-
INDIVIDUAL INVESTORS AND CORPORATE EARNINGS
A DISSERTATION
SUBMITTED TO THE GRADUATE SCHOOL OF BUSINESS
AND THE COMMITTEE ON GRADUATE STUDIES
OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
Daniel Jeffrey Taylor
November 2010
-
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at:
http://purl.stanford.edu/kc064tf8516
© 2011 by Daniel Taylor. All Rights Reserved.
Re-distributed by Stanford University under license with the
author.
This work is licensed under a Creative Commons
Attribution-Noncommercial 3.0 United States License.
ii
http://creativecommons.org/licenses/by-nc/3.0/us/http://creativecommons.org/licenses/by-nc/3.0/us/http://purl.stanford.edu/kc064tf8516
-
I certify that I have read this dissertation and that, in my
opinion, it is fully adequatein scope and quality as a dissertation
for the degree of Doctor of Philosophy.
David Larcker, Primary Adviser
I certify that I have read this dissertation and that, in my
opinion, it is fully adequatein scope and quality as a dissertation
for the degree of Doctor of Philosophy.
William Beaver
I certify that I have read this dissertation and that, in my
opinion, it is fully adequatein scope and quality as a dissertation
for the degree of Doctor of Philosophy.
Charles Lee
Approved for the Stanford University Committee on Graduate
Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission
of this dissertation in electronic format. An original signed hard
copy of the signature page is on file inUniversity Archives.
iii
-
iv
ABSTRACT
This dissertation comprises two papers on the trading of
individual investors
around earnings announcements:
1. This study examines the effect of earnings announcements on
individual
investors’ trading decisions and their trading profits.
Consistent with
earnings news informing the trading decisions of individual
investors, I
find that earnings announcements are associated with significant
increases
in individual investor market participation, and that these
increases persist
even after controlling for the information in prices. Moreover,
and in
contrast to the conventional wisdom that disclosure benefits
unsophisticated investors at the expense of more sophisticated
investors, I
find that individuals’ trades around earnings announcements
earn
economically and statistically significant losses, and that
these losses are
significantly greater than the losses of non-announcement
trades.
Consistent with these losses resulting from inefficient
information
processing, I find the higher the information content of the
earnings
announcement the greater the loss, and that increased losses
around
earnings announcements are concentrated among those
individual
investors who are not classified as affluent or active traders.
Given the
limited information processing ability of individual investors,
the results
suggest a more nuanced view of the welfare effects of
disclosure.
-
v
2. This study examines the effect of contrarian retail trades on
the pricing of
earnings information. Consistent with price pressure from
contrarian retail
trades delaying the adjustment of prices to earnings
information, I find that
the negative price drift accompanying bad news is largest when
retail
investors buy on bad news, and that the positive price drift
accompanying
good news is largest when retail investors sell on good news.
These
findings are consistent with the correlated trading of retail
investors
around earnings announcements causing a delayed price adjustment
which
manifests as drift.
-
vi
ACKNOWLEDGMENTS
I thank Maureen McNichols and Bill Beaver—for encouraging me to
pursue
research on individual investors early in my doctoral studies,
and for detailed comments
on multiple versions of manuscripts, Dave Larcker—for countless
hours of advice on
research and life and what must be over a thousand dollars in
coffee, Charles Lee—for
exciting me about research in this area and for helping me to
understand its implications
and broader contribution, Ro Verrecchia—for helping me to focus
what was a descriptive
paper into one more closely tied to theory and the disclosure
literature, and Ian Gow—
for providing brutally honest feedback on a timely basis. I also
thank Mary Barth, George
Foster, Alan Jagolinzer, Gaizka Ormazabal, and Itamar Simonson
for advice and
encouragement, Stefan Nagel for sharing code, and Terry Odean
for sharing data. Also, I
am indebted to Jennifer Francis and Richard Willis, for their
willingness to advise a
starry-eyed graduate student in economics, and for introducing
me to accounting
research.
Finally, I thank my family for all of their support. I thank my
wife, Erin, for
tolerating late nights at the office, constant
absent-mindedness, and for cheering me up
when times got tough, and I thank my parents, Lou and Terry, and
grandparents,
Dorothy, George, and Johanna, for always encouraging me to
pursue my passions.
-
vii
TABLE OF CONTENTS
List of Tables
......................................................................................................................
x
Introduction..........................................................................................................................1
Chapter 1: Individual Investors, Earnings Announcements, and
Systematic Mistakes:
Can Trading on Information Make You Worse
Off?.....................................................5
1.
Introduction..............................................................................................................5
2. Hypothesis Development and Related Literature
..................................................11
2.1. Market Participation around Earnings Announcements
..............................11
2.2. Trade
Profitability........................................................................................11
3. Variable Measurement and Sample Selection
.......................................................19
3.1. Measures of Market
Participation................................................................19
3.2. Measures of Trading
Profits.........................................................................22
3.2.1. Implied Profits
....................................................................................22
3.2.2. Realized Trading Profits
.....................................................................23
3.3. Sample and Descriptive
Statistics................................................................26
3.3.1. Retail Broker Database
.......................................................................26
3.3.2. Primary Sample and Descriptive Statistics
.........................................29
4. Research
Design.....................................................................................................31
4.1. Individual Investor Market Participation
.....................................................31
4.1.1. Classic Event Study
............................................................................31
4.1.2. Relative Informativeness of
Earnings.................................................32
4.1.3. Do Earnings Inform Individual Investors Incremental to
the
Information in
Prices..................................................................................34
4.2. Individual Investor Trading
Profits..............................................................31
4.2.1. Changes in Trading Profits Around Earnings
Announcements..........35
4.2.2. Trade Profitability and the Information Content of
Earnings.............37
4.2.3. Who Loses Around Earnings Announcements?
.................................38
5.
Results....................................................................................................................41
-
viii
5.1. Individual Investor Market Participation Around
Earnings.........................41
5.2. Individual Investor Trading Profits Around Earnings
Announcements ......45
5.3. Cross-Sectional Variation in Trading Profits Around
Earnings
Announcements................................................................................................47
6. Conclusion
.............................................................................................................51
Chapter 2: Retail Investors and the Adjustment of Stock Prices
to Earnings
Information
..................................................................................................................52
1.
Introduction............................................................................................................52
2. Related
Literature...................................................................................................58
2.1. Retail Investors
............................................................................................58
2.2. Post-Earnings Announcement Drift
.............................................................61
3. Research Design and
Predictions...........................................................................64
3.1. Primary Asset Pricing
Tests.........................................................................64
3.1.1. Portfolio Sorts
.....................................................................................65
3.1.2. Return Regressions
.............................................................................67
3.2. Additional
Analyses.....................................................................................68
3.2.1. Speed of Price
Adjustment..................................................................68
3.2.2. Additional Determinants of
Drift........................................................69
3.2.3. Limited
Arbitrage................................................................................71
3.3. Measures of Primary
Variables....................................................................73
3.3.1. Measures of Earnings Surprise
...........................................................73
3.3.2. Measures of Contrarian Trade
............................................................73
4. Sample and Descriptive
Statistics..........................................................................75
4.1. Retail Investor Database
..............................................................................75
4.2. Primary
Sample............................................................................................77
4.3. Descriptive
Statistics....................................................................................78
5.
Results....................................................................................................................80
5.1. Primary Asset Pricing
Tests.........................................................................80
5.1.1. Portfolio Sorts
.....................................................................................80
5.1.2. Return Regressions
.............................................................................82
-
ix
5.2. Additional
Analyses.....................................................................................83
5.2.1. Speed of Price
Adjustment..................................................................83
5.2.2. Additional Determinants of
Drift........................................................84
5.2.3. Limited
Arbitrage................................................................................86
6. Robustness
.............................................................................................................80
6.1. Alternative Measures of Earnings Surprise
.................................................87
6.2. Abnormal Return
Measurement...................................................................88
6.3. Investor Attention Explanations
..................................................................89
6.4. Omitted Variable and Risk-based Explanations
..........................................91
7. Conclusion
.............................................................................................................92
Appendix A: Tables
...........................................................................................................93
Bibliography
....................................................................................................................107
-
x
LIST OF TABLES
Number Page
Table 1: Descriptive Statistics
...........................................................................................93
Table 2: Market Participation Around Earnings Announcements
.....................................95
Table 3: Individual Investor Trading Volume: Controlling for
Returns............................97
Table 4: Implied Trading Profits Around Earnings Announcements
................................99
Table 5: Realized Trading Profits Around Earnings
Announcements.............................100
Table 6: Trading Profits and Announcement Period Information
...................................102
Table 7: Trading Profits Around Earnings Announcements and
Investor
Demographics
................................................................................................104
Table 8: Retail Investor Database
....................................................................................107
Table 9: Descriptive Statistics - Sample
Variables..........................................................108
Table 10: Portfolio
Sorts..................................................................................................110
Table 11: Future Return
Regressions...............................................................................111
Table 12: Future Return Regressions: Multi-Quarter Evidence
......................................112
Table 13: Portfolio Sorts: Controlling for Additional
Factors.........................................113
Table 14: Future Return Regressions: Controlling for Additional
Variables ..................115
Table 15: Portfolio Sorts: Effects of Limited
Arbitrages.................................................116
Table 16: Portfolio Returns: Analyst Forecast
Errors......................................................117
-
1
INTRODUCTION
The primary theme of my work is individual investors. A large
and growing
literature examines the trading decisions of individual
investors and how such decisions
affect asset prices. Within this literature, one stream of
research examines the
performance of individual investors’ stock portfolios. Studies
in this stream find that
individual investors’ portfolios consistently underperform the
market. A second stream
of research in this literature takes individual investors’
trades as given and examines how
these trades affect asset prices. Studies in this stream
conclude that individual investors’
trades negatively predict asset prices. Relatively less research
examines the information
individual investors use in their investment decisions and how
they use such information.
Surprisingly, the existing literature remains silent on the
accounting information
that individuals use in their investment decisions and whether
the use of such information
leads to better investment decisions. Traditional models of
disclosure assume
homogenous information processing capabilities, and predict that
information events
reduce information asymmetry in the marketplace and benefit
unsophisticated, individual
investors. However, more nuanced models of disclosure in which
investors have
heterogeneous information processing abilities or heterogeneous
levels of
overconfidence, predict that information events may benefit
sophisticated investors at the
expense of individual investors—who are limited in their ability
to process information or
process it inefficiently (i.e. display overconfidence). While
the latter characterization of
individual investors is broadly consistent with evidence in the
experimental literature, the
-
2
implication that trades based on accounting information may lead
to worse outcomes for
unsophisticated investors is inconsistent with the conventional
wisdom in the disclosure
literature that more information leads to more informed
investing decisions.
I investigate these issues in chapter one. In particular, I
examine whether
individual investors use earnings information in their trading
decisions and whether the
use of such information leads to more informed trading decisions
(i.e. whether such
trades perform better). Consistent with individual investors
using earning information in
their trading decisions, I find a pronounced spike in individual
investor trading volume,
the number of individual investors participating in the equity
market, and the number of
first-time individual investors participating in the equity
market around earnings
announcements. In relative terms, the increase in volume and
market participation
around earnings announcements is greater than the increase in
volume and market
participation on days with similar price changes but no earnings
announcements. This
suggests that for the same amount of value-relevant information,
individuals place
disproportionate weight on earnings information.
However, despite such pronounced increases in trading activity,
individual
investors’ trades around earnings announcement not only earn
negative market adjusted
returns (even before considering transaction costs), but also
earn lower returns than non-
announcement trades. This suggest that individual investors
would be better off (from a
wealth standpoint) had they not traded around the information
disclosure. Moreover, and
consistent with these losses resulting from inefficient
information processing, I find that
announcement period losses are increasing in the amount of
value-relevant information
-
3
released on the announcement day and decreasing in the
individuals’ wealth and
historical trading frequency.
The findings of chapter one suggest that individual investors
make systematic
mistakes when trading on earnings information, and that earnings
announcements
facilitate a wealth transfer from individual investors to more
sophisticated investors. This
is in stark contrast to the conventional wisdom in the
disclosure literature, which argues
that disclosure of information should level the information
playing field. Instead the
results are consistent with more nuanced theories of disclosure,
in which the disclosure of
information benefits sophisticated investors at the expense of
unsophisticated investors.
At a minimum, the results suggest that individuals should avoid
trading around corporate
disclosures because other investors can more efficiently process
information.
Having shown a pronounced increase in individual investor
trading around
earning announcements, in chapter two, I examine the effect of
such trades on asset
prices. A long-enduring anomaly in the capital markets
literature is the tendency for
prices to drift in the direction of the earnings surprise up to
one year following the
earnings announcement. Two broad classes of explanations for
this drift have been
proposed in the literature: risk, and the trading of
unsophisticated investors. Building on
this literature, and the results of chapter one, I conjecture
that when individuals trade in
aggregate in the direction opposite the earning surprise, their
trades exert sufficient price
pressure to delay the adjustment of prices to earnings
information, giving rise to price
drift. Consistent with the contrarian trade of individual
investors delaying price
adjustment, I find that prices drift in the direction of the
earnings surprise only in
-
4
instances where individuals in net traded in the direction
opposite the earnings surprise,
and that these trades delay price adjustment only in those
stocks with high limits to
arbitrage. This supports the conjecture that post-earnings
announcement drift is related to
the trading of unsophisticated investors, and that, when
combined with limits to arbitrage,
the correlated trading of individual investors can cause prices
to temporarily deviate from
fundamental values.
Collectively, the evidence in chapter one suggests that
individual investors make
systematic mistakes when trading around earnings announcements,
and the evidence in
chapter two suggest that the correlated mistakes of individual
investors affect asset prices
around earnings announcements.
-
5
CHAPTER 1: INDIVIDUAL INVESTORS, EARNINGS ANNOUNCEMENTS, AND
SYSTEMATIC MISTAKES: CAN TRADING ON INFORMATION MAKE YOU
WORSE OFF?
1. INTRODUCTION
The SEC is charged with ensuring “fair” disclosure, where “fair”
is taken to mean
reducing the extent to which certain investors are at an
informational disadvantage, or
“leveling the information playing field.” An integral part of
the SEC’s mission is
protecting the welfare of individual investors (Cox, 2006). The
FASB, while not
explicitly charged with protecting individual investors, has
also shown concern for
individual investors.1 Many SEC and FASB deliberations
implicitly or explicitly reflect
concerns about whether individual investors’ trading decisions
would be adversely
affected by changes in disclosure policy. For example, in late
1990s, the FASB sought
comments on whether “[g]iven [efficient] markets[,] would any
disservice be done to the
interest of individual investors by allowing professional
investors access to more
extensive information?” (Bloomfield, Libby, and Nelson, 1999,
emphasis added).
Similarly, much of the debate over two-tiered financial
reporting reflects concerns that
providing individual investors’ with abbreviated financial
statements would adversely
affect their trading decisions (e.g., Bushman, Gigler, and
Indejikian, 2002). Relatedly,
Regulation Fair Disclosure (Reg FD) prohibits isolated
disclosure of non-public
information to a subset of market participants (e.g., analysts)
on the basis of “fairness” to
1 The FASB states that the objectives of financial reporting
“stem primarily from the informational needs of external users who
lack the authority to prescribe the financial information they want
from an enterprise and therefore must use the information that
management communicates to them.” (SFAC 1, paragraph 28).
-
6
individual investors (SEC, 2000). Despite the strong regulatory
and standard setting focus
on individual investors, and decades of research on how
accounting information affects
the stock market, empirical evidence on how accounting
information affects individual
investors’ trading decisions and trading profits is largely
absent from the literature (see
Bamber, Barron, and Stevens, 2008 for a review).2
Much of the recent empirical literature focuses on the relation
between accounting
information and the cost of capital. However, the effect of
accounting information on the
cost of capital can be very different from the effect of
accounting information on investor
welfare (e.g., Gao, 2010). While many extant disclosure models
make explicit predictions
regarding the effect of accounting disclosures on the trading
profits of unsophisticated
investors (see Verrecchia, 2001 for a review), there is scant
empirical evidence on this
relation. This is an important gap in the literature, because if
regulators require
disclosure under the auspices that such disclosure “levels the
information playing field”
and benefits individual investors, it is important to
demonstrate that individual investors
use the information in the disclosure and that this information
improves their trading
decisions.
Using data on the actual trades of individual investors, this
study examines
whether and how individual investors process earnings
information. In particular, this
study focuses on two fundamental, unresolved issues: (i) whether
earnings information
informs the trading decisions of individual investors
incremental to the (earnings)
2 In fact, the absence of research on individual investors is
one reason some argue that market-based studies of accounting
information are of limited use to regulators (Holthausen and Watts,
2001, p. 27). With regard to the experimental studies, Kachelmeier
and King (2002) suggest research on how individual investors
process accounting information has been limited by the perception
that such research is not relevant if individuals do not affect
asset prices (see Libby, Bloomfield, and Nelson, 2002 for a
review).
-
7
information in prices, and (ii) whether individual investors
benefit from trading around
earnings announcements.
To investigate whether individual investors use earnings
information in their
trading decisions incremental to the information in prices, I
examine whether individual
investor participation in equity markets increases around
earnings announcements and
whether this increase persists after controlling for past and
contemporaneous returns.
While early studies find increased small trade volume around
earnings announcements
(e.g., Lee, 1992), using the actual trades of individual
investors, recent behavioral finance
research suggests that individuals chase returns and trade
following extreme price
movements (e.g., Grinblatt and Keloharju, 2000; Barber and
Odean, 2008). This literature
argues that increased individual trading around earnings
announcements is the result of
individuals trading on extreme price changes, and not
information in the earnings
announcement per se. While subtle, this is an important
distinction with implications for
disclosure regulation. For example, in justifying Reg FD on the
basis of “fairness” to
individual investors, regulators implicitly assume that
individual investors trade on the
information communicated through disclosure above and beyond the
information
impounded in prices. If individual investors trade solely on the
information in prices, then
for the purposes of individual decision-making, it does not
matter how information is
disseminated, or to whom, only that it is quickly and fully
reflected in prices.
To investigate whether individual investors benefit from trading
around earnings
announcements, I compare the profitability of individual
investors’ trades around
earnings announcements to that of their non-announcement trades.
I decompose trading
-
8
profits into transaction costs (inclusive of the bid/ask spread)
and gross returns (i.e.,
trading profits excluding transaction costs) and examine each
component separately.3
While Odean (1999) documents that trades by individual investors
lose money on
average, how accounting information in general—and earnings
information
specifically—affects individual investors’ trading profits
remains an open question. The
conventional wisdom in the disclosure literature is that
disclosure decreases the
information rents of informed traders and increases liquidity
(e.g., Verrecchia, 2001;
Leuz and Wysocki, 2009). In this regard, the conventional wisdom
predicts that
accounting information affects individuals’ trading profits via
a reduction in trading
frictions. With regard to earnings announcements, this suggests
that individuals’
announcement period trades are more profitable than
non-announcement trades because
of a reduction in transaction costs.
However, more nuanced models of disclosure suggest two
circumstances in which
individuals’ announcement period trades may actually be less
profitable than non-
announcement trades. First, if earnings announcements increase
the information rents of
more sophisticated traders, then individuals’ announcement
period trades will have
higher transaction costs (e.g., Kim and Verrecchia, 1994, 1997).
Second, individuals may
process earnings information inefficiently. If efficient and
inefficient information
processors trade on information in an earnings announcement,
efficient information
processors benefit at the expense of the inefficient information
processors (e.g., Fischer
3 Throughout the paper when I use the term “trading frictions”
or “transaction costs” I refer not only to the fixed costs of
trading (i.e. commission paid to the broker) but also the bid/ask
spread implicitly paid by the trader.
-
9
and Verrecchia, 1999). This suggests announcement period trades
are less profitable than
non-announcement trades even after excluding transaction costs.
Whether the
conventional wisdom, or these more nuanced models best describe
the effect of earnings
announcements on individual investors is an unanswered
question.
The findings of this study are as follows. First, I find
individual investor
participation in equity markets measured as (i) individual
trading volume, (ii) the number
of individual investors trading, and (iii) the number of
individuals trading in the firm’s
shares for the first time, increases around earnings
announcements. Additional analysis
suggests that individual volume on the day of the announcement
accounts for 13.27% of
the variation in individual volume over the quarter, more than
twice that of the average
non-announcement day. These findings suggest earnings
announcements are associated
with a pronounced increase in individual investor trading
activity. In addition, using two
distinct sets of tests, I find that individual investors trade
on the information in the
earnings incremental to the information in prices. First, I sort
earnings announcements
into quintiles based on both the magnitude of pre-announcement
and announcement
period returns. I find significant increases in market
participation across all quintiles.
Second, I match each announcement day with a non-announcement
day in the same firm,
based on the magnitude of past and contemporaneous returns.
Using this within-firm
matched sample design, I find earnings announcements have
significantly higher levels of
individual investor market participation than corresponding
non-announcement days with
similar past and contemporaneous returns. Taken together, these
results suggest
-
10
individual investors trade on earnings information incremental
to the information in
prices.
Second, I find that individuals’ trades around earnings
announcements earn
economically and statistically significant losses, and that
these losses are significantly
greater than the losses of non-announcement trades.
Investigating the source of these
losses, I find evidence that increased losses are the result of
both increased transaction
costs and inefficient information processing. In particular, and
consistent with earnings
announcements increasing the information rents of more informed
traders, I find that
individuals’ announcement period trades incur larger transaction
costs than non-
announcements trades. Additionally, I find (1) that announcement
period trades earn
larger losses than non-announcement trades even after excluding
transaction costs, (2)
that these increased losses are concentrated in firms for which
earnings reveal a large
amount of new information, and (3) that these losses are
concentrated among individual
investors who are not affluent and are not active traders. These
findings suggest losses
around earnings announcements are also attributable to
inefficient information
processing.
This study contributes to the literature by providing a detailed
examination of the
effect of earnings announcements on individual investors’ equity
market participation and
trading profits. Collectively, the results challenge the notion
that disclosure necessarily
“levels the information playing field.” Instead, the results
suggest that around earnings
announcements sophisticated investors gain at the expense of
individual investors. The
results are consistent with extant analytical models in which
disclosure increases the
-
11
information rents of more informed investors, and facilitates a
wealth transfer from
inefficient to efficient information processors (e.g., Kim and
Verrecchia, 1994, 1997;
Fischer and Verrecchia, 1999). Given the limited information
processing ability of
individual investors, the results suggest a more nuanced view of
the welfare effects of
disclosure.
The remainder of the paper proceeds as follows. Section 2
discusses the related
literature and develops the hypotheses. Section 3 describes the
sample and measurement
of key variables. Section 4 discusses the research design.
Section 5 presents the results,
and Section 6 concludes.
2. HYPOTHESIS DEVELOPMENT AND RELATED LITERATURE
2.1. MARKET PARTICIPATION AROUND EARNINGS ANNOUNCEMENTS
Beginning with Beaver (1968), much of the accounting literature
examines the
effect of earnings announcements on trading volume (see Bamber,
Barron, and Stevens,
2009 for a review). Papers in this literature examine whether
the volume reaction to
earnings announcements varies across firms (e.g., Bamber, 1987),
across time (e.g.,
Landsman and Maydew, 2002), and test whether these variations
are consistent with
investors trading on private information (e.g., Barron Harris
and Stanford, 2005).4 Within
this volume literature, one stream of research focuses on how
the reaction to earnings
announcements varies across market participants. For example,
using TAQ data, Lee
(1992) finds increased small trade volume around earnings
announcements. Similarly,
4 For parsimony, I focus only on studies examining volume around
earnings announcements rather than other disclosures.
-
12
Bhattacharya (2001) finds that small trade volume at the time of
the earnings
announcement is increasing in the magnitude of forecast errors
from a seasonal random
walk model of earnings.5 In a recent study, Dey and Radhakrishna
(2007) use data on
actual orders from all individual investors on the NYSE for 144
firms over a three-month
period (i.e., TORQ data) and find a surge in volume across all
market participants at the
time of the announcement. They also find that individual
investors account for 30.1%
(11.2%) of all trades (volume) around earnings
announcements.
Collectively, the results of this literature suggest that
individual investor volume
increases around earnings announcements. However, it is unclear
why. On the one hand,
if individuals have (or perceive that they have) private
information not impounded in
prices, trading around the announcement may reflect speculation.
On the other hand, if
individuals view the market as fully reflecting all information,
trading activity may be
explained by risk preferences. For example, if earnings contain
information about a
firm’s risk (e.g., Beaver, Kettler, and Scholes, 1970; Fama and
French, 1993), then trades
around the announcement may reflect timely decisions with regard
to long-term asset
allocation (i.e. portfolio rebalancing based on new information
about risk). Still another
reason why individuals may trade around earnings announcement is
that trading during
such periods minimizes the rents paid to more informed traders
(e.g., Admati and
Pfleiderer, 1988).
5 Additionally, Battalio and Mendenhall (2005) find the
direction of small trades around the earnings announcement has a
higher correlation with seasonal random walk forecast errors than
analyst forecast errors, Shanthikumar (2004) shows this correlation
is increasing in the forecast error from prior quarters, and
Bhattacharya et al. (2007) show this correlation is larger for
pro-forma earnings rather than GAAP earnings.
-
13
Recent behavioral finance research suggests increased individual
investor trading
activity around earnings announcements may not be the result of
individuals trading on
the information in earnings per se. For example, using the
actual trades of individual
investors, prior work finds that individuals tend to trade more
in stocks with extreme
returns over the past three months (e.g., Grinblatt and
Keloharju, 2000). Similarly, prior
work also finds that individuals display an “attention effect”
buying on days immediately
following extreme price changes (e.g., Barber and Odean, 2008).
Because many stocks
experience extreme price changes prior to the earnings
announcement (e.g., Ball and
Brown, 1968), these findings raise the possibility that
increased individual trading around
earnings announcements may be the result of individuals trading
on information in prices
(i.e. extreme returns), rather than the information in the
earnings announcement. If
individuals do not trade on information in accounting
disclosures incremental to the
information in prices, then the disclosure itself is not
relevant for their decision making,
only the information impounded in prices. In contrast, if
earnings inform the trading
decisions of individuals incremental to prices, it suggests that
individuals base their
trading decisions on information other than price.
I extend this literature by testing various explanations for
increased individual
investor trade around earnings announcements. First, I test
whether individuals trade
around earnings announcements after controlling for the
information in prices. If
individuals trade on earnings incremental to the information in
prices, then I expect to
find increased individual trading volume around the earnings
announcement even after
controlling for the information in pre-announcement and
announcement period returns.
-
14
Additionally, I expect to find that individual volume around
earnings announcements is
greater than the individual volume around non-announcement days
with similar past and
contemporaneous returns. Second, I test whether individual
trades around earnings
announcements are motivated by a reduction in trading frictions.
If increased individual
trade around earnings announcements is motivated by reduced
trading frictions then I
expect to find transaction costs (inclusive of the bid-ask
spread) are lower around
earnings announcements.
2.2. TRADE PROFITABILITY
A growing literature in behavioral finance examines the
portfolio choices and
investment performance of individual investors (see Campbell,
2006 for a review).
Barber and Odean (2000) find that investors at a large discount
broker tilt their portfolios
toward small value stocks, pay 3% in commissions and 1% in
bid-ask spread, and that
their portfolios earn significant abnormal returns of –3.7% net
of transaction costs.6
Odean (1999) examines the profitability of individual investor
trades (as opposed to
portfolio positions) and finds that stocks sold by individuals
earn higher subsequent
returns than stocks bought by individuals. Examining
cross-sectional variation in trading
profits, Seru, Shumway, and Stoffman (2009) find that more
experienced investors are
more profitable. Linnainmaa (2009) finds that the market orders
of individual investors
earn higher returns than limit orders, consistent with traders
that demand immediacy
6 Prior to including transaction costs, Barber and Odean (2000)
find that market-adjusted and Fama-French adjusted portfolio
returns of individual investors is insignificantly different from
zero (Table 2, p. 787). Using a comprehensive dataset of all
individual investors in Taiwan, Barber et al. (2009) find that the
annual net losses of individual investors is equivalent to 2.2% of
Taiwan’s GDP.
-
15
having superior information or skill, and Grinblatt, Keloharju,
and Linnainmaa (2009)
show that the trader’s intelligence quotient (IQ) is positively
related to trading profits.
However, what effect (if any) accounting information has on
individuals’ trading
profits is unclear. Some argue that if markets are
informationally efficient, then investors
who rely on public accounting information are “price protected,”
in which case
accounting information—so long as it is priced efficiently—will
have no effect on trading
profits (e.g., Kothari, Ramanna, and Skinner, 2009).7 However,
this “price protection”
argument takes prices as given and ignores the effect of
accounting information on
trading frictions and the formation of equilibrium prices (e.g.,
Grossman and Stiglitz,
1980). In this regard, the extant disclosure literature suggests
a more nuanced view of
capital markets, in which accounting information can affect
individuals’ trading profits.
The conventional wisdom is that disclosure “benefits the
uninformed at the
expense of the informed” and results in a decrease in the rents
to private information and
an increase in liquidity.8 Under this view, earnings
announcements reduce the
information rents of informed investors, such that trades of
individual investors around
earnings announcements should be more profitable than their
non-announcement trades.
In contrast to the conventional wisdom, several studies suggest
the possibility that
earnings announcements facilitate a wealth transfer from
individual investors to more
sophisticated investors.
7 This argument dates back to the early literature on market
efficiency and its implications for accounting standard setters
(e.g., Beaver and Demski, 1974; Gonedes and Dopuch, 1974). 8 See
Leuz and Wysocki (2009) for a review of the disclosure literature
and its effect on market liquidity and adverse selection.
-
16
There are two non-mutually exclusive reasons to suspect
individual investors’
trades around earnings announcements may actually be less
profitable than their non-
announcements trades. First, the conventional wisdom is
predicated on the notion that
disclosure reduces private information. However, Indejikian
(1991) shows that when
information processing is costly, disclosure is associated with
an increase in private
information collection.9 Similarly, Kim and Verrecchia (1994,
1997) show that if
investors have private contextual information useful in
conjunction with the information
in the disclosure (e.g., private information about the
persistence of earnings), then
disclosure, and earnings announcements specifically, will
increase the information rents
of sophisticated investors. Consistent with increased
information rents at the time of the
announcement, Lee, Mucklow, and Ready (1993) find that the
bid/ask spread increases
on the day of and the day following the announcement, and
Krinsky and Lee (1996) find
a significant increase in the adverse selection component of the
spread around earnings
announcements, but no change in total bid/ask spread.
Second, the conventional wisdom is predicated on the notion that
unsophisticated
investors are efficient information processors. Even in the
absence of information
asymmetry, if individual investors process information
inefficiently, then their
information-based trades may be less profitable then their
non-information based trades.
Considering public disclosure in the context of efficient
(rational) and inefficient
(overconfident) traders, Fischer and Verrecchia (1999) show that
if the market is
composed of a sufficient number of rational traders, earnings
announcements will
9 Indjejikian (1991) suggests that investors who face higher
information processing costs will rely more on commonly observed
information sources (e.g., prices).
-
17
facilitate a wealth transfer from inefficient information
processors (i.e., non-Bayesians) to
the efficient information processors (i.e., Bayesians). While
individuals are known to
inefficiently process earnings information in laboratory
settings (e.g., Maines and Hand,
1996) it remains unclear what effect (if any) this has on their
trading profits. Whether the
“price protection” argument, the conventional wisdom, or these
more nuanced models
best describe the effect of earnings announcements on individual
investors is an
unanswered question.
Hints of the relation between individual investors’ trading
profits and earnings
information can be seen in a few recent studies. Elliott, Hodge,
and Jackson (2008)
examine survey data from 339 non-professional investors, and
find that investors’ self-
reported trading profits are decreasing in the ratio of the
number of times the investor
used information in SEC filings to the number of times the
investor used information
reported by Market Guide, Value Line, and Standard and Poor’s.
Although not focusing
on earnings announcements, Asthana et al. (2004) use the
five-day buy-and-hold return to
small trades on TAQ as a measure of the trading profits of
unsophisticated investors, and
find 10-K filings on EDGAR are associated with increased
profits. Also using this
measure of trading profits, Miller (2008) finds the
profitability of small trades around 10-
K dates is negatively related to measures of 10-K “readability,”
and Mikhail et al. (2007)
find that small trades placed around analyst recommendation earn
losses. Each of these
studies is careful to acknowledge the limitations of using TAQ
data to infer trade
direction and sophistication of traders, and computing profits
assuming a five-day
holding period.
-
18
More recently, Hirshleifer et al. (2009) and Kaniel et al.
(2009) find that
individual investors’ net purchases around earnings
announcements negatively predict
asset prices. While primarily interested in the relation to
post-earnings announcement
drift, the results of these studies suggest that individual
investors’ trades around earnings
announcements earn losses. However, given that prior work
documents a negative
correlation between individual investor trades and returns in
general (e.g., Odean, 1999;
Barber, Odean, and Zhu, 2009), it is unclear whether the results
of Hirshleifer et al.
(2009) and Kaniel et al. (2009) are unique to trades around
earnings announcements.
While the results in prior work may hint that trades around
earnings announcements lose
money, relevant to this study is whether the trades around
earnings announcements are
more or less profitable than non-announcement trades, and
whether changes in
profitability are due to changes in trading frictions or
inefficient information processing.
To the best of my knowledge this is the first study to address
these questions.
This study extends this literature by examining the effect of
earnings
announcements on the trading profits of individual investors
using data on the actual
trades of individual investors, with a particular focus on
whether differences in trade
profitability results from trading frictions or inefficient
information processing. If
changes in profits around earnings announcements are due to
changes in trading frictions
(e.g., Kim and Verrecchia, 1994, 1997), I expect to find
earnings announcements are
associated with the amount individual investors pay in bid/ask
spread, and unrelated to
trade profitability after excluding such transaction costs. If
changes in profits around
earnings announcement are associated with inefficient
information processing (e.g.,
-
19
Fischer and Verrecchia, 1999), I expect to find announcement
period trades earn lower
profits (larger losses) than non-announcement trades even after
excluding transaction
costs, that the higher the information content of the earnings
announcement the larger the
loss, and that changes in profits are most pronounced among
those investors least likely
to understand the valuation implications of earnings news.
3. VARIABLE MEASUREMENT AND SAMPLE SELECTION
3.1. MEASURES OF MARKET PARTICIPATION
In this section I develop a model that yields testable
predictions regarding the
relationship between debt contracting, monitoring of the
financial reporting process and
properties of the reported financial numbers, such as
conservatism. While, as discussed
above, much of the literature on debt covenants has focused on
signaling and
renegotiation, I focus on a setting with symmetric information
at the time of contracting
and essentially no renegotiation. While these features are
undoubtedly important in
practice (Dichev and Skinner, 2002; Gârleanu and Zwiebel, 2008;
Nini, Smith, and Sufi,
2008), the imperfection of accounting information, incomplete
contracting and costly
renegotiation are also likely to be important. With regard to
renegotiation, the model I use
here might be viewed as a limiting case in which renegotiation
is sufficiently costly to
effectively rule it out.
My primary measure of individual investor market participation
is individual
investor trading volume. Similar to prior research on total
volume (e.g., Landsman and
-
20
Maydew, 2002), I compute abnormal individual investor volume
(AbIV) for firm i in
quarter q on date t as
qi
tqitqi
tqi
IVEIVAbIV
,
,,,,
,,
][
σ
−
= (1)
where IVi,q,t is individual trading volume for firm i, in
quarter q on date t scaled by shares
outstanding, E[IVi,q,t] is the predicted value from a modified
market model of individual
volume estimated for each firm-quarter over the sixty-one
trading days surrounding the
earnings announcement and excluding the five trading days
surrounding the
announcement (i.e. t = -30…-6 , 6…30), and σi,q is the standard
deviation of market
model residuals for firm i in quarter q. Specifically I
estimate
IVi,q,t = αi,q + β1i,q Vm
t + β2i,q IVm
t + ε i,q,t t = –30…–6 , 6…30 (2)
Vm is market-wide total volume scaled by shares outstanding, and
IVm is market-wide
individual volume scaled by shares outstanding. Market-wide
total volume (Vm) controls
for market-wide factors that affect volume (e.g. macro-economic
news), and market-wide
individual investor volume (IVm ) controls for changes in
individual investor volume that
are the result of correlated individual biases (e.g., individual
investor sentiment),
individual specific news events (e.g., the release of quarterly
account statements), or a
difference in the relevance of macro-economic news between
individual and non-
individual investors (e.g., a change in the discount rates).
Since the regression is
estimated for every firm-quarter, the intercept serves as a
firm-quarter fixed effect and
filters out any unobserved cross-sectional variation either
across firms or across quarters,
-
21
leaving only intra-quarter temporal variation.10 These
modifications to the standard
market model (e.g., Beaver, 1968) ensure that market-wide
variation in individual
volume is filtered out and that abnormal individual volume is
cross-sectionally
comparable (i.e., similar mean and variance). For comparison, I
use the same procedure
to compute abnormal total volume (AbV) but omit the market-wide
individual volume
(IVm) term in equation (2).
In order to address concerns that changes in individual volume
around earnings
announcements may be driven by a small subset of individual
investors, I also use two
non-volume based measures of market participation. The first,
IndInv, is the number of
individual investors trading in firm i, in quarter q on date t,
scaled by the number of
individuals on the dataset. The second, NewIndInv, is based on
the number of new
individual investors trading in the stock (i.e. those
individuals who were not previous
shareholders) scaled by the number of individuals on the
dataset. As with individual
volume, I compute the abnormal number of individual investors
(AbII) and abnormal
number of new individual investors (AbNII) relative to a
modified market model. For the
former, the market model includes market-wide total volume (Vm)
on day t and the sum
of IndInv across all firms on day t. For the latter, the market
model includes market-wide
total volume (Vm) on day t and the sum of NewIndInv across all
firms on day t.
10 The difference in average individual volume for across firms
will be reflected in the intercepts. For example, if individual
volume in Firm A is on average higher than Firm B, Firm A’s
intercept will be larger than Firm B’s. Similarly, within-firm
differences in average individual volume will also be reflected in
the intercept. For example, if individual volume in Firm A in
quarter q is higher than that in quarter r, Firm A’s quarter q
intercept will be higher than quarter r.
-
22
3.2. MEASURES OF TRADING PROFITS
Much of the prior research on individuals investors focuses on
the performance of
their portfolio holdings rather than the profitability of
specific trades (e.g., Barber and
Odean, 2000; Barber and Odean, 2001; Barber et al., 2009). In
contrast, my hypotheses
are in regard to the profitability of individuals’ trades around
earnings announcements,
rather than the overall performance of their portfolios. Thus,
testing my hypotheses
requires a trade-specific measure of risk-adjusted trading
profits. I use two
complementary methods to estimate risk-adjusted trading
profits.11 The first method uses
no information about the trade other than the trade date and
trade direction, and estimates
trading profits as the abnormal return in the direction of the
trade assuming a constant
holding period. The second method is more elaborate and takes
advantage of detailed
trade-specific information. The second method uses the
time-series of trades, trade
quantities, trade prices, and trade dates, and computes actual
holding periods and profits
assuming trades are allocated on a first-in-first-out basis.
3.2.1. Implied Profits
The first method I use to estimate risk-adjusted trading profits
is a direct
application of techniques in the insider trading literature
(e.g., Jagolinzer et al., 2009).
For each individual j, I net all trades in firm i on date t.12 I
then calculate abnormal
returns to each net trade as the intercept from the three factor
Fama-French (1993) model
11 Profits are denominated in returns rather than dollars, as
returns are cross-sectionally comparable and allow for the
application of standard risk adjustment techniques. The
disadvantage of using a return measure is that it does not take
into account variation in the amount invested. However, I adjust
for this in my analyses by weighting observations by the dollar
value of the trade. 12 This analysis ignores trades that are opened
and closed within the same day. Such trades account for less than
0.5% of the sample.
-
23
estimated over each of three different horizons: the 50, 100,
and 150 trading days
following the trade.
(Ri - Rf) = α + β1 (Rmkt - Rf) + β2 SMB + β3 HML + ε (3)
Ri is the daily return for firm i, Rf is the daily risk-free
rate; Rmkt is the CRSP value-
weighted market return, and SMB and HML are the size and
book-to-market factors
(Fama and French, 1993). Implied profits for purchases (sales)
are calculated as the
estimated α (-α) multiplied by the number of days used to
estimate the model (i.e., 50,
100, or 150). Computing trade-specific abnormal returns in this
manner controls for
differences in risk across trades (i.e. trade-specific factor
loadings) and controls for the
tendency of individuals to tilt their portfolios toward smaller
value-oriented firms (e.g.,
Barber and Odean, 2000). This procedure results in an estimate
of the risk-adjusted
trading profits to the net trade of individual j in firm i on
day t over the 50 trading days
(ImpProfit50), 100 trading days (ImpProfit100), and 150 trading
days (Improfit150)
following the trade.
3.2.2. Realized Profits
The second method I use to estimate trading profits, uses data
on actual trade
prices, quantities, and holding periods. As before, for each
individual j, I net all trades in
firm i on date t. If individual j places multiple trades in firm
i on date t, I calculate the
price of the net trade (POpen) as the weighted average of trade
prices on that day. For
example, on day t if individual j buys 1 share at $1 and then
buys 2 shares at $2.50, the
net trade will be 3 (1+2) shares at $2 (1/3 x $1 + 2/3 x $2.50).
I then calculate the first
date in which the position taken by individual j on date t is
fully offset, using first-in-
-
24
first-out (FIFO) to match trades and compute realized holding
periods. Any trades not
fully offset over the sample period (e.g., an investor buys 1
share but does not sell it prior
to the end of the sample) are closed out on the last day of the
sample (i.e. profits are
measured over the sample period). I estimate risk-adjusted
trading profits, net of
transaction costs (ProfitNeti,j,t), according to the
formula:
ProfitNet IP
PP
P
PPSZBM
t
SZBM
t
SZBM
T
Open
t
Open
t
Closed
T
−−
−= – Commission (4)
t is the day of the trade, or the day the trade is “opened.” T
is the day the trade is fully
offset, or “closed”, computed according to FIFO. POpen is the
price of the trade on date t
defined above, PClosed is the price at which the trade is
closed. If the trade on date t is
closed out over a number of subsequent trades, PClosed is the
weighted average price of the
subsequent trades. For example, if the investor buys 100 shares
at $1 on date t, sells 50
shares at $1.10 on date t+1, and sells 50 shares at $1.20 on
date t+2, the holding period is
two days (T – t = 2) and the closing price is $1.15 (50/100 x
$1.1 + 50/100 x $1.20).
Computing closing price in this way, accounts for any partial
“cashing out”. PSZBM is the
price of the respective (5x5) size and book-to-market portfolio
on day t based on the size
and book-to-market ratio of the firm measured as of the prior
quarter-end. I is an
indicator variables equal –1 if the trade is a sale and 1 if it
is a purchase. Commission is
the total amount paid to the broker in trading commission
divided by the dollar value of
the trade (i.e. measured in percent).
In addition to ProfitNet, I also construct an alternative
measure of realized trading
profits, Profit, which excludes trading commission paid to the
broker (Commission).
-
25
Because realized holding periods vary by trade, and can be as
short as a day, estimating
risk-adjusted trading profits over the holding period using the
Fama and French (1993)
model, as in equation (3), is not practicable (i.e. estimation
of equation (3) requires at
minimum four observations). Instead, equation (4) measures
trading profits relative to an
equivalent investment in the respective (5x5) size and
book-to-market portfolio.
Computing trade-specific abnormal returns in this manner
controls for differences in
holding periods and controls for the tendency of individuals to
tilt their portfolios toward
smaller value-oriented stocks.
Following prior literature (e.g., Barber and Odean, 2000, 2001)
I decompose net
trading profits (ProfitNet) into three components: the amount
paid in trading commission
(Commission), the amount paid in bid-ask spread (Sprd), and
gross trading profits
(ProfitGross). Commission measures the trading commission paid
to the broker as a
percent of the dollar value of the trade and is defined as
above. Sprd measures the
amount of bid-ask spread paid by investor as a percent of the
dollar value of the trade and
is defined as in Barber and Odean (2000, 2001) as
Sprd IP
POpen
t
t
−= 1 (5)
where Pt is the closing price on date t, POpen is the price of
the trade on date t, and I is an
indicator variables equal –1 if the trade is a sale and 1 if it
is a purchase.13 Sprd includes
any intraday return of the trade. If the closing price is the
equilibrium price, then Sprd
measures the bid-ask spread paid by the individual without
error. For example, if the
13 The literature on large block trades uses a similar measure,
replacing trade price with price immediate prior to the trade
(e.g., Holthausen, Leftwich, and Mayers, 1987; LaPlante and
Muscarella, 1997).
-
26
equilibrium price is $4.90 with a posted ask of $5.00 and the
investor buys at the
prevailing ask, the investor effectively paid 2% (1 –
$4.90/$5.00) in spread. Note that if
the investor subsequently sells at $4.90 he would realize a
profit of –2%. However, in this
case the loss is purely the result of the prevailing bid-ask
spread at the time of the trade.
ProfitGross measures the gross trading profit and excludes the
amount paid in
commission to the broker and the amount paid in bid-ask spread.
Gross trading profits are
computed as in equation (4) using the closing price on the day
of the trade, rather than the
actual trade price (i.e., substituting Pt for POpen).
3.3. SAMPLE AND DESCRIPTIVE STATISTICS
3.3.1. Retail Broker Database
This study uses a database of individual investor trades and
demographic data
provided by a popular retail broker. The database covers the
daily trades of 158,006
accounts from January 1991 through November 1996, inclusive. Of
these, 126,460
accounts placed trades over the period, 101,581 placed trades in
U.S. common stock
(SHRCD 10, 11, or 12 on CRSP), and the remaining 24,879 traded
exclusively in mutual
funds, close-end funds, bonds, or other securities. The accounts
in this database were
randomly selected from the broker’s total client-base of 1.25
million households, which
represents approximately 4% of the population of individual
investors over the period.14
The total value of accounts in this database is approximately
$8.83 billion, and that the
14 The retail broker does not provide equity research services.
See Barber and Odean (2000) and Kumar and Lee (2006) for more
details on this database.
-
27
average account value is just over $55,000. In total, French
(2008) reports that U.S
households directly held 27.2% of all public equity at the end
of 1996.15
For each trade the sample contains the account number, date,
trade price, quantity
of shares traded, trade direction (i.e. buy or sell), and
trading commission paid to the
broker. Prior work on individual investors uses trade size
observed on the TAQ tapes to
infer the trades of individuals, deeming “small trades” (
-
28
Ramadorai, and Schwartz, 2007).18 Third, this study documents a
pronounced increase in
individual investor trade size around earnings announcements. In
this regard, using a
simple cutoff rule (e.g.,
-
29
3.3.2. Primary Sample and Descriptive Statistics
I construct my primary sample using the retail broker database,
CRSP, and
Compustat Quarterly. To be included in the sample, a firm must
appear on the retail
broker database, Compustat Quarterly, and have common stock on
CRSP (SHRCD equal
to 10, 11, or 12). Additionally, in each quarter, I require net
income, an earnings
announcement date on Compustat within three months of the
quarter-end, and market
value and book value of equity at the end of the prior quarter.
Finally, I require positive
individual investor trading volume on at least one day over the
quarter, and trading
volume and returns on CRSP over the thirty trading days prior to
the earnings
announcement. The resulting sample contains 63,250 firm-quarters
and 1,224,218 daily
trades for 91,066 accounts.19
Panel A of Table 1 reports descriptive statistics for firms in
the sample.
Consistent with Barber and Odean (2000), Panel A shows that
individual investors tend
to transact in small stocks (median market value of $139
million) and value stocks (mean
and median book-to-market ratio of 0.5 and 0.47). Over the
sixty-one day period
centered on the earnings announcement, trading volume is on
average 31% of shares
outstanding, and trading volume for individuals in the sample is
on average 0.06% of
shares outstanding. On average 19.36 individual accounts trade
in the firm’s shares over
the period, and of those 7.37 accounts did not previously hold
the firm’s shares.
Consistent with returns leading earnings, over the sixty trading
days prior to the earnings
announcement the average (unsigned) abnormal return is 1.28%
(14.50%). With regard
19 Measures of trade profitability are computed prior to
imposing these sample requirements. In this manner, sample
requirement do not affect the matching of trades using FIFO.
-
30
to earnings announcement, the average (median) announcement
contains good (bad)
news. The mean (median) announcement period return is 0.77%
(-0.19%). Consistent
with earnings containing significant information, the average
(median) unsigned
announcement period return is 8.65% (5.66%).
Panel B of Table 1 reports descriptive statistics for the trades
in the sample.
Similar to Barber and Odean (2000), most of the trades in the
sample are buys (54%), the
average (median) trade size is $12,471 ($5,375), and the average
holding period is 180
(91) trading days.20 Panel C of Table 1 reports descriptive
statistics on the trading profits
for each account in the sample. Similar to Barber and Odean
(2000), the average account
pays 1.87% in commission to the broker and 0.51% in bid/ask
spread.21 To put these
costs in context, French (2008) suggests the cost of investing
in an active managed
mutual fund over the period is 1.46% and the cost of investing
in a passive index fund
over the period is 0.15%. This suggests individuals incur a
significant cost to trading,
and that this cost even exceeds that of investing in an actively
managed mutual fund.
Consistent with prior work, and the average individual investor
being uninformed,
Panel C shows that individual investors consistently lose money.
After including
transaction costs (trading commission and bid/ask spread), the
average (median) account
earns –6.91% (–1.55%) in trading profits After excluding trading
commission, the
average (median) account earns –5.04% (–0.10%), and after
excluding trading
commission and the amount paid in bid/ask spread, the average
(median) account earns –
20 Barber and Odean (2000) report 54.94% of trade in their
sample are buys and the average trade size is $12,332.46. They do
not calculate holding periods. 21 Barber and Odean (2000) report
that the average trade costs 1.50% in commission (3% round-trip)
and 0.5% in spread (1% round trip).
-
31
4.33% (0.34%). Notably, even after including transaction costs,
the 75th percentile of
trading profits exceeds 20% (75th percentile of ProfitNet is
21.03%). This suggests that,
when ranked on trading profits, the top quartile of accounts
earn very large profits. This
is consistent with prior work that finds the performance of
individual investors
significantly with characteristics of the investor (i.e.,
gender, experience, wealth, etc.),
and suggests the need to control for investor-specific effects
in subsequent analysis.
4. RESEARCH DESIGN
4.1. INDIVIDUAL INVESTOR MARKET PARTICIPATION
4.1.1. Classic Event Study
To test whether earnings announcements are associated with an
increase in
individual investor market participation, I regress each measure
of abnormal market
participation on three announcement period indicator variables.
Pooling across all firms
(i) and the sixty-one trading days centered on the earnings
announcement (i.e., ∈t [–
30,+30]), I estimate the regression
Xi,t = θ0+ θ1 ANNCi,t + θ2 POSTANNCi,t + θ3 PREANNCi,t + ηi,t
(6)
where X is one of the three measures of abnormal market
participation (either AbIV, AbII,
or AbNII), ANNC is an indicator variable equal one on the day of
and the day following
the announcement (i.e. ∈t [0,+1]) and zero otherwise, POSTANNC
is an indicator
variable equal one for ∈t [+2,+5] and zero otherwise, and
PREANNC is an indicator
-
32
variable equal one for ∈t [–1,–5] and zero otherwise.22 If
earnings announcements
inform the trading decisions of individual investors I expect θ1
and θ2 are positive. Since
each of the measures of market participation is standardized by
the mean of the non-
announcement period, θ0 equals zero by construction. While
equation (3) may seem
parsimonious, recall that the measures of abnormal market
participation filter out market-
wide movements, and include firm-quarter fixed effects to filter
out the effects of cross-
sectional determinants of market participation that do not vary
within a firm-quarter (e.g.,
book-to-market).
4.1.2. Relative In formativeness of Earnings
While the classic event-study analysis is used throughout the
trading volume
literature, Ball and Shivakumar (2008, BS) argue that it can
provide a “misleading
impression” about the relative importance of earnings for the
overall informational
environment of the firm. While BS examine the relative
importance of earnings for the
overall equity market, I next examine the relative importance of
earnings for individual
investors. To do so, I adapt the research design of BS to my
setting. In particular, rather
than focus on equity returns and total trading volume, as BS do,
I adapt their tests to
individual investor trading volume.
22 While Compustat reports the date of the earnings
announcement, it does not report whether the announcement occurs
after the market closed, in which case the next trading day after
the announcement would be “day zero”. Prior work (e.g., Patell and
Wolfson, 1982; Francis Pagach and Stephan, 1992; Berkman and
Truong, 2009) suggests a significant number of earnings
announcements occur after the market has closed. Notably, Berkman
and Truong (2009) show that failure to correct for such
announcements can lead to significant downward bias in calculating
announcement day volume and returns. To overcome this bias, they
suggest defining the announcement day (the ANNC indicator) to
include both the day of and the day after the announcement.
Inferences are unchanged if ANNC is defined relative to the day of
the announcement. In subsequent analyses I calculate market
participation and trading profits separately for each event
day.
-
33
Fraction of Volume. First, I measure the relative importance of
earnings for individual
investors by calculating individual trading volume on the
earnings announcement as a
fraction of total individual trading volume over the sixty-one
day period centered on the
earnings announcement. I then test whether the fraction of
volume on the announcement
day is significantly different from what one would expect on a
non-announcement day. I
use a simple Monte Carlo simulation to provide an estimate of
the appropriate null.23 In
particular, for each firm-quarter I randomly draw a day outside
the five- day
announcement window and compute the fraction of volume over the
sixty-one day period
that occurs on that day. I repeat this procedure 1,000 times and
use the mean value as the
null hypothesis. I then test whether the fraction of volume
occurring on the
announcement day is significantly different from the null using
the empirical distribution
of values.
Explanatory Power. I also examine the fraction of the variation
in volume explained by
announcement period volume. In particular I regress total volume
over the sixty-one day
period on volume on the announcement day (scaling both variables
by average shares
outstanding). The adjusted-R2 from this regression measures the
fraction of volume over
the period explained by volume on the announcement day.24 Again,
I use a simulation to
provide an estimate of the appropriate null. In particular, I
re-estimate this regression but
using volume on a randomly selected day outside five-day
announcement window as the
independent variable. I repeat this procedure 1,000 times to get
the empirical distribution
23 BS test whether the fraction of trading volume on the
announcement day is significantly different from what one would
expect if trading volume is uniformly distributed over all days (in
this setting 1/61). 24 The adjusted-R2 from this regression measure
the extent to which volume on the announcement day explains volume
over sixty-one day period. The higher the R2, the higher the
relative importance of earnings for the overall information
environment of the firm.
-
34
of R2 under the null hypothesis that the announcement day has
similar explanatory power
to a non-announcement day, and use the mean value as the null
hypothesis. I then test
whether the explanatory power of the earnings announcement is
significantly different
from the null using the empirical distribution of values.
4.1.3. Do Earnings Inform Individual Investors Incremental to
the Information
in Prices?
I use two complementary research designs to test whether
earnings
announcements inform individual investors incremental to the
information in prices.
These two research designs test whether earnings inform
individual investors holding
constant the information in past and contemporaneous returns. In
other words, whether
increased individual investor market participation around
earnings announcements is
simply the result of investors trading on extreme returns.
In the first design, I compute buy-and-hold abnormal returns
relative to the
respective (5x5) size and book-to-market portfolio over the
sixty trading days prior to the
announcement (PastRet) and over the eleven trading days centered
on the announcement
(AnncRet).25 I then sort each earnings announcement into
quintiles based on the
respective unsigned abnormal return (AbsPreRet and AbsAnncRet
respectively).26 For
each sample partition, I re-estimate equation (6). If individual
investors trade on earnings
25 Stocks are matched to each of the twenty five (5x5) Fama and
French (1993) book-to-market portfolios based on market value and
book-to-market value at the end of the prior quarter. 26
Buy-and-hold abnormal returns capture the net revision in the
market’s prior about the value of the firm over the period of
interest. I use unsigned abnormal returns because my predictions
pertain to the total amount of information incorporated into
prices, not whether that information was good or bad news. For
example, the attention effect documented by Barber and Odean (2008)
is in regards to extreme returns, regardless of sign.
-
35
information incremental to the information in prices, I predict
θ1 and θ2 are positive in
every sample partition.
In the second design, I employ a within-firm matched pair
design. In particular, I
calculate the equivalent of AbsPastRet and AbsAnncRet for every
non-announcement
date. I then match each announcement day (treatment sample) with
a corresponding non-
announcement day, excluding the five trading days before and
after the announcement,
for the same firm based on values of AbsPastRet and AbsAnncRet
(control sample).27
This procedure yields a one-to-one, within-firm match between
announcement days and
non-announcement days based on the magnitude of past and
contemporaneous returns. I
then test for a difference in abnormal market participation
between the treatment and the
control samples. If earnings announcements inform individual
investors incremental to
the information in price (i.e. incremental to the extreme
returns around earnings
announcements), I expect earning announcements are associated
with higher levels of
individual investor equity market participation even when
compared to days with a
similar amount of return news.
4.2. INDIVIDUAL INVESTOR TRADING PROFITS
4.2.1. Changes in Trading Profits Around Earnings
Announcements?
To test for a systematic change in individual investors’ trading
profits around
earnings announcements, I regress estimated trading profits on
three announcement
period indicator variables. In particular, I estimate the
regression
27 The within-firm matched pair design is implemented by
selecting the non-announcement day that minimizing the Euclidean
distance between the announcement day and all other
non-announcement days within a given firm. Using a within-firm
matched pair design control for the effect of any firm-specific
determinants of abnormal volume.
-
36
Xi,j,t = θ0+ θ1 ANNCi,t + θ2 POSTANNCi,t + θ3 PREANNCi,t +
ηi,j,t (7)
where X is a measure of the risk-adjusted trading profit to the
trade of individual j in firm
i on day t, and ANNC, POSTANNC, and PREANNC are as previously
defined. Similar to
Barber and Odean (2000, 2001), I estimate equation (7) using
weighted-least-squares,
where each trade is weighted by its size (in dollars).28
Additionally, to control for cross-
sectional variation in trading profits attributable to
unobserved individual investor
characteristics (i.e. skill, risk aversion, etc.), I also
estimate equation (6) including
investor-specific fixed effects. For example, if investor j’s
non-announcement trades
earn below (above) average returns and investor j also trades
more around earnings
announcements than other individuals, then we might expect
trades around earnings
announcements to earn below (above) average returns. If this is
the case, then after
including investor-specific fixed effects, the coefficient on
the announcement period
indicators will be insignificant. In the investor-specific fixed
effect model, the
announcement period indicators test whether the profitability of
announcement period
trades is different from the profitably of non-announcement
trades after adjusting for the
expected profitability of the specific investor who placed the
trade.
I estimate two versions of equation (7) and the
investor-specific fixed effect
model. In the first version, X is one of three measures of
implied trading profits (i.e.
ImpProfit50, ImpProfit100, and ImpProfit150). In the second
version, X is one of three
measures of realized trading profits or transaction costs: net
profits including commission
28 Because the analysis is conducted at the individual-level,
and profits are denominated in returns, it is important to control
for differences in the dollar value of the trade. For example, if a
$500 trade earns a 10% return and a $2000 trade earns a 1% return,
then the net (or total) return is 2.8% (1/5 x 10% + 4/5 x 1%). The
announcement period indicators represent the difference in net
return between announcement and non-announcement period trade.
-
37
and bid/ask spread (ProfitNet), profits excluding commission
(Profit), gross profits
excluding commission and the bid/ask spread (ProfitGross), and
the amount of bid/ask
spread paid by the individual (Sprd). If disclosure “benefits
the uninformed at the
expense of the informed” via a reduction in private information,
then I expect
individuals’ announcement period trades are more profitable than
their non-
announcement trades, and thus θ1 and θ2 are positive when the
dependent variable
measures trading profits (e.g., ProfitNet) and negative when the
dependent variable is the
amount of bid/ask spread paid by the individual (Sprd). In
contrast, if earnings
announcements are associated with an increase in private
information and/or individuals
inefficiently process earnings information, then I expect
individual investors’
announcement period trades are less profitable that their
non-announcement trades, and
thus θ1 and θ2 are negative (positive) when the dependent
variable measures trading
profits (amount of bid/ask spread paid by the individual).
Moreover, if earnings
announcements are associated with changes in trade profitability
simply because of a
change in transaction costs (i.e. a change in the bid/ask
spread), then I expect: (1) the
amount of bid/ask spread paid by individual investors, Sprd,
changes around earnings
announcements, and (2) trading profits excluding such
transaction costs, ProfitGross, do
not change around earnings announcements.
4.2.2. Trade Profitability and the Information Content of
Earnings?
If individual investors’ announcement period trades are less
profitable than their
non-announcement trades, and the decline in trade profitability
is incremental to changes
in transaction costs, it raises the possibility that increased
losses could be the result of
-
38
inefficient information processing (e.g., Maines and Hand, 1996;
Fischer and Verrecchia,
1999). To investigate whether inefficient information processing
contributes to changes
in individual investors’ trading profits around earnings
announcements, I sort earnings
announcements into quintiles based on the amount of information
priced around the
earnings announcement. I use the absolute value of returns in
excess of the respective
(5x5) size and book-to-market portfolio over the eleven day
window centered on the
earnings announcement (AbsAnncRet) to measure the amount of
information impounded
in prices over the announcement period.29 I then re-estimate
equation (7) for each
quintile. If increased losses around earnings announcements are
the result of inefficient
information processing, then the larger the amount of
information, the greater should be
the loss. Moreover, if individual’s non-announcement trades are
also correlated with the
information in earnings, then the greater the announcement
return, lower the profitability
of non-announcement trades.
4.2.3. Who Loses Around Earnings Announcements?
An alternative approach to testing whether increased losses
around earnings
announcements are the result of inefficient information
processing, is to investigate
whether these losses vary in a predicted manner with the
characteristics of the individual
investor. Prior work suggests that wealthy individuals
outperform the market (e.g.,
Yitzhaki, 1987), that experienced investors outperform
inexperienced investors (e.g.,
29 By “the amount of information” I mean the extent to which the
market revises its priors. Using announcement period returns to
measure the information in earnings allows for heterogeneity in
model of expected earnings and heterogeneity in the quality of
reported earnings (e.g., Ecker et al., 2006, Ball and Shivakumar,
2008). For example, despite a large forecast error, reported
earnings might be of such low quality that the earnings surprise
actually contains very little information. Using the announcement
period return controls for this possibility, and is consistent with
prior research that shows announcement period returns are a
function of both the magni