Institutional Order Illiquidity and Expected Stock Returns

Qiuyang Chen*

Huu Nhan Duong

Manapon Limkriangkrai

This Version: 10th September 2016

JEL Classifications: G10, G20, G24

Keywords: Institutional Price Impact; Order Illiquidity; Asset Pricing

Authors are from the Department of Banking and Finance, School of Business and Economics, Monash University, Melbourne, Australia.

*Corresponding author: Qiuyang Chen, Department of Banking and Finance, Monash Business School, Monash University, 900 Dandenong Road, Caulfield East, 3145, Australia. Telephone: (+613) 9903 4078; Email: mars.chen@monash.edu.

Acknowledgements: We are grateful to the Securities Industry Research Centre of Asia-Pacific (SIRCA) for providing the data used in our study. We are also grateful for helpful comments from Avanidhar Subrahmanyam, Henk Berkman, Petko Kalev, Phil Gray, Te-Feng Chen, Eric Lam, Clark Liu, Talis Putnins, and Stephen Brown. All remaining errors are our own.

Institutional Order Illiquidity and Expected Stock Returns

Abstract

This study proposes a new approach for estimating the adverse selection component of

illiquidity, and links the new measures to asset pricing. Motivated by Kyle’s (1985) price impact

model, we decompose the aggregate price impact into institutional and individual components by

conditioning order flows on the identity of different investor classes. The asset pricing analysis

shows that the positive illiquidity premium is predominantly driven by the institutional order

illiquidity, and that individual order illiquidity plays no role in explaining stock returns. Further

analyses on the buy- and sell- component of institutional/individual order illiquidity suggest that

institutional sell-order illiquidity is the most significantly priced price impact variable. The

significant pricing of institutional order illiquidity is supported by both the information and

liquidity channels of institutional trading.

JEL Classifications: G10, G20, G24

Keywords: Institutional Price Impact; Order Illiquidity; Asset Pricing

i

Preface

Paper title: Institutional Order Illiquidity and Expected Stock Returns

Supervisors: Dr. Huu Nhan Duong and Dr. Manapon Limkriangkrai

Extant literature on the return-illiquidity relation based on the adverse selection cost uses

illiquidity measures that assume a symmetric relation between order flows and price changes

(see measures, such as Glosten and Harris, 1988; Brennan and Subrahmanyam, 1996; Huang and

Stoll, 1997; and Madhavan Richardson, and Roomans, 1997). The price impact approach of

estimating the adverse selection cost is supported by Kyle’s (1985) seminal theory. However,

one embedded assumption in Kyle’s model is that the market maker cannot distinguish different

types of traders (i.e., informed traders and noise traders) behind the bulk order submission in

each round of the auction. As a result, the lambda (i.e., the inverse measure of the market depth)

not only measures the price impact of informed traders, but also takes into account of the bid-ask

spread generated by noise traders. Hence, the aggregate lambda measure may cast some concerns

regarding the true effect of the adverse selection premium required by liquidity providers on

asset returns. In this study, we propose a new approach for estimating the adverse-selection

component of illiquidity and relate the new measures to asset prices.

ii

1. Introduction

The relation between liquidity and returns is one of the most researched areas in the finance

literature. Several studies demonstrate the importance of liquidity as a determinant of expected

returns (see, for example, Amihud and Mendelson, 1986; Brennan and Subrahmanyam, 1996;

Amihud, 2002; Jones, 2002; and Chordia, Huh and Subrahmanyam, 2009 ).1 One important

strand of the liquidity pricing literature focuses on the role of the adverse selection component of

illiquidity in explaining stock returns. As illustrated in Bagehot (1971), Kyle (1985) and Glosten

and Milgrom (1985), the adverse selection (i.e., the information asymmetry) is considered to be

the primary source of illiquidity due to the presence of informed traders. Subsequently, Brennan

and Subrahmanyam (1996) and Chordia, Huh, and Subrahmanyam (2009) find that the adverse

selection cost of illiquidity is priced in the cross-section of expected stock returns with large

economic significance.

Prior studies on the pricing of the adverse selection cost of illiquidity principally use

price impact measures that assume a symmetric relation between order flows from different

groups of investors and price changes. In this study, we allow for the asymmetric relation

between of order flows from institutional and individual investors groups and price changes. By

doing so, we provide the first empirical evidence on the asset-pricing implications of institutional

and individual price impacts. The distinction between institutional and individual order flow is

crucial, since two classes of investors differ substantially in terms of their possessions of

information (Nofsinger and Sias, 1999; Grinblatt and Keloharju, 2000; Griffin, Harris, and

Topaloglu, 2003; and Linnainmaa and Saar, 2012). Therefore, the institutional and individual

order illiquidity (i.e., the ‘lambda’) provide stronger empirical underpinnings to the theoretical

implications for the effects of the adverse selection cost of illiquidity on asset prices.

We utilize a unique intraday dataset that contains the identities of brokers involved in

every transaction in the Australian Securities Exchange (ASX) over an extended sample period

from January 1996 to December 2012. Motivated by Linnainmaa and Saar (2012) and Fong,

Gallagher and Lee (2014), we classify institutional and individual order flows as trades executed

1 For further evidence on the relation between liquidity and expected returns, see, among others, Datar, Naik and Radcliffe (1998), Jacoby, Fowler and Gottesman (2000), Easley, Hvidkjaer, and O’Hara (2002), Huh (2014), and Chung and Huh (2016).

1

through institutional and retail brokers, separately. To compare the relative importance between

institutional and individual lambdas in asset pricing, we use the empirical asset pricing

framework developed by Brennan, Chordia, and Subrahmanyam (1998). That is, individual stock

returns are adjusted for Fama and French (1993) factors as well as Carhart (1997) momentum

factor to alleviate the errors-in-variable problem in cross-sectional regressions. Moreover, we

also use the weighted least-squares (WLS) regression recommended by Asparouhova,

Bessembinder and Kalcheva’s (2010) to correct for potential market microstructure biases in our

asset pricing analyses.

Our key findings are summarized as follows: First, we show that institutional lambda is

priced more significantly than individual lambda in the cross-section of expected stock returns.

The prominence of institutional lambda is robust to the estimations of alternative price impact

models, different trade-size classifications of order flows, and different subsample periods. More

importantly, after controlling for the effect of institutional lambda, individual lambda is not

significantly priced in stock returns. Furthermore, the pricing of institutional lambda is also

economically significant. For the largest firm size quintile, the long-short portfolio sorted on

institutional lambda yields a return spread of 4.68% per annum. Our finding is consistent with

the theoretical implications in Kyle (1985), Glosten and Milgrom (1985), Easley and O’Hara

(1987), Glosten and Harris (1988), and Chordia et al. (2009), which suggest that only informed

traders observe the fundamental value of the risky security, and that the impact of institutional

trades generates a permanent price change as new information is incorporated into security

prices.

In order to dissect the underlying sources behind the prominence of institutional lambda,

we decompose the lambda into more fundamental elements by using the asymmetric illiquidity

framework developed by Brennan, Chordia, Subrahmanyam, and Tong (2012) and Brennan,

Huh, and Subrahmanyam (2013). Specifically, institutional and individual lambdas are

decomposed into their corresponding buy and sell components. When we regress stock returns

on these four lambda components (i.e., institutional and individual buy/sell lambdas), we find

that the institutional sell lambda is the most significantly priced illiquidity variable. This finding

is consistent with the early theoretical work on institutional asset fire sales by Shleifer and

2

Vishny (1997) and Gromb and Vayanos (2002). Their theoretical models suggest that when

financial constraints, such as margin calls and investor redemptions, are imposed on institutional

investors, they can be forced to liquidate their positions prematurely. The adverse price impact

induced by institutional sell-offs significantly reduce liquidity in the market and result in further

price reduction.2 Thus, institutional sell-side orders are particularly sensitive in illiquid stocks,

and as such, institutional investors are willing to pay higher illiquidity premium on their sell-side

orders.

Having established a strong positive relation between institutional lambda and stock

returns, we further investigate the potential channels behind this relation. We posit that the effect

of institutional lambda on stock returns can be attributed to both the information and liquidity

channels of institutional trading. The information and liquidity hypotheses are examined

extensively in the literature. Holthausen, Leftwich, and Mayers (1990) and Chan and Lakonishok

(1993) find that most of the price impact induced by institutional block trading is permanent, and

the permanent price effect is driven by the information content of institutional trades. On the

other hand, Lakonishok, Shleifer, and Vishny (1992), and Nofsinger and Sias (1999) document a

temporary liquidity impact of institutional trading on stock returns, and the liquidity effect tends

to dissipate quickly in the short-run.

We utilize two types of events to test the information and liquidity channels of

institutional lambda effect, namely, the unscheduled corporate announcement and the

announcement of quarterly S&P/ASX 200 index deletion. To test the information channel, we

investigate the relation between the pre-announcement institutional/individual lambda and the

three-day announcement-period cumulative abnormal returns. We focus on unscheduled

corporate announcements, given that the absence of predetermined announcement dates makes

the announcement returns more difficult to be inferred from the public information. This unique

feature of the unscheduled announcement allows us to detect the information content of

institutional/individual trading induced price impact. Our analyses show that the pre-

announcement institutional buy lambda (sell lambda) has a positive (negative) relation with the

three-day announcement return. In contrast, the pre-announcement individual buy and sell 2 Brunnermeier and Pedersen (2009) show that a negative shock on the aggregate market liquidity can lead to losses on liquidity providers’ initial positions (e.g., market makers and other financial intermediaries), thus, forcing them to continue to sell. Therefore, on average, institutions pay higher illiquidity premium to exit their positions.

3

lambdas have no effect on the forthcoming announcement return. These results support the

information channel on the effect of institutional lambda on stock returns.

The information effect of institutional lambda does not rule out the channel of the

liquidity effect. The short-term illiquidity premium often arises from the difficulty of

institutional investors locating the willing counterparties whom they can transact with

immediately. To test for the liquidity channel, we focus on the institutional trading surrounding

the announcement of quarterly S&P/ASX 200 index deletion. The announcement date of index

deletion is publicly accessible and certain institutional investors (e.g., index funds) are forced to

rebalance their portfolios around the announcement date. Hence, this provides us an ideal setting

to gauge the behavior of the price pressure generated by institutional trading when changes to the

index composition are announced. More importantly, the announcement of the index rebalance

itself is information-free, therefore, the trading induced price pressure surrounding the

announcement date reflects the institutional demand for liquidity immediacy. In the analysis, we

observe an immediate and sharp decrease (increase) in institutional buy (sell) lambda upon the

announcement of quarterly S&P/ASX 200 index deletion. In contrast, individual buy and sell

lambdas do not exhibit any particular patterns after the index deletion list is announced. Hence,

the contemporaneous relation between the announcement of index deletion and changes in

institutional lambda provides evidence supporting the liquidity hypothesis.

Our study makes two important contributions to the literature on the pricing of the

adverse selection cost of illiquidity. First, to the best of our knowledge, this study is the first to

examine the asset-pricing implications of institutional versus individual price impacts. Standard

price impact models assume a symmetric relation between the aggregate order flow and price

changes (see, for example, Kyle, 1985; Glosten and Harris, 1988; Brennan and Subrahmanyam,

1996; Huang and Stoll, 1997; and Madhavan et al., 1997). The lambda measure estimated by the

conventional approach does not differentiate the investor identity behind the order submission,

hence, the aggregate price impact measure could potentially obscure the true effect of the

adverse selection component of illiquidity on asset prices. By allowing an asymmetric relation

between institutional and individual order flow and price changes, this study shows that the

4

adverse selection cost induced by institutional price impact is the predominant driver behind the

illiquidity premium.

Second, our paper also sheds light on the relative importance of buy- and sell-order

illiquidity for different classes of investors in asset pricing. As highlighted by Brennan et al.

(2012) and Brennan et al. (2013), the equilibrium rates of return are more sensitive to seller-

initiated trades, but not sensitive to buyer-initiated trades. Our empirical results further extend

this emerging literature by showing that while both institutional and individual sell lambdas are

priced more significantly than the corresponding buy lambdas, the institutional sell lambda is the

most significantly priced variable among all other components of the adverse selection cost of

illiquidity. Furthermore, our findings of information and liquidity channels of institutional

lambda effect also complements the literature on the influence of institutional trading on stock

prices (see, for example, Holthausen et al., 1990; Lakonishok et al., 1992; Chan and Lakonishok,

1993; and Nofsinger and Sias, 1999).

The remainder of the paper is structured as follows: Section 2 outlines the theoretical

motivation and empirical estimations for our price impact model. Section 3 describes the data

and sample. Section 4 reports portfolio sorting analyses. Section 5 presents the result of asset

pricing tests. Section 6 investigates potential channels behind the prominence of institutional

lambda. Section 7 concludes the paper.

2. The Estimation of Institutional and Individual Lambdas

The first subsection explains the intuition behind the proposed lambda estimates, while the

second subsection introduces the price impact model applied in this study to estimate lambdas.

2.1 The Theoretical Motivation of Kyle’s (1985) Lambda

In Kyle’s (1985) model, the process of trading is modelled as a two-period auction, and a risky

security is traded among three types of traders: risk-neutral informed traders who have superior

information about the fundamental value of the risky security, liquidity traders (i.e., noise

5

traders) who trade securities for idiosyncratic or liquidity reasons (i.e., their demands are

exogenous), and a risk-neutral market maker who faces a perfect competition. The competitive

market maker sets a market clearing price that absorbs the net demand from other traders, so that

the expected profits conditioning on observing the aggregate order flow is zero. The trading

begins at time 0, and the risky security is liquidated at time 1. The ex post liquidation value of

the risky security (i.e., denoted by ) at time 1 is:

(1)

where is the liquidation value known to all agents at time 0, is the innovation of the payoff

known to informed traders, however, informed traders observe with a noisy signal, . The

quantity traded by informed traders is denoted by , and the quantity traded by liquidity trader is

denoted by . The random variables , , and are normally distributed with mean zero, and

assumed to be mutually independent.

Kyle (1985) shows that there exists a unique equilibrium if both profit maximization and

market efficiency conditions are held. In particular, the market efficiency condition requires the

clearing price set by the market maker to satisfy:

(2a)

(2b)

where is the net quantity (i.e., the net order flow) submitted by both informed traders and liquidity

traders at time 0. Since market maker earns zero expected profits conditioning on the bulk order flow, a

linear pricing rule is assumed to be applied by the market maker:

(3a)

(3b)

where is the classic inverse measure of market depth (i.e., the price impact of the net order flow on

price changes). Equation (3b) implies that the market maker cannot observe the specific quantities

traded by the informed traders and liquidity traders separately. As a result, price changes are always the

6

outcome of order flow innovations in each round of the auction. In addition, the profit maximization

condition in Kyle (1985) gives rise to a linear equilibrium in a sequential auction environment as follows:

(4a)

(4b)

(5)

where is the change in prices between two consecutive trades (i.e., ,

denotes the aggregate price impact of the combined order flow from both informed and liquidity

traders. Hence, Equation (5) is the original Kyle’s (1985) price impact model.

To relax the assumption that the market maker only observes the aggregate order flow ,

we allow asymmetric price responses to informed order flow and noise order flow separately:

. (6)

where is the price impact generated from informed order flow (i.e., adverse selection

lambda), and is the price impact produced by noise order flow (i.e., noise lambda). Given

that only contains new information about the fundamental value of the risky security, we

should expect that the adverse selection component of illiquidity is purely driven by . In

contrast, liquidity traders trade the risky security for idiosyncratic reasons, hence, on average,

should have no effect on security prices.

2.2 The Empirical Estimation of Adverse Selection and Noise Lambdas

In this subsection, we introduce the empirical technique in estimating the adverse selection

lambda and the noise lambda . Since Kyle (1985), subsequent studies on price impact

models suggest that the bid-ask spread is comprised of the adverse selection and noninformation

components (see, for example, Glosten and Harris, 1988; Lin, Sanger and Booth, 1995; Huang

7

and Stoll, 1997, and Madhavan et al., 1997).3 In order to achieve a sensible measure of adverse

selection component of illiquidity, we also account for the noninformation component of illiquidity in

the model. Specifically, we adopt the Brennan and Subrahmanyam (1996) model (i.e., which parallels to

the Glosten and Harris (1988) approach):

. (7)

Let denotes the sign of the incoming order flow at time ( for a buyer initiated trade and

for a seller initiated trade), the noninformation component of illiquidity is denoted by , the

aggregate adverse selection cost is denoted by , and is the public information signal observed

by the liquidity provider at time .

In order to decompose into the adverse selection lambda and the noise lambda ,

we use institutional order flow as the proxy for the informed order flow , and individual

(or retail) order flow as the proxy for the noise order flow . The choice of using

institutional investors as the proxy for informed traders and individual investors as the proxy for

liquidity/noise traders is justified by the evidence in existing literature on institutional/retail

trading. For instance, ample amount of evidence shows that stocks bought by retail investors

underperform stocks sold by retail investors (Odean, 1998; Barber and Odean, 2000, 2002;

Barber, Odean and Zhu, 2009, and Barber, Lee, Liu and Odean, 2009). This indicates that, on

average, retail investors trade stocks for idiosyncratic reasons, and their order flows do not

convey new information in stock prices. Moreover, Foucault et al. (2011) find that there is a

significant positive relation between retail trading and idiosyncratic volatility; hence, on average,

retail investors behave like noise traders. In contrast, there are also empirical evidence which

suggests that institutional investors usually are better informed and more sophisticated than

3 More specifically, the noninformation component further consists of order processing and inventory components. The consensus from prior literature is that permanent price changes can entirely be attributed to adverse selection, while the impact of inventory risk is temporary and reversed quickly (Glosten and Harris, 1988; and Madhavan et al., 1997). However, a recent study by Chung and Huh (2016) shows that the noninformation component of price impact also plays an important role in explaining the cross-sectional stock returns.

8

individual investors (see, for example, Grinblatt and Keloharju, 2000; Barber, Lee, Liu, and

Odean, 2009; and Beohmer and Kelley, 2009).

To extract institutional and individual order flows from the aggregate order flows , we

utilize the approach similar to that of Fong, Gallagher and Lee (2014). Specifically, we classify

brokers into institutional and retail brokers using information from brokers’ websites and Factiva

search.4 Institutional order flows are trades executed through institutional brokers, and individual

order flows are trades executed through retail brokers. For those trades that cannot be identified

as either institutional trades or individual trades, we classify them as unclassified order flows.

The method of identifying institutional and individual order flows through broker ID

classification is supported by Linnainmaa and Saar (2012). They suggest that the broker ID is an

effective tool to infer the identity of different types of traders behind the order submission.5

Therefore, Equation (7) is modified to allow asymmetric price responses to institutional and individual

order flows:

, (8)

where is the price impact generated by institutional order flows (i.e., the adverse selection

lambda), is the price impact produced by individual order flows (i.e., the noise lambda), and

is the price impact of unclassified order flows which will then be discarded in asset pricing

analyses.6 Based on Equation (8), the ordinary least squares (OLS) is performed each month for

each stock to estimate all parameters, and is treated as an error term.

3. Data

4 Details on broker ID classifications are covered in Section 3. 5 There are concerns regarding the possibility that institutional investors may use multiple brokerage accounts to camouflage the information content of their order flows. However, both Linnainmaa and Saar (2012) and Fong et al. (2014) find that that magnitude of multiple brokerage account usage is small. Hence, different broker trades are representative of the underling clientele trading.6 The sum of , and equals to the aggregate order flow in Equation (7). In addition, it should be noted that the

unclassified order flow only accounts for 13% of the overall order flows.

9

3.1 Data and Sample Collection

Our sample for institutional/individual lambda estimations includes ordinary shares listed on the

Australian Securities Exchange (ASX) over the period January 1996 to December 2012. We use

intraday transaction data from the Australian Equities Tick History Service (AETHS), supplied

by the Securities Industry Research Centre of Asia-Pacific (SIRCA). The AETHS database

provides details on every order submitted to the central limit order book including stock ticker

code, order type (i.e., submission, execution, revision, and cancellation), date and time, order

price, order volume (i.e., numbers of shares), order direction (i.e., buy or sell). We are able to

track the order from its initial submission to any revision, cancellation or execution, provided

that there is a unique identification number assigned to each new order. In addition, we classify

trades into buyer-initiated and seller-initiated trades based on the directions of the market orders,

thus, we do not rely on Lee and Ready’s (1991) algorithm to infer the trade direction. Finally, for

every trade, our data provide information on the buying and selling brokers associated with the

trades.7

In order to classify brokers into institutional and retail brokers, we first match the broker

IDs in AETHS with the broker name list on the IRESS system. We filter out trades initiated by

institutional brokers, and then sort the remaining trades into those initiated by retail brokers. To

classify brokers, we follow a manual procedure similar to that used in Fong et al. (2014). That is,

we first check the broker’s website, their mission statement, the company overview, and

especially the “about us” section to identity their primary customer group. If the broker does not

have a website, we use Factiva to research for news items related to the broker, and we then

classify them accordingly. For those brokers that do not have sufficient information on their

website or Factiva search, we classify them as unclassified brokers. Finally, to avoid extreme

illiquidity stocks, we only include stocks with at least 330 trades per month or 15 trades per day

on average.8

7 SIRCA issued a notification on February 3, 2016 that restricts the use of broker IDs on orders and orders placed on the Centre Point order matching system (dark pool) after October 2010. These restrictions do not apply to our paper as we do not require access to broker IDs on order data. We only use broker IDs on trade data, which are still available in SIRCA or other data provider, such as IRESS. 8 We also used other liquidity filters, for example, 110 trades and 220 trades per month to estimate institutional and individual lambdas. The alterative lambdas yield similar results in asset pricing analyses, and therefore are not included in the paper.

10

The monthly stock returns, market capitalization, number of share outstanding are

obtained from the Share Price and Price Relative (SPPR) database. The daily stock returns and

share turnover variables are sourced from the SPPR daily. The accounting variable data such as

book value of total shareholder equity is obtained from Morningstar (formally known as Aspect

Huntley). The SPPR group ticker code is used to merge the intraday transaction data from

AETHS with stock return data from SPPR as well as accounting data from Morningstar. Finally,

we collect unscheduled corporate announcements data from the Australian Corporate

Announcement (ACA) database, available on SIRCA, for our analyses on the relation between

pre-announcement lambdas and unscheduled announcement returns.

3.2 Sample Characteristics

Panel A of Table 1 presents the summary statistics for the estimates of institutional and

individual lambdas. Institutional and individual lambdas are estimated by applying Equation (8),

and estimates are winsorized at the 1st and 99th percentiles. The first noticeable feature is that the

sample mean of institutional lambda is greater than that of individual lambda by approximately

17%. The mean-difference-test on null hypothesis that the mean of institutional lambda equals to

the mean of individual lambda is rejected at the 1% level. The magnitude of institutional lambda

exceeds that of individual lambda at 25th percentile, median, and 75th percentile. This indicates

that the institutional lambda’s greater magnitude is not driven by any particular lambda

subsamples. Moreover, the standard deviation of institutional lambda is also greater than the

standard deviation of individual lambda, suggesting that institutional lambda is more volatility

on average. Panel A also reports the average t-statistics for institutional and individual lambda

estimates and percentage of estimates that exceed t-value of 1.96. The averages of t-statistics

suggest that the estimates of both institutional and individual lambdas are highly significant.

Furthermore, institutional lambda has 78% of estimates that are significant at the 5% level. In

contrast, only 69% of individual lambda estimates that are significant at the conventional level.

Panel B of Table 1 provides the average of institutional and individual lambdas in up and

down markets separately. Motivated by the evidence in Brennan et al. (2013) that the illiquidity

in down markets is priced more significantly than the illiquidity in up markets, we present the

11

summary statistics separately for months in which the current month’s value-weighted market

return is positive (negative) as the up (down) market.9 It is evident that the difference in lambda

magnitude between up and down market is greater for institutional lambda than that for

individual lambda.10 This suggests that institutional investors prefer to avoid being trapped in an

illiquid stock in a falling market, and thus are willing to pay a major price concession in the

down market (i.e., they are facing a higher price impact for their trades). In contrast, there is only

small difference between individual lambda in up and down markets. The drastic increase of

institutional lambda in down markets can be explained by the relation between funding liquidity

and liquidity risk. Brunnermeier and Pedersen (2009) theoretically show that the impairment of

financial intermediaries’ capital holdings during the down market has a spillover effect on

institutions’ funding liquidity and will further result downward liquidity spirals.

Panel C of Table 1 reports the mean of institutional and individual lambdas for five size

groupings sorted on firms’ market capitalization. First, the mean of institutional lambda always

exceeds the mean of individual lambda within each size quintile portfolio. The mean-difference-

test on the null hypothesis that institutional lambda equals to individual lambda is rejected at the

1% level for all size quintiles. Second, both institutional and individual lambdas decrease

monotonically with firm size. Further, the difference between institutional and individual

lambdas is higher for bigger firms, indicating that institutional price impact is higher than that of

individuals’ in large size firms. This is because large size stocks are generally more liquid, only

institutional trades can have an impact on the bid-ask spread. This result is similar to that

documented in Linnainmaa and Saar (2012), they find that only trades from institutional

investors have a price impact on the bid-ask spreads for large stocks, whereas the price impact

induced by retail brokers’ order flows is negligible.

[Insert Table 1]

9 In terms of the aggregate market illiquidity level, Chordia, Roll, and Subrahmanyam (2001) find that illiquidity is greater in down markets.10 The sharp contrast between institutional and individual lambda in up and down markets can be partially explained by the liquidity co-movement documented in Chordia, Roll, and Subrahmanyam (2000). Recent studies suggest that the market liquidity co-movement is partially driven by institutional investors’ demand for liquidity in down markets (Mitchell and Pulvino, 2012). The fact that the difference for individual lambda between up and down markets is trivial indicates that the demand-side liquidity co-movement is mainly driven by institutions’ demand for liquidity in crisis periods.

12

4. Returns on Portfolio Sorts

Similar to Chordia, Huh and Subrahmanyam (2009), we first sort stocks into five size quintiles,

and subsequently each size quintile is sorted into five lambda quintiles. This sorting procedure

results in the total of 25 portfolios. We then compute value-weighted portfolio returns based on

the market capitalization from the end of previous month. The size and lambda groups are

denoted as SIZEi and Illqi (where ). For consistency, stocks used for the portfolio

analyses are limited to those used for our main regression analyses in the subsequent section. In

addition, we also compute the intercepts from the time-series regression of 25 value-weighted

portfolio returns (in excess of 13-week Treasury notes rate) regressed on the Fama-French-

Carhart four (FFC) factors.11

[Insert Table 2]

Table 2 reports the average of value-weighted monthly quintile portfolio returns for both

institutional and individual lambdas. It can be seen that there is a strong positive return-

institutional lambda relation across all five size quintiles. The average monthly returns in the

month following portfolio formation increase monotonically from the lowest institutional lambda

portfolio (i.e., the most liquid portfolio) to the highest institutional lambda portfolio (i.e., the

most illiquid portfolio). The return spreads between highest (Illiq 5) and lowest (Illiq 1)

institutional lambda are all statistically significant at the 5% level. The economic significance of

institutional lambda can be interpreted based on the return spread of the biggest size quintile,

which is 39 bps per month (i.e., 4.68% per annum).12 Panel B reports the average of value-

weighted monthly returns for individual lambda. The return spreads of individual lambda hedge

portfolios are only significant for the first three size quintiles. Moreover, the return spread of

individual lambda hedge portfolio in size quintile five is small in magnitude, suggesting that the

effect of individual lambda on stock returns is predominately concentrated in small stocks.

11 We also compute the equal-weighted mean returns for 25 portfolios formed by sorting stocks into lambdas and firm size which leads to similar results. Above-mentioned portfolio sorting results are available upon request.12 The average market capitalization for size quintile five is approximately $10 billion. In contrast, the average market capitalization for size quintile one is around $56 million. The characteristics of size quintile five represent the most investable stocks in the ASX; hence, our interpretation of economic significance will focus solely on size quintile five.

13

5. Asset Pricing Regressions

In this section, we examine the role of institutional and individual lambda and their relative

importance in asset pricing. The first subsection introduces the empirical methodology for asset

pricing tests. The second subsection reports the main results, and third subsection presents

subsample analyses as robustness check.

5.1 Empirical Asset Pricing Methodology

Our asset pricing methodology closely follows Brennan, Chordia, and Subrahmanyam (1998)

and Chordia, Huh and Subrahmanyam (2009). Specifically, to alleviate the error-in-variable

problem, the individual stock returns are adjusted for Fama-French-Carhart (FFC) factors — i.e.

market ( ), size ( ), book-to-market ( ), and momentum ( ) factors (Fama

and French, 1993; Carhart, 1997) based on two methods.13 The risk-adjusted return, and

are then used as dependent variables for the Fama-MacBeth (1973) cross-sectional regression.

The t-statistics of average slope coefficients are computed with Newey-West (1987) standard errors.

The regression specification follows Brennan et al. (1998):

1

(9)

where or 2, is the lagged lambda variable for stock (i.e., or which is

estimated from Equation (8)), and are lagged control variables for stock . The choice of

control variables is based on well-known cross-sectional stock return determinants. Follows the

spirit of Fama and French (1993), Jegadeesh and Titman (1993), Amihud (2002), Ang, Hodrick,

Xing and Zhang (2006), and Cheng, Hameed, Subrahmanyam, and Titman (2016), the control

variables are defined as follows:

13 In the first method, we apply the static sample fitting adjustment for monthly stock return . For the second method, we

utilize a dynamic sample fitting as the adjustment for individual stock return . In particular, the time-series rolling factor

loadings, are estimated for all stocks each month over the entire sample period using past 36 months return observations (with at least 24 months of past returns).

14

1) SIZE: measured as the natural logarithm of the market capitalization.

2) BM: measured as the natural logarithm of the book value of the firm’s equity to its

market value of equity.

3) SH_TURN: measured as the natural logarithm of the company’s share turnover rate

which is computed as the trading volume divided by the total number of shares

outstanding.

4) REV: measured as the one-month short-term return reversal.14

5) RET2-3: measured as buy- and-hold return on stock from month to .

6) RET4-6: measured as buy- and-hold return on stock from month to .

7) RET7-12: measured as buy- and-hold return on stock from month to .

8) IVOL: idiosyncratic volatility is measured as the volatility of the idiosyncratic return (

). The idiosyncratic return (i.e. the residual term) is computed by regressing daily stock

return on a value-weighted market index and daily Fama-French factors over a maximum

of 250 days ending on December 31 of year t.

9) AMIHUD: measured as the average daily ratio of the absolute stock return to the dollar

trading volume within the month t.

The book-to-market ratio are constructed from SPPR monthly and Morningstar accounting files.

Amihud (2002) illiquidity measure, idiosyncratic volatility (i.e., IVOL), and share turnover (i.e.,

SH_TURN) are constructed using both SPPR daily and monthly data. Other firm characteristic

variables (i.e., SIZE, RET2-3 to RET7-12) are compiled from SPPR monthly file.

5.2 Empirical Results

In this section, we examine whether institutional and individual lambdas have any explanatory

power on stock returns at the individual stock level. Based on Equation (9), the Fama-MacBeth

(1973) cross-sectional regression is performed each month. Table 3 reports the time-series

14 A recent study by Cheng et al. (2016) shows that the institutional selling pressure in response to past stock price declines is not immediately accommodated by market makers, which, in turn leads to the short-term return reversal. To ensure the premium of institutional order illiquidity is not driven by the effect short-term reversal, we include it in all our regressions.

15

averages of the coefficients of institutional/individual lambda and a group of control variables.

To correct for autocorrelation, t-statistics are computed based on Newey-West (1987) standard

errors. Panels A and B of Table 3 report the regression results for institutional and individual

lambdas on both raw and adjusted individual stock returns. First, based on the result from the

regression of raw excess stock returns (i.e., EXSRET0), the average coefficient of institutional

lambda is positive and statistically significant at the 1% level after controlling for effects of other

return determinants. Similarly, for EXSRET0 model specification, individual lambda also has

significant explanatory power on stock returns even after controlling for other firm

characteristics. However, it is evident that the magnitudes of both coefficient and t-value of

institutional lambda are substantially higher than those of individual lambda.

In addition, moving from raw returns to risk-adjusted returns in second and third columns

of Table 3 (i.e., EXSRET1 and EXSRET2), the explanatory power of both institutional and

individual lambdas attenuates substantially. This pattern suggests that the pricing of institutional

and individual lambdas is subject to systematic risks. In particular, individual lambda loses its

significance in EXSRET2 regression specification, indicating that the effect of individual lambda

on returns can be fully captured by FFC risk factors.

The signs of the coefficients on control variables in the Fama-Macbeth regressions are

consistent with prior literature. The negative coefficient on the firm size and positive coefficient

on book-to-market ratio are consistent with the size and value effects documented in Fama and

French (1992, 1993). The average coefficients of momentum variables are also positive and

significant. This is consistent with the price momentum documented in Jegadeesh and Titman

(1993). The negative coefficient on one-month short-term reversal is consistent with the reversal

effect documented in Jegadeesh (1990). It is worth noting that the other two alternative liquidity

proxies – share turnover and Amihud (2002) measure are less significant compared to lambda

variables. This suggests that the low-frequency-based illiquidity proxies do not provide much

adverse selection related information to the asset pricing model after the high-frequency-based

illiquidity is controlled for, which is consistent with the findings in Huh (2014).

Given that the significance of individual lambda is biased towards small stocks, a related

concern is whether the market microstructure biases could affect our results. Asparouhova,

16

Bessembinder and Kalcheva (2010) show that the estimated illiquidity premium produced by the

standard cross-sectional regression suffers upward bias in t-statistics in the presence of bid-ask

bounce. More importantly, the bid-ask bounce bias is more pervasive in small stocks, hence, we

address this issue by performing the weighted least-squares (WLS) cross-sectional regression.

The prior-month gross return (i.e. one plus the return in month ) is used as the weighting

choice for WLS. Panels C and D of Table 3 show that the magnitude of both the coefficients and

t-values of institutional and individual lambdas has decreased after adjusting the market

microstructure bias. However, overall results are similar to those reported in Panels A and B.

Specifically, institutional lambda maintains its significance in all three model specifications,

whereas individual lambda is not priced in stock returns when EXSRET2 is used as the dependent

variable.

[Insert Table 3]

To gain insight into which component of lambda (i.e., institutional or individual) is the

key driver behind the pricing of the adverse selection cost of illiquidity in stock returns, we

perform a horserace between institutional and individual lambda in Table 4. Panel A of Table 4

shows individual lambda is never priced in stocks returns in the presence of institutional lambda,

meanwhile institutional lambda is highly significant across all regression specifications. In

comparison to the regression results in Table 3, the coefficient of individual lambda reduces

substantially with the inclusion of institutional lambda, whereas the coefficient magnitude (as

well as t-value) of institutional lambda is qualitatively similar. This is a strong evidence

supporting our hypothesis that the adverse selection premium embedded in illiquidity cost is

predominantly driven by the information component (i.e., proxied by institutional order flows)

of Kyle’s lambda. On the other hand, the noise component (i.e., proxied by individual order

flows) of Kyle’s lambda plays no role in explaining stock returns. Given that the WLS

specification is a more stringent hurdle for claiming the pricing of illiquidity, we repeat the

analyses with WLS in Table 4 Panel B. Similar to Panel A, when both institutional and

individual lambdas are included in the regression, only institutional lambda is significantly

17

priced, suggesting that the adverse selection premium is mainly driven by institutional order

flows.

[Insert Table 4]

5.3 Robustness Tests

In summary, the main results in section 5.2 present two important findings. First, the institutional

lambda has the most significant effect on the cross-sectional stock returns. This is evident in the

anomalous return patterns of portfolios sorted on institutional lambda as well as the significant

pricing of lambda in risk-adjusted Fama-MacBeth (1973) regressions. In contrast, the effect of

individual lambda is mainly concentrated in small stocks and can be explained by the FFC risk

factors. Second, the positive relation between stock returns and the adverse selection component

of illiquidity is mainly driven by the institutional trading induced price impact. In this section,

we show our results are robust to different subsample periods, alternative price impact

estimations of institutional and individual lambdas, and different trade-size classifications of

order flows. For brevity, the results of above-mentioned robustness tests are available upon

request.

6. Underlying Channels of the Institutional Lambda Effect

In this section, we investigate the potential explanations behind the asymmetric pricing between

institutional and individual lambdas in stock returns. In order to dissect the underlying sources

behind the institutional/individual effect, we use two unique events to investigate the underlying

channels behind the prominence of institutional lambda over individual lambda in Section 6.1.15

15 In an unreported Fama-MacBeth (1973) cross-sectional regression analysis, we show that sell lambda is always priced more significantly than that of buy lambda. This asymmetric pricing relation holds both at institutional and individual levels. In addition, the institutional sell lambda has the greatest effect on returns among four lambda variables examined in the horserace. In order to estimate buy and sell lambdas for institutional and individual order flows, we modify Equation (8) in the spirit of Brennan et al. (2012) to estimate the institutional/individual buy- and sell-order illiquidity. Specifically, and are split

into corresponding buy and sell order flows by conditioning on trade directions, i.e., is institutional buy order

flow, is institutional sell order flow, is individual buy order flow, and is individual sell order flow.

18

6.1 Information and Liquidity Channels of the Institutional Lambda Effect

Following both the theoretical and empirical literature on institutional trading, we hypothesize

that the significance of institutional lambda is potentially caused by both information and

liquidity channels of institutional trading induced price impact (Holthausen et al., 1990;

Lakonishok et al., 1992; Chan and Lakonishok, 1993; Saar, 2001 and Sias et al., 2006). The

influential work by Chan and Lakonishok (1993) argues that the large institutional trades can

cause permanent price changes if the trades themselves reveal private information that is not yet

incorporated into stock price. The information effect of institutional trading will lead to a

positive and significant adverse selection cost. On the other hand, the short-term illiquidity

premium is often caused by the difficulty of institutional investors in locating the willing

counterparties whom they can transact with immediately. The efforts by institutional investors to

attract counterparties is then translated in the form of major price concessions on their trades

(Kaniel et al., 2008, and Campbell et al., 2009). To gain insight into these two potential channels,

we use two unique events to investigate the information and liquidity hypotheses of institutional

lambda effect, namely, the unscheduled corporate announcement and the announcement of

quarterly S&P/ASX 200 index deletion.

6.1.1 The Information Channel

To test the information channel, we investigate the relation between the pre-announcement

period abnormal individual/institutional lambda and the forthcoming three-day cumulative

abnormal returns. If the significance of institutional lambda is driven by the private information

content of their trades, we expect the pre-announcement institutional lambda to have predictive

power on the forthcoming announcement returns. We collect all corporate announcements data

over the period of January 1, 1996 to December 31, 2012 from the Australian Corporate

Announcement (ACA) database via SIRCA.16 We only focus on the unscheduled

announcements, such as the open-market share buy-back, director appointment/resignation, and

the intention to make takeover bid. Given that the unscheduled announcement does not have a

16 We require 60 trading days prior to the announcement to compute the benchmark level lambda, therefore, the starting point of the sample is April 1996.

19

predetermined date, the rumor induced trading is less likely, and thus provides us a cleaner

setting to examine the information content of institutional lambda.

Following Khan and Lu (2013) and Hao (2015), we first compute the benchmark average

lambda over the window [-60, -11]. We then subtract the benchmark lambda level from the daily

lambda during the pre-announcement window [-10, -2]. We denote the difference as the

abnormal level lambda . The model specification is given by:

1

(10)

where is one of the four lambda variables for stock (i.e.,

and ) in day . To test whether the pre-announcement

abnormal lambdas have any predictive power on the forthcoming announcement returns, we run the

following panel regression:

1

(11)

where the dependent variable is the cumulative abnormal return around the three-day

announcement window, for stock . The value-weighted market return is used as

the benchmark to calculate . To ensure that the result is not driven by other illiquidity

proxies, we compute the average share turnover in natural logarithm and average daily Amihud

ratio over the window [-60, -2] prior to the announcements. To control for the potential price

run-up leading up to the announcements, we compute the buy-and-hold market adjusted returns

over the window [-60, -2], as well as the standard deviation of daily stock return over the

window [-60, -2] prior to the announcements. We also control for firm characteristics, such as

size, book-to-market ratio, stock price, together with industry fixed effect and year fixed effect

.

20

Table 5 reports the results of panel regression of the three-day announcement return on

the pre-announcement abnormal lambda . Following Thomson (2011), all model

specifications apply two-way cluster-robust standard errors that are clustered by firm and event.

It can be seen that institutional buy lambda (institutional sell lambda) has a positive (negative)

relationship with the unscheduled announcements’ returns, and both variables are significant at

the 1% level. In contrast, none of the abnormal individual lambdas have predictive power on

, suggesting that individual pre-announcement trades do not contain any private

information. Our result is consistent with the finding in prior literature that institutional investors

are the more informed (see, for example, Grinblatt and Keloharju, 2000; Barber, Lee, Liu, and

Odean, 2009; and Beohmer and Kelley, 2009). The difference in predictive power between

institutional and individual lambdas on the forthcoming announcement returns provides evidence

to support the information hypothesis. The information channel of institutional price impact

implies that the prominence of institutional lambda is a form of adverse selection premium

required by uninformed traders to compensate for losses on trading with informed traders. As a

result, this will lead to a positive and significant institutional lambda effect on stock returns.

[Insert Table 5]

6.1.2 The Liquidity Channel

The information channel of institutional lambda does not rule out the possibility of the liquidity

channel. To directly examine the liquidity effect of institutional lambda is challenging, given that

the detailed institutional portfolio holding data are generally unavailable. As a result, we utilize a

unique event where the institutional investors with indexing approach are forced to rebalance

their portfolios without fundamental reasons. The announcement of quarterly S&P/ASX 200

index deletion provides us a unique setting to gauge the liquidity-motived institutional lambda

effect, given that the rebalance of index itself is information-free. This unique characteristic of

index deletion allows us to disentangle the liquidity effect from the information effect.

21

We manually collect all the announcement dates of the S&P/ASX 200 index quarterly

rebalancing from the announcement files which can be retrieved from the ASX’s official

website. The index deletion is announced to the public on the 10th business day (after the market

closed) prior to the effective rebalance date.17 Hence, we define our event date (i.e., effective

announcement date) as the next trading day after the S&P announcement date. We further

remove firms which were deleted from the index for reasons other than the breach of market

capitalization (e.g., spin-off, M&A and other corporate events). In each quarter over the period

September 2000 to December 2012, we compute the averages of individual and institutional

lambdas on the deleted S&P/ASX 200 stocks over the window [-10, +10] surrounding the event

date.18

Figure 1 plots the average and 95% confidence intervals for institutional buy/sell lambda

and individual buy/sell lambda surrounding the announcement of index deletion. Panel A of

Figure 1 shows that there is a sharp decrease in institutional buy lambda upon the announcement

of the deletion. Institutional buy lambda bounces back to the pre-announcement level two days

after the announcement. However, this rebound of institutional buy lambda is short-lived, this is

evident in the V-shape pattern in the post-announcement window. Panel B of Figure 1 shows that

there is a sharp increase in institutional sell lambda upon the deletion announcement. The

increase in institutional sell lambda on the announcement corresponds to the decrease in

institutional buy lambda in Panel A. The sharp surge in institutional sell lambda indicates that

there is an increased institutional demand to sell the removed stocks for minimizing their

tracking errors. However, institutional sell lambda quickly rebounds to the pre-announcement

level approximately six days after the announcement.

In contrast, Panels C and D reveal that there is no particular pattern in individual buy and

sell lambdas surrounding the announcement. For instance, there is only a small decrease in

individual lambda upon the announcement of deletion. Moreover, the movement of individual

sell lambda appears to be random surrounding the announcement. Given that the 95% confidence

17 Although the announcement of index rebalance is usually scheduled on the 10 th business day prior to the effective date of rebalancing, however, the actual announcement day may vary. Hence, we manually collect all actual announcement dates of ASX 200 quarterly rebalancing from the announcement files which can be retrieved from the ASX official website by typing in the announcement code “ZSP”: http://www.asx.com.au/asx/statistics/announcements.do?by=companyName&companyName.18 The S&P/ASX 200 Index was launched on 31 March 2000, however, the earliest available announcement file is only available from 15th September 2000. Therefore, we restrict our sample over the period of September 2000 to December 2012.

22

intervals of individual lambdas are substantially wider compare to that of institutional lambdas,

indicating that the averages of individual lambda are quite uninformative (i.e., higher standard

errors) around the announcement. Therefore, the overall result of the index deletion rebalance

does provide an indirect support for the liquidity channel of institutional lambda effect.

[Insert Figure 1]

7. Conclusion

One important strand of the asset pricing literature links the adverse selection cost of illiquidity

to asset prices (see, for example, Brennan and Subrahmanyam, 1996; and Chordia et al., 2009).

However, previous research on the return-illiquidity relation based on the adverse selection cost

assumes a symmetric relation between order flow and price changes. In this study, we propose a

novel method to estimate the adverse selection component of illiquidity, and link the new

measures to asset pricing. By using a comprehensive broker identity stamped intraday trading

data, this study is the first to show that the adverse selection cost of illiquidity is predominantly

driven by the institutional trading induced price impact (i.e., institutional lambda). In contrast,

the price impact generated by individual/retail trading (i.e., individual lambda) plays no role in

explaining stock returns. We further examine the information and liquidity channels for

institutional lambda effect. We find that the pre-announcement institutional buy (sell) lambda is

positively (negatively) related to the announcement-period abnormal returns for unscheduled

announcements. We also observe a sharp decrease (increase) in the institutional buy (sell)

lambda upon the announcement of index deletions. These findings imply that the underlying

mechanisms of the significant institutional lambda effect on returns is in fact driven by both

information and liquidity effects.

23

References

Amihud, Y. (2002). Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets, 5(1), 31-56.

Amihud, Y., & Mendelson, H. (1986). Asset pricing and the bid-ask spread. Journal of Financial Economics, 17(2), 223-249.

Ang, A., Hodrick, R. J., Xing, Y., & Zhang, X. (2006). The cross-section of volatility and expected returns. Journal of Finance, 61(1), 259-299.

Asparouhova, E., Bessembinder, H., & Kalcheva, I. (2010). Liquidity biases in asset pricing tests. Journal of Financial Economics, 96(2), 215-237.

Bagehot, W. (1971). The only game in town. Financial Analysts Journal, 27(2), 12-14.

Barber, B. M., & Odean, T. (2000). Trading is hazardous to your wealth: The common stock investment performance of individual investors. Journal of Finance, 55(2), 773-806.

Barber, B. M., & Odean, T. (2002). Online investors: do the slow die first?. Review of Financial Studies, 15(2), 455-488.

Barber, B. M., Lee, Y. T., Liu, Y. J., & Odean, T. (2009). Just how much do individual investors lose by trading?. Review of Financial studies, 22(2), 609-632.

Barber, B. M., Odean, T., & Zhu, N. (2009). Do retail trades move markets?. Review of Financial Studies, 22(1), 151-186.

Boehmer, E., & Kelley, E. K. (2009). Institutional investors and the informational efficiency of prices. Review of Financial Studies, 22(9), 3563-3594.

Brennan, M. J., & Subrahmanyam, A. (1996). Market microstructure and asset pricing: On the compensation for illiquidity in stock returns. Journal of Financial Economics, 41(3), 441-464.

Brennan, M. J., Chordia, T., & Subrahmanyam, A. (1998). Alternative factor specifications, security characteristics, and the cross-section of expected stock returns. Journal of Financial Economics, 49(3), 345-373.

Brennan, M. J., Chordia, T., Subrahmanyam, A., & Tong, Q. (2012). Sell-order liquidity and the cross-section of expected stock returns. Journal of Financial Economics, 105(3), 523-541.

24

Brennan, M. J., Huh, S. W., & Subrahmanyam, A. (2013). An analysis of the Amihud illiquidity premium. Review of Asset Pricing Studies, 3(1), 133-176.

Brunnermeier, M. K., & Pedersen, L. H. (2009). Market liquidity and funding liquidity.  Review of Financial studies, 22(6), 2201-2238.

Campbell, J. Y., Ramadorai, T., & Schwartz, A. (2009). Caught on tape: Institutional trading, stock returns, and earnings announcements. Journal of Financial Economics, 92(1), 66-91.

Carhart, M. M. (1997). On persistence in mutual fund performance. Journal of Finance, 52(1), 57-82.

Chan, L. K. C., & Lakonishok, J. (1993). Institutional trades and intraday stock price behavior. Journal of Financial Economics, 33(2), 173-200.

Cheng, S., Hameed, A., Subrahmanyam, A., & Titman, S. (2016). Short-term reversals: The effects of institutional exits and past returns. Journal of Financial and Quantitative Analysis, Forthcoming.

Chordia, T., Huh, S. W., & Subrahmanyam, A. (2009). Theory-based illiquidity and asset pricing. Review of Financial Studies, 22(9), 3629-3668.

Chordia, T., Roll, R., & Subrahmanyam, A. (2000). Commonality in liquidity. Journal of Financial Economics, 56(1), 3-28.

Chordia, T., Roll, R., & Subrahmanyam, A. (2001). Market liquidity and trading activity. Journal of Finance, 56(2), 501-530.

Chung, K. H., & Huh, S. W. (2016). The noninformation cost of trading and its relative importance in asset pricing. Review of Asset Pricing Studies, Forthcoming.

Datar, V. T., Y Naik, N., & Radcliffe, R. (1998). Liquidity and stock returns: An alternative test. Journal of Financial Markets, 1(2), 203-219.

Easley, D., & O'Hara, M. (1987). Price, trade size, and information in securities markets. Journal of Financial Economics, 19(1), 69-90.

Easley, D., Hvidkjaer, S., & O’Hara, M. (2002). Is information risk a determinant of asset returns?. Journal of Finance, 57(5), 2185-2221.

Fama, E. F., & French, K. R. (1992). The cross-section of expected stock returns. Journal of Finance, 47(2), 427-465.

Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3-56.

25

Fama, E. F., & MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical Tests.  Journal of Political Economy, 81(3), 607-636.

Fong, K. L., Gallagher, D. R., & Lee, A. D. (2014). Individual investors and broker types. Journal of Financial & Quantitative Analysis, 49(2), 431-451.

Foucault, T., Sraer, D., & Thesmar, D. J. (2011). Individual investors and volatility. Journal of Finance, 66(4), 1369-1406.

Glosten, L. R., & Harris, L. E. (1988). Estimating the components of the bid/ask spread. Journal of Financial Economics, 21(1), 123-142.

Glosten, L. R., & Milgrom, P. R. (1985). Bid, ask and transaction prices in a specialist market with heterogeneously informed traders. Journal of Financial Economics, 14(1), 71-100.

Griffin, J. M., Harris, J. H., & Topaloglu, S. (2003). The dynamics of institutional and individual trading. Journal of Finance, 58(6), 2285-2320.

Grinblatt, M., & Keloharju, M. (2000). The investment behavior and performance of various investor types: a study of Finland's unique data set. Journal of Financial Economics, 55(1), 43-67.

Gromb, D., & Vayanos, D. (2002). Equilibrium and welfare in markets with financially constrained arbitrageurs. Journal of Financial Economics, 66(2), 361-407.

Hao, Q. (2016). Is there information leakage prior to share repurchase announcements? Evidence from daily options trading. Journal of Financial Markets, 27(1), 79-101.

He, P. W., Jarnecic, E., & Liu, Y. (2015). The determinants of alternative trading venue market share: Global evidence from the introduction of Chi-X. Journal of Financial Markets, 22(1), 27-49.

Holthausen, R. W., Leftwich, R. W., & Mayers, D. (1990). Large-block transactions, the speed of response, and temporary and permanent stock-price effects. Journal of Financial Economics, 26(1), 71-95.

Huang, R. D., & Stoll, H. R. (1997). The components of the bid-ask spread: A general approach. Review of Financial Studies, 10(4), 995-1034.

Huh, S. W. (2014). The price impact and asset pricing. Journal of Financial Markets, 19(1), 1-38.

Jacoby, G., Fowler, D. J., & Gottesman, A. A. (2000). The capital asset pricing model and the liquidity effect: A theoretical approach. Journal of Financial Markets, 3(1), 69-81.

26

Jegadeesh, N. (1990). Evidence of predictable behavior of security returns. Journal of Finance, 45(3), 881-898.

Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of Finance, 48(1), 65-91.

Jones, C., (2002). A century of stock market liquidity and trading costs. Working Paper, Columbia University.

Kaniel, R., Saar, G., & Titman, S. (2008). Individual investor trading and stock returns. Journal of Finance, 63(1), 273-310.

Khan, M., & Lu, H. (2013). Do short sellers front-run insider sales?. Accounting Review, 88(5), 1743-1768.

Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica, 53(6), 1315-1335.

Lakonishok, J., Shleifer, A., & Vishny, R. W. (1992). The impact of institutional trading on stock prices. Journal of Financial Economics, 32(1), 23-43.

Lee, C., & Ready, M. J. (1991). Inferring trade direction from intraday data. Journal of Finance, 46(2), 733-746.

Linnainmaa, J. T., & Saar, G. (2012). Lack of anonymity and the inference from order flow. Review of Financial Studies, 25(5), 1414-1456.

Madhavan, A., Richardson, M., & Roomans, M. (1997). Why do security prices change? A transaction-level analysis of NYSE stocks. Review of Financial Studies, 10(4), 1035-1064.

Mitchell, M., & Pulvino, T. (2012). Arbitrage crashes and the speed of capital. Journal of Financial Economics, 104(3), 469-490.

Newey, W. K., & West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3), 703-708.

Nofsinger, J. R., & Sias, R. W. (1999). Herding and feedback trading by institutional and individual investors. Journal of Finance, 54(6), 2263-2295.

Odean, T. (1998). Volume, volatility, price, and profit when all traders are above average. Journal of Finance, 53(6), 1887-1934.

Saar, G. (2001). Price impact asymmetry of block trades: An institutional trading explanation. Review of Financial Studies, 14(4), 1153-1181.

Shleifer, A., & Vishny, R. W. (1997). The limits of arbitrage. Journal of Finance, 52(1), 35-55.

27

Sias, R. W., Starks, L. T., & Titman, S. (2006). Changes in institutional ownership and stock returns: Assessment and Methodology. Journal of Business, 79(6), 2869-2910.

Thompson, S. B. (2011). Simple formulas for standard errors that cluster by both firm and time. Journal of Financial Economics, 99(1), 1-10.

Table 1

Summary statisticsThe sample consists of common stocks listed on the ASX over the period of January 1, 1996, to December 31, 2012. Panel A presents the descriptive statistics of institutional and individual lambdas represent institutional order illiquidity and individual order illiquidity, respectively. The institutional and individual lambdas (i.e., Inst_lam and Ind_lam) are estimated for each stock in each month by regressing price changes on signed institutional and individual order flows, separately. The aggregate order flow is classified into institutional and individual components based on the broker IDs associated with each trade. The institutional and individual lambdas are then divided by stock prices from previous month and scaled up by 10 4. The table reports the number of observations, mean, 25th percentile, median, 75th percentile, standard deviation, and the percentage of institutional and individual lambdas with t-statistics exceed 1.96 and associated average t-statistics. The mean difference tests are conducted on the null hypothesis that the mean of institutional lambda equals to the mean of individual lambda. Panel B presents the mean and standard deviation of lambda estimations when market monthly excess value-weighted return is positive and negative. Panel C presents the mean of institutional and individual lambdas sorted by the market capitalization. The mean difference tests are conducted on the null hypothesis that the mean of institutional lambda equals to the mean of individual lambda, and the associated p-value is reported for each size quintile.

Panel A: Institutional and Individual lambdas  

Obs. Mean 25th Median 75th Std.Dev. %(t >1.96) t-stats

Inst_lam 69606 0.00794 0.00053 0.00251 0.00885 0.01415 78% 4.72

Ind_lam 69606 0.00677 0.00023 0.00149 0.00686 0.01343 69% 4.15

H0: Inst_lam = Ind_lam p-value < 0.0001          

Panel B: Sorted by market return at time t

  Mkt(t) > 0   Mkt(t) < 0  

Mean Std.Dev. Mean Std.Dev.

Inst_lam 0.00788 0.01397 0.00802 0.01427

Ind_lam   0.00678 0.01320     0.00676 0.01357  

H0: Inst_lam = Ind_lam p-value < 0.0001     p-value < 0.0001  

Panel C: Sorted by firm size

Small Size 2 Size 3 Size 4 Big

Inst_lam 0.01811 0.00899 0.00635 0.00479 0.00236

Ind_lam 0.01685 0.00734 0.00426 0.00270 0.00113

p-value   <0.0001 <0.0001 <0.0001 <0.0001 <0.0001      

28

Table 2

Average returns to lambda portfoliosThis table reports the average of value-weighted monthly returns for the 25 portfolios formed on firm size and institutional/individual lambda. The institutional and individual lambdas (i.e., Inst_lam and Ind_lam) are estimated for each stock each month by regressing price changes on signed institutional and individual order flows, separately. The aggregate order flow is classified into institutional and individual components based on the broker IDs associated each trade. The stocks are first sorted into five size quintiles, subsequently five lambda quintiles are further sorted within each size quintile. Panel A reports the time-series average returns for five institutional lambda differential portfolios. Panel B reports the time-series average returns for five individual lambda differential portfolios. Specifically, the return spreads are differences between highest illiquidity portfolio (Illiq 5) and lowest illiquidity portfolio (Illiq 1). The t-statistics are computed with Newey-West (1987) standard errors that are reported in parentheses. The sample covers the period from period January 1996, to December 2012.

Panel A: Institutional Lambda Average Returns Panel B: Individual Lambda Average Returns

Inst_lam Illiq 1 Illiq 2 Illiq 3 Illiq 4 Illiq 5 H-L Ind_lam Illiq 1 Illiq 2 Illiq 3 Illiq 4 Illiq 5 H-LSIZE 1 (S) 0.74 1.11 1.27 1.56 3.18 2.43 SIZE 1 (S) 0.85 1.32 1.98 2.57 3.48 2.65

(1.37) (2.40) (2.57) (3.45) (4.60) (5.07) (1.90) (2.29) (3.99) (3.50) (5.13) (5.43)

SIZE 2 0.83 0.97 1.16 1.50 2.45 1.62 SIZE 2 0.71 1.09 1.36 1.51 2.38 1.68

(1.14) (1.26) (1.59) (1.63) (3.09) (3.16) (0.50) (1.33) (1.97) (2.00) (3.05) (3.00)

SIZE 3 0.12 0.42 0.57 0.69 1.43 1.31 SIZE 3 0.19 0.25 0.63 0.52 0.98 0.78

(0.19) (0.67) (1.07) (1.09) (2.52) (2.41) (0.16) (0.43) (1.23) (0.98) (2.46) (2.35)

SIZE 4 0.26 0.46 0.55 0.65 0.88 0.61 SIZE 4 0.46 0.44 0.61 0.77 0.90 0.43

(0.54) (1.03) (1.25) (1.29) (2.36) (2.28) (0.98) (1.02) (1.26) (1.50) (2.18) (0.86)

SIZE 5 (B) 0.40 0.54 0.51 0.61 0.79 0.39 SIZE 5 (B) 0.24 0.22 0.45 0.34 0.48 0.21

  (1.23) (1.57) (1.50) (1.68) (2.11) (2.05)   (0.48) (0.43) (0.94) (0.81) (1.27) (0.82)

29

Table 3

Stock-level cross-sectional regressionsThis table reports the time-series average of individual stock coefficients from the Fama-MacBeth (1973) cross-sectional regression. In each month from 1996 to 2012, a regression is estimated with stock's return in excess of the 13-week Treasury notes rate (i.e. EXSRET0) and the Fama-French-Carhart risk-adjusted returns (i.e. EXSRET1 and EXSRET2) as dependent variables. SIZE denotes the natural logarithm of the market capitalization. BM denotes the natural logarithm of the book-to-market ratio. SH_TURN represents the natural logarithm of the share turnover. REV is the one-month short-term reversal, which is defined as the stock return over the prior month. RET2-3, RET4-6, and RET7-12 are the buy-and-hold returns on stock over the t - 2 to t - 3, t - 4 to t - 6, and t - 7 to t – 12 months prior to current month t. IVOL is the idiosyncratic volatility. Amihud is the average daily ratio of the absolute stock return to the dollar trading volume within the month. The institutional and individual lambdas (i.e., Inst_lam and Ind_lam) are estimated for each stock in each month by regressing price changes on signed institutional and individual order flows, separately. The aggregate order flow is classified into institutional and individual components based on the broker IDs associated with each trade. Panel A reports the regression results for institutional lambda. Panel B reports regression results for individual lambda. Panel C reports the results of weighted least-square (WLS) cross-sectional regressions for institutional lambda. Panel D reports the results of WLS cross-sectional regressions for individual lambda. Following Asparouhova, Bessembinder and Kalcheva (2010), the prior-month gross return (one plus the return in month t - 1) is employed as weighting variable for the WLS regression. All lambdas are divided by the stock price from previous month and then scaled up by 104. The t-statistics are computed by using Newey-West (1987) standard errors, with * represents statistical significance at the 10% level, ** represents statistical significance at the 5% level, *** represents statistical significance at the 1% level.

Panel A: Institutional Lambda Panel B: Individual Lambda

Expla. Var EXSRET0 EXSRET1 EXSRET2 Expla. Var EXSRET0 EXSRET1 EXSRET2

Intercept 0.075*** 0.074** 0.032 Intercept 0.102*** 0.101*** 0.053*

(3.41) (2.26) (1.20) (4.02) (2.81) (1.81)

Inst_lam 10.020*** 9.699*** 8.959*** Ind_lam 3.697** 3.574** 2.897

(9.17) (8.38) (6.91) (2.21) (2.06) (1.58)

SIZE -0.005*** -0.005*** -0.004*** SIZE -0.006*** -0.006*** -0.004***

(-5.56) (-3.94) (-2.96) (-5.68) (-4.90) (-3.57)

BM 0.016*** 0.014*** 0.012*** BM 0.016*** 0.014*** 0.011***

(5.50) (4.68) (4.06) (5.59) (4.83) (4.07)

SH_TURN -0.006*** -0.006*** -0.005*** SH_TURN -0.006*** -0.006*** -0.005***

(-3.92) (-3.68) (-3.15) (-4.16) (-3.81) (-3.66)

REV -0.020** -0.019** -0.018** REV -0.022** -0.020** 0.018**

(-2.29) (-2.19) (-2.03) (-2.44) (-2.39) (-2.09)

RET2_3 0.034*** 0.033*** 0.032*** RET2_3 0.038*** 0.035*** 0.027***

(5.50) (5.45) (5.20) (5.73) (5.19) (4.84)

RET4_6 0.008 0.004 0.003 RET4_6 0.007 0.006 0.004

(1.37) (0.77) (0.56) (1.22) (1.02) (0.84)

30

RET7_12 0.011*** 0.011*** 0.008** RET7_12 0.009*** 0.009** 0.007**

(3.24) (2.81) (2.53) (2.74) (2.34) (2.31)

IVOL -0.043*** -0.037** -0.029** IVOL -0.048*** -0.041*** -0.033**

(-2.65) (-2.15) (-2.01) (-2.71) (-2.65) (-2.09)

Amihud 0.018*** 0.016*** 0.012** Amihud 0.024*** 0.021*** 0.016***

(3.12) (2.68) (2.27) (4.23) (3.67) (3.47)

R2 0.170 0.141 0.120 R2 0.166 0.135 0.113

Panel C: Institutional Lambda WLS Panel D: Individual Lambda WLS

Expla. Var EXSRET0 EXSRET1 EXSRET2 Expla. Var EXSRET0 EXSRET1 EXSRET2

Intercept 0.088*** 0.072*** 0.040 Intercept 0.118*** 0.099*** 0.064**

(3.58) (3.07) (1.43) (3.91) (3.15) (2.07)

Inst_lam 9.441*** 9.196*** 8.526*** Ind_lam 3.420** 3.336** 2.528

(7.91) (7.28) (6.66) (2.07) (1.97) (1.46)

SIZE -0.005*** -0.005*** -0.003** SIZE -0.007*** -0.006*** -0.004***

(-3.81) (-3.66) (-2.58) (-4.61) (-4.28) (-3.18)

BM 0.014*** 0.013*** 0.012*** BM 0.016*** 0.014*** 0.011***

(4.74) (4.56) (3.88) (5.46) (4.76) (3.82)

SH_TURN -0.006*** -0.006*** -0.005*** SH_TURN -0.006*** -0.006*** -0.005***

(-3.88) (-3.61) (-3.13) (-3.99) (-3.63) (-3.60)

REV -0.022** -0.020** -0.018** REV -0.022** -0.021** -0.019**

(-2.31) (-2.18) (-2.06) (-2.23) (-2.21) (-2.08)

RET2_3 0.035*** 0.034*** 0.032*** RET2_3 0.036*** 0.032*** 0.029***

(5.69) (5.61) (5.29) (5.59) (5.20) (4.96)

RET4_6 0.009 0.006 0.004 RET4_6 0.008 0.007 0.005

(1.58) (1.01) (0.87) (1.39) (1.18) (1.06)

RET7_12 0.012*** 0.009*** 0.008*** RET7_12 0.010*** 0.010*** 0.007**

(3.35) (2.82) (2.76) (3.11) (2.70) (2.47)

IVOL -0.043** -0.043** -0.033* IVOL -0.048** -0.047** -0.030**

(-2.21) (-2.13) (-1.76) (-2.41) (-2.33) (-2.05)

Amihud 0.017*** 0.014** 0.011* Amihud 0.023*** 0.020*** 0.018***

(3.23) (2.38) (1.94) (3.61) (3.26) (3.19)

R2 0.163 0.136 0.117 R2 0.157 0.131 0.110

31

Table 4

Stock-level cross-sectional regressions: Institutional lambda versus individual lambdaThis table reports the time-series average of individual stock coefficients from the Fama-MacBeth (1973) cross-sectional regression. In each month from 1996 to 2012, a regression is estimated with stock's return in excess of the 13-week Treasury notes rate (i.e. EXSRET0) and the Fama-French-Carhart risk-adjusted returns (i.e. EXSRET1 and EXSRET2) as dependent variables. SIZE denotes the natural logarithm of the market capitalization. BM denotes the natural logarithm of the book-to-market ratio. SH_TURN represents the natural logarithm of the share turnover. REV is the one-month short-term reversal, which is defined as the stock return over the prior month. RET2-3, RET4-6, and RET7-12 are the buy-and-hold returns on stock over the t - 2 to t - 3, t - 4 to t - 6, and t - 7 to t – 12 months prior to current month t. IVOL is the idiosyncratic volatility. Amihud is the average daily ratio of the absolute stock return to the dollar trading volume within the month. The institutional and individual lambdas (i.e., Inst_lam and Ind_lam) are estimated for each stock in each month by regressing price changes on signed institutional and individual order flows, separately. The aggregate order flow is classified into institutional and individual components based on the broker IDs associated with each trade. Panel A reports the regression results for institutional lambda versus individual lambda. Panel B reports the results of weighted least-square (WLS) cross-sectional regressions for institutional lambda versus individual lambda. Following Asparouhova, Bessembinder and Kalcheva (2010), the prior-month gross return (one plus the return in month t - 1) is employed as weighting variable for the WLS regression. All lambdas are divided by the stock price from previous month and then scaled up by 104. The t-statistics are computed by using Newey-West (1987) standard errors, with * represents statistical significance at the 10% level, ** represents statistical significance at the 5% level, *** represents statistical significance at the 1% level.

Panel A: Institutional Vs Individual Panel B: Institutional Vs Individual WLS

Expla. Var EXSRET0 EXSRET1 EXSRET2 Expla. Var EXSRET0 EXSRET1 EXSRET2

Intercept 0.060** 0.059* 0.025 Intercept 0.066** 0.062** 0.035

(2.42) (1.69) (0.90) (2.52) (2.03) (1.22)

Inst_lam 9.797*** 9.361*** 8.443*** Inst_lam 9.385*** 9.022*** 8.090***

(7.46) (7.21) (6.32) (7.36) (7.15) (5.98)

Ind_lam 1.849 1.644 1.030 Ind_lam 1.804 1.496 0.790

(1.22) (0.93) (0.87) (1.10) (0.86) (0.68)

SIZE -0.005*** -0.004*** -0.004*** SIZE -0.004*** -0.004*** -0.003**

(-4.23) (-3.85) (-2.75) (-4.07) (-3.11) (-2.28)

BM 0.016*** 0.014*** 0.012*** BM 0.014*** 0.013*** 0.011***

(5.62) (4.78) (3.91) (4.68) (4.55) (3.71)

SH_TURN -0.006*** -0.005*** -0.005*** SH_TURN -0.006*** -0.005*** -0.005***

(-3.94) (-3.56) (-3.14) (-3.87) (-3.48) (-3.13)

REV -0.021** -0.020** -0.018** REV -0.022** -0.020** -0.017*

(-2.30) (-2.20) (-1.98) (-2.26) (-2.12) (-1.83)

RET2_3 0.034*** 0.032*** 0.030*** RET2_3 0.035*** 0.032*** 0.031***

(5.21) (5.05) (4.68) (5.53) (5.25) (5.18)

RET4_6 0.008 0.005 0.004 RET4_6 0.009 0.006 0.005

32

(1.37) (0.93) (0.74) (1.58) (1.09) (0.96)

RET7_12 0.011*** 0.010*** 0.008** RET7_12 0.012*** 0.010*** 0.008***

(3.20) (2.70) (2.51) (3.09) (2.92) (2.71)

IVOL -0.059*** -0.059*** -0.042** IVOL -0.054*** -0.054*** -0.029**

(-2.78) (-2.74) (-2.13) (-2.66) (-2.60) (-2.03)

Amihud 0.020*** 0.017*** 0.015*** Amihud 0.019*** 0.016** 0.015**

(3.27) (3.12) (2.72) (3.01) (2.53) (2.45)

R2 0.179 0.150 0.128 R2 0.172 0.145 0.123

33

Table 5

Regression of announcement returns on pre-announcement abnormal level lambdasThis table reports the pooled panel regression of firm’s announcement cumulative abnormal returns on pre-announcement abnormal level lambdas and control variables. The dependent variable is the three-day [-1, +1]

cumulative abnormal return (CAR_3) around the announcement date . The SPPR value-weighted market

return is used as the benchmark for computing the CAR_3. SIZE denotes the natural logarithm of the market capitalization. BM is the natural logarithm of the book-to-market ratio. Price is the average stock price in

natural logarithm over the window [-60, -2] prior to the announcement date . SH_TURN represents the

average share turnover in natural logarithm over the window [-60, -2] prior to the announcement date .

Amihud is the average daily ratio of the absolute stock return to the dollar trading volume over the window [-

60, -2] prior to the announcement date . PastReturn is the buy-and-hold market-adjusted return over the

window [-60, -2] prior to the announcement date . ReturnStd is the standard deviation of daily stock return

over the window [-60, -2] prior to the announcement date . AL_Inst_buy/AL_Inst_sell is the pre-

announcement abnormal level institutional buy/sell lambda over the window [-10, -2]. AL_Ind_buy/AL_Ind_sell is the pre-announcement abnormal level individual buy/sell lambda over the window [-10, -2]. The t-statistics are computed by using Thompson (2011) robust standard errors clustered by firm and event.

Unscheduled Announcements

Expla. Var CAR_3 CAR_3 Expla. Var CAR_3 CAR_3

Intercept 0.1275*** 0.1198*** Intercept 0.1151*** 0.1077***

(3.63) (2.88) (3.43) (2.71)

AL_Inst_buy 0.1016** 0.0969* AL_Inst_sell -0.1396*** -0.1330***

(2.01) (1.92) (-2.86) (-2.78)

AL_Ind_buy 0.0326 0.0213 AL_Ind_sell -0.0341 -0.0102

(0.60) (0.39) (-0.48) (-0.13)

SIZE -0.0026 -0.0021 SIZE -0.0023 -0.0019

(-1.10) (-0.82) (-1.02) (-0.77)

BM -0.0086 -0.0033 BM -0.0071 -0.0031

(-0.93) (-0.36) (-0.76) (-0.32)

Price 0.0030 0.0028 Price 0.0025 0.0023

(1.32) (1.23) (1.10) (1.06)

SH_TURN -0.6753 -0.4772 SH_TURN -0.5538 -0.3509

(-0.95) (-0.68) (-0.81) (-0.52)

PastReturn -0.0191** -0.0158* PastReturn -0.0188** -0.0156*

(-2.31) (-1.95) (-2.28) (-1.91)

ReturnStd 0.3548*** 0.3401*** ReturnStd 0.3504*** 0.3363***

(2.68) (2.62) (2.66) (2.59)

Amihud 0.0641 0.0547 Amihud 0.0617 0.0514

(1.27) (1.09) (1.19) (1.06)

34

Year FE No Yes Year FE No Yes

Industry FE No Yes Industry FE No Yes

R2 0.0328 0.0559 R2 0.0349 0.0582

Fig. 1. Individual and institutional lambdas around the S&P/ASX 200 Index quarterly rebalancing announcement.This figure shows the averages and 95% confidence intervals for individual buy/sell lambda, institutional buy/sell lambda on removed stocks around the S&P/ASX 200 Index quarterly rebalancing announcement. The horizontal axis represents trading days surrounding the announcement date. Panel A presents the average magnitude of institutional buy lambda on removed stocks around the rebalance announcement date. Panel B presents the average magnitude of institutional sell lambda on removed stocks around the rebalance announcement date. Panel C presents the average magnitude of individual buy lambda on removed stocks around the rebalance announcement date. Panel D presents the average magnitude of individual sell lambda on removed stocks around the rebalance announcement date. The sample period includes all S&P/ASX 200 Index quarterly rebalancing removal over the period September 2000 to December 2012.

35

36

37