
a ten-year period (1993–2002).
Other studies explore the return/order flow relation, but around specific events or over short periods of time.
Sias (1997) analyzes order imbalances in the context of institutional buying and selling of closed-end funds;
Lauterbach and Ben-Zion (1993) and Blume, MacKinlay, and Terker (1989) study order imbalances around
the 1987 crash; and Lee (1992) does the same around earnings announcements. Chan and Fong (2000)
investigate the impact of order imbalances on the relation between stock volatility and volume, using data for
about six months. Hasbrouck and Seppi (2001) and Brown, Walsh, and Yuen (1997) study order imbalances
for 30 and 20 stocks, over one and two years, respectively. In comparison, we study return predictability from
order flows in a comprehensive sample of actively traded stocks over more than 2,000 trading days.
A separate line of research explores a link between liquidity and stock returns by way of a premium
demanded by investors for trading in illiquid stocks. For example, Amihud and Mendelson (1986) and Jacoby,
Fowler, and Gottesman (2000) provide theoretical arguments and empirical evidence to support the existence
of liquidity premia in the cross-section. In addition, Jones (2001) and Amihud (2002) show that liquidity
predicts expected returns in the time-series, and Pastor and Stambaugh (2003) and Acharya and Pedersen
(2005) find that expected stock returns are cross-sectionally related to liquidity risk. In our paper, we explore a
distinctly different link between liquidity and asset prices by asking whether liquidity is associated with an
enhanced degree of intraday market efficiency.
Our investigation can be best cast in terms of competing hypotheses on how liquidity and market efficiency
may be related. Note first that illiquidity does not necessarily imply any return predictability from order flows.
In Glosten and Milgrom (1985) or Kyle (1985), for instance, markets are illiquid because prices for buying and
selling shares are different, but prices also are martingales because market makers are risk-neutral. Thus, in
these models, current prices and order flows do not predict future returns. Departing from these paradigms,
there are at least two ways in which return predictability from order flows can arise. First, if market makers
have limited risk-bearing capacity, persistent asymmetric orders will lead to predictable returns due to
temporary deviations of prices from fundamental values. Agents such as floor brokers and floor traders who
are able to detect such deviations may submit arbitrage trades, which may speed up the convergence of prices
to fundamental values. However, the proclivity of these traders to submit arbitrage orders would be negatively
influenced by illiquidity. This mechanism creates a link between liquidity and efficiency. Alternatively, risk-
neutral but cognitively challenged market makers might misreact to the information content of the order flow
(Barberis, Shleifer, and Vishny, 1998) and thereby fail to eliminate return predictability. The mispricing would
create an incentive for outside agents to gather information about order flow. If they do so and trade on such
information, market makers will face increased adverse selection, and as a result the market may be less liquid.
At the same time, prices may be more efficient because more information is impounded into the stock price.
This argument suggests that increased efficiency may be associated with less liquidity. Our analysis may then
be viewed as a test of these competing hypotheses.
For stocks that traded every day on the NYSE from 1993 to 2002, our empirical results include the
following: (a) Intraday market efficiency is intimately linked to daily liquidity; specifically, the forecastability
of returns from lagged order flows is significantly diminished on days when the market is more liquid;

1
‘‘One after another, LTCM’s partners, calling in from Tokyo and London, reported that their markets had dried up. There were no
buyers, no sellers. It was all but impossible to maneuver out of large trading bets.’’—Wall Street Journal, November 16, 1998.
2
Huang and Stoll (1994) show that intraday returns are predictable from past returns, but do not relate this phenomenon to liquidity.
Boehmer, Kelley, and Pirinsky (2005) consider the impact of institutional investors on intraday price discovery.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 251

(b) efficiency has improved with the reduction in the minimum tick size, which corresponds to a decrease in
bid-ask spreads; (c) variance ratio tests using intraday and daily midquote returns indicate that prices were
closer to a random walk benchmark in the more liquid decimal regimes than in the regimes with higher
minimum tick sizes.
The notion of efficient markets (Fama, 1970) emphasizes a lack of return predictability as the criterion for
efficiency, while the microstructure literature emphasizes a distinctly separate measure of financial market
quality, namely, the amount of private information reflected in prices. As Kyle (1985) points out, even semi-
strong efficient prices can reflect varying degrees of private information. Intuitively, markets that become
more liquid due to secular events, such as a reduction in tick size, may incorporate private information more
readily because an exogenous decrease in trading costs may stimulate trading on information about
fundamentals (Admati and Pfleiderer, 1988). We provide evidence of this phenomenon by reporting changes
across three tick size regimes in two information-associated statistics, (a) open-to-close versus close-to-open
return variances and (b) first-order daily return autocorrelations. The idea is to check whether improved
liquidity associated with an exogenous reduction in the tick size enhances informational efficiency by
encouraging trading on private information.
We find that open-to-close/close-to-open variance ratios increased and first-order autocorrelations
decreased (strongly for smaller firms) as the tick size fell. As French and Roll (1986) suggest, an increase in
these variance ratios could be caused either by more mispricing or by increased informational efficiency.
Increased mispricing after a tick size reduction should be associated with higher serial correlation of returns.
However, return autocorrelations decrease with the tick size. Thus, the evidence points to an increase in
private information trading in the lower tick size regimes.
Overall, the data suggest that the liquidity improvement following the tick size reductions has not only
accompanied an increase in the efficiency of accommodating order flows, but has also enhanced informational
efficiency by allowing better incorporation of private information into prices.
The rest of the paper is organized as follows. Section 2 develops our economic arguments and describes the data.
Section 3 presents the intraday evidence. Section 4 presents the variance ratio analysis and Section 5 concludes.

2. Economic reasoning and data description

2.1. Motivation

There are at least two competing reasons for why market efficiency may be related to liquidity. First,
suppose that market makers have limited risk-bearing capacity and/or face inventory financing constraints. In
this situation, price pressures induced by incoming order flows may cause midquotes to temporarily diverge
from fundamental (‘‘full-information’’) values. This implies that returns may be predictable from past order
flows over short horizons.3 Agents who continuously monitor the market, such as day traders, floor traders, or
floor brokers trading for their own account, may detect the divergence of midquotes from true asset values.
They may then submit arbitrage orders to profit from the temporary deviation of midquotes from their full-
information counterparts.4
Note that if the orders of arbitrageurs arrive in sufficient quantities and in a timely manner, they would tend
to quickly reduce market makers’ excess inventories and thus cause midquotes to adjust speedily to initial
imbalance shocks. This adjustment would tend to diminish return predictability, as argued by Chordia, Roll,
and Subrahmanyam (2005). However, the inclination of arbitrageurs to submit such orders would be greater
when the bid-ask spread, a measure of illiquidity, is low.5 This argument suggests that more liquid markets
should exhibit less pronounced return predictability and vice versa.
3
See Stoll (1978). Chordia and Subrahmanyam (2004) show that price pressures due to positively autocorrelated imbalances induce a
positive relation between returns and lagged imbalances.
4
These orders would likely be either marketable limit or market orders to assure speedy execution since the deviation of quote midpoints
from full information values would be short-lived in actively traded stocks.
5
See Peterson and Sirri (2002) for detailed documentation of the effective spreads paid by market orders as well as marketable limit
orders, and for evidence that trades are sensitive to such costs of transacting. Brennan and Subrahmanyam (1998) also show that trading
strategies are affected by illiquidity.
 ARTICLE IN PRESS
252 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

Although the above argument suggests that liquidity reduces the predictability of returns from order flow,
there are competing hypotheses. Suppose that market makers underreact to order flow because of cognitive
limitations (Barberis, Shleifer, and Vishny, 1998). In this situation, other agents may find it profitable to
collect information about order flow and trade on it, thereby increasing efficiency by causing prices to
react more fully to the order flow. The presence of such agents, however, would increase the adverse
selection problem of the market maker, which means that market liquidity could decrease. Thus, in
this scenario, illiquidity would be accompanied by greater market efficiency. And there is yet a third
hypothesis, that if outsider assistance is not needed and rational (designated and de facto) market makers
promptly absorb imbalances by updating quotes, then illiquidity may bear no relation with return
predictability from order flows. Our empirical work can be viewed as resolving the tension among these three
arguments.

2.2. Imbalance and return data

Since the serial dependence in returns is close to zero for active stocks over a daily horizon,6 an investigation
of the relation between liquidity and the efficiency-creating process must focus on intraday trading. An initial
research choice involves the length of the intraday interval over which returns and order flows are measured.
We decide to focus on five-minute intervals for the following two reasons. First, though shorter intervals are
technically possible, nontrading becomes an issue. Second, though longer intervals are feasible, short-lived
market inefficiencies would probably be less conspicuous over such intervals, since a predictive relation
between order imbalances and future returns is unlikely to last very long. Five minutes seems like a good
compromise.7 Also, we exclude small stocks from our study because of the inherent difficulty in measuring
serial dependence at the five-minute horizon when trading is infrequent.
The sample is selected as follows. At the beginning of each year from 1993–2002, all NYSE firms listed on
CRSP are sorted on market capitalization, and the largest 500 are noted. Of these, 193 traded every day on the
NYSE over the entire period and are retained in the sample. The 1993–2002 period is selected because (a)
transactions data are available from the TAQ (Trade and Automated Quotations) database recorded by the
Exchange for this period, and (b) this period spans significant changes in the minimum tick size, which was
reduced from $1/8 to $1/16 during 1997 and was reduced further to one cent by January 2001.8 The tick size
changes are exogenous and provide a natural experiment to test the impact of liquidity on return
predictability. Thus, we hope to discern changes in the price formation process during the years preceding and
following these events. Future investigations should extend the analysis to smaller firms and other years,
exchanges, and countries.
To calculate order imbalances, we apply the Lee and Ready (1991) algorithm, which requires that each
transaction be matched to a bid-ask quote.9 We first filter the trade and quote transactions data as in Chordia,
Roll and Subrahmanyam (2001). Then, following Lee and Ready (1991), from 1993 to 1998 inclusive, we set
the matching quote to be the first quote at least five seconds prior to the trade. Due to a generally accepted
decline in reporting errors in recent years (see, for example, Madhavan, Richardson, and Roomans, 2002),
after 1998 the matching quote is simply the first quote prior to the trades. We apply the Lee and Ready (1991)
method to the matched trade-quote sample and thereby obtain an estimate of whether a particular trade was
buyer- or seller-initiated. Order imbalance for each stock over any time interval can then be calculated
variously as the number of buyer- less the number of seller-initiated trades divided by the total number of
trades (OIB#), or the dollars paid by buyer-initiators less the dollars received by seller-initiators divided by the
6
See, for example, Chordia, Roll, and Subrahmanyam (2005, p. 273).
7
Andersen, Bollerslev, Diebold, and Ebens (2001) also use a five-minute interval when computing daily variances from high frequency
data.
8
Ball and Chordia (2001) show that for the largest stocks, more than half the bid-ask spread in 1996 was due to the impact of rounding
onto the tick grid. Also, Chordia and Subrahmanyam (1995) argue that a reduction in the tick size would result in more competition and
less payment for order flow, thus causing orders to flow to the least-cost providers of market making services.
9
The Lee–Ready algorithm classifies a trade as buyer- (seller-) initiated if it is closer to the ask (bid) of the prevailing quote. If the trade is
exactly at the midpoint of the quote, a ‘‘tick test’’ is used whereby the trade is classified as buyer- (seller-) initiated if the last price change
prior to the trade is positive (negative). Note that a limit order is most often the passive side of the trade; i.e., the noninitiator.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 253

total dollars traded (OIB$).10 The first of these order imbalance measures disregards the size of the trade,
counting small orders equally with large orders. The second measure weights large orders more heavily. Our
results are similar using either measure; therefore, for the most part, we focus on OIB$.
Order imbalances are computed over all trades within each time interval. For example, contemporaneous
OIB$ during the five minutes ending at 9:50 a.m. contains the dollar value of buyer-initiated trades less the
dollar value of seller-initiated trades between 9:45:01 a.m. and 9:50:00 a.m. The corresponding lagged five-
minute OIB$ involves the accumulation between 9:40:01 a.m. and 9:45:00 a.m.
We now turn to the calculation of returns over the short (five-minute) intervals that are the focus of our study.
In computing such returns, several methods can potentially be used. One alternative is to use transaction prices
to compute returns, but such returns are affected by bid-ask bounce. Another alternative is to compute returns
from midquotes that are matched to the last transaction during a particular interval. However, because of
heterogeneities in trading frequencies across stocks, this procedure would have led to returns being computed over
intervals of different lengths for different stocks, confounding the interpretation of the results. Keeping these issues
in mind, we present results from returns that are computed by using the midpoints of the quotes prevailing at the
end of each five-minute interval. When calculations involve lagged values, the first five-minute interval of each
trading day is discarded because it would have been related to a lagged interval from the previous trading day.

2.3. Liquidity data

Measures of aggregate illiquidity are constructed from individual firm bid/ask spreads. The effective spread
is simply twice the absolute difference between the transaction price and the prevailing quote midpoint, and is
obtained from the trade-quote matched sample described in the previous subsection. In a robustness check
reported later, we also use the quoted spread, which is computed as the difference between the transaction-
matched ask and bid quotes. We compute the market-wide illiquidity indicators as follows. First, for each
stock, we calculate the mean effective and quoted spreads over each trading day.11 These are then value-
weighted and averaged across stocks (with market capitalization at the end of the previous year used to
calculate weights) to obtain the aggregate measures.
The sample spans three distinct liquidity subperiods: (i) January 4, 1993–June 23, 1997; (ii) June 24
1997–January 28, 2001; and (iii) January 29, 2001–December 31, 2002. These correspond to the eighth,
sixteenth, and decimal tick size regimes, respectively. Table 1 presents summary statistics associated with the
illiquidity, imbalance, and return measures separately for each of the tick size regimes.
We find that illiquidity has declined sharply across the three periods; thus, the average effective spread was
12 cents in the eighths regime, eight cents in the sixteenths regime, and three cents in the decimal regime. The
order imbalance in number of transactions is fairly stable across the three regimes. While the average dollar
imbalance is higher in the decimal period than in the other two regimes,12 average returns in the post-decimal
period were mediocre relative to the prior period. This result is intriguing because, based on earlier research
that documents a strong imbalance-return relation (e.g., Chordia, Roll, and Subrahmanyam, 2002), we would
expect patterns in order flows to mimic patterns in returns. However, the increased imbalance in the decimal
period may be due to individual investors who trend-chase by extrapolating returns from the prior bull market
(Griffin, Nardari, and Stulz, 2007). These trades would be uninformative and consequently be unrelated to
long-horizon returns.
10
The scaling of imbalance allows for the notion that large firms would have large dollar imbalances simply because they would have
larger volume. However, this raises the concern that the results relating imbalance to returns may be influenced by a relation between
returns and total volume (see, e.g., Gervais, Kaniel, and Mingelgrin, 2001), which appears in the denominator of our measure. We also
experimented with an alternative scaling method, which involves scaling the dollar imbalance by the average trading volume across five-
minute periods throughout the relevant time period. Results from this alternative scaling method are substantively similar to those
presented here.
11
Note that the alternative liquidity proxies of Amihud (2002) and Pastor and Stambaugh (2003) are not readily computable as daily
measures.
12
The sample sizes are equal to the number of five-minute trading intervals during each period, excluding missing observations for all
stocks (e.g., the market was closed after the terrorist attack on September 11, 2001) or, occasionally, when one of the three size groups (to
be reported upon later) is missing a five-minute interval; approximately 0.06% of the total number of five-minute intervals is associated
with missing data. The retained intervals are identical in every calculation.
 ARTICLE IN PRESS
254 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

Table 1
Summary statistics
Quoted and effective spreads (QSPR and ESPR, respectively) and order imbalances (OIB) are from the NYSE TAQ database. Spreads
are averaged across the day for each stock and then value weighted across stocks using market capitalizations at the beginning of each
calendar year. OIB# is the number of buyer-initiated less the number of seller-initiated trades divided by the total number of trades during
a five-minute trading interval. OIB$ is the total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by
the dollar volume of trading during a five-minute trading interval. N is the number of trading days and n is the number of five-minute
trading intervals. The sample spans the years 1993–2002 inclusive, and consists of all stocks that traded every day on the NYSE during the
sample period. The eighths regime spans January 1, 1993–June 23, 1997, the sixteenths June 24, 1997–January 28, 2001, and the decimal
regime spans the remainder of the sample period.

QSPR($) ESPR($) OIB# OIB$

Entire sample Mean 0.1350 0.0896 0.0594 0.0622

N ¼ 2519 Median 0.1447 0.0916 0.0613 0.0705
n ¼ 195,447 Standard Deviation 0.0464 0.0314 0.0925 0.251
Eighths regime Mean 0.1742 0.1176 0.0568 0.0458
N ¼ 1,130 Median 0.1710 0.1175 0.0576 0.0520
n ¼ 87,822 Standard Deviation 0.0093 0.0031 0.166 0.293
Sixteenths regime Mean 0.1305 0.0838 0.0598 0.0705
N ¼ 908 Median 0.1275 0.0842 0.0618 0.0763
n ¼ 70,446 Standard Deviation 0.0130 0.0069 0.139 0.219
Decimal regime Mean 0.0515 0.0349 0.0658 0.0852
N ¼ 481 Median 0.0489 0.0333 0.0663 0.0880
n ¼ 37,179 Standard Deviation 0.0114 0.0073 0.114 0.192

2.4. Construction of five-minute return/imbalance portfolios

Recall that we use the midpoint of the bid-ask quotes prevailing at the end of the time interval to construct
our returns. By using quotes that are not matched to specific trades, we mitigate the problem of
nonsynchronous trading. However, if the quotes are updated too infrequently (for example, if they tend to be
updated properly only when a trade occurs), the effect of nonsynchronous trading may not be fully removed.
In this situation, stocks trading with a lag could induce a spurious apparent forecasting relation between
portfolio imbalances and future portfolio returns. Although we could potentially use individual security data,
which would not be plagued by nonsynchronicity, these data are excessively noisy.13 To alleviate problems
with noisy individual stock data, we conduct the analysis using value-weighted portfolios (with market
capitalizations at the end of the previous year used to calculate weights). In addition, to address any residual
problems arising from nonsynchronous trading, we exclude stocks that did not trade at time t1 from the
portfolio constructed at time t.
Table 2 presents correlations among five-minute returns and order imbalances. Returns exhibit high
contemporaneous correlations with both imbalance measures, and the correlations are reasonably stable
across the three liquidity subperiods.

3. The intraday evidence

3.1. Basic regressions

Table 3 reports a simple regression of five-minute returns on past five-minute order imbalances measured by
the number of transactions (OIB#) and by dollars (OIB$). Consistent with Chordia, Roll, and Subrahmanyam
(2005), lagged OIB is a significant predictor of five-minute returns. This result holds for both measures of
imbalance, though the coefficient is larger for OIB#. A one-standard deviation move in OIB$ increases
13
In any case, we consider results from individual stock-level analyses in Section 3.3.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 255

Table 2
Contemporaneous correlations for five-minute trading intervals
Returns and order imbalances (OIB) are from the NYSE TAQ database. Five-minute returns are computed using the midpoints of the
first and last quotes within each five- minute trading interval; n is the number of five-minute intervals. OIB# is the number of buyer-
initiated less the number of seller-initiated trades divided by the total number of trades during a five-minute interval. OIB$ is the total
dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading during a five-
minute interval. The sample spans the years 1993 to 2002 inclusive and consists of all stocks that traded every day on the NYSE during the
sample period. The eighths regime spans January 1, 1993 to June 23, 1997, the sixteenths June 24, 1997 to January 28, 2001, and the
decimal regime spans the remainder of the sample period.

Return OIB#

Entire Sample OIB# 0.534

n ¼ 195,447 OIB$ 0.504 0.681

Eighths Regime OIB# 0.503

n ¼ 87,822 OIB$ 0.498 0.666
Sixteenths Regime OIB# 0.650
n ¼ 70,446 OIB$ 0.618 0.731
Decimal Regime OIB# 0.595
n ¼ 37,179 OIB$ 0.575 0.630

Table 3
Predictive regressions of five-minute returns on lagged order imbalance, 1993–2002
Returns and order imbalances (OIB) are from the NYSE TAQ database. Five-minute returns are computed using the mid-points of the
first and last quotes within each five- minute trading interval. The return in interval t is denoted Returnt. The variable OIB#t is the number
of buyer-initiated less the number of seller-initiated trades divided by the total number of trades during five-minute interval t. OIB$t is the
total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading during five-
minute interval t. The sample spans the 1993 through 2002 inclusive and consists of all stocks that traded every day on the NYSE during
the sample period. The first five-minute interval of each day is used only in the lagged regressors, and the sample size is 192,928.

Dependent Variable: Returnt

Coefficient t-Statistic

Intercept 0.0021 973

OIB#t1 0.0426 30.75
Adjusted R2 0.0049
Intercept 0.0010 4.83
OIB$t1 0.0259 31.40
Adjusted R2 0.0051

five-minute returns by about 0.1%, which is substantial, given the time interval (five minutes). The
explanatory power of the regression is small, however, with the R2 being only about 0.5%.
To gain insight regarding how market efficiency has changed over time, predictive regressions identical to
those in Table 3 are computed for each calendar month in the ten-year sample. Fig. 1 plots the resulting R2’s
and t-statistics for lagged OIB$. Clearly, the degree of NYSE market efficiency has improved over this decade.
The R2 has moved from a peak of about 8% in the eighths regime to virtually zero by the end of the sample
period. The t-statistic on imbalance has also gone from around 12 at the start of the sample to considerably
less than two towards the end.
We now turn to the interaction of day-to-day liquidity with market efficiency. For parsimony, we focus on
dollar imbalances (OIB$) and the effective spread (ESPR) measure of illiquidity.14 These are probably the
more informative choices because OIB$ provides the economic magnitude of the order imbalance, and the

14
Using OIB# yields similar conclusions.
 ARTICLE IN PRESS
256 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

9% 14
8% 12
7%
10

T-Statistic for Prediction

6%
Predictive R-square

5% 8

4% 6
T-statistic
3% 4
2%
R-square 2
1%
0% 0
Eighths Regime Sixteenths Regime Decimal Regime
-1% -2
Ju 3
Ja 3
Ju 4
Ja 4
Ju 5
Ja 5

Ju 6
Ja 6
Ju 7
Ja 7
Ju 8
Ja 8
Ju 9
Ja 9
Ju 0
Ja 0
Ju 1
Ja 1
Ju 2
2
-9
l-9

9
l-9

9
l-9

9
l-9

9
l-9

9
l-9

9
l-9

0
l-0

0
l-0

0
l-0
n-

n-

n-

n-

n-

n-

n-

n-

n-
n
Ja

Fig. 1. Market inefficiency trend, NYSE, 1993–2002. Five-minute return predictions using lagged (by five minutes) dollar order
imbalance.

0.200

0.175

0.150

0.125
ESPR ($)

0.100

0.075

0.050

0.025
Eighths Regime Sixteenths Regime Decimal Regime
0.000
04 -97
04 -93
04 l-93

04 -94
04 l-94

04 -95
04 l-95

04 -96
04 l-96

-J 7
04 -98
04 l-98

04 -99
04 l-99

04 -00
04 l-00

04 -01
04 l-01

04 -02

2
04 l-9

-0
an

an

an

an

an

an

an

an

an

ul
an
u

u
-J

-J

-J

-J

-J

-J

-J

-J

-J

-J
-J

-J

-J

-J

-J

-J

-J

-J
-J
04

Fig. 2. Value-weighted daily average effective spread, NYSE, 1993–2002.

effective spread is closer to the actual transaction costs incurred by traders. Fig. 2 plots the time series of the
effective spread (ESPR). Clearly, ESPR has experienced three distinct regimes corresponding to subperiods
for the eighth, sixteenth, and decimal minimum tick sizes. Moreover, ESPR appears to be trending to some
extent within each subperiod. In order to examine the interaction of liquidity with return predictability, we
need to separate liquid periods from illiquid ones within each regime. Based on analyses by Blume,
MacKinlay, and Terker (1989), and Cox and Peterson (1994), and on experiences with specific events such as
the LTCM crisis, we surmise that the effect of illiquidity on trading activity (and arbitrage activity in
particular) is likely to be particularly pronounced during days of abnormally low liquidity. We therefore define
‘‘low liquidity’’ days as those when the linearly detrended effective spread for that regime is at least one
standard deviation above the mean.
Table 4 reports summary statistics for effective spreads on low and high liquidity days. The numbers of
illiquid days as a percentage of the total number of trading days are 16%, 13%, and 19%, in the eighth,
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 257

Table 4
Distribution of liquid and illiquid days
Effective spreads (ESPR) are obtained from the NYSE TAQ database. Spreads are averaged across the day for each stock and then
value weighted across stocks using market capitalizations at the beginning of each calendar year. Illiquid days are those on which the daily
effective spread is at least one standard deviation above the detrended expected effective spread for the trading day. The detrending is done
separately for each tick size regime. The sample spans the years 1993–2002 inclusive and consists of all stocks that traded every day on the
NYSE during the sample period. The eighths regime spans January 1, 1993–June 23, 1997, the sixteenths June 24, 1997–January 28, 2001,
and the decimal regime spans the remainder of the sample period.

ESPR on liquid days ESPR on illiquid days

Eighths regime Mean 0.1168 0.1227

Median 0.1170 0.1219
Standard Deviation 0.0022 0.0029
Number of days 976 154
Sixteenths regime Mean 0.0819 0.0986
Median 0.0833 0.0962
Standard Deviation 0.0072 0.0117
Number of days 804 104
Decimal regime Mean 0.0328 0.0462
Median 0.0321 0.0480
Standard Deviation 0.0049 0.0077
Number of days 404 77

sixteenth, and decimal regimes, respectively, and these proportions are reasonably close to each other. The
mean effective spread on illiquid days is 120% of that on liquid days in the sixteenth regime and the
corresponding ratio is 141% for the decimal regime. However, this ratio is only about 105% for the eighths
regime, possibly due to the fact that the larger tick sizes may have caused the measured effective spread to be
higher than its true value due to the impact of rounding,15 leaving less scope for a significant upward
movement of the spread on illiquid days relative to normal days.
To estimate the influence of liquidity on market efficiency, a low-liquidity dummy (ILD) is interacted with
the basic explanatory variable (lagged OIB$) already used in the regressions of Table 3. Thus, the interaction
variable, OIB$t1 * ILD, equals OIB$t1 on days with low liquidity and zero otherwise. Table 5 presents the
results of regressions that include the interaction variable.
For the portfolio that aggregates all sample firms, the coefficients on OIB$t1 * ILD are positive
and significant in all subperiods, and those on OIB$t1 are significant in the two earlier regimes.
The coefficient estimates of OIB$t1 have decreased steadily from 0.030 in the eighths regime to 0.006 in the
decimal regime. The significantly positive coefficients on OIB$t1 * ILD suggest that the ability of lagged
OIB$ to predict returns increases during periods of illiquidity. Thus, liquidity enhances efficiency. Indeed,
during the sixteenth and decimal subperiods, the coefficients on OIB$t1 * ILD are more than twice as
large as those on OIB$t1 alone. Hence, in these later subperiods, which are generally more efficient on
average, illiquidity has a relatively stronger inhibiting influence. For instance, during the eighths regime a
one-standard deviation increase in lagged OIB$ causes a 0.6% increase in returns when markets are liquid
and a 0.8% increase in returns when markets are illiquid. During the decimal regime the impact on returns of a
one standard deviation increase in lagged OIB$ is 0.1% during liquid periods and 0.4% during illiquid
periods.
A decline in the significance and explanatory power of the regressions in Table 5 accompanies the general
improvement in liquidity over the sample period. Adjusted R2’s drop across the three tick size regimes from
around 4% in the eighths regime to less than 1% in the decimal regime. Overall, the evidence suggests that the
secular changes in spreads accompanying tick size reductions (Bessembinder, 2003) have been accompanied by
a marked increase in the degree of market efficiency.

15
See Ball and Chordia (2001).
 ARTICLE IN PRESS
258 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

Table 5
Predictive regressions of five-minute returns on lagged order imbalances, and lagged order imbalances interacted with a dummy variable
for liquidity regimes
Returns and order imbalances (OIB) are from the NYSE TAQ database. The five-minute return, Returnt, is the dependent variable in all
regressions; it is computed using the mid-points of the first and last quotes within five-minute trading interval t. OIB#t1 is the number of
buyer-initiated less the number of seller-initiated trades divided by the total number of trades during five-minute trading interval t1.
OIB$t1 is the total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading
during five-minute trading interval t1. The illiquidity dummy, ILD, is 1.0 when the daily effective spread is at least one standard
deviation above the detrended expected effective spread for the trading day, otherwise zero. The sample spans the years 1993 to 2002
inclusive and consists of all stocks that traded every day on the NYSE during the sample period. The eighths regime spans January 1, 1993
to June 23, 1997, the sixteenths June 24, 1997 to January 28, 2001, and the decimal regime spans the remainder of the sample period. The
sample size is represented by n. (The first interval of each trading day is used only in the lagged regressors.)

Coefficient t-Statistic

Eighths Regime (n ¼ 86,692) OIB$t1 0.0298 45.53

OIB$t1 * ILD 0.0241 13.64
Intercept 0.0010 5.62
Adjusted R2 0.0348
Sixteenths Regime (n ¼ 69,538) OIB$t1 0.0148 7.92
OIB$t1 * ILD 0.0517 9.53
Intercept 0.0012 2.99
Adjusted R2 0.0032

Decimal Regime (n ¼ 36698) OIB$t1 0.0059 1.55

OIB$t1 * ILD 0.0278 3.23
Intercept 0.0003 0.35
Adjusted R2 0.0005

3.2. Size-based subsamples

It is of interest to explore patterns in the liquidity-efficiency relation across groups stratified by firm size.
Table 6 presents results for subsamples formed by ranking firms by market capitalization at the beginning of
the year, dividing the ranked firms into thirds, and then calculating value-weighted returns, imbalances, and
detrended effective spreads within each tierce. Again, the coefficients on OIB$t1 and OIB$t1 * ILD are
positive and significant in eight of the nine cases corresponding to the three firm size groups and the three tick
size regimes. The only exception is that for the large firms during the decimal regime, when only the interactive
variable is significant.
The coefficient estimates on lagged OIB$ across the size groups for the eighths regime decline from 0.029 for
the large firms to 0.012 for the small firms.16 Across the tick size regimes, the coefficients for the large firms
show a monotonic decline with the tick size, as compared to the smaller firms. The coefficient for the smaller
size groups relative to those for the large caps is about two to ten times larger in the sixteenths and decimal
regimes. The coefficient patterns suggest that the efficiency of the large cap sector has increased with the
reduction in minimum tick size and also imply that in the smaller tick size regimes, the degree of market
efficiency is greater for larger firms. Perhaps this finding is explained by the eighth tick size having been more
binding on the larger firms (Ball and Chordia, 2001). Thus, the greater decrease in spreads for large firms
accompanying tick size reductions may account for increased market efficiency in the decimal regimes for
these firms.

16
An interesting aspect of the regressions is that during the eighths regime the coefficient of lagged imbalance alone is greater for larger
firms. This highlights the issues arising from cross-sectional comparisons. For example, Chordia and Subrahmanyam (2004) find that
imbalance autocorrelations are greater for larger firms (they attribute this finding to greater influence of institutional herding in large
firms). Since autocorrelated imbalances are the key driver of return forecastability in Chordia and Subrahmanyam’s (2004) setting,
stronger serial correlations in imbalances in their model imply stronger return predictability from order flows. Thus, the higher imbalance
autocorrelations for larger firms may explain why the lagged imbalance coefficients in the return regressions are larger for such firms.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 259

Table 6
Predictive regressions of five-minute returns on lagged order imbalances, and lagged order imbalances interacted with a dummy variable
for liquidity regimes, by firm-size
Returns and order imbalances (OIB) are from the NYSE TAQ database. The five-minute return, Returnt, is the dependent variable in all
regressions; it is computed using the midpoints of the first and last quotes within five-minute trading interval t. OIB#t1 is the number of
buyer-initiated less the number of seller-initiated trades divided by the total number of trades during five-minute trading interval t1.
OIB$t1 is the total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading
during five-minute trading interval t1. The illiquidity dummy, ILD, is 1.0 when the daily effective spread is at least one standard
deviation above the detrended expected effective spread for the trading day, otherwise zero. The sample spans the years 1993–2002
inclusive and consists of all stocks that traded every day on the NYSE during the sample period. The eighths regime spans January 1,
1993–June 23, 1997, the sixteenths June 24, 1997–January 28, 2001, and the decimal regime spans the remainder of the sample period. Each
firm size category includes one-third of all sample firms. The first interval of each trading day is used only in the lagged regressors. The
sample size is the number of available five-minute intervals in each subperiod, as reported in Table 1, less the number of days (the first
interval of each trading day is used only in the lagged regressors).

Large Firms Mid-Cap Firms Small Firms

Coefficient t-Statistic Coefficient t-Statistic Coefficient t-Statistic

Eighths Regime OIB$t1 0.0293 41.47 0.0192 37.79 0.0115 32.29

OIB$t1 * ILD 0.0222 11.15 0.0097 7.58 0.0103 11.39
Intercept 0.0010 4.68 0.0006 4.12 0.0000 0.21
Adjusted R2 0.0277 0.0227 0.0194
Sixteenths Regime OIB$t1 0.0094 4.63 0.0228 17.67 0.0207 21.39
OIB$t1 * ILD 0.0514 8.62 0.0299 8.68 0.0318 12.37
Intercept 0.0008 1.82 0.0019 6.15 0.0015 5.79
Adjusted R2 0.0019 0.0079 0.0129
Decimal Regime OIB$t1 0.0029 0.74 0.0111 3.44 0.0245 8.84
OIB$t1 * ILD 0.0287 3.11 0.0187 2.74 0.0259 4.45
Intercept 0.0002 0.20 0.0005 0.82 0.0022 3.63
Adjusted R2 0.0003 0.0007 0.0039

3.3. Robustness checks

We now present the results of additional analyses that are intended to test the interpretation and reliability
of our results.

3.3.1. Issues surrounding interpretation of results

In interpreting the interaction coefficient on OIB and illiquidity, one possible issue is that an exogenous
shock might cause extreme order imbalances and simultaneously reduce liquidity. We could be picking up the
effect of the shock rather than capturing the role of liquidity in assisting the establishment of market efficiency.
To investigate this possibility, we compute absolute order imbalances within liquid and illiquid periods for
each of the three size groups, across the tick size regimes. In each of the nine cases, the absolute imbalances
differed only minimally across liquid and illiquid periods (the difference is less than 1% in every case). This
finding suggests that the identified illiquid periods are not capturing periods of abnormally high (absolute)
order imbalances. Moreover, there is no reason to believe that illiquidity is jointly determined with signed
imbalances. Thus, when markets are illiquid, one may expect to see fewer buy as well as fewer sell orders, but
this has no implication for net order flow, which is our right-hand variable.
As a further check against an unknown contemporaneous influence on return predictability and illiquidity,
we try an alternative definition of illiquid periods based on the effective spread on the previous day
(as opposed to the contemporaneous spread), and find that this does not have any material impact upon the
central results. This result is to be expected, given the high autocorrelation in liquidity documented by
Chordia, Sarkar, and Subrahmanyam (2005).
To shed additional light on the finding that increased liquidity on a given day is associated with more
efficient markets on the next day, we conduct Granger causality tests, which relate our efficiency metric to
 ARTICLE IN PRESS
260 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

illiquidity. In this process, we first perform regressions of returns on lagged imbalances on a daily basis.
Specifically, we estimate one time-series regression of the type in Table 3 for each day in our sample, using all
five-minute intervals available for that day. Next, we used the imbalance coefficient and our illiquidity
measure (i.e., the effective spread) in a vector autoregression. Thus, we estimate a bivariate system in which the
daily imbalance coefficient and the daily effective spread are each modeled as linear functions of lagged
coefficients and lagged values of the spread.17 We then perform our causality tests using the resulting
coefficient estimates. We find that the p-value for the null that the effective spread Granger-causes the lagged
imbalance coefficient is 0.001, whereas that for the reverse hypothesis that the coefficient causes the spread is
0.141. The respective p-values for the (former, latter) causality tests within tick size regimes were (0.073, 0.202),
(0.039, 0.105), and (0.032, 0.417) for the eighth, sixteenth, and decimal subperiods, respectively. Thus, in each
case there is evidence that the effective spread Granger-causes the coefficient in the regression of returns on
lagged imbalance, but there is no evidence of econometric causality running from the regression coefficient to
the spread. This finding supports our interpretation that liquidity causes markets to be more efficient.
Since previous work (Chordia and Subrahmanyam, 2004) suggests that autocorrelated imbalances may be a
key source of predictability of future returns from imbalances, one concern is that our results may be
influenced by the behavior of imbalance autocorrelations over time. It is therefore worth considering how
imbalance autocorrelations have behaved across the three tick size regimes. We find that the first-order serial
correlations in OIB$ are 0.266, 0.186, and 0.187 during the eighths, sixteenths, and decimal regimes,
respectively. While there is some decrease in autocorrelations during the smaller tick regimes, it is not enough
to reduce the R2 from about 8% at the beginning of the sample (Fig. 1) to virtually zero in the decimal period.
Also, from the sixteenth to the decimal period, the R2 and predictive t-statistic dropped further with no
reduction in the autocorrelation of imbalance. The evidence thus supports the hypothesis that increased
liquidity allows outsiders to more effectively assist the designated dealer (i.e., the specialist) in absorbing
investor demands, which assists in the establishment of efficiency.
Our basic contention is that return predictability from order flows is diminished during liquid periods
because market makers receive greater assistance from outsiders in these periods. Thus, we propose that return
predictability from order flows arises because of the limited ability of rational market makers to absorb
outsider demands, rather than due to any cognitive limitation on the part of market makers. To bolster this
notion, we decompose order imbalances into expected and unexpected components using an AR(12) model
for OIB.18 We would expect rational market makers using Bayesian updating to largely incorporate the effects
of predictable imbalances in prior periods, implying that the predictability of returns from lagged order
flows should arise principally from the unexpected component of order flow in any given period. Results
of predictive return regressions using the expected and unexpected components of OIB are presented in
Table 7.19
In general, the results show that the predictability of returns emanates largely from the unexpected
component of OIB. In almost every case, only the innovation coefficients are significant. While there is one
instance in the eighths regime where the coefficient of the predictable component is significant, its magnitude is
smaller than that of the corresponding innovation coefficient. Also, there is reason to believe that the
significance of the predictable component may have arisen due to the large tick size, which may have caused a
sluggish price response to order imbalances in the eighths regime. Overall, the evidence suggests that market
makers react rationally to order flow and that in liquid periods they receive assistance from outsiders, which
mitigates return predictability from order flow.
It also is worth obtaining some perspective on the role of institutional frictions, such as price continuity
rules, that may preclude NYSE specialists from responding quickly to large imbalances and thus influence
return predictability. In this regard, note that our central hypothesis is not that returns are predictable from
order flows (this has been established in earlier literature, e.g., Chordia and Subrahmanyam, 2004). Rather, we
17
We use the Akaike information criterion to select the lag length based on the corresponding vector autoregression. Use of the
alternative Schwarz criterion leads to similar results. The lag length used ranges from two to eight across the four autoregressions for the
full sample and the three tick size regimes.
18
We try a number of alternative lag structures going from three to 15 lags, and find qualitatively similar results.
19
In the remainder of this subsection and the next one, we revert to using the effective spread as a measure of illiquidity, though the
results are qualitatively identical if quoted spreads are used.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 261

Table 7
Predictive regressions of five-minute returns on lagged expected and unexpected order imbalances, and lagged order imbalances interacted
with a dummy variable for liquidity regimes
Returns and order imbalances (OIB) are from the NYSE TAQ database. The five-minute return, Returnt, is the dependent variable in all
regressions; it is computed using the mid-points of the first and last quotes within five-minute trading interval t. OIB is measured as the
total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading during five-
minute trading interval t1.
OIB is decomposed using an AR(12) model into unexpected (OIBI) and expected (OIBP) components. The illiquidity dummy, ILD, is
1.0 when the daily effective spread is at least one standard deviation above the detrended expected effective spread for the trading day,
otherwise zero. The sample spans the years 1993 to 2002 inclusive and consists of all stocks that traded every day on the NYSE during the
sample period. The eighths regime spans January 1, 1993–June 23, 1997, the sixteenths June 24, 1997–January 28, 2001, and the decimal
regime spans the remainder of the sample period. The sample size is the number of available five-minute intervals in each sub-period, as
reported in Table 1, less the number of days (the first interval of each trading day is used only in the lagged regressors).

Coefficient t-Statistic

Eighths Regime OIBI$t1 0.0311 39.27

OIBI$t1 * ILD 0.0237 10.87
OIBP$t1 0.0158 5.90
OIBP$t1 * ILD 0.0038 0.58
Intercept 0.0000 0.12
Adjusted R2 0.0257
Sixteenths Regime OIBI$t1 0.0131 7.51
OIBI$t1 * ILD 0.0310 6.15
OIBP$t1 0.0012 0.16
OIBP$t1 * ILD 0.0131 1.80
Intercept 0.0003 0.50
Adjusted R2 0.0016
Decimal Regime OIBI$t1 0.0023 0.60
OIBI$t1 * ILD 0.0254 2.67
OIBP$t1 0.0157 1.03
OIBP$t1 * ILD 0.0195 1.03
Intercept 0.0005 0.33
Adjusted R2 0.0003

argue that liquidity enhances efficiency, a claim that would seem to be worth investigating even in the presence
of price continuity rules. Nonetheless, one possible way to address this issue is to use NASDAQ stocks, which
are not subject to price continuity rules. For these stocks, however, there are issues related to measurement of
liquidity due to the lack of odd-eighths quotes in the earlier years of our sample period, which reduces the
time-series variation in spreads (Christie and Schultz, 1994). Furthermore, many NASDAQ stocks are not
actively traded, which further complicates any analysis of intraday efficiency. We address these issues by
taking the largest ten NASDAQ stocks ranked by market capitalization at the end of 2000. We then apply our
portfolio technique and run regressions analogous to those in Table 5 for the sample from January 29, 2001 to
December 31, 2002 (the NYSE post-decimal period). The results are similar to those in Table 5; while order
imbalance itself is not significant, its interaction with the illiquid days indicator is significant with a magnitude
(t-statistic) of 0.259 (15.13), and both numbers are much larger than the corresponding ones in Table 5. This
suggests that even in the absence of price continuity rules, liquidity enhances efficiency, and the effect appears
to be stronger in NASDAQ stocks.

3.3.2. Alternative illiquidity specifications

Our liquidity indicator has hitherto been the effective spread, which measures the deviation of transaction
prices from quote midpoints. In a sense, this measure is related to the contemporaneous price impact of a
trade. While there is no reason to expect a mechanical relation between the contemporaneous price impact and
future return predictability from past order imbalances, there may still be a residual concern about relating a
transaction-price based liquidity measure to the return-imbalance relation. Thus, for robustness, we also try
an alternative illiquidity indicator, namely, the quoted spread (QSPR). This measure, being the difference
 ARTICLE IN PRESS
262 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

between quoted ask and quoted bid prices, is not mechanically related to the transaction price response to
order flow. We calculate the variable ILD based on QSPR and run regressions analogous to those in Tables 5
and 6. The results are qualitatively unaltered relative to those in the earlier tables. In particular, return
predictability is diminished during liquid days in every instance. This lends confidence that our previous results
are robust to alternative measures of liquidity.
Another robustness check involves examining alternative procedures for computing illiquidity innovations.
Our previous detrending procedure is a parsimonious but subjective choice. We examine an alternative
procedure whereby the effective spread is regressed on linear as well as quadratic trend terms. In addition,
following Roll, Schwartz, and Subrahmanyam (2007), we include indicator variables for calendar effects, such
as day-of-the-week and month-of-the-year regularities, and for trading days that precede major holidays
(Christmas, New Year’s Day, Independence Day, and Thanksgiving). Illiquidity innovations are defined as the
residual from this augmented regression. Note that adjustment for the preceding variables can potentially
work against finding support for our main thesis. This is because markets may be more efficient during days of
the week when the market is more liquid, and adjusting for calendar regularities would negate this effect.
Nonetheless, the results from this alternative adjustment procedure, presented in Table 8, show that the
significance of the variables improves slightly, but the results are otherwise unchanged, indicating that
previous results are robust to the illiquidity adjustment procedure.

3.3.3. Individual stock regressions

Our final check addresses the fact that we use portfolios rather than individual stocks (as do Brennan,
Jegadeesh, and Swaminathan, 1993; Lo and MacKinlay, 1990, in their studies of market efficiency). To
examine the impact of the portfolio method, we also calculate regressions for individual stocks using illiquid
days defined relative to the effective spread for each stock. The regressions are analogous to those in Table 5,
but we also control for contemporaneous and lagged market return to reduce cross-sectional correlations in

Table 8
Predictive regressions of five-minute returns on lagged order imbalances, and lagged order imbalances interacted with a dummy variable
for liquidity regimes, using an expanded model of illiquidity innovations
Returns and order imbalances (OIB) are from the NYSE TAQ database. The five-minute return, Returnt, is the dependent variable in all
regressions; it is computed using the mid-points of the first and last quotes within five-minute trading interval t. OIB#t1 is the number of
buyer-initiated less the number of seller-initiated trades divided by the total number of trades during five-minute trading interval t1.
OIB$t1 is the total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading
during five-minute trading interval t1. The illiquidity dummy, ILD, is 1.0 when the daily effective spread is at least one standard
deviation above the adjusted expected effective spread for the trading day, otherwise zero. The adjustment consists of regressing effective
spreads on day of the week, month of the year, linear and quadratic trend terms, and a dummy variable for whether a day preceded a
holiday (Independence Day, Christmas Day, or Thanksgiving). The sample spans the years 1993 to 2002 inclusive and consists of all stocks
that traded every day on the NYSE during the sample period. The eighths regime spans January 1, 1993 to June 23, 1997, the sixteenths
June 24, 1997 to January 28, 2001, and the decimal regime spans the remainder of the sample period. The sample size is the number of
available five-minute intervals in each sub-period, as reported in Table 1, less the number of days (the first interval of each trading day is
used only in the lagged regressors).

Coefficient t-statistic

Eighths Regime OIB$t1 0.0304 46.46

OIB$t1 * ILD 0.0195 11.04
Intercept 0.0010 5.67
Adjusted R2 0.0341
Sixteenths Regime OIB$t1 0.0138 7.38
OIB$t1 * ILD 0.0577 10.79
Intercept 0.0011 2.80
Adjusted R2 0.0035

Decimal Regime OIB$t1 0.0059 1.59

OIB$t1 * ILD 0.0330 3.55
Intercept 0.0003 0.36
Adjusted R2 0.0005
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 263

Table 9
Individual Stock Predictive regressions of five-minute returns on lagged order imbalances, and lagged order imbalances interacted with a
dummy variable for liquidity regimes
Returns and order imbalances (OIB) are from the NYSE TAQ database. The five-minute return, Returnt, is the dependent variable in all
regressions; it is computed using the mid-points of the first and last quotes within five-minute trading interval t. OIB#t1 is the number of
buyer-initiated less the number of seller-initiated trades divided by the total number of trades during five-minute trading interval t1.
OIB$t1 is the total dollars paid by buyer-initiators less the total dollars received by seller-initiators divided by the dollar volume of trading
during five-minute trading interval t1. The illiquidity dummy, ILD, is 1.0 when the daily effective spread of a stock is at least one
standard deviation above the adjusted expected effective spread for the trading day, otherwise zero. The sample spans the years 1993–2002
inclusive and consists of all stocks that traded every day on the NYSE during the sample period. The regression includes the
contemporaneous and lagged equally-weighted average of all sample returns (the coefficients of these variables are not reported for
brevity). The eighths regime spans January 1, 1993–June 23, 1997, the sixteenths June 24, 1997–January 28, 2001, and the decimal regime
spans the remainder of the sample period. The first interval of each trading day is used only in the lagged regressors. The t-statistics are
adjusted for cross-correlations in the residuals. All coefficients are multiplied by 1000.

Coefficient t-statistic

Eighths Regime OIB$t1 0.1369 54.04

OIB$t1 * ILD 0.0318 6.17
Intercept 0.0041 8.26
Average adjusted R2 0.389
Sixteenths Regime OIB$t1 0.1130 31.69
OIB$t1 * ILD 0.0471 5.42
Intercept 0.0024 4.31
Adjusted R2 0.333
Decimal Regime OIB$t1 0.0154 1.62
OIB$t1 * ILD 0.0518 2.65
Intercept 0.0006 0.35
Adjusted R2 0.367

the regression residuals.20 We also account for any remaining cross-correlation by correcting the t-statistic of
the mean coefficient across all individual stock regressions. This correction, mentioned in Footnote 8 of
Chordia, Roll, and Subrahmanyam (2000), assumes that the residual cross-correlation and the residual
variance are homogeneous across stocks. Under those assumptions, the standard error of the cross-sectional
mean is inflated by [1+(N1)r]1/2, where N is the number of regressions and r is the common cross-
correlation in the residuals,21 which is proxied by the average residual cross-correlation across all pairs of the
individual stock regressions. The cross-sectional average regression coefficients are presented in Table 9. As
can be seen, the significance pattern of the coefficients mimics that in Table 5, though the magnitude of the
coefficients is smaller. Again, results for the size-stratified portfolios are not presented for brevity but are
broadly consistent with those in Table 6.
Overall, based on all of the evidence, the evidence largely supports the hypothesis that efficiency-creating
activity is enhanced when the market is liquid. In the next section, we show how variance ratios can be used to
shed further light on the relation between liquidity and efficiency.

4. Variance ratios

4.1. Deviations from a random walk benchmark across the tick size regimes

The preceding analysis focuses on the predictive relation between returns and order flows intraday as an
inverse measure of market efficiency and, implicitly, the quality of a financial market. An alternative approach
20
The market return is defined simply as the equally-weighted average of returns across all stocks within a particular five minute interval.
21
The formula of Chordia, Roll, and Subrahmanyam (2000) contains a typographical error. Specifically, it contains an extraneous
digit, 2.
 ARTICLE IN PRESS
264 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

involves the comparison of short- and long-horizon variance ratios. This approach is based on the observation
that for a random walk price process, the variance of long-horizon returns is q times the variance
of short horizon returns, where q is the number of short horizon intervals in the longer horizon.22 Deviations
from a random walk (efficient-market) benchmark can arise because inventory control activity can
induce return serial correlation (Grossman and Miller, 1988). But the deviation would be mitigated if
arbitrageurs assist the market maker in absorbing outside orders, and our hypothesis contends that
there should be smaller deviations from random walk benchmarks during more liquid regimes when trading is
less costly.
We consider the ratio of midquote return variances computed from five-minute intervals and from open-to-
close of trading days. In computing this variance ratio, the five-minute return variance is multiplied by the
number of five-minute intervals in a trading day. For a random walk, this scaled variance ratio would
converge to unity in large samples.
Panel A of Table 10 reports the five-minute/open-close variance ratios for each tick size regime and for each
size-based portfolio. Consistent with intuition, the variance ratios are closer to unity for larger firms,
indicating that the prices for these firms conform more closely to random walks. The point estimates of the
variance ratios are lowest for the decimal regimes. In addition, the variance ratios for small and midcap firms
in the decimal regime are significantly lower relative to the corresponding ones in the eighths regime. This
indicates that the extent of ‘‘noise’’ induced by the trading process is smaller in the decimal regime, especially
for the small and midcap firms. Given the secular decrease in bid-ask spreads brought about by the reduction
in the tick size, the evidence accords with our hypothesis that deviations from random walk benchmarks
should be reduced when markets are more liquid.

4.2. Open-close/close-open return variance ratios and return autocorrelations

The preceding analysis implicitly assumes deviations from a random walk benchmark to be a metric for the
noise introduced by the trading process. Lower levels of such noise implicitly imply higher market quality.
However, the microstructure literature also considers the degree to which prices reflect private information
about fundamentals as an indicator of a financial market’s quality. This measure is generally termed
informational efficiency. Thus, even if returns are not predictable from past public information, they can
reflect varying degrees of private information, as implied by Kyle’s (1985) model.
It is reasonable to expect more liquid markets to be accompanied by greater informational efficiency. For
example, if a smaller tick size allows agents to trade on smaller pieces of information, one would expect more
informed trading as the tick size decreases. Hence, informational efficiency should progressively increase
across the three tick size regimes. This is not the only possible hypothesis, however. If a reduction in tick sizes
adversely affects institutional execution costs due to decreased depth (Jones and Lipson, 2001), informational
efficiency could be adversely affected because of reduced trading on information by large institutions.
We use the ratios of open-to-close and close-to-open return variances to provide some evidence regarding
these hypotheses. French and Roll (1986) relate these ratios to the amount of information incorporated into
prices. They show that (per hour) open-close return variances are greater than (per hour) close-open ones and
offer three potential explanations for this finding: (i) incorporation of private information during trading
hours, (ii) mispricing caused by investor misreaction or market frictions and microstructure noise induced by
bid-ask bounce, and (iii) greater incorporation of public information into prices during trading hours. They
reject (iii) because the variance ratios are not significantly different on business days when the stock market is
closed. They conclude that the other two components help explain the higher ratio during market trading
hours, with (i) being the dominant factor. Relying on their conclusions, the variance ratio measures
either mispricing or the incorporation of private information into prices during trading hours. Thus,
the variance ratio is possibly linked to the informativeness or efficiency of the pricing system in the sense of
Kyle (1985).
We seek to discern whether variance ratios have changed across the three tick size regimes, and if so,
whether the pattern of changes reveals which of the two components identified by French and Roll (1986) is
22
See, for example, Lo and MacKinlay (1990).
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 265

Table 10
Daily variance ratios and autocorrelations
Panel A presents the ratio of five minute return variance to open-to-close return variance (scaled by the number of five minute intervals
in a day). Panel B presents (open-to-close)C(close-to-open) per hour return variance ratios, based on mid-quote returns, across three tick
size regimes (eighths, sixteenths, and decimals). Panel C presents daily first order return autocorrelations across these regimes. The sample
spans the years 1993 to 2002 inclusive, and consists of all stocks that traded every day on the NYSE during the sample period. The eighths
regime spans January 1, 1993 to June 23, 1997, the sixteenths June 24, 1997 to January 28, 2001, and the decimal regime spans the
remainder of the sample period. In Panel C, p-values for the relevant numbers being statistically different from zero are provided in
parentheses. A * (**) in Panel A (Panel B) indicates that the relevant number is statistically higher than the corresponding number in the
eighths (sixteenths) regime at the 5% level. A * (y) in Panels A and C indicates that the relevant number is statistically different from the
corresponding number in the eighths regime at the 5% (10%) level. The statistical conclusions for Panels A and B derive from bootstrap
analyses described in the text.

Eighths Sixteenths Decimals

Panel A: Five minutes/daily variance ratios

Large firms 1.21 1.21 1.10
Mid-cap firms 1.81 1.82 1.36*
Small firms 2.28 2.38 1.79y
Panel B: Per hour open/close variance ratios
Large firms 6.17 7.14 6.69
Mid-cap firms 11.85 7.88 18.01**
Small firms 8.95 15.33 22.16*

Panel C: First order autocorrelations of daily returns

Large firms 0.0572 0.0228 0.0039
(0.055) (0.493) (0.932)
Mid-cap firms 0.0924 0.0697 0.0533
(0.002) (0.036) (0.244)
Small firms 0.2121 0.0814* 0.0339*
(0.000) (0.014) (0.458)

relatively more important.23 Panel B of Table 10 presents the (open-close)C(close-open) per hour variance
ratios for the three tick size regimes and the three size groups. To obtain per hour variances, open-close and
close-open returns are first used separately to compute raw variances. Then, each computed variance is divided
by the total number of calendar hours over the subsample in question.
As shown in the table, all per hour variance ratios are much higher than unity. To test statistical
significance, we compute a nonparametric sign test from paired open-close and close-open squared returns,
each normalized by the number of hours in their respective periods. The computed t-statistics far exceed two
over the nine comparisons corresponding to the variance ratios in Table 10, thereby indicating a very high
level of statistical probability that open-close volatility far exceeds close-open volatility. This finding is
consistent with that of French and Roll (1986).
Table 10 also reveals that all of the variance ratios in the decimal regime are greater than the corresponding
ratios in the eighths regime. We employ a bootstrap analysis to check the statistical significance of these
differences. In the bootstrap, pseudo regimes, with the same numbers of observations as in the original
data, are formed by randomly sorting actual dates, with replacement, and then by computing a ratio of
ratios, i.e., (open-close)C(close-open) variance ratio in a lower pseudo tick size regime divided by the
(open-close)C(close-open) variance ratio in the pseudo eighths regime. This procedure is repeated 5000 times,
and the actual observed ratio of the variance ratios is compared to the relevant fractile of the bootstrapped
distribution. The tests indicate that the per hour (open-close)C(close-open) variance ratios for midcap firms

23
We also estimate variance ratios across liquid and illiquid days (defined as in the previous section) and find that return variances are
consistently higher on illiquid days. On the one hand, this is consistent with the notion that high illiquidity increases asset price fluctuations
(Subrahmanyam, 1994). However, there is an endogeneity problem because exogenous volatility shocks (unrelated to private information)
could cause reduced liquidity due to greater inventory risk. Because of the resulting interpretational problem, we do not present details of
these empirical results.
 ARTICLE IN PRESS
266 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

during the decimal regime are statistically greater than the corresponding ratios in the sixteenth regime; the
same conclusion applies for smallcap firms in the decimal regime vis-à-vis the eighths regime. The point
estimates indicate that the variance ratios for the small and midcap firms generally increased in the decimal
period relative to the other periods.
As mentioned earlier the higher variance ratios can be attributed either to an increase in mispricing or to an
increase in privately informed trading that results in the incorporation of more information into prices when
the market is open. To provide some suggestive evidence that distinguishes between these two possibilities, we
consider the first-order daily return autocorrelations (based on the midpoint of the last quote in effect at the
end of a trading day) across the three tick regimes. French and Roll (1986) argue that the absolute levels of
these autocorrelations should be positively related to microstructural frictions or misreactions of investors to
new information, and thus capture mispricing.24
The autocorrelations are presented in Panel C of Table 10 (together with p-values for whether they are
statistically different from zero). They are positive in all but one (statistically insignificant) case, and they
generally decrease across the regimes. For example, none of the autocorrelations is significant in the decimal
regime, whereas five of the six autocorrelations in the other two regimes are positive and significant. Further,
the lower autocorrelations in the sixteenths and decimal regimes for the smallest firms are statistically different
from the corresponding one in the eighths regime. In considering these autocorrelations, note that
nonsynchronous trading and bid-ask bounce are not materially relevant in our context, because firms are
included only if they traded every day and end-of-day midquotes are used to compute returns. The positive
autocorrelations are thus consistent with investor underreaction to information (Barberis, Shleifer, and
Vishny, 1998), and/or with incomplete adjustment to autocorrelated order flow owing to market maker risk
aversion (Chordia and Subrahmanyam, 2004). Either of these two phenomena can be termed ‘‘mispricing,’’ in
the sense of French and Roll (1986). Such mispricing may arise due to behavioral biases (the former case) or
insufficient market making capacity (the latter case).
Regardless of the cause, if such mispricing were driving the increase in variance ratios across time,
autocorrelations should have increased along with variance ratios as the tick size decreased; but there is no
evidence of this increase. In fact, there is reliable evidence that the opposite transpired for smaller firms.
Consequently, the decrease in the variance ratios is consistent with the more effective incorporation of private
information into prices in the lower tick regimes, especially for smaller firms. A plausible rationale for this
phenomenon is that with lower bid-ask spreads, informed traders might find it worthwhile to trade on private
information even with modest profit potential.25
Overall, all of the evidence is consistent with our proposed hypotheses. Increases in liquidity facilitate
efficiency via two distinct channels. First, return predictability from order flows is diminished during
periods of high liquidity because arbitrageurs are better able to assist specialists in absorbing order flows
during such periods. Second, a reduction in the minimum price change allows for the collection of more
information that, in turn, increases informational efficiency by allowing prices to reflect more information
about fundamentals.

5. Conclusions

In an efficient market, return predictability from past information should be short-lived and minimal. Given
the evidence that such predictability does exist in the short run, understanding its time variation and its
relation to other financial market attributes, such as liquidity, are of fundamental importance. Based on this
motivation, we examine how the predictive relation between returns and order flow varies over time and across
different liquidity regimes.
24
In Kyle (1985), where market makers have unlimited market making capacity, autocorrelations are zero. However, nonzero
autocorrelations obtain in irrational investor models such as Daniel, Hirshleifer, and Subrahmanyam (1998) as well as in the inventory
model of Grossman and Miller (1988).
25
The increased informed trading, in turn, could potentially reduce equilibrium liquidity (increase spreads) in the post-decimal period.
Our variance ratio results in Table 10, coupled with the finding that actual spreads generally decreased over time (Fig. 2), suggest that
while decreased tick sizes have allowed for increased informed trading, the effect of this increase on illiquidity has not been sufficient to
cause spreads to rise to the levels during the eighths regime.
 ARTICLE IN PRESS
T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268 267

Our analysis relies on a continuous series of short-horizon returns for a comprehensive sample of all NYSE
stocks that traded every day during the ten-year period from 1993 to 2002. Return predictability from order
flows has declined substantially over time with reductions in the minimum tick size. Such predictability is
markedly diminished during liquid periods within each tick regime. Prices are closer to random walk
benchmarks during the more recent decimal tick size regime than in earlier ones. The overall evidence is
consistent with the hypothesis that increased arbitrage activity during more liquid periods enhances market
efficiency.
Distinct from the Fama (1970) notion that efficiency implies a lack of return predictability, the
microstructure literature also considers informational efficiency, which is defined as the amount of private
information revealed in prices. We shed light on this measure of a financial market’s quality by considering
patterns in per hour open-to-close/close-to-open variance ratios. Variance ratios generally have increased
while first- order return autocorrelations have declined as the minimum tick size was reduced; this pattern is
particularly strong for smaller firms. This suggests that the observed increase in variance ratios is not due to
increased mispricing, but to more private information being reflected in prices following the tick size
reductions. In sum, this evidence is consistent with greater liquidity engendering a higher degree of
informational efficiency.
An extension of this analysis would be to study the relation between return predictability and illiquidity for
fixed income and currency markets. Considering the forecastability of returns from order flows prior to
important announcements (when the market might be particularly illiquid due to impending uncertainty)
would also be interesting. Pursuit of such topics appears to be a worthwhile agenda for future research.

References

Acharya, V., Pedersen, L., 2005. Asset pricing with liquidity risk. Journal of Financial Economics 77, 385–410.
Admati, A., Pfleiderer, P., 1988. A theory of intraday patterns: volume and price variability. Review of Financial Studies 1, 3–40.
Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets 5, 31–56.
Amihud, Y., Mendelson, H., 1986. Asset pricing and the bid-ask spread. Journal of Financial Economics 17, 223–249.
Andersen, T.G., Bollerslev, T., Diebold, F.X., Ebens, H., 2001. The distribution of realized stock return volatility. Journal of Financial
Economics 61, 43–76.
Ball, C., Chordia, T., 2001. True spreads and equilibrium prices. Journal of Finance 56, 1801–1835.
Barberis, N., Shleifer, A., Vishny, R., 1998. A model of investor sentiment. Journal of Financial Economics 49, 307–343.
Bessembinder, H., 2003. Trade execution costs and market quality after decimalization. Journal of Financial and Quantitative Analysis 38,
747–777.
Blume, M., MacKinlay, A., Terker, B., 1989. Order imbalances and stock price movements on October 19 and 20, 1987. Journal of
Finance 44, 827–848.
Boehmer, E., Kelley, E., Pirinsky, C., 2005. Instititutional investors and the informational efficiency of prices. working paper, Texas A&M
University.
Brennan, M., Jegadeesh, N., Swaminathan, B., 1993. Investment analysis and the adjustment of stock prices to common information.
Review of Financial Studies 6, 799–824.
Brennan, M., Subrahmanyam, A., 1998. The determinants of average trade size. Journal of Business 71, 1–25.
Brown, P., Walsh, D., Yuen, A., 1997. The interaction between order imbalance and stock price. Pacific-Basin Finance Journal 5, 539–557.
Chan, K., Fong, W., 2000. Trade size, order imbalance, and the volatility–volume relation. Journal Of Financial Economics 57, 247–273.
Chordia, T., Roll, R., Subrahmanyam, A., 2000. Commonality in liquidity. Journal of Financial Economics 56, 3–28.
Chordia, T., Roll, R., Subrahmanyam, A., 2001. Market liquidity and trading activity. Journal of Finance 56, 501–530.
Chordia, T., Roll, R., Subrahmanyam, A., 2002. Order imbalance, liquidity and market returns. Journal of Financial Economics 65,
111–130.
Chordia, T., Roll, R., Subrahmanyam, A., 2005. Evidence on the speed of convergence to market efficiency. Journal of Financial
Economics 76, 271–292.
Chordia, T., Sarkar, A., Subrahmanyam, A., 2005. An empirical analysis of stock and bond market liquidity. Review of Financial Studies
18, 85–129.
Chordia, T., Subrahmanyam, A., 1995. Market making, the tick size, and payment-for-order flow: theory and evidence. Journal of
Business 68, 543–575.
Chordia, T., Subrahmanyam, A., 2004. Order imbalance and individual stock returns: Theory and evidence. Journal of Financial
Economics 72, 485–518.
Christie, W., Schultz, P., 1994. Why do NASDAQ market makers avoid odd-eighth quotes? Journal of Finance 49, 1813–1840.
Cox, D., Peterson, D., 1994. Stock returns following large one-day declines: evidence on short-term reversals and longer-term
performance. Journal of Finance 49, 255–267.
 ARTICLE IN PRESS
268 T. Chordia et al. / Journal of Financial Economics 87 (2008) 249–268

Cushing, D., Madhavan, A., 2000. Stock returns and trading at the close. Journal of Financial Markets 3, 45–67.
Daniel, K., Hirshleifer, D., Subrahmanyam, A., 1998. Investor psychology and security market under- and overreactions. Journal of
Finance 53, 1839–1885.
Epps, T., 1979. Comovements in stock prices in the very short run. Journal of the American Statistical Association 74, 291–298.
Fama, E., 1970. Efficient capital markets: a review of theory and empirical work. Journal of Finance 25, 383–417.
French, K., Roll, R., 1986. Stock return variances: the arrival of information and the reaction of traders. Journal of Financial Economics
17, 5–26.
Gervais, S., Kaniel, R., Mingelgrin, D., 2001. The high-volume return premium. Journal of Finance 56, 877–919.
Glosten, L., Milgrom, P., 1985. Bid, ask and transaction prices in a specialist market with heterogeneously informed traders. Journal of
Financial Economics 14, 71–100.
Griffin, J., Nardari, F., Stulz, R., 2007. Do investors trade more when stocks have performed well? Evidence from 46 countries. Review of
Financial Studies 20, 905–951.
Grossman, S., Miller, M., 1988. Liquidity and market structure. Journal of Finance 43, 617–633.
Hasbrouck, J., Seppi, D., 2001. Common factors in prices, order flows and liquidity. Journal of Financial Economics 59, 383–411.
Hillmer, S., Yu, P., 1979. The market speed of adjustment to new information. Journal of Financial Economics 7, 321–345.
Huang, R., Stoll, H., 1994. Market microstructure and stock return predictions. Review of Financial Studies 7, 179–213.
Jacoby, G., Fowler, D., Gottesman, A., 2000. The capital asset pricing model and the liquidity effect: a theoretical approach. Journal of
Financial Markets 3, 69–81.
Jones, C., 2001. A century of stock market liquidity and trading costs. working paper, Columbia University, New York, NY.
Jones, C., Lipson, M., 2001. Sixteenths: Direct evidence on institutional execution costs. Journal of Financial Economics 59, 253–278.
Kyle, A., 1985. Continuous auctions and insider trading. Econometrica 53, 1315–1335.
Lauterbach, B., Ben-Zion, U., 1993. Stock market crashes and the performance of circuit breakers: empirical evidence. Journal of Finance
48, 1909–1925.
Lee, C., 1992. Earnings news and small traders: an intraday analysis. Journal of Accounting and Economics 15, 265–302.
Lee, C., Ready, M., 1991. Inferring trade direction from intraday data. Journal of Finance 46, 733–747.
Lo, A., MacKinlay, C., 1990. When are contrarian profits due to stock market overreaction? Review of Financial Studies 3, 175–205.
Madhavan, A., Richardson, M., Roomans, M., 2002. Why do security prices change? A transaction-level analysis of NYSE stocks. Review
of Financial Studies 10, 1035–1064.
Pastor, L., Stambaugh, R., 2003. Liquidity risk and expected stock returns. Journal of Political Economy 113, 642–685.
Patell, J., Wolfson, M., 1984. The intraday speed of adjustment of stock prices to earnings and dividend announcements. Journal of
Financial Economics 13, 223–252.
Peterson, M., Sirri, E., 2002. Order submission strategy and the curious case of marketable limit orders. Journal of Financial and
Quantitative Analysis 37, 221–241.
Roll, R., Schwartz, E., Subrahmanyam, A., 2007. Liquidity and the law of one price: the case of the futures/cash basis. Journal of Finance
62, 2201–2234.
Sias, R., 1997. Price pressure and the role of institutional investors in closed-end funds. Journal of Financial Research 20, 211–229.
Stoll, H., 1978. The supply of dealer services in securities markets. Journal of Finance 33, 1133–1151.
Subrahmanyam, A., 1994. Circuit breakers and market volatility: a theoretical perspective. Journal of Finance 49, 237–254.