You are on page 1of 12

International Review of Financial Analysis 19 (2010) 181–192

Contents lists available at ScienceDirect

International Review of Financial Analysis

New evidence on the relation between stock liquidity and measures of


trading activity
Daniel Chai a, Robert Faff b,⁎, Philip Gharghori a
a
Department of Accounting and Finance, Monash University, Clayton, Victoria 3800, Australia
b
UQ Business School, University of Queensland, St Lucia, Queensland, 4072, Australia

a r t i c l e i n f o a b s t r a c t

Article history: The goal of this paper is to examine two empirical issues regarding stock liquidity: (1) to what degree are
Received 19 February 2009 different liquidity proxies correlated? and (2) how are different liquidity proxies related to stocks' trading
Received in revised form 10 February 2010 characteristics? Answers to these questions will help us better understand whether there are common
Accepted 28 February 2010
sources of liquidity. This has considerable implications for studying stock liquidity, since selecting an
Available online 11 March 2010
appropriate proxy for liquidity is an important issue in empirical research design. Using data from the
JEL classification:
Australian equity market, our results confirm prior research that stocks' trading characteristics are important
G10 determinants of liquidity. Though the relationships are generally consistent with expectations, some proxies
do react differently to certain trading characteristics. This finding is consistent with the contention that
Keywords: liquidity is a multifaceted concept and each alternative proxy may only capture a certain aspect of liquidity.
Stock liquidity © 2010 Elsevier Inc. All rights reserved.
Trading characteristics
Australian evidence

1. Introduction Australian trading mechanism is different from that of the US market


primarily because of the absence of market makers and the fact that
Liquidity has long been an important issue for securities traded in public limit orders provide liquidity to the market and establish the
financial markets. A certain level of liquidity is necessary for securities bid and ask prices. This market characteristic provides a more
to be traded in the quantities required in a timely fashion without any transparent trading environment to market participants, as they
price discount. What are the factors that influence stock liquidity? have the ability to observe recent trades. In recent years, order-driven
Early studies such as Tinic (1972), Benson and Hagerman (1974), and limited order book market structures have increased rapidly
Branch and Freed (1977) and Stoll (1978) investigate the influence of because of improvements in information technology and financial
both stock and market characteristics on bid–ask spreads, which is market deregulation. As noted in Brown and Zhang (1997), markets
often used as a proxy for the trading cost of immediate transactions that allow limit orders tend to have a lower execution-price risk and
(Demsetz, 1968). Stock characteristics include stock price, trading have a higher level of liquidity. The Australian market provides us the
volume, number of trades and return volatility. Market characteristics opportunity to examine the role of liquidity in financial markets
include market structure and competition. Stoll (2000) and Chordia, across different stock exchange mechanisms.
Roll and Subrahmanyam (2000) extend the early studies by employ- In this paper, we extend the literature by employing six different
ing various alternative proxies for liquidity. Similar to the previous liquidity proxies constructed from low-frequency data. The proxies are
studies, stocks' trading characteristics such as stock price, trading proportional bid–ask spread, stock turnover, the illiquidity ratio from
volume and volatility are found to be important determinants of the Amihud (2002), the return reversal measure from Pastor and
spread and other liquidity proxies. Stambaugh (2003), the zero return measure (proportion of zero daily
As with most of the finance literature, the role of stocks' trading returns) from Lesmond, Ogden and Trzcinka (1999), and turnover-
characteristics on liquidity has been focused on quote-driven markets, adjusted number of zero daily volumes from Liu (2006). These liquidity
namely, the NYSE, AMEX and Nasdaq. The purpose of this paper is to proxies are widely used in asset pricing research and each of them
fill the gap in the literature with respect to order-driven systems by represents different liquidity dimensions and trading behaviour. By
examining the relationship between stock liquidity and stocks' examining the relationships between these liquidity proxies and stocks'
trading characteristics in the order-driven market of Australia. The trading characteristics, the current research seeks to understand the
extent to which there are common determinants of stock liquidity.
Specifically, we ask how are these proxies related to each other and how
⁎ Corresponding author. are they related to stocks' trading characteristics? A finding that they are
E-mail address: r.faff@business.uq.edu.au (R. Faff). not related in the same way will support the view that liquidity is a

1057-5219/$ – see front matter © 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.irfa.2010.02.005
182 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

complex multidimensional concept. The results also help us to The reciprocal of turnover is often used to represent the average
understand the trading aspects that a particular proxy is capturing. holding period of securities (e.g. see Atkins and Dyl, 1997). The
The efforts are significant as liquidity plays an important role in asset smaller the turnover rate, the longer the average holding period.
pricing, and the selection of liquidity proxies in a research design would According to the clientele effect of Amihud and Mendelson (1986),
have considerable influence on empirical results. The aims of this paper assets with higher spreads have relatively longer expected holding
are in the spirit of Stoll (2000), who asserts that understanding the periods. Thus, turnover should be negatively related to spread and
source of liquidity is important for asset pricing research. should be positively related to liquidity. Datar et al. (1998) show that
When using these liquidity proxies, we have to be aware of their stock returns are strongly negatively related to the turnover ratio. In
limitations. Liu (2006) notes that existing liquidity proxies have a contrast, Lee and Swaminathan (2000) question the use of turnover as
limited ability to capture liquidity risk and they might be poor measures, a liquidity proxy. They find some evidence that turnover is not highly
even in the particular dimension they intend to capture. Moreover, correlated with firm size or the relative bid–ask spread. Further, they
liquidity proxies constructed from high-frequency (microstructure) show that turnover is related to stocks' past performance — stocks
data are more precise than those created from low-frequency data that are perceived to or have growth potential (growth stocks) have a
(Hasbrouck, 2009). Recently, Goyenko, Holden and Trzcinka (2009) higher turnover rate than stocks that are relatively undervalued
demonstrate that liquidity proxies constructed from low-frequency (value stocks). However, Chan and Faff (2003) find evidence against
data are comparable to high-frequency measures. Their results give us the value/growth argument. They find a strong negative relationship
some confidence on the use of the liquidity proxies employed in the between turnover and stock returns from 1990 to 1999 in the
current study. Australian market. The significant relationship remains even after
The remainder of the paper is organised as follows. Section 2 controlling for size, book-to-market, beta and momentum.
presents the literature review and statement of hypotheses. We first
introduce the liquidity proxies, then the theoretical background of the 2.1.2. Illiquidity ratio
relationship between trading characteristics and liquidity. Section 3 The illiquidity ratio, proposed by Amihud (2002), is defined as the
discusses the methodology and data sources. Section 4 presents the daily absolute return of a stock divided by its trading volume on
empirical results and Section 5 concludes. that day. The monthly illiquidity ratio is obtained from the following
equation:
2. Literature review and statement of hypotheses
D
2.1. Liquidity proxies 1 i;t
Illiquidi;t = ∑ jr j = vi;d;t ð2Þ
Di;t d = 1 i;d;t
Kyle (1985, p.1316) points out that “market liquidity is a slippery
and elusive concept, in part because it encompasses a number of
where ri,d,t is the absolute return for stock i on day d in month t, and vi,d,t
transactional properties of markets. These include tightness, depth, and
is the trading volume in millions of dollars for stock i on day d in month t
resiliency.” Tightness refers to the cost of transactions, such as the bid–
and D is the number of daily observations for stock i in month t.
ask spread. Depth represents the ability of the market to absorb a large
The illiquidity ratio captures the price movement associated with
quantity of trade without having a large impact on price. Resiliency is
trading volume or the price impact of the order flow. Stocks are
defined as the speed with which the prices bounce back to equilibrium
regarded as illiquid (liquid) if small (large) trades cause high (small)
following a large trade. Black (1971) earlier suggested another
price movement. The illiquidity ratio follows Kyle's (1985) concept of
dimension of liquidity, immediacy, which represents the speed with
liquidity, which examines the relationship between price movement
which buy or sell orders can be executed. Liquidity proxies used in the
and net order flow and is commonly known as price impact or Kyle's λ.
literature can therefore be classified into these four categories. While
Amihud (2002) demonstrates that the illiquidity ratio is positively
these dimensions are to some degree overlapping, there is no single
related to stock returns in the cross-section. The illiquidity ratio
liquidity proxy that fully captures all dimensions. This paper employs six
overcomes the shortcoming of stock turnover in that it incorporates
widely used liquidity proxies that are computed on a monthly basis
price movement.2 Hasbrouck (2009) compares various price impact
using both daily and monthly price information. The proxies are stock
measures constructed from daily and microstructure data. He finds
turnover, the illiquidity ratio, the return reversal measure, proportional
that the Amihud illiquidity ratio is most strongly correlated with
spread, the zero return measure and turnover-adjusted number of zero
microstructure-based price impact measures. Goyenko et al. (2009)
daily volumes. Each is discussed in turn below.
also find that the Amihud measure does a good job in capturing price
impact.
2.1.1. Stock turnover
Similar to other liquidity measures, the illiquidity ratio also has some
Stock turnover is the ratio of the number of shares traded to the
limitations. As Grossman and Miller (1988) point out, the illiquidity
number of shares outstanding:
ratio is usually obtained based on averaged price changes and averaged
trading volume from the past. Therefore, it cannot answer the question
turnoveri;t = VOLi;t = sharei;t ð1Þ of how the stock price is influenced by the sudden arrival of a large trade.
Further, it cannot distinguish whether the price fluctuations are due to
the lack of liquidity or the arrival of new information. A stock might
where VOLi,t is the total trading volume for stock i in month t and
experience large price fluctuations not because it is illiquid but because
sharei,t is the number of shares outstanding for stock i in month t.
new information arrives frequently. To make the liquidity proxies
Volume data for each stock is collected on a daily basis, while data on
consistent, we flip the sign of the illiquidity ratio so that it is a measure of
the number of shares outstanding is collected on a monthly basis.1
liquidity.
1
Our method of calculating stock turnover over a month is similar to that used in
Chan and Faff (2003) and Datar et al. (1998). An alternative specification of Eq. (1) is
2
to use daily shares outstanding. However, these data are not readily available to us Assume that stock A and stock B have the same turnover ratio over the same
over our sample period. We calculated an alternative measure of stock turnover by period, but stock A has a price movement of 3% and stock B has a price movement of
averaging the current and prior month's shares outstanding to assess the robustness of 6%. According to the turnover ratio, both stocks have the same level of liquidity.
our results. These unreported findings, which are available on request from the However, in reality stock A may be more liquid than stock B since for stock A, the same
authors, demonstrate that our inferences are robust to this alternative measure. level of trading volume induces lower price movement.
D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192 183

2.1.3. Return reversal measure who are better informed than they are. However, bid–ask spreads are
The return reversal measure, developed by Pastor and Stambaugh not special to quote-driven markets. Glosten (1994) shows that bid–
(2003), captures the price changes associated with order flow. ask spreads are positively associated with adverse selection costs in an
Following Pastor and Stambaugh (2003), the monthly return reversal order-driven trading environment, suggesting that the bid–ask spread
measure for security i is obtained by running the following OLS represents the cost of supplying immediacy. This finding is supported
regression: by Handa, Schwartz and Tiwari (1998) who claim that the bid–ask
spread is a natural property of an order-driven market because
e e
ri;t + 1 = γ0 + γ1 ri;t + λ½signðri;t Þ × voli;t  + εi;t ð3Þ market participants are willing to pay for price certainty.
Despite being a widely used liquidity proxy, the bid–ask spread
where rei,t+ 1 is the excess return with respect to the value-weighted has certain shortcomings. As noted in Grossman and Miller (1988),
market index return for firm i on day t + 1, ri,t is the return for firm i on buying and selling do not appear simultaneously but are randomly
day t, sign(rei,t) is the sign of the excess return with respect to the separated in time. The transaction price may occur outside or within
value-weighted market index return for firm i on day t, and voli,t is the the bid and ask prices. This is because when buying securities, limit
trading volume in millions of dollars for firm i on day t. The coefficient order customers may be willing to pay more than the bid price to
λ measures the expected return reversal for a given trading volume. minimise the price risk associated with waiting. Therefore, the spread
The return reversal measure is in line with the concept of the price cannot serve as a reliable measure of the trading cost. Brennan and
impact (resiliency) dimension of liquidity. The basic idea in this model Subrahmanyam (1996) also argue that the bid–ask spread is a noisy
is that an asset is considered liquid if it is able to absorb large amounts measure for the same reason. Further, large price stocks normally
of trading quickly without a big change in its price. In a quote-driven have wider spreads, but this does not really mean that they are
environment, the return reversal is a form of compensation to market illiquid. The proportional spread overcomes the problem that quoted
makers for bearing higher risk when facing selling (buying) pressures spread is an increasing function of stock price. For consistency, we flip
from liquidity or non-informational trades. Thus, lambda in Eq. (3) the sign of the proportional spread to make it a liquidity measure.
should be negative and larger in absolute value when liquidity is low.
However, the absence of dealers on the Australian Securities Exchange 2.1.5. Zero return measure
(ASX) suggests that the return reversal effect may be less relevant, The proportion of zero daily returns observed over the relevant
when compared to markets such as NYSE. month is calculated as:
When using the return reversal measure, we have to be cautious
about the role that volume can play in current and lagged returns. zeroi;t = zeroreturni;t = tradingdayi;t ð5Þ
Llorente, Michaely, Saar, and Wang (2002) show that stocks accompa-
nied with a high degree of information-motivated trades produce more
where zeroreturni,t is the number of zero daily return days for stock i in
volume-related return continuations. This return continuation would
month t, and tradingdayi,t is the number of trading days for stock i in
weaken the volume-related return reversal. Recently, Goyenko et al.
month t.
(2009) find that the return reversal measure has a low correlation with
Lesmond et al. (1999) argue that zero returns occur when the
microstructure-based price impact measures and sometimes has the
transaction cost threshold is not exceeded for marginal traders who
wrong sign. It is important to note that Pastor and Stambaugh (2003) do
may be informed or uninformed. For informed traders, they are likely
not apply the lambda directly for individual stocks, but use a portfolio
to reduce their desired trades or choose not to trade when
approach instead. Moreover, they employ data filtering rules that are
information is not sufficient to compensate the cost of transacting.
likely to reduce the occurrence of extreme values for their measure.
Therefore, there will be no price movement from the previous day. For
Pastor and Stambaugh (2003) exclude stocks with less than 15 days of
liquidity traders, they will generally choose not to trade if liquidity is
return observations in a given month. This criterion could exclude quite
low and the transaction costs are high. As a result, price will also
a few small firms in the Australian market. Therefore, we set a weaker
remain unchanged. Lesmond et al. (1999) show that the zero return
rule that firms only need to have at least 10 return observations in a
measure is negatively related with firm size and is positively related
given month.3
with spread measures, which is consistent with the transaction cost
effect on stock returns. Bekaert, Harvey and Lundblad (2007) find that
2.1.4. Proportional spread
the zero return measure is to some degree correlated with other
The proportional bid–ask spread for stock i in month t is given by:
liquidity measures such as stock turnover and bid–ask spreads across
D emerging markets. They also find evidence that this measure
1 i;t A B A B
pspreadi;t = ∑ ðp −pi;d;t Þ = ð0:5pi;d;t + 0:5pi;d;t Þ ð4Þ significantly predicts future returns in emerging equity markets.
Di;t d = 1 i;d;t
The zero return measure is readily obtainable since it only requires
daily stock price information. However, there are some limitations
where pAi,t (pBi,t) is the daily closing ask (bid) prices for stock i on day d
with the zero return measure. As noted in Bekaert et al. (2007), daily
in month t and D is the number of daily observations for stock i in
zero returns may occur because of a lack of information flow. Smaller
month t.
stocks may automatically show a higher level of non-trading
The proportional spread is a widely used measure of asset liquidity
compared to larger stocks. Further, the zero return measure ignores
in the literature. Early research focuses on quoted bid–ask spread
price fluctuation within a day, and therefore, it cannot truly represent
(difference between the ask price and the bid price), which is a direct
the trading behaviour of stocks. To convert the measure into a
measure of transaction costs. Microstructure theory decomposes the
liquidity proxy, it is modified as 1 − zeroi,t.
spread into two components. The first component allows market
makers to generate revenue from order flow to cover inventory costs
2.1.6. Turnover-adjusted number of zero daily volumes
and order processing fees. The second is an adverse selection
Liu (2006) proposes a new liquidity measure that aims to capture
component, which arises because market makers are facing investors
multiple dimensions of liquidity and places a particular focus on
trading speed. The liquidity measure is defined as:

3
This filtering rule does not affect our results. The results are robust when either 5  
1 = ðturnoveri;t Þ 21
or 15 return observations are the minimums imposed for inclusion. This will be LMi;t = NoZVi;t + × ð6Þ
discussed further in Section 3. Deflator NoTDt
184 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

where NoZVi,t is the number of zero daily trading volumes for stock i in according to some optimal policy. Demsetz (1968) shows that the
month t; turnoveri,t is the stock turnover rate for stock i in month t difference between the bid and ask price (bid–ask spread) can
obtained from Eq. (1); NoTDt is the total number of trading days in the represent the trading cost for such transactions. Under this view, the
market in month t; and the deflator4 is set to 480,000 as suggested in spread serves as a (inverse) proxy for stock liquidity. If the spread is
Liu (2006). Volume data for each stock are collected on a daily basis. important, it is essential to understand the sources of the spread. After
NoZV serves as an indicator of liquidity — the lower the number of Demsetz (1968), studies such as Tinic (1972), Benson and Hagerman
zero daily trading volumes, the more frequent the trade and, thus, the (1974), Branch and Freed (1977), Stoll (1978) and Stoll and Whaley
more liquid the stock. It reflects the continuity of trading and potential (1983) examine variables that could influence the spread. The
delay in executing a trade (Liu, 2006). To some extent, the turnover variables can be divided into three categories: (1) variables that
component captures the notion of how much quantity has been have an effect on the cost of positioning for an individual security (i.e.
traded. It acts as a tiebreaker when two stocks have the same number trading characteristics), (2) variables associated with the structure of
of zero daily trading volumes. Therefore, stocks that LM classifies as the institutions that provide liquidity (e.g. capitalisation of the
most liquid are those that are traded frequently and have large portfolio and securities included in the portfolio), and (3) variables
turnover over the relevant month. Multiplication by the factor 21/ that have an influence on the profit margin of the specialists (e.g.
NoTD standardizes the number of trading days in a month to 21 and market structure and competition). The empirical results show that
therefore, makes LM comparable over time. variables related to stocks' trading characteristics are most influential.
In sum, LM uses the number of zero daily trading volumes to Recent studies such as Chordia, Roll and Subrahmanyam (2000) and
identify the liquidity of stocks, and then it relies on turnover to Stoll (2000) extend early studies by employing different alternative
distinguish between stocks that have the same level of liquidity as liquidity proxies. Using a cross-sectional regression approach, Chordia
classified by the number of zero daily volumes. The role of NoZV is also et al. (2000) find that variables such as stock price, volatility and trading
similar to the number of zero daily returns in Lesmond et al. (1999). volume are related to the quoted spread, the proportional spread,
Therefore, LM also reflects the trading cost dimension of liquidity. The quoted depth, the effective spread and the proportional effective spread.
turnover component in LM overcomes the potential size effect in the Except stock price, the influence of volume and volatility are consistent
number of zero daily trading volumes. However, it has a short- through different liquidity proxies.
coming similar to the turnover measure — it does not consider stock Stoll (2000) uses ten different liquidity proxies constructed from
price movement. For consistency, we flip the sign of LM so that it is a microstructure data to examine whether they are related in the same
measure of liquidity. way to stocks' trading characteristics. The purpose of his investigation
is to understand the common sources of different liquidity measures.
2.1.7. Summary of liquidity proxies His sample includes stocks from both the NYSE/AMEX and Nasdaq
In sum, each of the six liquidity proxies captures a different dimension over a three-month period from December 1997 to February 1998.
of liquidity. Proportional spread and the zero return measure can be The liquidity proxies include various spread measures, covariance of
categorised as measures of tightness since both proxies reflect trading price changes, price impact measures, and the volatility at the opening
costs. The illiquidity ratio represents the price movement associated with of the market. He first demonstrates that liquidity measures that
trading volume and therefore, it is related to depth/price impact. The capture different dimensions of liquidity are not strongly correlated.
return reversal measure captures resiliency. Finally, stock turnover and He then performs cross-sectional regressions having different
the turnover-adjusted number of zero daily volumes represent immedi- liquidity proxies regressed on stock price, return variance, trading
acy because both proxies reflect trading speed and trading frequency. volume and number of trades. These trading characteristics are
There is no consensus on which is the superior liquidity proxy. It is found to be important determinants of liquidity. However, not every
also important to note that while they are all plausible liquidity proxies, liquidity proxy is related in the same way to the estimated charac-
each of them captures a particular element of liquidity and trading teristics. The results imply that liquidity is a multifaceted concept.
behaviour. For example, Stoll (2000) points out that spreads capture real Moreover, the influence of stock price on price impact measures is
friction and price impact measures represent informational friction. different in the two markets. Stoll (2000) argues that this finding may
Further, since our liquidity proxies are obtained from low-frequency indicate that information is incorporated into stock prices differently
data instead of intraday data, they are likely to contain some noise. in the two markets.
Nevertheless, as the liquidity proxies that we employ address the four The underlying principle on the relationship between liquidity and
separate dimensions of liquidity, they can help us to understand the role stock characteristics is based on order execution and inventory
that liquidity plays in financial markets. control (Stoll, 2000). Large trading volume reduces the risk of carrying
In the subsequent analysis, we refer to stock turnover as inventory for a period of time, which should increase stock liquidity.
TURNOVER; the Amihud (2002) illiquid ratio as AMIHUD, the return Higher return volatility increases the risk of holding inventory, and it
reversal measure as PS; proportional bid–ask spread as PBA; the zero should have a negative effect on stock liquidity. Stock price controls
measure as ZERO; and the turnover-adjusted number of zero daily the effects of price discreteness and can be used as a proxy for risk, as
volumes proposed by Liu (2006) as LM. low price stocks tend to be riskier. The transaction costs of small size
firms should also be higher than for large size firms (Stoll and Whaley,
2.2. Determinants of liquidity 1983). Based on this reasoning, our testable hypotheses are:

Hypothesis 1a. Price per share is expected to be positively related to


In this section, we briefly review the literature to provide a context liquidity.
for the current study. Stock liquidity can be measured by the trading
cost for an immediate transaction. Traders need a price concession Hypothesis 1b. Trading volume is expected to be positively related to
when buying the asset immediately and they can then sell the asset liquidity.

Hypothesis 1c. Return volatility is expected to be negatively related


to liquidity.
4
The reciprocal of turnover produces a wide range of numbers that can be very
large or very small, depending on the magnitude of the monthly turnover rate. The
We refer to price per share at the end of each month as PRICE;
deflator is used to ensure that 0 b [(1/xmonthturnover)/deflator] b 1, so stocks that have return volatility of daily stock returns in each month as VARIANCE;
the same number of zero daily volumes (NoZV) can be further differentiated. and trading volume (aggregated in each month) as VOLUME.
D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192 185

2.3. Other potential determinants Finally, a stock must have at least 10 valid return observations in a
given month.7 Panel A of Table 1 shows market capitalisations and
We incorporate two additional variables into our analysis — coverage of our sample versus the population in the CRIF database.
absolute monthly stock return and the thin trading measure proposed From the table, on average, our sample covers more than 60% of total
by Beedles, Dodd and Officer (1988). stocks available from 1995 to 2005. The coverage is lower from 1991
Absolute monthly stock return (ABSR) can be treated as an to 1994 with less than 40% coverage in 1991 and 1992. As expected,
alternative measure of volatility and may be used as a measure of our data filtering rules have excluded quite a few small stocks.
information flow.5 The advantage of this measure is that it is simple to Compared with the CRIF sample, our minimum market capitalisation
calculate, particularly in comparison to conventional volatility mea- is higher and the median market capitalisation is more than double.
sures. Similar to volatility, ABSR should have a negative influence on Nevertheless, our sample is representative of the Australian market
liquidity. ABSR is measured as the absolute value of a stock's monthly given that, on average, it comprises more than 80% of the market's
return. total capitalisation.
Beedles et al. (1988) use the difference between the last price Month end bid and ask prices are not available for all stocks in our
date and last trading date in a month to create a crude proxy for sample. In order to calculate the spread, we revert to a subset of stocks
the proportion of missing daily returns. Specifically, the measure is that have bid–ask prices. Panel B of Table 1 shows the availability of
defined as: the bid–ask spread data. Comparing the number of companies in
Panel A and Panel B, the sample coverage is about 10 percentage
BEEDLES = f100−½100 = ðn + 1Þg = 100 ð7Þ points lower in the latter. In our empirical analysis, we use the sample
presented in Panel B of Table 1 only in spread related results.
where n is the difference in time (measured in days) between the last
3.2. Fama–MacBeth cross-sectional regressions
price date and last trading date in each month from the Centre for
Research in Finance database (CRIF). For example, if a stock is traded on
To examine whether trading characteristics are significant deter-
the last trading day of the month, n = 0 and BEEDLES = 0 (0%) — that is,
minants of liquidity, following Stoll (2000), we run cross-sectional
it is assumed that the stock traded every day. As another example, if the
regressions using monthly data according to:
stock's last trade was on the second last trading day, n = 1 and
BEEDLES = 0.5 (50%) — that is, it is assumed that the stock traded every K
second day. Liquidityjt = α0t + ∑ βkt TradingActivityjkt + εjt ð8Þ
Beedles et al. (1998) note that this measure is not an ideal proxy k=1

for liquidity. It is possible that a stock has no difference between the


last price date and the last trading date in a month throughout its where j = 1,2,…,N; t = 1,2,…,T; Liquidity is the liquidity proxy for
trading history. Nevertheless, they find that small size portfolios have stock j in month t; TradingActivity represents each trading charac-
a larger proportion of daily missing returns, indicating that small teristic variable such as stock price, volatility, and trading volume. The
stocks are less liquid. Their results show that liquidity is a partial research method involves three stages. In stage one, the dependent
explanation for the size effect. Since the BEEDLES measure aims to variable (each liquidity proxy) and independent variables (trading
capture the thin trading aspect of stock illiquidity, it should be characteristics) required for the cross-sectional regression are con-
negatively related with liquidity. Thus, out testable hypotheses for structed. We then estimate different versions of Eq. (8) each month
these two variables are: from January 1991 to September 2006 in stage two. Following Stoll
(2000) and Chordia et al. (2000), the regressions are estimated using
Hypothesis 1d. Absolute monthly return is expected to be negatively OLS. In stage three, we compute the average coefficient and variance
related to liquidity. from the time series of the cross-sectional regression coefficients
generated in stage two.
Hypothesis 1e. Beedles' thin trading measure is expected to be
We follow the Fama and MacBeth (1973) methodology and apply
negatively related to liquidity.
equal weight to all slope coefficients in estimating the average
coefficient.8 Chordia et al. (2000) note that the coefficients from the
3. Data and methodology
estimated regression are likely to be correlated across time if both
the dependent variable and independent variables are not returns.
3.1. Data sources and sample coverage

The analysis in this paper is carried out at the monthly level from
January 1991 to September 2006. The data come from two main sources.
Daily stock information such as stock price, trading volume and the
7
value-weighted market index are obtained from the Securities Industry We use this rule for three reasons. First, the return reversal measure requires at least four
return observations within a month. Second, we need about the same number of daily return
Research Centre of Asia-Pacific (SIRCA). Month end bid and ask prices observations in order to calculate the standard deviation (volatility) of a stock's return. Third,
are also obtained from SIRCA. Company information such as market the liquidity proxies can be estimated with less noise when a certain number of observations
capitalisation, number of shares outstanding and monthly stock return are required. However, as will be shown in Table 1, the filtering criteria skew our sample
is obtained from the Centre for Research in Finance (CRIF) database. towards larger firms. As a robustness check to ensure that our results are not influenced by
firm size, we impose a weaker restriction that stocks only need to have 5 valid return
In the calculation of the liquidity proxies, a daily observation is
observations in a month. This would allow us to include more small firms. As a further
excluded if the stock did not trade on the current or previous day.6 robustness check, we look at results in a sub-period (e.g. after year 1995) where the data
coverage is higher. Similar results are obtained under different filtering rules. Other data
filtering criteria such as the impact of low price stocks or IPOs were also considered.
Specifically, the influence of trading characteristics remain similar if we exclude stocks with
price below 10 cents or if stocks were IPOs for the calendar year. The results are available from
5
The idea is similar to using the absolute value of daily stock returns to proxy daily return the authors upon request.
8
volatility (e.g. Duffee, 1995; Ding, Granger and Engle, 1993; Chordia, Shivakumar and We chose the Fama–MacBeth procedure over the pooled time-series cross-section
Subrahmanyam, 2004). approach because the error terms in the pooled regression are likely to be cross-
6
As an alternative way to deal with the non-trading issue, we assign a zero return to sectionally correlated over time (Cochrane, 2001). The standard error will be biased
the current trading day if the stock did not trade on the previous day. The results are when errors are correlated. Petersen (2009) refers to this as a time effect and the
robust and are available upon request from the authors. Fama–MacBeth procedure is designed to address this effect.
186 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

Table 1
Comparison of market capitalisation and sample coverage. At the end of each year, we calculate summary statistics of stocks' market capitalisation in our sample and in the CRIF database.
Min, median and mean are in $millions. Max and total market capitalisation are in $billions. Panel A (B) presents the results for the sample without (with) bid–ask spread data.

Market capitalisation — sample Market capitalisation — all stocks (CRIF) Coverage percentage

Year Number of Min. Max. Median Mean Total Number of Min. Max. Median Mean Total Number of Market
firms ($M) ($B) ($M) ($M) ($B) firms ($M) ($B) ($M) ($M) ($B) firms cap.

Panel A: without bid–ask spread data


1991 302 1.36 21.51 72.55 618.43 186.77 926 0.05 21.55 8.98 267.10 247.34 32.61 75.51
1992 328 1.08 21.52 80.23 595.38 195.28 893 0.02 27.29 10.75 301.45 269.19 36.73 72.54
1993 565 1.45 29.12 50.92 532.76 301.01 1008 0.05 29.12 19.57 367.81 370.75 56.05 81.19
1994 446 2.12 33.60 94.62 594.16 265.00 1099 0.17 33.60 19.13 309.39 340.02 40.58 77.94
1995 580 1.69 37.03 58.46 563.76 326.98 1105 0.46 37.03 20.56 358.75 396.42 52.49 82.48
1996 750 0.72 35.60 49.24 552.19 414.14 1125 0.47 35.60 26.04 411.39 462.82 66.67 89.48
1997 621 1.63 30.26 73.42 734.38 456.05 1145 0.31 30.26 23.12 445.97 510.64 54.24 89.31
1998 626 0.65 35.62 83.76 898.36 562.37 1128 0.31 35.62 21.56 532.53 600.70 55.50 93.62
1999 841 1.59 53.17 54.47 764.07 642.58 1207 0.68 53.17 31.37 583.23 703.96 69.68 91.28
2000 862 1.15 43.70 42.02 764.40 658.91 1315 0.95 54.03 24.16 564.93 742.88 65.55 88.70
2001 772 0.92 49.74 56.35 922.05 711.82 1290 0.04 49.74 21.75 579.93 748.11 59.84 95.15
2002 726 0.89 48.18 62.08 916.78 665.58 1307 0.17 48.18 20.35 528.18 690.34 55.55 96.41
2003 965 1.45 45.76 57.36 803.16 775.05 1358 0.16 45.76 29.93 618.56 840.01 71.06 92.27
2004 1119 1.05 54.95 59.00 879.47 984.13 1488 0.58 54.95 33.88 701.55 1043.90 75.20 94.27
2005 1112 1.94 81.67 71.84 1038.59 1154.91 1614 0.65 81.67 34.95 763.19 1231.79 68.90 93.76

Panel B: with bid–ask spread data


1991 245 1.36 21.51 84.62 696.40 170.62 926 0.05 21.55 8.98 267.10 247.34 26.46 68.98
1992 268 1.08 21.52 91.28 654.79 175.48 893 0.02 27.29 10.75 301.45 269.19 30.01 65.19
1993 459 1.45 29.13 65.29 602.77 276.67 1008 0.05 29.12 19.57 367.81 370.75 45.54 74.62
1994 368 2.12 33.61 95.41 645.97 237.72 1099 0.17 33.60 19.13 309.39 340.02 33.48 69.91
1995 501 1.69 37.03 58.53 590.22 295.70 1105 0.46 37.03 20.56 358.75 396.42 45.34 74.59
1996 652 1.45 35.60 46.13 573.69 374.04 1125 0.47 35.60 26.04 411.39 462.82 57.96 80.82
1997 551 1.63 30.26 64.27 753.53 415.19 1145 0.31 30.26 23.12 445.97 510.64 48.12 81.31
1998 551 0.65 35.62 72.97 885.02 487.65 1128 0.31 35.62 21.56 532.53 600.70 48.85 81.18
1999 724 1.59 53.17 48.27 829.89 600.84 1207 0.68 53.17 31.37 583.23 703.96 59.98 85.35
2000 690 1.15 43.70 44.86 876.68 604.91 1315 0.95 54.03 24.16 564.93 742.88 52.47 81.43
2001 592 0.92 49.74 63.45 1047.32 620.01 1290 0.04 49.74 21.75 579.93 748.11 45.89 82.88
2002 544 0.89 48.18 67.86 1008.21 548.47 1307 0.17 48.18 20.35 528.18 690.34 41.62 79.45
2003 675 2.18 45.76 69.97 930.94 628.38 1358 0.16 45.76 29.93 618.56 840.01 49.71 74.81
2004 781 1.05 54.95 68.75 958.26 748.40 1488 0.58 54.95 33.88 701.55 1043.90 52.49 71.69
2005 774 2.53 81.67 79.24 1131.35 875.66 1614 0.65 81.67 34.95 763.19 1231.79 47.96 71.09

Cochrane (2001) also suggests that ignoring such correlation could have last price date and the last trading date. This difference is, however, zero
a large impact on the inferences made. Following Cochrane (2001), the for most of the stocks. Stocks are deemed to have serious non-
adjusted standard
pffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffierror is obtained by multiplying the standard error synchronous trading problems if the difference is greater than zero. The
by 1 + p = 1−p, where p is the first-order auto-correlation of the summary statistics reflect the nature of the BEEDLES measure. Both
estimated coefficients across the sample period. liquidity proxies and trading characteristic variables display consider-
able positive skewness with the sample means exceeding the medians.
Therefore, in our empirical analysis we take the logarithmic transfor-
4. Empirical results mation of PRICE and VOLUME. VARIANCE, ABSR and BEEDLES are not
transformed because they may be zero for some stocks.
4.1. Descriptive statistics

4.1.1. Basic results 4.1.2. Size groups


Panel A of Table 2 reports the time-series average of the cross- To understand the distributions of the liquidity proxies and trading
sectional statistics of the liquidity proxies and stocks' trading characteristics further, we separate the sample into size groups. In each
characteristics. Recall that we flipped the sign of AMIHUD, LM and month, stocks are first sorted based on their current month's market
PBA to make them liquidity measures. Therefore, a negative sign for capitalisation and are then separated equally into three size groups.
the statistics (except standard deviation) is expected for AMIHUD, LM Panels B, C and D of Table 2 report summary statistics for small,
and PBA. Among the six liquidity proxies, AMIHUD exhibits the largest medium and large size firms, respectively. In general, small firms have
standard deviation. PS has a wide range of values from positive to the lowest average price and the highest return volatility. The mean
negative. We expect the value of PS to be negative and larger in value of PBA, LM, AMIHUD and ZERO increases from the small firm to
absolute value when liquidity is low. This is due to the return reversal the large firm sample, which indicates that small firms are less liquid
effect — the greater the order flow the greater the change in the compared to medium-size and large firms. The mean value of PS does
subsequent expected return. TURNOVER, ZERO and PBA have a lower not exhibit a strong pattern across size groups. The small size group
variation compared to the other proxies. has the largest average TURNOVER and it decreases dramatically from
The lowest priced stock in our sample is less than 10 cents, while the small to medium-size firms. The difference is smaller between
highest is above 60 dollars. This suggests that our sample comprises medium-size and large firms. This finding is in contrast to the results
stocks across a wide range of the size spectrum. It is worth mentioning on the other liquidity proxies and could be explained by the results
that BEEDLES is zero in minimum, lower quartile, median and upper on VOLUME. On average, small and large firms have a higher mean,
quartile. Recall that BEEDLES is calculated as the difference between the lower quartile, median and upper quartile VOLUME than medium-
D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192 187

Table 2
Summary statistics. Mean, minimum, lower quartile, median, upper quartile, maximum and standard deviation are computed each month cross-sectionally and then averaged across
time from January 1991 to September 2006. AMIHUD is the liquidity ratio from Amihud (2002). PS is the return reversal measure from Pastor and Stambaugh (2003). TURNOVER is
the stock turnover rate. ZERO is the zero return measure from Lesmond et al. (1999), but is modified as (1 — the proportion of zero daily returns). LM is the turnover-adjusted zero
daily volumes from Liu (2006). PBA is the proportional bid–ask spread. PRICE is stock price at the end of each month. VARIANCE is the variance of stock returns. VOLUME is the
aggregate trading volume. BEEDLES is the thin trading measure proposed by Beedles et al. (1988). ABSR is the absolute stock monthly return. All the variables are computed monthly.
VOLUME is measured in $millions. VARIANCE is multiplied by 100. Stocks are sorted based on their current month's market capitalisation in each month and then allocated into three
equal groups: small, medium and large. Panel A presents the results for the entire sample whereas Panels B, C and D present the results for the small firm, medium firm and large firm
sample, respectively. The F-tests (Kruskal–Wallis (KW) tests) are based on the null hypothesis that the mean (median) values of indicated variables are equal across the three size
groups. P-values are reported for both the F and KW tests in Panel E.

Mean Min Lower quartile Median quartile Upper quartile Max Standard deviation

Panel A: all firms sample


(a) Liquidity proxies
AMIHUD − 1.7954 − 101.5758 − 1.5021 − 0.5063 − 0.1393 − 0.0013 5.7740
PS 0.0017 − 6.1719 − 0.0341 0.0001 0.0349 7.1513 0.5337
TURNOVER 0.0453 0.0001 0.0127 0.0253 0.0496 1.1577 0.0771
ZERO 0.6953 0.0672 0.5782 0.7148 0.8339 0.9997 0.1825
LM − 1.6424 − 9.5488 − 2.8411 − 0.4471 − 0.0367 − 0.0366 2.1404
PBA − 0.0380 − 0.3280 − 0.0518 − 0.0273 − 0.0139 − 0.0010 0.0364
(b) Trading characteristics
PRICE 2.0738 0.0083 0.2160 0.7511 2.1586 65.9903 4.4142
VARIANCE 0.3271 0.0006 0.0288 0.0889 0.2653 30.8573 1.6245
VOLUME 9.2401 0.0215 0.9462 2.5707 7.6052 360.9786 23.471
BEEDLES 0.0360 0.0000 0.0000 0.0000 0.0000 0.8961 0.1397
ABSR 0.1207 0.0000 0.0308 0.0730 0.1510 2.1485 0.1689

Panel B: small firms sub-sample


(a) Liquidity proxies
AMIHUD − 2.3620 − 80.4151 − 2.0294 − 0.8491 − 0.3511 − 0.0142 6.7288
PS 0.0017 − 3.4636 − 0.0695 0.0005 0.0712 3.4516 0.4938
TURNOVER 0.0665 0.0031 0.0208 0.0388 0.0753 1.0462 0.1013
ZERO 0.6049 0.0967 0.4911 0.6175 0.7324 0.9775 0.1737
LM − 2.8784 − 9.2273 − 4.5741 − 2.5626 − 0.8673 − 0.0366 2.2809
PBA − 0.0702 − 0.3245 − 0.0843 − 0.0610 − 0.0442 − 0.0149 0.0417
(b) Trading characteristics
PRICE 0.2490 0.0085 0.0736 0.1506 0.2922 4.0553 0.3761
VARIANCE 0.6749 0.0067 0.1614 0.3148 0.6137 22.1583 1.9321
VOLUME 6.9347 0.0830 1.0396 2.3370 5.6645 208.5090 18.1314
BEEDLES 0.0702 0.0000 0.0000 0.0000 0.0106 0.8347 0.1863
ABSR 0.1785 0.0000 0.0578 0.1250 0.2309 1.6503 0.2031

Panel C: medium-size firms sub-sample


(a) Liquidity proxies
AMIHUD − 1.9128 − 48.8644 − 1.8415 − 0.7317 − 0.2793 − 0.0147 4.4204
PS 0.0014 − 2.7278 − 0.0501 0.0000 0.0493 3.1471 0.4091
TURNOVER 0.0393 0.0013 0.0105 0.0209 0.0427 0.6605 0.0647
ZERO 0.6735 0.1420 0.5767 0.6909 0.7908 0.9823 0.1606
LM − 1.5560 − 8.4810 − 2.4928 − 0.6508 − 0.0367 − 0.0366 1.9767
PBA − 0.0311 − 0.1359 − 0.0379 − 0.0267 − 0.0192 − 0.0073 0.0180
(b) Trading characteristics
PRICE 1.1602 0.0518 0.3976 0.7680 1.4076 12.6119 1.3726
VARIANCE 0.2390 0.0019 0.0379 0.0875 0.1876 11.1451 0.9867
VOLUME 6.1721 0.0432 0.6335 1.6396 4.6953 225.2294 18.7033
BEEDLES 0.0298 0.0000 0.0000 0.0000 0.0000 0.7693 0.1242
ABSR 0.1142 0.0000 0.0312 0.0713 0.1428 1.5104 0.1585

Panel D: Large firms sub-sample


(a) Liquidity proxies
AMIHUD − 1.1090 − 55.9482 − 0.5179 − 0.1248 − 0.0312 − 0.0021 4.5980
PS 0.0019 − 4.2067 − 0.0103 0.0001 0.0104 4.8410 0.5253
TURNOVER 0.0299 0.0001 0.0113 0.0232 0.0378 0.3115 0.0326
ZERO 0.8078 0.2541 0.7268 0.8342 0.9176 0.9995 0.1426
LM − 0.4887 − 7.8230 − 0.1245 − 0.0367 − 0.0367 − 0.0366 1.2443
PBA − 0.0127 − 0.0727 − 0.0163 − 0.0103 − 0.0059 − 0.0010 0.0102
(b) Trading characteristics
PRICE 4.8355 0.3371 1.5668 2.8399 5.3850 65.6813 6.6372
VARIANCE 0.0619 0.0016 0.0147 0.0269 0.0514 3.0517 0.2403
VOLUME 14.5892 0.0273 1.6569 5.0866 16.1032 250.1999 26.6769
BEEDLES 0.0068 0.0000 0.0000 0.0000 0.0000 0.5950 0.0572
ABSR 0.0691 0.0000 0.0217 0.0479 0.0896 0.7547 0.0835

Panel E: P-values for tests of equality of means and medians

F-tests KW
KW tests
tests

(a) Liquidity proxies


AMIHUD 0.00** 0.00**
PS 0.99 0.91
TURNOVER 0.00** 0.00**

(continued on next page)


188 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

Table 2 (continued)
Panel E: P-values for tests of equality of means
Mean Min and medians Lower quartile Median quartile Upper quartile Max Standard deviation
F-tests KW tests
(a) Liquidity proxies
ZERO 0.00** 0.00**
LM 0.00** 0.00**
PBA 0.00** 0.00**
(b) Trading characteristics
PRICE 0.00** 0.00**
VARIANCE 0.00** 0.00**
VOLUME 0.00** 0.00**
BEEDLES 0.00** NA
ABSR 0.00** 0.00**

** and * denote significance at the 1% and 5% levels, respectively.

sized firms do. This suggests that small and large firms are traded report that the characteristic of their liquidity measure is sensitive to the
more actively than medium-size firms are. Small firms generally have specification of the model given in Eq. (3).11 Further, Panel A shows that
a lower total number of shares outstanding. As a result, high turnover LM tends to have better correlations with all the other proxies. This
rates are observed among small firms. Since trading volume plays an finding is consistent with the intuition that LM captures multiple
important role in estimating TURNOVER and PS, it is not surprising to dimensions of liquidity (Liu, 2006).
see a non-linear pattern across size groups. The relationship between Consistent with the findings in Table 2, though small in magnitude,
size and TURNOVER is also similar to those found in prior Australian there is a negative relationship between TURNOVER and SIZE. This result
studies such as Chan and Faff (2003). Chan and Faff (2003) find that is similar to Chan and Faff (2003), who find that there is no relationship
size is not correlated with turnover. between stock turnover and size, and the correlation is −0.064 in the
To compare the difference in values formally, we conduct F-tests cross-section in the Australian market. VOLUME is positively related
(Kruskal–Wallis (KW) tests) to examine whether the mean (median) with all liquidity proxies except for PS. As expected, PRICE is highly
values of an indicated variable are statistically different across the correlated with SIZE and VARIANCE is correlated with ABSR.
three size groups. The p-values of these two tests are reported in Panel Table 2 suggests that firm size has some influence on the relationship
E. The null hypothesis of equal means (medians) is rejected for all the between liquidity and trading characteristics. Accordingly, we consider
liquidity proxies except PS, which provides evidence that liquidity to what extent the correlations are impacted by different size groups.
varies across the size groups. Regarding trading characteristics, there Panels B, C and D of Table 3 report the cross-sectional correlations for the
is also evidence that the mean and median values vary with firm size. small, medium-size and large firms, respectively. The correlations are
On average, large firms have higher stock prices, larger trading generally consistent with the results in Panel A. Both AMIHUD and
volumes and lower return volatility. It should be noted that we cannot TURNOVER have stronger correlations with other liquidity proxies (and
perform the KW tests on BEEDLES because it has zero median values size) in each size group. Comparing across the size groups, the
in each of the size groups. In sum, results in Table 2 support the view correlations are stronger between liquidity proxies for larger firms.
that smaller firms are less liquid when compared to larger firms. This result implies that large firms are priced more efficiently than
medium-size and small firms. Moreover, the weak correlations between
4.2. Correlations the liquidity proxies observed in Panel A might be influenced by the
pattern of trading volume across the size groups as shown in Table 2. Our
In Table 3, we report the average monthly cross-sectional correlations results also suggest that while these proxies all capture liquidity to some
between the liquidity proxies and trading characteristics, covering the extent, they do not necessarily behave similarly because they capture
period 1991 to 2006. In Panel A, the correlations are shown for the full different dimensions of liquidity.
sample. We expect that the correlations of liquidity proxies that capture
similar trading behaviours would be stronger. Table 3 shows that the 4.3. Regression results
correlations between liquidity proxies are generally low in the cross-
section except for LM with PBA, ZERO with LM, and ZERO with PBA. Both 4.3.1. Basic results
ZERO and PBA capture a similar dimension (tightness) of liquidity so their Table 3 shows that the correlations among the employed liquidity
higher correlation is not surprising. The component of zero trading proxies are generally low suggesting that the cross-sectional behaviour
volumes in LM is also similar to the concept of ZERO so again, the higher of these proxies is quite divergent. In this section, we run cross-sectional
correlation is expected. These three liquidity proxies also exhibit strong regressions to understand the sources of liquidity more fully. The
correlations with SIZE. regression results are presented in Table 4. According to our hypotheses,
Except for LM, both AMIHUD and TURNOVER have low correlations we expect that stock price and trading volume (volatility) are positively
with the other proxies. The weak correlation between AMIHUD and (negatively) related with stock liquidity.12 From Table 4, except for PS,
TURNOVER is similar to Sadka (2006), who finds that stock turnover has trading characteristics are significantly related with the liquidity
a low correlation with price impact measures and firm size in the US proxies. Their influences are generally consistent with the hypotheses
market, indicating that volume may not always be a good proxy for and their impacts are large and highly significant. The impact of the
liquidity.9 PS is generally uncorrelated with the liquidity proxies used in trading characteristics on PBA and LM is consistent with our hypotheses.
our sample. This finding is also similar to Eckbo and Norli (2002) and Firms with large prices, large trading volumes and low volatilities tend
Goyenko et al. (2009) in the US setting.10 Pastor and Stambaugh (2003)
11
Pastor and Stambaugh (2003) point out that one can consider 24 alternative
specifications of their model. For example, the dependent variable can be either the
excess or the total stock return. They report that the correlations between liquidity
9
Sadka (2006) finds that the cross-sectional correlation between both the Amihud series generated from different specifications are low.
12
illiquidity ratio and firm size with stock turnover is close to zero. We do not include size in our regression analysis because it is highly correlated
10
For example, Eckbo and Norli (2002) find the average time-series correlation with stock price. Recall that we modified ZERO and flipped the sign of AMIHUD, LM
across firms is 0.019 between PS and stock turnover and − 0.023 between PS and and PBA to make them liquidity measures. The results are interpreted based on the
proportional quoted spread. modified liquidity proxies.
D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192 189

Table 3
Correlation matrix of liquidity proxies and trading characteristics. AMIHUD is the liquidity ratio from Amihud (2002). PS is the return reversal measure from Pastor and Stambaugh
(2003). TO is the stock turnover rate. ZERO is the zero return measure from Lesmond et al. (1999), but is modified as (1 — the proportion of zero daily returns). LM is the turnover-
adjusted zero daily volumes from Liu (2006). PBA is the proportional bid–ask spread. The sign of AMIHUD, LM and PBA was flipped to make them represent liquidity. PRICE is stock
price at the end of each month. VARIANCE is the variance of daily stock returns. VOLUME is the aggregate trading volume. BEEDLES is the thin trading measure proposed by Beedles
et al. (1988). ABSR is the absolute stock monthly return. All the variables are computed monthly. VOLUME is measured in $millions. PRICE, VOLUME and SIZE are logarithmic
transformed. The correlations reported in the table are the time-series average of the monthly cross-sectional correlations. Stocks are sorted based on their current month's market
capitalisation in each month and then allocated into three equal groups: small, medium and large. Panel A presents the correlations for the entire sample whereas Panels B, C and D
present the correlations for the small firm, medium firm and large firm sample, respectively.

AMIHUD PS TO ZERO LM PBA PRICE VARIANCE VOLUME SIZE ABSR BEEDLES

Panel A: all firms sample


AMIHUD 1
PS − 0.0313 1
TO 0.0981 0.0003 1
ZERO 0.0413 0.0082 0.0505 1
LM 0.2793 0.0023 0.1017 0.3593 1
PBA 0.1974 −0.0076 −0.0737 0.4170 0.5338 1
PRICE 0.1534 −0.0116 −0.0449 0.3483 0.4272 0.6960 1
VARIANCE − 0.0967 0.0034 0.2487 − 0.0762 − 0.1740 −0.5365 −0.3837 1
VOLUME 0.3558 0.0001 0.4331 0.2313 0.4761 0.1529 −0.1465 0.1050 1
SIZE 0.1001 −0.0011 −0.2011 0.5047 0.4703 0.6851 0.8319 − 0.3279 0.2689 1
ABSR − 0.0258 0.0063 0.3754 0.0171 − 0.0822 −0.3038 −0.2791 0.3776 0.1306 −0.2869 1
BEEDLES − 0.0861 0.0022 −0.0328 − 0.1430 − 0.3635 −0.1987 −0.1357 0.0639 − 0.1496 −0.1805 0.0086 1

Panel B: small firms sub-sample


AMIHUD 1
PS − 0.0084 1
TO 0.1284 −0.0009 1
ZERO − 0.0848 0.0151 0.1566 1
LM 0.2091 0.0036 0.2595 0.1417 1
PBA 0.1884 −0.0074 0.0844 0.1907 0.3741 1
PRICE − 0.1495 0.0026 −0.0286 0.3123 − 0.0006 0.4541 1
VARIANCE − 0.1142 0.0067 0.1813 0.0585 − 0.0647 −0.4997 −0.3078 1
VOLUME 0.3285 0.0004 0.5412 − 0.0148 0.4658 −0.0072 −0.6127 0.2257 1
SIZE 0.0630 0.0052 −0.1033 0.1618 0.2915 0.5092 0.4956 − 0.2525 − 0.0506 1
ABSR − 0.0070 0.0124 0.3183 0.1623 0.0442 −0.1837 −0.0685 0.3164 0.2122 −0.1067 1

Panel C: medium-size firms sub-sample


AMIHUD 1
PS − 0.0213 1
TO 0.1745 0.0026 1
ZERO − 0.0143 0.0200 0.1805 1
LM 0.3285 0.0135 0.2398 0.2241 1
PBA 0.2114 −0.0121 0.1154 0.1571 0.3423 1
PRICE − 0.2305 0.0032 −0.1439 0.1781 − 0.1744 0.4176 1
VARIANCE − 0.0193 0.0091 0.2881 0.1552 0.0380 −0.4110 −0.3342 1
VOLUME 0.4075 0.0019 0.5957 0.1156 0.5261 0.0127 −0.6433 0.3053 1
SIZE 0.0145 −0.0019 −0.1117 0.1280 0.1310 0.3774 0.4756 − 0.2094 − 0.0524 1
ABSR 0.0388 0.0103 0.3753 0.2018 0.0972 −0.2000 −0.1589 0.4778 0.2934 −0.1156 1

Panel D: large firms sub-sample


AMIHUD 1
PS − 0.0274 1
TO 0.2082 0.0031 1
ZERO 0.1219 0.0009 0.2247 1
LM 0.4511 −0.0010 0.2072 0.3034 1
PBA 0.4558 0.0037 0.3259 0.4441 0.5446 1
PRICE − 0.1585 0.0045 −0.0560 0.3943 − 0.0503 0.3584 1
VARIANCE − 0.1082 −0.0086 0.2129 0.0780 − 0.0531 −0.3277 −0.2292 1
VOLUME 0.4663 0.0053 0.5940 0.3452 0.5090 0.5527 −0.1719 0.0279 1
SIZE 0.0554 0.0022 0.0906 0.4718 0.1623 0.4983 0.5647 − 0.1786 0.4913 1
ABSR − 0.0351 −0.0076 0.2595 0.1135 − 0.0039 −0.1575 −0.0974 0.4729 0.0585 −0.1088 1

to have higher PBA and LM. The result for PBA is also consistent with associated with trading volume is larger when the prices are large. A
Chordia et al. (2000), Stoll (2000) and Aitken and Frino (1996). possible explanation is as follows. In Table 2, we find some evidence that
For ZERO, the sign of the coefficients is consistent with our trading volumes are higher for small and large firms (as opposed to
hypotheses except for VARIANCE and ABSR. Our results indicate that medium-size firms). If a large stock experiences a large price movement,
stocks with higher return fluctuations tend to have a lower proportion of given the same level of trading volume, it would have a larger price
zero daily returns (i.e. higher liquidity). In general, the price movement impact. Our finding is similar to that in Stoll (2000). Stoll (2000) finds that
for a stock will be low if its proportion of zero daily returns is high and so firms with large prices tend to have higher price impacts on the Nasdaq.
our findings are not surprising. ABSR has the same influence since it is an He argues that this result may be influenced by the information that is
alternative measure of volatility. Moreover, firms with large prices and incorporated into price.
large trading volumes tend to have fewer zero daily returns. TURNOVER displays a somewhat different cross-sectional pattern
The positive (negative) coefficient of VOLUME (VARIANCE) in the than the other liquidity measures. The results show that firms with
AMIHUD regressions indicates that price impacts are smaller for firms smaller prices and higher volatilities tend to have higher turnover rates.
with large volumes and they are larger when the volatility is high. This These results are not surprising given the findings in Table 2. Small firms
finding is consistent with our hypotheses and Stoll (2000). The negative have higher TURNOVER than medium-size and large firms. The
coefficient of PRICE is puzzling. It indicates that the price movement relationship between TURNOVER and BEELDES may look odd at first
190 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

Table 4
Fama–MacBeth cross-sectional regressions — full sample. Individual stock liquidity proxies are regressed cross-sectionally each month on the concurrent month's closing price
(PRICE), the variance of daily returns for the concurrent month (VARIANCE), monthly aggregate dollar trading volume (VOLUME), monthly absolute stock return (ABSR) and
Beedles' thin trading measure (BEEDLES). PRICE and VOLUME are logarithmic transformed. AMIHUD is the liquidity ratio from Amihud (2002). PS is the return reversal measure from
Pastor and Stambaugh (2003). TURNOVER is the stock turnover rate. ZERO is the zero return measure from Lesmond et al. (1999), but is modified as (1 — the proportion of zero daily
returns). LM is the turnover-adjusted zero daily volumes from Liu (2006). PBA is the proportional bid–ask spread. The sign of AMIHUD, LM and PBA was flipped to make them
represent liquidity. The coefficients are averaged across 189 trading months in the sample. T-statistics are adjusted for first-order auto-correlation and are reported in parentheses
below the coefficient estimates.

Intercept PRICE VARIANCE VOLUME ABSR BEEDLES Adj. R2

Hypothesised sign + − + − −

AMIHUD − 20.2155 − 0.2955 − 158.1188 2.9346 0.1612


(− 38.14)** (− 6.10)** (− 8.44)** (35.58)**
− 19.9357 − 0.4005 − 159.5008 2.9145 − 1.1611 − 1.1331 0.1665
(− 37.05)** (− 8.86)** (− 8.00)** (34.76)** (− 5.12)** (− 5.74)**
PS 0.0311 0.0075 0.8012 − 0.0050 0.0125
(0.61) (1.43) (1.10) (− 0.67)
0.0299 0.0087 0.3065 − 0.0053 0.0376 0.0051 0.0163
(0.59) (1.46) (0.41) (− 0.69) (1.97)* (0.22)
ZERO 0.1934 0.1626 4.7084 0.0817 0.3871
(2.57)** (7.21)** (10.85)** (7.73)**
0.2019 0.1669 3.8426 0.0785 0.1445 − 0.0302 0.4005
(2.83)** (6.93)** (9.31)** (7.44)** (19.04)** (− 7.76)**
TURNOVER − 0.2445 − 0.0065 2.6707 0.0438 0.3897
(− 9.30)** (− 3.93)** (8.08)* * (10.27)**
− 0.2340 − 0.0032 1.4173 0.0403 0.1258 0.0063 0.3413
(− 9.91)** (− 1.79) (6.21)** (11.00)** (14.79)** (4.20)**
LM − 12.367 1.1055 − 37.3496 1.7250 0.3799
(− 40.29)** (25.60)** (− 11.80)** (35.43)**
− 11.4635 0.978 − 34.5721 1.6074 − 0.2549 − 3.7646 0.4506
(− 37.96)** (27.18)** (− 11.18)** (34.05)** (− 4.36)** (− 60.22)**
PBA − 0.1142 0.0309 − 2.7036 0.0135 0.6536
(− 33.69)** (25.61)** (− 11.57)** (25.88)**
− 0.1100 0.0296 − 2.6774 0.0131 − 0.0135 − 0.0168 0.6676
(− 31.57)** (25.00)** (− 12.01)** (24.40)** (− 6.77)** (− 9.91)**

** and * denote significance at the 1% and 5% levels, respectively.

Table 5 Table 6
Fama–MacBeth cross-sectional regressions — small firm sample. Individual stock liquidity Fama–MacBeth cross-sectional regressions — medium-size firm sample. Individual stock
proxies are regressed cross-sectionally each month on the concurrent month's closing price liquidity proxies are regressed cross-sectionally each month on the concurrent month's
(PRICE), the variance of daily returns for the concurrent month (VARIANCE), monthly closing price (PRICE), the variance of daily returns for the concurrent month (VARIANCE),
aggregate dollar trading volume (VOLUME) and monthly absolute stock return (ABSR). PRICE monthly aggregate dollar trading volume (VOLUME) and monthly absolute stock return
and VOLUME are logarithmic transformed. AMIHUD is the liquidity ratio from Amihud (ABSR). PRICE and VOLUME are logarithmic transformed. AMIHUD is the liquidity ratio from
(2002). PS is the return reversal measure from Pastor and Stambaugh (2003). TURNOVER is Amihud (2002). PS is the return reversal measure from Pastor and Stambaugh (2003).
the stock turnover rate. ZERO is the zero return measure from Lesmond et al. (1999), but is TURNOVER is the stock turnover rate. ZERO is the zero return measure from Lesmond et al.
modified as (1 — the proportion of zero daily returns). LM is the turnover-adjusted zero daily (1999), but is modified as (1 — the proportion of zero daily returns). LM is the turnover-
volumes from Liu (2006). PBA is the proportional bid–ask spread. The sign of AMIHUD, LM adjusted zero daily volumes from Liu (2006). PBA is the proportional bid–ask spread. The sign
and PBA was flipped to make them represent liquidity. The coefficients are averaged across of AMIHUD, LM and PBA was flipped to make them represent liquidity. The coefficients are
189 trading months in the sample. T-statistics are adjusted for first-order auto-correlation averaged across 189 trading months in the sample. T-statistics are adjusted for first-order
and are reported in parentheses below the coefficient estimates. auto-correlation and are reported in parentheses below the coefficient estimates.

Intercept PRICE VARIANCE VOLUME ABSR Adj. R2 Intercept PRICE VARIANCE VOLUME ABSR Adj. R2

Hypothesised sign + − + − Hypothesised sign + − + −

AMIHUD − 26.5802 0.4805 −156.7351 4.0127 0.1760 AMIHUD − 21.1798 0.1176 −310.7806 3.1992 0.1992
(− 26.96)** (4.02)** (−7.98)** (24.55)** (− 30.83)** (0.99) (−12.92)** (28.81)**
− 26.8475 0.5101 −156.7451 4.0822 −0.9152 0.1783 − 21.519 0.1427 −306.9715 3.2738 −1.2216 0.2016
(− 27.40)** (4.32)** (−7.96)** (25.18)** (− 3.09)** (− 29.96)** (1.14) (−11.62)** (27.77)** (− 2.67)**
PS − 0.0005 0.0041 0.6785 0.0001 0.0326 PS − 0.0018 0.0133 5.0375 −0.0004 0.0254
(− 0.00) (0.32) (0.72) (0.01) (− 0.03) (1.74) (2.46)* (−0.05)
0.0004 0.003 0.3547 −0.0007 0.0334 0.0348 0.006 0.0099 4.4921 −0.0021 0.0193 0.0281
(0.00) (0.24) (0.34) (−0.04) (1.26) (0.11) (1.25) (1.47) (−0.24) (− 0.45)
ZERO 0.2426 0.2168 4.5105 0.0827 0.2394 ZERO 0.0873 0.1903 17.0387 0.096 0.2092
(6.37)** (7.06)** (9.43)** (15.32)** (3.87)** (9.28)** (11.74)** (29.80)**
0.2709 0.2105 4.0334 0.0752 0.0975 0.2477 0.1165 0.1857 15.3763 0.0896 0.1183 0.2156
(6.65)** (7.06)** (8.31)** (12.48)** (12.42)** (4.76)** (9.38)** (10.92)** (25.32)** (10.85)**
TURNOVER − 0.7362 0.1111 2.0877 0.1387 0.4847 TURNOVER − 0.4494 0.0629 5.0269 0.0784 0.5093
(− 16.89)** (16.51)** (7.75)** (18.20)** (− 10.10)** (11.13)** (8.79)** (10.72)**
− 0.6913 0.1012 1.5129 0.1288 0.0717 0.5131 − 0.4191 0.0571 3.3282 0.0727 0.0617 0.5387
(− 17.21)** (16.51)** (6.78)** (18.53)** (9.76)** (− 10.79)** (12.11)** (6.27)** (11.58)** (10.38)**
LM − 20.5198 2.0748 −29.0121 3.0781 0.3472 LM − 15.4610 1.1357 −81.3121 2.3011 0.3181
(− 87.09)** (43.03)** (−9.99)** (79.30)** (− 44.08)** (35.50)** (−10.65)** (41.60)**
− 21.2028 2.2625 −22.8765 3.2281 −0.6724 0.374 − 15.8823 1.2284 −70.7545 2.3791 −0.3394 0.3393
(− 92.66)** (41.35)** (−8.54)** (84.47)** (− 7.73)** (− 48.32)** (36.82)** (−9.40)** (46.77)** (− 2.92)**
PBA − 0.2184 0.0549 −2.6395 0.0323 0.5310 PBA − 0.1240 0.0281 −3.4188 0.0163 0.4521
(− 30.60)** (29.84)** (−12.88)** (28.39)** (− 32.29)** (37.76)** (−12.39)** (25.77)**
− 0.2275 0.0564 −2.4560 0.0345 −0.0265 0.5516 − 0.1270 0.0284 −3.1708 0.0169 −0.0159 0.4697
(− 32.89)** (30.76)** (−13.13)** (30.86)** (− 13.45)** (− 31.85)** (38.99)** (−11.97)** (25.81)** (− 8.46)**

** and * denote significance at the 1% and 5% levels, respectively. ** and * denote significance at the 1% and 5% levels, respectively.
D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192 191

glance, but this relationship is also consistent with the findings in Table 2. Table 7
Non-zero values of BEEDLES are mainly concentrated in small firms, Fama–MacBeth cross-sectional regressions — large firm sample. Individual stock liquidity
proxies are regressed cross-sectionally each month on the concurrent month's closing price
which also have higher TURNOVER on average. (PRICE), the variance of daily returns for the concurrent month (VARIANCE), monthly
We find no relationship between PS and trading characteristics. aggregate dollar trading volume (VOLUME) and monthly absolute stock return (ABSR). PRICE
This indicates that the sources of the return reversal measure are not and VOLUME are logarithmic transformed. AMIHUD is the liquidity ratio from Amihud
related to stocks' trading characteristics. It is important to note that (2002). PS is the return reversal measure from Pastor and Stambaugh (2003). TURNOVER is
the stock turnover rate. ZERO is the zero return measure from Lesmond et al. (1999), but is
Pastor and Stambaugh (2003) did not assign their liquidity measure to
modified as (1 — the proportion of zero daily returns). LM is the turnover-adjusted zero daily
individual stocks. They argue that while their liquidity measure is volumes from Liu (2006). PBA is the proportional bid–ask spread. The sign of AMIHUD, LM
imprecise for individual stocks, their market-wide liquidity measure and PBA was flipped to make them represent liquidity. The coefficients are averaged across
becomes more precise as the sample size increases. Therefore, we 189 trading months in the sample. T-statistics are adjusted for first-order auto-correlation
have to be cautious about applying PS to individual stocks. and are reported in parentheses below the coefficient estimates.

Stoll (2000) points out that if the relationships with stock character- Intercept PRICE VARIANCE VOLUME ABSR Adj. R2
istics are the same using different liquidity proxies, liquidity is simply a Hypothesised sign + − + −
one-dimensional concept. If not, liquidity is a multifaceted concept. Our
AMIHUD −17.632 −1.5249 −1061.1613 2.6711 0.2574
results support the notion that liquidity is multidimensional, and that it (−25.77)** (− 9.75)** (− 10.43)** (24.13)**
can be viewed from different aspects of trading behaviour. −17.5842 −1.5511 −1151.2270 2.6696 −0.0608 0.2617
(−26.43)** (− 9.66)** (− 9.97)** (24.55)** (−0.11)
PS 0.0559 0.0372 18.0408 −0.0123 0.0583
4.3.2. Robustness check: size groups (0.52) (1.59) (1.31) (−0.71)
As a robustness check, the influences of trading characteristics on 0.061 0.0369 21.901 −0.0128 −0.0573 0.0634
liquidity across the size groups are examined. The BEEDLES measure is (0.54) (1.59) (1.10) (−0.72) (−0.49)
ZERO 0.1199 0.1858 46.5322 0.0881 0.386
excluded because all stocks within a size group have zero values for (2.33)* (56.42)** (11.58)** (11.91)**
BEEDLES in some months. Tables 5, 6 and 7 report regression results 0.1166 0.1866 43.0804 0.0875 0.1285 0.3925
for small, medium and large size firms, respectively. The results in the (2.31)* (57.90)** (8.84)** (11.94)** (8.63)**
TURNOVER −0.1424 0.0070 10.1079 0.0245 0.4295
size groups are similar to those found in the main analysis. The (−11.14)** (8.70)** (7.58)** (12.40)**
influences of trading characteristics on ZERO, PBA and LM remain −0.1402 0.0068 7.0283 0.0237 0.0646 0.4699
consistent across the size groups. Trading characteristics continue to (−10.86)** (7.46)** (5.81)** (12.34)** (10.73)**
LM −6.2956 0.0904 −115.5866 0.8804 0.2473
show no explanatory power on PS. (−23.30)** (5.01)** (− 6.18)** (21.41)**
PRICE becomes positively related with TURNOVER in the size groups, −6.3265 0.0722 −120.3095 0.8885 0.0669 0.2699
which is consistent with expectations. This indicates that stocks with (−22.24)** (4.46)** (− 6.23)** (20.47)** (0.57)
PBA −0.0739 0.0097 −4.5445 0.0088 0.5721
higher prices have higher turnover rates in each size category. The
(−21.42)** (24.22)** (− 8.26)** (18.58)**
relationship between PRICE and AMIHUD is consistent with our −0.0736 0.0097 −4.3551 0.0088 −0.0054 0.5775
hypothesis in the small size group. Stocks with lower prices tend to (−21.73)** (23.91)** (− 10.00)** (18.76)** (−4.09)**
have higher price impacts. The relationship between PRICE and AMIHUD ** and * denote significance at the 1% and 5% levels, respectively.
disappears in the medium-size group and becomes negative for large
firms. This indicates that the negative relationship between PRICE and
AMIHUD observed in Table 4 may be driven by large size firms.

5. Conclusion important for the other proxies. This also implies that the return
reversal effect is small in the Australian market. In addition, we find
The role of liquidity in generating stock returns has been examined in that small and large firms have a higher average volume than medium-
the Australian equity market (e.g. see Chan and Faff, 2003; Marshall and size firms do. This suggests that trading may be more focused on small
Young, 2003). In this paper, we take a step further and examine the and large firms in Australia. However, we are not claiming that higher
relationships between liquidity proxies and the influences of trading trading volume necessarily indicates higher liquidity.13 In addition,
characteristics on stock liquidity in the pure order-driven market of based on other liquidity proxies, small firms do not necessarily have
Australia. Prior studies that have examined similar research issues mainly higher liquidity than medium-sized firms do.
focus on liquidity proxies constructed from microstructure data. There are Notably, we have been silent on the question of what is the “best”
many situations in which little or no microstructure data are available and liquidity proxy. This research issue is beyond the scope of the current
in these cases, we have much less guidance on the best way to proceed. study. However, as noted in Goyenko et al. (2009), the selection of
This gap in the literature motivates the current research to use liquidity liquidity proxies in an empirical design depends on what exactly one
proxies constructed from low-frequency data. In this paper, we examine wants to capture. Our results support their assertion, as liquidity is
the relationship between liquidity and stocks' trading characteristics, such multidimensional and can be captured by different measures of trading
as price per share, return volatility and trading volume in the Australian activity. The current study shows that through firm trading character-
market. We mainly focus on trading characteristic variables because data istics, we can better understand the sources of liquidity.
for these variables are relatively accessible, and they have been shown to
be important sources of liquidity.
We find that the correlations among the employed liquidity proxies Acknowledgements
are low in the cross-section. This implies that they represent different
dimensions of liquidity and a certain type of trading behaviour. Stoll The helpful comments of an anonymous referee and the financial
(2000) also notes that different liquidity measures need not be support provided by an ARC Linkage Grant (LP0453913) are gratefully
correlated for similar reasons. Consistent with the literature, trading acknowledged.
characteristics are important determinants of liquidity. However, their
relationships with stock turnover and the return reversal measure
exhibit a somewhat different pattern than the other liquidity proxies. 13
Pastor and Stambaugh (2003) note that it is often the case that volume is high
In particular, the return reversal measure does not depend on stocks' when liquidity is low. This is especially the case during periods of financial crisis or
trading characteristics. This result suggests that the source of the stress. The problem with using trading volume as a proxy for liquidity is also discussed
return reversal measure is not related to stock characteristics that are in Chordia et al. (2002).
192 D. Chai et al. / International Review of Financial Analysis 19 (2010) 181–192

References Duffee, G. R. (1995). Stock returns and volatility a firm-level analysis. Journal of
Financial Economics, 37(3), 399−420.
Aitken, M., & Frino, A. (1996). The determinants of market bid ask spreads on the Eckbo, E. B., & Norli, O. (2002). Pervasive liquidity risk. Dartmouth College Working
Australian stock exchange: Cross-sectional analysis. Accounting and Finance, 36(1), Paper.
51−64. Fama, E. F., & MacBeth, J. D. (1973). Risk, return and equilibrium empirical test. Journal
Amihud, Y. (2002). Illiquidity and stock returns: Cross section and time series effects. of Political Economy, 81(3), 607−636.
Journal of Financial Markets, 5(1), 31−56. Glosten, L. R. (1994). Equilibrium in an electronic open limit order book. Journal of
Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid–ask spread. Journal of Finance, 49(5), 1127−1161.
Financial Economics, 17(2), 223−249. Goyenko, R., Holden, G. W., & Trzcinka, C. A. (2009). Do liquidity measures measure
Atkins, A. B., & Dyl, E. A. (1997). Transactions costs and holding periods for common liquidity? Journal of Financial Economics, 92(2), 153−181.
stocks. Journal of Finance, 52(1), 309−325. Grossman, S., & Miller, M. (1988). Liquidity and market structure. Journal of Finance, 21(3),
Beedles, W., Dodd, P., & Officer, R. R. (1988). Regularities in Australian share returns. 617−633.
Australian Journal of Management, 13(1), 1−29. Handa, P., Schwartz, R., & Tiwari, A. (1998). The ecology of an order-driven market.
Bekaert, G., Harvey, C. R., & Lundblad, C. (2007). Liquidity and expected returns: Lessons Journal of Portfolio Management, 24(2), 47−55.
from emerging markets. Review of Financial Studies, 20(6), 1783−1831. Hasbrouck, J. (2009). Trading costs and returns for U.S. equities: Estimating effective
Benson, G. J., & Hagerman, R. L. (1974). Determinants of bid–asked spreads in the over- costs from daily data. Journal of Finance, 64(3), 1445−1477.
the-counter market. Journal of Financial Economics, 1(4), 353−364. Kyle, A. (1985). Continuous auctions and insider trading. Econometrica, 53(6), 1315−1335.
Black, F. (1971). Towards a fully automated exchange: Part 1. Financial Analyst Journal, Lee, C. M., & Swaminathan, B. (2000). Price momentum and trading volume. Journal of
27(4), 29−34. Finance, 55(5), 2017−2069.
Branch, B., & Freed, W. (1977). Bid–asked spreads on the AMEX and the big board. Lesmond, D. A., Ogden, J. P., & Trzcinka, C. A. (1999). A new estimate of transaction costs.
Journal of Finance, 32(1), 159−163. Review of Financial Studies, 12(5), 1113−1141.
Brennan, M. J., & Subrahmanyam, A. (1996). Market microstructure and asset pricing: Liu, W. (2006). A liquidity-augmented capital asset pricing model. Journal of Financial
On the compensation for illiquidity in stock returns. Journal of Financial Economics, Economics, 82(3), 631−671.
41(3), 441−464. Llorente, G., Michaely, R., Saar, G., & Wang, J. (2002). Dynamic volume–return relation
Brown, D. P., & Zhang, Z. M. (1997). Market orders and market efficiency. Journal of of individual stocks. Review of Financial Studies, 15(4), 1005−1047.
Finance, 52(1), 277−308. Marshall, B., & Young, M. (2003). Liquidity and stock returns in pure order-driven
Chan, H. W. H., & Faff, R. W. (2003). An investigation into the role of liquidity in asset markets: Evidence from the Australian stock market. International Review of
pricing: Australian evidence. Pacific-Basin Finance Journal, 11(5), 555−572. Financial Analysis, 12(2), 173−188.
Chordia, T., Roll, R., & Subrahmanyam, A. (2000). Commonality in liquidity. Journal of Pastor, L., & Stambaugh, R. F. (2003). Liquidity risk and expected stock returns. Journal
Financial Economics, 56(1), 3−28. of Political Economy, 111(3), 642−685.
Chordia, T., Roll, R., & Subrahmanyam, A. (2002). Order imbalance, liquidity and market Petersen, M. A. (2009). Estimating standard errors in finance panel data sets:
returns. Journal of Financial Economics, 65(1), 111−130. Comparing approaches. Review of Financial Studies, 22(1), 435−480.
Chordia, T., Shivakumar, L., & Subrahmanyam, A. (2004). Liquidity dynamics across Sadka, R. (2006). Momentum and post-earnings-announcement drift anomalies: The
small and large firms. Economic Notes, 33(1), 111−143. role of liquidity risk. Journal of Financial Economics, 80(2), 209−249.
Cochrane, J. (2001). Asset pricing. New Jersey: Princeton University Press. Stoll, H. (1978). The pricing of security dealer services: An empirical study of NASDAQ
Datar, V., Naik, N. Y., & Radcliffe, R. (1998). Liquidity and stock returns: An alternative stocks. Journal of Finance, 33(4), 1153−1172.
test. Journal of Financial Markets, 1(2), 203−219. Stoll, H. (2000). Friction. Journal of Finance, 55(4), 1479−1514.
Demsetz, H. (1968). The costs of transacting. Quarterly Journal of Economics, 82(1), Stoll, H., & Whaley, R. (1983). Transaction costs and the size effect. Journal of Financial
33−53. Economics, 12(1), 57−79.
Ding, Z., Granger, C. W., & Engle, R. F. (1993). A long memory property of stock market Tinic, S. (1972). The economics of liquidity services. Quarterly Journal of Economics, 86(1),
returns and a new model. Journal of Empirical Finance, 1(1), 83−106. 79−93.

You might also like