The Pennsylvania State University
The Graduate School
Smeal College of Business
THE EFFECT OF EARNINGS ANNOUNCEMENTS
ON TRADING OUTCOMES FOR DIFFERENT
INVESTOR CLASSES
A Dissertation in
Business Administration
by
James Dale Vincent
Copyright 2010 by James Dale Vincent
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Doctor of Philosophy
August 2010
ii
The dissertation of James Dale Vincent was reviewed and approved* by the following:
Orie E. Barron
Professor of Accounting
Dissertation Adviser
Chair of Committee
Paul Fischer
Professor of Accounting
Karl Muller
Associate Professor of Accounting
Laura C. Field
Associate Professor of Finance
*Signatures are on file in the Graduate School.
iii
Abstract
Theory suggests that earnings announcements can either increase or decrease the level of
information asymmetry between investors. I investigate this by testing the effect of
earnings announcements on the relative ability of small and large investors to trade
advantageously. On average, I find that large traders’ returns increase following earnings
announcements at the expense of small traders, indicating an increase in information
asymmetry following earnings disclosures, at least in the short term. I also find that
several proxies for the availability of private information are associated with the
advantage of large traders following announcements. I also find weak evidence that large
investors are at a greater advantage when earnings surprises are large. However, I find no
evidence to support an effect from the presence of certain types of shareholders at
earnings announcements. Finally, I test the hypothesis of Chae (2005) that volume
patterns around earnings announcement are partially the result of less informed traders
delaying their trades until after the announcement to achieve a more level informational
playing field. Results are inconsistent with this hypothesis.
iv
TABLE OF CONTENTS
List of Figures……………………………………………………………….v
List of Tables………………………………………………………….……vi
Acknowledgements……………………………………………………...…vii
Chapter 1. INTRODUCTION………………………….……………………1
Chapter 2. RELATED LITERATURE AND HYPOTHESIS
DEVELOPMENT……………………………………………………..…….4
Literature Related to Information Asymmetry Following Earnings
Announcements………………………………………………………4
Behavior of Small and Large Traders………….……………………10
Chapter 3. VARIABLES AND RESEARCH
DESIGN……………………………………………………………………14
Chapter 4. DATA AND RESULTS…………..……………………………24
Data Procedures………………………………………..……………24
Descriptive Statistics…………………………………..……………26
Main Results……………………………………...…………………26
Sensitivity Analyses…………………………………………………44
Chapter 5. CONCLUSION……………………………...…………………48
Appendix……………………………………………………………..…….51
References………………………………………………….………………53
v
List of Figures
Figure 1……………………………………………………………….……15
vi
List of Tables
Table 1……………………………………………………………………27
Table 2……………………………………………………..……………..28
Table 3……………………………………………………………………30
Table 4……………………………………………………………..……..32
Table 5……………………………………………………………………35
Table 6……………………………………………………………………37
Table 7…………………………………………………………...……….38
Table 8…………………………………………………………………....40
Table 9………………………………………………………..…………..41
Table 10…………………………………………………………………..45
Table 11…………………………………………………………….…….47
vii
Acknowledgements
I have received an enormous amount of help and advice in the process of
producing this dissertation. I wish to thank my committee: Orie Barron, Paul
Fischer, Karl Muller, and Laura Field, who have been extremely patient and
helpful throughout the process. As an advisor and committee chair, Orie
Barron has been the best a doctoral student could hope for: insightful,
understanding, and supportive. I will always appreciate the work he has done
to help me to become a better thinker and a better researcher.
I also received many helpful comments from students and faculty at the
various schools I visited during the job search process, with Linda Bamber
and Donal Byard providing especially thorough and useful feedback. I
would also like to thank the students and faculty at Penn State, who have
been very helpful and supportive. I would like to give special thanks to
James Sinclair, who has been extremely giving of his time in helping me
through multiple revisions. More so than most, James understands that
support and criticism are complimentary, and I will always appreciate his
well thought out critiques of my work.
Finally, it would not be possible for me to be here without the support of my
family. My mother has been a great inspiration to me throughout my life and
her love and support has been a great help to me during the last few years.
My wife Amanda has also been a tremendous source of love and support
through some difficult times. She was always a wonderful companion and
helped to relieve my stress even as she was working to complete her own
dissertation. Her grace and warmth are an inspiration to me, and her strength
and fortitude throughout her pregnancy and the birth of our son are a
testament to her character. I love her deeply, and I’ll always remember her
support for me during this time. Lastly, I want to thank my father, who
encouraged me in my academic career and always took great interest in my
advancement through the degree. It is my deepest sorrow that he was not
able to see the end of the process. I dedicate this work to him.
1
1. Introduction
The theoretical literature has modeled the effect of earnings disclosures on
information asymmetry between investors in several different ways.1 On the one hand,
disclosures have been seen as reducing or eliminating uncertainty about some future
event, causing the better informed to lose some or all of their advantage over the
uninformed (e.g. Kyle, 1986; Glosten and Milgrom, 1985; Admadi and Pfleiderer, 1988).
On the other hand, disclosures can spur the generation of new private information about a
firm’s future state (Kim and Verrecchia, 1994, 1997). These two ideas are not mutually
exclusive; disclosures can simultaneously resolve uncertainty about some aspect of a
firm’s future earning power while also giving better informed investors additional inputs
from which they can refine their beliefs about future payouts. The purpose of this paper is
to examine how earnings announcements change the relative advantage of the informed
versus the uninformed, and which firm and market attributes serve to change this
dynamic.
Prior studies have examined the effect of earnings announcements on information
asymmetry using bid-ask spreads (Yohn, 1998; Libby, Mathieu and Robb, 2002), the
adverse selection component of spreads (Krinsky and Lee, 1996), the price impact of
trades (Brooks, 1996), and the probability of informed trade (Brown, Hillegeist and Lo,
2008). This paper differs from this prior work in that, rather than measuring a proxy for
the presence of information asymmetry, I measure a likely outcome of that asymmetry.
To examine the effect of information asymmetry on investors, I examine all trades
made in the stock of each firm in my sample, and calculate the three day returns that
1 In this study, I consider information asymmetry between investors, which is distinct from the information
asymmetry between the firm and outsiders. See Agarwal and O’Hara (2007) for a discussion of the
differences between these two forms of information asymmetry.
2
could be earned by the initiator of each trade. I then classify the trade size as small,
medium, and large, with the small group proxying for uninformed investors and the large
group proxying for better informed investors.2 I then compute the overall imputed returns
for each group for each trading day. Finally, I compare the relationship between small
and large returns overall with the small and large returns shortly following earnings
announcements, to see how these disclosures affect the relative returns earned by the two
groups.
Results show that overall, there is an increase in large traders’ profits relative to
small traders’ profits in the period immediately following an earnings announcement. I
also run my tests using control variables for the direction and magnitude of earnings
surprises, to mitigate concerns that results may be related to PEAD trading strategies.
Results remain the same even with the inclusion of these earnings surprise control
variables. I also examine some firm and announcement level characteristics that may be
related to the production of private information. I find that large traders are at a greater
advantage when trading in the shares of firms with large analyst following, suggesting
that large traders are better able to produce private information following earnings
announcements when incentives to produce such information is higher. Furthermore,
reduction of private information, as measured by the change in analysts’ consensus, is
associated with a reduced advantage for large traders following announcements.
However, a similar interaction using firm size, which has been used as a proxy for the
incentive to generate private information in prior literature, does not produce significant
results.
2 Following prior research, I do not use the medium group, as it is more difficult to make inferences about
the composition of this group.
3
Additionally, I test the effect of several features of earnings announcements on
the ability of large traders to achieve better returns following earnings announcements. I
find weak evidence that large earnings surprises (as measured by the difference between
analysts’ expectations and actual earnings) increases large traders’ advantage following
earnings announcements. However, I find no evidence that random walk earnings
surprises affect their advantage, nor any differential effect between positive and negative
surprises, or surprises where the analyst expectation surprise is a different sign than the
seasonal random walk earnings surprise.
I also test the effect of the presence of certain types of institutional traders on
large traders’ advantage. While, generally speaking, the presence of transitory institutions
is associated with better relative returns for small traders, and the presence of dedicated
investors is associated with better relative returns for large traders, there is no incremental
effect following earnings announcements. So while it appears that investor mix does have
a significant effect on the relative returns of more informed and less informed traders, I
find no evidence that features of institutional ownership are associated with differential
returns shortly following earnings announcements.
I also use this data to test the discretionary liquidity trader hypothesis in Chae
(2005) concerning patterns of trading around earnings announcements. This hypothesis
posits that the patterns of trading observed around earnings announcements (decrease in
volume in the period immediately prior to the announcement followed by a large increase
in volume immediately following the announcement) are due, at least in part, to
uninformed traders who wish to time their trades advantageously. According to this
hypothesis, uninformed traders with some discretion in the timing of their exogenously
4
determined liquidity trades will wish to delay trade until after the earnings
announcement, to reduce the adverse selection problem of trading against informed
investors. This implies that uninformed traders who trade immediately following the
announcement should enjoy better returns, relative to informed traders, than uninformed
traders who trade immediately prior to the earnings announcement. I test this proposition
directly by examining the difference in relative trade performance both prior to and
following earnings announcements. While I find that large traders do better than small
traders following earnings announcements, I find no difference in their relative trading
performance immediately prior to the announcement. This evidence suggests that
discretionary liquidity trade timing is unlikely to explain earnings announcement trading
patterns.
The remainder of this paper proceeds as follows. In section 2, I briefly discuss
some of the background literature and the rationale for the tests being performed. I then
discuss the construction of the variables and the research design I use to test the
hypotheses in section 3. In section 4, I discuss the data and results, and I conclude with
section 5.
2. Related Literature and Hypothesis Development
2.1 Literature Related to Information Asymmetry Following Earnings Announcements
Theoretically, there are two distinct approaches to modeling the effect of information
disclosure on information asymmetry between investors. In models such as Kim and
Verrecchia (1991) and Foster and Viswanathan (1990), some investors have private
signals concerning an upcoming announcement. The announcement supplants the private
5
information, making their private signals no longer useful, and information asymmetry is
alleviated. In contrast, in other models (Kim and Verrecchia 1994; Kim and Verrecchia
1997) information disclosures can spur the generation of new, private information, which
increases asymmetry. Although different models emphasize different features of
information announcements, these two effects are not mutually exclusive, and in fact,
complex disclosures such as earnings announcements are likely to substitute for some
privately held information while simultaneously stimulating new private information
among some investors, as in Kim and Verrecchia (1997). These two countervailing
effects make it difficult to predict ex-ante the effect of accounting earnings
announcements on information asymmetry between investors.
Researchers have used a number of proxies for information asymmetry in
studying its relationship with earnings disclosures, with bid-ask spreads, or a spread-
based construct being the most common. Other proxies include analyst consensus,
probability of informed trade (PIN), and the price impact of block trades. No general
consensus has emerged in this literature as to how the announcements affect asymmetry.
One of the first papers to study the issue of information asymmetry around
earnings announcements is Morse and Ushman (1983). This study examines bid-ask
spreads around quarterly earnings announcements, looking at each day from ten days
prior to the announcement to ten days following the announcement. These days are
individually compared to days that are further from the announcement date. No
significant differences are found for any date in this window, although this non-result
may be an artifact of the sample: 378 quarterly announcements for 25 firms over four
years. It’s also questionable how applicable this sample would be to the present day, as
6
trading volume has increased dramatically, and spreads fallen greatly in the years since
their sample ends in 1976.
Lee, Mucklow, and Ready (1994) provide a more modern and thorough
examination of the behavior of spreads (and market depth) around earnings
announcements. They show that liquidity decreases prior to earnings announcements
(and prior to large surprises especially) and remains low in the wake of the announcement
for a full day (i.e. spreads are higher than normal and quoted depth is lower than normal.)
This effect is greatly mitigated, however, when volume is controlled for, in this case
being limited to a liquidity decrease for only the first half hour following the
announcement. The authors note, however, that the effect of controlling for volume is not
perfectly clear theoretically.
Brooks (1996) approaches the question using the Hasbrouck (1991)
decomposition of price variance. Price movements are decomposed by vector
autoregression into a random walk component and a stationary component, and a
measure of the proportion of private to public information is created. The author shows
that this measure is significantly lower on earnings announcement days, although it oddly
seems to be lower on the two previous days as well. Overall, Brooks (1996) interprets his
results as being consistent with earnings announcements having an information leveling
effect on asymmetry.
Krinsky and Lee (1996) examine the issue by using a decomposition of bid-ask
spreads into its component parts. They note that prior literature examining spreads around
announcements is difficult to interpret as components of the spread other than adverse
selection may be changing in announcement periods as well. Upon decomposition, they
7
find that different spread components tend to move in different directions simultaneously,
and they attribute the previously mixed results in the literature to this fact. They find that
while the adverse selection component of the spread increases following (and prior to) an
announcement, while the inventory holding cost and order processing cost components
decrease. They interpret their results as being consistent with information asymmetry
between investors increasing following earnings announcements. While this is an
interesting result which addresses some of the difficulties of prior research with spreads,
there have recently been a number of papers critical of the many various methods used to
decompose elements of the spread. Van Ness, Van Ness and Warr (2001) examine five
different measures of adverse selection, and find that they are uncorrelated with other
measures of information asymmetry such as analyst forecast error and market to book.
The measures are also only weakly correlated with each other and with measures of the
presence of informed trade such as analyst following and institutional ownership. Neal
and Wheatley (1998) document the puzzling result that different estimations of adverse
selection components of the spread show consistently high values of adverse selection in
closed-end mutual funds, which by their nature have little to no information asymmetry
problems. These studies suggest that interpretations of decomposition based research
remain difficult.
Barclay and Dunbar (1996) use yet another proxy for information asymmetry, the
price impact of block trades, in addition to examining effective spreads (and a spread
decomposition) around announcements. While they expect that prices will be more
sensitive to block trades at times of higher asymmetry, they find no evidence that any
window around the announcement has a different sensitivity than their baseline period.
8
Similarly, they show no significant difference between their measure of spreads prior to
and following the announcement. They conclude that they find “no evidence to support
the hypothesis that the uninformed can trade on more favorable terms by varying the
timing of their trades in relation to a quarterly earnings announcement.”
Yohn (1998) reaches a different conclusion. Testing a sample of 1,989 quarterly
announcements between 1988 and 1990, she finds that spreads remain higher than normal
for both the day of the announcement and the day following the announcement. She
further documents a positive relationship between this reaction and the size of the price
reaction to the earnings announcement. She also notes that the size of the spread increase
prior to the announcement is increasing in both firm size and analyst following.
Analyst forecast information is used by Barron, Byard and Kim (2002) to
examine the issue from a different angle. Using the Barron et. al. (1998) estimation of the
ratio of analysts’ private information to analysts’ common information, they show that,
on average, earnings announcements tend to increase the proportion of idiosyncratic,
private information in analysts’ forecasts about future earnings. This suggests that
earnings announcements can be used by sophisticated investors to increase their private
information about future firm performance.
While the studies mentioned above have focused on U.S. equity markets, Libby,
Mathieu, and Robb (2002) examine earnings announcements on the Toronto Stock
Exchange. They examine spreads and quoted depth simultaneously, noting that both are
methods used by market makers to provide or withdraw liquidity. They find that liquidity
seems to increase following the announcements in their sample, although the window
being tested is large (thirty days prior to and following the announcement.) As a result, it
9
is difficult to make inferences about the short term effect of earnings announcements as a
result of this study.
Prior research relating to information asymmetry following earnings
announcements has also been used to make inferences about trading volume patterns at
announcement time. In traditional models of trading in the presence of information
asymmetry, such as Kyle (1985) and Glosten and Milgrom (1985), uninformed traders
have exogenous liquidity needs, and are unable to choose the timing of their trades. Other
models examine the effect of timing on trading outcomes for relatively uninformed
investors. In Admati and Pfleiderer (1988), some uninformed liquidity traders can
exercise discretion over the timing of their trades. This leads to trade clustering, as the
uninformed benefit from the lower sensitivity of price to trading when volume is high.
The informed traders likewise benefit from the presence of the uninformed liquidity
traders, which drives volume even higher. Chae (2005) hypothesizes that discretionary
liquidity trading may explain some of the patterns that are observed around earnings
announcements: slightly lower volume immediately prior to the announcement and very
high volume immediately following it. Examining the differences between scheduled and
unscheduled (e.g. merger) announcements, Chae (2005) notes that scheduled
announcements have greater volume reductions prior to the announcement and larger
increases in volume following the announcement than unscheduled announcements do.
They conclude that this is consistent with uninformed traders delaying their trades until
after the announcement, leading to pent-up demand that is released after the
announcement. Additionally, Chae (2005) finds stronger results when some proxies for
information asymmetry are high, implying that less informed investors are more likely to
10
delay their trades until after the announcement, when asymmetry attenuates. If, however,
information asymmetry tends to increase following earnings announcements, this calls
into question the discretionary liquidity trader hypothesis.
2.2 Behavior of Small and Large Traders
In operationalizing my tests of information asymmetry, I divide investors between
small and large traders, a proxy for the less informed and better informed investor
classes. Two streams of literature examine the behavior of smaller, less sophisticated
investors versus larger, more sophisticated ones. The first looks directly at the long term
performance of trades made by different investor classes or of stocks that are heavily
purchased or sold by individuals. The second uses the inferred net buying or selling
activities of investors to see how large and small traders react differently to various
events. Among the first stream, Barber and Odean (2000) use data from a discount
brokerage to examine the performance of individual investors and find that they
underperform the market as a whole, and that frequent traders incur large transaction
costs without any appreciable increase in performance. Hvidkjaer (2008) finds that shares
favored by small investors experience declines in both the near and long term. Barber et.
al. (2008) show that individuals trading in the Taiwan market lose systematically to
institutions and foreign players, and that their losses stem largely from active trading (i.e.,
buy or sell orders rather than limit orders). In contrast, Kaniel, Saar, and Titman (2008)
use NYSE data to examine the performance of stocks that are heavily bought or sold by
individual investors, and finds that stocks that are heavily purchased by individuals have
higher future returns. Kaniel et. al. (2009) do a similar analysis around earnings
11
announcements and find that stocks that are heavily purchased by individuals prior to an
announcement experience better returns in the coming days and months. They attribute
this to both individuals’ provision of liquidity to institutions, and superior private
information possessed by individuals.
Asthana, Balsam and Sankaraguruswamy (2004, hereafter ABS) examine
the issue of relative trade profitability of small and large traders around 10-K filing dates.
They find that small traders’ trades tend to be less profitable in the period around 10-K
filings, relative to other times, and that large traders’ trades tend to be more profitable
around these filings, though this effect is mitigated when firms file on EDGAR. There are
several ways in which the results in this paper differ from the findings in ABS. First of
all, the profitability measure in ABS is calculated as the product of net-buying of a
particular size group and the five day price change in days -1 to +4 surrounding the filing.
As a result, each buy (or sell) in the event period is taken equally, regardless of the actual
executed trading price. Thus, a trade made in anticipation of the 10-K filing is treated the
same as a trade made several days following the filing, after which market price may
have partially or fully incorporated the information in the filing.3
Secondly, the trading window examined in ABS incorporates the day prior
to the 10-K filing in addition to the period immediately following the filing. We are
therefore unable to assert whether the difference demonstrated in event versus non-event
periods shown in ABS is due to trading in anticipation of the filing or due to reaction to
3 To examine the impact of these differing methods on results, I also run my main test using a modified
version of the ABS method. In this untabulated test, I calculate my dependent variable by multiplying the
netbuying activity of the small and large groups by the overall three day price swing following the day of
calculation. For both the LG variable and the LG*EAday interaction term, this modified version of the
ABS test yields no statistically significant coefficients. This demonstrates how the increased precision of
the research design in this paper can yield different conclusions from those based on aggregations of
buying and selling activity.
12
information in the filing itself. As my research question relates to the differential use of
accounting disclosures by small and large traders, I limit my sample to the period after
the earnings announcement.
Finally, my study differs from ABS in that 10-K filings are fundamentally
different from earnings announcements. Except in the unusual cases where 10-K’s are
filed prior to an earnings announcement (Stice, 1991), 10-K filings take place after a
large portion of the financial information contained in the 10-K has already been
disseminated. Strong market reactions to 10-K filings are rare (Easton and Zmijewski,
1993) and more likely to be the result of a discrepancy with previously disclosed
financials or information contained in a footnote disclosure which was not discussed at
the earnings announcement. Since the relevant disclosures are relatively obscure, this
type of environment is likelier to favor sophisticated traders than the relatively more
straightforward earnings announcements. In other words, it is reasonable to expect, ex-
ante, that trading following earnings announcements would be a more favorable
environment for small investors than trading following 10-K filings.
The second stream of literature uses the imbalance in trade orders to make
inferences about the differing behavior of small and large investors. Lee (1992) found
that whereas large traders tend to buy on good surprises around earnings announcements
and sell on downside surprises, small traders buy on either good or bad news.
Bhattacharya (2001) shows that small traders’ abnormal trading activity is associated
with a random walk model of earnings expectation, while the same does not hold true for
large traders. Battalio and Mendenhall (2005) extend Bhattacharya (2001) by using the
inferred sign of the trade to establish a net buying/selling position for small and large
13
investors. They find similar results: small traders seem to respond to earnings
announcements based on a random walk model, while large traders respond based on
analyst expectations. Other papers have looked at the role of small investors in causing
PEAD (Hirshleifer et. al. 2008; Ayers, Li, Yeung 2009) and their interpretation of analyst
recommendations (Malmendier and Shanthikumar, 2007). While these studies document
interesting regularities in the behavior of different types of investors, they do not speak
directly to the relative trading performance of these investors following earnings
announcements.
The discussion above shows that small and large investors tend to demonstrate
systematically different trading behaviors. And while there are several different notions
about the way that disclosures of important financial information affects information
asymmetry, we know very little about how earnings announcements translate into trading
outcomes for different investor classes. This paper is an attempt to understand how
earnings announcements affect the actual trading performance of different sized
investors. Do smaller investors benefit, on average, from the revelation of formerly
private information that takes place at earnings announcements? Or does the advantage
that accrues to larger, more sophisticated investors who are able to generate private
event-period information in the run-up to earnings announcements overwhelm any
leveling effect that the announcement might have? As it is possible that either effect may
predominate, I do not form a directional hypothesis. Instead, I proceed by first testing the
sample as a whole to identify any general tendency in one particular direction.
As theory demonstrates, private information production following earnings
announcement is likely to be due to private information gathered prior to the
14
announcement in anticipation of it. I therefore also gather a number of firm and
announcement characteristics that are likely related to private information production,
and interact these variables with my measure of information asymmetry. I follow Yohn
(1998) in examining the effect of both firm size and analyst following on my main
variable of interest. The first variable I examine is firm size. Atiase (1985) proposes that
market participants will generate more private information before earnings
announcements when firm size is large, suggesting that large investors will benefit more
from earnings announcements of larger firms. By conditioning on size, I test the extent to
which the earnings announcement effect is greater or smaller for large versus small firms.
I also test the effect of analyst coverage on the relative performance of small
versus large investors. Prior literature has seen analysts as either exacerbating (Brennan
and Subrahmanyam, 1995) or alleviating (Roulstone, 2003) information asymmetry
problems. I examine the effect of analyst coverage on my variables of interest, as the
information that analysts generate may affect the ability of different investor groups to
interpret earnings disclosures.
Finally, change in analysts’ consensus (as defined in Barron et. al. 1998) has been
shown in several studies to be associated with private information production (Barron,
Byard and Kim, 2002; Botosan, Plumlee and Xie, 2004; Barron Harris and Stanford,
2005). The consensus variable measures the ratio of common to private information
available to analysts in making earnings forecasts. If consensus about future earnings
decreases around an earnings announcement, this suggests that the announcement has
stimulated the production of new private information.
3. Variables and Research Design
15
To investigate my research question, I use a measure of imputed short term
profitability of trade, PROF. For each actual trade made for firms in my sample, I
compare the actual transaction price with closing price three days later.4 This serves as a
proxy for the ability of the investor initiating the trade to anticipate short term price
movement. This is similar to the method Asthana, Balsam and Sankaragurswamy (2004)
use to establish the potential short term profitability of trades. In their paper, they assess
the potential profitability of buying or selling a specific security, and multiply this return
by the net-buying or net-selling of the small investor group. This method assigns small
investors, as a whole, a certain short-term potential profitability, with the intention
Trader
Category
Trade
Type
Execution
Price
Number
of Shares
3-Day
Forward Price
Imputed
Return
Small Buy 11.51 100 11.45 -0.0052
Large Buy 11.53 5,800 11.45 -0.0070
Small Sell 11.53 200 11.45 0.0070
Large Sell 11.50 8,000 11.45 0.0044
PROF for small investors for firm-day: 0.0029
PROF for large investors for firm-day: -0.0004
Figure 1: Simplified example of construction of PROF for a single firm-day. Each actual trade execution price is
compared to the closing price 3 days forward to construct an imputed return for that trade. Then, the trades are value-
weighted within the trader category to arrive at a PROF for each category for the day.
of comparing their investment performance across disclosure regimes. In this paper, I
construct a relative measure of profitability between the large and small groups by
comparing each signed order (inferred using the Lee and Ready (1991) algorithm5) to the
closing price three trading days later. I then value-weight the trades within the group to
4 I use the midpoint of closing bid and ask quotes as closing price, to avoid the “bid-ask bounce” problem.
5 I use a one second rule to alleviate the concerns raised by Henker and Wang (2006).
16
construct a return, and compare the returns that could potentially be earned by large
versus small traders, and how that relationship changes in the wake of earnings
announcements. This process is illustrated in a simplified example in Figure 1.
My measure of trade profitability has several advantages. Like the measure in
Asthana, Balsam and Sankaragurswamy (2004), it gives a sense of the ability of the
different groups of investors to assess the impact of earnings news in the short-term. But
it also adds an additional layer of precision by examining each trade individually. 6
By
comparing the short-term price benchmark to actual trading prices, PROF also measures
the ability of investors to buy and sell at advantageous prices.
In order to partition trades into informed and uninformed, I use a trade size cutoff.
Although many different cutoff points (and methods) have been used in the literature, I
follow Bhattacharya (2001) in using a $5,000 cutoff for small trades and a $50,000 cutoff
for large trades. I use these cutoffs because the rather conservative dollar values provide
the best assurance that trades in each category truly represent trades from informed or
uninformed investors. All trades between $5,000 and $50,000 are eliminated as
ambiguous. If a stock’s price, however, is above $50, then single lot trades are still
considered to be small trades, since trades of under $5,000 would not be possible without
trading odd lots. Stocks priced above $100, however, are thrown out of my sample, as it
no longer becomes possible to transact in those stocks at prices that unambiguously
represent uninformed investors. I also perform a sensitivity test, described in section 4.4,
to address concerns about changes in small and large trade patterns over time.
6 This allows the measure to capture not only the effect of intraday variation in prices, which can be large
around earnings announcements, but also some of the transactions costs incurred by traders (via the bid-ask
spread and price concessions.)
17
My main question of interest is in the effect of earnings announcements on the
relative ability of small and large investors to trade advantageously. Prior research has
shown that abnormal returns occur around earnings announcements; this effect has been
attributed to both behavioral bias (DeBondt and Thaler, 1985) and risk shifting (Ball and
Kothari, 1991). Due to the known patterns of abnormal returns around earnings
announcements, my tests do not focus on the effect of the earnings announcement on
each investor class, but rather on the relative impact of earnings announcements on the
two groups. In other words, there are many reasons why the short term returns following
an earnings announcement might vary, but the differential impact on small versus large
traders should best capture the effect of the announcement on the change in informedness
between the two groups. To investigate the effects of the difference between the two
investor groups, I use a stacked regression with one observation for each imputed three
day return for each investor type. I estimate the following equation for my main test:
(1)
Where:
PROFij = The imputed 3 day return on firm-day i of trades initiated by trader category j,
expressed in percentage terms.
EAdayi = An indicator variable equal to 1 if firm-day i is the day of or the day following
an earnings announcement, and 0 otherwise.
LGj = An indicator variable equal to 1 if trader category j is large, 0 otherwise.
This specification allows me to directly compare the differential effect of earnings
announcements on the trading of large versus small investors. A positive estimation of α3
would indicate that large traders improve their trading performance during the EAday
period relative to small traders. This result would be consistent with earnings
announcements, on average, causing an increase in information asymmetry between
18
investors. If α3 is negative, on the other hand, this would be indicative of an improvement
in the performance of small traders relative to large, consistent with earnings
announcements serving primarily to reduce information asymmetry between investors.
I include a number of controls for other features of the trading environment which
may cause differences in returns around earnings announcements. RET3i measures the
decile ranked magnitude of the three day return after the observation day (firm day i to
firm day i+3). Since absolute returns following earnings announcements are larger, on
average, than returns on non-announcement days, this control picks up the portion of
variation in PROF that is simply a result of the magnitude of the price movement.
Without the inclusion of this control, any results could be attributable to changes in return
magnitudes, rather than in the underlying abilities of different investor classes.
NB3ij is the net-buying activity for investor class j in the three days prior to the
observation (firm day i-3 to firm day i). This controls for the possibility that trading
outcomes may be partially the result of short term reversals of prior trading positions.
Kaniel et al. (2009) examine the behavior of individual investors using a proprietary data
set that identifies trade initiators and conclude that some individuals tend to unwind their
positions following the announcement. This control will help to alleviate concerns that
patterns in the data may be the result of prior trading rather than earnings announcement
information.
Other controls include Q4i, a dummy which equals 1 if the earnings
announcement is a fourth quarter and zero otherwise, and BA%i, a measure of the daily
bid-ask spread, scaled by closing price. I also include a number of controls designed to
describe the mix of traders, which could potentially affect the performance of the two
19
different groups. LPROPi and SPROPi describe the proportion of total trades on firm-day
i made by large and small investors, respectively. LVOLPROPi and SVOLPROPi
measure the proportion of trades by the two groups, and LAVGi and SAVGi is the
average number of shares traded by each group on firm-day i. These variables control for
changes in investor mix and trading activity around earnings announcements.7
One potentially confounding effect is post earnings announcement drift. If
different investor types are differentially using PEAD-related trading strategies, their
returns following earnings announcements may not reflect the degree to which they are
using the announcement to generate private information. This presents a difficulty
because the research question of this paper centers on the level of private information
about a firm that different investor groups possess. If some traders are better able to
understand and exploit PEAD, then results may pick up this difference rather than
differences due to information based trading. As there is some evidence that institutional
investors exploit PEAD (Ke and Ramalingegowda 2005) I control for the magnitude of
the earnings surprise. But since earnings surprise is also information that informed
investors can use to make investment decisions in conjunction with their private
information, I estimate model (1) both with and without the surprise controls.
I use four different controls for earnings surprise. Given that small and large
traders have been shown to respond differently to surprises calculated using the random
walk or analyst forecast method (Bhattacharya, 2001), I include both as controls. I also
partition the variables into negative and positive surprises, as the effect of PEAD can be
different for negative versus positive surprises. SURP_RW_POSi represents the positive
random walk earnings surprise of the earnings announcement closest to firm-day i, with
7 Results are similar whether these variables are included or not.
20
SURP_RW_NEG representing negative random walk earnings surprises.
SURP_AN_POS and SURP_AN_NEG represent the corresponding variables for earnings
surprises calculated using analyst forecasts.
I also examine the effect of two firm-level characteristics on the relationship
explored in formula (1), firm size and analyst following. I do this by interacting these
variables with my main variables of interest:
(2)
Where PI is one of three variables designed to proxy for the amount of private
information held by investors. MVE_RANK is the decile rank of firm size, measured as
market value of equity. NUMAN is the number of analysts following the firm prior to the
earnings announcement. And C_CONS is the change in analyst consensus, as defined in
Barron et. al. (1998). Appendix A provides detailed variable definitions.
Finally, I test the discretionary liquidity trader hypothesis of Chae (2005). While
the results of model (1) have some implications for this hypothesis, (1) by construction
compares the post-earnings announcement period to all other periods. It is possible,
however, that the period immediately prior to the announcement may also be
fundamentally different from other periods. If large investors had a greater advantage in
the pre-announcement period than in the post-announcement period, it would still be
advantageous for uninformed investors with a short discretionary window to delay their
trade until after an earnings announcement. I test the pre announcement period with the
following model:
(3)
21
This is identical to model (1), except for the substitution of EApre for EAday. EAprei is
an indicator variable which is equal to 1 if firm-day i is one of the two days prior to the
earnings announcement (days -1 or -2). A positive coefficient estimate for α3 would
indicate that the pre announcement period is more advantageous to large traders than
other periods. A negative coefficient would indicate that this period is more advantageous
for small traders.
Next, I test my sample to examine the effect of certain earnings features on the
ability of better informed traders to achieve higher returns following an earnings
announcement. The first attribute of earnings that I test is the absolute magnitude of the
surprise. Yohn (1998) finds that this is positively associated with spread following
earnings announcements, suggesting that large surprises are more likely to result in
information asymmetries between investors. This notion also has intuitive appeal, as
more sophisticated investors are more likely to be in a position to correctly interpret the
meaning of an earnings surprise. For this test, and other tests of earnings attribute effects,
I restrict my sample to days of, or days following (if the day of the announcement is not
available) the earnings announcement. This is due to the fact that these tests are designed
to test the difference between different types of announcements. As a result, I test the
following model:
….(4)
where ABS_EARN is the measure of absolute earnings surprise, based on either analysts’
expectations or a seasonal random walk. In addition to this continuous measure, I also
22
test to see if there are differences between announcements where earnings surprises are
extreme, versus announcements with zero surprise, or very little surprise:
(5)
where EXTREME = 1 if the absolute earnings surprise is in the 5th
quintile among my
sample, and EXTREME = 0 if the absolute earnings surprise is in the 1st quintile among
my sample. All other observations are omitted to afford a direct comparison between the
highest and lowest earnings surprise quintiles.
Next, I examine whether the profitability of trade for larger investors is greater
following positive versus negative earnings surprises. Brown, Hillegeist, and Lo (2009)
provide some evidence that positive and negative earnings surprises may affect the arrival
rate of private information differently, thus having a differing effect on the profitability of
different types of traders. Furthermore, prior studies have shown that investors react
differently to negative versus positive surprises (Hayn, 1995; Matsumoto 2002) I
therefore test the following model:
(6)
Where POS = 1 if the earnings surprise is positive and POS = 0 if it is negative. Due to
the fact that I use two different types of earnings surprise in this paper, and they are not
always in concordance (i.e. one will be negative while the other is positive, on occasion),
I only use observations for (6) where both estimations of earnings surprise are in
agreement. All other observations, including those with zero earnings surprise, are
omitted from the estimation of (6).
Next, I examine the case where there is disagreement between the two measures
of earnings surprise. This is potentially interesting because Battalio and Mendenhall
23
(2005) and Bhattacharya (2001) both provide evidence that small traders are more likely
to anchor off seasonal random walk expectations of earnings, while large traders appear
to react more to analysts’ expectations of earnings. Thus, when one construct of earnings
surprise is positive and the other negative, the reactions of the two groups may be most
different for these announcements, and this may be a time at which small traders’
information (i.e. prior year’s quarterly earnings) is most out of date. I examine this
question by testing the following model:
(7)
Where DISAGREE = 1 if the two measures of earnings surprise have opposite signs, and
DISAGREE = 0 if the two measures have the same sign. Other observations are omitted.
Next, I examine the effect of shareholder attributes on large traders’ advantage.
Since large trades are likely to be institutional, it is plausible that the types of institutions
which trade in a firm are likely to affect the relative profitability of large trades. Ali et. al.
(2004) find that patterns of changes in institutional holdings have some predictive power
for returns around earnings announcements, and Matsumoto (2002) finds that greater
ownership by transient institutional investors increases management’s incentives to meet
earnings forecasts. I use the classification scheme developed in Bushee (2001) to classify
13f holdings of firms in my sample as held by transient institutions, quasi-indexing
institutions, and dedicated institutions.8 For each observation in my sample, I then use the
proportional holdings of each type of institution for the quarter prior to the nearest
earnings announcement to estimate the following:
8 Institutions are classified based on portfolio turnover and diversity. Transient investors, for example,
display a large amount of turnover and do not tend to own similar clusters of stocks.
24
(8)
where INV = the proportion of outstanding shares owned by a type of institution in the
quarter prior to the nearest earnings announcement. I expect that the presence of both
dedicated and transient ownership will be positively associated with large traders’
profitability following earnings announcements, as these are the types of institutions
which are more likely to seek an advantage from firms’ accounting statements. Quasi-
indexers, on the other hand, are more likely to employ informationless trading strategies,
and I form no prediction regarding their presence on my main effect of interest.
4. Data and Results
4.1 Data Procedures
My sample includes all ordinary equity securities for domestic, non-ADR firms
(CRSP shrcd 10 or 11) for the years 1994 to 2007 inclusive. I further restrict my sample
to only firms listed on the NYSE to avoid difficulties in comparing microstructure
techniques such as the Lee-Ready algorithm across different exchanges.9 For each
security in my sample, I use the TAQ database to infer the direction of trade initiation. I
eliminate all trades that are not routed through the NYSE (i.e. regional exchanges) as
variations in the timing of quotes from these exchanges can make comparison with
NYSE quotes and trades problematic. Trades with a negative price or a price more than
$5 different from the previous price are eliminated.10
I further eliminate quotes where the
bid is equal to or greater than the ask price, where the spread is greater than $5, where
depth is negative, or where changes in the bid or ask are more than $5 from the previous
9 Additionally, as noted by Atkins and Dyl (1997) and Anderson and Dyl (2005), double counting of trades
makes work with NASDAQ at the transaction level difficult. And Ellis, Michaely, and O’Hara (2000) point
out issues related to using the Lee-Ready algorithm with NASDAQ. 10
These observations are likely errors in the database.
25
bid or ask. I then match the trades with the quotes, using the Lee-Ready (1991) algorithm
(with the one second delay suggested by Henker and Wang (2006)) and infer the
direction of the trade initiator by comparing executed price with the prevailing quote.
I then compute the short term imputed profit for each trade in my sample by
comparing the actual execution price to the midpoint of the closing bid and ask quotes
three days hence. Trade direction is used to determine the sign of the imputed return.
Investors are also classified as small, medium, or large at this point based on the dollar
value of the trade. All medium-sized trades are discarded. For the small and large traders,
PROFij is computed by taking the value-weighted average of imputed three-day return for
that day within that investor class.
The resulting dataset is then merged with Compustat to obtain earnings
announcement dates and some firm variables. Firms without a minimum of five earnings
announcement dates are dropped from the sample. To prevent serial correlation due to
overlapping return periods, I drop two thirds of my observations. To maximize the
number of earnings announcement observations, I anchor on earnings announcement
days, using the following procedure. For each observation with two earnings
announcement days (i.e. day 0 and day +1), I randomly choose one day to begin the
process of culling overlapping return periods.11
After this, I begin dropping observations
around the anchor date. For example, if day +1 relative to the earnings announcement is
my anchor date, I would delete days +2 and + 3, as well as days 0 and -1. I then continue
through the days that have the relevant earnings announcement date as the nearest
earnings announcement date, keeping days +4 and -2 in this example, and deleting days
11
In the case where only one of these two days is available, I use that day as the anchor.
26
+5 and +6, -3 and -4, and so forth. This ensures that there are no overlaps between the
many different calculated return periods in my sample.
4.2 Descriptive Statistics
A total of 2,883 unique firms representing 1,594,692 non-overlapping trading
days are in the sample. 156,239 sample days are in the post-earnings announcement
period, i.e. they are days 0 or 1 relative to the earnings announcement. Almost all of these
days have some small trader activity, and 87% of days have some larger trade activity as
well. Both mean and median PROF is negative for both large and small traders, although
the mean is somewhat lower for small traders. Negative PROF is intuitive, as most of the
three-day windows involve little price movement, but the entry or exit price of the
security includes the cost of the spread.
Table 2 shows a correlation matrix consisting of the main variables used in the
tests. PROFij is positively correlated with EAdayi, NB3i, MV_RANKi and NUMANALi.
As expected, PROFij is significantly negatively correlated with BA%, indicating that
higher spreads reduce short term returns.
4.3 Main Results
Results of estimating (1) are shown in Table 3. The coefficients for both LGj and
EAdayi*LGj are positive and significant at conventional levels. This indicates that large
traders enjoy an advantage in their short term returns compared to small traders in both
the period following earnings announcements and at other times as well. The 0.022
coefficient for EAdayi*LGj indicates that earnings announcements cause large traders’
advantage over small traders to more than double from the 0.014 percent advantage
implied by the coefficient for LGj. An increase of 0.022% over a three day window is an
27
TABLE 1
Descriptive Statistics for a Sample of NYSE firm trading days for calendar years
1994 - 2007
PROF is defined as the value weighted imputed 3-day return for each investor class for the firm-day.
LPROP and SPROP are the proportion of the total number of large trades and small trades, respectively, to
the total number of trades in the day. LVOLPROP and SVOLPROP are the proportion of the total volume
composed of large trades and small trades in a day, respectively. LAVG and SAVG are the average number
of shares per large trade and per small trade, respectively. SURP is the absolute value of the earnings
surprise, defined as the difference between the analyst consensus (defined as the average of forecasts
initiated or reiterated in the 45 days prior to the announcement) and IBES actuals. If this number is
unavailable, I use the difference between reported earnings and the seasonal random walk. EAday is an
indicator variable equal to 1 if the day is an earnings announcement day (or the day following an
announcement) and zero otherwise. LG is an indicator variable equal to 1 if the investor group is large, and
zero otherwise.
Panel A:
Sample Description
Total Number of Firms in Sample: 2,883
Total Number of Earnings Announcement Firm-Days 156,239
Total Number of Firm-Days 1,594,692
Firm-Days with Calculable Large Trader Profit 1,388,730
Firm-Days with Calculable Small Trader Profit 1,578,662
Panel B:
Variable Descriptives
Variable Name N Mean Median Std 25th
75th
PROF (LG=1) 1 1,388,730 -0.0365 -0.0565 1.652 -0.5734 0.4630
PROF (LG=0) 1,578,662 -0.0818 -0.0517 1.388 -0.4564 0.2865
RET3 1,577,698 0.0336 0.0203 1.024 0.0087 0.0400
NB3 1,570,341 0.0611 0.0857 0.276 -0.0667 0.2174
SURP_RW_POS 1,546,347 0.0098 0.0012 0.033 0 0.0057
SURP_RW_NEG 1,547,578 -0.0093 0 0.036 -0.0051 0
SURP_AN_POS 1,225,012 0.0013 0.0002 0.003 0 0.0012
SURP_AN_NEG 1,225,021 -0.0017 0 0.009 -0.0002 0
LPROP 1,355,122 0.2494 0.1786 0.229 0.0426 0.4219
SPROP 1,355,122 0.3632 0.3262 0.202 0.1964 0.5000
LVOLPROP 1,355,122 0.6054 0.6636 0.304 0.3377 0.8948
SVOLPROP 1,355,122 0.0927 0.0456 0.112 0.0146 0.1304
LAVG 1,388,730 5,391 3,566 9,452 2,201 5,942
SAVG 1,372,700 145 100 113.5 100 150
28
TABLE 2
Correlation Matrix of Selected Variables
Variables are defined in Appendix A. Pearson correlations are on the top right, Spearman correlation on the bottom left. P-values are displayed below
correlations.
PROF EAday LG RET3 NB3 BA% SURP_AN
_POS
SURP_AN
_NEG
MV_RANK NUMAN C_CONS
PROF - 0.0102
0.000
0.0149
0.000
0.0012
0.038
0.0228
0.000
-0.0619
0.000
-0.0099
0.000
0.0266
0.000
0.0456
0.000
0.0092
0.000
0.0021
0.013
EADay 0.0104
0.000 -
0.0048
0.000
0.0019
0.001
0.0011
0.071
0.0081
0.000
0.0101
0.000
-0.0053
0.000
-0.0087
0.000
-0.0029
0.001
0.0000
0.972
LG -0.0078
0.000
0.0023
0.009 -
-0.0023
0.000
0.0313
0.000
-0.0468
0.000
-0.0238
0.000
0.0565
0.000
0.0899
0.000
0.0116
0.000
0.0002
0.826
RET3 -0.0012
0.1923
0.0484
0.000
-0.0023
0.009 -
-0.0048
0.000
0.0257
0.000
0.0035
0.000
-0.0155
0.000
-0.0127
0.000
0.189
0.000
-0.0153
0.000
NB3 0.0161
0.000
0.0018
0.038
0.0638
0.000
-0.0174
0.000 -
-0.1818
0.000
-0.0585
0.000
0.0883
0.000
0.2157
0.000
0.0902
0.000
0.0036
0.000
BA% -0.0624
0.000
0.0113
0.000
-0.0039
0.000
0.1678
0.000
-0.1139
0.000 -
0.0999
0.000
-0.2655
0.000
-0.4797
0.000
-0.1593
0.000
-0.0195
0.000
SURP_AN_POS 0.0057
0.000
0.0168
0.000
-0.0063
0.000
0.0163
0.000
-0.0169
0.000
-0.0498
0.000 -
0.0811
0.000
-0.1667
0.000
-0.0742
0.000
0.0007
0.000
SURP_AN_NEG 0.0099
0.000
-0.0000
0.964
0.0035
0.000
0.0025
0.0042
0.0477
0.000
-0.0223
0.000
0.5775
0.000 -
0.2604
0.000
0.0503
0.000
0.0197
0.000
MV_RANK 0.0409
0.000
-0.0057
0.000
0.0207
0.000
-0.1240
0.000
0.1625
0.000
-0.3822
0.000
-0.0602
0.000
0.0760
0.000
- 0.5256
0.000
0.0028
0.001
NUMAN 0.0208
0.000
-0.0053
0.000
0.0126
0.000
0.0169
0.000
0.0873
0.000
-0.1936
0.000
-0.0486
0.000
0.0674
0.000
0.5323
0.000
- 0.028
0.000
C_CONS 0.0034
0.000
0.0004
0.635
0.0003
0.6895
-0.0113
0.000
0.0025
0.004
-0.0204
0.000
0.0119
0.000
0.0359
0.000
0.0036
0.000
0.0333
0.000 -
29
economically significant excess return for such a small period of time. Assuming 252
trading days in a year, excess return of 0.022% in each 3 day period would lead to an
annual excess return of 1.87%. These results are consistent with earnings announcements
on average causing an increase in short-term information asymmetry that benefits better-
informed investors. If we set all control variables to their median values, the difference
between small and large profitability following earnings announcements is 0.034% over
three days, equivalent to a 2.90% annual return.
The second column of Table 3 shows results with the inclusion of controls for the
magnitude and direction of the earnings surprise (each variable is included individually,
as well as interacted with EAdayi and LGj individually, although only the three way
interaction term is shown for brevity.) Interestingly, while none of the surprise interaction
terms are statistically significant, their inclusion increases the magnitude of the
coefficient for EAdayi*LGj and decreases its standard error. Thus, to the extent that
investors do trade strictly on PEAD following an earnings announcement, such trading
the short term returns of investors, as expected. Higher levels of price movement are
associated with higher individual trading returns, and this relationship becomes stronger
on earnings announcement days. Prior net-buying seems to also affect individual short
term returns, indicating that unwinding of prior positions does seem to affect the
profitability of trade, although it is interesting to note that this relationship does not
change in the period following earnings announcements. As expected, bid ask spreads
have a negative impact on short-term returns.
Results for the incremental effect of information variables are shown in Table 4.
The first column of panel A shows the results of inclusion of a firm size interaction with
30
TABLE 3
Results of Regressions of Short Term Trader Profitability on Trader Characteristics
and Information Variables
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day (only one out of each three potential trading days is used, to avoid overlapping
return periods.) Some interactions are omitted for brevity (all earnings surprise variables are also interacted
with EAday and LG individually). Firm fixed effects are included. All standard errors are clustered by date
using the Rogers (1993) method. Coefficients in italics are statistically significant at the 5% level. All tests
are two-tailed
Variable Name Estimate T-Stat Estimate T-Stat
Intercept 0.00941 0.96 0.01605 1.57
EAday -0.01217 -1.65 -0.02208 -2.78
LG 0.01436 4.94 0.00396 3.22
EAday*LG 0.02184 2.38 0.03425 3.22
RET3 0.00744 7.53 0.00677 6.61
RET3*LG 0.00098 1.08 0.00084 0.91
RET3*EAday 0.00990 4.82 0.01068 5.00
RET3*EAday*LG 0.00038 0.16 -0.00144 -0.58
NB3 0.02302 3.82 0.02071 3.22
NB3*EAday 0.00501 0.20 0.02131 0.82
NB3*EAday*LG -0.04209 -1.44 -0.04430 -1.43
Q4*EAday -0.00724 -0.74 -0.00893 -0.87
BA% -2.22494 -13.05 -2.15328 -12.02
SURP_RW_POS*EAday*LG -0.67895 -1.89
SURP_RW_NEG*EAday*LG 0.30154 0.96
SURP_AN_POS*EAday*LG 1.32583 0.48
SURP_AN_NEG*EAday*LG -0.85744 -0.63
LPROP 0.06914 3.22 0.07499 3.35
SPROP -0.04726 -2.45 -0.02859 -1.42
LVOLPROP -0.07251 -5.19 -0.07436 -5.11
SVOLPROP -0.03219 -0.87 -0.04992 -1.32
LAVG 0.00000 -0.05 0.00000 -0.06
SAVG -0.00022 -6.51 -0.00024 -6.35
Number of observations: 2,558,530 2,277,370
31
the main variables of interest. The coefficient for MV_RANK is positive and significant,
indicating that trading profits are higher overall for large firms, although the negative
does not appear to substantially explain the relative improvement in trading performance
that large traders experience following an earnings announcement.
Regarding the control variables, the magnitude of the price change seems to affect
coefficient for MV_RANK*LG indicates that increases in size tend to benefit smaller
traders. The coefficient for the three way interaction term MV_RANK*LG*EAday is not
statistically significant, nor is the coefficient for MV_RANK*EAday, indicating, contrary
to expectations, that firm size does not seem to affect the difference between small and
large traders’ profits following earnings announcements.
Column 2 of Panel A shows the results of interactions with the number of analysts
following a firm prior to the earnings announcement. Here results show that more analyst
following does affect trading around earnings announcements, with large traders
benefiting from an increase in analyst coverage, relative to small traders. Interestingly,
however, the opposite effect obtains in non-earnings announcement periods, although the
magnitude of the effect is smaller. This indicates that, while heavily followed stocks are,
ceteris paribus, a worse trading environment for large traders, these stocks are
nonetheless more advantageous for large traders following earnings announcements.
Finally, in Panel B, we see the results of interactions with C_CONS, or change in
analyst consensus. If analyst consensus decreases following an earnings announcement,
this implies that the ratio of private information to common information has increased,
which would tend to benefit sophisticated investors who can generate private
information. Results are consistent with this notion, as the coefficient on
32
TABLE 4
Effect of Firm-Level Variables on the Relative Advantage of Large Traders
Following Earnings Announcements
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day (only one out of each three potential trading days is used, to avoid overlapping
return periods.) Some interactions are omitted for brevity (all earnings surprise variables are also interacted
with EAday and LG individually). All standard errors are clustered by date using the Rogers (1993)
method. Coefficients in italics are statistically significant at the 5% level. All tests are two-tailed.
Panel A:
Incremental effect of firm size and analyst following
PI = MV_RANK PI = NUMAN
Variable Name Estimate T-Stat Estimate T-Stat
Intercept 0.1477 1.42 0.0248 1.67
EAday -0.0203 -2.45 -0.0212 -1.41
LG 0.0057 1.78 0.0170 3.56
EAday*LG 0.0334 3.10 0.0097 0.57
PI 0.0000 1.05 -0.0002 -0.33
PI*LG -0.0000 -5.00 -0.0011 -2.81
PI*EAday -0.0000 -1.24 -0.0009 -0.76
PI*EADay*LG 0.0000 0.67 0.0027 2.09
RET3 0.0068 6.64 0.0058 4.45
RET3*LG 0.0008 0.86 -0.0004 -0.40
RET3*EAday 0.0106 4.85 0.0123 4.53
RET3*EAday*LG -0.0014 -0.57 -0.0015 -0.51
NB3 0.0207 3.23 0.0152 1.64
NB3*EAday 0.0217 0.84 0.0661 1.56
NB3*EAday*LG -0.0439 -1.42 -0.0501 -1.00
Q4*EAday -0.0092 -0.89 -0.0054 -0.43
BA% -2.1529 -12.01 -1.4961 -6.00
SURP_RW_POS*EAday*LG -0.6768 -1.88 -0.3407 -0.64
SURP_RW_NEG*EAday*LG 0.2999 0.96 0.0091 0.02
SURP_AN_POS*EAday*LG 1.3578 0.49 -1.7486 -0.42
SURP_AN_NEG*EAday*LG -0.8674 -0.63 -0.7284 -0.30
LPROP 0.0753 3.28 0.0789 2.76
SPROP -0.0284 -1.40 0.0057 0.20
LVOLPROP -0.0744 -5.08 -0.0857 -4.43
SVOLPROP -0.0499 -1.32 -0.0821 -1.55
LAVG -0.0000 -0.05 0.0000 1.19
SAVG -0.0002 -6.35 -0.0003 -4.63
Number of observations: 2,277,746 1,300,341
33
Panel B:
Incremental effect of change in analyst consensus
PI = C_CONS
Variable Name Estimate T-Stat
Intercept 0.0202 1.45
EAday -0.0485 -3.67
LG 0.0087 1.80
EAday*LG 0.0584 3.37
PI 0.0003 0.83
PI*LG -0.0003 0.83
PI*EAday 0.0042 2.45
PI*EADay*LG -0.0059 -2.48
RET3 0.0057 4.45
RET3*LG -0.0005 -0.47
RET3*EAday 0.0124 4.55
RET3*EAday*LG -0.0015 -0.52
NB3 0.0147 1.59
NB3*EAday 0.0659 1.54
NB3*EAday*LG -0.0475 -0.95
Q4*EAday -0.0053 -0.43
BA% -1.4881 -5.97
SURP_RW_POS*EAday*LG -0.3585 -0.67
SURP_RW_NEG*EAday*LG 0.0129 0.03
SURP_AN_POS*EAday*LG -2.1094 -0.51
SURP_AN_NEG*EAday*LG -0.5253 -0.22
LPROP 0.0816 2.88
SPROP 0.0073 0.25
LVOLPROP -0.0859 -4.45
SVOLPROP -0.0825 -1.56
LAVG 0.0000 -4.63
SAVG -0.0071 -0.07
Number of observations: 1,300,461
34
C_CONS*EAday*LG is negative. This indicates that if an earnings announcement
produces information that increases the amount of private, idiosyncratic information in
analysts’ forecasts about future earnings (a decrease in consensus), large investors’
advantage over small investors following the announcement is greater. This is consistent
with the proposition of this paper that the relative profitability of trade between large and
small traders following earnings announcements is driven by differential levels of private
information.
Table 5 details the effect of earnings surprises on the profitability of large traders.
As the sample from which this is drawn includes only post earnings announcement days,
the coefficient of interest is the interaction between the surprise term and the LG
indicator. When absolute surprise is measured using analysts’ expectations, large trader
profitability is positive, but only significant at the 10% level. This is weak evidence that
large traders’ advantages increase when earnings surprise, as measured by analysts’
expectations, is high. When earnings is measured as a seasonal random walk, the effect
on large trade profitability not statistically significant. Note that, consistent with prior
results, LG is positive and significant, indicating that large traders are more profitable
than small traders on post earnings announcement days. I also test both types of earnings
surprise simultaneously to see if the effect persists in the presence of both types of
earnings surprise. Interestingly, the results stay the same when both types of surprise are
present, consistent with the notion in Battalio and Mendenhall (2005) extend
Bhattacharya (2001) that the different investor types are focused on different types of
earnings surprise.
35
TABLE 5
Effect of Earnings Surprises on Relative Advantage of Large Traders
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day; in these tests only earnings announcement days (day 0 or day +1 relative to the
earnings announcement) are used. Firm fixed effects are included. All standard errors are clustered by date
using the Rogers (1993) method. Coefficients in italics are statistically significant at the 5% level. All tests
are two-tailed.
Using Analysts’
Expectations Surprise Using Random Walk
Surprise
Variable Name Estimate T-Stat Estimate T-Stat
Intercept -0.0971 -11.38 -0.0973 -11.38
LG 0.0349 3.38 0.0363 3.48
ABSOLUTE_ANAL -0.0041 -0.56
ABSOLUTE_ANAL*LG 0.0156 1.65
ABSOLUTE_RAND 0.0345 0.61
ABSOLUTE_RAND*LG -.14532 -1.45
RET3 0.0163 6.80 0.0162 6.74
RET3*LG -0.0006 -0.22 -0.0002 -0.07
NB3 0.0741 2.65 0.0757 2.71
NB3*LG -0.0533 -1.59 -0.0562 -1.67
Q4*LG 0.0061 0.33 0.0066 0.36
Number of observations: 217,632 217,632
Both Types of Surprise
Variable Name Estimate T-Stat
Intercept -0.097 -11.37
LG 0.036 3.51
ABSOLUTE_ANAL -0.004 -0.58
ABSOLUTE_ANAL*LG 0.016 1.70
ABSOLUTE_RAND 0.036 0.64
ABSOLUTE_RAND*LG -0.153 -1.53
RET3 0.016 6.74
RET3*LG -0.000 -0.09
NB3 0.075 2.70
NB3*LG -0.056 -1.67
Q4*LG 0.006 0.34
Number of observations: 217,632
36
In Table 6, I turn to the effect of extreme earnings surprises. This sample includes
only observations that are in the highest and lowest absolute earnings surprise quintiles in
my sample. While there is weak evidence (at the 10% level) that the profitability of trade
initiators in general increases when analysts’ expectation earnings surprises are extreme,
there is no evidence that this gain is captured mostly by large traders (the coefficient for
EXTREME*LG is not significant). This suggests that the effect documented in Table 5 is
not confined to the most extreme earnings surprises. Again, when surprise is defined
using the seasonal random walk, no effect is discernable.
The difference between positive and negative earnings surprises is documented in Table
7. Interestingly, trade initiator profitability in general is significantly higher when
earnings surprises are positive, but again there is no incremental effect for large traders
specifically. One difficulty of this specification used in this paper is that earnings
surprises will effect short term security returns. Therefore, it is likely that the inclusion of
the RET3*LG control variable in this estimation, and in all the earnings characteristics
estimations in Tables 5-8, may take away from the main effect of interest. But given that
large security returns are likely to be mechanically associated with higher profitability for
large traders, I continue to include RET3 as a control variable.
My final test of the effect of earnings characteristics on short term trade
profitability is in Table 8. The term DISAGREE measures instances in which the
analysts’ forecasts measure of earnings surprise is a different sign than the random walk
measure of earnings surprise. The coefficient for DISAGREE is negative and significant,
indicating that overall short term returns for trade initiators are lower when these surprise
measures yield different conclusions. Again, however, the interaction term
37
TABLE 6
Effect of Extreme Earnings Surprises on Relative Advantage of Large Traders
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day; in these tests only earnings announcement days (day 0 or day +1 relative to the
earnings announcement) are used. Firm fixed effects are included. Sample consists of observations that are
in either the first or fifth quintile of absolute earnings surprises. Observations in the fifth (highest) earnings
surprise quintile are coded EXTREME=1, whereas those in the first quintile are coded EXTREME=0. All
standard errors are clustered by date using the Rogers (1993) method. Coefficients in italics are statistically
significant at the 5% level. All tests are two-tailed.
Analysts’ Expectation
Surprise Random Walk Surprise
Variable Name Estimate T-Stat Estimate T-Stat
Intercept -0.1252 -7.83 -0.0956 -6.44
LG 0.0372 1.77 0.0562 2.99
EXTREME 0.0407 1.93 -0.0188 -1.03
EXTREME*LG 0.0192 0.81 0.0094 0.43
RET3 0.0140 4.50 0.0139 4.45
RET3*LG -0.0034 -0.82 -0.0059 -1.45
NB3 0.0563 1.37 0.0692 1.60
NB3*LG -0.0653 -1.24 -0.0479 -0.90
Q4*LG 0.0196 0.61 0.0089 0.29
Number of observations: 87,055 87,064
38
TABLE 7
Effect of Positive versus Negative Earnings Surprises on Relative Advantage of
Large Traders
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day; in these tests only earnings announcement days (day 0 or day +1 relative to the
earnings announcement) are used. Firm fixed effects are included. Sample consists of observations in
which there is concordance between analysts’ expectations earnings surprise and random walk earnings
surprise. If both measures of surprise are positive, then POS=1. If both are negative then POS=0. All other
observations are discarded. All standard errors are clustered by date using the Rogers (1993) method.
Coefficients in italics are statistically significant at the 5% level. All tests are two-tailed.
Variable Name Estimate T-Stat
Intercept -0.1639 -13.64
LG 0.0359 2.21
POS 0.0871 6.81
POS*LG -0.0101 -0.57
RET3 0.0204 7.72
RET3*LG 0.0012 0.37
NB3 0.0712 2.36
NB3*LG -0.0718 -1.84
Q4*LG 0.0031 0.14
Number of observations: 144,773
39
DISAGREE*LG yields no firm results. So while disagreement between measures of
earnings surprise seems to affect the trading environment following earnings
announcements, I cannot conclude that it affects the relative advantage of the better
informed over the less informed. Overall, the evidence on the effect of earnings attributes
on large trader advantage indicates there is some advantage to large traders when
absolute analysts’ forecast earnings surprise is high, but other aspects of earnings
examined here do not seem to affect this advantage. This may be due to the inability of
this research design to separate the effect of earnings announcements while
simultaneously controlling for the price effect of that announcement.
I examine the effect of the presence of different types of institutional traders in
Table 9. Contrary to my expectations, transient institutional ownership is positively
associated with profitability in general, but negatively associated with large trader
profitability on all days. This indicates that the profitability of large trades tends to be
lower when transient institutional ownership is high. It’s unknown whether this is due to
the trading activities of the transient institutions themselves or some other associated
attribute of their ownership. However, transient ownership seems to have no particular
effect on earnings announcement days as opposed to other days. So the effect of
ownership appears to be a general one, and not necessarily related to interpretation of
financial statements following earnings disclosures.
The second column of Table 9 shows the results for dedicated institutional
ownership. Contrary to the results for transient ownership, greater dedicated ownership is
associated with better short term trade profitability for large traders. This is in line with
expectations, as dedicated investors have invested heavily in understanding a particular
40
TABLE 8
Effect of Differing Earnings Surprise Signals on the Relative Advantage of Large
Traders
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day; in these tests only earnings announcement days (day 0 or day +1 relative to the
earnings announcement) are used. Firm fixed effects are included. Sample consists of observations in
which there are differences in earnings surprise signals between analysts’ expectations earnings surprise
and random walk earnings surprise. If one measure of surprise is positive and the other is negative, then
DISAGREE=1. If both are either positive or negative, then DISAGREE=0. All other observations are
discarded. All standard errors are clustered by date using the Rogers (1993) method. Coefficients in italics
are statistically significant at the 5% level. All tests are two-tailed.
Variable Name Estimate T-Stat
Intercept -0.0847 -8.65
LG 0.0296 2.32
DISAGREE -0.0325 -3.34
DISAGREE*LG -0.0010 -0.07
RET3 0.0175 6.95
RET3*LG 0.0005 0.18
NB3 0.0909 3.08
NB3*LG -0.0493 -1.35
Q4*LG 0.0105 0.53
Number of observations: 181,833
41
TABLE 9
Effect of Investor Mix on the Relative Advantage of Large Traders
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day (only one out of each three potential trading days is used, to avoid overlapping
return periods.) INV represents the percentage of shares outstanding by three different types of 13f
reporting institutions in the quarter prior to the nearest earnings announcement. All standard errors are
clustered by firm using the Rogers (1993) method. Coefficients in italics are statistically significant at the
5% level. All tests are two-tailed.
INV = Transient Investor
Percentage INV = Dedicated
Investor Percentage
Variable Name Estimate T-Stat Estimate T-Stat
Intercept .0161 1.95 .0456 5.59
EAday -0.163 -1.69 -0.014 -1.48
LG 0.031 7.81 0.002 0.42
EAday*LG 0.018 1.42 0.019 1.40
INV 0.151 11.96 -0.062 -4.13
INV*LG -0.125 -6.98 0.117 4.72
INV*EAday 0.003 0.09 -0.013 -0.29
INV*EADay*LG 0.015 0.25 0.014 0.23
RET3 0.007 12.38 0.008 14.40
RET3*LG 0.000 0.50 -0.000 -0.44
RET3*EAday 0.011 5.78 0.011 5.82
RET3*EAday*LG 0.001 0.59 0.002 0.67
NB3 0.082 11.78 0.089 12.70
NB3*EAday -0.021 -0.85 -0.022 -0.88
NB3*LG -0.072 -7.85 -0.079 -8.52
NB3*EAday*LG 0.001 0.05 0.003 0.08
Q4*EAday -0.012 -1.09 -0.012 -1.03
BA% -2.553 -16.32 -2.581 -16.37
SURP_RW_POS*EAday*LG -0.567 -1.41 -0.564 -1.40
SURP_RW_NEG*EAday*LG 0.281 0.76 0.280 0.76
SURP_AN_POS*EAday*LG 2.261 0.70 2.306 0.71
SURP_AN_NEG*EAday*LG -1.124 -0.73 -1.116 -0.73
LPROP 0.109 8.08 0.104 7.68
SPROP -0.063 -3.09 -0.053 -2.61
LVOLPROP -0.097 -7.33 -0.099 -7.48
SVOLPROP -0.008 -0.19 -0.035 -0.87
LAVG 0.000 1.72 0.000 1.92
SAVG -.0003 -9.34 -0.000 -9.73
Number of observations: 1,885,015 1,885,015
42
INV = Quasi-Indexer Investor
Percentage
Variable Name Estimate T-Stat
Intercept -0.001 -0.16
EAday -0.001 -0.11
LG 0.040 7.41
EAday*LG 0.026 1.45
INV 0.092 10.49
INV*LG -0.067 -5.86
INV*EAday -0.040 -1.63
INV*EADay*LG -0.015 -0.40
RET3 0.009 14.85
RET3*LG -0.001 -0.73
RET3*EAday 0.011 5.77
RET3*EAday*LG 0.002 0.69
NB3 0.087 12.40
NB3*EAday -0.020 -0.80
NB3*LG -0.076 -8.28
NB3*EAday*LG 0.003 0.10
Q4*EAday -0.012 -1.05
BA% -2.411 -14.83
SURP_RW_POS*EAday*LG -0.570 -1.42
SURP_RW_NEG*EAday*LG 0.283 0.77
SURP_AN_POS*EAday*LG 2.229 0.69
SURP_AN_NEG*EAday*LG -1.098 -0.72
LPROP 0.101 7.51
SPROP -0.047 -2.33
LVOLPROP -0.092 -6.94
SVOLPROP -0.050 -1.27
LAVG 0.000 1.76
SAVG -0.000 -9.57
Number of observations: 1,885,015
43
firm. Again, however, there is no discernable effect on post earnings announcement days.
The final column in Table 9 shows the effect of quasi-indexer ownership on large trader
advantage. Similarly to transient ownership, large trades tend to be relatively less
profitable as quasi-indexer ownership increases. This may be due to the tendency of these
institutions to make large, informationless trades. As with the other institutional types,
however, this test shows no significant effect of ownership mix on earnings
announcement days, or large trader advantage on those days.
Next, I turn to the discretionary liquidity trader hypothesis of Chae (2005). This
argues that patterns in trading volume around earnings announcements can be explained
by the “pent up demand” of liquidity traders who wish to delay their trade until after the
earnings announcement, as that will reduce their chances of trading against informed
traders and cause them to experience lower average returns. While the evidence presented
in table 3 would seem to be inconsistent with this notion, it is possible that the conditions
for small traders are even worse immediately prior to an earnings announcement than
immediately following the announcement. If this is the case, it still may be rational for
uninformed traders who have exogenous liquidity needs to delay their trade until the
announcement has been made. I test this proposition by running an estimation comparing
small and large trade performance in the period immediately prior to the earnings
announcement. Results of this test are presented in table 10.
If conditions for small traders are poor in this period, we would expect that the
coefficient for EApre*LG would be positive and significant. Instead, the coefficient is
not statistically significant, nor is the coefficient for EApre alone, indicating that there is
no evidence to support the notion that the period prior to the announcement is a worse
44
trading environment for small traders. Combined with the results in table 3, these results
are consistent with the view that uninformed investors would put themselves at a
disadvantage by waiting for the post-announcement period to execute their trades, the
opposite of the discretionary liquidity trader hypothesis.
4.4 Sensitivity Analyses
While the use of small and large trade size to proxy for less and more
sophisticated investors is well established in the literature, it has also been noted that the
proxy may be becoming less reliable over time, as electronic trading innovations make
order splitting by institutions easier and less costly (Kaniel, Saar, Titman, 2008). While
order splitting has been a potential strategy for masking informed trade for many years
(Barclay and Warner, 1993) there is some evidence that its use has increased over time.
Puckett and Yan (2008), using a large database of actual trades from 840 different
institutions show a steady decline in median trade size from 60,030 in 1999 to 14,232 in
2005. And Campbell et. al. (2009) show that very small trades are correlated with
changes in institutional holdings, suggesting that some institutions are splitting their
trades into very small portions. Finally, Hvidkaajer (2008) finds that in his sample,
correlation between small trading direction and institutional holdings is negative until
2002, at which point it turns slightly positive.
I do two additional analyses to address these concerns. First, I partition my
sample into an early and a late period. The early period (1994 – 2001) should have fewer
concerns about proxy noise due to excessive institutional order splitting, compared to the
later period (2002 – 2007). I rerun my tests on both periods in order to examine any
45
TABLE 10
Relative Performance of Small and Large Traders Prior to the Earnings
Announcement
Variables are defined in Appendix A. Stacked regression consists of one observation per trader type (small
or large) per trading day (only one out of each three potential trading days is used, to avoid overlapping
return periods.) Some interactions are omitted for brevity (all earnings surprise variables are also interacted
with EAday and LG individually). All standard errors are clustered by date using the Rogers (1993)
method. Coefficients in italics are statistically significant at the 5% level. All tests are two-tailed.
Variable Name Estimate T-Stat
Intercept 0.0277 2.51
EApre 0.0078 0.71
LG 0.0133 4.60
EApre*LG 0.0117 0.81
RET3 0.2738 3.21
RET3*LG -0.1258 -1.73
RET3*EApre 0.0775 0.31
RET3*EApre*LG -0.0370 -0.12
NB3 0.0199 2.88
NB3*EApre 0.1253 3.33
NB3*EApre*LG -0.0792 -1.76
Q4*EApre -0.0283 -1.92
BA% -2.0311 -10.79
LPROP 0.0798 3.31
SPROP 0.0166 0.76
LVOLPROP -0.0654 -4.25
SVOLPROP -0.0898 -2.30
LAVG -0.0000 -0.14
SAVG -0.0003 -7.10
Number of observations: 2,014,305
46
differences. Abbreviated results of estimating (1) for these two time periods are shown in
Panel A of Table 11. For the early period, results are consistent with the entire sample
results reported in Table 3. The coefficients for LG and LG*EAday are both positive and
significant, and both are larger in magnitude than the combined results. For the latter
period, however, the coefficient on LG becomes negative and significant, with the
coefficient for EAday*LG remaining positive but only at a marginal level of significance.
These results are consistent with a change in trading patterns of informed and uninformed
investors over time, although I cannot rule out the possibility of other influencing trends
over time.
To better understand the changing nature of trade over time, I present trends in
order size for my sample in Panel B. In examining the proportion of large and small
trades both by number and by trading volume, a clear trend emerges away from large
trades and towards small ones. While there are likely other causes for this trend besides
order splitting (such as the emergence of online brokerages and day trading) the
magnitude of the change points towards wide-spread changes in institutional trading
habits. This is consistent with the data presented in Chordia, Roll and Subrahmanyam
(2008) who find a similar trend, and conclude that institutional order splitting is the most
likely cause.
Secondly, I adopt a different technique for categorizing large and small traders
based on evidence presented in Campbell et. al. (2009). This paper examines the
correlation between different trade sizes and changes in reported institutional
shareholdings. While large trades tend to be positively associated with institutional
ownership, and most small trades negatively associated with institutional ownership, very
47
TABLE 11
Sensitivity Test and Trends in Small/Large Trading Patterns
Variables are defined in Appendix A. Selected estimates from estimation of equation (1) for different
sample periods.
Panel A:
1994-2001 2002-2007
Variable Name Estimate T-Stat Estimate T-Stat
EAday -0.0184 -1.43 -0.0048 -0.46
LG 0.0393 10.64 -0.0261 -8.13
EAday*LG 0.0508 2.91 0.0280 1.89
Panel B:
Average values by year of selected variables
Year LPROP SPROP LVOLPROP SVOLPROP
1994 0.390 0.270 0.808 0.040
1995 0.400 0.264 0.811 0.039
1996 0.414 0.252 0.816 0.036
1997 0.426 0.238 0.812 0.035
1998 0.415 0.244 0.800 0.038
1999 0.365 0.292 0.779 0.047
2000 0.359 0.307 0.771 0.052
2001 0.133 0.362 0.553 0.066
2002 0.086 0.450 0.553 0.108
2003 0.058 0.500 0.379 0.148
2004 0.056 0.482 0.348 0.152
2005 0.053 0.483 0.317 0.163
2006 0.042 0.517 0.269 0.194
2007 0.022 0.646 0.156 0.317
48
small trades (below $2,000) show a positive association, suggesting that many of these
trades represent institutional order splitting. I therefore re-run my main analyses using
negatively associated with institutional ownership, respectively. I classify trades between
$2,000 and $10,000 as small12
, and trades greater than or equal to $40,000 as large.
Untabulated results from this alternate specification are nearly identical to the previously
reported results. This suggests that the reported results are robust to alternate
specifications designed to reduce the amount of investor type misclassification.
Additionally, I test my data for the presence of multicollinearity. Computing
variance inflation factors for all the variables used in my main tests shows that none have
a VIF of over 10. Several, however, are relatively high, with SPROP, LPROP,
SVOLPROP, and LVOLPROP all being greater than seven. This is intuitive, as the
proportion of small traders is highly negatively correlated with the proportion of large
traders. As these are control variables, the relatively high VIF’s are not a concern, but in
any case removing them from the tests does not change the results. All other variables
have VIF scores under 3.5.
5. Conclusion
In this paper I test the effect of earnings announcements on the relative ability of
small and large investors to trade advantageously. While many studies have examined the
issue of the effect of earnings announcements on information asymmetry, using various
proxies for asymmetry, a clear consensus has yet to emerge. I measure information
12
Since the great majority of trades take place in round lots of 100 shares, eliminating trades under $2,000
will only eliminate trades for shares with a price of $20 or less. To address the possibility that very small
order splitting takes place for shares with a price above $20, I eliminate all single lot trades from this
analysis.
49
asymmetry between large and small investors (a proxy for informed and uninformed
investors) by examining how well those investors’ actual trades predict short-term price
movement. Results show that, on average, large traders tend to improve their trading
performance following an earnings announcement more than small traders. This effect is
strongest for firms which are very large and have large analyst followings, and for
earnings announcements which produce more private information about future earnings.
There is some evidence that large traders improve their relative performance when
earnings surprises (as measured by analysts’ forecasts) are high. Other attributes of
earnings surprises, however, fail to yield significant results. Similarly, examination of
attributes of intuitional ownership fails to find evidence that certain types of institutions
are more likely to increase the advantage of large trading on post earnings announcement
days.
This paper contributes to two different streams of literature. The first is the broad
category of research on the effect of earnings announcements on information asymmetry.
This paper differs from prior work in that it examines the consequences of information
asymmetry, rather than relying on estimations of the probability of informed trade.
Regulatory concern relating to accounting disclosures is focused on the consequences of
accounting disclosures on the trading behavior of investors. This paper examines this
issue directly by analyzing the short-term returns of actual trades.
Secondly, this paper contributes to the literature on volume patterns around
earnings announcements. Prior literature has suggested that one possible explanation for
the large amount of trading volume following earnings announcements is the “pent up
demand” of uninformed traders who were waiting for an information-leveling event
50
before executing their trades. Results in this paper suggest, however, that informed
traders increase their advantage following earnings announcements, while no such
advantage can be demonstrated in the period immediately preceding the announcement.
Collectively, these results are inconsistent with the idea that earnings announcement
trading volume patters are caused by the type of trade-shifting described in Chae (2005).
51
Appendix – Variable Definitions
PROFij The three day imputed profit earned by investor group j on
firm-day i. For group j (large or small), PROF is the value
weighted average of returns earned by buying or selling at
each actual transaction price, and reversing this position three
days later (the midpoint of daily close bid-ask three trading
days forward.) PROF is then multiplied by 100 to express it in
percentage terms, and winsorized at the 1% and 99% levels.
RET3i The decile ranked absolute value of the security’s return over
the three trading day period from i to i+3
NB3ij The net-buying activity of investor group j in the three days
prior to day i. Net buying is computed as the number of shares
bought by group j on firm-day i less the number of shares
sold, divided by the total number of shares transacted by
group j on firm-day i.
BA%i Bid-ask spread for firm-day i, expressed in percentage terms.
Computed as the closing ask less the closing bid, divided by
closing price on firm-day i.
LGij An indicator variable equaling 1 if investor group j is the large
investor group on firm-day i, and 0 otherwise.
EAdayi An indicator variable equaling 1 if firm-day i is in the post
earnings announcement period (days 0 or 1 relative to the
earnings announcement.)
EAprei An indicator variable equaling 1 if firm-day i is in the pre
earnings announcement period (days -1 or -2 relative to the
earnings announcement.)
SURP_RW_POSi Positive seasonal random walk earnings surprise. Seasonal
random walk is computed as the difference between earnings
per share in the earnings announcement closest to firm-day i
less earnings per share four quarters prior (adjusted for stock
transactions). If the surprise is positive, SURP_RW_POS
takes its value. Otherwise, it takes a zero value.
SURP_RW_NEGi Negative seasonal random walk earnings surprise. The
random walk earnings surprise is calculated as above, if it is
negative, then SURP_RW_NEGi takes its value. Otherwise, it
takes a zero value.
SURP_AN_POSi Positive analyst expectation earnings surprise. Surprise is
defined as the difference between actual EPS (as defined by
IBES) and the most recent forecast or forecast update prior to
the earnings announcement closest to day i. If an
announcement has no forecasts or updates within the 30 days
prior to the announcement, it is not assigned a value. If the
surprise is positive, SURP_AN_POS takes its value.
Otherwise, it takes a zero value.
SURP_AN_NEGi Negative analyst expectation earnings surprise. The analyst
52
expectation earnings surprise is calculated as above, if it is
negative, then SURP_AN_NEGi takes its value. Otherwise, it
takes a zero value.
MV_RANKi Decile rank of market value (defined as market price at the
end of the prior quarter multiplied by shares outstanding, as
defined by CRSP.
NUMANi Number of analysts following a firm. Defined as the number
of unique analyst codes in IBES (excluding 0000) identified
as making earnings forecasts for the announcement closest to
firm-day i.
Q4i An indicator variable that takes the value 1 if the earnings
announcement closest to firm-day i is for the firm’s fourth
fiscal quarter, and 0 otherwise.
C_CONSi Decile ranked change in analyst consensus for one quarter
ahead earnings. Consensus is defined as in Barron, Harris and
Stanford (2005) where Consensus = 1 – D/V. D is defined as
analyst forecast dispersion, and V is defined as the squared
differences in analysts’ forecasts:
Where FCi is analyst i’s earnings forecast, EPS is actual
earnings per share, and n is the number of forecasting
analysts. The measure represents the ratio of commonly held
information to private information. An increase (positive
C_CONS) represents a decrease in the proportion of private
information due to information released in the earnings
announcement closest to day i.
LPROPi The proportion of trades made by large traders to total trades
on firm-day i.
SPROPi The proportion of trades made by small traders to total trades
on firm-day i.
LVOLPROPi The proportion of trading volume generated by large traders
to total trading volume on firm-day i.
SVOLPROPi The proportion of trading volume generated by small traders
to total trading volume on firm-day i.
LAVGi The average number of shares traded per large trader
transaction on firm-day i.
SAVGi The average number of shares traded per small trader
transaction on firm-day i.
53
References
Admati, A. and P. Pfleiderer. 1988. “A theory of intraday patterns: Volume and price
variability.” Review of Financial Studies. 1(1): 3-40.
Ali, A., Durtschi, C., Lev, B. and M. Trombley. 2004. “Changes in institutional
ownership and subsequent earnings announcement abnormal returns.” Journal of
Accounting, Auditing, and Finance. 19(3): 221-248.
Argawal, P. and M. O’Hara, 2007. “Information risk and capital structure.” Cornell
University working paper.
Anderson, A. and E. Dyl, 2005. “Market structure and trading volume.” Journal of
Financial Research, 28(1): 115-131.
Asthana, S., S. Balsam, and S. Sankaraguruswamy. 2004. “Differential response of small
versus large investors to 10-K filings on EDGAR.” Accounting Review. 79(3): 571-589.
Atiase, R. 1985. “Predisclosure information, firm capitalization, and security price
behavior around earnings announcements.” Journal of Accounting Research. 23(1): 21-
36.
Atkins, A. and E. Dyl, 1997. “Market structure and reported trading volume, NASDAQ
versus the NYSE.” Journal of Financial Research, 20(3): 291-304.
Ayers, B., Li, O., and E. Yeung. 2009. “Investor trading and the post earnings
announcement drift.” University of Georgia working paper.
Ball, R., and S.P. Kothari, 1991. “Security returns around earnings announcements.”
Accounting Review, 66(4): 718-738.
Barber, B., and T. Odean. 2000. “Trading is hazardous to your wealth: the common stock
investment performance of individual investors.” Journal of Finance. 55(2): 773-806.
Barber, B., Y. Lee, Y. Liu, and T. Odean. 2008. “Just how much to individual investors
lose by trading?” Review of Financial Studies, forthcoming.
Barclay, M. and C. Dunbar, 1996. “Private information and the costs of trading around
quarterly earnings announcements.” Financial Analysts’ Journal, 52(6): 75-84.
Barclay, M., and J. Warner, 1993. “Stealth Trading and Volatility: Which Trades Move
Prices.” Journal of Financial Economics. 34(3): 281-305.
Barron, O., Kim, O., Lim, S., and D. Stevens, 1998. “Using Analysts’ Forecasts to
Measure Properties of Analysts’ Information Environment.” Accounting Review. 73(4):
421-433.
54
Barron, O., D. Byard, and O. Kim. 2002. “Changes in analysts’ information around
earnings announcements.” Accounting Review. 77(4): 821-846.
Barron, O., D. Harris, and M. Stanford. 2005. “Evidence that investors trade on private
event-period information around earnings announcements.” Accounting Review. 80(2):
403-421.
Battalio, R., and R. Mendenhall. 2005. “Earnings expectations, investor trade size, and
anomalous returns around earnings announcements.” Journal of Financial Economics.
77(2): 289-319.
Bhattacharya, N. 2001. “Investors’ trade size and trading responses around earnings
announcements: an empirical investigation.” Accounting Review. 76(2): 221-244.
Botosan, C., Plumlee, M., and Y. Xie. 2004. “The role of information precision in
determining the cost of equity capital.” Review of Accounting Studies, 9(2-3): 233-259.
Brennan, M. and A. Subrahmanyam, 1995. “Investment analysis and price formation in
securities markets.” Journal of Financial Economics, 38: 361-381.
Brooks, R. 1996. “Changes in asymmetric information at earnings and dividend
announcements.” Journal of Business, Finance & Accounting, 23(3): 359-378.
Brown, S., S. Hillegeist, and K. Lo, 2008. “The effect of earnings surprises on
information asymmetry.” University of Maryland working paper.
Bushee, B., 2001. “Do institutional investors prefer near-term earnings over long-run
value?” Contemporary Accounting Research, 18(2): 207-246.
Callahan, C., C. Lee, and T. Yohn, 1997. “Accounting information and bid-ask spreads.”
Accounting Horizons. 11(4): 50-60.
Campbell, J., Ramadorai, T., and Schwartz, A. 2009. “Caught on tape: Institutional
trading, stock returns, and earnings announcements.” Journal of Financial Economics.
92(1): 66-91.
Chae, J. 2005. “Trading volume, information asymmetry, and timing information.”
Journal of Finance. 60(1): 413-442.
Chordia, T., Roll, R., and A. Subrahmanyam. 2008. “Why has trading volume
increased?” Emory University working paper.
De Bondt, W. and R. Thaler, 1985. “Does the stock market overreact?” Journal of
Finance, 40(3): 793-805.
55
Easton, P. and M. Zmijewski. 1993. “SEC form 10-K/10-Q reports and annual reports to
shareholders: reporting lags and squared market model prediction errors.” Journal of
Accounting Research. 31(1): 113-129.
Ellis, K., R. Michaely and M. O’Hara. 2000. “The accuracy of trade classification rules:
evidence from Nasdaq.” Journal of Financial and Quantitative Analysis. 35(4): 529-551.
Foster, F., and S. Viswanathan. 1990. “A theory of the interday variations in volume,
variance, and trading costs in securities markets.” Review of Financial Studies. 3(4): 593-
624.
Glosten, L., and P. Milgrom, 1985. “Bid, ask and transaction prices in a specialist market
with heterogeneously informed traders.” Journal of Financial Economics, 14: 71-100.
Hasbrouck, J. 1991. “Measuring the information content of stock trades.” Journal of
Finance. 46: 179-207.
Hayn, C., “The Information Content of Losses.” Journal of Accounting and Economics,
20: 125-153.
Henker, T. and J. Wang. 2006. “On the importance of timing specifications in market
microstructure research.” Journal of Financial Markets. 9: 162-179.
Hirshleifer, D., J. Myers, L. Myers, and S. Teoh. 2008. “Do individual investors cause
post-earnings announcement drift? Direct evidence from personal trades.” Accounting
Review, 83(6): 1521-1554.
Hvidkjaer, S. 2008. “Small trades and the cross-section of stock returns.” Review of
Financial Studies, 21(3): 1123-1151.
Kandel, E. and N. Pearson. 1995. Differential interpetation of public signals and trade in
speculative markets. Journal of Political Economy 103 (4): 831-72.
Kaniel, R., G. Saar, and S. Titman. 2008. “Individual investor trading and stock returns.”
Journal of Finance, 63(1): 273-310.
Kaniel, R., Liu, S., Saar, G., and S. Titman, 2009. “Individual investor trading and return
patterns around earnings announcements.” Duke university working paper.
Ke, B. and S. Ramalingegowda, 2005. “Do institutional investors exploit the post-
earnings announcement drift?” Journal of Accounting and Economics. 39: 25-53.
Kim, O., and R. Verrecchia, 1991. “Trading Volume and Price Reactions to Public
Announcements.” Journal of Accounting Research, 29(2): 302-321.
56
Kim, O. and R. Verrecchia. 1994. “Market liquidity and volume around earnings
announcements.” Journal of Accounting and Economics. 17: 41-67.
Kim, O. and R. Verrecchia. 1997. “Pre-announcement and event-period private
information.” Journal of Accounting and Economics. 24: 395-419.
Krinsky, I., and J. Lee. 1996. “Earnings announcements and the components of the bid-
ask spread.” Journal of Finance. 51(4): 1523-1535.
Kyle, A. 1985. “Continuous auctions and insider trading.” Econometrica. 53: 1315-1335.
Lee, C. 1992. “Earnings news and small traders: an intraday analysis.” Journal of
Accounting and Economics. 15(2-3): 265-302.
Lee, C., and M. Ready. 1991. “Inferring trade direction from intraday data.” Journal of
Finance. 46(2): 733-746.
Lee, C., B. Mukclow, and M. Ready. 1993. “Spreads, depths, and the impact of earnings
information: an intraday analysis.” Review of Financial Studies. 6: 345-374.
Libby, T., R. Mathieu, and S. Robb, 2002. “Earnings announcements and information
asymmetry: an intra-day analysis.” Contemporary Accounting Research, 19(3): 449-472.
Malmendier, U., and D. Shanthikumar. 2007. “Are small investors naïve about
incentives?” Journal of Financial Economics. 85(2): 457-489.
Matsumoto, D., 2002. “Management’s Incentives to Avoid Negative Earnings Surprises.”
Accounting Review. 77(3): 283-514.
Morse, D. and N. Ushman, 1983. “The effect of information announcements on the
market microstructure.” Accounting Review, 58(2): 247-258.
Neal, R. and S. Wheatley, 1998. “Adverse selection and bid-ask spreads: evidence from
closed-end funds.” Journal of Financial Markets, 1: 121-149.
Puckett, A. and X. Yan. 2008. “The interim trading skills of institutional investors.”
University of Missouri Working Paper.
Roulstone, D. 2003. “Analyst following and market liquidity.” Contemporary Accounting
Research, 20(3): 552-578.
Stice, E. 1991. “The market reaction to 10-K and 10-Q filings and to subsequent the wall
street journal earnings announcements.” The Accounting Review, 66(1): 42-55.
Van Ness, B., R. Van Ness, and R. Warr, 2001. “How well do adverse selection
components measure adverse selection?” Financial Management, 30(3): 77-98.
57
Yohn, T. 1998. “Information asymmetry around earnings announcements.” Review of
Quantitative Finance and Accounting. 11: 165-182.
James D. Vincent Vita Pennsylvania State University WK: (814)-865-0573
Department of Accounting Cell: (607)-727-8904
371B Business Building Email: [email protected]
University Park, PA 16802
Education
Ph.D. Accounting, Pennsylvania State University Expected 2010
M.S. Accounting, Binghamton University 2005
B.A. History, Grinnell College 1997
Research Interests
Voluntary disclosure, Investor behavior, Trading volume, Information Economics,
Liquidity.
Research
Working Papers
“The Effect of Earnings Announcements on Trading Outcomes for Different Investor
Classes” Job market paper.
“Large and Small Investors’ Reactions to the Change in the Degree of Private
Information in the Market” with O. Barron and S. Chung. Currently under review at
Journal of Accounting Research
“Market Sentiment and the Reaction to Analyst Recommendation Changes” with O.
Barron and S. Chung.
Works in Progress
“Information Dissemination in Earnings Press Releases Versus Conference Calls”
“Liquidity and Disclosure Levels in Earnings Press Releases” with O. Barron and D.
Byard.
“Pro-Forma Earnings Disclosures and Information Quality”
Related Work Experience
Research Assistant and Instructor, Penn State University 2005-Present
Program Accountant, BAE Systems 2004-2005
Cost Accountant, BAE Systems 2003-2004