International Journal of Managerial Finance

Do investors herd intraday in Australian equities?

Julia Henker Thomas Henker Anna Mitsios
Article information:
To cite this document:
Julia Henker Thomas Henker Anna Mitsios, (2006),"Do investors herd intraday in Australian equities?",
International Journal of Managerial Finance, Vol. 2 Iss 3 pp. 196 - 219
Permanent link to this document:
http://dx.doi.org/10.1108/17439130610676475
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

Downloaded on: 17 December 2014, At: 09:03 (PT)

References: this document contains references to 36 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 718 times since 2006*
Users who downloaded this article also downloaded:
Spyros Spyrou, (2013),"Herding in financial markets: a review of the literature", Review of Behavioural
Finance, Vol. 5 Iss 2 pp. 175-194 http://dx.doi.org/10.1108/RBF-02-2013-0009

Access to this document was granted through an Emerald subscription provided by 277067 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.

*Related content and download information correct at time of download.

 The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1743-9132.htm

IJMF
2,3 Do investors herd intraday in
Australian equities?
Julia Henker and Thomas Henker
196 School of Banking and Finance, University of New South Wales, Sydney,
Australia, and
Anna Mitsios
Investment Banking Group, Macquarie Bank, Sydney, Australia
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

Abstract
Purpose – The purpose of this research is to consider whether market wide herding occurs intraday.
Design/methodology/approach – Using the 1995 Christie and Huang and the 2000 Chang et al.
models, the paper tests whether market wide and industry sector herding occurs intraday in the
Australian equities market.
Findings – Neither market wide nor industry sector herding occurs intraday.
Research limitations/implications – Both herding measures focus on one specific type of
herding, herding evidenced by changes in the cross-sectional return distribution. Therefore the
herding measures are ill suited to capture the effects of period specific abnormally high or low market
returns and they can also capture herding of market participants or groups of market participants only
in as far as it manifests itself in security specific returns.
Originality/value – No previous studies have considered the possibility of intraday herding in
equities markets. Even if there is little evidence of herding over longer time periods, market frictions
and inefficiencies continue to be exploited at least anecdotally by traders with very short time horizons
to the detriment of longer term investors.
Keywords Equity capital, Finance, Australia, Investors
Paper type Research paper

Herding among investors is a common behavioral explanation for the excess

variability and short-term trends observed in financial markets. Herding is difficult to
measure, in part because a common definition requires knowledge of investors’ reasons
for trading. Some researchers, led by Christie and Huang (1995), develop measures to
directly test for the impact of herding behavior on asset prices. However, these studies
employ relatively coarse data frequencies, prompting the critique that they may miss
intra-interval herding. We address this gap in the literature. We use high frequency
intraday data to test for market wide and industry sector herding. We find that
participants in the Australian equity market do not herd, but rather discriminate
between securities as predicted by the rational asset-pricing paradigm.

1. Introduction
The tendency for humans to imitate the behavior of others is a phenomenon observed
International Journal of Managerial in a myriad of social contexts, including the actions of agents in economic settings.
Finance Immitative behavior in capital markets is called “herding”. Nofsinger and Sias (1999,
Vol. 2 No. 3, 2006
pp. 196-219
q Emerald Group Publishing Limited
1743-9132
The authors wish to thank an anonymous referee and Arnold R. Cowan for helpful comments,
DOI 10.1108/17439130610676475 and SIRCA and the ASX for providing the data.
 p. 2263) define herding as “a group of investors trading in the same direction over a Do investors
period of time”, while Banerjee (1992, p. 798) proposes a herd involves “everybody herd intraday?
doing what everyone else is doing even when their private information suggests doing
something else”.
Anecdotal evidence suggests that herding behavior is prevalent in financial
markets, evoking concerns about potential mispricing of individual assets. Most
empirical research, however, finds weak or no evidence of herding in the market prices 197
of developed capital markets. We hypothesize that these studies do not find herding
because the interval they consider is too long, ranging from daily to quarterly. Trading
in developed markets is vigorous and continuous throughout the trading day.
Day-trading has become an occupation and traders on a trading floor can see what one
another buy and sell; experienced electronic screen traders recognize one another’s
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

trading. All market participants are continuously aware of general movements in the
market. Furthermore, information revealed during a day, or especially during the
longer intervals used in some studies, may overshadow any information obtained from
observing the positions of other fund managers (Radalj and McAleer, 1993). Even if
there is little evidence of herding over longer time periods market frictions and
inefficiencies continue to be exploited at least anecdotally by traders with very short
time horizons to the detriment of longer term investors. To determine whether there is
such herding in developed markets we must search for it during the trading day.
We investigate whether herding is present in the market as a whole or within
specific industry sectors by adopting a definition of herding, proposed by Christie and
Huang (1995) and used by Chang et al. (2000), as the process through which market
participants trade based on the collective actions of the market rather than their private
expectations. That is, we investigate a market wide herding in which investors follow
the performance of the market and buy or sell without regard for the individual
characteristics of the stocks. While this definition differs from that in which a
subgroup of investors imitate one another’s actions, our type of herding also inevitably
results in the mispricing of individual assets due to the suppression of investors’
equilibrium beliefs. We also divide our sample into industry sectors to search for
herding effects within industries. In all cases, except that of the Property Trust
industry sector, we find that investors do retain the proclivity to judge stocks
individually. We can assert that market wide herding does not occur intraday in the
Australian equity market nor within the industry sectors.
The remainder of the paper is organized as follows. The next section describes
previous efforts to model and empirically identify market wide herding behavior. The
third section describes the theoretical models and the empirical equivalents that we
estimate. Next we describe our intraday data set and the variable specifications.
Section five reports our results, and section six concludes.

2. Herding measures in cross-sectional data

The existence of herd behavior among specific participants in speculative markets is the
subject of a number of studies[1]. A difficulty encountered in many of these studies is
distinguishing between irrational herding behavior and homogeneous expectations, both
of which could lead an investor to make the same trades as another. Christie and Huang
(1995) take another approach and consider aggregate market herding in equity return
data. They measure the market impact of herding by considering the dispersion or the
 IJMF cross-sectional standard deviation (CSSD) of returns[2]. The rationale for the use of this
2,3 dispersion measure is that if market wide herding occurs, returns on individual stocks
will be more than usually clustered around the market return as investors suppress their
private opinion in favor of the market consensus. Traditional asset pricing theory
predicts that the dispersion of returns increases with the aggregate market return due to
varying stock sensitivities to market returns. Since dispersion measures the average
198 proximity of individual returns to the mean, when all stock returns move in perfect
unison with the market, dispersion is zero. When individual returns differ from the
market return, however, the level of dispersion increases. Christie and Huang (1995)
contend that if investors ignore the idiosyncratic features of stocks, security returns will
be “swept along” with the market, resulting in a lower than average level of dispersion
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

during periods characterized by large market movements. Using daily and monthly
returns on US equities, Christie and Huang (1995) find a higher level of dispersion around
the market return during large price movements, evidence against herding.
Chang et al. (2000) modify the Christie and Huang (1995) model. They use the
cross-sectional absolute standard deviation (CSAD) of returns as a measure of
dispersion to find herding in the USA, Hong Kong, Japanese, South Korean and
Taiwanese markets. Their model suggests that if market participants herd around
indicators, a nonlinear relationship will result between the absolute standard deviation
of returns and the average market return during periods of large price movements. By
including an additional regression parameter, Chang et al. (2000) develop a more
sensitive means of identifying herding than that of Christie and Huang (1995). They
use their model to examine individual returns on a monthly basis and find a significant
nonlinear relationship between equity return dispersion and the underlying market
price movement of the South Korean and Taiwanese markets. They do not, however,
find evidence to support the presence of herding in the developed markets of the USA,
Hong Kong, and Japan.
The data frequency of these studies precludes the discovery of herding that occurs
within the trading day. The obvious response to this criticism is to consider intraday
data. In a recent study, Gleason et al. (2004) use intraday US Exchange Traded Funds
(ETF) data with the Christie and Huang (1995) and Chang et al. (2000) models to
examine whether traders herd during periods of extreme market movements. They
find no evidence of herding in this specialized market. However, ETFs are basket
securities, which display different dispersion characteristics than shares. We examine
a traditional equity market on an hourly basis for evidence of market wide herding.
We extend the analysis to examine industry sectors for evidence of herding behavior.
Sharma et al. (2004) argue that herding would be more likely within industry sectors than
across the entire market, and Moskowitz and Grinblatt (1999) provide evidence of a
strong industry momentum effect in the industry components of stock returns,
supporting the idea of herding at an industry level. However, aside from Christie and
Huang’s (1995) study and Sharma et al.’s (2004) investigation of herding in new economy
stocks during the period 1998 to 2000, literature considering the tendency for investors to
herd within specific industries is scant. We sub-set our data set by industry code and test
for intraday sector wide herding over a broad range of industries.
Additional motivation for our use of high frequency data comes from the volatility
literature. The fat tails of the distribution of stock returns correspond to large
fluctuations in prices or “bursts of volatility”. These fluctuations, difficult to explain in
 terms of variations in fundamental economic variables (Shiller, 1989), and not Do investors
necessarily related to the arrival of information (Cutler et al., 1989), could be explained herd intraday?
as herding. If a large number of agents co-ordinate their actions, the imbalance between
buy and sell orders will cause a substantial price change (Bouchaud, 2002). Bouchaud
(2002) presents a dynamic model of herding that accounts for volatility clustering by
describing the collective behavior of a set of traders exchanging information but
having heterogeneous opinions. During times when imitation is strong, a discontinuity 199
appears in the evolution of opinion, denoted a “crash time”. In a crash time, a portion of
the population simultaneously changes opinion, leading to a succession of bull and
bear markets with strong non-Gaussian behavior.
Volatility clustering is not limited to daily time series. It has been studied in the
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

context of intraday returns, strengthening the proposition that herding is an intraday

phenomenon. Mian and Adam (2001) examine the dynamic behavior of volatility of the
intraday equity index returns of the Australian Stock Exchange (ASX). They find that
the volatility of Australian equities follows an L-shaped curve over the trading day,
distinct from the U-shaped pattern documented by previous studies in other markets.
They also find that with increasing frequencies, the distribution of returns becomes
more non-normal. Fama (1976) and Oldfield and Rogalski (1980) also show that return
distributions become increasingly non-normal at increasingly fine intervals.
Cont and Bouchard (2000) provide a quantitative link between the fat tails observed
in the distribution of stock market returns and herding behavior, asserting that the
failure of purely statistical explanations to account for the presence of heavy tails
suggests the existence of a more fundamental market mechanism common to all
speculative markets. Their model indicates that excess kurtosis increases as the degree
of herding among market participants increases. Furthermore, the distribution of
returns is more Gaussian when price changes over longer intervals are considered,
again implying that herding behavior may be more prevalent when high frequency
data are analyzed.
Past empirical studies of herding behavior consider herding among specific market
participants; however, few studies investigate herding within the aggregate equity
market. Furthermore, those studies that do focus on market wide herding in individual
stocks generally use daily or lower frequency data, thereby missing any herding that
occurs during the trading day. Our paper contributes to existing literature on
aggregate market herding by exploring this gap in previous research.

3. Identifying market wide herding

We adapt two existing models to examine the intraday herding behavior of market
participants in the Australian equity market. The first model from Christie and Huang
(1995) derives a theoretical relationship between cross-sectional dispersion (CSDt),
portfolio volatility and average volatility. The second model, by Chang et al. (2000),
though based on the same logic, adds additional power to the empirical tests.
Both herding measures focus on one specific type of herding, herding evidenced by
changes in the cross-sectional return distribution. Therefore the herding measures are
ill suited to capture the effects of period specific abnormally high or low market returns
and they can also capture herding of market participants or groups of market
participants only in as far as it manifests itself in security specific returns.
 IJMF Christie and Huang (1995) define the cross-sectional dispersion at time t as:
2,3 X
n
CSDt ¼ wit ðr it 2 r pt Þ2 ð1Þ
i¼1

where rit (rpt) is the return of security i (portfolio p) for time t and wit is the weight of
200 each stock i in portfolio p at time t. When all securities in the portfolio move in concert
CSDt is zero; conversely, CSDt is large when the distribution of rit is dispersed. That is,
CSDt quantifies the average proximity of individual returns to the realized average
(Chang et al., 2000). Taking expectations and performing several transformations (see
Appendix 1 for the derivation) yields the following relationship between equity return
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

dispersions, average volatility, and portfolio volatility:

X
n
PVOL ¼ AVOL 2 EðCSDÞ þ wi E½r i
2 2 E½r p
2 ð2Þ
i¼1

If the average volatility of securities comprising the portfolio is assumed to be

exogenous, then the volatility of the portfolio will be an increasing function of the
average volatility of component securities, while portfolio volatility will be negatively
related to the expected cross-sectional dispersion E[CSD] of component security returns.
An increase in portfolio volatility should generate a decrease in the dispersion of returns.
If portfolio volatility is assumed to be exogenous, then E[CSD] is positively related to the
average volatility of securities. If we define market wide herding to be when all securities
in the (market) portfolio move together, then during periods where herding behavior
prevails average volatility will be low and therefore dispersion will also be low.
Christie and Huang (1995) use this decomposition to arrive at a test for herding
under extreme market conditions, where herding is defined as traders ignoring their
private assessment of individual assets and following the trend of the overall market.
Thus, if herding occurs, individual returns will converge to the aggregate market
return, resulting in decreased dispersion of individual returns from the market return
(Gleason et al., 2004). The findings of Nofsinger and Sias (1999) indicate that
institutions herd together and trade with the momentum of the market on days when
there are large movements in the aggregate market[3]. Goetzman (1995) expands on
this result when he describes investor behavior during market fads and crashes. A fad
occurs when stock prices are apparently moving together more than normal and stock
returns across the entire market are increasing. During a fad, the cross-sectional
variation would be low. Goetzman (1995) argues that this reduced variation would also
be expected with investor mass pessimism, for instance, during a panic or crash. In
markets not marked by either euphoria or pessimism the cross-sectional variation
would be expected to be higher. Therefore, we examine trading intervals characterized
by large swings in average prices, in which a lower than expected level of
cross-sectional variation would indicate herding.
Rational asset pricing models and herding behavior propose distinct predictions
regarding the behavior of the cross-sectional standard deviation of returns during
periods of markets stress. Rational asset pricing models predict that during extreme
market movements, the differing sensitivities of individual securities to the market
return will result in an increase in dispersion. The existence of market wide herding
 behavior, in contrast, dictates that dispersions would be relatively low in the presence Do investors
of large market movements. We test for herding, or alternatively for rational asset herd intraday?
pricing, by estimating the following empirical specification:

CSSDt ¼ a þ b L DLt þ b U DU
t þ 1t ð3Þ

where CSSDt is the cross-sectional standard deviation at time t, and is defined as: 201
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 1 X n
CSSDt ¼ t ðr it 2 r pt Þ2 ð4Þ
n 2 1 i¼1
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

and DLt ¼ 1 if the market return on day t lies in the extreme[4] lower tail of the
distribution and zero otherwise; DU t ¼ 1 if the market return on day t lies in the
extreme upper tail of the return distribution and zero otherwise.
The a coefficient captures the average level of dispersion of the sample, excluding
the regions covered by the two dummy variables. The dummy variables act to capture
differences in investor behavior during extreme up or down versus relatively normal
markets. Herding is indicated by significantly negative coefficients for these dummy
variables. Conversely, rational asset pricing models predict significantly positive
coefficients for b L and b U.
Because non-normality of returns and the fat tails of return distributions affect
standard deviation metrics more than they affect absolute deviation measures, and
because standard deviations are inherently more sensitive to outliers than are mean
absolute deviations, Chang et al. (2000) propose an alternate model to identify herding.
In the framework of the conditional version of the capital-asset pricing model (CAPM),
the absolute value of the deviation from the tth period portfolio expected return (AVD)
of security i’s expected return in period t is defined as:

AVDi;t ¼ jbi 2 bm jE t ½r m 2 r j
ð5Þ

Forming the arithmetic mean of the AVDi,t’s over all assets i results in a measure of the
average absolute value deviation, a proxy for the expected cross-sectional absolute
value deviation, E[CSAD]:

1X n
E½CSADt
¼ jbi 2 bm jE t ½r m 2 r f
ð6Þ
n i¼1

If all market participants estimate prices according to the conditional CAPM, i.e. in the
absence of herding, the E[CSAD] will be positively and linearly related to the expected
market return Et[rm,t]. Taking the first and second derivatives of E[CSAD] with respect
to the market return reveals the increasing and linear relation between dispersion and
the time-varying expected market return as predicted by rational asset pricing models:

›E½CSADt
1 X N
¼ jbi 2 bm j . 0 ð7Þ
›E t ½r m
N i¼1
 IJMF ›2 E½CSADt

¼0 ð8Þ
2,3 ›E t ½r m
2

However, when market participants herd, expected security returns will not deviate far
from the expected overall market return. Thus, expected returns will no longer satisfy
202 the fundamental conditional CAPM equation and E[CSADt] will not increase linearly
Et[rm,t], rather, the relationship will become nonlinearly increasing or even decreasing.
To test the model represented by equation (6), we replace the expected quantities
with the realized return of our market portfolio rp,t and the realized cross-sectional
absolute deviation of returns CSADt and estimate the model:
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

CSADt ¼ a þ g1 jr p;t j þ g2 ½r p;t
2 þ 1t ð9Þ

We predict that market participants are more likely to herd during periods of large
price movements. Thus, we would expect a less than proportional increase in CSADt
for extreme values of rp,t. The coefficient g2 captures any nonlinear relations between
CSADt and rp,t.
Chang et al. (2000) report an asymmetry in investors’ reactions to news. They show
that the rate of increase in dispersion as a function of aggregate market returns is
higher when the market is advancing than when it is declining, reflecting an increased
likelihood for herding to occur during periods of down-market stress. McQueen et al.
(1996) report related results, finding that of the common NYSE stocks they study, both
large and small stocks react quickly to negative macroeconomic news, but small stocks
adjust to positive news about the economy with a delay. This lagged response to good
news but not bad news is evident in monthly and weekly return series. Consistent with
this current and lagged asymmetry, Lamoureux and Panikkath (1994) find that the
CSSD of daily returns is asymmetric between large up and down swings in the market.
Studies of the behavior of institutional investors also document this asymmetry.
Grinblatt et al. (1995) find that fund managers follow asymmetric momentum
investment strategies only after good news. Keim and Madhavan (1995) find that
institutional traders spread buy orders over longer periods than equivalent sell orders.
The relative impatience on the part of sellers may be due to a higher perceived risk of
failing to sell in a declining market. Furthermore, few investors have a natural short
position, and many face restrictions on short sales. Thus, while traders can choose
among many potential assets to buy, they usually limit their sales to those assets they
already own (Keim and Madhavan, 1995).
An asymmetric market reaction to good news and bad news is also a characteristic
of intraday returns. Lockwood and McInish (1990) find that intraday returns are
significantly more volatile during bear markets than during bull markets. We test for
an intraday asymmetric reaction to news with both of the models we estimate. For the
Christie and Huang (1995) model, we examine the relationship between the coefficient
estimates b L and b U. For the Chang et al. (2000) model, an expected asymmetric
reaction to news would be indicated by a rejection of the null hypothesis that g UP 1 ¼
g DOWN
1 and g UP
2 ¼ g DOWN
2 , with the prediction that the rate of increase of CSAD in an
up-market will be greater than that in a down-market.
 4. Data and estimation technique Do investors
The data for this study were complied by the Securities Industry Research Centre of herd intraday?
Asia-Pacific (SIRCA) from the Stock Exchange Automated Trading System (SEATS) of
the Australian Stock Exchange (ASX). The original sample comprised the 200 largest
ASX stocks by market capitalization for the year 2001 to 2002. Stocks trading for less
than A$0.50 and stocks that trade on fewer than 200 trading days are removed from the
sample. The resulting sample of 160 stocks includes the most actively traded stocks on 203
the ASX and all stocks with a market capitalization of more than A$0.5 billion[5].
The prices are observed at five hourly intervals of the trading day, starting at 10:30
a.m. We exclude the first half hour of trading from the analysis due to the nature of the
sequential opening auction procedure used on the ASX[6]. This data structure also
excludes the last half hour of the trading day, which avoids capturing volatility
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

associated with the close. To enable a comparison to other studies, we also perform the
analysis on a daily interval from 10.30 a.m. to 3.30 p.m.[7].
The midpoint return is defined as the natural logarithm of the average of the bid
and ask quotes at the end of the interval divided by that at the beginning of the
interval. The market return is defined as the equally weighted average of the midpoint
returns of all the firms in the sample. We use an equally weighted portfolio return
rather than a value weighted return for our investigation because the market
capitalization of the Australian equities market is heavily concentrated. The largest 20
companies represent 46 percent of the equity market capitalization in Australia. The
equally weighted portfolio emphasizes the impact of smaller capitalization stocks.
Lakonishok et al. (1992) and Wermers (1999), measuring herding of fund managers,
suggest that herding may be more prevalent for small company stocks. While their
measure differs from the one we use, we might intuitively expect more herding in
smaller stocks if information about them is less readily available.
The cross-sectional standard deviation of returns as defined by Christie and Huang
(1995) is measured as:

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N
uX
u ðr i;t 2 r p;t Þ2
u
t i¼1
CSSDt ¼ ð10Þ
N 21

where ri,t is the observed return on firm i and rp,t is the market portfolio return or the
cross-sectional average of the N firms in the portfolio. The cross-sectional absolute
deviation of returns as defined by Chang et al. (2000) is expressed as:

1X N
CSADt ¼ jbi 2 bp kr i;t 2 r f j ð11Þ
N i¼1

where bi is the time-invariant systematic risk of the security, bp is the systematic risk
of the equally weighted portfolio, ri,t is the observed return on firm i and rf is the
risk-free rate of interest. In Appendix 2 we show that (11) is equivalent to:
 IJMF 1X N

2,3 CSADt ¼ jr i;t 2 r p;t j ð12Þ

N i¼1

We estimate (12) and thereby avoid estimating individual security betas.

Finally, to test for industry-wide herding, we form ten equally weighted industry
204 portfolios for each of the hourly intervals. Each firm in the sample is assigned to one of
ten economic sectors as defined by the Global Industry Classification Standard (GICS).
To ensure that both the intraday and corresponding daily portfolios contained at least
20 securities in any particular day, four sectors, the Energy, Healthcare,
Telecommunications, and Utilities industries, are excluded from the intraday data
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

set and five, the aforementioned four plus the Consumer Discretionary industry, are
excluded from the daily data set. We then estimate the models specified in equations (3)
and (9) for each of the remaining industry sectors.
We use the Generalized Method of Moments (GMM) of Hansen (1982) to estimate the
parameters because the dispersion variables used as the dependent variable in our
models display significant levels of serial correlation and a non-normal distribution.
We use the regressors themselves to identify the parameters in the models.

5. Results
Our study supports the predictions of rational asset pricing models. With one possible
exception, we find no evidence of market wide herding with either model, regardless of
whether we consider intraday or daily data, the entire market or sectors. Table I
summarizes the hourly return, volume, closing price, and market capitalization
statistics of individual stocks within the sample. The sample spans the years 2001 and
2002 and consists of observations from 160 frequently traded stocks from the ASX 200
index. It consists of 62,641 (476,638) daily (intradaily) observations where at least one
trade took place during the period. The average price is approximately $5.20 and the
average market capitalization of stocks within the sample is $3,405 million. In Table II,
we report univariate statistics for the intraday and daily market returns and the CSAD
and CSSD of returns. As expected, the magnitude of the dispersion measures is higher
for the daily data than for the intraday data. The intraday CSAD dispersion measure
displays a lower mean and less variability than the intraday CSSD dispersion measure,
confirming the findings of previous literature which indicates that standard deviation
metrics are inherently more sensitive to outliers than are mean absolute deviations
(Granger and Ding, 1993). To test for stationarity in the dispersion series, we conduct
the Dickey and Fuller (1979) test. The null hypothesis of a unit root is rejected, however
the autocorrelation evident in the dispersion series supports the choice of GMM as the
parameter estimation technique.
Table III reports the GMM parameter estimates and the heteroskedasticity
consistent t-statistics generated from the estimation of equation (3), the Christie and
Huang (1995) model, using the upper and lower 2 percent and 5 percent of the market
return distributions as the extreme price movements. The coefficients bL and bU,
which capture the change in investor behavior associated with extreme downward and
upward movements, are positive and significant at the 1 percent[8] level, indicating
that the CSSD increases during periods of market stress. This result supports a rational
asset-pricing regime. The b L and b U for the intraday data are smaller than the
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

Sample Variable N Mean Standard deviation Minimum Maximum

Intraday rt (%) 467,638 20.0055 0.6973 2 23.6103 12.8712

Price ($) 467,638 5.64 7.11 0.50 52.66
Volume ($) 467,638 163,411 522,603 0 50,296,638
Market cap ($millions) 467,638 3,405 7,248 36 49,746
Daily rt (%) 62,641 20.0190 1.4495 2 20.8372 15.6948
Price ($) 62,641 5.20 7.40 0.50 52.66
Volume ($) 62,641 7,121,476 18,567,243 0 542,567,990
Market cap ($millions) 62,641 3,405 7,248 36 49,746
Notes: Reports the intraday and daily mean, standard deviation and the maximum and minimum values of individual stock returns (rt), volume, the
closing price of individual stocks and the market capitalization over the sample period. The stock return variable is defined as the quote midpoint return.
The market capitalization variable is the market capitalization prevailing on 1 January of each yearly period in the sample. The sample consists of 160 of
the top (by market capitalization) 200 stocks on the ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and
illiquid (no trading for 50 days) stocks from the sample
herd intraday?
Do investors

and market capitalization

returns, price, volume,
Summary statistics of
205

Table I.
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2,3

CSSD
206
IJMF

Table II.

market return, CSAD and

Summary statistics of the
Mean Standard deviation Minimum Maximum Serial correlation at lag
Sample Variable N % % % % 1 2 3 5 20 DF test

Intraday
rn,t 2,515 20.0054 0.1148 2 0.9321 0.74163 0.1048 0.1048 0.0870 0.0725 20.0409 2 42.783
CSADt 2,515 0.3917 0.1308 0.1236 1.6660 0.4347 20.1394 2 0.1453 0.7504 0.7072 2 31.514
CSSDt 2,515 0.6597 0.2512 0.2076 2.4823 0.3946 20.0486 2 0.0553 0.6064 0.5749 2 33.049
Daily
rn,t 504 20.0233 0.1465 2 0.5622 0.6922 20.0005 0.0773 2 0.0227 20.0873 0.0009 2 22.542
CSADt 504 0.4947 0.1218 0.2573 1.5181 0.4253 0.4290 0.5134 0.2890 0.0550 2 14.226
CSSDt 504 0.8140 0.2211 0.3870 2.3102 0.2884 0.3168 0.3333 0.1745 0.0568 2 16.615
Notes: Reports the intraday and daily mean, standard deviation and the maximum and minimum values of the market return (rp,t), the cross-sectional
absolute deviation (CSADt) and the cross sectional standard deviation (CSSDt). In addition, the serial correlation of rp,t, CSADt and CSSDt is reported for
lags 1, 2, 3, 5 and 20 along with test statistics of the augmented Dickey-Fuller test. The sample consists of 160 of the top (by market capitalization) 200
stocks on the ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and illiquid (no trading for 50 days) stocks from
the sample
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2% criterion 5% criterion
Sample a bL bU Adj. R 2 t-stat. a bL bU Adj. R 2 t-stat.

Intraday 0.0065 0.0042 0.0026 0.1119 (5.56) * 0.0064 0.0027 0.0018 0.1095 (5.86)
(144.53) * * (7.27) * * (4.71) * * (143.73) * * (8.36) * * (6.51) * *
Daily 0.0079 0.0056 0.0031 0.1684 (1.78) 0.0078 0.0038 0.0021 0.1707 (2.10)
(79.63) * * (3.96) * * (2.53) * (86.05) * * (3.91) * * (3.17) * *
Notes: Reports the estimated coefficients of the following regression model using hourly and daily returns: CSSDt ¼ a þ b L DLt þ b U DU t þ 1t , where
DLt ðDUt Þ equals one if the market return at time interval t lies in the extreme lower (upper) tail of the distribution and zero otherwise. The 2% and 5%
criteria refer to the percentage of observations in the upper and lower tails of the market return distribution used to define extreme price movements. The
t-test tests the null hypothesis that bU ¼bL in each of the three cases. A total of 2,515 dispersion measures are used in the intraday regression while 505
are used in the daily estimation. The sample consists of 160 of the top (by market capitalization) 200 stocks on the ASX during the period 2001-2002. We
exclude low priced (daily closing price less than $0.50) and illiquid (no trading for 50 days) stocks from the sample. Parameters are estimated using GMM.
Heteroskedasticity-consistent t-statistics are reported in parentheses. *( * *) Indicates the coefficient is significant at the 5% (1%) level
herd intraday?
Do investors

during periods of market

Regression results of the

stress
intraday and daily CSSD
Table III.
207
 IJMF corresponding daily coefficients, indicating that intraday data generate lower levels of
2,3 dispersion, but these data also fail to generate price behavior consistent with herding.
It is of interest to note that in comparing the parameter estimates generated across the
two criteria used to define market stress, the parameter estimates are smaller for the
sample using the 5 percent criterion than they are for the sample using the 2 percent
criterion. This outcome is consistent with the findings of Christie and Huang (1995) that
208 the predictions of the rational asset-pricing model are better satisfied as the sample is
confined to more extreme market movements. It is contrary to our expectation that
dispersions would decrease when the market is subject to greater levels of stress.
A comparison of the coefficients b L and b U suggests that the increase in dispersion
resulting from market stress is consistently greater in down markets than in up
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

markets. The t-test for parameter equality confirms that this difference is significant at
the 5 percent level for the intraday data. The greater increase in dispersion during
down-markets than up-markets is contrary to our expectations. It appears that market
participants follow the pattern of trading established by aggregate market activity
during up-market phases rather than during down-market periods, suggesting that
booming markets aggregate less information and could therefore be more fragile than
market downturns. This finding challenges the assertion of Chang et al. (2000) that
there is an increased likelihood that herding will occur during periods of down-market
stress. Our findings are, however, consistent with those of Hwang and Salmon (2001),
who provide evidence that both the USA and UK markets exhibit herding towards the
market portfolio during bullish periods, with the greatest level of herding observed just
prior to the Russian Crisis of 1998. Similarly, Choe et al. (1999) document herding by
foreign investors in the Korean market prior to the 1996-1997 period of economic crisis
and find that herding is lower during the crisis period.
Table IV presents the results of the empirical estimation of the Chang et al. (2000)
model on the entire data sample, representing the whole distribution of market
sentiment, with the regression CSADt ¼ a þ g1 jRm;t j þ g2 ½Rm;t
2 þ 1t (equation (9).
This model includes a quadratic term, allowing for the possibility of a nonlinear
component to changes in dispersion. The coefficient of the absolute market return, g1,

Sample a g1 g2 Adjusted R 2

Intraday 0.0032 0.8711 17.1696 0.3371

(79.83) * * (12.19) * * (0.72)
Daily 0.0043 0.4109 92.6208 0.3814
(41.18) * * (2.15) * (1.98) *
Notes: Reports the estimated 2 coefficients of the following regression model:
CSADt ¼ a þ g1 jrp;t j þ g2 r p;t þ1t , using hourly and daily returns, where CSADt refers to the
cross-sectional absolute deviation, jrp,tj refers
2 to the absolute value of an equally-weighted realized
return of all available securities and r p;t is the squared value of this return. A total of 2,515
Table IV. dispersion measures are used in the intraday regression, while 505 are used in the daily estimation.
Regression results of the The sample consists of 160 of the top (by market capitalization) 200 stocks on the ASX during the
intraday and daily CSAD period 2001-2002. We exclude low priced (daily closing price less than $0.50) and illiquid (no trading
on the linear and squared for 50 days) stocks from the sample. Parameters are estimated using GMM.
term of the market Heteroskedasticity-consistent t-statistics are presented in parentheses. *( * *) Indicates the coefficient
portfolio return is significant at the 5% (1%) level
 is positive and significant, indicating that CSAD increases with the absolute market Do investors
return. Figure 1 illustrates this relation graphically, showing a nearly perfect cone herd intraday?
shape. If market wide herding behavior were present, at the outer edges, i.e. for the
extreme market returns, the figure would droop, pulling the cone down into, for
example, the shape of a pair of wings. The coefficient of the squared market return, g2,
for the intraday regression is not statistically significant, which indicates an absence of
herding behavior at an intraday level and suggests that traders trade away from the 209
market consensus during periods of market stress. The extra dispersion associated
with the daily data is reflected in the statistical significance of the g2 coefficient.
Table V reports the results for the estimation of equation (9) when the sample is
divided into up (down) markets, defined as the positive (negative) half of the market
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

return distribution. The coefficients of the linear term, g1, are positive and significant for
the intraday data and for the up-market daily data. The coefficients of the squared market
return, g2, are not statistically significant, with the exception of the extremely positive
coefficient for the daily down-market. These results provide strong support for the
rational asset pricing prediction that there is a linear relation between CSAD and the
absolute market return and that specifically the level of dispersion increases with the
market return. Furthermore, the statistically insignificant g UP2 and statistically significant
g UP
1 estimates support the linear relation between CSAD and the market return as
hypothesized by rational asset pricing theory. This finding is consistent with the positive
bU coefficient estimated with the Christie and Huang (1995) model, providing further
evidence that market wide herding is not a feature of the Australian equity market.
The rate of increase in the CSAD associated with a down-market, as captured by
g DOWN
2 , is larger than that of an up-market as captured by g UP 2 . The larger magnitude
of the g DOWN
2 coefficient relative to the g UP
2 coefficient provides possible evidence that
investors display asymmetric reactions to good and bad macroeconomic news. The

Figure 1.
Relationship between the
cross sectional absolute
deviation and the
corresponding equally
weighted market return
using intraday returns
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2,3

210
IJMF

Table V.

portfolio return
term of the market
Regression results of the

on the linear and squared

intraday and daily CSAD
CSADDOWN
t ¼ a þ gDOWN
1 jr DOWN
p;t jþ
UP UP UP UP 2
gDOWN
2 ½r DOWN
p;t
2 þ 1t CSADUP
t ¼ a þ g1 jr p;t j þ g2 ½r p;t
þ 1t
Test statistics
Sample a gDOWN
1 g DOWN
2 Adj. R 2 a g UP
1 g UP
2 Adj. R 2 F1 F2

Intraday 0.0032 0.8390 31.6691 0.3784 0.0032 0.9305 28.6449 0.28983 (20.57) (2 0.77)
(52.59) * * (2.43) * (0.93) (58.84) * * (9.97) * * (20.25)
Daily 0.0046 0.0307 195.6979 0.4470 0.0040 0.8195 1.9316 0.3277 (2.76) * (2 1.61) *
(33.53) * * (0.12) (2.35) * (27.75) * * (4.54) * * (0.05)
Notes: Reports the estimated coefficients of the quoted regression models using hourly and daily return data CSADt refers to the cross sectional absolute
deviation, jr DOWN
p;t jðjr UP
p;t jÞ refers to the absolute value of an equally-weighted realized return of all available securities at interval t when the market is down
2 2
(up) and ½rDOWN
p;t
ð½jr UP
p;t j
Þ is the squared value of this return. The down (up) market comprises the lower (upper) 50% percentile of observations where
DOWN DOWN
r p;t , 0ðrp;t . 0Þ. F1 tests gUP 1 2 g1 while F2 tests gUP
2 2 g2 . The sample consists of 160 of the top (by market capitalization) 200 stocks on the
ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and illiquid (no trading for 50 days) stocks from the sample.
Parameters are estimated using GMM. Heteroskedasticity-consistent t-statistics are presented in parentheses. *( * *) Indicates the coefficient is significant
at the 5% (1%) level
 F-test for the null hypotheses that g UP1 ¼ g1
DOWN
and g UP
2 ¼ g2
DOWN
is rejected for the Do investors
daily sample, though not for the intraday. Thus we find mixed support for the herd intraday?
expectation of asymmetric reaction and stronger herding in down-markets.
Although we fail to find intraday market wide herding, investors might follow one
another’s choices within industry sectors. To test for this possibility, we allocate each of
the companies in the sample to an industry sector and estimate the models for those
sectors. Table VI provides the average CSAD of each industry sector, its associated 211
standard deviation, and the average number of firms used to compute these statistics for
the period. For both the daily and the intraday data the Property Trusts Sector displays
the lowest level of dispersion, while the Industrial Sector has the highest average and
standard deviation of dispersion. The low level of dispersion evidenced in the Property
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

Trust returns reflects the consistent performance of the sector during the sample period.
Historically, Property Trusts are about 40 percent less volatile than general equities.
Table VII presents the intraday (daily) regression estimates of the Christie and
Huang (1995) model, CSSDt ¼ a þ b L DLt þ b U DU t þ 1t (equation (3), across the
various industry sectors. Under the 2 percent and 5 percent market stress criteria, the
parameters bL and bU, which capture the change in investor behavior associated with
extreme price movements, are positive and statistically significant at the 1 percent
level. The 5 percent criterion generates smaller estimates than the 2 percent criterion
across all industries. That is, dispersions are higher the more extreme the industry
portfolio return. The increase in dispersions indicates that individual security returns
do not herd around their industry average during periods of market stress. By
comparing the magnitude of the coefficients across industries, we note that the
Property Trusts Sector has the lowest average dispersion, as captured by the
coefficient a, and the smallest increase in dispersion during periods of market stress, as
reflected by the relative magnitude of the coefficients bL and bU.
Table VIII reports the coefficients of the Chang et al. (2000) model CSADt ¼
a þ g1 jRm;t j þ g2 ½Rm;t
2 þ 1t (equation (9) using hourly (daily) returns for the industry
portfolios in up- and down-markets. The negative and statistically significant g2
coefficient of the Property Trust sector in both the daily and intraday estimation
reflects a violation of the prediction of rational asset pricing. We observe a nonlinear
relation between the CSAD and the absolute return for this sector, with the CSAD
increasing at a decreasing rate as individuals suppress their own opinions in favor of
the Property Trust Industry consensus. The occurrence of herding in the Property
Trust sector could be a sample specific anomaly. Alternatively, it could be attributed to
the perception of the sector as less risky than most industries. The boom and bust of
the information technology bubble during the late nineties could have resulted in
investors and fund managers moving out of the technology sector and into the
Property Trust sector, consistent with Bikhchandani and Sharma’s (2000, p. 279)
portrayal of investors and fund managers as herds, which, at the first sign of trouble,
“flee to safer havens”. Furthermore, this explanation is consistent with the explanation
for industry momentum provided by Moskowitz and Grinblatt (1999).
The results of the nonlinear regression specification, which indicate herding exists
in the Property Trust sector, conflict with those of the dummy variable regression
reported in Table VII, where herding behavior is not evident. This inconsistency can be
attributed to the nature of the dummy variable regression model, which requires a
greater magnitude of nonlinearity in the return dispersion and mean return
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2,3

212
IJMF

Table VI.

and daily CSAD by

industry classifications
Comparison of intraday
Industry N Mean Standard deviation Minimum Maximum Average firm #
% % % %

Intraday
Materials 2,515 0.4215 0.1709 0.0905 2.0160 24
Industrials 2,515 0.4814 0.1969 0.0868 1.9112 29
Consumer Discretionary 1,963 0.3845 0.1855 0.0682 2.1849 20
Consumer Staples 2,026 0.3843 0.1691 0.0683 2.0174 21
Property Trusts 2,499 0.2139 0.1176 0.0084 1.0167 26
Financial £ Property Trusts 2,515 0.3796 0.1609 0.0945 2.2549 28
Daily
Materials 485 0.9772 0.2817 0.4377 2.4189 17
Industrials 322 1.0532 0.3347 0.4433 2.4012 16
Consumer Staples 447 0.9371 0.3319 0.2684 2.9563 16
Property Trusts 504 0.4531 0.1571 0.1027 1.2215 21
Financial £ Property Trusts 504 0.8985 0.2673 0.3631 3.0349 20
Notes: This table reports the daily and intraday mean, standard deviation, and the maximum and minimum values of each industry cross-sectional
absolute deviation (CSADt) as well as the average number of firms in each industry for the corresponding sample periods. The sample consists of 160 of
the top (by market capitalization) 200 stocks on the ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and
illiquid (no trading for 50 days) stocks from the sample. To ensure that both the intraday and corresponding period daily portfolios contained at least 20
securities in any particular day, the Energy, Healthcare, Telecommunications, and Utilities industries were excluded from the intraday and daily data set,
and the Consumer Discretionary industry was excluded from the daily data set
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2% criterion 5% criterion
Industry a bL bU Adj. R 2 F-test a bL bU Adj. R 2 F-test
Panel A: intraday
Materials 0.0062 0.0053 0.0038 0.1049 (0.73) 0.0061 0.0036 0.0032 0.1335 (0.40)
(110.26) * * (5.57) * * * (8.01) * * (111.95) * * (7.61) * * (9.36) * *
Industrials 0.0073 0.0062 0.0052 0.1127 (0.75) 0.0071 0.0049 0.0040 0.1623 (11.91) * *
(100.03) * * (9.19) * * (5.79) * * (103.07) * * (9.46) * * (10.77) * *
Cons. discretionary 0.0055 0.0089 0.0038 0.1964 (11.91) * * 0.0053 0.0054 0.0030 0.1838 (9.11) * *
(84.81) * * (7.37) * * (4.40) * * (86.26) * * (8.32) * * (6.89) * *
Cons. staples 0.0055 0.0046 0.0053 0.1328 (0.33) 0.0054 0.0030 0.0034 0.1328 (0.46)
(89.42) * * (5.98) * * (6.28) * * (87.73) * * (7.34) * * (8.08) * *
Property Trusts 0.0034 0.0025 0.0032 0.1334 (1.74) 0.0033 0.0018 0.0022 0.1293 (2.44)
(85.32) * * (8. 16) * * (6.73) * * (83.13) * * (11.26) * *8 * (9.47) * *
Financial £ Property Trusts 0.0055 0.0055 0.0051 0.1332 (1.48) 0.0054 0.0037 0.0003 0.1371 (0.00)
(93.21) * * (6.18) * * (5.99) * * (92.53) * * (7.90) * * (7.75) * *
Panel B: daily
Materials 0.0133 0.0045 0.0042 0.0351 (0.01) 0.0131 0.0032 0.0037 0.0536 (0.12)
(61.60) * * (1.66) (2.86) * * (60.60) * * (2.41) * (3.29) * *
Industrials 0.0143 0.0107 0.0067 0.0000 (1.04) 0.0142 0.0055 0.0030 0.0659 (1.21)
(62.15) * * (4.13) * * (2.32) * (48.99) * * (3.00) * * (2.18) *
Cons. staples 0.0128 0.0063 0.0058 0.0552 (11.91) * * 0.0124 0.0047 0.0063 0.1153 (9.11) * *
(53.15) * * (2.35) * (2.01) * (53.60) * * (3.00) * * (4.71) * *
Property Trusts 0.0062 0.0023 0.0025 0.0416 (0.01) 0.0062 0.0016 0.0016 0.0412 (0.02)
(48.43) * * (2.87) * * (3.32) * * (46.34) * * (4.33) * * (3.70) * *

Financial £ Property Trusts 0.0124 0.0016 0.0077 0.0642 (4.57) * * 0.0121 0.0043 0.0054 0.1140 (0.35)
(59.95) * * (1.92) (2.82) * * (61.95) * * (3.60) * * (3.77) * *
L U
Notes: Panel A reports the estimated coefficients of the following regression model: CSSDt ¼ a þ b L DLt þ b U DU t þ 1t , where Dt ðDt Þ equals one if the
market return at interval t lies in the extreme lower (upper) tail of the distribution and zero otherwise. The 2% and 5% criteria refer to the percentage of
observations in the upper and lower tail of the market return distribution used to define extreme price movements. The sample consists of 160 of the top
(by market capitalization) 200 stocks on the ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and illiquid (no
trading for 50 days) stocks from the sample. Panel B uses daily data to estimate the same regression model. To ensure that both the intraday and
corresponding period daily industry portfolios contained at least 20 securities in any particular day, the Energy, Healthcare, Telecommunications, and
Utilities industries were excluded from the intraday data set, and the Consumer Discretionary industry was excluded from the daily data set. The F-test
tests the null hypothesis that bU ¼ bL . Parameters are estimated using GMM. Heteroskedasticity-consistent t-statistics are presented in parentheses.
*( * *) indicates the coefficient is significant at the 5% (1%) level
herd intraday?

by industry classification
Do investors

during periods of market

Regression results of the

stress
intraday and daily CSSD
Table VII.
213
 Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

2,3

214
IJMF

Table VIII.

in up and down-markets
Regression results of the

by industry classification
intraday and daily CSAD
CSADDOWN
t ¼ a þ gDOWN
1 2 jr DOWN
p;t jþ UP UP UP UP 2
gDOWN ½r DOWN
þ 1t CSADUP
t ¼ a þ g1 jr p;t j þ g2 ½r p;t
þ 1t
2 p;t Test statistics
Industry a g DOWN
1 g DOWN
2 Adj. R 2 a g UP
1 g UP
2 Adj. R 2 F1 F2
Panel A: intraday
Materials 0.0033 0.5884 40.8360 0.3737 0.0032 0.7435 2.1670 0.2782 (1.41) (21.74)
(46.83) * * (9.58) * * (4.29) * * (37.63) * * (8.11) * * (0.11)
Industrials 0.0035 1.1158 27.2978 0.3696 0.0036 0.8918 22.5487 0.2942 (20.35) (0.16)
(43.92) * * (14.83) * * (20.78) (33.53) * * (6.94) * * (20.09)
Cons.discr. 0.0028 0.5618 53.5054 0.5085 0.0030 0.4542 65.1080 0.3343 (0.36) (20.70)
(34.91) * * (7.67) * * (4.52) * * (27.39) * * (3.55) * * (2.44) *
Cons. staples 0.0094 0.8023 21.6462 0.2989 0.0029 0.6810 28.4457 0.3206 (20.75) (0.78)
(32.57) * * (8.73) * * (20.10) (28.08) * * (5.03) * * (0.83)
Property Trusts 0.0012 1.2496 297.9141 0.4988 0.0011 1.2863 257.5678 0.5748 (0.39) (1.58)
(24.84) * * (17.36) * * (24.57) * * (24.33) * * (23.02) * * (24.21) * *
Financial £ Property Trusts 0.0028 0.6465 16.5512 0.3688 0.0028 0.7308 24.9382 0.3032 (0.94) (21.84)
(42.19) * * (10.33) * * (1.89) (39.57) * * (11.56) * * (20.87)
Panel B: daily
Materials 0.0094 20.190 38.9459 0.1617 0.0089 0.2185 8.8098 0.0940 (1.79) (21.45)
(26.14) * * (21.34) (3.48) * * (23.40) * * (1.17) (0.49)
Industrials 0.0094 0.3671 0.8844 0.1932 0.0088 0.3174 7.3425 0.1789 (20.24)
(20.87) * * (2.74) * * (0.21) (18.46) * * (1.93) (0.72)
Cons. staples 0.0083 0.1621 14.3011 0.2198 0.0081 0.2934 16.9105 0.1749 (0.13) (0.60)
(26.65) * * (1.55) * (4.05) * * (18.58) * * (1.43) (0.83)
Property Trusts 0.0034 0.6617 237.8068 0.2443 0.0033 0.6227 224.9816 0.3062 (0.74) (20.27)
(16.55) * * (6.04) * * (22.72) (20.84) * * (7.26) * * (23.42) * *
Financial £ Property Trusts 0.0082 0.1700 8.2419 0.1354 0.0082 20.1382 55.4591 0.3436 (2.17) * (21.31)
(32.02) * * (2.00) * (1.99) * (20.68) * * (20.61) (2.41) *
Panel A reports the estimated coefficients of the quoted regression models using hourly return data. Panel B reports the corresponding coefficients using
daily return data. CSADt refers to the cross sectional absolute deviation, jr DOWN
p;t ðjr UP
p;t jÞ refers to the absolute value of an equally-weighted realized return
DOWN 2 2
of all available securities at interval t when the market is down (up) and ½r p;t
ð½rUP p;t
Þ is the squared value of this return. The down (up) market
comprises the lower (upper) 50% percentile of observations where rp;t , 0ðrp;t . 0Þ. The sample consists of 160 of the top (by market capitalization) 200
stocks on the ASX during the period 2001-2002. We exclude low priced (daily closing price less than $0.50) and illiquid (no trading for 50 days) stocks from
the sample. To ensure that both the intraday and corresponding period daily portfolios contained at least 20 securities in any particular day, the Energy,
Healthcare, Telecommunications, and Utilities industries were excluded from the intraday and daily data set, and the Consumer Discretionary industry
DOWN
was excluded from the daily data set. The F test tests the difference in the up-market and down-market coefficients. F1 tests gUP 1 2 g1 while F2 tests
DOWN
gUP
2 2 g2 . Parameters are estimated using GMM. Heteroskedasticity-consistent t-statistics are presented in parentheses. *( * *) indicates the
coefficient is significant at the 5% (1%) level
 relationship to find evidence of herding. If we consider the regression specification Do investors
given by equation (9) for the up-market using intraday returns, the concave relation herd intraday?
estimated indicates that CSAD reaches its maximum value at the point where r p;t ¼
2ðg1 =2g2 Þ or 1.12 percent. Thus, as rp,t increases over the range where the realized
average daily returns are less (greater) than this threshold value, dispersion trends up
(down). The dummy variable regression model, CSSDt ¼ a þ b L DLt þ b U DU t þ 1t ,
requires that some, if not all, the market return values during periods of market stress 215
fall within the region where the dispersion measure trends downwards. If this would
not occur, the dummy variable coefficient bU would not be negative.

6. Conclusion
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

In general, our results favor the rational asset pricing argument in both the daily and
intraday analysis, as well as for all but one of the industry sectors. We find no evidence
of herding intraday for the entire market or for industry sectors, addressing an
important critique of the results of earlier studies. All indications support the assertion
that information is communicated efficiently to investors in the Australian equity
market. Furthermore, our results imply that market participants have a high level of
firm specific information, consistent with the findings of Bessembinder et al. (1996).
Avery and Zemesky (1998) note that herding behavior will not occur when uncertainty
within the stock market is only about the value of the underlying investment, and in
such cases, the stock market price will be informationally efficient. However, when an
additional element of uncertainty is contributed to the market, namely uncertainty
about the accuracy of information possessed by market participants, the stock price
will no longer be efficient and herd behavior arises even among rational investors.
It should be noted that the approach used to detect herding in this study, as argued
by Richards (1999), looks for evidence of a particular form of herding, i.e. herding
towards the market portfolio, and considers it only in the asset specific component of
returns. It does not consider herding manifesting itself in period specific market
returns, i.e. bubbles, and considers herding by market participants or groups of market
participants only in as far as it manifests itself in security specific returns of stocks.
Thus, the absence of evidence for this particular form of herding should not be
construed as indicating that other types of herding do not exist.

Notes
1. See, for example, Lakonishok et al. (1992), Grinblatt et al. (1995), Nofsinger and Sias (1999),
Sias (2004), and Wermers (1999) for studies investigating herding among institutional
investors.
2. For variants of the dispersion measure, refer to McEnally and Todd (1992), McInish and
Wood (1990) and Bessembinder et al. (1996).
3. Large movements in the stock market are defined as times when the absolute value of the
return of the market portfolio exceeds two percent.
4. We define extreme market movements as the lower and higher 2 percent and 5 percent tails
of the market distribution.
5. The same analyses were applied to samples comprising the largest (by market
capitalisation) 200 and the largest 100 firms listed on the ASX (no share price limitations
imposed); results are quantitatively similar and available from the authors on request.
 IJMF 6. Opening times are staggered depending on the starting letter of the ASX code for the
security, so most stocks do not have a trade determined opening price at 10.00 a.m.. Entering
2,3 orders during the opening, whether for the market as a whole or for an individual secuity, is
not allowed. As a robustness check, we performed the same analysis for intervals starting at
10.10 a.m. (by which time trading in most stocks has commenced under normal market
conditions). The results are consistent with those reported here and are available from the
authors on request.
216 7. The daily analysis was also performed on open to close returns. The results, available from
the authors on request, are similar to those of the 10.30 a.m. to 3.30 p.m. “day”.
8. The coefficient for b U with daily data is significant only at the 5 percent level.
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

References
Avery, C. and Zemsky, P. (1998), “Multidimensional uncertainty and herd behavior in financial
markets”, American Economic Review, Vol. 88, pp. 724-48.
Banerjee, A. (1992), “A simple model of herd behavior”, Quarterly Journal of Economics, Vol. 107,
pp. 797-817.
Bessembinder, H., Chan, K. and Seguin, P. (1996), “An empirical examination of information,
differences of opinion and trading activity”, Journal of Financial Economics, Vol. 40,
pp. 105-34.
Bikhchandani, S. and Sharma, S. (2000), “Herd behavior in financial markets”, IMF Staff Papers,
Vol. 47, pp. 279-310.
Bouchaud, J.P. (2002), “An introduction to statistical finance”, Physica A, Vol. 313, pp. 238-51.
Chang, E.C., Cheng, J.W. and Khorana, A. (2000), “An examination of herd behavior in equity
markets: an international perspective”, Journal of Banking and Finance, Vol. 24,
pp. 1651-79.
Choe, H., Kho, B. and Stulz, R.M. (1999), “Do foreign investors destabilize stock markets? The
Korean experience in 1997”, Journal of Financial Economics, Vol. 54, pp. 227-64.
Christie, W.G. and Huang, R.D. (1995), “Following the pied piper: do individual returns herd
around the market?”, Financial Analysts Journal, Vol. 51, pp. 31-7.
Christie, W.G. and Huang, R.D. (1995), “Relative performance measurement: the information in
equity return dispersions”, working paper, Owen Graduate School of Management,
Nashville, TN.
Cont, R. and Bouchaud, J.P. (2000), “Herd behavior and aggregate fluctuations in financial
markets”, Macroeconomic Dynamics, Vol. 4, pp. 170-96.
Cutler, D.M., Poterba, J.M. and Summers, L. (1989), “What moves stock prices?”, Journal of
Portfolio Management, Vol. 15, pp. 4-12.
Dickey, D.A. and Fuller, W.A. (1979), “Distribution of the estimators for autoregressive time
series with a unit root”, Journal of American Statistical Association, Vol. 74, pp. 427-31.
Fama, E.F. (1976), Foundations of Finance, Basic Books, New York, NY.
Gleason, K.C., Mathur, I. and Peterson, M.A. (2004), “Analysis of intraday herding behavior
among the sector ETFs”, Journal of Empirical Finance, Vol. 11, pp. 681-94.
Goetzman, W. (1995), “Discussion: on fads, crashes and asymmetric information”,
Anglo-American Finance Systems: Institutions and Markets in the 20th Century, Irwin
Publishers, Chicago, IL.
Granger, C.W.J. and Ding, Z. (1993), “Some properties of absolute return: an alternative measure
of risk”, working paper, University of California, San Diego, CA.
 Grinblatt, M., Titman, S. and Wermers, R. (1995), “Momentum investment strategies, portfolio Do investors
performance and herding: a study of mutual fund behavio”, American Economic Review,
Vol. 85, pp. 1088-105. herd intraday?
Hansen, L.P. (1982), “Large sample properties of generalized method of moment estimators”,
Econometrica, Vol. 50, pp. 1029-54.
Hwang, S. and Salmon, M. (2001), “A new measure of herding and empirical evidence”, working
paper, City University Business School, London. 217
Keim, D.B. and Madhavan, A. (1995), “Anatomy of the trading process: empirical evidence on the
behavior of institutional investors”, Journal of Financial Economics, Vol. 37, pp. 371-98.
Lakonishok, J., Shleifer, A. and Vishny, R.W. (1992), “The impact of institutional trading on stock
prices”, Journal of Financial Economics, Vol. 32, pp. 23-43.
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

Lamoureux, D. and Panikkath, S. (1994), “Variations in stock returns: asymmetries and other
patterns”, working paper, John M. Olin School of Business, St Louis, MO.
Lockwood, L.J. and McInish, T.H. (1990), “Tests of stability for variances and means of
overnight/intraday returns during bull and bear markets”, Journal of Banking and
Finance, Vol. 14, pp. 1243-53.
McEnally, W. and Todd, R. (1992), “Cross-sectional variation in common stock returns”,
Financial Analysts Journal, Vol. 48, pp. 59-63.
McInish, T.H. and Wood, R.A. (1990), “A transactions data analysis of common stock returns
during 1980-1984”, Journal of Banking and Finance, Vol. 14, pp. 99-112.
McQueen, G., Pinegar, M.A. and Thorley, S. (1996), “Delayed reaction to good news and the cross
autocorrelation of portfolio returns”, Journal of Finance, Vol. 51, pp. 889-919.
Mian, G.M. and Adam, C.M. (2001), “Volatility dynamics in high frequency financial data: an
empirical investigation of the Australian equity returns”, Applied Financial Economics,
Vol. 11, pp. 341-52.
Moskowitz, T.J. and Grinblatt, M. (1999), “Do industries explain momentum?”, Journal of
Finance, Vol. 54, pp. 1249-90.
Nofsinger, J.R. and Sias, R.W. (1999), “Herding and feedback trading by institutional and
individual investors”, Journal of Finance, Vol. 54, pp. 2263-95.
Oldfield, G.S. Jr and Rogalski, R.J. (1980), “A theory of common stock returns over trading and
non-trading periods”, Journal of Finance, Vol. 35, pp. 729-51.
Radalj, M. and McAleer, M. (1993), “Herding, information cascades and volatility spillovers in
futures markets”, working paper, University of Western Australia, Perth.
Richards, A. (1999), “Idiosyncratic risk: an empirical analysis with implications for the risk of
relative-value trading strategies”, working paper, IMF 99/148.
Sharma, V., Easterwood, J.C. and Kumar, R. (2004), “Did institutional investors herd into new
economy stocks?”, working paper, Pamplin College of Business, Blacksburg, VA.
Sias, R.W. (2004), “Institutional herding”, Review of Financial Studies, Vol. 17 No. 1, pp. 165-206.
Shiller, R. (1989), Market Volatility, MIT Press, Cambridge, MA.
Wermers, R. (1999), “Mutual fund herding and the impact on stock prices”, Journal of Finance,
Vol. 54, pp. 581-622.
 IJMF Appendix 1
We are indebted to William Christie for the following derivation, which was omitted from
2,3 Christie and Huang (1995). Begin with:
X
n
CSDt ¼ wit ðrit 2 rpt Þ2 ðA1Þ
i¼1

218 Drop the time subscript for convenience and take the expectations of equation (A1) to get:
" #
Xn h i
EðCSDÞ ¼ E wi r 2 E r 2p
2
ðA1bÞ
i¼1
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

The first term on the right hand side of equation (A2) can be rewritten as:
" #
X n Xn   X n   2 
2
E wi r i ¼ wi E r2i ¼ wi s 2 ðr i Þ þ E r i ðA1cÞ
i¼1 i¼1 i¼1
X
n
where s 2 ðri Þ is the variance of security i’s returns. Define wi ½s 2 ðri Þ
as the weighted average
i¼1
volatility of the component securities of portfolio p and the above equation can be rewritten as:
" #
X
n Xn  2
2
E wi ri ¼ AVOL þ wi E r i ðA1dÞ
i¼1 i¼1

The second term in equation (A1b) can also be rewritten using the method outlined above:
h i  2  2
E r2p ¼ s 2 ðrp Þ þ E rp ¼ PVOL þ E r p ðA1eÞ

Substitute equations (A1d) and (A1e) into equation (A1b):

  X
n  2  2
E CSD ¼ AVOL 2 PVOL þ wi E r i 2E r p ðA1fÞ
i¼1

Finally, solve for PVOL to yield the following relationship between equity return dispersions,
average volatility, and portfolio volatility:
  Xn  2  2
PVOL 2 AVOL 2 E CSD þ wi E r i 2E rp ðA2Þ
i¼1

Appendix 2
Expressions 11 and 12 can be shown to be equal by considering the following set of equations for
each stock i:
r i;t ¼ rf þ bi ðr m 2 rf Þ ðA3aÞ
and similarly,
rm;t ¼ rf þ bm ðrm 2 rf Þ ðA3bÞ
hence,
 r i;t 2 r m;t ¼ r f þ bi ðrm 2 rf Þ 2 bm ðrm 2 rf Þ 2 rf ðA3cÞ Do investors
and herd intraday?
r i;t 2 r m;t ¼ ðbi 2 bm Þðr m 2 rf Þ ðA4Þ

219
Corresponding author
Julia Henker can be contacted at: j.henker@unsw.edu.au
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com

Or visit our web site for further details: www.emeraldinsight.com/reprints
 This article has been cited by:

1. Asma Mobarek, Sabur Mollah, Kevin Keasey. 2014. A cross-country analysis of herd behavior in Europe.
Journal of International Financial Markets, Institutions and Money 32, 107-127. [CrossRef]
2. Juan Yao, Chuanchan Ma, William Peng He. 2014. Investor herding behaviour of Chinese stock market.
International Review of Economics & Finance 29, 12-29. [CrossRef]
3. Spyros Spyrou. 2013. Herding in financial markets: a review of the literature. Review of Behavioural Finance
5:2, 175-194. [Abstract] [Full Text] [PDF]
4. Bartosz Gębka, Mark E. Wohar. 2013. International herding: Does it differ across sectors?. Journal of
International Financial Markets, Institutions and Money 23, 55-84. [CrossRef]
5. Natividad Blasco, Pilar Corredor, Sandra Ferreruela. 2012. Market sentiment: a key factor of investors’
imitative behaviour. Accounting & Finance 52, 663-689. [CrossRef]
Downloaded by Rochester Institute of Technology At 09:03 17 December 2014 (PT)

6. Natividad Blasco, Pilar Corredor, Sandra Ferreruela. 2012. Does herding affect volatility? Implications for
the Spanish stock market. Quantitative Finance 12, 311-327. [CrossRef]
7. N Blasco, P Corredor, S Ferreruela. 2011. Detecting intentional herding: what lies beneath intraday data
in the Spanish stock market. Journal of the Operational Research Society 62, 1056-1066. [CrossRef]
8. Douglas M. Patterson, Vivek Sharma. 2010. THE INCIDENCE OF INFORMATIONAL CASCADES
AND THE BEHAVIOR OF TRADE INTERARRIVAL TIMES DURING THE STOCK MARKET
BUBBLE. Macroeconomic Dynamics 14, 111-136. [CrossRef]
9. Natividad Blasco de las Heras, Pilar Corredor Casado, Sandra Ferreruela Garcés. 2009. Generadores de
comportamiento imitador en el mercado de valores español. Spanish Journal of Finance and Accounting /
Revista Española de Financiación y Contabilidad 38, 265-291. [CrossRef]