Essays in Capital Markets

advertisement
Essays in Capital Markets
by
vesley S. Chan
B.A. Economics
Princeton University, 1995
SUBMITTED TO THE SLOAN SCHOOL OF MANAGEMENT
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
AT THE
MASSACHUSSETTS INSTITUTE OF TECHNOLOGY
JUNE 2002
©2002 Wesley S. Chan. All rights reserved.
The author hereby grants to MIT permission to reproduce and distribute publicly
paper and electronic copies of this thesis document in whole or in part.
Signature of Author:
V
A
.d9'
yT
'(
'I -
X,
aI
Sloan School of Management
June 1, 2002
Certified by:
e-
Accepted by:
S. P. Kothari
Professor of Accounting
Thesis Supervisor
~ ~~~Dvn
~ I ~ ~ ~INSTITUnTEd
MASSACHUSETTS~
is
e
w
MASSACHUSETTS
INSTITUTE
OFTECHNOLOGY
JUN 2
2002
LIBRARIES
,,-4 !~~~~~C 44,, -TX' 4:,,.-
Professor of Management Science
Chairman, Committee for Graduate Students
ARCHIVES
2
Essays in Capital Markets
by
Wesley S. Chan
Submitted to the Sloan School of Management on June 1. 2002
in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy
ABSTRACT
This thesis consists of three chapters, each about a separate aspect of how investors respond to
information in equity markets.
The first chapter concerns news and stock returns. Using a comprehensive database of headlines
about individual companies, I examine monthly returns following public news. I compare them to
stocks with similar returns, but no identifiable public news. There is a difference between the two
sets. I find strong drift after bad news. Investors seem to react slowly to this information. I also
find reversal after extreme price movements unaccompanied by public news. The separate patterns
appear even after adjustments for risk exposure and other effects. They are, however, mainly seen
in smaller, more illiquid stocks. These findings support some integrated theories of investor overand underreaction.
The second chapter is joint work with Richard Frankel and S. P. Kothari. Models based on psychology can explain momentum and reversal in stock returns, but may be overfitted to data. We
examine a typical basis for these models, representativeness, in which individuals predict the future based on how closely past outcomes fit certain categories. We use accounting performance to
mimic possible investor-defined categories for firm performance. We test the idea that investors
predictably bias their expectations about future operations by using these categories. We find little
evidence that the sequence or trend of past accounting performance is related to future returns,
and is therefore unlikely to bias investor expectations.
The third chapter concerns how informational advantage differs between institutional investors.
Slow information diffusion can cause return momentum. Institutions are thought to be more informed than individuals, and should eliminate return predictability. However, higher institutional
ownership is associated with more momentum. Therefore, institutions either herd on returns or
can have information before individuals. I find evidence of the latter. However, the effects are
economically small, suggesting that aggregate data obscures differences between institutions. I
divide institutions by trading aggressiveness. Aggressive institutions are more responsive to recent
returns, and a strategy mimicking their trades generates even better performance. This confirms
that some investors are more informed than others, but do not eliminate return predictability.
Thesis Supervisor: S. P. Kothari
Title: Professor of Accounting
Other Members of Dissertation Committee:
Andrew Lo, Professor of Finance
Jonathan Lewellen, Professor of Finance
Sendhil Mullainathan, Professor of Economics
4
ACKNOWLEDGEMENTS
This thesis has an eclectic parentage. Its genesis was many conversations I had with Sendhil
Mullainathan. His doctoral class on the economics of uncertainty opened a fascinating new world
of behavioral economics to me. From watching him work, I learned to push the boundaries of orthudoxy. His unceasingly generous comments and encouragement gave me energy to move forward
with all sorts of projects when I lacked motivation.
Three other advisors were "present at the creation" of this thesis. S. P. Kothari guided me through
many statistical tests and along the way trained me to be careful and thorough in handling data.
From him I learned to be skeptical about results, but open-minded about possibilities. I relied very
much on Jonathan Lewellen's time and attention. He was supportive of my approach even when I
doubted it myself, and his detailed comments, on substance and presentation, made my work much
better. There has been no better guide and friend from day to day through the agony and ecstasy
of writing this thesis. Andrew Lo has always believed that I had potential even in the earliest days
of my time at MIT. His enthusiasm is contagious, and I credit my curiosity and desire to "think
big" about the field to him.
I would like to thank Ken French for giving me many excellent comments and criticism. Although
he left MIT before I had the chance to finish this research, his strong influence will be clear to any
reader. Kent Daniel also gave me many suggestions, and his excitement over the topic spurred me
on towards further progress. Jay Ritter was a generous discussant for the first chapter, and John
Liew and Tim Crack also provided a fresh perspective by sharing their thoughts with me. I have
been highly influenced by my fellow students, in particular Joon Chae, Li Jin, Stefan Nagel, and
Oguzhan Ozbas.
Finally, my parents and in-laws have given me much love and support during this journey. Many
friends from Park Street Church also brightened my days. But the person to whom I owe the most,
and with whom I share all the fruits of my labors, is my loving wife, Theresa. She inspired me to
do my best, and in doing so, to give glory to God.
Contents
1 Stock Price Reaction to News and No-News
1.1 Introduction ............
1.2
8
. o
Literature review.
...
1.3.3
Data description.
1.4 Results.
Raw returns .
1.4.2
Size and book-to-market
. ...............
.... ... .......
...
.
8
11
........ . 13
........ . 15
..... ...........
....
16
........... ..... 20
....
....................
1.4.1
. . . . . . . . .. .... . .o
.... ....... ......
......13
1.3.1 Portfolio formation.
Test procedure.
.
.... ... .......
1.3 Methodology .................
1.3.2
o. o
tched
portfolio mELtched
.......... ...... 20
....
returns ...............
. ...............
1.5 Other adjustments
1.5.1
Buy and hold average abnormal returns (BHAARs)
1.5.2
Ranking on event month abnormal returns
1.5.3
Weighting stocks by frequency of news within the event month . .
5
..............
...................
22
24
....
24
25
25
CONTENTS
6
1.5.4
Liquidity: the effect of low priced stocks ......
1.5.5
A more restrictive definition of news .........
..
...
26
28
1.5.6 The effect of size ....................
29
1.5.7
Subperiod analysis.
30
1.5.8
The relationship between trading volume and news .
31
1.5.9
Risk changes caused by news .............
33
1.5.10 When does drift occur? ................
1.6 Discussion of results and implications
35
............
38
1.7 Conclusion ...........................
Chapter 1: Tables and Figures
41
. ..
. ..
. . . . . ..... . . . . . . . . . . . ..
. . ..
2 Trends and Sequences in Financial Performance
43
56
2.1 Introduction ......................
. . . . . . . . . . . . . . . . . . .
56
2.2
Background and Research Design Issues ......
....................
60
2.3
Data and Tests ....................
....................
62
2.3.1
Testing Procedure
2.3.2
Data Summary ................
2.4 Results
............
..............
....
....
...
...
62.....
.... .. ..... .. . 67
.....
.........
...
.....
..
..
...
. . . 69......
.... . ..... 69
.....
2.4.1
Results for Short-Horizon Momentum . . .
2.4.2
Results for Long-window.R.eversal
2.4.3
Results for Consistency of Performance and Subsequent Returns .......
......... .. .. 71
......
71
.
CONTENTS
7
2.5 Conclusion
..............
77
Chapter 2: Tables and Figures
......
79
3 Highly, Somewhat, and Un-Informed Investors
3.1 Introduction
................
.........................
3.2 Literature review ................
3.3 Data and summary statistics
100
...................
3.4.1
102
...............................
3.4 Aggregate institutional ownership and return predictability
104
..............
How institutions change their holdings ...................
.. 112
3.5.1
What motivates the transactions of different institutions?
3.5.2
Are more aggressive institutions managers trading on information? ......
Conclusion .........................................
Chapter 3: Tables and Figures
Bibliography
.................................
105
1...
109
3.5 Institutions split by measures of aggressiveness ...................
3.6
100
..........
. 118
120
122
124
141
Chapter 1
Stock Price Reaction to News and No-News:
Drift and Reversal After Headlines
1.1
Introduction
There is a large amount of evidence that stock prices are predictable. In the last decade, studies
have shown that stock returns exhibit reversal at weekly and three to five year intervals, and drift
over 12-month periods.'
Some research shows that stock prices appear to drift after important
corporate events for several months. 2 This suggests that drift is driven by underreaction to information. However, there are many days when financial markets move dramatically, but without any
apparent economic news. In other words, there appeaxs to be "excess volatility" in asset prices.3
This suggests that investors overreact to unobserved stimuli. These two phenomena raise some
interesting questions. Do returns after major public news and returns after large price movements
(in the absence of public news) differ? And if so, what can this difference tell us about how investors
'See, for example, DeBondt and Thaler (1985), Jegadeesh (1990), Lo and MacKinlay (1990), and Jegadeesh and
Titman (1993).
2
Kothari and Warner (1997), Fama (1998), and Daniel, Hirshleifer, and Subrahmanyam (1998) review the literature
on returns after various corporate events. I describe specific studies below.
3
For example, Shiller (1981) concludes that stock prices are too volatile to be explained by dividend changes.
Excess volatility studies typically look at the link between news stories in he media and stock price movements.
Although I deal with longer horizons and do not look at volatility, I share the same sources as two prominent
members of the literature, Roll (1988) and Mitchell and Mulherin (1994).
8
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
9
respond to information?
Using a database of stories about companies from major news sources, I look at monthly stock
returns after two sources of stimuli. The first is public news, which is identifiable from headlines and
extreme concurrent monthly returns. The second is large price movements unaccompanied by any
identifiable news. Each month, I form portfolios of stocks by each source, and follow momentum
trading strategies. I examine if there is subsequent drift or reversal, against the alternative of no
abnormal returns.
I find that stocks with news exhibit momentum, while stocks without news do not.
In
particular, stocks that had bad public news display negative drift for up to 12 months. Less drift
is found for stocks with good news. I interpret this to mean that prices are slow to reflect bad
public news. Furthermore, stocks that had no news stories in the event month tend to reverse in
the subsequent month. The reversal is statistically significant, even after controlling for size and
book-to-market. This is consistent with investor overreaction to spurious price movements. It is
also consistent with bid-ask bounce, although I attempt to control for this. I also find that the
effects diminish, but are present, when one eliminates low priced stocks, and are stronger among
smaller, more illiquid stocks than larger ones. A possible explanation is that some investors are
slow to react to information, and transaction costs prevent arbitrageurs from eliminating the lag.
The fact that most drift occurs after low returns reinforces this view, since shorting stocks is more
expensive than buying them. I also show that most bad news drift occurs in subsequent months
without earnings announcements.
My results fit two old strains of thought among investment practitioners, which have gained
an academic following. First, investors are slow to respond to valid information, causing drift.
Second, investors overreact to price shocks, causing "excess" trading volume and volatility and
leading to reversal. The results are also consistent with a richer set of theories (detailed below)
that try to explain short-run underreaction and long-run overreaction in terms of investor behavior.
The goal of this chapter is to deepen our understanding of how information flows drive
anomalies in three ways. First, I sample all forms of news. Fama (1998) suspects that the abnormal
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
10
reaction literature focuses only on events that show interesting results. Other events that are similar
but have no unusual patterns are unreported. My dataset is free of selection bias. I am able to see
if underreaction or overreaction remains a feature of the data by looking at a wider class of events
than has been previously examined. 4
Second, I distinguish between return patterns after news events and after price shocks that do
not appear to be news motivated. This adds to our understanding of momentum strategy payoffs.
These have not been conditioned on the incidence of news in typical studies, yet are thought to
arise because of different investor responses to public and private signals. Specifically, three major
theories seek to explain momentum and reversal. Daniel, Hirshleifer, and Subrahmanyam (1998)
(hereafter, "DHS") use overconfidence and biased self-attribution to model investor behavior. The
result is that investors hold too strongly to their own information, and discount public signals.
Barberis, Shleifer, and Vishny (1998) ("BSV") rely on conservatism and the representativeness
heuristic. They hypothesize that investors change sentiment about future company earnings based
on the past stream of realizations. Hong and Stein (1999) ("HS") present a model not tied to
specific psychological biases, with two classes of traders. One group ignores the news, but reacts
to prices. The result is initial underreaction and subsequent overreaction. Naturally, all three
theories generate momentum and reversal, but they differ in some ways. DHS state that there will
be underreaction to public information and overreaction to private information. BSV state that
investors will over- or underreact to news depending on the stream of past news. HS state that
investors will underreact to news and overreact to pure (non-information based) price movements.
Since it is difficult to find price movements that have no component of private signals ex ante, the
assumptions of DHS and HS are hard to separate empirically. I test the assumption of differential
responses to information by separating stocks by news incidence using a headline database. While
there are some differences in timing, the results are generally consistent with the DHS idea that
investors ignore the balance of the headlines (i.e., they pay attention only to news that supports
4
One study that takes a similar approach in a different direction is Pritamani and Singal (2001). They collect
daily news stories from the Wall Street Journal and Dow Jones News Wire for a subset of stocks in 1990-1992 that
had extreme returns, and find both positive and negative abnormal return drift for up to 20 days after a news story.
Their results are not directly comparable to mine since they use strict filters for trading volume, volatility, size, and
price that results in a subset of about 1% of the NYSE/AMEX universe.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AiNDuNO-NEWSv
1
their prior) while they overreact to private signals embedded in pure price shocks. However, the
results are even more supportive of the HS idea that some groups of investors are slow to react
to news, while others are feedback traders. This helps us know what sort of information causes
investors to change their expectations, and improves our understanding of their behavior.
Third, I examine when post-news drift occurs. In asset markets, arbitrage is a powerful
force against non-risk-related predictability. However, in some cases noise trader risk or frictions
can limit arbitrage (see, for example, Shleifer and Vishny (1997)). Recent research suggests that
various informational or transactional frictions can have a major effect of asset prices. By looking
at whether or not the drift occurs when more information is revealed, I provide indirect evidence
on frictions. Since most drift happens in the absence of later news, I conclude that frictions slow
the diffusion of information.
The chapter proceeds as follows. Section 2 of the chapter outlines previous research into
investor reactions, reversal, and drift. Section 3 describes my dataset and testing methodology. I
present my results in Section 4 and some extensions in Section 5. Section 6 discusses what my
results say about different theories of investor behavior, and how they relate to other findings
concerning the effect of information on returns. Finally, Section 7 concludes.
1.2
Literature review
Despite forty years of research by financial economists, the debate continues over how fast information is incorporated into prices. In this section, I describe evidence of predictability in returns.
Most of the research on stock returns after specific news items supports the idea of underreaction, which is defined as average post-event abnormal returns of the same sign as event date
returns (abnormal or raw). The main examples include signaling events5 and scheduled news re5
Dividend initiations and omissions are covered by Michaely, Thaler, and Womack (1995). Stock splits could also
fall in this category, examined recently by Ikenberry and Ramnath (2002), with similar conclusions.
CIIAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
12
leases.6 Investors also seem to be slow to react to capital structure changes7 and ignore the personal
investments of managers themselves.8
Important evidence that contradicts the view that investors underreact include results for
acquiring firms in mergers in Agrawal, Jaffe, and Mandelker (1992) and proxy fights in Ikenberry and
Lakonishok (1993), apparent reversal for new exchange listings in Dharan and Ikenberry (1995), and
a host of different return patterns for IPOs depending on the horizon in Ritter (1991). Barber and
Lyon (1997) and Kothari and Warner (1997) question the conclusions of event studies by explaining
ways in which some statistical tests used in the above research lack power since the standard errors
are understated. Fama (1998) observes that the above patterns present no consensus on investor
reactions, and some disappear entirely after accounting for size and book-to-market effects.9
Some returns can be predicted without public news. Jegadeesh and Titman (1993) find
multi-month momentum and DcBondt and Thaler (1985) find multi-year reversal. The success of
technical momentum strategies, in particular, is very puzzling from an efficient markets perspective.
Therefore, such strategies have been strongly linked to boundedly rational investor behavior by
some researchers. Momentum is robust across subperiods and appears in other markets.'
is also distinct from post-earnings drift and
reversal.1 1
It
Hong, Lim, and Stein (2000) find that
momentum is strongest in stocks that have no analyst coverage. They interpret this to mean that
research analysts play an important role in disseminating information.
Finally, there is evidence that investors overreact to price movements and trade more than
they should. It appears that the act of trading increases volatility. French and Roll (1986) find
6
Bernard and Thomas (1990) and others show drift after earnings surprises for up to 12 months after the initial
surprise. Michaely and Womack (1999) find a lag in response to changes in analyst recommendations. Womnack
(1996) documents an asymmetric lagged price response after changes in analyst recommendations, for a set of large,
liquid stocks.
7Ikenberry, Lakonishok, and Vermnaelen(1995) find drift after tender offers, and Loughranl and Ritter (1995) find
it after seasoned equity offerings. Gompers and Lerner (1998) show drift after venture capital distributions.
sSeyhun (1997) finds profits to mimicking the large trades of insiders. See also Lakonishok and Lee (2001), who
find that predictive power is mostly restricted to buys in smaller stocks.
9
Loughran and Ritter (2000) have an opposite interpretation based on the same fact.
' 0 Grundy and Martin (2001) show that, after accounting for potential risk factor exposures, momentum exists from
the 1920s to the present. Rouwenhorst (1998) shows that momentum occurs in other countries.
"LLeeand Swaminathan (2000) show momentum is linked to reversal, conditional on trading volume. They also
look at it in the context of earnings drift, as do Chan, Jegadeesh, and Lakonishok (1996).
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
13
that the variance of stock returns is larger when the market is open than when it is closed, even
when similar amounts of information are released. Cutler, Poterba, and Summers (1989) look
at the relations between extreme market-wide returns and major business stories from the New
York Times. They conclude that neither economic variables nor news stories can fully explain
extreme aggregate price movements. Roll (1988) looks at the R-squared for regressions of daily and
monthly stock returns on CAPM and APT factors and finds that much of the variance in returns
is unexplained.
12
In sum, many would describe underreaction to news as a "pervasive regularity" ,13 but others
would dispute this claim, noting that the results are inconclusive and the methodology problematic.
Furthermore, negative return autocorrelation at very short and long lags confounds the perceived
pattern of drift. Some interpret this as evidence of overreaction.
1.3
Methodology
Do drift or reversal patterns occur consistently after news? To summarize my approach, I collect
all stocks in a given month that had at least one news story. I rank all such stocks by monthly
raw returns and select the top and bottom terciles. I refer to these two sets as "news winners"
and "news losers", respectively. I then examine abnormal returns for up to 36 months after the
initial headline month. To determine whether predictable drift or reversal occurs after pure price
movements, I repeat the test above for "no-news" stocks, the.se that had no headline in a given
month. In the following sections, I describe each of these steps in more detail.
1.3.1
Portfolio formation
Each month, I separate firms that had one or more news stories from those that did not. I then
divide these news stocks by performance. Using the Center for Research in Security Prices (CRSP)
monthly data series (with delisting returns), I rank news stocks each month by raw return. To
12
Mitchell and Mulherin (1994) show that while news moves the market, the relationship is not very strong.
'3 See Barberis, Shlecifer,and Vishny (1998), abstract.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
14
be included in the ranking, the stock must have traded during the month. I pick the top and
bottom thirds as my "good news" and "bad news" groups, respectively. Terciles yield diversified
portfolios where non-news related characteristics are less important, since there are few sample
stocks in the earliest periods. On the other hand, some "bad news" stocks have positive returns
when the return breakpoints are positive. I use months for comparison with momentum studies
and to reduce microstructure problems that are present in daily or weekly data.
Each month, I also use the news return breakpoints to select a group of winner and loser
stocks from among the monthly no-news set. No-news stock returns could reflect reactions to
private signals, news not covered by my sources, or supply and demand shocks. One can also
think of the no-news portfolio as a benchmark for the news portfolio, since they have similar event
date returns. This helps us to understand stock behavior after public announcements versus pure
price movements. Each month I also sort all subset stocks by returns alone and pick the top and
bottom thirds as winners and losers, respectively. This is the "all" set. I use a different set of
return breakpoints to separate these winners and losers because I want to see how a pure 1-month
momentum strategy would do. I continue to use thirds, however, to make the all results roughly
comparable to the news and no-news returns. Again, each stock in the no-news and all groups
must trade during the formation month to be included.
The additional analysis of all and no-news stocks will also help me address some problems with
long-run event studies identified by Barber and Lyon (1997) and Kothari and Warner (1997). For
instance, most cumulative and buy-and-hold abnormal returns appear positive in random samples. 14
This causes the tests to have low power. Also, various data requirements for a sample bias the
abnormal returns. Both the news and no-news samples could suffer from these problems. However,
the difference between the two sets of returns should still tell us something definitive about how
news affects stocks, under the hypothesis that misspecification affects both samples in more or less
the same way.
141mainly test cumulative average abnormal returns (CARs), although I discuss one set of results for buy-and-hold
average abnormal returns (BHAARs). Kothari and Warner's simulation results indicate that BHAARs can be more
misleading than CARs. Cumulation bias due to bid-ask spread is mitigated in my CARs, since monthly returns
exhibit less bid-ask bounce than weekly returns. Roll (1983) describes this l.roblem.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
1.3.2
15
Test procedure
I form monthly equal-weighted portfolios of the winner and loser stocks. Portfolios can be easily
interpreted as trading strategies. I calculate overlapping returns using a standard rolling portfolio
method as in Jegadeesh and Titman (1993) and Fama (1998). As an example, suppose we want
to look at how good news affects returns over four months. At the end of each calendar month,
we calculate the abnormal return for all stocks that fell into the news winner category in the last
month. We then average the abnormal returns for the calendar month across stocks to get the
abnormal return on a portfolio of last month's news winners. For the same calendar month, we also
calculate the abnormal return on portfolios of news winners from two, three, and four months ago
and average all four resulting portfolio returns. This average tracks the calendar month performance
of a news winner strategy that holds a series of portfolios formed in the last month as well as the
previous three months. I repeat this process every calendar month to get a time-series of returns.
Following Fama (p. 295):
The time-series variation of the monthly abnormal return on this portfolio accurately
captures the effects of the correlation of returns across event stocks missed by the model
for expected returns. The mean and variance of the time series of abnormal portfolio
returns can be used to test the average monthly response of the prices of event stocks
for [four months]... following the event.
In this case, the "event" is a high return, conditioned on one or more headlines. I follow the same
steps for different horizons (one to 36 months after the event) and for other sets of stocks (winners
and losers, all, news and no-news).
Most previous research averages the returns of the component portfolios each month. This
average can be interpreted as the payoff to a strategy constructed using overlapping portfolios. For
example, in the four-month rolling portfolio strategy, the calendar month t payoff would be the
average of the time t returns on the four overlapping portfolios formed from months t - 1 to t - 4.
In this study, I present summed, instead of averaged, returns. This makes it easier to see how a
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
16
strategy performs over time. However, it makes it harder to frame as a practical trading strategy.
Throughout the analysis, if one wants to see average monthly returns as in Jegadeesh and Titman
(1993), simply divide my average cumulative returns by the post-event horizon over which they are
cumulated. The statistical significance will not change, of course.
To summarize the degree of drift or reversal, I also use a long-short strategy where past "good
news" stocks are held with positive weights, offset by short positions in "bad news" stocks. This is
repeated for all and no-news stocks.
The test statistics are simply the time-series average of calendar month returns divided by
the time-series standard error. The test statistic for cumulative abnormal returns (CARs) should
be distributed unit normal if there is no systematic abnormal performance. For good news, positive
CARs indicate post event drift (consistent with underreaction), and negative CARs indicate reversal
(consistent with overreaction); vice versa for bad news. How do I calculate CARs? Daniel and
Titman (1997) suggest that size and book-to-market characteristics are better predictors of future
returns than factor betas. Therefore, I subtract the contemporaneous returns of size and book-tomarket matched portfolios. Some stocks are lost due to my matching criteria, described more fully
in the next section.l5 Note that I make no adjustment for momentum. I want to test reactions to
news, a possible cause of momentum.
1.3.3
Data description
To examine stock price reactions to public news, I need to know when information was released.
I use the Dow Jones Interactive Publications Library of past newspapers, periodicals, and newswires. This database has abstracts and articles from many sources, going back to before 1980.
However, some sources are only available after being archived in electronic format. To get around
the problem of spotty data, I select only those publications with over 500,000 current subscribers,
151 have also regressed summed rolling portfolio excess returns on contemporaneous factors (appropriately scaled)
according to the CAPM and Fama-French 3-factor models. The constant term is cumulative alpha. The results are
very similar to those presented below. The regression method increases the profitability of the news strategy and
decreases the losses to the no-news strategy when compared with the portfolio matching approach.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
17
daily publication, and stories available over as much of the 1980-2000 period as possible.l 6 For
each company in my set, I hand-collect all dates when the stock was mentioned in the headline
or lead paragraph of an article from the sources. To reduce over counting news about the same
subject from multiple sources, I note only if there was news on a particular day, not how many
stories appeared. I do not include magazines, since it is difficult to say on which day or week they
became publicly available. Also not covered are investment newsletters, analyst reports, and other
sources not available to the broadest audience.
There are more sources in the later part of the 1980s and 1990s. As a result, I could miss a
larger fraction of news events early in my sample period. However, by far the sources with the most
complete coverage across time and stocks are the Dow Jones newswires. This source does not suffer
from gaps in coverage, and it is the best approximation of public news for traders. Furthermore, a
stock only needs a single news story over the month to be selected for the news set, which reduces
the chance that later periods (with headlines on many days in a month) dominate the set of news
events.
Since data retrieval is time consuming and labor intensive, I focus on a random subset of
approximately one-quarter of all CRSP stocks. The result is a set of over 4200 stocks, with 766 in
existence at the end of January 1980 and over 1500 at the end of December 2000. Table
shows
counts of stocks in subsets in December of each year. The all set is roughly the union of the news
and no-news sets, for both winners and losers. About half of my subset of stocks has some news
in each month. The proportion ranges from 40% at the start of the period to 60% at the end of
the period. On average less than 5% have news on more than five days in a month, although that
percentage increases through time. The increasing number of days with news is consistent with
improving media, coverage. The numerous news stocks each month also suggests that headlines do
not consist solely of previously studied corporate actions.
16The resulting list of data sources, with their coverage dates, follows: the Wall Street Journal (all editions)
from 1980-present, Associated Press Newswire from 1985, the Chicago Tribune from 1989, The Globe and .Mail (for
coverage of a few Canadian companies) from 1977, Gannett New Service from 1987, the Los Angeles Times from
1985, the New York Times from 1980, the Washington Post from 1984, USA Today from 1987, and all Dow Jones
newswires from 1979. The results are virtually unchanged, even in later periods, using only the Dow Jones newswires
and Wall Street Journal.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
Panel B presents correlations of news citations with selected firm characteristics.
18
Stocks
with headlines are larger and one might expect them to exhibit fewer asset-pricing anomalies than
no-news stocks. Cross-sectionally, the correlations of log market value on log citations per month
average 0.37 over time. News citations per month have a weak positive correlation coefficient of
0.01 with returns. The occurrence of headlines is more strongly related to turnover; the average
correlation is 0.16. I conclude that headlines do not seem to favor good news (denoted by high
returns).
Also, depending on the interpretation of turnover, liquid stocks attract more media
attention or news causes more trading.
Table 2 presents winner and loser summary statistics for December each year. Winners tend
to be larger than losers. News stocks are larger than all momentum stocks, which in turn are
larger than no-news stocks. Most selected stocks would be considered small-cap, although some
of the winners and news stocks might be classified as mid-caps. One should note that no-news
stocks might be more subject to microstructure movements since they are typically very small.
These averages conceal large variations, but are an appropriate way of viewing the portfolio since
I equal-weight observations.
News, no-news and all portfolios have similar event month (time t) returns as shown in the
last six columns of Table 2. Winner or loser portfolios are not very concentrated by industry.
I classify all portfolio stocks by the 20 industries used by Grinblatt and Moskowitz (1999), and
calculate the cross sectional Herfindahl index for each month.l7 The monthly Herfindahl averages
(not shown) are remarkably uniform across news/no-news and winner/loser categories, at about
16%. Given an average of 18 industries per portfolio each month, this implies that a single industry
should not dominate the analysis.
Table 3 shows some details of news stories for a sample midcap firm, Jacobs Engineering
(ticker JEC), for 1983-1986. Every news month is displayed. Winner, loser, or "neutral" designations within the set of news months, and the contemporaneous return, are shown in the left
columns. This table highlights some features of the data. First, many news "events" are not cor17
My Herfindahl index is 2 1 P2, where Pit is the percentage of stocks in industry i in month t. This is a measure
of the industry concentration of the portfolio each month.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
19
porate actions or pre-scheduled earnings releases. These include capital spending announcements,
blockholder sales and purchases, and new contracts. Second, there are some months when a reading
of the headline does not reveal if the news was good or bad. Since judging the text of stories is
subjective, it is wise to rely on the market reaction to filter "good" and "bad" news. Third, the
winner and loser categories are broad because I use thirds to divide firms by returns. Fourth, there
is potential mixing between news and no-news events, as some headlines may not appear to contain
any economically relevant information. This will reduce the distinctions between the two sets, but
is necessary if we are to avoid picking and choosing important
news.
How frequently do stocks have news? Three-quarters of the stocks have news on only 72
months or less. Less than two percent have news on 216 or more months from January 1980 to
December 2000. Given that most firms exist for only a few years, however, it is better to ask what
percentage of their sample existence in months do they have news. I construct a histogram (not
shown) of stocks by percent of months they had headlines out of all months they existed in the
sample. Only 14% of the total have news on 90% or more of the months that they existed from
1980-2000. Slightly under half have news about half the time they existed from 1980-2000 (30% to
70% of their sample lifespan). About 8% have news on only 10% or less of the months they existed.
Thus, most stocks have fairly frequent periods of both news and no news.
The incidence of news is not autocorrelated. A single stock can switch from being a news
winner to a news loser several times in a year. The transition probabilities of stocks in each of
the news/no-news winner and loser groups (not shown) confirm this. News losers are slightly more
likely to repeat as losers (news or no-news). News stocks have a 60% chance of having more news in
subsequent months (be it good, bad, or neutral), and no-news stocks have a 40% chance. However,
the average proportion of stocks in the four categories (news winner, news loser, no-news winner,
no-news loser) switching into another category over subsequent post-formation months is roughly
equal. Therefore any post-news patterns are likely due to reactions to single news events, not the
accumulated reaction to multiple related news items.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
1.4
20
Results
I present raw returns first, size and B/M adjusted returns second, and various results for adjusted
datasets last. In all cases, rolling portfolios are used.
1.4.1
Raw returns
Panel A of Table 4 shows cumulative returns to the long-short zero investment strategy, out to
three years after the event month. Separating stocks on news incidence causes dramatic differences
even in first month returns. While there are few statistically significant signs that the long-short
strategy is profitable for all and no-news sets, the news set returns nearly 5% in the first twelve
months. Returns are negative in the first months, especially for the no-news strategy, which loses
1.83%. This is in line with the results of Lo and MacKinlay (1990), who show positive returns to
a short-term contrarian strategy up to one month. It takes the all strategy almost half a year to
recover from the effects of the t + 1 reversal. In contrast, news stocks experience less reversal in
month t + 1, and also have more drift than all stocks for most of the following year. There are
also some large negative returns beyond the 12-month horizon for news stocks, although they are
not enough to eliminate the early drift. The difference between news, no-news, and all returns is
statistically significant in the first 12 months.' 8
Month-by-month news returns (not shown) are larger three, six, nine, and 12 months after the
event month, which suggests that post-earnings drift can be a driver of the long-short returns. My
' 8As is generally the case for all of the following subsets, long-run returns seem to exhibit reversal around the
two-year mark, so that long-short strategy gains are almost eliminated. However, after 12 months there is virtually
no difference between news and no-news monthly returns. I include long-term returns to see if short-term effects are
transitory, and I cannot rule that out. However, I am reluctant to draw further inferences from them, for several
reasons. First, there is the chance that the expected returns models I use are misspecified. Barber and Lyon (1997)
and Kothari and Warner (1997) show that this becomes more of a problem as time goes on. However, this is less of a
problem in the short term, and I generally find zero abnormal returns in most months beyond the first few. Second,
in my 19 year sample period there are only six completely non-overlapping 3-year returns, a very small sample.
Overlapping returns do not necessarily improve the quality of statistical inferences at very long horizons. Third, it is
conceptually harder to justify long-run movements in stock returns as a response to publicly available news than it is
to explain short-term movements, especially when intervening periods show no particular abnormal return pattern.
I present cumulative returns out to the third year, however, for the interested reader.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
21
news set contains about 90% of stocks in the CRSP subsample that make earnings announcements
(as recorded on IBES and Compustat) in a given month. Therefore, whatever results I find could
be largely driven by the earnings drift phenomenon. Later, I eliminate earnings announcements
from my sample and redo the analysis. As discussed below, earnings announcement returns are
important, but the news drift remains economically and statistically significant even after excluding
them.
A long-short strategy using no-news stocks loses money in the first month, and has essentially
zero profits thereafter. The pattern of returns is consistent with an interpretation of no-news shocks
as having a temporary component, due to overreaction. It is also consistent with microstructure
effects like bid-ask bounce. To examine this possibility, I wait one week after forming portfolios
before investing in the strategy. This procedure is typically used in momentum strategies to reduce
influence of short-term microstructure movements on subsequent cumulative returns. The results
in Table 4, Panel B, show that waiting a week lessens the magnitude of reversal for no-news stocks,
but does not eliminate it. In contrast, the news long-short strategy is even more profitable and has
no reversal in the first month.
Undoubtedly, the stronger pattern is that of drift after news events. First month reversal
for no-news stocks is economically and statistically less significant. Skipping a week may not
eliminate all of the microstructure effects, and one might still have doubts that the reversal is due
to overreaction. However, skipping an entire month would make it impossible for me to study
any short-term effects (although doing so strengthens my post-news drift findings). I continue to
comment on first month effects since they appear in most later adjustments. The no-news reversal
pattern is fairly robust, if not large, but is hard to separate from small stocks.
How is one-month news momentum related to the longer-horizon raw return momentum
strategies documented by Jegadeesh and Titman (1993)? One could hypothesize that multi-month
raw return momentum simply aggregates news drift and no-news reversal. The no-news stocks
in such a strategy obscure the effect of news. One question is how much the firm composition
of the standard 6-month strategy overlaps with that of a news or no-news one-month strategy.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
22
To address this, I create a six-month rolling portfolio strategy from my subset of CRSP stocks,
1980-2000. Again, I divide winners and losers by thirds. I then split winners and losers into stocks
that had news in the last month (and would therefore likely appear in a news long-short strategy)
and those that didn't (and would likely appear in a no-news strategy). One month no-news stocks
make up about 40-50% of the stocks in the 6-month strategy. However, including these stocks
substantially recudes momentum profitability. Excluding no-news stocks generates returns 40-50%
higher at six to 12 months horizons. This is consistent with the idea that standard momentum
simply reflects news drift and some extra no-news noise. 19
In subsequent tables and charts, I do not refer to the all stock strategy since its component
stocks are an even mix of news and no-news stocks. The results for the all set are similar in
magnitude and sign to those of the entire CRSP database in the same period for almost all horizons.
1.4.2
Size and book-to-market portfolio matched returns
I next describe the method for adjusting returns for size and book-to-market (B/M). I merge all
stocks in the CRSP database with book value2 0 using a method outlined by Fama and French
(1992). For June of each year t, all CRSP stocks are formed into 25 portfolios by size at the end of
June of year t and B/M at the end of December of year t - 1. I use market value from December
and accounting book value for the fiscal year ending in year t - 1 for B/M. Only stocks on the
NYSE with positive book values are used to calculate size and B/M breakpoints. The resulting
portfolios are then equal weighted, and I calculate 25 sets of monthly returns. 2 1
At the end of June every year, I pick only stocks from these 25 portfolios that match those
from my subsample of over 4,200 stocks. I lose about 20% of the original sample stocks each month,
with slightly more lost in the beginning of the period (23% in January 1980), and less in the later
19
As another sign that news and standard momentum reflect the same phenomenon, both the news and standard
momentum strategies backfire in January. In contrast, most no-news reversal occurs in January and is long lasting
(although it is still statistically significant in non-January months). A blended strategy combining all stocks regardless
of news shows January reversal, consistent with previous work.
20
Data item 60 on the Compustat tapes. I use the CRSP/Compustat merged database.
21I construct my own size and B/M portfolios to be consistent with how I measure size and B/M for individual
stocks. My portfolio returns are over 90% correlated with those from Ken French's website.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
23
dates (14% in December 2000). This is due to the merging criteria, since I require data from the
previous year as well as each June. On average, the resulting stocks are slightly larger than those
without the size and B/M requirements. Averaged through time, the result is about 17% fewer
news and 23% fewer no-news winners and losers.
I then subtract the size and B/M matched portfolio return from stock returns each month.
I cumulate and test the resulting time series of adjusted returns as before. Finally, I skip the first
week after portfolio formation before investing. As in Table 4, Panel B, this is meant to mitigate
microstructure effects.
I present size and B/M adjusted data in Table 5. In general, the results are the same.
Post-news drift is clear from the positive returns to the news long-short strategy. The 12 month
cumulated abnormal return for news winners is 1.3% (statistically significant at the 10% level) and
news losers is -2.6% (significant at the 1% level). The reversal for the no-news winner group is
statistically significant at the 5% level. However, the first mor.th returns are very small relative
to the event month run-up, at -0.2% vs. 16.7% for no-news stocks. From Panel C, one can see
that no-news losers show the same pattern of reversal followed by zero abnormal returns. They
gain back 0.9% following a 13.9% drop. News stocks, however, experience no reversal in the first
-month. The first month no-news reversal, for both winners and losers, implies that large price
swings contain an element of overreaction.
The difference between news and no-news returns is biggest for the losers. There is almost no
difference between the two winner sets, except for the first few months. Almost all later adjustments
(which tend to lessen the impact of the smallest stocks) confirm that news losers have drift, but
news winners do not. In particular, any news winner continuation is due to post earnings drift.
These facts support the view that investors primarily underreact to bad news.
In summary, the results of the size and B/M adjustment give further weight to the interpretation of underreaction to news. The evidence suggests an asymmetric response to information.
Risk changes are unlikely to explain the entire story. The CAR spreads I have found are around
4% by month 12. Abarbanell and Bernard (1992) find size adjusted CARs of 8% for a strategy of
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
24
longing positive and shorting negative earnings surprise stocks from 1976-1986, and Bernard and
Thomas (1990) find long-short CARs of between 4% and 10%. The studies use quintiles and deciles,
respectively, while I use thirds. Various horizon momentum strategies also return anywhere from
8% to 12% a year. Therefore, my results are reasonable when compared to those of other studies.
1.5 Other adjustments
In this section, I adjust the methodology and sample, to explore further the patterns I have found.
For comparison, I continue to use size and B/M adjusted returns like those in Table 5, skipping
the first week after formation before investing. The results for long-short strategies are shown in
Table 6.
1.5.1
Buy and hold average abnormal returns (BHAARs)
CARs are the sum of period-by-period average returns of all stocks in the portfolio. The trading
strategy is effectively re-balanced each month, which is economically costly. It also increases the
effects of small stocks, particularly after long periods of time. This is because each month, I equal
weight all positions, even those that shrank a lot in the previous month. Buy-and-hold average
abnormal returns (BHAARs) reflect the profits to a more feasible trading strategy and put more
emphasis on relatively larger stocks.
To get BHAARs each month, I calculate the buy-and-hold return to each stock in the portfolio, skipping a week after formation. I then subtract the buy-and-hold return over the same
horizon of a matched size and B/M portfolio. In each calendar month, I average these abnormal
returns across portfolios. Since the time-series of returns is overlapping, I calculate t-statistics
using Newey-West standard errors. Again, the results (Table 6, Panel A) are little changed. The
news long-short strategy is profitable, while the no-news long-short strategy is unprofitable. The
difference is statistically significant in all months. The no-news strategy loses less over time, but
the news strategy profits are a bit larger vs. those of Table 5. Winners and losers for both news and
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
25
no-news strategies tend to have higher returns at longer horizons. However, the difference in news
and no-news loser cumulative returns is statistically significant at the 1% level for all months, while
the difference for winners is not beyond the first three months. Also, both no-news winners and
losers reverse and earn increasing compounded negative and positive abnormal returns to the third
post-formation month, respectively, while neither news winners nor losers do. The three months it
takes for no-news reversal to end makes it more likely that this is not due to a trading shock.
1.5.2
Ranking on event month abnormal returns
Most studies measure the abnormal return around the event for each stock. The interpretation
is that the event month idiosyncratic returns reflect firm-specific information, and subsequent
abnormal returns show investor under- or overreaction to such news. I repeat the analysis above,
ranking on event month size and B/M adjusted returns instead of raw returns. The results (Table
6, Panel B) are essentially unchanged. Idiosyncratic news drift is less pronounced. Again, news
losers play a larger role in the difference between news and no-news portfolios. They return 2.4% (t-statistic -2.5) at twelve months. Stocks ranked by no-news idiosyncratic returns also show
strong reversal in the first month, while news stocks do not reverse. Both no-news winners and
losers contribute to this. The cumulative return difference between news and no-news losers is
statistically significant at the 1% level for all months, but not for winners beyond the first three
months.
1.5.3
Weighting stocks by frequency of news within the event month
The impact of an announcement can be complex and professional investment analysts and reporters
might need time to discover the full story. Therefore, stocks with news over several days should
show less drift. One way to observe the effects of multiple headlines would be to weight each stock
in a month by the number of days it had a headline when forming nws portfolios. If there is
less drift, we can conclude that investors underreact less to many headlines than to a few news
stories. An alternative view might be that investors' underreaction is proportional to the amount
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
26
of information they receive, which would mean that more headlines implies more underreaction.
I repeat the size and B/M adjusted analysis, weighting news stocks in the portfolios by the
number of days of headlines within the event month. Again, the results (Table 6, Panel C) are
largely unchanged. Long-short returns show a pattern of drift for news stocks. Almost all of the
post-news drift is from news losers, who return -4.0% by month 12 (t-statistic -3.7). News stocks
show no reversal in the first month. No-news portfolio returns are of course unchanged from Table
5. The differences between news and no-news strategy returns are large and statistically significant
at the 1% level in the first month, and beyond for losers.
It is not surprising that there is little change from previous results, since extreme return stocks
probably have more news. Weighting by number of headlines, however, does reduce the influence
of the smallest stocks, since headline incidence and size are correlated. Given these findings, it is
difficult to say that having more news makes investors less likely to underreact to information. It
still seems that there is a delay in price adjustment, regardless of the number of days of coverage
a company has.
1.5.4 Liquidity: the effect of low priced stocks
It might not be profitable to attempt to "arbitrage away" apparent underreaction, since much of
the drift seems to be driven by smaller stocks. These tend to be more illiquid, and have higher
direct transactions costs as a percentage of any position. Large transactions would probably have
a large price impact. This might explain why the drift effect seems .o persist, although not why it
arises in the first place. One way to see how liquidity affects the drift pattern is to exclude those
stocks that have high direct transactions costs. I repeat the size and B/M adjusted analysis of
Table 5, but eliminate all stocks with prices of $5 or less from my sample. My remaining sample
should consist of more liquid stocks, since price is related to ease of buying or selling.
Dropping low-priced stocks further reduces the sample. In the CRSP database for this period,
on average, 32% of observations (stocks in all months) are priced at $5 or below. The no-news set
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
27
contains more low priced stocks; 25% of my news winners and losers are low-priced, while 46% and
38% of no-news losers and winners, respectively, fall out of my size and B/M matched subsample.
The resulting portfolios of winners and losers have less extreme returns in the event month. Loser
returns average about -10% in the eent month for news and no-news portfolios. Winner returns
average about 13%. Also, my remaining sample consists of larger stocks. Loser news and no-news
portfolios average about $1.02 billion and $318 million, respectively, while winner news and no-news
portfolios average $1.16 billion and $425 million. These are much larger averages than those used
in the original analysis.2 2
The CARs of the reduced set (shown in Table 6, Panel D) are similar to those of Table 5.
News drift is still present, and no-news stock returns are closer to zero. The difference between
news and no-news strategy payoffs is smaller, but still statistically significant at all horizons. Again,
the news loser drift drives all of the news strategy profitability. News loser 12-month cumulative
returns are -3.3% (t-statistic -4.2), which is much different from the no-news loser cumulative return
of close to zero. News winners have little drift (at 12 months, they return 1.0%, t-statistic 1.2).
Moreover, no-news stocks continue to have reversal. The pattern, however, is much weaker and
confined to losers. No-news winners return 0.0% in the first month (t-statistic
0.2) and no-news
losers return 0.3% (t-statistic 2.8). The difference between news and no-news sets is economically
and statistically large only for losers. Therefore, first month no-news reversal is driven by the losers
in higher priced stocks.
Even after removing relatively illiquid stocks, news losers continue to show economically
and statistically significant negative returns for up to 12 months after formation. This drives the
profitability of the news momentum strategy. However, reversal is much weaker, and it is difficult
to tell if this is due to liquidity constraints or investor behavior. These probably reinforce one
another since individuals, who tend to be less informed, are more dominant shareholders in less
liquid small stocks. The fact that news-related patterns diminish when accounting for liquidity
22
The summary statistics shown in Table 2 are not exactly comparable to those I describe here. Table 2 shows
statistics for all of my stocks, not for the subset that can be matched with size and B/M portfolios. However, the
characteristics of the size and B/M matched group (regardless of price) are not very different from those of Table 2,
except that they are somewhat larger in size.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
28
(but do not disappear) suggests that frictions may play a role in slowing news. I explore this in
more detail below.
1.5.5
A more restrictive definition of news
I count news stocks as all firms that had one or more headlines in a month. This simple definition
does not rely on personal judgement to distinguish different forms of public information, which
helps us avoid selection bias. Moreover, the news/no-news differences yield interesting results even
without adding anything else to the analysis.
However, there are two problems with this division. First, I undoubtedly mix a lot of "nonews" stocks with news stocks, since not all headlines say something that investors would find
informative. Second, since the quantity of news increases considerably over time, the number of
no-news stocks declines. This weakens any no-news patterns. One way around these problems is
to define news stocks by additional criteria, such as abnormally high share turnover. This filter
relies on the idea that investors will trade in larger quantities than they usually do when news is
truly noteworthy. However, it has some drawbacks. First, we must believe that true public news
has been noticed by many traders or a few heavily trading parties. In doing so, however, we rely
on the trading behavior of investors to tell us what is noteworthy, when most bounded-rationality
hypotheses start from the idea that investors cannot make this distinction. Second, in adopting a
more restrictive definition of news, we may overcompensate and force important news events into
the no-news category, thus clouding no-news results.
Nevertheless, I reclassify news stocks as those firms that experienced both a headline and
abnormally high share turnover. The latter is defined as turnover in the 3 days around headlines
that falls in the top third of daily share turnover over the three months prior to the formation
month. All other firms are considered no-news stocks. Again, I use formation month returns to
differentiate between good news and bad news. This results in many more no-news stocks, and
fewer news stocks. For instance, I begin with about 200 no-news winner stocks in January 1980
and end with around 270 in 2000. There are about 100 to 200 no-news losers over the entire period.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
29
I have about 30 news stocks in the early 1980s and 350 in the late 1990s, for both winners and
losers. The new definition also substantially increases the time-series average size of the news and
no-news portfolios, to $825 million and $935 million for news losers and winners, and $248 million
and $416 million for no-news losers and winners. Industry concentration falls a bit for portfolios,
with the average Herfindahl index around 13% and the number of industries about the same.
In Table 7, I present news and no-news strategy returns (size and B/M adjusted, skipping the
first post-formation week) for the more restrictive definition of news. Not much changes compared
to Table 5. The news strategy is slightly more profitable and the no-news strategy is slightly less
unprofitable. Losers mostly drive news drift. Both no-news winners and no-news losers reverse
in the first month. Overall, the news drift and no-news reversal observed in previous sections
are robust to other definitions of news. A tighter Siiter for news tends to strengthen the news
drift results slightly, while weakening the no-news results slightly. However, the same general
patterns persist, implying that investors appear to ignore even obvious activity of other traders
that highlights important headlines.
1.5.6
The effect of size
Underreaction to news seems stronger in lower priced, more illiquid stocks. Just how small are the
stcks that exhibit these patterns? One way to find out is to divide the sample each month into
size quintiles. There are problems doing this under the broadest definition of news. If any stock
with a headline in a month is part of the news set, then there are not enough no-news stocks in
the largest quintiles for comparison. The more restrictive turnover-based news definition mitigates
this problem, although there are still a few months with few or zero no-news stocks in the largest
size groupings.
I use the news definition in Table 7 for this analysis. I divide stocks at the end of each
June based on month-end market values using NYSE breakpoints. Table 8 shows how the longshort news and no-news strategies perform for different size quintiles of stocks. Time-series average
numbers of stocks in the resulting news and no-news groups are also shown in the first columns of
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
30
each group. One can see that in the smaller quintiles, there are more no-news than news-stocks,
while the situation is reversed for larger stocks.
There are several patterns to note. First, news drift is economically and statistically significant into the third quintile. No-news reversal is also present in the two smallest quintiles. It
appears that no-news stocks begin to show drift in later months, but this is probably due to the
fact that some headline stocks without high share turnover got classified as no-news due to the
modified definition. For this reason, it is useful to look at the difference between news and no-news
strategies. Note that the third column is not the exact arithmetic difference between news and
no-news returns, since some months have zero no-news stocks. These returns clearly show that
news drift and no-news reversal are strongest in the two smallest quintiles. This supports the
interpretation drift and reversal are related to ease of trading, liquidity, attention, institutional
ownership, or other factors related to size.
1.5.7
Subperiod analysis
It is possible that any investor underreaction fell in recent years with the advent of new sources
of information. Broadening stock market investment may also have changed any reaction that was
present in the 1980s. I split my sample into two subperiods. Table 9 shows size and B/M adjusted
returns for 1980-1990 (on the left) and 1991-2000 (on the right). Again, all returns are cumulated
after skipping one week between portfolio formation and investment. Long-short, winner, and loser
strategy results are in Panels A, B, and C, respectively. Since there are many more headlines in
the 1990s, the results will be less clear since some news stocks may in fact have no news under this
definition.
One can see that the drift after news is much larger in the earlier period, although there
are still statistically significant and economically sizeable returns of over 3% at 12 months for the
news strategy in the 1991-2000 period. Abnormal returns tend to be more negative in the earlier
subperiod and more positive for the latter subperiod. In Panel B there is less difference between
news and no-news winners for either period, except for the first few months. For losers in Panel
I
_ __
I___
__
_
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
31
C, there is more of a difference between news and no-news in both decades. The difference in
cumulative abnormal returns between the two periods could be attributed to the performance of
the size and B/M factors. In fact, a 3-factor regression of strategy returns for each subperiod shows
that R-square values are smaller in the latter decade. This could mean that the 3-factor model is
less well specified in the latter period. However, the patterns of relative magnitudes and signs are
more important than the point estimates of abnormal returns. The difference between news and
no-news set returns is consistent for both periods. For both winner and losers in both decades,
the same general patterns hold. There is pronounced reversal for no-news stocks and evidence of
drift in news stocks, mostly for the losers. Although underreaction might have diminished in recent
years, it remains even in most recent years. There is also evidence of overreaction in the reversals
of the first month.
1.5.8
The relationship between trading volume and news
Lee and Swaminathan (2000) find that high trading volume stocks experience more momentum
and high volume tends to attenuate drift in winners but strengthens it in losers. They suggest a
"momentum life cycle" explanation, which is intended to link drift and reversal over the evolution
of stocks. How does this relate to the news findings here? News losers do indeed have higher share
turnover than no-news losers, and they drift more, but the hypothesis does not fit exactly for news
winners, who show no reversal.
To examine this relationship further, I use a Fama-MacBeth approach in Table 10. Each
month from 1980 to 2000, I regress the one, three, six, and 12 month cumulated
subset stocks (skipping the first week) on various past firm characteristics.
returns of my
I then average the
coefficients across months, and calculate standard errors from this time series of coefficients. In
each case, I wait a week after characteristic calculation before investing. In regressions where the
dependent variable is cumulative returns over several months, I use Newey-West standard errors
with a lag equal to the number of months over which these cumulative returns overlap. Since I
include share turnover among the regressors, I exclude all NASDAQ stocks to avoid problems with
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
32
inflated trading volume due to double-counted dealer trades. This is also consistent with how Lee
and Swaminathan handle volume.
The top panel of Table 10 shows the results from a baseline regression. The preliminary
explanatory variables are log formation month B/M, log market value (from the formation monthend), past month return, an indicator for the occurrence of headlines in the formation month, and
an interaction term for news and past return. The size and B/M terms account for other factors that
affect the cross-section of returns. The interaction term shows how the incidence of news increases
the percent of last month's return that is carried over into future returns. The results confirm
what we know about news. First, the positive and significant B/M coefficient confirms the value
effect. The past return coefficient is negative, larger in absolute value, and statistically significant.
This indicates that unconditional stock returns tend to reverse for up to three subsequent months.
The news interaction term, however, shows that this reversal is mostly ill no-news stocks, and that
firms with news continue to have similar performance in future months relative to other stocks.
At longer horizons, the incidence of news creates strong continuation, dwarfing the effects of the
return reversal (compare the average coefficient on the news and return interaction term with that
for past returns alone).
At the bottom of Table 10, I show the regression coefficients for the same group of explanatory
variables, with additional terms for turnover and turnover interacted with past return. The results
of Lee and Swaminathan suggest that the coefficient for turnover should be negative, since higher
turnover depresses winners and forecasts more bad performance for losers. This is also consistent
with a liquidity interpretation of turnover. For example, Datar, Naik, and Radcliffe (1998) use the
same turnover measure as a proxy for liquidity, and find that there is a liquidity price premium
even after adjustments for other determinants of stock returns. Furthermore, Lee and Swaminathan
suggest that turnover should enhance momentum, so we expect the coefficient on the turnover and
return interaction term to be positive.
As reported in Table 10, all of these predictions hold. Turnover has a negative coefficient,
and the turnover/return interaction term is positive. However, since I measure turnover over the
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
33
formation month and not over a longer period as in Lee and Swaminathan, my results for trading
volume are not directly comparable to theirs and should be treated with caution. More importantly,
it appears that the news coefficients are barely changed, suggesting that the news momentum effect
is generally separate from the effects of trading volume.
1.5.9
Risk changes caused by news
As mentioned before, bad news can increase risk and drive up expected future returns. However,
this does not explain multi-month drift patterns. Losers continue to underperform, even though
a risk explanation would say that they should have higher post-event returns since they become
riskier. The same paradox, in reverse, holds for winners.
In my analysis of cumulative abnormal returns, I have already accounted for some changes in
known "risk" factors, and still found statistically significant drift for news stocks and reversal for
no-news stocks. However, Table 11 shows the evolution of month-by-month 3-factor loadings and
alphas for winner and loser portfolios, news and no-news sets. These loadings are from a time series
regression of portfolio excess returns each post-formation month on contemporaneous Fama-French
factors in calendar time. Specifically, I regress:
Ri - rf =
i + /3i(Rm
- rf) + yi(SMB) + 6i(HML)+ Ei
(1.1)
where Ri is the portfolio return i months after formation, for i = (0, 1, 2, 3, ..., 36), and 3i, -i, and
Ji are the coefficients on the Fama-French market, size, and B/M portfolio returns from the same
calendar months. Note that the portfolio returns are unadjusted and non-rolling. To analyze the
difference between winners and losers, news and no-news sets, I also conduct the regression on
various differenced portfolios (not shown). T-statistics are calculated from robust standard errors.
First, the month-by-month alphas from the regression confirm the general patterns of no-news
reversal in month one and news loser drift to month 12. Second, within winner or loser categories,
news and no-news portfolios have similar factor loadings at inception and over the course of the
strategy. A 3-factor regression on the difference between news and no-news loser portfolios at
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
34
t = 0 (not displayed) shows no statistically significant differences between the two groupings for all
factor loadings. Furthermore, the difference is statistically indistinguishable from zero in almost
all subsequent months, except that news stocks achieve higher betas than no-news ones. The same
holds for news and no-news winners, except that the size coefficient of the news minus no-news
portfolio is 0.17 '(t-statistic 2.06) at formation. However, this differenced portfolio SMB coefficient
becomes negative in most post-formation months. Overall, there are no consistent differences in
factor loadings aside from market beta. One might say that news stocks gain market risk over time.
This makes the similarity in their post-formation returns more unusual.
Are factor loadings for winners much different from those for losers, within news or no-news
categories? No. For the news group, only the formation month market beta on the portfolio of
winners minus losers (not shown) is statistically different from zero (0.29, t-statistic 2.24). However,
any differences disappear in months after formation. For no-news stocks, winners have higher betas
than losers at formation (the beta of the no-news winner-loser portfolio is 0.37, t-statistic 3.14), but
all other winner and loser factor loadings are indistinguishable. Even the difference in beta quickly
dwindles until the portfolio of no-news winners minus losers attains nearly zero factor loadings
across the board. In summary, loser and winner stocks are similar over time for news and no-news
groups. If anything, they tend to become more and less risky, respectively, which is the opposite
of what one would expect from observing the drift patterns. Therefore, the abnormal returns are
unlikely to be caused by changes in risk.
Finally, the 3-factor model does a good job of describing returns. R-square statistics for my
portfolios are around 0.7 to 0.8, similar to other diversified portfolios like mutual funds that have
R-squares of about 0.8 or more.2 3
23
0ne could add the "momentum factor" to the regression. As mentioned before, news drift is a possible cause
of momentum. Therefore, it is reasonable that it could be strongly related, but not completely explained by, an
unconditional momentum portfolio that includes no-news stocks. Using the additional momentum "factor portfolio"
from Ken French's website (UMD), I repeat the month-by-month time-series regressions above. Almost all of the
loadings on Market-RF, SMB and HML are the same. As expected, the news strategy has a statistically significant
positive loading on UMD (around 0.5 in post-formation months), and the strategy alphas are cut in half, but are still
statistically different from zero. The no-news set also has a smaller, but statistically significant loading on the UMD
factor. R-squares for the regression rise only slightly when compared to those in Table 10. Therefore, while the news
long-short strategy is related to traditional momentum, it is not entirely explained by it, which is consistent with
unconditional momentum reflecting both news drift and noise.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
35
1.5.10 When does drift occur?
The results above indicate that smaller stocks seem to underreact (mostly to bad news). Why might
this be the case? There are two potentially overlapping explanations. The first is that investors
simply have different attitudes to good and bad news. They form incorrect expectations about
future performance, and consistently underreact to bad news. Another explanation would be that
frictions of some sort prevent some information, particularly bad news, from being impounded into
the stock price. The frictions would likely include short sales constraints, since other costs such as
bid-ask spreads or noise trader risk cannot explain an asymmetric drift pattern. Note that short
sales constraints might explain the persistence of drift, but not why it exists in the first place. For
most holders of stocks, it makes more sense to sell shortly after negative headlines are public, rather
than wait for a predictable 4% loss.
One way to examine these stories would be to look at how much of the news drift occurs on
months with subsequent news. This approach is similar to the analysis of La Porta, Lakonishok,
Shleifer, and Vishny (1997) (LLSV), who show that the market appears to be positively surprised
by the earnings of value stocks and often disappointed by the earnings of growth stocks (i.e. post
earnings drift is stronger in value stocks). If investors form systematically incorrect expectations
about performance, they will be serially surprised. One might expect to see that most post-badnews drift comes in months with major news. Alternatively, if investors face difficulty in selling
their stock or other frictions, one might expect to see proportionally more post-bad-news drift in
months without major news. Investors could liquidate positions over an extended period of time
to reduce transactions costs. If the main cost is short-selling, there will be less extended drift for
news winners than for news losers in non-news months.
Earnings announcements are times when we know that major news was released. One way to
see if the profitability
of a news strategy comes when information is made public is to see how much
of the return occurs during earnings announcements. However, we know from other research that
stocks making earnings announcements tend to show drift on subsequent earnings announcement
days. Before proceeding, we must first account for this phenomenon.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
36
Earnings announcement stocks constitute about a third of all news stocks each month. Do
the responses to earnings announcements drive my esults? To answer this question, I repeat
the size and B/M adjusted analysis of Table 5, but exclude all stocks that had a known earnings
announcement in the exventmonth from my set of observations.2 4 A few no-news stocks (not shown)
with apparently unpublicized earnings announcements are dropped.
Even after excluding earnings months, the results are comparable to those in Table 5. I show
only the non-earnings-announcement-related news strategy returns in the first group of columns in
Table 12. Long-short adjusted profits to a news strategy are smaller (around 3% twelve months after
formation vs. nearly 4% with earnings announcement stocks included) but still large and statistically
significant at the 1% level. The difference between news and no-news long-short strategy is still
economically large and statistically significant at 12 months when compared to Table 5. News
winners experience a reversal (-0.2% vs. 0% for Table 5 news winner returns in the first month) and
have an adjusted cumulative return of -0.2% 12 months after formation (vs. 1.3%). No-news winners
show little change (unsurprising, since they contain few earnings announcement stocks), and the
difference between news and no-news winners is generally smaller. News losers excluding earnings
announcement stocks have lower returns than those in Table 5, Panel C. They reach a cumulative
12-month return of -3.3% (t-statistic -2.8). No-news losers are largely unchanged. The difference
between news and no-news returns is larger for losers, reaching nearly -6.4% in month 12 vs. the
-5.5% found in Table 5 (t-statistics of-5.4 vs. -5.3). In conclusion, post earnings announcement drift
is important but does not drive all of the underreaction I have found. Excluding stocks that had
earnings announcements eliminates any trace of post-news-winner drift. Investors do not appear
to underreact to good news, aside from positive earnings announcements.
Having first accounted for post-earnings drift, we can now explore if there is "post-news"
drift for non-earnings-announcement-related stocks. Mechanically, for each post-formation month,
24I first check to see if my sample exhibits earnings momentum. I select all stocks that had an earnings announcement in each month, and repeat the size and B/M adjusted analysis of section 4.2. Similar studies usually use deciles
instead of thirds. I find large and statistically significant abnormal returns of 0.9%, 0.3%, 0.3%, and 0.7% in the
third, sixth, ninth, and twelfth month after the announcement, respectively. The cumulative twelve month abnormal
return is over 3.8%. Therefore my sample does seem to exhibit earnings momentum.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
37
I take the strategy returns from above and zero out all of the positions for stocks that did not
have an earnings announcement.
I preserve the weights in all other stocks. I also repeat the
zeroing procedure for stocks in subsequent mon'ths that did have earnings announcements to get
the cumulative returns attributable to stocks without news. The sum of the two sets of cumulative
returns will equal the results above.
The second and third groups of columns in Table 12 presents results for this decomposition.
Most of the drift comes from non-earnings announcement months. Panel C shows that almost all of
the news loser drift comes from months without major news, supporting the frictional story. The
subsequent CARs from counting only those news loser stocks with earnings announcements are
indistinguishable from zero. It seems that investors are not particularly responsive to subsequent
news about stocks that fell into the news loser category. However, there is continued price pressure
in other months, suggesting that someone is selling shares after bad news, even in the absence of
more bad news.
The results for the winner leg of the long-short strategy (Panel B) are harder to interpret.
News winners continue to rise in subsequent months when they announce earnings. Yet "news
surprise" is almost entirely cancelled out by the negative returns in non-announcement months.
The result is the small CARs for news winners that we observed earlier. However, the magnitudes
of the winner movements are small when compared to those of the loser drift.
In summary, I find some signs that frictions increase bad news drift by slowing the incorporation of information into prices. The decomposition of returns for news winners implies a more
complicated story, however.2 5 Hong, Lim, and Stein (2000) find that smaller stocks with little
size-adjusted analyst coverage experience the most momentum, driven mostly by losers. They propose that investors are slow to react to bad news (defined as an unconditional negative return)
unless they have "help" from Wall Street analysts. My conclusions support this idea. One crucial
difference, however, is that I find that investors are slow to respond to public news. In other words,
the underreaction appears to result not from barriers to "knowing" news, but barriers to "unders25
News returns decomposed by news and non-news (not just earnings or non-earnings) months yields similar
findings.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
38
tanding" it. One possibility is that analysts may not give investors private information, but may
help them digest public news.
1.6
Discussion of results and implications
This study has documented that stocks with public news in a given month experience momentum.
Those that do not have public news show no momentum; if anything, they tend to reverse if they
had a large price movement. What do these findings add to our understanding of momentum and
reversal?
Although the strategy here uses a one-month horizon to distinguish winners from losers, it
overlaps with longer-horizon momentum strategies used in previous research. First, the general
patterns of momentum are the same: winners beat losers for up to 12 months, the patterns are
robust to risk adjustments, the strategy backfires in January, it is stronger in smaller stocks, and
holds in different time periods. It is harder to extend the one-month news results to longer horizons
because one would need to decide how to weight news over six or 12 months (and no-news stocks
are rare over long periods), but there is considerable overlap in stocks selected by both strategies.
More importantly, there is a theoretical link between news motivated drift and momentum.
One goal of this study is to test if there is underreaction to public news. This is an assumption of
most behavioral theories, and is supported here. The fact that the "control" set of no-news stocks
tends to reverse also supports the view that investors overreact to signals that are not informative.
This evidence of differential responses to news and no-news is broadly consistent with all three
models (DHS, BSV, HS) mentioned in the introduction.
The general point that they make is
supported: both underreaction and overreaction feature in investor responses to stimuli. There are
some distinctions and qualifications, however. One difficulty in relating my results to the DHS,
BSV, and HS theories lies in mapping "news" and "no-news" shocks to the signals featured in
their models. However, the general point that they make is supported: both underreaction and
overreaction feature in investor responses to stimuli.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
39
Delayed overreaction drives the DHS model.26 A central theme of the DHS analysis is that
"stc 'k prices overreact to private information signals and underreact to public signals" (see DHS,
p. 1841). Specifically, momentum results when investors overreact to certain signals for a period
of time and ignore others, which inevitably leads to reversal. The question is: what is considered
private and public information? If private information includes things investors read in the newspaper, then the growth and shrinkage of news drift supports their view. However, the fact that
most drift occurs in the absence of subsequent confirmatory signals (see Table 11) is at odds with
one feature of their model, that overreaction gets worse as more information confirms initial beliefs.
If private information is captured by no-news, then the one-month no-news reversal would reflect
a correction of overreaction. DHS, however, model much longer horizon reversal. Instead, I find
very short horizon reversal. Again, while this feature of their model is not supported, there is a
difference in responses that is broadly in line with the assumptions of their theory.
The news/no-news split, combined with evidence on frictions in the previous section, fits
the HS model relatively well. Interestingly, this is the framework that has the least grounding of
the three in specific aspects of psychology. The empirical results show that people do not appear
to read the news headlines, while some people seem to over-react to price movements. Also, it
seems to take a long time for news in headlines to affect prices. These are all features of the
HS model. The distinction between public news and no-news seems to be tailored to test the HS
"news-watchers" and "momentum traders" assumptions. One apparent difference is that in the HS
model, newswatchers receive (and act on) private signals, not public ones. They are private in the
sense that newswatchers only act on a fraction of information at any given time, and ignore the
rest. This gives the model a long-lasting drift because information diffuses slowly. However, the
definition of "private signals" is not necessarily at odds with drift after public news. The model
is really meant to "capture the idea that information moves gradually across the newswatcher
population" (HS, p. 2148). Like DHS, HS never really define what these perceived private signals
are in real life. It could be several pieces of related news, released sequentially. It could also be a
26
Strictly speaking, in the DHS model, drift reflects "continuing overreaction". I use the term "underreaction" to
denote that prices move only partly to the level they eventually reach at a future date; after that date the price may
fall again. The only difference between the terms is the horizon on which an investor should condition.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
40
single story, which somehow is revealed across investor groups slowly. My finding that news is not
autocorrelated supports the latter interpretation. Therefore, public news and private signals are
not necessarily different.2 7 Furthermore, the fact that post-news drift occurs in the absence of other
news supports the idea that information diffuses slowly. The HS interpretation of overreaction also
fits the horizons of no-news reversal. Depending on the calibration of various aspects of their model,
reversal could take place over many horizons, but most plausible values show that overreaction is
very short term (within a few months), and the correction of the overreaction much more gradual
(see their Figures 1-3).
In the BSV model, the sequence of signals is important to determining the amount of over- or
under-reaction. Since I do not form portfolios beyond one month of returns, I cannot directly test
this assumption. BSV argue that momentum is stronger after a period of contradictory headlines,
and reversal more prevalent after a run of reinforcing news stories. Therefore, both momentum and
reversal could result from the same type of signals, in a different order. While this may be true, the
strong news and no-news return difference suggests that one does not need a time-series pattern of
signals to predict returns. Furthermore, the BSV model is extremely difficult to test given all the
unspecified parameters such as an appropriate horizon and what constitutes a particular sequence.
However, in the context of news headlines, one could look at returns around headlines over the
past quarter or year, and classify them as good or bad relative to some benchmark. One could
classify the chain of news returns by some pre-determined algorithms, and analyze momentum
profits. These are tasks that require more thought and theoretical guidance.
Finally, none of the three models makes any provision for asymmetry in returns, nor do they
explicitly say why the patterns should be stronger in smaller stocks than larger ones. This clearly
suggests that new versions of behavioral theories should be richer, perhaps by incorporating specific
frictions and heterogeneous investors.
Recent work by Daniel and Titman (2001) and Cohen, Gompers, and Vuolteenaho (2001)
also enrich our understanding of momentum and reversal as responses to different types of infor27
For example, Hong, Lim, and Stein (2000) effectively assume that analysts (who can be fairly public figures)
produce "private information" in their test of the slow diffusion assumption of HS.
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
41
mation. They focus on the specific meaning of information rather than how people hear it. Daniel
and Titman decompose accounting and market variables into expectations about "tangible" and
"intangible" growth rates, and provide evidence that investors overreact in the long run to the
second. Unlike me, they find no evidence of underreaction.
This is because their "tangible" news
and my "public" news are not exactly the same. The tangible/intangible differentiation relies on
valuation levels and longer horizons instead of changes in expectations over a single month, and
the results are not directly comparable to the findings here. Cohen et al. use vector autoregressions
(also based, in part, on accounting variables) to show that individual investors underreact to that
part of returns that predicts cash flows. Their "cash flow news" is conceptually similar to my
public news and their conclusions are almost the same. Since I measure information directly from
its source, however, the test presented here is arguably more direct. On the other hand, it is also
less specific than theirs about the nature of the information. They go a step further and show that
most underreaction is due to non-institutions (those with less than $100 million in assets), which
helps us understand who might react to news (and prevent drift from occurring) and who might
not.
1.7
Conclusion
I have examined various views of investor reaction to news in an integrated framework. Using
a comprehensive sample of headlines for a large, randomly selected
oup of firms, I test the
hypothesis that stocks exhibit no abnormal return after public news. This is not the case. Stocks
that experienced negative returns concurrent with the incidence of a news story continued to
underperform their size, B/M, and event return matched peers. Stocks that experienced good news
show less drift. On the other hand, extreme return stocks that had no news headlines for a given
month experienced reversal in the subsequent month and little abnormal performance after that.
The post-event drift is mainly after bad news and is very robust. The conclusion of overreaction
is somewhat weaker, since liquidity effects can drive the reversal of returns. However, the reversal
continues to appear after waiting a week to pursue a no-news long-short strategy. Ranking by
CHAPTER 1. STOCK PRICE REACTION TO NEWS AND NO-NEWS
42
idiosyncratic risk by adjusting for size and B/M characteristics does not eliminate these results.
Neither does weighting by number of news stories or excluding earnings announcements. Buy-andhold abnormal returns display the same pattern of news drift and first month no-news reversal.
Drift patterns become less evident as one moves up size quintiles, implying that underreaction
is mostly confined to small stocks. They also seem stronger for low-priced stocks, although the
results hold for higher-priced stocks, too. There is evidence that the relations are weaker, but still
economically significant, in more recent years.
These results seem to confirm some assumptions of the DHS model of investor behavior or
the HS model of two classes of investors. Investors appear to underreact to public signals and
overreact to perceived private signals. The stronger finding is for the news stocks. This noteworthy
result is more understandable if one considers different types of investors. Most of the drift is on
the downside among smaller, probably illiquid stocks. Furthermore, most negative drift happens
over many months even without new information in the press. This support te idea that more
sophisticated investors cannot arbitrage away the pattern, since shorting is more expensive than
buying. Thus, it appears that both bounded rationality and frictions, far from being minor factors
in asset pricing, instead interact to create relatively long-lasting anomalies.
CHAPTER 1: TABLES AND FIGURES
.
43
0c) 000M))
t W.- N N
0o 0r- N (0D
o r-0 0 M)
r -- (, oo
CC
(.
rC~
l CSC~l
Nr r-0
8 I
0
1
0
.0
M
E
3 ; _
O
U)
L a.-
""',
C
E-
e
E,
2
2
n
e
z
O 0)-
r - O NN
~~~~~~b~MON
tU0W (r-W 1--r -
v
-
-
-
-
cV)
N N C
N 0o 0DC)
0- 9
4.
v
v M
M
A
1f.Nl
3~
C
.-
i
I
0U., ¢
sr
I·s
30
.
.j
0
0
a)
to
a0
M
co
.r
O
0_
.E0 -2.E
o
.
a.
E
Co
2010
o
0
IL
o
o
E 2
zM
Xe
C
0
Cu
~B
W
(C0)
M
4. Lo
(0(0(0
_4
1-4.
NN
MC
N
C
z
Ia
3
a
0)
-
OO
0
N ) n (00
O CO
0)
N N N N
c)
)0
.
c')Ct')
N0
4 4.
W
(0
C
(
_ - -
N WW
- oO C-0
4.N40
V.
Oco
1-
I .4
C
0c
O
)CuO
DC- U)0)
N N C N N
N
CO
sIn
-0
6o
vu
t
N
(0
4C
0a
I-
0
co
Cu
S
a
.2
a
zCs
U
2U)
C
i
.2
.5
C
0 0t- 0- N1O
Nw(DmN0
1.4. 01N
0( 0 0 s -(
r a
r1. ON -00
n 4.4
1.-
Z
D
D
0
CO
_a
M 00C)(0
A VqMC
0CO
(D 0
0-CO-o u . %M0
O
)) o -CCcoOInr)C N
c
V
v
M. Ct V.t V 1 o 1D
_
_
-_ a
OD 1
04
M
-
1-_
0) 0) 0) 0) 90w 00C0O0 0
La
_
_
V.-
00
w
--
0
-
__
) 0
-_
C)(0
N ) MA)
0) ) )0
.
V.
_-
1-
0)
0)0)) 0
,,
o
0
N 0)
Q)Q)Q)Q~~Q)4
1-1_1_V_
(0
0
cV)
C0)0
I
D. q0
C)
C- 4- D C N(0t2C-Dfl.-(0
w(0(0000
0
ao
(0(0c C--
n
.0
E
z
c
00
Z
a
.(
0C.10
(001)
04.OC
;mvvvv
Fo Vzi
M
C
0II
5
(0
00
C
's00
11
o
M
4.
0
z
0 -0
-
m
0)
M
Ou ) t-O
44
2L))4.N(0(00)4-0 4 0 )t tON0)
C
G3 Lg.
8
04
(0_U0
1-C'Nr-NO
4.4'4(0
. 4)(0(0o00
. 4.
( 010(0t(0(
4. 4.
4
iA
nO
3C
0s
0)
5.
0
0
co
( N W
*-1-1-1-1-1-1-1-
LL
E 2
0
D 0
S
C'
(U
0) F.N
0)
co
O)
Cv)
I-
8
o
1- (O N
C
o
U)
-E 0
r- m)
N 0 0000)
C
5
.0
C>
l
(0 ( 0 O N 1 N
-0 4 0t
) O0 .
t D 0(
0
. .( 0N0Nw-CC1)N(C)
~~N (1N NNN N'-NN'-N
r-~~ON~~~~~O
_N"Mm"Lon""~~r
0
C)
CD
0
00m
_.s0
,
r rFz
co
.E
Cu
0.a
0
0M
.s)
0) a '2
3) M
D
%
E
co
1= E
CHAPTER 1: TABLES AND FIGURES
..,t
.0
-.C
0C
C
-0
C
4
L.
a,
00
CoOC
44
00 N NrfC" 0
0 U)
-
'-
N I-
o
c C Vtr- -
6
C)
2
2
2
Co 9
C z1
3:
· a,$~r
a
W$Zm
Sj C0'ao:
o
E=
(
.-w 0)
-c:c
-c
.4
0
Q)
o .~,'·-
a
0
.
0 -'
~0
· ~.CU
z
2a
?iO.
0
-
0
'-
L
C
V - --V VXN
't -t- NM
N NN LO
'--
6
0
6
i
N
CDCUD
N (0 U)
''
99999999999i
C
,i
6
C,
~
M
LO
V C
v- v -v-
C. ~-
.
0'--
-
6 6 6 61 6 6O 6 6 6 6 6 6 6 6 6 6 6
66 6
a
-j0 2
4.6
:~
0:
-
' -
-)0 'C
i-N J- r666
6
iiiiac
2
E
0
'
' - '
e C CVC N CDr- N
C
0E
·.c"CU-- .. 0
Eoc
Co
0
C
U)
co
6
-
N
'
"it
C
0~~~
-
0000
0O-'---
I'l0
0-
rN
6
lx
o o~~~~
,
co
0
r
E
oc
CEo
Co
N
N
000
0
a C00
0V N
~ ~N 0N
N ' - 000
N C
N CD
C%
666666666666666666ooo
0
I)
C N I- r- _ -U
0 '-Vt'-'N C)
Mv
- '-N
V
to
N
r- Vv- NC
'
'
N
v-v-
m
M
C
L
C
O
*u
: '
0C .
F
o
oco C
w
0) c0
00
CDCD
Lo M
0)0
~
O~ ~ ~0
~
~
~~
O
I
~
~ ~
~
~ ~
tD
0o ' -O0 cO V -
Z
Z
0:cr 00
0
wCC.
oco
c
o0
.C
~
0
IC
OE~o P
E
zZ6
CE0
C z
r~
~0
-N NN ~
N
~0~Br~~~
~ ~~~Or
~0~~~rN
CD
N
(
~ ~o~~~o~~o~ ~
co
co
Wr
a
cn
0
4
C
W)) 0 =
.6~ 2~ ~
Ct'~
.0
WE
I-C
00
EU
0
m
E,
C
0)
0 0 C'-
co
CU55516
0 cm
0
I d - C)CD
0 0MO N0)aD'0O )C lD
N
C 0 ) S ( 0)
'C~
oD0
0
)
00
O-
0
N'-
0D
Z
LOz
3Co
0
C)
N
'
__-
0
N N CVN
-- -
.=04
Cl)
D
N C c)
O-
)
z
CE ch gm w
c~
Jao,
u30~·-E~~~
C)
'N
1!
4
o
0
Z
E
'-
-
'
N
N'
-
M CO
N
V
O
0
._
m
O'- N tC)V CDC 0 0O' _N
CU
0
0~
~0OaOo
)0)a0000
o_ __
__o
~
0CV
v
V) CDCD-0 O
_
__
o
)
c
.E g
p
CHAPTER 1: TABLES AND FIGURES
45
Er
aC
4)
V0
<
.
0..C
0
E
i
Ec=
0
E:
coE.
o
(0
En Q0
i
0
.-.C r
E "
0.
C
a
E
E
0c
0
Co
--
4
a;
e-
E
CL
Iv)
6
>
10
.
0.
Q-
0
D0
0 j
2 0.
0-7C
_
V
=0
0
80°o
E
0)E
0
a0
-*
E w
oj:
LI.
0
-0M
2
E
S
E -2 1
-4)
E0U-)
2rL
WCL(D
L-
- Ew5
-
-
im l' UL CDI-
|
l&
e
0
I
-J
"W *0C2
0
w w
C
_
,O
10
0a 00
00 8 ~ CO
COO .0< 00
=~~~o=
0.E._
> Mri(D3.o
-i0300
0
v0
Ca -
OC
lzc
0
E
>C
04
..
>W0
> 1iW
,' C2
00
0
li00
g
0. Mco
0 o .ao
(D2
.o_0'E.
0
C
E
a0>
(C
*X
Ae
2
.C
.0=
0>
U-
- 0-M 0
R
E
QW
W
Cqo 0, )M
) iO
U v- v 0e-
. Io
0r.
. .I. .o
!
:E
C
D
2
0t· p
-B
0C0) C
>a
.-
0
00~·
t
.
D
D
w!Ct
"0
E
L-a,
Mt a,
aL-,>.>1
0 3 3 D
<
2
z0 L
v-o
oo
!~~~~~~~C
.
>
0 c
9
0 a)
2 W
Da
W
'3
<Q 73 DEE
0 =2
:2 C)
>
M,
0
(
- U)Q
-0 0
~
rrO~~~~~g
L
&
L
, -5: 0 e 0D
>
_3
2W
L
0 00
300
LL
0)
0 0
O CS)
C,
)
_2C
)0.
0
0
a)
0) CD O aw
C
CC C
O
2
O
0
00
CCO
0 0 0 0 0
C c
C
0
.
0- 0
0 cCa
-
S z~-~
-
rr
cC
a 0
0
CHAPTER 1: TABLES AND FIGURES
46
Table 4: Cumulative Long-Short Returns (%), 1980-2000, at Different Horizons.
This table shows the summed raw returns for rolling portfolios over several holding periods. Each month,
all stocks within a subsample are ranked by their performance. Stocks in the top and bottom thirds are
held in the same portfolio with positive and negative weights, respectively. This portfolio formation
process is conducted on three sets of stocks: 1) an "all" subset of randomly selected CRSP database
stocks, 2) a "news" group consisting of "all" stocks that had at least 1 news headline during the month,
and 3) a "no-news" group of "all" stocks without a news headline for the month. The resulting long-short
portfolios are then aggregated into larger portfolios with overlapping positions. Overlapping portfolio
returns are summed to get cumulative retums. Panel A shows the average cumulative returns and tstatistics to immediately investing after portfolio formation, and Panel B shows the results to waiting a
week after formation before investing. All months are weighted equally in the time-series average. Only
returns from January 1980 to December 2000 are used in performance calculations.
Mos. after Portfolio
Formation
All Stocks
Average
t-statistic
News Stocks
Average
t-statistic
No-News Stocks
Average
t-statistic
Panel A: Immediate Investment After portfolio Formation
1
3
6
9
12
24
36
-0.98 %
-4.18
-0.37
0.45
1.74
2.48
0.88
-1.15
-0.74
0.59
1.68
2.18
0.43
-0.38
-0.33 %
0.86
2.11
3.78
4.65
3.49
0.90
-1.34
-1.83 %
-6.92
1.67
2.80
3.63
4.15
1.78
0.31
-1.98
-1.82
-0.96
-0.53
-2.74
-3.94
-3.74
-2.26
-0.91
-0.45
-1.36
-1.43
Panel B: Waiting 1 Week After Portfolio Formation Before Investment
1
3
6
9
12
24
36
-0.15 %
0.44
1.21
2.44
3.18
2.06
0.05
-0.73
1.00
1.72
2.50
2.99
1.07
0.02
0.35 %
1.49
2.76
4.36
5.39
4.72
2.31
1.71
3.32
3.92
4.38
4.96
2.41
0.76
-0.80 %
-0.80
-0.74
0.18
0.58
-1.07
-2.57
-3.71
-1.67
-0.95
0.18
0.51
-0.54
-0.90
CHAPTER 1: TABLES AND FIGURES
47
Table 5: Cumulative Size & B/M Adjusted Returns (%), 1980-2000, Skipping 1st Week
This table shows the summed size and B/M adjusted returns for rolling portfolios at various horizons.
Portfolios are formed for two subsamples of stocks: 1) a "news" set consisting of all stocks that had
at least one news headline during the month, and 2) a "no-news" set of all stocks without a headline
for the month. Each month, all stocks within the subsample are ranked by returns. Stocks in the
top and bottom thirds are held in an equal-weighted portfolio with positive and negative weights,
respectively. One week is skipped between portfolio formation and investment. Rolling portfolios are
used to calculate standard errors. Panel A shows the average cumulative returns and t-statistics to
the long-short strategy, Panel B shows the results for winners (top third), and Panel C shows the
results for losers (bottom third). All months are equally weighted in the time-series average.
Mos. After Portfolio
Formation
News Stocks
Average
t-statistic
No-News Stocks
Average
t-statistic
Difference
Average
t-statistic
0.03 %
0.91
1.70
3.12
3.93
3.42
0.97
Panel A: Long-Short Strategy
0.20
-1.14 %
-5.75
2.44
-1.60
-3.78
2.94
-1.58
-2.34
3.99
-1.09
-1.23
4.43
-0.70
-0.69
2.38
-2.33
-1.44
0.48
-3.63
-1.70
1.18 %
2.51
3.28
4.21
4.62
5.75
4.60
7.01
7.37
6.01
6.10
5.62
4.59
3.03
formation date
1
3
6
9
12
24
36
16.60 %
0.00
0.32
0.42
0.99
1.32
1.84
2.28
Panel B: Winner Portfolio
57.83
16.67 %
63.70
0.00
-0.24
-1.98
1.26
-0.12
-0.39
0.94
0.44
0.92
1.59
1.42
2.24
1.69
2.23
2.90
1.44
4.24
3.25
1.26
6.35
3.35
-0.07 %
0.24
0.43
-0.01
-0.43
-0.91
-2.40
-4.07
-0.36
2.00
1.66
-0.03
-0.63
-1.01
-1.57
-1.77
formation date
1
3
6
9
12
24
36
-13.89 %
-0.03
-0.60
-1.28
-2.13
-2.60
-1.58
1.31
Panel C: Loser Portfolio
-83.37
-13.90 %
-86.73
-0.28
0.91
7.17
-2.09
1.48
5.01
-2.45
2.02
3.68
-2.77
2.52
3.22
-2.65
2.93
2.94
-0.80
6.57
3.50
0.43
9.98
3.79
0.01 %
-0.94
-2.08
-3.29
-4.64
-5.53
-8.15
-8.67
0.14
-6.83
-6.23
-5.45
-5.59
-5.25
-4.07
-3.04
1
3
6
9
12
24
36
CHAPTER 1: TABLES AND FIGURES
0C
0
0 )
CD
I)
w
a
00
aU
o.
0
CU
CM
(A
z
0.
0M
0
oCUC
C, ¢ C CC
tD 2
2
00
75C
0',
E-'f
0.!
EBi
,c ._ c
0
;
0i
._
. (
'0
.C
10
(P)
0
z
C
M
pc (> go * -O
cu
CD
1M
10. i .0.0 . cq
.D0
up,
0
o a
2a
0'- 100'---CO
c
;,
V-:6 V.:oi 6
A
0-
N CD
)0 N 0)
N~ 0(0
0D)10 CV)
C O -CV CM
CV
.5
C
C
IL
0.
Ln
C
oiOcn
ON ci
O
f)0
C'
EC
c
a a ac a ac
u c c uco
o
C
IL
0.
0)
in
O C D 0( d
U)
o00'D~VcV)Q
Ni Xj uz u
ws cN
t-
(
CDC
C4
r Ocd ?
' 00O-C-O
CD 0N
) E:
O*
)
aD
a CS U Ca
A
E
E
m
:._ c)CUo C:0.
I
et
10)
CU
M
Ek
'-
-
S.-
0
04
cf)
C) N)
T)
0
0
0
*
.0.
:r
A
10
0
N CV1 1Nq10
'-
C)cp0edofqr
0c
r;
d 0IN
e cnjciC(
uj 1cp
CU
D
C'JsCtJN0N
00
CU
S
*J Cv) X (0tj
)Cj
0)
S
C
0
:
.)
S
41
o
cI
0
.0
)
-0 .
t0
.-
0
0
CU
0
1r
-n0
Sc
: C
C)
CC .C)
4) C tV
Co
l-
U,0
'CU .n
CO
0)
10
C
.,
00O
C
v,
o
..
cnu)CJ 0CDC 0
t
0
1 CO
o-v-
-c
) -
-
Va C O C M <D O
m
0
-
W _
o
CU
V
N
Co
I.
2o2 c' ' o
a eg
0 '-CDO
C
2
z
c
C
v- 0 10 (0
N
-
.M
co
C
a)
W
az
2
L-)
.o
(Cc
C
O
M0-
c9 oD
l,
Ci
,S
a) (c
V) 11
Q
0 -0 C C
-0
CU
00
4
ID
D 0
Ur
C
C
0
1 ) c <D
a) cnr cs cnO
)
x
C
0-
a)
0
0
a O
,
rl`
0
Q
o (
C
0
o
C
E
M
p
0
48
ao
I
a
a
co.
.s v . o. cj .
CDo 00 oN
CV
c
S
a
C
E
C
N
O
>
C
E
-C
0,
>
S
a.
coi
. .
-CCL
E
C
D
0,
CD
C) 10 0OJ
o0-
o
CV)
a:> CV
0) (0Or
)d-C
cV (0 (0
C)
C
a(LI
cn0
r o cs 0
-Q
3CUC
n E :0
cU 0
,' ,'a
0 0-a
35
CU
4.
..:C
=M
I
O1D
0 X
0
N
tCV
,
V C
C') .)0
-cCD0 CA c
0o
I
CHAPTER 1: TABLES AND FIGURES
49
Table 7: Cumulative Size & BIM Adjusted Returns (%), 1980-2000, Skipping 1st Week
Using a More Restrictive Definition of News
This table shows the summed size and B/M adjusted returns for rolling portfolios at various horizons.
Portfolios are formed for two subsamples of stocks each month: 1) a "news" set consisting of all stocks
that had at least one headline with high share turnover in the three days around the headline, and 2) a "no-
news" set of all other stocks. Each month, all stocks within the subsample are ranked by returns. Stocks
in the top and bottom thirds are held in an equal-weighted portfolio with positive and negative weights,
respectively. One week is skipped between portfolio formation and investment. Rolling portfolios are used
to calculate standard errors. Panel A shows the average cumulative returns and t-statistics to the longshort strategy for both sets, Panel B shows the results for winners (top third), and Panel C shows the
results for losers (bottom third). All months are equal-weighted in the time-series average.
Mos. After Portfolio
Formation
News Stocks
Averge
t-statistic
No-News Stocks
Average
t-statistic
Difference
Average
t-statistic
1
0.21 %
Panel A: Long-Short Strategy
1.09
-0.93 %
-5.08
1.14 %
6.88
3
1.18
2.73
-1.26
-3.28
2.44
6.85
6
9
12
24
2.05
3.63
3.93
4.18
3.12
4.19
3.93
3.00
-0.91
-0.19
0.61
-0.47
-1.53
-0.23
0.70
-0.34
2.96
3.81
3.32
4.65
5.28
5.25
3.78
3.62
36
2.28
1.31
-1.95
-1.04
4.23
2.61
Panel B: Winner Portfolio
formation date
19.84 %
53.33
17.50 %
62.51
2.34 %
9.57
1
3
6
9
0.12
0.54
0.53
1.18
1.00
1.70
0.94
1.52
-0.37
-0.42
0.25
0.96
-3.23
-1.63
0.57
1.64
0.50
0.95
0.28
0.22
4.07
3.58
0.59
0.32
12
24
36
1.25
2.23
3.11
1.29
1.40
1.44
1.82
3.64
5.46
2.59
3.04
3.28
-0.56
-1.41
-2.35
-0.66
-0.95
-1.13
formation date
1
-13.83 %
-0.09
-0.85 %
-0.64
-7.49
-5.27
Panel C: Loser Portfolio
-68.15
-0.66
-12.99 %
0.56
-78.33
5.24
3
6
-0.64
-1.52
-2.08
-2.85
0.85
1.16
3.26
2.49
-1.48
-2.68
-4.98
-5.39
9
-2.45
-3.13
1.15
1.70
-3.60
-5.24
12
24
36
-2.68
-1.95
0.82
-2.73
-1.02
0.29
1.21
4.11
7.41
1.44
2.55
3.14
-3.89
-6.07
-6.58
-4.68
-4.11
-3.17
CHAPTER 1: TABLES AND FIGURES
50
Table 8: Size-Split Cumulative Size & BIM Adjusted Returns (%),1980-2000, Skipping 1st Week,
Using a More Restrictive Definition of News
This table shows the summed size and B/M adjusted returns for rolling portfolios at various horizons,
for five size quintiles. Portfolios are formed for two subsamples of stocks: 1) a "news" set consisting
of all stocks that had at least one headline with high share turnover during the month, and 2) a "no-
news" set of all other stocks. Stocks in the top and bottom thirds (by news return breakpoints) are
held in an equal-weighted
portfolio with positive and negative weights, respectively.
One week is
skipped between portfolio formation and investment. Rolling portfolios are used to calculate standard
errors. News and no-news long-short portfolio returns are shown for all five size quintiles, in descending
order from largest stocks to smallest. All months are weighted equally in the time-series average.
Time-series average counts of stocks in each modi;ied group are also shown.
Size
Mos. After
Quintile
Formation
Count
1
3
6
9
12
38
1
41
5 (Large)
4
News Stocks
3
6
9
12
3
0.02 % 0.06
0.29
0.21
-0.40
-0.46
0.12
0.16
0.58
0.39
20
0.59
-0.05
0.90
3.17
2.41
31
1.70
-0.08
0.87
2.44
1.59
49
-0.75
1.01
1.4-2
6
2.18
3.61
4.05
2.56
2.42
0.05
1.57
3.67
5.26
7.63
0.18
2.56
4.00
4.10
5.20
90
0.21
0.76
2.45
501
1.31
1.98
3.47
4.01
2.41
3.34
3.28
1
74
3
6
9
12
1 (Small)
No-News Stocks
Count
-0.26
9
12
2
t-stat
3
1
49
Average
1
3
6
9
12
244
1.91
Average
t-stat
-0.44 % -0.87
Difference
Average
0.69 %
-0.51
-0.38
0.69
1.42
-0.47
-0.24
0.31
0.54
1.04
0.17
-0.24
-1.07
-0.46
-0.95
-0.44
0.67
1.34
2.93
-0.59
0.95
0.28
-0.10
-0.41
0.22
1.28
2.32
4.04
-0.55
-0.01
0.62
t-stat
1.25
0.91
0.11
-0.13
-0.47
1.83
0.32
-0.09
1.06
-0.34
0.97
1.51
1.72
-0.60
-1.16
0.37
1.46
2.08
0.13
0.76
0.89
1.38
2.83
-006
0.31
0.86
0.64
0.80
-0.03
-2.16
0.59
1.54
2.51
2.73
3.82
1.69
2.36
2.91
2.32
2.92
1.26
2.94
3.38
4.35
3.96
4.82
6.13
4.34
4.36
3.16
-0.03
1.11
1.48
2.61
3.71
2.55
-1.06
-1.64
-1.40
-0.78
-0.01
-5.20
3.32
-3.90
-2.10
-0.86
-0.01
-
'-If
CHAPTER 1: TABLES AND FIGURES
U)
0)
o..
r
Cl
.
C
0
0
·
O 0>
CD0
c
a
OC E
.
N O)c ) Ct mt
_
f) cno
Cc a) o~)
(c
10 CO - 0)CN
c c) c c
a)
sCOW
i-
cc
If
a
52
U: C O O U)
d) O O oCvi n~
cO
LOl NCi C15
c0 0
O
O O-oo
0 - C00
O 0
0 c
CD CiC~
10 N
9C, cX
i-'o Cnc
a COCul
.- a,~D
o
a
CO~(
)
a,
c-½
'2-4U)
CU3
- CD
coa,
C
L
C
o
M
()
an
-
C-
,) e )-
0
00
CD
D.
co .
ES
· o-
0)
Irl z
.
Cc o
c
o
U)
cm
is
V
aa i
1
'C -
*c-
CC O
m ,-W
-za
;n
C U>
mn~-EEo
· .' I-0
3" ,o
0C
L-fr~.-.. 0>m.M
S- a)
o.C
:
U) CUO9
Ei-d-1dt ) NNN . a0
a ;O aOCO
ct
C 6 6i
N
r~ r->
~, L~
rco
c c
a-o
o 00.-coaommM'
IE
-o'
rC
OaL ,c
E
a-0
m
j m
c
.
. ~
. : " ,:
ra,~~~
L) aEOaE
cO r
CCO)O
O0
_O
OU) C0o
0C
CO
C
U
4)
ao
a,, zZ i.
C
a) Z
c
E 0D
o:!
to
c~~ o 0)
c C) C
c ao a
(D
C
O O 0n a o
a
a
cOU)
9
L 1s m m CD cO
01°0
u)
**
· o ,--
aM N o
(LU t) CO
CD
CO)
i--
CD
a, N
.. C
t)
C0
0
09Nt) a10(0
a
0o a
-
aD
C0 04
0co
U)
tNO
CaN
co
0( 7( 0 uv
oC
ON
cl
O O 0O
.
.
e
1s C) cD)
o LoE
cd
If
(
cp V c
dm~
c co
d
0pC uq
ZV
7C
,
E(a5<r
d fl oeN co3 ca o
O
i
Lo-
O
Cm )
"t
C)
. C)
O. cO N
oA5
4
U.
0)
V7 c CR OR cl
CD CO 1NCO
- ) Q
!rC c)
0 )
cao
(c0( )
, co C
La r
m
m
~-0 0
E 0)
a, E, a,~
u
CO
c
CE 0CD. aC
(D
..,
) . C0
OLO O
.04 M
N
00R
I-
NdN0O
E
6
_Y
O
O
LO
CR
lp cy
CO Ct
C1 I'llV
(
CNi
_ 0)C; C;
C U) c
a,
m
a
a)
O0
0 -0 4m-..
...=- co0co
0
..
.
._
>
E
~
co
* 000l
N1
(Nec i-ac) eO
E CO
oC N
l-
cmo c
a0
C
-o
.>
o C U
N ~ c:
'r
~
O.~-"O'
c ~ 0, ~ aC
t
q~ (DOCD
._!
c C0
a
cs
n .
ZU)
3:
0
0..
-
- %c, O,C cN A
co
---
In
_
O
CfII
co
N
CO
Occ
Qh
af d' | . 2
C,)
o
:
)
a
CLc
aw
a oI
",.t IP cpq
N
0)
,.. <.
0 CDCD O
f
C a)co 2x
v)
Cl cn
CU) n
c Ct)
cT
ITs
t) CtD
o o0O
csi
N
i- 0(
CHAPTER 1: TABLES AND FIGURES
53
E
Co
U)
Lo
0O
cC
On
_E3°
r:
ru
,
E
>
_
NV
-
c
(a
ODj
q
uo
>_°.
Eg
co
co0Q)
o
Co
)
0)C
oo cql
CI
6
-
O
Ci
0
,_Cio
o0 N
vs
Cla
E ---
.c
.
=
3
CIJ
CD
o
yc
)
M*
O
O
.
I
0'Y
°
.
N
0O
0
os
os
o;
O
a
os
,
o,
NO
o .
0 r- E
e
0
sCo
o
. >' C .
,
0
0
0
o~
o a,
c' -
,
0
00 O--
E
C
ia
U
w
3:0
.-r
m E
EC
o,')
'i-
CD
CO
Oc
c
- CU-
0c
-
co
2
30
o
-0
oO C.
o0
0
Os
E
I-
od
C
'z
o: O
O
Osq
UO) s
O
6
.
a.:
-
-
0
co
CD
0c
9N
C).
0nC
-
0
o . O
q
0 -0C
O
0
0
O
d,
d
o
a
CM
Cj
tO
O
O0)
-D <o
0) 2
0)
2
a
-
C>
o
Qo o
C)
·
C
dC
CD)
d
o
0
ir
e
E
E
0
E
: m
o co
c0
-
o
oo
o
d
No N
gc
Ž 0
-4
C0
o
o
o
0
0
0
0
0
0
-
C)
(D
O
> oE
-o
'*
c
0
of
CM
N
Os
o t.
6
o
N
*
zj o'
0. · - 4)
cc
°- ( ,
.0
t,,-
O
r-.ff
I--,
s
C0
,. r
E
E
0
N
=
-C>C
t-.,
0
00
,,
E
E
E
C
N
d
CHAPTER 1: TABLES AND FIGURES
54
%
ag ~a
.E
fl oo(6 cico C5 6
o
C5
r
fl- co c
I-
a
E
a) -9
Oa)
-J
LD
O
:E
"0
C
co
"0
a
0)
666666
C> t QO
O
a
U
C CD CMC N
c
act
'°Cs
E
'o
D a,
.E
3
a: *
.r
c
a
oE
o
I)0ti
E
Ta,
0
.C
tL~
eU
(U CU
zr 00
E
0
g0 D'
-
4,
a
.0U'O
m._o a
0 ':S0
a
yE
0
a
o0)
co
a
4-0
)-
0
0
z
z
z6
6
O
o
5
0
0C
D
tj
CU
U
C)
.§
.tD
c ,8
6-
0a
CU
D O
Ir.
C -
Q
a)
.
-j
CD
TO
c
co
.L
O 0 o
E .
U)
a)
a
0
0
z
C
c8
17
CDO- CD 0
CD
I a
Cc
6a66
ciJ
I
I
0I
-i
0 cO 0
N
cto
V V:
r
O O
cn ci r~l ci rl.-
-
T
7 r· V7
C0i
C C
) N
N NCD
666666
C4 (P C4 C4 CN 0
ce
(j
r'r:i
N
C
<
VD T O
M
"-: c; O
11Z
6 CO
6 cO
t
C
Cl
iiE
U")t 0)
O C6ui CNc
N o c V)
0
z
N 0) 0) ) 0 0)
oo
o o o
;
60
C6
1:> t
COf)C
0 U)M
0
-
v
C
I
I
M a
0
u
O r00
I)
O
C
03c~
I
Iq O)
It
0000C l
-O0-
CU
CU
l) !-')
E.
·
a!
(
d
V- - 04 -
C
a:
a,
E
M
0
3
<d
'- C)I)0)000
0
C
V5
0a)
b
I
c o
EO:
o.
II
C
ao
0 oW
C
0e
0 -.
C
0
0 )E
6D
"It 000co0
co U) v- U)
o C!
0
c
aE
a
Vr
0~ cn
tR OR
0 cn
0: oCe
0R0
c, C
D
0
E
t-
le _ 0
a
0000 0
z
-'
0 Ds
*
r- CQ CR
o:
C)
(A
0
.D
c
oD
W
0
1.
:
O
V
o oi
0
C,,
CD
a
c
N
co
Im C
0
O 00000a
CM0
1rV C!
C C40 ci ci
Ca:
a:
"I
co2a
-
c o
A
)a tD
aD, a
:0
CV)O
)
)
I
_
0
D
.,D
.0
q) @ 0
E
_
r-o
N
0'
0100
o o o
V
0 c
C
C) N
VI,
I CHAPTER 1: TABLES AND FIGURES
55
Table 12: Non-Earnings-Announcement-Related
News Stock Returns
Decomposed by EA and non-EA Months, 1980-2000, Skipping I st Week
This table shows the cumulative size and B/M adjusted returns to buying 1 month past "news"
winners and shorting 1 month past "news" losers. I exclude all stocks that had an earnings
announcement in the formation month. The returns to this strategy are shown in the first set of
columns. The cumulative payoffs to the non-eamings-announcement-related news strategy are
divided into returns that 1) would accrue had only those strategy stocks with earnings
announcements in a subsequent month been held, and 2) returns from holding only those strategy
stocks which had no earnings announcement in a given month. For the long-short strategy, stocks in
the top and bottom thirds are held in an equal-weighted portfolio with positive and negative weights,
respectively. The resulting portfolios consist of overlapping positions. Panel A shows the average
cumulative returns and t-statistics to the long-short strategy for both sets, Panel B shows the results
for winners, and Panel C shows the results for losers.
Months After Portfolio
Formation
Non-EA-Related
News Stocks
Avg.
t-stat
Decomposition of Subsequent Returns:
in EA Months
Non-EA Months
Avg.
t-stat
Avg.
t-stat
Panel A: Long-Short Strategy
-0.37
0.01 %
0.05
1.18
0.13
0.83
1.76
-0.01
-0.03
2.94
0.12
0.37
2.96
-0.03
-0.08
1.85
-0.54
-0.86
0.39
-1.55
-1.73
1
3
6
9
12
24
36
-0.07 %
0.51
1.22
2.67
3.13
2.88
0.84
1
3
6
9
12
24
36
-0.15 %
-0.17
-0.34
-0.11
-0.16
-0.34
-0.33
Panel B: Winner Portfolio
-1.30
0.07 %
-0.55
0.18
-0.63
0.36
-0.15
0.52
-0.16
0.70
-0.23
1.61
-0.16
2.40
1
3
6
9
12
24
36
-0.08 %
-0.68
-1.56
-2.78
-3.29
-3.22
-1.17
Panel C: Loser Portfolio
-0.52
0.07 %
-2.05
0.05
-2.55
0.37
-3.14
0.40
-2.80
0.73
-1.43
2.15
-0.34
3.94
-0.08 %
0.38
1.23
2.55
3.17
3.42
2.39
-0.52
1.11
2.24
3.43
3.77
2.76
1.41
1.28
1.51
1.77
1.82
1.92
2.54
2.55
-0.23 %
-0.35
-0.69
-0.63
-0.85
-1.95
-2.73
-2.48
-1.49
-1.66
-1.09
-1.18
-1.64
-1.60
0.85
0.37
1.47
1.16
1.38
2.33
2.83
-0.15 %
-0.73
-1.92
-3.19
-4.02
-5.37
-5.12
-1.32
-2.71
-3.91
-4.47
-4.47
-3.05
-1.91
Chapter 2
Trends and Sequences in Financial Performance:
A Test of Behavioral Theories
2.1
Introduction
Several studies document momentum in stock returns at horizons ranging from three to twelve
months and reversals at longer horizons.' The interpretation of these profits, particularly the longhorizon performance, has been widely debated.2 Some argue that these findings provide strong
evidence of market inefficiency while others argue these returns represent compensation for risk.
The desire of the former to add rigor to the inefficiency argument has led to the proliferation of
theories based on psychological biases. The purpose of this analysis is two fold: first, we identify and
put structure on a pervasive behavioral phenomenon that underlies many of these theories. Second,
we use financial performance to construct out-of-sample tests for the presence of this phenomenon.
Our research is motivated by the observation that behavioral theories are constructed to explain
observed phenomenon using a seemingly boundless set of psychological biases. This creates the
'See, for example, Jegadeesh and Titman (1993), .Jegadeesh and Titman (2001), DeBondt and Thaler (1985) and
DeBondt and Thaler (1987).
2
E.g. Fama (1998) and Ball, Kothari, and Shanken (1995).
56
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
57
potential for 'theory dredging'.3 Thus, assessing the predictive ability of behavioral hypothesis
using out-of-sample data is important. Furthermore, by identifying pervasive psychological biases
and forming empirically testable hypotheses, we can aid behavioral theorists to isolate the key
biases affecting asset-pricing. In this respect, we are motivated by the comments of Fama (1991)
on the usefulness of the single factor model. Fama argues the model "does the job expected of
a good model. In rejecting it, repeatedly, our understanding of asset-pricing is enhanced". In a
similar spirit, we attempt to refine behavioral asset-pricing explanations by testing a basic empirical
prediction that underlies many of these theories.
Our tests examine the relation between past trends and sequences in financial performance and
future returns. Specifically, we measure long-term and medium-term past performance using five
years of annual and one year of quarterly accounting data. We investigate the relation between
consistency of these past performance measures and returns both during and after subsequent
confirming and disconfirming accounting performance. For example, we test whether return momentum is related to the magnitude and consistency of quarterly net income growth in the prior
four quarters. We then examine momentum following an additional quarter whose performance
confirms or disconfirms the trend in the prior four quarters.
These tests are based on a fundamental aspect of human behavior. Individuals are thought to
make biased judgements under uncertainty because limited time and cognitive resources lead them
to apply heuristics (Hirshleifer, 2001). A central heuristic that underlies many behavioral theories
is representativeness (see Daniel, Hirshleifer, and Teoh (2002)). Representativeness can be defined
as the tendency for subjects to classify things into discrete groups based on similar characteristics.
Kahneman and Tversky (1974) note that by focusing in similarities, subjects can diverge from
Bayesian reasoning in a number of ways. First, subjects fail to consider base rates. For example,
they may think a rock is gold because of its color and weight and in so doing fail to consider the
low probability of finding gold. Second, subjects fail to incorporate sample size or the precision
3
See Rubinstein (2001), Hirshlcifer (2001), and Shiller (1999). According to Miller (1986), "That we abstract
from all these stories in building our models is not because the stories are uninteresting but because they may be too
interesting and thereby distract us from the pervasive market forces that should be our principal concern". Fama
(1991) uses the term 'theory dredging' to describe the practice for overfitting theories to empirical observation.
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
58
of qualitative information in their classifications and predictions. Therefore, they can confidently
believe two companies have significantly different financial prospects despite a limited sample of
prior performance. Finally, in their desire to maintain distinct categories, subjects making predictions fail to realize extreme observations are unlikely to be repeated. Thus after a history of
outstanding performance, investors are likely to be disappointed when future performance regresses
to the mean. In sum, representativeness implies sequences of past performance cause investors to
place a firm into a category, and form predictably biased expectations about future performance.
The centrality of the representative heuristic to behavioral theory can be seen by the number of
specific biases that exemplify its logic.4 For example, in the "halo effect", individuals observing
a positive characteristic of a firm form expectations about other characteristics. The "clustering
illusion" and the "hot hand" misconceptions predict that investors seeing a sequence of repeated
returns will incorrectly characterize them as following a trend. Consistent with this bias, Sirri and
Tufano (1998) find increased flows into mutual funds with exceptional past performance. Lakonishok, Shleifer, and Vishny (1994), cite the base rate bias and the tendency of investors to make
categorical predictions to explain the profitability of contrarian investment strategies.
Other behavioral biases concern how representative categories are established in people's minds.
For example, the availability bias (see Kahneman and Tversky (1974)) concerns the ways subjects
gather decision relevant information. According to the availability bias know as "salience", more
recent and more dramatic events have a greater effect on the subject's priors. Similarly, "framing"
can determine how an individual utilizes relevant information to make a decision (see Kahneman
and Tversky (1986)). Based on the above discussions, we conclude that representativeness can
lead investors to extrapolate incorrectly existing trends (see Daniel, Hirshleifer, and Teoh (2002)).
Such extrapolation can lead to reversal once incorrect conclusions are corrected. Recent behavioral
theories have tried more rigorously to examine the interaction between momentum and reversal by
4
For a more extensive exposition and discussion of the predictable errors in judgement related to categorical
thinking see Mullainathan (2001) and Rabin (2001). Mullainathan's model differs by having agents that do not
update beliefs gradually, but instead shift their beliefs from category to category more abruptly. Throughout this
chapter, we refer to "categories" as those groupings that feature in the representativeness heuristic, and not the
groupings in his model.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
59
relying on representative thinking by investors. For example, Barberis, Shleifer, and Vishny (1998)
(hereafter BSV) explicitly combine representativeness with conservatism in a model where investors
try to predict a firm's earnings., Representativeness causes investors to classify firms into two sets
based on their operating performance: trending or mean-reverting. Investors change their beliefs
about a firm's class based on the observed pattern of results. If the pattern is steadily growing,
they are more likely to believe the firm is trending. If the pattern is volatile, they believe the firm
is mean-reverting. In fact, the true dividend-generating process is a random walk, which leads to
predictable biases.
Consistent with the model of BSV, Kahneman and Tversky (1974) note that consistency of past
data affects the formation of categories: "People express more confidence in predicting the final
grade-point average of a student whose first-year record consists entirely of B's than in predicting
the grade-point average of a student whose first-year record includes many A's and C's". Moreover,
the importance of 'salience' in category formation suggests that consistent performance is likely to
influence investor judgements because repeated, positive information should have increased 'availability'. Therefore, we examine whether the consistency of past performance in the subperiods that
comprise the overall trend has an effect on subsequent momentum/reversal.
Specifically, theory
suggests that if accounting performance is salient, it should be related to predictable returns. In
addition to the well-known earnings surprise literature, 5 other research on accounting data also supports this view. For example, Barth, Elliott, and Finn (1999) find firms with patterns of increasing
income have higher earnings multiples than other firms.
Aside from providing a new and unexplored context for testing theory, our results provide evidence on how investor expectations incorporate the times series of accounting results. Accounting
statements are arguably the largest source of information about firm performance that investors
possess. Thus any explanation of anomalies based on investor behavior must include them. We
separate firms by both the trend and the consistency of their past operating performance to see if
we can predict returns. This is meant to test theories about how investors use financial statements
to form their expectations about future performance. Our analysis of accounting data also adds to
5
See Ball and Brown (1968) and Beaver (1968).
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
60
the growing literature on the importance of different types of financial news and how presentation
of that news affects judgments.
We find no evidence that investors over-extrapolate the operating performance at long horizons.
Abnormal performance in the year after five-year periods of extraordinary high or low growth is
insignificant. We find some evidence that investors do underreact to a one-year trend in accounting
performance, but this phenomenon does not appear to be distinct from post-earnings announcement
drift. Finally, we find no evidence that the consistency of firm performance influences expectations.
Thus, investors do not appear to form biased expectations about future firm performance based
solely on patterns or trends in past growth. These results present a challenge to the entire class of
representativeness-based theories. While investors may form biased expectations about future firm
growth rates using information outside of accounting statements, to date behavioral theories have
not made this distinction.
The remainder of the chapter is organized as follows. Section 2 of the chapter discusses background
and research design issues. Section 3 covers the data we use and our specific test methodology.
Section 4 shows results, and Section 5 concludes.
2.2
Background and Research Design Issues
We directly examine whether various sequences of accounting results forecast momentum and reversal, at different horizons. We hypothesize that investors use accounting information to form
biased expectations of future growth rates. Because theory does not suggest which measure of
financial performance is most salient for investors, we measure financial performance in three ways:
sales, operating income, and net income. Furthermore, theory does not specify what horizon is
appropriate,
so we use condition on both 12 months and five years.
Tests of specific investor biases tend to followa defined procedure. First, researchers define a model
of investor expectations about dividends or earnings. Second, researchers specify a model of how
dividends truly evolve. Third, researchers derive the pattern of errors in expectations, and finally,
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
61
show that these predict patterns of stock returns. 6 To clarify our experimental design, these steps
are explained below.
We assume investor expectations are a function of the past four quarter (five-year) financial performance. Based on the representativeness heuristic, we assume that investors use a given sequence
of quarterly (annual) growth rates to place a firm into one of three categories:
1. Perceived high growth firms. These firms have the highest growth rates and most consistently
positive quarterly (annual) results relative to their peers. Firms in this category are expected
to continue growing, and are awarded a higher valuation than other firms.
2. Perceived distressed/low growth firms. These firms are characterized by a sequence of steadily
falling (and possibly negative) sequence of growth. This category is expected to continue to
display unfavorable relative performance and a low valuation.
3. Perceived mean-reverting firms. These firms are characterized by a cyclical pattern of growth.
Investors in such firms will likely believe that firm performance will revert in the near term.
They may therefore underweight recent public signals.
The high growth and distressed categories resemble the growth and value dichotomy that has
received much attention in research and the financial press. However, in the present case, the
overall trend and the subperiod sequence in financial performance are important. Other categories
may exist, but because theory provides little guidance for identifying additional categories we
focus on these cases to increase the power of our tests. We make limited assumptions about actual
underlying growth rates. They may be random or have seasonal drift. Our tests simply require that
actual growth rates contain more variability or are less predictable than investors expect. Returns
are correlated with the differences between expected and realized performance.7 For example, if
earnings expectations for the high (low) growth group are too optimistic (pessimistic) then future
excess returns will eventually turn negative (positive) as investors realize their classification is
6
See Bernard and Thomas (1990) and DeChow and Sloan (1997) for examples of this approach when testing naive
extrapolation hypotheses of earnings growth.
7
See Ball and Brown (1968).
_
_
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
62
incorrect.8 Thus, stocks in the distressed and growth categories will eventually produce return
reversal. Moreover, firms with more consistent prior performance should experience more reversal.
Consistent performance should be more salient and thus lead to more definitive classification by
investors and greater reversal when investors learn prior classifications were incorrect.
Firms generate momentum if they fall into the third category. In this case, investors seeing the
beginnings of a trend will discount it in the belief that performance will revert. As they are surprised
by future results, a drift in returns develops. BSV and Rabin (2001) rely on this form of belief in
the "law of small numbers" to generate momentum.
2.3
Data and Tests
2.3.1 , Testing Procedure
In this section, we outline our tests and review their underlying logic. In this analysis, we calculate
operating growth rates over two periods, one year (4 rolling quarters) and five years (using annual
data). As mentioned above, we use three measures of operating performance: sales, net income,
and operating income. Sales growth is computed as follows:
(St - S)/S
(2.1)
S is sales per share, t represents the ending period, and s the starting (base) period, over which
growth rates are calculated. Depending on the time horizon, s can be four quarters or five years
before t. One drawback of sales per share is that it may have little relation to underlying profitability
and the relation may vary across firms and industries. Therefore, if investors focus on profitability,
sales will not measure variation in the key driver of stock prices. The second measure of operating
performance is change in net income scaled by base period assets:
(NIt - NI)/A
8
(2.2)
Lakonishok, Shleifer, and Vishny (1994) and La Porta, Lakonishok, Shllleifer,and Vishny (1997) use this model
of extrapolation.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
63
NI is net income per share in the period and A is assets per share. This operating growth statistic
is roughly equivalent to change in the return on assets. Assets are used in the denominator to scale
the income of large firms and to enable computation of changes in periods where net income is
negative. Simple earnings-per-share growth measures would not be meaningful in these contexts.
The third measure is exactly as above, but utilizes operating income after depreciation per share
instead of net income per share:
(Ot -OI)/A 8
(2.3)
Operating income is used to prevent large one-time items from affecting our measure of operating
performance. In this analysis, "operating", "financial", and "accounting" performance all refer to
growth as captured in accounting statements.
Finally, we provide comparable results for a control group of firms sorted by prior returns. High
growth (top quintile) firms by the return metric are typically called "winners", and low growth
(bottom quintile) firms are called "losers". Empirical evidence that five-year winners underperform
five year losers (see DeBondt and Thaler (1985)) and one-year winners outperform one-year losers
(see Jegadeesh and Titman (1995)) provides much of the motivation for research into apparent under
and over-reaction of investors to information. Therefore we provide evidence of momentum and
reversal associated with prior returns for comparison with predictability after contemporaneously
measured operating performance.
Using these operating growth statistics, our goal is to answer three questions:
1. Is medium-term (one year) operating performance associated with momentum in returns?
2. Is long-term (five year) operating performance associated with reversal in returns?
3. Does the does the consistency of prior operating performance affect the bias in investors
expectations and therefore explain reversal or momentum in returns?
The most common behavioral explanation for momentum is that investors underreact to information, causing a lag of many periods for it to be reflected in returns (e.g., Bernard and Thomas
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
64
(1990)). Because returns reflect information for a variety of sources, our tests of financial performance can be viewed as an attempt to isolate the particular type of information that investors are
slow to assimilate. Cohen, Gompers, and Vuolteenaho (2001) argue that investors underreact to information about cash flows, or that part of the return which predicts future operating performance.
A related but simpler idea is that investors underreact to past operating performance, as opposed
to information about purely future growth (Daniel and Titman (2001) provide evidence that the
latter is important to investor misperception). Under this framework, investors do not fully digest
accounting performance, and so changes in operating growth should forecast returns in the same
direction over the next several months. Our test is based on this hypothesis. As mentioned above,
we postulate that investors are more likely to underreact if they classify firms into a mean-reverting
(or at least unlikely to trend) category.
We select all firms each calendar period with data on the specified measure of operating growth.
In the case of our momentum tests (the first question), we use all firms in the Compustat quarterly
database from 1976-2000 that had at least seven past quarters of data. 9 Returns for each firmquarter observation are computed after the end of each March, June, September, and December
and operating performance is computed based on data from the previous calendar quarter. The
one-quarter lag is designed to ensure that performance data is publicly available prior to return
computation. Our operating performance sort is based upon the trend of past growth from one
year to the next, where year is defined as a non-overlapping four-quarter period. To be precise,
operating performance is computed by taking:
[(St+ St-i + St-2 + St-3) - (St-4 + St-5 + St-6 + St-7)]
(2.4)
(St- 4 + St-5 + St- 6 + St-7)
for the sales-per-share measure, and
(NIt + NIt_1 + NIt-2 + NIt-3) - (NIt-4 + NIt_5 + NIt-6 + NIt- 7)
(2.5)
At-4
for the net income measure. At- 4 represents assets four calendar quarters before the current quarter
t. A similar method is used to compute the operating income measure. Firms in the top quintile
9
This requirement provides the data necessary for the consistency tests discussed below. To control for seasonality
of performance, we calculate growth relative to year-ago numbers each quarter.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
65
by growth in a given quarter are labeled "high growth" firms, and those in the bottom quintile are
"low growth" firms.
To test whether investors underreact to improvements in operating performance (question one
above), we construct a trading strategy that involves buying and selling the equal-weighted portfolios of high and low-growth firms, respectively. We then hold these portfolios without rebalancing
for three, six, nine, and 12-month horizons and refer to returns produced by this strategy as "longshort" returns. Positive profits to such a strategy imply that investors do not rationally adjust to
information contained in publicly available accounting statements, and therefore returns continue
to drift in the same direction as past operating growth. To see if "accounting momentum" effects
are separate from other known predictors of returns, we regress the time-series of returns on the
the three Fama-French factors (the market return- Riskfree rate, and the HML "value" factormimicking and SMB "size" factor-mimicking portfolios). Pure return momentum should simply be
a noisy proxy for accounting momentum under the hypothesis that investors underreact to past
operating growth. To account for return momentum, we add a fourth momentum factor (UMD
or "up minus down", courtesy of Ken French, from his website). Our approach is similar to the
decomposition that Carhart (1997) used for mutual fund returns by style. Finally, stocks for our
comparison return momentum portfolios are selected each calendar quarter end based on returns
over the past twelve months.
We examine whether operating trends influence investor expectations and future returns over longer
horizons (question two above), in the similar manner. Returns can reverse at longer horizons if the
representativeness bias leads investors to incorrectly extrapolate past growth rates too far into the
future. La Porta, Lakonishok, Shleifer, and Vishny (1997) provide evidence for this view. To test
this idea, each year from 1975 to 2000 we select all firms that had at least five years of past data
from the Compustat Annual data file.'0° Each June, we rank these firms by year t to t - 5 growth
(using numbers available from fiscal year-ends falling in the previous calendar year), and form
l°We select this date range to match the quarterly sample period. Note that we require firms to have five years of
past data. This reduces survivorship bias resulting from Standard and Poors's incorporation of back-filled data on
existing NASDAQ stocks in the mid-seventies.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
66
portfolios of the top and bottom quintiles by each measure as before. We hold these portfolios for
the next three to 12 months. We then examine the returns of the long-short strategy for evidence
of reversal related to prior operating performance. Such evidence implies that investors extrapolate
past growth rates too far into the future, and suffer subsequent surprise or disappointment. As in
the quarterly tests, we control for size, book-to-market, and price momentum effects using timeseries regression. Finally, our comparison group of firms sorted by past five-year returns is selected
each December and held over the next twelve months.
To test for the existence of effects related to consistency of past performance, we rank firms within
each performance quintile by consistency of performance in the sub intervals of the ranking period.
For the one-year (five-year) sample, we examine performance in each of the four quarters (five
years). Consistency ranks for a given firm are determined by the number of quarters (years) in
which that firm experiences above median year-on-year growth relative to the entire cross section
of firms. Those top growth quintile firms with above median growth in all four quarters (five years)
within the past one-year (five-year) period are labeled "consistent" growers. Those top quintile
firms with only two-or-fewer quarters (three-or-fewer years) of above median growth are called
"inconsistent" growers. We repeat the process for bottom quintile firms, so that firms with four
quarters (five years) of below median growth are "consistent" and firms with two-or-fewer quarters
(three-or-fewer years) of below median growth are "inconsistent"''." We then observe returns for
these subgroups. We perform this two-way consistency/performance sort because investors are
expected to definitively categorize firms as high and low growers when past performance is more
consistent. If growth is inconsistent, the firm is more likely to fall into a mean-reverting category.
Therefore we expect greater momentum in firms with inconsistent growth patterns in the short-run
as investors realize that growth is not very mean-reverting. We also expect more reversal in firms
with consistent growth patterns in the long-run as investors realize growth is not trending.
Another way to determine if the past pattern (as opposed to trend) of growth affects investor
expectations is to examine investor reaction when a firm's recent performance contradicts past per' We use these consistency categories to ensure adequate observations within each porfolio after a two-way sort.
Changing the number of periods used to define a consistency category does not alter the results.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
67
formance. For example, more consistent high growth firms with a subsequent low growth quarter
will disappoint investors, but these investors may be slower to assimilate information that contradicts a prior categorization. There should be less investor resistance to new information for firms
that are less consistent. Similarly, consistent firms should experience less drift after a quarter with
confirming growth, while inconsistent firms should experience more as investors slowly revise their
expectations to include the possibility of a trend.
To test the relation between the consistency of prior performance and the investor reaction to contradictory and confirming results, we begin with the two-way sort by growth quintile and consistency.
Within these groupings, we observe long-short returns for consistent firms during and after confirming or disconfirming quarters (years). Under the hypothesis that investors form expectations on
past performance, consistent firms should experience less momentum after an additional quarter of
confirming performance because the new information fits the well-established category into which
investors have placed the firm. In addition, consistent firms should experience more reversal after
an additional quarter of contradictory information, because investors resist the implication that
the firm is not trending.
2.3.2
Data Summary
Summary statistics of the one-year and five-year data sets are shown in Table 1. Counts of the
five-year sets of stocks in selected years are displayed in Panel A of Table 1. Time series average
proportions of the set of firms with five-years of past data that fall into consistent and inconsistent
groups are shown in Panel B, as is average market value in millions. Note that the total number
of firms falling into high or low growth quintiles in a given year is 20%, by construction.
As one can see, there are roughly equal numbers of firms with five years of past reported sales, net
income over assets, and operating income over assets. Of these, roughly five percent fall into the
most consistent high and low growth groups on average each year, although more firms fall into
consistent low growth by operating income than fall into consistent high growth. About twice as
many firms fall into the inconsistent high and low growth groups. Consistent high and low growth
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
68
firms are much larger and smaller in size, respectively, than their inconsistent counterparts.
Of
course high growth firms by returns are much larger than low return firms.
Panels C and D of Table 1 show the same summary statistics as Panels A and B, but for the
set of firms with seven quarters of past data (for calculations of four year-on-year growth rates).
Once can see that there are many more firms, in general, by this criteria (since it does not require
firms to have existed for such a long period). Also, the proportion of firms falling into consistent
groups within high and low growth classifications is larger, but roughly balances the inconsistent
groups, for each of the four growth measures. Finally, the average sizes of firms falling into high
and low growth groups, by consistency, are much less extreme than those for the five-year groups.
In summary, our data sets are relatively balanced between consistent and inconsistent groups. For
the five-year set, consistent high growers are much larger in size than inconsistent high growers,
yet they make up a smaller proportion of stocks. Consistent low growers are very small, but not
much smaller than inconsistent low growers. For the one-year set, the same patterns hold, although
consistent firms are relatively more common than inconsistent ones, suggesting that performance
is the autocorrelated over shorter time frames. There is less dispersion by size among consistency
groups for this set.
Table 2 shows how our consistency statistics are related across firms. Panel A reports the time
series average of the cross-sectional correlation of firm consistency ranks (across the four measures of
growth), market values, five-year growth rates, and future returns. One can see that the consistency
statistics are fairly correlated across measures, but they are far from perfect substitutes. All are
correlated with past returns and market values. The OIAD measure is closely related to the net
income measure. However, it is striking how much less the operating measures of consistency and
growth co-move with the return based ones. This points to the fact that there is a difference
between return-based predictability and accounting predictability. Panel B tells a similar tale,
although with the one-year set.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
2.4
2.4.1
69
Results
Results for Short-Horizon Momentum
Table 3 displays how trends in operating performance over the past rolling seven-quarter window
predict future returns. It shows the returns to an equal-weighted portfolio of top-quintile growers,
bottom-quintile growers, and a strategy of buying the former and shorting the latter. Raw returns
are on the left, followed by three-factor time series regression constants (middle), and four-factor
regression constants (right). Portfolio performance is grouped in rows by growth metric, beginning
with sales per share growth, followed by change in net income divided by base year assets, and
change in OIAD divided by base year assets.
As mentioned before, we present returns to a pure price momentum strategy for comparison at the
bottom of the table. We see the previously documented momentum effect in returns. Firms with
high past 12-month returns (top quintile) exhibit significantly higher future returns over the next
3, 6, and 9-month horizons than firms with low past 12-month returns. This outperformance is
robust to three-factor controls, consistent with what we know about price momentum. Of course,
once we also include a fourth momentum factor (UMD) on the right-hand side of the regression,
unexplained momentum strategy profits are indistinguishable from zero.
When one looks at returns for the three operating measures, one can see that past momentum in
accounting performance also predicts future returns. One appears to be able to earn around four
percent a year by following a strategy of buying past high growers and selling past low growers
every calendar quarter. This result holds for whether income is defined as operating income or
net income. Furthermore, the returns generated by a long/short investment strategy based on an
income sort are similar in significance and magnitude to those generated by a returns-based sort
and, the time-series regression constants are significant event when they axe computed using a four
factor model that includes momentum. This result indicates that accounting-based momentum
effect is separate from the return-based momentum effect.
We find weaker return prediction results when past sales growth is used as the as the basis for
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
70
portfolio formation. Both the magnitude (one percent or less) and the statistical significance of
these long-short returns are small. This could be due to the limitations of sales growth as a measure
of investor gains discussed above. On the whole, this accounting momentum is unusual, but not
surprising if one subscribes to the view that investors do not fully digest information. We interpret
these results as confirming underreaction in investor behavior on out-of-sample data.
The accounting-based momentum results could be due to post-earnings-announcement drift documented by Ball and Brown (1968), Foster, Olsen, and Shevlin (1984), and Bernard and Thomas
(1989). We re-run the long-short strategies for the rolling four-quarter horizon, as in Table 3, but
eliminate all stocks that would have also been selected by an earnings-surprise long-short strategy.
This second filter screens out stocks that had returns in a three-day window around the earnings
announcement in the top or bottom quintiles of all such returns for all stocks within the latest
quarter.
Completely eliminating overlapping stocks is an extreme a way of seeing if accounting
momentum is distinct from previous findings. Both earnings surprise and accounting momentum
could be theoretically driven by the same underreaction to operating results, and therefore would
complement each other. However, using this harsh filter may let us see how much investors underreact to short-term surprises (seen in an earnings surprise filter) vs. to longer horizon trends (seen
in the accounting momentum results).
When we eliminate the stocks that would also have been selected by an earnings surprise strategy based on announcement date returns, we find that the accounting-based momentum al but
disappears. Table 4 displays returns for the same categories as Table 3, but for the filtered set
of stocks. The first two groups of columns suggest that stocks exhibit a persistent return in the
same direction as operating performance for the next quarter. Again, this is mainly for the NI and
OIAD measures. This suggests that investors underreact to longer-term trends in performance in
addition to being repeatedly surprised by earnings announcements. However, the drift is generally
less than in Table 3, and of much shorter duration (one quarter only). When we control for pure
price momentum, we find that accounting momentum fades away entirely. This reveals that in
the absence of earnings surprise in the most recent quarter, there is little evidence that investors
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
71
underreact to medium-term trends in performance.
2.4.2
Results for Long-window Reversal
At five-year horizons, we confirm return reversal of DeBondt and Thaler (1985). High return firms
over the past five years ending each December underperform low return firms for a substantial
amount over the next twelve months beginning in January of the next year. Controlling for the three
Fama-French factors cuts the return difference roughly in half, but the gap remains economically
sizeable.
However, we are unable to document reversal in returns when portfolios are formed based on accounting performance. For the net income increase measure, we find that subsequent returns measured
over 3, 6, 9, and 12-month horizons are actually higher for the highest income growth quintile compared to the lowest income growth quintile. However, the results shown in Table 5 indicate that
this difference is not consistently significant across controls. The same holds for the OIAD increase
measure, as well as for sales per share growth. Given these generally inconsistent subsequent returns
across horizons, we conclude that there is no evidence that simple trend extrapolations of operating
performance play a role in investor expectations. If investors are overreacting to information and
causing reversal at a later date, they are not overreacting to accounting information. 12
2.4.3
Results for Consistency of Performance and Subsequent Returns
Our next tests examine the effect of the pattern of prior performance on subsequent returns.
Specifically,we investigate whether consistent prior accounting performance is associated with less
momentum and more reversal than inconsistent prior performance over the five year and four
quarter horizons. Our methodology for sorting high and low growth firms by consistency was
detailed in the previous "Testing Procedure" section. To summarize, we classify all stocks in each
12Daniel and Titman (2001) report a similar disconnect between past fundamental valuations and future returns.
They do, however, document the consequently stronger negative relationship between expectations of future growth
(encapsulated in that part of market valuation that is orthogonal to current fundamental value) and future returns.
They attribute this to investor overreaction to intangible information, rather than tangible (i.e. fundamental or
operating) information.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
72
quarter as above or below median growers (relative to the entire cross section of firms). Firms are
classified as "consistent" if they had growth of the same direction, above or below median, as the
five-year (four-quarter) trend in every one of the past five years (four of the past four quarters,
year-on-year). Firms are classified as "inconsistent" if they had growth in the same direction as
the five-year (four-quarter) trend in only three or fewer of the past five years (two or fewer of the
past four quarters). For comparison, we repeat the same consistency separations using stock-price
returns over the same two horizons.
To recap the general idea of the various consistency-based hypotheses, investors form stronger
expectations about the likely path of a firm's future growth is they see very consistent performance
as opposed to more volatile results. In the case of longer horizons where overreaction is the featured
hypothesis, investors are more likely to over-extrapolate trends that they see consistently. In the
case of shorter horizons where underreaction is thought to play a role, they are more likely to spot
trends if they see consistent evidence of it in the recent past, and therefore be less surprised. This
can be applied to our four-quarter and five-year strategies. A strategy of buying and selling fiveyear consistent high and low growth firms will yield bigger losses than one applied to inconsistent
growers. A strategy of buying and selling four-quarter high and low growth firms will achieve higher
profits with inconsistent firms than with consistent ones.
The results of the short-term (one-year quarterly) and long-term (five-year annual) tests are presented in Tables 6 and 7, respectively. Table 6 shows the results when consistency of prior performance
is measured on a quarterly basis over the prior year. In this table, rows labeled "consistent" ("inconsistent") display returns to buying and selling the top and bottom quintiles of consistently
(inconsistently) performing firms. These rows therefore show how much return drift occurs for
consistent and inconsistent stocks. The rows labeled "difference" show the gap in returns between
the more and less consistent long-short strategies. The rows are organized into groups based on
the sales growth, net income, and OIAD measures of growth. Firms selected by returns are used
for comparison in the bottom grouping as before. Returning to the predictions of some theories,
all the entries in the "difference" row should be negative. This is not the case. Table 6 shows that
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
73
firms with inconsistent prior performance have less return momentum than those with consistent
performance, contrary to the predictions of the consistency-based theories. In line with Table 3,
there is no drift for firms sorted by sales per share growth. However, even the drift that exists in
the other two accounting-based growth measures is greater in consistent stocks. To be sure, the
difference between consistent and inconsistent drift is rarely statistically significant across horizons
and measures. It is always positive, however, which implies that investors underreact even more
to trending performance that they have seen repeatedly than to flash-in-the-pan results. This is
contrary to the idea of underreaction.
In Table 7, we repeat the tests in Table 6 using annual performance over the previous five years
to measure consistency. Again, according to representativeness-based theories, all entries in the
"difference" rows should be negative since investors should form extremely biased expectations
about future performance for the consistent growth set of stocks. As with the quarterly measurement period, this is not the case. For almost every horizon, and every measure of growth, the
post-portfolio formation returns of consistent firms are statistically indistinguishable from those
of inconsistent ones. In Table 5, we found no evidence that investors used accounting trends to
form biased expectations about future growth. Table 7 reaffirms that the past pattern of operating
results does not matter either.
In sum, these tests for the effects of consistency find no evidence of representativeness or "law' of
small numbers"-based behavioral biases by investors. After using a number of measures of firm
performance, and looking across two different conditioning horizons, we find no sign that the pattern
of performance influences investor expectations.
As an additional test of whether or not the consistency of results affects expectations, we focus on
returns after a marginal period that confirms or disconfirms previously consistent (or inconsistent)
performance. What are the predictions of behavioral theory?
1. In general terms, more consistent growth causes investors to form strong opinions that future
growth will be in the same direction. Confirming signals do not lead them to change their
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
74
opinion. Disconfirming signals are initially discounted, leading to reversal.
2. Less consistent growth causes investors to believe that future growth will be mean-reverting,
or at least lack a trend. Therefore confirming signals will be discounted, leading to drift, as
investors bet that the results were a fluke. Disconfirming signals do not lead them to change
their initial opinion.
The disconfirming signal in our test is an extra quarter (year) of operating growth in the opposite
direction as the trend for the past four-quarter (five-year) data set. Recall that we measure consistency over both past one-year and past five-year periods. Note that both of the above predictions
could apply to each horizon. However, drift is stronger when conditioning over four-quarters and
reversal is more evident over five years. Thus, point 1 is potentially a better description of what
happens over five years, and point 2 should apply more to one year.
Let us examine the one-year period in more detail. Consistency hypotheses suggest that steadier
growth trends make underreaction less likely. Marginal signals that confirm the trend in investors'
minds should have little positive effect. Marginal signals that contradict the trend could cause
several periods of reversal, although this is not likely to be the case since there is little empirical
evidence of reversal at this horizon. For inconsistently growing firms, the situation is reversed.
Investors should have less strong opinions about future growth trends (in fact, in some models, they
expect them to mean-revert). Marginal trend-confirming quarters herald the start of a period of
underreaction as investor expectations slowly adjust to information revealed by the trend. Marginal
trend-disconfirming quarters don't tell investors anything new. Therefore, inconsistent firms should
display more drift that consistent ones after confirming signals. They should also display less
reversal after disconfirming signals, if reversal occurs at all.
Similar logic can be applied to the five-year set. Consistency hypotheses suggest that steadier
growth trends make reversal more likely. Marginal signals that confirm the trend in investors'
minds should have little positive effect. Marginal signals that contradict the trend could cause
several periods of reversal as investors adjust their expectations. For inconsistently growing firms,
the situation is reversed. Investors should have less strong opinions about future growth trends
CHAPTER 2. TRENDS AND SEQUENCESIN FINANCIAL PERFORMANCE
75
(in fact, in some models, they expect them to mean-revert). Marginal trend-confirming quarters
are less likely to cause investors to over-extrapolate performance. Marginal trend-disconfirming
quarters don't cause as much adjustment in expectations.
Therefore, inconsistent firms should
display less reversal that consistent ones after disconfirming signals. They should also display more
drift after confirming signals, although this is less likely since little drift has been found at this
horizon.
If one takes the specific predictions together, one can formulate the following test. We form a
long-short strategy of buying high growth firms and selling low growth firms for different subsets
of stocks. This strategy should be less profitable for consistent firms than for inconsistent ones
during a confirming period, no matter the conditioning horizon. It should be less profitable for
consistent firms than for inconsistent ones during a disconfirming period, no matter the horizon.
Furthermore, such profitability patterns should continue to hold in subsequent periods.
We consider the one-year horizon first in Table 8. There are four panels in the table. Panel A
shows results for firms sorted on sales per share growth. Panel B shows firms sorted by NI/A, and
Panel C shows firms separated by OIAD. Finally, Panel D displays comparable results for firms
selected by past returns. Within each section, we display long-short strategy profits to firms during
("return in qtr. t") and after they experience a confirming and disconfirming quarter (subsequent
returns over 3, 6, 9, and 12 months). Columns are grouped together to show raw returns, 3-factor
time-series regression alphas, and 4-factor alphas. The rows of each section correspond to profits
for all firms with relevant data ("all"), consistent, and inconsistent. Te
"difference" row shows
the gap in long-short profits between consistent and inconsistent sets over time. According the
predictions of consistency based-theories as outlined above, this row should be negative in every
case.
Before discussing the difference row, we note that confirming accounting performance always generates contemporaneous positive returns and disconfirming performance generates negative returns.
This is to be expected. However, note that for the NI/A and OIAD measures, both these returns
trend for up to a year after the marginal quarter. The returns are also fairly robust to 3- and
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
76
4-factor adjustments. The results are more dramatic after disconfirming quarters, which generate
losses of anywhere from -2% to -6% at 12 months. This highlights the existence of some accounting
momentum documented in Table 3, although we found that to be closely tied to earnings surprise
in Table 4. For the sales per share measure, only disconfirming information causes persistent and
economically sizeable reversal (-5% to -6%), which is also in line with our previous results. By
contrast with the operating growth measures, a single quarter of confirming returns does little to
affect subsequent returns.
We next examine whether consistency of prior performance affects the magnitude of this momentum
and reversal in the face of marginal information. In general, the results do not show any difference
in returns. Hardly any of the entries in the difference row are statistically significant at generally
accepted levels, and point estimates are often positive. Surprisingly, there is even little difference
in the marginal quarter, where one would expect new information would cause dramatic changes in
expectations. Three- and four-factor adjustments do not change this conclusion. Again, we confirm
the results of Table 6. There is no evidence that the past pattern of returns causes investors to
form biased expectations about future performance.
We run similar tests on the five-year sets of stocks, the only difference being we consider the effects
of a marginal year of operating performance on concurrent and subsequent year returns, instead
of quarters. The results are shown in Table 9. We do not find evidence that return in the year
after a confirming (or disconfirming) year is related to prior accounting performance. 13 We also
do not find evidence that focusing on consistent performers increases this reversal effect following
a disconfirraing year. The statistical weakness of the results can be attributed to having fewer
observations, as well as the long annual time frame over which we measure returns as a response to
accounting information. The point estimates are often positive, contrary to what we would expect
if consistency mattered for expectations. Note that the numbers in the difference row are often not
the exact arithmetic difference between consistent and inconsistent groups, since there are some
periods when we lack inconsistent firms that also have disconfirming marginal years.' 4
' 3 As in Table 5, our results do confirm the existence of significant reversal when prior performance is measured by
returns. As before, we find this is not the case for firms sorted by accounting performance.
14Lack of observations is sometimes an issue because of the long time frame over which we measure returns. For
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
77
Overall we interpret cur consistency and subsequent performance tests as providing little evidence
for behavioral effects related to representativeness. In general, though we have tried to document
an effect across a variety of performance measures and horizons, the pattern of prior performance
does not seem to add significantly to momentum and reversal documented by prior research.
2.5
Conclusion
Many stories about investor behavior rely on some form of the representativeness heuristic, the
rule that investors use to judge if an outcome belongs to a category based on how similar it is to
the category. This rule is sometimes inaccurate, and leads them to form biased expectations future
outcomes. In a typical financial model, investors mentally misplace firms into various groups based
on the past performance, and are subsequently surprised or disappointed in predictable ways. This
is reflected in returns. We use accounting data to test the idea that investors classify firms into
groups. We use trends and sequences of accounting performance to separate firms into high and
low growth and to divide them by consistency of growth patterns. The advantage of this approach
is that we use a specific, traceable source of information to model possible investor categories in
a simple and straightforward way. Furthermore, our approach provides out-of-sample tests of the
idea that investors under or over-react to past information. Finally, we use diffrrent horizons and
growth metrics to allow for different information investors could use.
We find evidence of multi-month momentum in returns after accounting performance. However,
this is substantially reduced when we account for earnings surprise effects. We find no support
for multi-year reversal related to accounting performance. Finally, we find little evidence that
conditioning on consistency of past growth rates improves return predictability. It appears that the
sequence of past accounting performance is not related to future returns, and therefore is unlikely
to bias investor expectations.
some difference rows, we require long-lived firms that continue to survive for a marginal year and display very specific
patterns of results. There is no clear way to avoid this problem, however, since we have already defined growth and
consistency to assure adequately diversified portfolios in Table 5.
CHAPTER 2. TRENDS AND SEQUENCES IN FINANCIAL PERFORMANCE
78
Overall, these results suggest that multi-month momentum and long-term reversal are not due
to investors categorizing stocks incorrectly. They do not seem to extrapolate the growth rates of
firms too far into the future. Nor do they seem to underreact to incipient trends in performance.
The past pattern of performance has no bearing on how likely they are to be surprised by future
performance. All of these conclusions cast doubt on the representativeness heuristic-based theories
of bounded investor rationality.
One could conclude that representativeness has no place in describing investor behavior. However,
the predictability of returns remains an interesting and problematic fact potentially at odds with
fully efficient markets. Another lesson is that we need better models of human behavior. It is
possible that investors do indeed think in categories, and that we have not found the correct'
categories, metrics, or horizons. Yet it is also clear that any biases they have are not as simple' as
those found in most behavioral models. Our point in documenting the latter is that researchers
need to refine such models if they hope to describe reality better.
CHAPTER 2: TABLES AND FIGURES
79
Table 1: Summary Statistics
This table shows some summary statistics of the firms used in the analysis, for selected periods.
Panel A displays counts of firms with sufficient Compustat and CRSP data to compute five-year
past returns and five-year past growth rates for three measures of operating performance. "Sales"
refers to the growth rate of sales per share. "NI/Assets" refers to change in net income per share,
divided by base year assets. OI/Assets" is a similar measure, but uses operating income after
depreciation in the numerator. Panel B shows the average measures of performance and market
value of firms across the samples shown in Panel A. These firms are broken out by consistency
and growth performance. "Consistent" firms have growth consistent with the five-year trend in each
of the past five years. Inconsistent" firms have growth consistent with the five-year trend in three
or fewer of five years. Panels C and D correspond to Panels A and B, but show counts and
averages for the set of firms with at least four quarters of past seasonally adjusted growth (i.e.
seven quarters of past data). In the quarterly case, "consistent" firms have growth consistent with
the one-year trend in each of the past four quarters. Inconsistent" firms have annual growth
consistent with the one-year trend ins two or fewer of the past four quarters. Consistency is
measured by comparing the firm's growth in the period to the median growth across all firms.
Panel A: Observations Meeting 5 Year Annual Data Requirements
Year
Number of Firms with 5 Years of Past Growth
Returns
Sales
NI/Assets
1975
1862
1559
1561
1552
1980
3997
1874
1883
1874
1985
4055
1724
1738
1726
1990
4583
1765
1777
1769
1995
4954
2140
2211
2169
2000
5396
2596
2648
2643
O/Assets
Panel B: Percentage of Observations and Market Value by Annual Category
Category
Consistent
high growth
Inconsistent
high growth
Inconsistent
low growth
Consistent
low growth
Time-Series Average % of Firms with 5
Years of Past Growth for:
Returns
Sales NI/Assets OilAssets
3.8%
5.5%
4.1%
4.7%
Market Value
Returns
3,469
Sales
1,712
Nl/Assets Oi/Assets
2,736
3,344
7.1%
6.5%
8.5%
10.5%
913
714
800
822
6.9%
6.9%
13.3%
2.8%
87
792
416
376
3.7%
5.0%
1.2%
17.2%
90
430
516
477
_
CHAPTER 2: TABLES AND FIGURES
80
Table 1, continued:
Panel C: Count of Observations Meeting 1 Year Quarterly Data Requirements.
Year
Number of Firms with 7 Quarters of Past Data
Returns
Sales
NI/Assets
1976
4898
1955
1948
455
1980
4495
2175
2187
1457
1984
5553
3716
3714
2033
1988
6310
4119
4184
2731
1992
6068
4188
4237
3337
1996
7630
5392
5489
4285
2000
7141
4908
5019
3833
OI/Assets
Panel D: Percentage of Observations and Market Value by Quarterly Category
Category
Consistent
high growth
Inconsistent
high growth
Inconsistent
low growth
Consistent
low growth
Time-Series Average % of Firms with 7
Quarters of Past Data for:
Returns
Sales
NI/Assets OI/Assets
5.6%
11.8%
8.4%
9.9%
Market Value
Returns
1,566
Sales
862
NI/Assets Ol/Assets
969
1,015
4.2%
4.0%
5.9%
5.0%
314
451
369
299
3.3%
3.7%
6.7%
6.7%
141
366
403
382
5.7%
11.5%
7.1%
7.4%
149
383
290
356
CHAPTER 2: TABLES AND FIGURES
81
Table 2: Average Cross-Sectional Correlations of Firm Characteristics for All Stocks
This table shows the time-series average cross-sectional correlations between various firm
characteristics and returns. Panels A and B display Pearson correlations for the set of firms with five
years and four quarters of past growth rates, respectively. Variables definitions follow. The sample
consists of all firms with necessary Compustat and CRSP data between 1970-1999.
Panel A: Set of Stocks with 5 Years of Past Data
NIC
NIC
1.00
OIC
RETC
SC
Retl
Ret2
MVAL
NIG
OIG
SG
0.73
0.20
0.34
0.23
0.00
0.07
0.16
0.18
0.02
OIC
RETC
SC
Retl
Ret2
MVAL
NIG
OIG
SG
1.00
0.20
0.42
0.24
0.00
0.07
0.15
0.20
0.03
1.00
0.17
0.52
0.00
0.40
0.07
0.09
0.01
1.00
0.19
0.00
0.11
0.09
0.16
0.07
1.00
-0.02
0.26
0.12
0.15
0.04
1.00
-0.04
0.00
0.00
0.00
1.00
0.04
0.04
0.00
1.00
0.89
0.39
1.00
0.45
1.00
Panel B: Set of Stocks with 7 Quarters of Past Data
NIC
OIC
RETC
SC
Retl
Ret2
MVAL
NIG
OIG
SG
NIC
OIC
RETC
SC
Retl
Ret2
MVAL
NIG
OIG
SG
1.00
0.77
0.28
0.36
0.29
0.02
0.09
0.21
0.25
0.06
1.00
0.25
0.45
0.27
0.03
0.08
0.20
0.29
0.09
1.00
0.19
0.60
0.03
0.28
0.09
0.12
0.02
1.00
0.18
0.01
0.14
0.08
0.18
0.13
1.00
0.01
0.18
0.11
0.16
0.04
1.00
-0.03
0.02
0.01
-0.01
1.00
0.02
0.04
0.00
1.00
0.68
0.18
1.00
0.21
1.00
Variables Definitions
NIC
OIC
RETC
SC
Retl
Ret2
MVAL
NIG
OIG
SG
Consistency of past 5-year (4-quarter) growth in net income/assets
Consistency of past 5-year (4-quarter) growth in operating income/assets
Consistency of past 5-year January to December (calendar quarter growth over 4-quarters)
annual (quarterly) returns
Consistency of' past 5-year (4-quarter) growth in sales per share
Total cumulative return over the past 5 years (4 quarters)
Total cumulative return in the 12 months from July of the next year
Market capitalization in millions in December of year
Endpoint-to-endpoint growth rate in net income/assets over 5 years (4 quarters)
Endpoint-to-endpoint growth rate in operating income/assets over 5 years (4 quarters)
Endpoint-to-endpoint growth rate in sales per share over the past 5 years (4 quarters)
CHAPTER 2: TABLES AND FIGURES
0)
.oc
C:. c::
C1 e(
mL
?,'oT
82
-, 0 ),_
0 -.
,o 0c:mm0
aCa
CCN
I
I
W4
¢,Dco
U CO
C
0
.
(Dcr( ::."
- r.'~0
..O*C~
~ CC~
C~ v r40 ~:
O
=
46t I
N- :
T7 l~.
N N~
N"-l
7
Ni C
-C
C%? .
'1Cs
0
0
*
0)
(O
1
N
r-
C
C
.2
~
Ci
0)
C
.C
I
c0
O
C a
v.C;
t
)
0
00
0
Ut
,
v00
00
.
.
..
.
N
"'w t
l-
.0
m su '-
.rM
cn
- (0r
0
a
4
'
i/j
rN
_ . , ,,,_~
O
a)
V~c>
0
ao
'~
·
C:
'
- o,
-.
,.
o m-_,.::::m
~
.C 0 0
tD--- ,oc O)
'..0. .
m
..,
(00
0N
cmO-i
q
v- V
co
oO
,
WE
."
0
0
0r
~
0M
.v
-b-
~0
C
_
0E 3 .0
.s.-*-cuc.O
k-
'
_
.0
0
C
-
·
J
0 > 0
_
.
--
~~
-.-
*-
aO
I-
'a
a
5
co
0
(D5CD
-a
O 'A
f-u
oO
O (D-~:
EC
OOo;
- kr
0'- Mr
0. 0 ,-'CcV
u 0 0 0 '-,u ."-a"
0-,, , o --0.?e
e
aa0
0
L ~,
C)- D a(D ( >4)
3 V
.
a0
ocotP-
O
aa a
0'
0'-!
a I
a a
ij N
v-
'Pp
0
c 0'5~ .,
(D
0
EQ)
E : 3 C.-> 0 0 , M.
0 U.
(
=
N
o0
08
I
a CD
C
nCC
0
.1
0_
0
1.
cmC,
p
6.
03)
*DO)
CD O"M 1
0'
0 CN.N
N
O-
Ns
r.
C) N
ocr
Wc
Nl (0
coc)
aD
0)
CrC'
AO
6 6
a
ri
C
M
M.
0-
M(
W tI-
O0Cs
9
OwIt
'--
.
v-
0
;
T'-0)
O
o0 1
a
:"
v-
0
V)
0-
0c(N
O6'-
(D C
Cc0
C
C!',
I C"
E
0
I.
C
-0
z
C
C
0)'cC
04
5
0
0)
5U
I,
a
vir
Woct
0csd MIV:@
2
(D
cm
0
0
COeo~
0
C! c)
oNMO
i
,c ,
f
csi C
~ ~~~~~r
o)
so
r cio
co-
biitz
Cc
q c
o u:
1..
0
0)
.r
.2'
.C
0
0)
0
U
C
a
a..
V-
40 W.
Oa M
0)
0
0
a
M O
C C;
a.
0a
0
I Iat
- I
N
O
0n 0 (O
I0
to
0
A
Y
0
C
CO
r.-
00
co )
9t11
M ;CD
E
co
,
Z.C
Q tu
ao
o
6 C,
L~ N aO
n
.--'
M
0
o
F
z COC
'i
0
......
M.E.
(D
-~cm-
I
M
.~~~~~~~~~C
,_
I
C
cD
0M
M::___=
0
0
o~
0
0 '?- E
0
a
CriN
00
c: ~ e.
c)
C5 C;
o'e;-,
m r" oz
c
·
e
uMd
ODO)¢W
c-'-
V) It
,r:
0
oII
m
.r-cD
"
COCc
U.
L_.
.
ci'c. (-:I
2,
0
r'- - . : -'m
r C Eo
c.
_
'tt C')M
0
0
ts
-::__'.'.-.Ea.
0.
· r_
O>
,
o =. E-.10
E~
a,)
,WO
LT
~- ,m.,EEa
,
o oE
c0
aMC
00.
0I.
e
'5
0)
.r0)
a
c
2-
E
0
a
CHAPTER 2: TABLES AND FIGURES
83
i6j
('5
C N ti d
4)
0,v0)
rI-'t*
do
Z~
o
CY
co:c
@g
.- Z
c0o
e
> D
t
I
I
NLC)
C*4'?
,
0
O
<
t5)
N d
CO
CO
NC
N
C
00 CS
cq
C
C
S) (4
n
C
o
tO
0o t
tN
!tz t
v-
CD
C-ei
ro
hu) CV
-
ucO
I
CD
U)
a , 'C
I
(
iti
oe
e09
ccoce
co
14,
t co
h
ou
U
N' -
E
a,
0t: v0) r
I
t-
I(V) °i Cs
~-LO
1v
a
0
I
CUe
LLC NC')
sii N
rC
oi- t I
a
o
C
Ilen
.co
vJ
t I
r0 r
"tg
rte
0U6
ONOs
U0)
o
0
2a,o
VO
0a
cs Cr
V)
0
01
NDO
,cn-
t:
c8I
tstQ
co tc
4d
C
O
CU
C
00 m
P
L
m
tD
0
ts
Ui
0
a
P-
O
cn 1..
v C
P
dI'
i Ni
J
C>
m
.,- el::'
CDO,
oo
'--C'?
!2
o-)
(U CO
51)
t
Nr
*5
'
0
C
0
Nco
tN
C
0t(
r-4
NVB
'tO
0
0
q tC
'
..
N
a,
(PO
tD ti
3
N
Br
("
N"
Ci lt
Y.
w
0.
0)
Ut
toW
r
0(A
'N-N
vCD7 04
Nlt
U)wO
cou
to
0
a3
0
EC
(A
0
0
N CO
5i1
tD
r-
._
,V)
0tco
.0)
Co
I
J)
0w)
m
SOj t;O
tL'
_ o
CM
O fi4
CI
0U
C
..
ei
0
IR
co
coN
t
no
0',)
r-N
st
Ni
tn
cs
e
I
fi
O
_
0
06a.
2
CJ)
,
Zw:
l
-
2
332
C
.0
i0
(5
0SO
ci
0
20)
-c
it
I
.)
o
t"
co
C
.1
0.
2CL
z
rx
-C
0
E
0oa
tom
Ct
la
X
ta
I(A
(5
L
co
twi
CHAPTER 2: TABLES AND FIGURES
Caa,
CO
.
a
u 0
-
CN 0)
a"mEd -
C v- aC a4
C
cn I
aC
0 , c , ~'0)
B) O
C
C
.N
~E
CCa.G a,
w =m O
D'~
_
84
3
_.r
V0)C
.
0:1
J:z
2 a.!.
o
V
ll cM
o~~~2E
c4
)
t5
0 Cm 0O
a,
o
mo
S
a0
0
'.
0
.'
a0=C ccoa~E
~oem-'L~~M
'P
C' .J
a a
a
00
C C 0=
o
a a
-
a
.
I
0P,',o
I
ao
C
v
I
C
a
a0
a..
CO
cor'C)
N C'
C\J
a .
.
co
0
a,
Oh
o0
C; sa
Io,
LL
,:C
?
ma
aR.
..
N
d C
Q)O'
I-.
oa,
a-:;
.C
Ii- I0
c' 0)
c
0C
0 a) cor-.
id
a
N
a a
N
cCMU)s
v00
a a
0) co
Oa
o o
fo c
ci
t
.-co
do
a
.
4
oci
a -Da
4
0
0tona
a a
a,
(UcO
I
.
C
0.
(
._
IeS
C
0
C'
O
a, O0). .~)
a,"
(D
.a - U-r
,,cn u
ai
r-:
cMN
I
Sa0
C
'-
G c0a a,.
CC
o C
co
a
ICO
r1
0co
C
C__
co=,EC
-r_:a
'a,
00~
. --wE.C
CI
a,0
.: .,.
0
C
o C
.
a, rsaC
idi
,.
e
I
qa.
2
0
O
I c;I;
a0
.C a.
0.
u,
C
a,
an
a (0 N '1J:
0
0C.'
aa
2a
a V-
O c'
a C
i4 (N
0a
Oa
a C(0 O.M
cTr.. - rC
.a
a, a,.
- e
C
a
MMv.
.a
EE
C
pa
co 4
20
I
a
z
co 0a
Ea
.C1)cV
I
aC
Oco
a to
vC CY
.
(O 0C'9j V-r*
a)<a O~
a a pn
ct
I-
.
n
I-
Ia
0
U
am
V)
6V
I-
o
cv)(O
o
C OO
'- "PF
OC '
4v
,
0
u
0
-. 0
00-
rC
>>4)( 0~~R
.F0
x
0Ez
a.
'It
.0
a0
ce
a
.~Na
'~
rr
co -
a,
a
u)
~-c
=N
·
--- "
,)C"~
4.5
4
0
Ct
0
0
5
0a,S
D
a
I'
I-
m U)a
16 E0
E .10-=~c o.t~
v~~ m~ c 0: (a=0
LE
rrrUS
rcO
9 a ma (Rr -aL
-W
m 0E
d Em
w v-
04l C
co -
N "i to) ,
t(OA
Q
0U
-I
0
Eo) co(O
t
o
i-I-
0
'Q,
E
u," U~ra - a-8
",9Ca-a
a a~ 0a
00CL
-aVu
O
t::
E
5
g
:.ma,,E:)O
a a
5
2
r.2 '
5
C
.5
I
a
cS
co
O,
co
I41
C
.cr
CD
r-o (V
toN'
t- 04
0~r N
Ci
o
'v
C;
FL
0-i
0
C
:.
._
a
a-
0
S
C
I
O
0)
9
E
z
It
a
:E
..a,
to
0
1)
v WI)
_OIn
Cs
-r
V
8
r ","t P'.
P
C
a
c
a
o
0
Ca
2d
0L
'S
c
Nc,
- i
0c
.20,
-C
._
~2)
.C0
tm
3
P:
ca
V5
1.
&- C
%
CHAPTER 2: TABLES AND FIGURES
i,
i6
N
C
0o0
S v
,
0
a
C
c
o0 It
eic
oC
.I
. ICY _C:
. *
0
CD C' M
0
CO
CC
I
1-4
C
0
.C
S
to C0<O
''-
a
I
J
e:¢
Cm,U O)
t
I
d·
0'7,
4
i6
CD
a
0o.
C
0)
0 a) TC)
I
.
o
0
C
ra (0 N (V
m
CD
C
0
0
I-
0.
00
..
0.
r-
..
CU
1
.
oC
aC
L N
*i
tn
CO0 0
I
Cl
C N
_)
0)r.
I I . oC
·
O
CD
I,
O
CO
o
I
CD
U) V0.
vsi
Co C)
O
o ta to tofC.
v_-
0
E
z
ia
m
s
0
C
q.
a
ta
E
0
2
N0
=D ae
c ao
. I
CD L-
8
Co
j
U)
CD
c
IO
Cs
a)
0
I
( 0
0
C00
0
.
u)o
C; Cs
0
a
's(D
I
CN
C
Ci
OC
.
.
0
C C
C (O
to
-sC0
C
CO t)
C~i
niI ol ,C*,
I C
I< I
0) -C0
0mC
Oa
00
CO
CM0C V
4)
0
II.
to C) c
C.,
F, O
I
s
2
z
C
0
a
CD
I
0)0
0
85
CD
f
NV2
v- C
cO
I00td
0
0
u
L
m
e)
0)
M N )
I
i C0 0
5
r-
rC"'
L0
(N
Irtoc')
" v
cOC:
C14~
ki
14
Ub
0
cI.
a.
0.
o ~"
lt1 X
0u
w 9
0'
n
CO
2
0
cvw.
4.
m o-
i/9
C O~
aIi
- C)
-i
ZI
ui
"cv,
t0CVr
X6
dCn
cm:
- C
O C)
ccr
0>
oc
-
0a)U)
V Cs
v
M
Q)
4
C*
to
a80C.
CD O
C-i j
i
.r
0
U
C
0h.
0
0-
0a.
in
CL
v; Nr
c
0
1
0
2 CO
5,
r>
.. @p
v-
V ft
1b
oo cli
ui
a,
W
%
OCO
COs
CDCO'
0 Ul)
ui -
sr
ui o,;-
. .
cic
Nw- .
m
20
0.
0
C.
0
6
0
0)
0
.'-c
0
C
C)
co
.
'F
0
C
5
0
2CD . . 00
1a
a
CHAPTER 2: TABLES AND FIGURES
c
EE -
i/j
Cv'C
c v- N a
1
C
.~~ ,~_`
CoDc-.
o:_,_
to
o C~~
U'-u4C
. =.W ud
I oCU
-Fa
oJ_
.0
io
O
a
#,f
0
if'
4
C
i'
0a
co
N
a n
I
I
cic
O
0
,O
6)a.
4)
an
0
0
C.,m
C
I
0
"It
I
o U)
0
2
o ~
--.
0
~m 4) Co,ai4
=U 02.: ,a. -
E) 0
cc
Z*M L.
Mv
c-o
V
0
1n0
(
CC
a)
(
rC
EC
"E-~.om%
OE
rWe
E~~LO0
C
4
a>d
a.
r- .M
M
N
vCV0)
CO V)
I-
cv'-.
I Ia
,
CCD'
,
0 c
c N
cd
a ov
(D 9·
4
CD
CQch
P~,
4
q
40 w
CO 0t
o
0b
0) O.
00 C)
.
o
0
O)
cC.
Ir) Itl
0-
4 '
C,,
C)N
1 i
w
. -
C"JN
Cs
0
C CD
a
(00
t
Oi
dI Cs
I
u ui
Ir
oU
a or.
r
o c
v
C
I
E
0
0
A-co
m
a(A
0
o
oC:)
V)
ci ci
coa (
v
I
_ 6I
c
*1
(C
C"
-
Q 0)
V:
C)
N
v. .
00to
t
a0C
Ul
3DI,
CD
N
dao
ot
(NN
U)
C
R
e0ci
r.
0
CM
V0
0
oD
c,
0)
0
C
0
0
0
D
eDto
N
DCtv,.
cl 114
i CN
edCr
IID C
00)N
i
Cn t c(0
_O
0
a..
0
id
C7
io
C
M N
U
5;
B
3
r CN
6C$ - i
riV dd
tar
0.
40)
D
C
to C
0
S
an
a
a o<
c
E
0
U-
..
c0
i
O :CsC
0
c
=EEEn
E
C
)I
0
to
M
C:'
O MMO -.
i-
O:
r N
C v- d C0
C
Cc
r)
0
'~":'
iij
0c
°
0)
a,3
I
0
.C
0
c
) " ,,,, I
(D r 0 A
or,-
0
0. = a
J~
v
,.:
8
L-
a)"t
C
O
11,
CU
ea'~
0 o m O- o '-0
~orE
~
0
i=1:0
cc
.~C
EL
W.,= Cr)
~'0
O"
G.0-c
L C fun)
0, - i ... v,
~ ~..~
'~"~Oa.mw.o.-
iL
m
N
C
U-
I,
0
CO
am
CD
.- D co
o crc
ao-- i~i_.
O
m E c.
M0L
a..3.
om
>L)0
o C
00
n lv --
I
CDIt U)
I
)
4n
Il
o
.
N
tOj-:
bI_
cn
o
"
a
*
NC
~l-c
C0
<w6 aE,
. W '.-
U) o)
UL)O)
I
c, Q
V-: -:
c0
a.)
0 Cv
NCU
cj '?
,.-- u '-0C
'i
6-
Pr
M4
U.
CD~
o '"o.o=wEorn
1 .- ~.u
eo)
.0
a..,
a.. . x~ a,
I
Cn co
W)
C.,
0 O o cEa.
C
0CU
a Cd
o0 m
CU~
~ ~~~~C
O§c m- o,"
.
0)
.
CU
-r:- .
0 ~.,,
m:CL
C~,,L,
~2-g,
0 -o
o0
a
a,
2L-
u)
I,
y~g·-CU
C
C
C
0
C
tU
0.
0
n
a.
0)
.2'
-C
5
C
o)
.9
nVc
a..
86
CHAPTER 2: TABLES AND FIGURES
87
P,
~0
csCbN
C
j
Co
Co
0)
Q(C~
a I
f%
II CO
a
0(
co
"
(a
i/i
sC (D
i-CD
-C
c~
d
1 T
f
$
,Z
CD)
ClCO:
co
I
10
0 0I)0).
I
o
r
o
06 v-
Vv
v0)
0
,.I
CO D
N(DC(l
)O
(c)
U oCo
a CD
I
<r
L0
t
(o
61 o
0
U-
q N
cOP
0
(1
m 1')
C
0e
6cO iN
4L
co
6 c'
N,
oI
a? 4
I
-cv
r C0 IIr- CN
Eo
C
t Xn ,.w,
u,
c 0 c
C"
(pcV~c cO c
1*t o.
m
(D I
W) I
Y-
C
1n-~
o
4
4
NE
i s
C
I
C
i
's
yO
e O
,
.(U
CO CCX
)O
cj
) C4
.a
Q
0
o
.0
II
I
0
0
(a
E
DC\I
E
U
o
-
0)
C
U'-
c
2
C
c
2
._
0)
CN'
co
00.
o
o
I3
CY
co l
N C*
co
cD
co
ICD7
t l)
c
Ci
a
tCD~
cr
CI
Oca
20
a
40
Cc v-
0
ct
aE
0
4C) 1
oRa
0
C)
IL
dd
c3
oh
oi
b n
5
L.CD
IO~v
uCL
-C
o9
o
j
,, F
0
r
0yC3)
-C
0)
o
0o
o p,
i0
oo.C
..c
y- C
r
^
0
03
Ocy
o
CO Gi
C
.r CO
LO o
.'_
I'-
a.
0
cn
Ia.
og
0
NO
P
0o
(6
OCD
'C
-.o
C
c
C,
o a'c
c0
W
eaar
V:0 GO9
C 'r
C
0
0
jca
C-
C
qC
O
~L~r
C"
(f)Z
iNr
CAD
0
U
·L
f O)
o
o
oo
o
10 2 X
0
E
ge
c;
2
I
C
to
a.
o
0
l
CHAPTER 2: TABLES AND FIGURES
88
c
:E
Ln
CV~
C
C
Ci
0
v:
.)
(0 0C(3
0
a .CD
0
0ci
0
I
I
"I
C
CD
c:
,cj
0
-
if N
2
C
t
iC
-
II
M
i
CD
C
e0
w
0
U)
LL
0
a7 1
I
cd
0
C
i,CV1i-9
CC
00 pi
f
m °
0
ZU
o
2 'o
0
-i02
0'-
oo
r-
nI
(1
m
o
CD
CuAC
)0
co Q
0) ci
o
4a)
w2s
0
C
0
0
2C
o 0
o0
to
Nj N
oo,
)
no
.- : tJ
Og
I
L,
..
r %C
CR
00
60
t,
N el
.
b
uC)
r
C
- g
Ul
m
O) W
sn
m
ro V
co
0
0
i
I-
I
or
O
C) Vo
E O
ad,
NC
c.
- (O
a, 0
2w
oM
X
ml
M'I
IO, Q.
0
h.
0
'-_ co
L
cn
n
Q
E
0 co
3FL
(a
1
a
o
0,
:,0
o
qocv
00C
C
co,
,ct
C;
V
Om
w
C)
6d cpco
q(
2
..
(A
:V-
e
c
to
m
0a
C
4,
0.
C
C
(0)
C
o
o
C
0C.
*
.
c
0
0
C
0
KE
a
CHAPTER 2: TABLES AND FIGURES
89
I
C
C
r-0m
CN
liI
cm
rD
8. I
m
0) 03
0
0
O CO
cOU')
00)
co
r-
6C C
e'j
(V C
co N
i'- t
0
_,
O0
I O
0OC
v
n
-
I-
9>
to
I'l
h~jN
rts
V
q
w CO
vi
iv
a
0
C.)
o) )
CU
U-
0
C.)
0)
o 4'?'
,.,:
GC'
01N
1- C)
0
sC
C
OrC s
C t
NC)
ch
:
O Cl
0~0
a0) c CO
L
C
m
C
a
1-
i0
Co
tCU
cCO C)
C')
CL co n Co
C
0.s
cCj~
~O
06'--
Coc
C.)
L C' CoN
c',I6
co
o
00 0
0m
C
0
.0
I-
oA
MvC%
tnC'
OC
co C
c4 "c
C
0
Ul
Co
cir 4
co
tv'
01
r=O
00N
cmco
Cmi
O- O
C
C O
O
LOW
CD
XtP
0·- (O
Cn
aDc'O
C9 cl
CO
_DC
__v
C)0
s
CC
.
.v
O
LL
Co0')
C
0
0
0
cI
oC
6i
I U)
6 :
C7 Ta:
o
6C
CO'~
CO
0C 00
NC
D?
tp)
0
t
11 qo
'll
0
0C(6
CV)
A
vi
ucO)
cc
CO
Nfv
tDCC
U :~
0f0 -
4
=dci
n 0)0
Cto
tn
gf_*
U>
i .
L
V3
Li
ci
CDV
C
.4
v
Ui ·
I
VC
0*
-0
%, 0-
5
.,I
1)
61
:
C C
C)
CNJ
t Cc
0c,
0 a: o) )
.-
Cici
0
2
0
OP
4
r00n
0'- 0 '
0)10
CU &
to
ij
io
:s
0i OC
04 CO
CJ O
C
,,-:
C
_O
co
s Cc
LO
O
1
It
s
.Cn
-
0
o
m
.rCD [VVF
C
0L
CC
Mv
io
tO N
CDo
00
E
C
(A
C
CO
to 3
tz
12
I
1
c v-
00
N'I
C
o '"
C
C
I
C
C=I:
a
o
0)U)
i
~0
V:O
dv tt)
CN
CNN
P
N
T0)
t t
CtO
-ti
r C
1C
0 c
ciNNi
(A
¶T
C104LO
t
" C"
Nn
N U)
(U
0
Ci
2
am
co
C"
MN
0-q C)
(D W)
N
eE v
(C
U)
t-
.q
200
')
C6 -
C.0
0)C
to
cNNi i
5
), N
i
U)
v,c')
O0
Cvi
t
-N
-
C
a
U
n9
cn
0
0
o
o
.
c
t~
0
.,
E0
-f
C
I
, ,
I0
0NC
c
cO
46c
0
C
C.
DC; 6
I.
Oo,
v- C%
53
0. 3z
0
a
0
-
3
N
1u
n3
C
c '3
CJ
*aI
m
F
0 II
II
-
coo
#_
0
0%
Cs
C
CU)
Sfl
LI)
N
c) Q
1
(D coNly
e-i a-
t
I
-C)
,,-O)~
E
CO-
0
C
m
t: 0 NV
C
.4..
D
CC
a
0
E
CNNV
C
0 0 0)CS4
U 3 C C4
C 50:
z0
_
Nsc
.cs-
I-
0
Cl
0.ts
Qr
C
0
C
O
CJ
c
_c C
T
0
0
U
C
e
.
h.
0
0
.0
c
0c
8C
-
IW
o
q.
0h.
CHAPTER 2: TABLES AND FIGURES
90
O
I"
ci
J
O
InO
1-
ci
a
0I
40
-0
0
)·-C
0
ol iN1OCt
>
(0
c;,
N
c1cs
C
C
"
M
1O
r
_1 Cs
0 co84
o
C
O
C '
f
0
E0
LO N
O
C
',
1-:
00
1 r
r
m0
O-
O)
?tO o
E
(n c
C
~-CS
Eco O-
ao
mw
V?
_6
.
(Owv-
oC
in
0?-t
Ec
E
_
(5
ci
OC;
0s.0
O
,V
- Eo
>
oe
O
0
S
C'j
I-
.D.
I.Eu
C co
cO
a
0
-J
w
_fP;
I4
s
C)
CO
la
O O
lI lo-
E
0
';~
C?
0
zU
0
iz
inr~.,s
3
0N
N C
l
ci
s O)
I
O10
QIC ,
R.,
co
Ur,
e
to
'-c
N1:
0ci
C
0
00
U,
oe
C
C
m
IL
I
I
-
C
(a
0u,
ce
to
In0
m0
a
E
0
U,
Q
C
9 IO
00
O
tn
W
£
C
0
CHAPTER 2: TABLES AND FIGURES
91
..c
i,
C CN
20
o)
r-o
V
o"
"
0
-KrJ
N
Cn C
C D tc~r
.c
L.
0
s
vi
U
co
0
rC
C-j
O.
1+c
a
z
U)
0) aZ
0
gIV
0
a
15
I-.
1-
..
uivco 9 11
0
r.:
0m
ta
0
0C
Cv,
ti3l? '
0
00 v
U)
c Di
, COO
':)c,,
O
Cs
vs j
cmw
i v
fl
ihC
i
'o C-i
r v-
ecvO
cn C-i
co o
(
U)
m o
1: c;
a3
*qoo
u~
c6 V..
c
I
cn
"g
og
C
OCIJ
a)OT
r0
-i
_
IV5
PI
a
0C.
ci ce
C? c
E
E
o
P-I
£
C
C
0n
c.
.
:E
N
aCt
0.
N )
(D CO .
CD 1q
e In co
Cd (N
n
C
-r,v M
Inca
cm v4
us
"It
O
o
F-
ciUc~)C
e,,
6
m (o
rllhcvC
-e
r-iN
0
Et
V
0
1)
U
b.
(A
I
)
8
_c
0
C
e0
0
0
I
0
Cr-co
0
ca
rI[
._
It
._
m-
CvU)
c, _- 'C
c
L
h.
E.
a.
-
o
It
oe
cn CbX
=Pi
I
L D
I. ,
*N
1
i
CI
5
I
0
Nh
o01
C
co
C
C-
N
*i
CII
C0
P
co
co 1
0
c.)
to
acg0C
U
o
aco-
nCU
0
co
a
C
IL
V
0)
(o
)
la
.o
I~o
0
0
0a
co
oi
.
0
2C
'n
)
co
0q.~o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~o~~~~~~~~~~
CN
a
1a
_=
I
U,
Ec
o6
cc C" o, o,
0 M6"
F
t
a
0
E
C
.
0
0
r
I
oU 4 U
m
C
L Z v-bc
O
'-
a)'-
CU
a1
V
COC
'
0
0c.
Ua
IIh
oa
4%
0
C
,U
u.
co
U,
-Z
O m
Cv1 o Co
0
in
Ilv
I.l
,,'
U)C
1
U,
co
a0
CO
I,
Cn r
CO
cnRc
0
L)
0
fR
to0
'
cr~h
2a
0a
c1 O
I
q vU. C-,
cN
c
0
W
c
0
c
V5
cr rn,
8
,
0
0
a
CHAPTER 2: TABLES AND FIGURES
92
0
D
E
E 'm vM(LC
'C::aVC00 0'Y-cN
0
c0 C oJr
..
.
00E:> E,.-~~.
-,,q{
U N
0
0
-
OEOSo
c..
0 c 0c
2)CDE
_.
om
U)- Vn
)0
u
.
ao
-.-
'.
x
- W0
-
op
".C
6.ia
0
c
a. .
00
00
%
=
C - 0 . (D .. .:_, 0'"C' C._ .
· 1r,-,~ c:u, DMQ
E
-M
0)
o
I-(J)YO
-l.:))·-E0
0 00
0co
crc
a
cA
Q
-c_CO-
0o
c
05
(D
' Ln
.
4,.
'
Oa.0
(D
c
m.·' c:m~
-_
o'V~P
CM Ct:mg3
5
r
0e
t'" C:
'3 :E 66~~~~~~~~~C
M
a. - 5= 4)L-Om O 0 cO
06
'__
e0
-O
_
O,
-..,-,r r=8a rM:A
m 0J=:
C
,'
o 0M
--.
E MTvcr0M W
r,-,Q: G)
00
G.
m5CD
,
m
)
,,.~G)
V.s
'
~
M>
E0CL
o .0
00
O0
c Mo
d
v-
llo
909
0 o0
o
t
CNo
i N
Oto
9 j
m e
Cv -
2 e
0)c'
O
C
i
9
0 0(S
"-
CD (0 D
oQ
Cv)
o c' oo C; 0,
6
v,
0
vz
I'-:
C
eiC
o0)N
0
r-
D
0)
0c
0
:
0
5D
to
e
U)
Ca
-C
C,
co
0
e
0)
: uz
99
(Dc
LO I' )
ca N
1W
v -t M
N
N0 . V
lC°,NI
cO.D
3
to
v
N4U)
qi c;
C c
v
I'
o
I
o O
(D)(Ca
1r
CD
9
ci
"-c,,'t
0.S0EL
E
4
0
c
8
0 Q;
U)
C
06
o
)
cn Cs
I
w
o; c
°,
ON
[email protected]
I)UO
O
N'- at
CO
Ir
O
0O
00N
C
C
0
c
0O0
e
0 CONm-
Nc
N c)
0
N co
99
ol
a.
y(O
NW
-0
oo
5
D
c
a
CsC
CDCN
co co
c
00
0
6.
oC
ci
-
q;io
n
6
cu t
Ca
rl
9- po
r-(P
r C~e~
mo
O
(0(
L
,. ,,
c2=~M0'. o
2··:
-0o )
0.05
WMC
It
00
0 0
CD
.
0
8
-
N'9 v9
0mEr0
g.0 _~E *-~Q Qj'-c>0-0Mw
c NN
o
0fCo
C
::
C
N
N
r ci
0
c9A
a
:.0 0L
E).
oo
c; 1e
~o
-o5
0 a)
6 .a
rC (
I-
0
CA
,. :/GIm:2
a .,_- C ,,,a
._~ CDct _cA
-Z:
m ~-P"
~ mc.c
_ ~
Lc- .. O
v 0~.cQ>
cCL
0M0,
C..:0C
-. C
006
cNVO
O,
99
CN
0V.-:2 .~0 . . ~; -, , 0
o.C:~~
j~ ~. - .0'_ C.C 2-. 0.' o
0 0
0 B
0 O
C
L.
~m~-.,_i,
'..MC.6
nmr C'~CDAo
- t.-~--'
,,.E.'nCMO=~~(
cs vtoCY
'
cr
In
E
r ,- 00
o"_
o,.~'~'~,j
C.0
E· 0U)Om
.) o,_Oo"CD- : m
m,
o-
m,
i~>
m'-D
m
DCi
C%
if
CS CD
6
4
u,
E _=
.
c
0.
0"CA o
;A
Lo
U
CN8
U0
U
o
s
U.
J-0.
cA
'CO
o0)o
on C
I
0
0
0L
a
)C.
~._c .
6o
.
a
0
C~
_ ._,0" " 0o~
~.cn.-rCU)~
0000...,00U6-gO (D
CU
co
o
CD
, OC, .5 C
,
.-.
0o
i-0
U)
-0
2
..,C~-.o e m"
0V0.0
O --oz0,-.
m :.. Q>
Eo0 C>
,-C
E 0.~
--CD
(l
c
o
CI.
t m
N
0
a.CIO
* -CD) m:: ..
CD E CD
D to
wo
U
0
6.
I
CHAPTER 2: TABLES AND FIGURES
0 co
I
CN 0 0)
(a
0
0)
cO
00
g
c3jC
03C
l)o
EN'-
cN
o
,,
m
to
oI
cO o
m
C? 9
0 O 00)
Ut~-
CT
CO
v
9
0
0' Do
0O
C
"
U)
c, U)
c0
0
oC
0ECD
O:R
CD
0L
CO
m
1)
I-
rlciJ0)moeU
0
O
9;
.
LU)
'-t
11
a
8
88
c;
I I
0 r
V-.
C' to CC
0
C)
LO6'*-
09 u
0
m
co
C
5D
I31
C
alh
U,
0n
ao
b
c)CD
0O
I
c,
w*m
cs
¢
0)
ax
a
mj
CC
N
in
c0
IC'
m
Y-N
CO
O 6
V
ciCs
aO'
e 8
C
clW eI C
PI co
le;,
e
I..
v-U)
0) co
ITU)
0
c0
n
cv
C'
NUjU
u
c
c)
*icu'
Tbz
CD
o)
D
cn
com
aD m)
cv c
C
NM
'T
v-
0 0
OC'
COO
'$U
CC')
i
Ni
C)
X
q6
aCD
0ce
"m
C
)
rcD
Q)1
e0
i
Ti
c i
@50
CO
.
a N°
aiCs
cr
Co9
am
~0
U)
cv))
80
E
c o ui
o00x NCOn VjsX8C6
Ui N
CO V
v
_
<D
._
U,
vi
U
0
8
U
$N-
v0
C:C
)
C
X
U)'54.
fo
0.
'U
°)
0)
'9
rv o
_
j
zz
a03
C
I
5
L
r0)
'a
0
l-
nQ
co0
L
o
C
vDo
AiL
CR- C7
10
In
eO
OM
ri
cn.
bCy mD
0
4U
U
0*?an )
"I o
(d
L
0
0
E
0
O)
CO a
a
t
Li
0)
a
nN
NOsi
_ '1-LI)
C4R
.' -
c· 'O s
0)
CO
C4
0
a
'
0
cr
5
. .
w
c
3
ji
C
ctO
0
7-
co
C.
L)
a0
o,
0
0
NO
-0"
c-i
Cs
U) fco )
COv
3 cqO
.CD)
CO03cn
P-'0
CO
C
CL
a)
)
v t)
cN
cfi
C ,6 O
z
0
0co
LO
0) C)
m
0
U)
r-.
0
to 41
-m
m'
0) 0) O
CO
' -00
0
CO
SN
C
ii0
cOo
Co (Z
co
co0O
r-co o~~~~~to~~~
LO
rC
lrO)
r.
0)
O co
0
0)
0
0-
C @
0
o
C0)o
0
li
0)
0
-CO
0)
0
0
q .
COi
0U
L(F:
u.
d·
¢Cso
0,
0)
0.
-
0)Uv) C
tCO
tO
t
U-
c;
o
J
'-0)
93
c
0
0c
in
"It
CHAPTER 2: TABLES AND FIGURES
cO
C1
.
F\~C
co
0
0
0a
I
.-
..
Q)O.
a
O~~di
CoD
C;
i
)O
'FN
N
Cdd
r
X6Ns~ .
r 0W
)h
CD
a
0'NCD
.
0
oC
u '"
.s C\
C9~d
o
u3
0_
'-N
Q)C4
O
0,9
cvi
0
0
I)
6o
(D
Ci
Z
0
0o
v
O
_
r E
i
t.-
o)
V
E
(i f
LOv
"
I- v
NOJ
I,:
i
O
(C
5t
m
o
Cn
Uo
in-
cp lr.q
N
Q
O
"J)
clNCd
0
co Ir
t.D
w 'C.
9;
t -i
cs
si
CD
ma eV
;
NF
C
0s
vv0
co
r-
o~
si
0
CCN
t(0
v-
-v
(v
0
9
CJ
N
c( 0 N
to
-
0·
m
c
a
LO
CY O0
(U
w
('-
cfN O
0w
coN
co
mc'o
I=
aL i CS
C -Jw
0
i K
m
o
v v
aD(6
N
cv
00
Q
NP.
t
QOr%..
NN Wl
0O ,
0)
O _Q
C6
0 X
cO
N
1
00
0
ci)
0) (" -
00
u,
0)(
0
(xN
Cr
E
CO
0
i
e0
.
0
CC\I
co
00L
0
C!
j.
.
CON
cvjN
Ds
-0 :
0
0
)
CD
.
94
1
I :
a
S
um
NNK
csi o''"i
ci,
O
0C
r'.ui
006
fls
N'-~
V'1o*=
CmC,)
cm cof
Y-
Ncc
-
c
I=
r(- vo
ll
-
C4 N
N
.
Q
hjr
cD
,IOtO
n
in
C4 Q
1's 0o
ci
*
Nr
nob-
060 CN_
adCN
i Os N
0
a
P
coF f
Clt0
"a
C49;
wC,)
C
NO'
n
L
2
n0
CN to
I',
_
(0
o
.D
n
C:i
ae
Vj
co
niC4¢
CON
0
(D
0
MtO
C:;
I)
rjr:
N
_
T-C 4
_
o~
0
Lo
eo
'AW
9; noc
~i:~~o
Nc
c
"tr (O
m
,6 C
'-
r- -
a) N
LO)
a
a
0
w
0
Nc pN
00)
-f
E
cv
J
I1-
0)
ow
04
*CO
O.
CU
E
3
v,
r
O)N
N'.atf
la1
e.I
0
co
N-N
coo
~-N
cN
CNN
(aD'.-
I vi
(0
t
C
0
C
0
o
(D'.
w
..
'5
no
0-
COi
oa c
C'4)
ac
C)
LO
aN a
LNcv
a
cEu
O *t .
Ca
NN
0
OV
CD
a
C
.0
u,
3)
5
0-
3(5
C
C
0,
ci,
*0
C
0
O
c
0
cC
0
a
0
0C
0)
C
'U
IL
44
co0
o
'IT Cj4
v
eo
C4
4n
OR
'
V
0
0
-
tDcn
c.i; .:
0
o
C9
P,
D
F.-
c v
V)a
COC
o0 o
a)
Nc
n
1.
crjwm
cO
o N
o No
0)
c,
'6
C
o
,s
C,
v)
N_
cr
r
CD)
ceo
0
e .
10 c
C'i
94
0c
(N Co cvn
N
E
0
LI
Cv
. .
cMc
,
m
u)
"9
m cv,
n
XD
a
A
UDC? r0
,- n
a) co
Sd,
>
CO
6
r-
cci
bo
o
Ca
0C
a0.
0
C
CHAPTER 2: TABLES AND FIGURES
h h
(D 0
v
Mv
cON
CC
:
co
toco
.
a0
0O
I
?
,t
C00
oco O
.
0
LO
-CO
q-O
0ccTM
Ct rT C!
a
o C
N'`;
-
csX
.O
L
0
NCi
. .
co
0 .
. ,7
Ce
-
to
1%
v'.
a
V)
- oN
<
U')
U2e
0
co
U.
99
Lc
l_()
"?)
Sl ' fl
co
0N
00
ci
t§ 0)
'Z·
'tP cio~
v-I
(3
.s
CO
ui -I
r
1 1
0 -
N'
co
cVi
O
Mm
C ,
'Tv*
,
IINc
L
O CO
u,
co
2
COND
t co
ci
*.
dS.
t v0
c
8110
[email protected]
0
I--
'r. V
I
C'V c
It0
N
0* d
P)
L6 cl
0)
t
.
9
ON
o,'aq
I
(1)
5o
(00
C)
0
A No
t6
- Nt
a)o
0 0;
dci-
CD
be
S
V-
'co
oC9
0~ 0
ej
na,
C CD
_
a
c
UC)
g)
co P.
a,
"
o g
L,
0
qi~OO
(M c0
00
4 .:
a
C
LaD )
r.
(o
O
7)
0 0
Mv
ic;Z
C
4C v
co
t
r
0 N
'CO
NI
O)C
CD (O
O
cDo
8co (O
le
0o-
(00
o
I:ci
c
9
C
C0 n
C;
i
C
COi -
M 0o
0
L
ru
0
C;
WC
d c
M
a)
E
.:
.
C',,
co r- C W_
T V)
C
UE
C,
e(O
0
(U
i-c0M
0)N
0)
NS
CS
I
oi
ci
0
'iv
o"
03
4
a
c OM'nO
117~ C
'TcP
to N O)
O
:n
A
O)
C' c0
coo
'trcC
5a)
n
t
'·-T-C
pCi
0
ao
CO C;
t
CO
II
0 0
t~
I
0a)
9
IDI
L
g
m
'-
Ti-
r-
0
6
't-
rM-"I N
* - O -a
CM
co
CO <
.
C
,Itco
'IT
U)
LOv5
INlicV
T-
t0
20
a)
-c
0
95
-
7
,
CJN
i
CC
ci
-
D M
- am
Ci
r
M) )
CO
"
v-
Od
00 )
c to
-C
o
z.
~(rc
C
r- o)~
N "R
COO
aDO
C
2o
)
,f
co
M3
1
0
c)
ci
3
ov:
COj
C!
o%
t
vo
cs
?
It
v-
J
!V
!
;z
U)
20
C(
cz
ui 0.
14' 0)
r w
5
11
I
N
N
cq 2o
m co
CO
0 ,V
CY ,-~r
N
I0
rrr CO '14.
v
Ni ci
i
tl-: O
[email protected]
Loto
DCr0)
v
a9
v- c
(C
°
0t
Nf
Q
N IX)
CC
o
6
1:. co
c
0
j-d'
ND t
fcc
o O)
o- oC
v) d')
o Co
Sto
(C; Cs
06oC
CO;
Ci
riC
SO
0
a,
C
.U
ez
to
0) CO
o N v-o
v-1
M
. .
ac
CO N
CO
t- N
MIm 1)
ci c6ciCs
i-'
".
hOS '
cr
C
a1)
a
C
(0
O
C
C
()
T)
O
C
E
p
a
C
a
E
0
Cto
a
D
0-it0_
1<
-
_
C
tn
C
C3
'3 C.
0
O
I
CHAPTER 2: TABLES AND FIGURES
CN
ejOD
r-c
cQc
,
(a:
0
0
0L
96
.X
coco
c)C
CD
q
£
t£0 r-'-
q(
?0
'-0,
v,'
r-0
Ci S
ui
V·
-z
00h
i-00
ao~o
Nci
o.
vO
£C J
cj
O5"
C'jC
N
""
C4
co
CI
0
CS
r- c
(EM
c'-
v
D)O
'w
CD
u,
u43
'h
C- a S
C?
CM'
o -E
CO
(4e
N
a (Vol
Sa
zw
S S
X
C
.
:
O CV
(N cO
N
0
E
0
0
t
c
0
4)
0
'U
m
mp
)
m
-_C ,
Ur)C0
rCV'
N
c
i
''k
V) Cs
c%
-:)
&
OC
i
M
ci
v0
ow
D
U) Co
_O
(N
f-TC
aD
o) z9
00
(l
C
O) U
o
*( q)
0
cr
04
n c al
-c
0,O
n
-
I
S C-~
II?
qo )
0
3-
av
C
iC')
ajh a
e m ur
4
Sq
C'V.:-
Itmt vo
40
LO
a in
CDm
a
U)
O
ia
0
O
-
S
(a0
s
(N'
r--
5!
N
CC
o,I C
Cv
N
*CD
0
an
r- I
O
c cn
0
K
i-I)
O6O
5
m
a
c-;
C6
(n
Vco-
OFV CC
U
CDCR
_SO.~
mi
po B
e- c
a)o0
,Rr
D
(D
C
C)o)
0U)
L
a
Iu
LL
V Go
co6O
Cv, Cs)0)
CIe
ED
.
CL
Z
C-
~
9
ul)
S j
ceo
4)
,
00
E
ax
8
e0
F
I
CHAPTER 2: TABLES AND FIGURES
97
Table 9: Long-Short Portfolio Returns (High-Low Growth) for Different Five-Year Growth
Measures and Horizons During and After Disconfirming/Confirming Years
This table displays the returns to buying and selling an equal-weighted portfolio of top and bottom
quintile stocks, respectively, in the year that subsequent operating performance was revealed, t, and the
following year, t+1. Quintiles are determined by sorting firms by growth measures every year. Panel A
shows returns for quintiles formed based on the percentage change in five-year sales-per-share. Panel
B shows returns for quintiles formed based on the change in five-year net income-per-share divided by
total assets-per-share in the initial year. Panel C shows returns for quintiles formed based on the
change in five-year operating income-per-share divided by total assets-per-share in the initial year.
Panel D shows returns for quintiles formed based on returns over the past five years. Within the top
and bottom growth quintiles, firm-year observations are considered "consistent" ("inconsistent") if all five
(three or fewer) years of sub period growth are consistent with the five-year trend. A given firm-year is
consistent with the high-growth (low growth) trend if it is above (below) the median growth of other
contemporaneous firm-years. The bottom rows show the difference in long-short returns between
consistency groups. The row labeled all" provides long-short returns for all top and bottom quintile
firms regardless of consistency of prior performance. The first group of columns displays raw returns,
the second alphas from a three-factor regression, and the third alphas from a four-factor regression. Tstatistics are shown in italics below portfolio returns. The sample period is 1970-1999.
Panel A: Operating Measure is Sales Growth
Confirming Yr.
All
Consistent
Inconsistent
Difference
Disconfirming Yr.
All
Consistent
Inconsistent
Difference
Raw Returns (%), Yrs:
3-Factor Alphas (%), Yrs:
4-Factor Aiphas (%), Yrs:
t
t+ 1
t
t+1
t
t+1
8.50
0.00
12.76
2.31
10.78
4.45
3.12
0.00
3.22
0.75
2.07
1.04
2.89
1.23
9.36
1.69
9.82
6.04
0.77
0.36
2.06
0.45
1.56
1.11
10.60
1.34
11.91
4.15
8.11
0.91
3.01
0.46
2.90
1.15
1.45
0.16
-7.70
-0.10
-2.55
-2.46
1.71
5.14
-1.83
-0.02
-0.63
-0.50
0.26
0.78
-11.59
-0.71
-8.36
0.36
-5.50
0.15
-6.08
-0.27
-2.81
0.09
-1.74
0.03
-6.96
-2.03
1.74
-0.22
0.36
-0.06
3.84
0.79
5.60
1.34
3.58
0.65
-12.08
-2.51
-8.52
1.78
-3.62
-0.74
-1.69
0.31
-7.99
-1.31
0.26
5.12
4.25
1.04
0.89
8.30
1.83
2,06
0.43
13.59
1.93
1.78
1.80
0.29
1:
CHAPTER 2: TABLES AND FIGURES
98
Table 9, continued:
Panel B: Operating Measure is Net Income/Total Assets
Confirming Yr.
All
Consistent
0.53
13.51
2.67
10.72
1.19
4.39
0.24
4.17
1.02
2.34
0.28
1.95
Inconsistent
Difference
Disconfirming
10.58
13.91
1.99
12.62
-3.02
0.36
1.77
2.05
0.32
1.40
-0.33
14.92
-1.11
12.36
3.69
9.54
-3.73
6.10
-0.38
3.88
1.02
1.73
-0.86
-12.97
8.14
1.54
-1.71
3.08
0.71
-2.27
1.91
0.23
-0.34
0.31
0.08
-25.86
-1.74
-19.08
-1.57
-20.46
-6.47
-7.87
-0.57
-4.91
-0.41
-4.23
-1.47
-29.44
-30.26
-26.49
-21.65
-14.63 ,
-13.52
-2.73
-1.94
-1.89
-0.86
-0.88
-0.64
-24.44
-0.80
-19.06
2.60
-20.21
-6.46
-6.37
-0.20
-3.23
0.43
-3.41
-1.20
-3.43
-27.36
-4.78
-22.12
9.61
-6.50
-0.28
-1.53
-0.28
-0.72
0.60
-0.26
s7.03
Yr.
All
Consistent
Inconsistent
Difference
Panel C: Operating Measure is Operating Income/Total Assets
Confirming Yr.
All
Consistent
Inconsistent
Difference
Disconfirming Yr.
All
Consistent
Inconsistent
Difference
8.18
2.01
10.38
3.50
7.64
3.93
3.42
1.04
2.57
1.69
1.42
1.07
3.45
5.60
10.05
6.07
12.84
7.70
0.92
2.08
1.89
2.59
1.79
1.86
9.91
0.39
6.69
4.68
1.15
0.26
3.70
0.16
1.81
1.85
0.22
0.07
-6.46
5.21
3.36
1.39
11.70
7.44
-1.45
1.67
0.76
0.43
1.87
1.45
-25.73
-0.37
-18.57
-2.09
-16.65
-1.96
-7.91
-0.11
-4.99
-0.49
-3.12
-0.41
-37.09
4.50
-10.81
-4.71
-7.29
-12.32
-2.46
0.39
-0.60
-0.28
-0.35
-0.95
-23.48
1.39
-15.22
6.57
-11.72
-2.24
-5.99
0.29
-3.02
0.83
-1.67
-0.38
-12.39
8.52
-0.66
-4.31
-2.16
-12.28
-0.95
0.83
-0.05
-0.31
-0.13
-1.47
CHAPTER 2: TABLES AND FIGURES
99
Table 9, continued:
Panel D: Performance Measure is Prior 12 Month Returns
Confirming Yr.
All
Consistent
Inconsistent
Difference
Disconfirming Yr.
All
Consistent
Inconsistent
Difference
73.20
-7.66
69.94
-2.56
65.65
-9.91
19.25
-1.81
20.03
-0.65
18.33
-2.03
64.70
-10.64
61.06
-8.28
57.26
-19.05
22.53
-1.96
33.64
-1.24
20.85
-2.28
80.96
-3.24
79.57
3.84
73.79
-4.36
14.54
-0.75
11.34
0.84
12.92
-0.81
-16.26
-7.40
-18.51
-12.12
-16.52
-14.69
-4.11
-1.62
-294
-1.79
-3.53
-2.23
-98.35
-11.37
-90.33
-13.12
-86.61
-8.90
-14.30
-4.43
-13.70
-5.41
-12.52
-2.27
-98.89
-11.90
-89.91
-14.17
-85.14
-8.83
-14.05
-2.54
-14.29
-2.62
-11.39
-1.35
-100.71
-10.50
-93.96
-10.15
-87.38
-8.41
-14.55
-3.85
-13.27
-3.12
-12.08
-1.54
1.82
-1.39
4.05
-4.02
2.24
-C.42
0.38
-0.32
0.96
-0.66
0.31
-0.07
Chapter 3
Highly, Somewhat, and Un-Informed Investors:
Institutional Aggressiveness and Information Flows in
Stocks
3.1
Introduction
The past twenty years has seen a dramatic rise in institutional money management. The prevailing
view is that this has made financial prices reflect more information, since institutions are more
informed than individuals. An alternative view, however, is that any given institution does not
have an appreciable edge over other investors. Moreover, delegating funds to managers creates all
sorts of agency problems. Some research shows that institutions reduce autocorrelation in returns,
so that they make prices more efficient. Other studies show they reduce autocorrelation in returns,
so that they make prices more efficient. To what extent are institutional trades related to informed
trading? And are active institutions more informed than less active ones?
Using the quarterly Spectrum database of 13F filings by institutions from 1980-2000, I test these
ideas. I find that institutional investors actually increase one and 12-month return drift. However,
this is not due to uninformed herding on past performance. Institutions condition their trades
100
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
10G1
across stocks more strongly on past returns. Yet they appear to earn small but reliable profits
from doing so, implying that they trade on information before individuals. This analysis confirms
that institutions as a group follow momentum strategies (among other characteristics that they
favor). It also supports the idea that, on average, such trading is informed and not merely feedback
investing.
As a further test of whether or not this idea, I separate institutions base-don an ex-ante measure
of how confident they are in their information. Those institutions that trade more actively appear
to use information contained in very recent returns when changing portfolio positions. They also
earn larger (pre-fee and transaction) profits from doing so than less active institutions.
I take
a new approach and divide managers into groups based upon two measures' of aggressiveness:
portfolio turnover and weighted-average change in percentage ownership of firms each quarter.
While imperfect, these rankings give us an independent proxy for how informed institutions believe
they are. Those managers that trade in and out of stocks in large quantity are more likely to do
so because they believe they have information, rather than for rebalancing reasons.
Separating institutions by a simple measure of confidence helps us in three ways. First, by looking
at disaggregated data, one can see if the conclusions about informed vs. uninformed investors still
hold for subsamples. It is widely thought that institutions are better investors than individuals
because they are able to take advantage of more opportunities.
However, the aggregate data
includes a diverse universe of dedicated hedge funds, multi-divisional investment banks, and index
funds, which clouds any conclusions. If informational advantage is evident for a large and growing
class such as all 13F filers, it must be even more clear for subgroups that are more aggressive in
pursuing investments. I find that the tests that indicate that institutions are more informed give
us even stronger results for more vs. less aggressive ones.
Second, looking at subgroups can give us a clearer idea of what motivates trading by different types
of institutional investors and individuals. Institutions favor stocks with certain characteristics.
However, by using subgroups, we can see what features appeal to active managers vs. more passive
institutions. I find that more aggressive institutions react very strongly to returns in the immediate
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS,
102
past month, but are not momentum traders. Less aggressive firms seem to react more strongly to
past 12-month returns, and less to 1-month ones. Finally, the most passive institutions do not react
much to past returns at all. This implies that the aggressiveness sort shows us how information
diffuses (in returns) across investor classes.
Third, do the trades of different types of institutions forecast returns? In other words, can they
predict performance because they are more informed? We can look at the returns to a trading
strategy that mimics how institutions (in aggregate, and within "aggressiveness" groups) change
their holdings each quarter. The fact that active institutions earn higher returns than passive ones
supports the intuition that they aggressively trade on information, at a much faster rate. One
gets a similar, but weaker, conclusion one gets from looking at the aggregate data and comparing
institutions vs. individuals.
In this analysis, I refer to "managers", "institutions", and "professional investors" interchangeably.
These terms are meant to distinguish institutions covered by the Spectrum database from "individuals". It is possible that smaller institutions are classified as individuals because of the 13F
filing restriction. However, one could argue that these smaller institutions share characteristics of
individuals, since they are likely to have smaller research budgets. As noted, I also distinguish
between more active and less active traders below.
Section 2 of the chapter reviews previous findings on institutions and their affect on asset prices.
Section 3 summarizes the data, Section 4 shows the analysis of the aggregate institutional data,
and Section 5 applies the same tests to the subgroups of institutions. The conclusion is in Section
6.
3.2
Literature review
Recent research shows that asset prices are predictable for extended periods into the future. This
could be due to transactional frictions, behavioral biases, legal or regulatory restrictions, or some
combination of the above. Several recent studies on the "limits of arbitrage" examine these features
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
103
of equity markets. Baker and Savasoglu (2002) look at returns to merger-arbitrage strategies from
1981 to 1996, and find that they are negatively related to deal completion risk and target size. They
also distinguish "arbitrageurs" from other institutions by looking at changes in ownership levels
during mergers. They find that limits to the supply of arbitrage capital have some influence on
returns to a merger arbitrage strategy. Mitchell, Pulvino, and Stafford (2002) find that uncertainty
about returns to an arbitrage strategy for 82 mis-valuations in the prices of parents and subsidiaries
is a significant risk.
These same limits to arbitrage can also allow differences between groups of investors to persist.
Various studies have looked at the differences between institutions and individuals. Jiambalvo,
Rajgopal, and Venkatachalam (2001) find that stocks with more institutional ownership appear
to have prices that reflect future earnings better than those with lower ownership. This is again
consistent with the idea that institutions in aggregate are more informed than others. Gompers
and Metrick (2001) show that levels of institutional ownership are positively related to returns, and
provide evidence that this might be related to the shrinking "size effect" in stocks. Sias and Starks
(1997) show that stocks with higher institutional ownership have high daily return autocorrelation.
This is consistent with the idea that institutions speed price adjustment to information. Del Guercio
(1996) shows that differences in bank and mutual fund investment guidelines cause their portfolios
to differ, with banks being much more conservative in their investments. Badrinath, Kale, and Noc
(1995) document that returns of high-institutional ownership stocks lead those of low-institutional
ownership stocks at weekly and monthly frequencies, supporting the idea that institutions are more
informed. Nofsinger and Sias (1999) use annual holdings data to show that institutional trades
are positively related to contemporaneous and lagged returns, and that stocks institutions buy
outperform those they sell.
Other studies look at specific types of institutions. Grinblatt, Titman, and Wcrmers (1995) find
that most mutual funds follow momentum strategies, and profit from them. They also find weaker
evidence of herding by these investors.
Lakonishok, Shleifer, and Vishny (1992) examine a large
sample of tax-exempt investment funds and find that they do not strongly herd together or pursue
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
104
positive-feedback strategies. Bushee (2001) introduces a measurement to classify institutions based on their past portfolio turnover, diversification, and momentum-following behavior. He finds
that short-term investors tend to overvalue near-term earnings and discount multi-year potential,
creating mispricing. Bushee and Noe (2000) use the same classification system to document that
companies that improve their disclosure quality attract a very active, but short-term group of investors. These investors then increase the stock's volatility. Badrinath and Wahal (2001) also divide
institutions by their entry into new stocks and exit from existing holdings, and find that institutions
are momentum traders when they enter stocks, but are contrarians when they exit. Dennis and
Strickland (2001) examine daily returns and the Spectrum database of 13F files, and find evidence
consistent with the idea that pension funds and mutual funds are positive feedback traders, since
stocks they hold react more strongly to market returns. They suggest that institutions overreact to
market drops. Dennis and Weston (2001) find that levels of institutional ownership are highly correlated with microstructure-based measures of informed trading activity. Chen, Hong, and Stein
(2001) combine short sales constraints with heterogeneous investors. Their model predicts that
stocks with lower numbers of investors will underperform those with more investors because the
pessimists' opinions are not fully registered in the stock price. They find substantial return difference between stocks in the top and bottom deciles of ownership breadth. Cohen, Gompers, and
Vuolteenaho (2001) decompose stocks returns into cash-flow and non-cash-flow components, and
show that institutions act on the former while trading against the latter, and therefore outperform
individuals.
3.3
Data and summary statistics
The institutional ownership data is from the Spectrum database of 13F filings. These cover all
entities that manage over $100 million in assets. 1 All positions in a stock held by such institutions
'This figure is fixed over the entire period from 1980-2000. Some entities could enter the database purely because
a rising market made them grow above the $100 million level for 3F filers. Such new entrants do not change the
character of the sample of institutions. Gompers and Metrick (2001) note that even if one inflates the $100 million
threshold level by market returns over time, only a very small percentage of institutional assets are excluded.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
105
are aggregated into a single number per institution.
Table 1 shows how institutional ownership changes over time for stocks divided by NYSE size
quintiles. Cross-sectional averages for selected year-ends of percentage shares held by 13F filers,
market values, and number of institutional holders are displayed. The bigger stocks tend to be
favored by larger institutions.
One can see that in all size quintiles, institutional ownership (in
both percentage and number) rises over the years. By the end of the period some the smaller
stocks have levels greater than those of the largest quintile at the beginning. The rate of increase
in ownership is greatest in the early 1980s, but has continued at a steady pace.
There are two other points to make about the patterns of Table 1. The time-series trend in levels
highlights why it may be better to analyze stocks based on absolute levels of percentage ownership
rather than cross-sectional percentiles. Stocks that institutions prefer should benefit from more
active scrutiny and liquid trading. It is likely that the absolute fraction of a firm that is held by
institutions matters more for returns than the relative level. Dividing firms by ownership percentiles
will obscure this. Stocks in the lowest percentile of ownership in more recent years are probably
very different from those of the 1980s, due to the growth of institutional investment.
Second, change in aggregate ownership of stocks from one quarter to the next is very slow. However,
there is substantial cross-sectional variation in levels and changes in levels of ownership of stocks
for specific institutions.
The aggregate figures obscure this fact, which suggests that managers
are heterogeneous in their trading behavior over time. Therefore, one benefits from analyzing the
differences among managers, something that I turn to in later sections.
3.4 Aggregate institutional ownership and return predictability
I begin the analysis by using aggregate institutional holdings data, to highlight the general effects
that large money managers play in transmitting, and responding to, information. While this
classification is crude (it combines all different sorts of institutions together), it will make the role
of institutions vs. individuals clear. Note that since I use all managers in the Spectrum database
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
106
as a proxy for institutions, one minus the percentage of shares owned by institutions equals the
percentage of individual ownership. Therefore, any coefficient for professional managers can be
converted to one for individuals, usually with the opposite sign. In this scheme, institutions behave
opposite to smaller investors.
The question we want to answer is: have institutions made stock prices more efficient in reflecting
information? One way to te,st this is to see if various forms of stock return predictability are
reduced by the presence of institutional owners. If less predictable returns indicate more efficient
prices, higher institutional ownership should reduce any relationship between future returns and
past variables. Depending on the form of predictability used, we can see how institutions react to
past prices, specific corporate events, or valuation levels. I focus on how managers change react
to past one-month and 12-month returns (which have been shown to predict future performance)
rather than other firm characteristics that are slower to change. Extreme returns can proxy for all
types of short to medium horizon information shocks.2
In Table 2 I present the results of Fama-MacBeth regressions of cumulated returns over three, six,
nine, and 12 months (excluding the first week to account for possible microstructure effects) on
various firm characteristics, including level of institutional ownership. Specifically, the dependent
variable is compounded return over months t + 1 to t + n in event time (where n is three, six,
nine, or 12). The regressors are the log of market value in millions of dollars, log of price, log
of the book-to-market ratio, and the percentage level of institutional ownership, all taken at the
end of month t. Again, note that since I use all managers in the Spectrum database as a proxy
for institutions, one minus the last number equals the percentage of individual ownership. I also
include share turnover over month t, returns over month t, and compounded returns over months
t - 13 to t - 1 in the regresscrs. These characteristics have been used on other research to predict
returns. I conduct cross-sectional regressions at the end of each March, June, September, and
2
The relationship between level of ownership and information efficiency in prices depends on the idea that institutions mostly move prices of stocks that they currently own. This assumption (often unexpressed in the literature)
contrasts with the idea that institutions buy or sell many shares of stocks they don't currently own. Since shorting
stocks is difficult (for various reasons), this may be reasonable for negative signals. Managers also probably pay more
attention to their holdings than to the universe of stocks that they don't own.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMED INVESTORS
107
December, and use the time-series of coefficients to get averages and Newey-West standard errors.
There are two specifications for each return period. The first specification (upper rows) shows how
compounded future returns are related to various firm characteristics. These coefficients are in
line with previous work on the cross-section of stock returns. Smaller, less liquid value stocks earn
higher returns. Consistent with a momentum effect, a small but significant part of past t - 13 to
t-1 returns carries into subsequent quarters. Returns in the immediate past month have much less
predictive power. Also, consistent with Gompers and Metrick (2001), firms with higher levels of
institutional ownership experience higher returns. A firm with 30% ownership at the end of month
t experiences. an extra 1% of returns in the next quarter, an extra 1.8% return over the next six
months, and an extra 2.8% by the next year. Gompers and Metrick attribute this to persistent
institutional preference for the same type of stocks they have always owned. This also means that
firms that are mostly owned by individuals have lower returns.
NWesee that past returns can forecast future cross-sectional returns from the first specification. Do
institutions eliminate this predictcability? The second specification includes two interaction terms
for the effects of institutional ownership on return predictability. The first shows the coefficient of
past month returns multiplied by the level of ownership. It is positive and statistically significant
for all future return periods. This means that firms with more institutional ownership have more
return autocorrelation than firms with less. To calibrate this unexpected result, a firm with 30%
ownership and a 2.5% return in the past month will experience 15 and 30 basis points of returns
over six and 12 months, respectively, above a firm with no ownership. The second interaction term
is past t - 13 to t - 1 returns multiplied by the percentage institutional ownership. As can be seen
from the coefficient, institutions
also seem to strengthen
12-month momentum.
A firm with a 15%
return over that period and 30% ownership at the end of month t earns an extra 26 and 46 basis
points over the next six and 12 months, respectively over a firm with no institutional ownership.
So far the evidence suggests that a bigger presence of professional managers increase the autocorrelation of returns in firms. This confirms the results of other research, for instance Nofsinger and
Sias (1999). There are two (not mutually exclusive) explanations for this institutional effect. First,
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
108
managers could be very fast to act on information that they have. They increase (decrease) their
positions before a rise (fall) in price, and therefore predict returns. Managers act as informed strategic traders, partially revealing new information through their trades. In this case, institutions
would be trading against individuals, and the former would gain while the latter suffers losses.
The other explanation is that institutions tend to herd, trading in a correlated manner aside from
any informational considerations. Thus they buy the stocks they previously bought, and sell those
they previously sold. In this case, institutions as a group should not make any persistent profits
over time, and can destabilize prices away from fundamental valuations as explored by DeLong,
Shleifer, Summers, and Waldman (1990). These two explanations do not necessarily conflict,3 but
if institutions have even a small ability to trade on information at monthly horizons, they should
earn profits.
Do institutions forecast returns by their trades? One quick way to find out is to create a trading
strategy that mimics their shifts in ownership. Each quarter end, I rank all stocks by the change in
institutional ownership that they experienced over the prior quarter. In Table 3 I present returns
to a strategy of buying the top tenth of stocks by change in institutional ownership and shorting
the bottom tenth of stocks. At the end of each March, June, September, and December, I hold
equal-weighted portfolios of top and bottom deciles, and skip one week between stock selection
and investment to account for microstructure biases. This is not a practical strategy in real life,
since 13F filings are not available immediately
after each quarter end. Panel A shows some average
characteristics of the top and bottom deciles of stocks by ownership change. Each decile portfolio
has a large number of firms in it, spread over many industries. 4 The stocks that institutions buy
tend to have had higher returns over the past quarter, and lower levels of ownership, relative to
the ones they sell. However, there is no systematic difference in market capitalization between the
two portfolios. On average, the stocks managers trade the most are above median size.
Given the large number of firms in the strategy, it is interesting to see that institutions can appa3In fact, one could argue that being able to predict the actions of other traders (at the start of a herding effect) is
important information. Most studies make a distinction between any forms of herding and informed trading, however,
for the sake of clarity.
41I divide stocks into 20 industries based on SIC codes, as in Grinblatt and Moskowitz (1999).
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
109
rently predict returns. Panel B shows the size and book-to-market adjusted returns to the strategy. 5
The stocks they buy outperform the stocks they sell by 1.8% after twelve months, on average, a
small but statistically significant gap. The magnitude is similar to that found by Cohen, Gompers,
and Vuolteenaho (2001), who use annual data to find that institutions outperform individuals by
a small but dependable 1.4% yearly. There is also little sign of reversal at various horizons. While
not conclusive by themselves, the results suggest that institutions can predict returns, which points
to the idea that they act on information before individuals. Can we see more direct evidence of
this?
3.4.1
How institutions change their holdings
Certain stock characteristics are associated with predictable return patterns. A first step in determining if different types of investors use information is to see what type of stocks they pursue.
From Gompers and Metrick (2001), we know that institutions are significant investors in large,
liquid stocks with low past returns. Furthermore, over time they are increasingly prevalent in value
firms. However, from the perspective
of information
diffusion, one would like to focus on what
determines the quarter-to-quarter change in holdings of stocks, and not the level. This will give us
a clue as to how sensitive institutions are to firm characteristics.
Specifically, since past returns can reflect changing expectations about firm cash flows, risk, and
the trading activities of other shareholders, they can proxy for many information flows. Similar
to the approach of Gompers and Metrick (2001), I regress changes in institutional ownership on
various firm characteristics in the previous quarter, and past returns. All variables are either in
terms of percentages or have been rescaled with logs. Table 4 shows the Fama-Macbeth time-series
averages of quarterly cross-sectional regression coefficients. The dependent variables are cumulative
change in aggregate institutional ownership over one (top row) to four (bottom row) quarters. I
use Newey-West standard errors to calculate statistical significance for those regressions where the
5
For each stock, I subtract the contemporaneous returns of a portfolio in the same size and B/M quintiles as each
firm. Size is based on month-end market capitalization of the previous June, and B/M is from the December prior
to each June. Month zero returns are not size and B/M adjusted.
CHAPTER 3. HIGHLY, SOMEWHAT, AND .1r
UN-INFORMED
INVESTORS
I
110
dependent variable overlaps from one period to the next. The regressors are: log of end-of-quarter
market value, in millions; shares traded divided by shares outstanding during the previous quarter,
separated for NYSE/AMEX firms and NASDAQ firms due to differences in reporting; log of end-ofquarter share price; log of end-of-quarter book-to-market ratio (as constructed in Fama and French
(1993)); returns in the last month of the past quarter; returns over the twelve months prior to
that; percentage level of aggregate institutional ownership at the end of last quarter; and standard
deviations of daily returns over the past quarter.
All of these variables have been found to be important to institutional ownership in previous
research, or important to future returns. Consistent with Gompers and Metrick (2001), we can
see that institutions as a group buy more of larger, higher priced firms. From the coefficients, for
a firm with a market capitalization of $1 billion, institutions will buy another 3.6% of a firm in
the next quarter, all else being equal. This buying reaches 6.7% by the fourth quarter. For a firm
with a $20 share price, they will buy another 1.5% of a firm in the next quarter, rising to 3.1%
twelve months later. They also prefer stocks that have had more trading volume recently, although
the turnover coefficient is much larger for NASDAQ stocks. However, this influence is short-lived
and economically small. While it appears from past research that institutions tend to herd into
similar stocks, they consistently sell (over several quarters) relatively large positions in firms that
have a big institutional presence, ceteris paribus. Given that the average firm is about 50% held
by institutions in 2000, this implies that institutions will sell 4.8% of it in the next quarter, and
9.5% of it by the next year. One must remember, however, that this selling is balanced by buying
associated with firm size and other characteristics that institutions as a group favor.6
How do past returns affect aggregate institutional trades? The size, price, institutional ownership,
and even volatility effects on changes in institutional ownership are all bigger than the effects of
past returns. However, institutions as a group do respond to price changes over the last month of
the past quarter and previous year. For example, as seem from the past return coefficients, a 10%
6
Why do institutions flee stocks that they own in large quantities, ceteris paribus? One reason might be that
managers seeking profitable investment opportunities may prefer stocks that have less competition. Or ownership
could be mean-reverting over time. This reasoning needs to be explored more.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
111
return in the past month prompts institutions to change their average ownership by 0.10% in the
first quarter and 0.20% over four quarters. The effects of returns over the four quarters preceding
the last one are even weaker, and less statistically significant. For example, a 30% return over that
period will lead institutions to increase their holdings by 0.03% over one quarter. This tiny effect
quickly dies out, however, in the succeeding quarters. 7
The responsiveness of aggregate institutional holdings to past returns suggests that managers update their positions to reflect newly revealed information. Institutions have a stronger reaction to
recent returns than to returns over longer horizons. Of course, institutions seem to react in a much
stronger way to other firm characteristics.8 However, even though the effects of past -eturns are
small, they are cannot be dismissed. Furthermore, the aggregate data is likely to conceal differences
between institutions, since everything from index funds to hedge funds is included in it.
The analysis of the aggregate Spectrum data leaves us with a few facts. First, institutions are associated with higher returns, and also more short-term (one month) and medium term (12 months)
momentum. This suggests that they either herd on past performance, can predict returns because
they are informed, or both. Second, a trading strategy based on their change in ownership of
stocks over quarters is profitable over various horizons. This effect is economically small, but statistically significant, and gives support to the idea that their trading reflects information. Finally,
institutions strongly react to various firm characteristics, but they also buy on high returns from
the past month, and sell on low returns. This suggests that they pay attention to recent marginal
information contained in returns. 9 Taken as a whole, the evidence is consistent with the idea that
institutions are informed traders, on average, relative to individuals. In the next section, I press
7
Contemporaneous returns would surely explain much of institutional trades, as Nofsinger and Sias (1999) have
shown. One would expect them to, since institutions conduct an increasing share of all transactions. To better answer
the question of causality, however, I focus on past variables that may forecast future changes in ownership. Most
of the characteristics I include in the regression are useful controls for future expected returns. In their analysis,
Nofsinger and Sias conclude that institutions probably trade on information more than individuals.
sCohen, Gompers, and Vuolteenaho (2001) also comment on the fact that institutions as a group change their
holdings very little in response to cash-flow news. They attribute this to their unwillingness to deviate from benchmarks.
9
Past returns are public knowledge. Therefore, the interpretation of institutional reaction to past returns as a
reaction to information means that institutions are using public information. In a perfectly efficient market, this
should not give them an advantage. An assumption of this analysis is that some groups of investors ignore past
returns. It seems that individuals certainly do.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMED INVESTORS
112
this intuition further by differentiating between institutional managers by other ex-ante proxies for
how informed they are. The proxy I choose is trading aggressiveness.
3.5
Institutions split by measures of aggressiveness
In this section, I divide managers by how informed they think they are, and then track the performance of stocks they trade over a quarter. In this study of how investors use information, we
want a measure that will divide "arbitrageurs" from "noise traders". Past research has shown that
individuals are likely to be in the latter category. But institutions surely differ in how informed
they are. The Spectrum database does not allow for much distinction between institutions. First,
it aggregates all the holdings of a stock across each institution, so that different divisions (e.g. risk
arbitrage and mutual funds at an investment bank) are lumped together. Second, it classifies institutions by only five very broad industries. For example, all institutions that are primarily mutual
fund complexes would be lumped into a single industry, even though they undoubtedly differ in
their investment styles.
Several studies have looked at investors in the context of style (momentum, contrarian, index) or
objective (growth, value). In this analysis, we are primarily interested in how investors respond to
information that they have, rather than how style affects returns. One simple indicator of whether
or not an investor is acting on perceived information is large shifts in holdings of stocks. One might
expect some trading for rebalancing purposes, but large trades are costlier. Therefore, investors
that shift their holdings a great deal from one quarter to the next are more likely to believe that
they have information, and behave aggressively.
Separating institutions based on position changes is cruder that separating institutions based on
style. For example, a high past return sends a signal to investors; some will want to buy, others
will want to sell. Both will change positions, and if they trade enough they will indicate that they
both feel more informed. Note that aggressive trading behavior is probably related to confidence
in strategy; whether or not it is truly related to an informational edge is an empirical question.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
113
It is likely that institutions that trade more aggressively do so because of a certain style of investing.
For instance, some mutual funds follow a heavily growth oriented, high transaction strategy. Managers of these funds may not necessarily be more informed than those who follow a low-turnover
approach. The latter will have fewer shifts in holdings from one quarter to the next, and so appear "less informed" by a turnover-based measure. In this case, we will not be able to distinguish
whether or not institutions act on recent information (about company fundamentals) or if they
follow one of the known predictable return patterns (such as the book-to-market, small stock, or
momentum effects). Previous research on mutual funds and institutions has found that the latter
is true. One could argue that such behavior shows institutions feel they are informed. However,
as will be discussed below, the most aggressive institutions do not appear to follow such patterns
alone. Furthermore, it is clear that aggressive institutions hope to profit from a certain type of
high-frequency information, so one can think about using transactions as a proxy for perceived
advantage in high-frequency information. l
Using this logic, I divide all institutions each quarter by two measures of aggressiveness based on
transactions within a quarter. The first measure is based on percent of the portfolio that was traded
in a quarter. For each manager, I first calculate the percentage of start-of-quarter portfolio value
that was traded during the quarter, where shares are assumed to have traded at the beginning of
the quarter:
Alt
N`P,t-l(Si,t1
Si,t-1)
(3.1)
where N is the number of stocks the manager owned at the start of the quarter, t is the end of
quarter date, and t - 1 is the start of quarter date, Pi,t is the price of a share of stock i at date t,
and Si,t is the number of shares of stock i held at time t. This statistic excludes all positions in
new stocks initiated during the quarter' I then calculate the percentage of end-of-quarter portfolio
value that was traded during the quarter, where shares are assumed to have traded at the end of
l°This approach is in a similar vein to that used by Bushee and Noe (2000). They identify three groups, one
of which is called "transient" because of their high portfolio turnover, high diversification, and momentum trading
style. However, as we shall see, aggressive institutions here are not directly analogous because they prefer other
characteristics and are not necessarily traditional momentum investors. Furthermore, Bushee and Noe focus on the
iiplications for volatility. This analysis focuses on whether or not institutions that are confident truly do better.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
114
the quarter:
_ S,,t_) 2
E=i Pi,tSi,t
This statistic excludes all positions that were sold to zero during the quarter.
A2t-
Z
P2 ,t
(3.2)
The measure of
aggressiveness, denoted WPOSt, is simply the average of these two statistics:
WPOS2t= Alt + A2t
(3.3)
I use the average since one cannot be sure if the manager traded shares at the start or the end of
the quarter.
This measure distinguishes institutions that trade aggressively from those that do not. Institutions
with a high WPOSt score have dramatically shuffled their portfolios relative to others, possibly
because of a liquidity shock, but more likely because they feel the need to act on perceived information within quarter t. Given transactions costs, a rational manager will probably shift holdings
a lot only to take advantage of a large opportunity. l l There are some disadvantages to using this
measure as a proxy for how informed an institution is. First, we assume that institutional managers
that believe they know something truly do. This will rank active fund managers very high, when it
is not clear that such managers are informed. Second, the measure is fairly crude, in that it could
rank an institution that makes a few large changes in holdings (for informational reasons) the same
as one that makes many small changes to many holdings (for instance, in portfolio rebalancing).
Nevertheless, the top ranking managers are still more likely to contain self-identified "informed"
managers. In each quarter t, I rank all institutions by WPOSt and divide them into quintiles. I
conduct analysis on the set of stocks that was owned by institutions within each quintile separately.
The second measure is based on the how much institutions changed their percentage ownership of
portfolio firms in a quarter. For each manager, I first calculate the change in percent of shares
outstanding owned for each firm the in the manager's portfolio. I then calculate the weighted
average of this number, across stocks, weighting by the start-of-quarter portfolio weights. This is
"Bushee (2001) uses a classification system based on portfolio turnover, diversification, and momentum trading
to separate institutions into transient, dedicated, and quasi-indexer groups. He uses this scheme to study whether or
not short-term investors influence firms' earnings management or research and development.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
115
equivalent to assuming the manager changed his ownership of all portfolio firms at the start of the
quarter:
Bit- Ei v I=1
Pi't-lSi,t-l (Sit/SOi't - Si't-1/sOi't-)2
-lt
EilV=Pi,t-1Si,t-l
(3.4)
where N is the number of stocks the manager owned at the start of the quarter, t is the end of
quarter date, and t - 1 is the start of quarter date, Pi,t is the price of a share of stock i at date t,
and Si,t is the number of shares of stock i held at time t, and SOi,t is the total number of shares
outstanding for firm i at time t. Again, this statistic excludes all positions in new stocks initiated
during the quarter. I create another statistic, this time averaging the change in ownership across
stocks using their end-of-quarter portfolio weights:
B2tB2t =
E/.
Z 1 Pi,tSi,t(Si,t/S0i,t
- Si,t-1/SOi,t-1)
EiN=1
Pi,t Si,t
2
-
(3-5)
Again, this statistic excludes all positions that were sold to zero during the quarter. The measure
of aggressiveness, denoted WOWNt, is simply the average of the two previous statistics:
WOWNt =
Blt + B2t
+ 2
(3.6)
A stock can have a large impact on WOWNt by either being a big proportion of the portfolio, or
experiencing a big change in ownership. This measure will distinguish institutions that want to
change their (possibly less liquid) positions from those that do not. Presumably such institutions
would only buy or sell large stakes in firms because they feel they have an informational advantage.
Each quarter, I split managers into quintiles based on WOWNt.
Note that since I re-sort institu-
tions each quarter, a manager could be classified as aggressive (top quintile) one period, and more
passive the next. 12
The first and second measures of aggressiveness are similar, in that they use institutional transactions to separate managers by how informed they feel they are. One advantage of the latter
measure is that it reduces the impact of small trades in large stocks, which are less likely to have
informational content. However, one drawback of this measure of aggressiveness is that it increases
' 2These dividing schemes give us six groups in each quarter: five sets of investors from most aggressive (by WOWN
or WPOS) to least, and individuals (individual investor ownership of a firm is one minus the share of institutions).
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
116
the influence of trades in smaller, more illiquid stocks. Since transactions in these firms are very
costly, it may make sense for institutions to publicly announce rebalancing trades, and to disguise
informational trades by trading smaller amounts. Therefore, the large changes in ownership that
WOWNt tracks could contain a mix of informational and liquidity transactions. This is less of a
problem for WPOSt because it is not weighted toward illiquid stocks. Managers ranking high on
WPOSt will face fewer problems of adverse selection when trading on information. Furthermore, a
high WOWNt score is likely to be biased towards larger institutions, since only they have enough
funds to trade substantial ownership stakes in many firms. The problem is that 13F data for these
institutions conceals different investment styles for various divisions and sub-managers.
Are managers stable over time in their aggressiveness groupings? If one believes that ownership
change by managers is related to their confidence in their ability or styles, then their rankings should
be fairly stable. Institutions rarely change their level of transactions from one quarter to the next.
By contrast, if one believes that informational advantage is transitory, a true ranking would be less
stable. To see if this is the case, I construct a transition probability table (Table 5) showing the
chances that an institution ends up in one aggressiveness quintile conditional on their starting in
another. Institutions sorted by WPOSt are in the left four columns, and sorted by WOWNt in the
right four columns. The probabilities do not all add to one, since some institutions drop out of the
sample for one reason or another. One can see from the table that institutions are fairly stable in
their classifications. Institutions that are highly aggressive have an 80% chance or more of remaining
in the top two ranks of our measure of self-identified informational advantage. The least aggressive
institutions, by contrast, are not as fixed in their categories over succeeding quarters. However,
they are highly unlikely to end up in the most aggressive classifications. Those institutions sorted
by change in ownership are more stable than those sorted by portfolio turnover. This suggests that
managers grouped by how much they change their ownership of firms retain their style of trading,
for reasons of either informational expertise or their own institutional characteristics (like size of
assets under management).
While institutions themselves are very likely to stay within their rankings from one quarter to the
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
117
next, they share the stocks they hold (not shown). For WPOSt, stocks that are held by the top
quintile have about an 80-90% chance of finding their way into the portfolios of each and every
aggressiveness quintile in all of the subsequent four quarters. This pattern continues until we reach
the lowest aggressiveness quintile, which holds stocks that are relatively more likely to remain in
the low WPOSt set. Stocks held by the lowest quintile have a 91% chance of staying in the group,
and a 76% chance of also being bought by the highest quintile. This tells us that active traders
hold firms that are attractive to all other groups as well. However, the most inactive managers
hold firms that are less attractive to high turnover institutions.
Stocks are highly mobile between WOWNt groups as well, with firms held by institutions in one
quintile at any quarter 70-90% likely to be found in other quintiles as well in subsequent quarters.
There is one difference, however. Stocks held by the top quintile of managers by change in ownership
are much more likely to remain in their set of stocks than to be owned by another quintile. In the
first quarter after ranking, stocks owned by the top aggressiveness quintile have a 92% probability
of remaining, but only a 72% chance of being picked up by the lowest quintile of managers. The
pattern disappears as we move down the rankings, until we find that stocks in the lowest quintile
of managers are almost equally likely to be found in other quintiles in later quarters. This tells us
that institutions that change their ownership of firms a lot concentrate their efforts in a group of
stocks that is not as attractive to other managers.
It is clear that WPOSt and WOWNt divide institutions into well-defined groups that are fairly
stable over time. The two measures of aggressiveness, however, differ in some ways. Highly aggressive institutions by portfolio turnover are more likely to concentrate on similar firms, while less
active institutions own a broader universe of stocks. On the other hand, highly aggressive institutions by ownership change are more likely to own a diverse set of firms, but less active institutions
avoid the firms that they hold. This pattern of ownership is somewhat surprising, since it suggests
that less active institutions by ownership change are relatively more concentrated in certain firms.
If managers who concentrate on a smaller universe have an informational advantage, the former
measure WPOSt may be better suited to capturing it.
_
I_
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMED INVESTORS
3.5.1
118
What motivates the transactions of different institutions?
In this section, I examine how stock characteristics influence ownership patterns for each of the
different aggressiveness groups. This time, the question we want to answer is: Do some institutions
respond to information faster than others? As in Table 4, I pay particular attention to how various
types of institutions respond to past returns at short and medium-term horizons, since returns are
a proxy for information revealed at those times.
We begin by looking at institutions separated by how much they shift their holdings within a
portfolio (WPOSt). I construct a measure of how institutions within a given quintile changed their
holdings of a particular stock. For each firm in each quarter, I aggregate the share of ownership
by quintile institutions. This gives me a time series of ownership statistics for the stock, which
roughly corresponds to how much of the firm was owned by aggressive institutions at each quarter:
Qt
OQj,t= Esij,t/sOj,t
i=l
(3.7)
where OQj,t denotes the ownership of stock j by quintile firms at quarter t, Qt is the number of
institutions in the aggressiveness quintile defined in quarter t, and Sij,t and SOj,t are the shares
held by institution i and shares outstanding of firm i at quarter t, respectively. The change in
ownership by quintile institutions over a quarter is simply this statistic minus its lagged value.
Note that the constituents of an aggressiveness quintile change over time. Therefore, this measure
of quintile ownership change does not the actually show how informed institutions at a given time
changed their ownership of a firm. It shows rather how a changing but well defined group of active
institutions altered their holdings. This avoids the difficulty in identifying and tracking the holdings
of numerous managers multiple times for a stock. Since most institutions are fairly stable across
quarters, the approximation should be adequate.
Table 6 shows the time-series average coefficients (multiplied by 100) for a cross-sectional regression
of OQj,t by institutions in each quintile on the same firm characteristics and past returns as in Table
4. The specification for most active institutions is at the top, and the least active at the bottom.
All t-statistics are calculated using Newey-West standard errors that account for overlapping time
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
119
series of change in ownership.
Various patterns are apparent when one looks across quintiles. All groups prefer larger firms,
although there is no clear monotonic relation among institutions from most to least aggressive. More
aggressive managers prefer less liquid firms, although this effect is economically small. However,
the largest differences between institutions appear in the last half of columns. Almost all of the
preference for value stocks comes from highly aggressive managers. Most importantly, however, is
the strong positive response that aggressive managers have towards returns in the previous month t.
A 10% return in the past month causes top quintile institutions to increase their percent ownership
by 0.14% in the first quarter. This small response is actually very laxge when compared to the
responses to other calibrated variables. Furthermore, there is a clear monotonic relation as one
moves down the quintiles. This is consistent with the interpretation of this coefficient as reflecting
a response to information. If more aggressive firms are quicker to act on information, then their
coefficients should be larger than those of less aggressive firms. Furthermore, the impact of returns
should be more immediate. We can see that this is the case in Table 6; the response of the top
quintile is large in the first couple quarters, and in fact diminishes afterwards. The coefficients for
the fourth quintile are also larger, but take longer to increase, and so on until the bottom quintile
has almost no response to recent returns.
The coefficient for past returns from month t - 13 to t - I is also consistent with a pattern of
information diffusion across institutions. The most aggressive firms trade against momentum, but
only after waiting at least four months. The next few quintiles, however, display a lagged response
similar to that for the one month past return. Given a 30% return from month t- 13 to t - 1,
the fourth quintile buys an extra 0.08% percent ownership in the first quarter, and peaks at 0.12%
ownership by the third quarter.
The third quintile is slower to buy, but peaks at 0.14% by the
fourth quarter. The fourth quintile shows even less response, but manages to buy 0.06% by the
fourth quarter. Finally, the bottom quintile does not respond in an economically meaningful way
to 12-month returns. Turning to the effect of other institutions, all quintiles of managers sell stocks
that have high overall levels of institutional ownership. The top quintile is most aggressive in doing
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
120
so. Given a firm with 50% institutional ownership, this group will sell 1.6% of its shares by the
next year.
Table 7 shows the same analysis for the set of institutions divided by change in percentage ownership
in firms (WOWNt).
The results are generally similar. Compared to those in Table 6, top quintile
institutions here show a much greater preference for large firms. Given a stocks with a billion dollar
market capitalization, top quintile managers will buy nearly 2% more of it after six months, and
2.8% by a year. This is much larger than the extra 1% of shares outstanding that top quintile
WPOSt institutions buy after 12 months. It is even more remarkable when considering that all
lower WOWNt
quintiles buy less of large firms when compared to their WPOSt counterparts.
Aggressive WOWNt institutions also show more of a preference for high-priced stocks, and slightly
less for value stocks. The same pattern of response to past one-month and 12 month returns is
evident, although top quintile stocks do not trade against momentum, and lower quintiles do not
trade as strongly with it. Finally, the most aggressive institutions in Table 7 sell high-institutional
ownership stocks more strongly than those in Table 6, ceteris paribus.
These tables confirm that the aggressiveness sorts do capture meaningful differences between institutions. Highly active managers react more strongly to recent returns than less active ones, by
both measures. The results are generally supportive of the idea that aggressive institutions respond
to recent signals. They do not appear to be pure momentum traders. The least active managers in
quarter t are less responsive to all firm characteristics, as expected. The major difference between
the two measures of aggressiveness is that highly active firms by WOWNt
prefer larger, higher
priced stocks than those by WPOSt.
3.5.2
Are more aggressive institutions managers trading on information?
So far we have seen that aggressive institutions react to very recent returns more than less aggressive
ones. This suggests that they are not purely momentum traders (nor are they solely devoted to
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMED INVESTORS
121
other predictable return patterns).13 Are the aggressive institutions indeed trading on information?
One way to find out is to look at the performance of the stocks they buy and sell. I repeat the
analysis of Table 3, where I mimic an institutional trading strategy by ranking all stocks by change
in ownership in a quarter. I buy the top tenth of stocks and sell the bottom tenth. I do this
for each quintile of institutions by WPOSt and WOWNVt and rank by change in ownership by
quintile managers only. The goal here is to see if they more aggressive institutions really are better
informed, rather than herding in an uninformative manner on past characteristics.
If so, they
should earn economically large and sustained profits. This does not mean that such institutions
are consistently profitable after transactions costs, although that is a possibility. Recall that the
measures of aggressiveness allow institutions to move between categories over quarters.
Table 8 shows the returns for an institutional strategy for each aggressiveness quintile by WPOSt.
The long-short cumulative size and B/M adjusted returns are in the first column, and the top (long)
and bottom (short) deciles are in the next set of columns. One can see immediately that the pattern
of returns in almost monotonically descending the less aggressive the set of institutions. The most
aggressive managers buy stocks that did well in the past month, and earn steadily growing profits.
On the other hand, the least aggressive institutions have basically fiat performance throughout
the subsequent twelve months. This supports the idea some institutions that believe they have
information truly do. Table 9 shows the same returns for institutions split by the measure of
ownership change, WOWINt. Aggressive institutions here also buy and sell stocks with high and
low past one-month returns, respectively. However, so do less aggressive institutions (although to
uneven degrees). Also surprising is the fact that not a single group of WOWNt institutions earns
reliable or sizeable profits. Although WOWNt does seem to be related to past one and 12 month
returns, it does not appear to be related to informational advantage.
The evidence on profitability of trades by aggressive institutions gives support to the idea that 1)
some investors are more informed than others, 2) they reveal some of their information in their
13In fact, Jegadccsh (1990) shows that a one-month past return strategy reverses the following month. Since the
most aggressive institutions respond to one-month returns and not to twelve-month momentum, one would expect
that they would lose money in this respect.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
122
transactions, and 3) thereby affect prices such that they speed information flow in the stock market.
In addition to the fact that individuals are not informed, those institutions that trade more (arind
therefore think they know something in a given quarter) appear more informed than those that
trade less. However, this statement is only true for the aggressiveness measure WPOSt, which is
based on the fraction of portfolio value that is shifted. The measure WOWNt is similar but does
not indicate that institutions that shift their ownership share of firms do so because of information
about future performance. Such changes in ownership might be related to liquidity trades, or
confined to large and diverse institutions that trade for numerous conflicting reasons.
3.6
Conclusion
Research into information diffusion across stock prices has shown that there are lags in predictable
returns. These patterns seem stronger in some situations than in others, implying that frictions or
other constraints may play a role. Nevertheless, such frictions only serve to highlight differences
between investors. Some are highly informed, and likely to arbitrage away any opportunity. Others
are less informed, and the likely source of opportunities in the first place. Who is the more informed
investor? What information do they act upon? And do they flatten the incidence of predictable
returns by speeding information flow? This analysis takes a step towards answering those questions.
First, I find that far from reducing one-month and 12-month return autocorrelation, institutions as
a group increase it. Although past research shows they are more informed, they do not eliminate
predictable patterns. Second, this occurs because institutions respond to past returns. Since their
trades are profitable, this suggests that they do so because of information that was revealed, and
not simply because they are feedback traders. Third, institutional investors that would identify
themselves as informed ex-ante appear to be so. Those institutions that are more aggressive in
their trading for a given quarter (and therefore are more confident in their information) respond
even more strongly to recent returns, and trade even more profitably. This provides another piece
of subsample evidence that some investors have an informational advantage, and that they release
their information into prices before the rest of the market fully incorporates it.
CHAPTER 3. HIGHLY, SOMEWHAT, AND UN-INFORMEDINVESTORS
123
In answer to the questions posed earlier in the section, we can say that aggressive institutions
appear to be more informed. They use information from the recent past, some of it publicly
revealed in prices. However, they do not flatten the incidence of predictable returns. In fact,
whatever "speeding up" they account for is not enough to get prices to where they should be
several period later. This leaves us with a question: if aggressive institutions spread information
about stock valuations, but only over a period of several quarters, why are other market participants
slow to follow?
CHAPTER 3: TABLES AND FIGURES
e:
cc
cf)
124
- fl
u~ c
rCO
V)
O
C
t
OC .0
OD
N
04
T
LO
CO
O
LO
CY
CD
O0
co
04 v-
O -3
cn
c
b
aa
.- ?
'
')
cu
U)
az
~r~
~~b~
CU~d
c
(D
w
v--
t--
O
II-
I ',r
cs
Cm co~ 0
CM
-N
C
O
t-
0
v
ff-
fr
"
C'
6
C~
i'.,.
C~
.C3 ~
CO
".
Pc
a CC)
C
I",.
Cf)
PO
a)
-
oZ
T-
-
CM
O 0C 4
<#
._N
m
-
0
co II)
o
.Nowc
{)E zti5
O~ c,0
~
o-6
CD
Co
nO
E 0N'DE
'o
L-
C
o
Ct
LOv- CO
covco
N
UC) C3~a
O
)
CM~~
a
CM
=
~1'
"
cD CDcf
C
o
)
)
v-
Ue) I-CO
r
r_
O-
Cr
C
Cc
C
v4r- I-s
CN
O
CD
*f U
Nco co
CM CD
N
CO
·-Oe
e
,-
co
m
e O t 0 0 OCOs OnnnON~~~cl
zj~aa)
E
No~
a
ti
0
CY _,
a) CO
3
M
'It
C
00
V"
0
C
O
r
O
O
CC
CI
s0
C6
>~t
_-
= oz.
EE
c§
.o
:3
I-
0
-)
CL
o
o ^M
G.)
(0
-Co
-
E
E
Q
0
ocp
-
-
C
"
fn
a,
cm
2en .-
Q)
0
co
a,
3
C
:3
a,
c'
- - 0
co
w
E E-
0-
E ° c"
m)
f
r2?
iw
E ° in
coEO
ZD
cu
E
·cc
-2
E
OM,
1!
E f~
j
Ie
cr
V5
C
C
i
G,
C
CHAPTER 3: TABLES AND FIGURES
126
Table 2: Impact of Institutional Ownership on Return Predictability
This table shows the time-series averages of cross-sectional regression coefficients. The dependent
variable is cumulative returns over various horizons, starting one week after the end of March, June,
September, and December for each year, 1980-2000. The regressors (listed below) are various firm
characteristics at the end of the past quarter, and returns measured over preceding months. Newey-West
standard errors are used to calculate t-statistics for the 2, 3, and 4-quarter cumulative return
specifications, since the dependent variables overlap across quarters.
Variable Definitions
MV
turn
turn2
price
B/M
Ret_0
past12
10
Natural log of past month's market value, in millions
NYSE and AMEX share turnover in month, 0 otherwise
NASDAQ share turnover, 0 otherwise
Natural log of price at end of past month
Natural og of book/market ratio in millions at end of past month
Return in past month
Return over 12 months before past month
Level of institutional ownership, in percent of shares outstanding
Horizon: intercept
3 mos.
6 mos.
9 mos.
12 mos.
MV
turn
tum2
price
B/M
Ret_0
past12
10
0.04
0.00
-0.08
-0.06
0.00
0.01
Q.Q0
0.02
0.04
2.46
-2.74
-4.20
-1.84
-0.20
3.25
0:38
5.71
5.68
Ret_0 past12
*10
*10
0.05
0.04
0.00
-0.08
-0.06
0.00
0.01
-0.01
0.01
0.03
0.08
0.04
2.53
-2.76
-4.24
-2.01
-0.22
3.48
-0.65
3.63
3.96
2.40
4.30
0.11
0.00
-0.18
-0.13
-0.01
0.01
0.03
0.03
0.06
3.16
-2.19
-5.87
-2.62
-0.98
3.62
1.74
5.13
6.03
0.12
0.00
-0.18
-0.14
-0.01
0.01
0.01
0.02
0.04
0.19
0.09
-2.19
-5.89
-2.79
-0.97
3.84
0.33
3.16
4.09
4.10
4.34
0.19
-0.01
-0.25
-0.18
-0.02
0.02
0.08
0.04
0.08
-1.60
-5.61
-2.61
-1.29
3.68
2.98
4.11
5.30
0.05
0.05
3.20
3.36
R2
0.05
0.05
0.19
-0.01
-0.25
-0.19
-0.02
0.02
0.03
0.02
0.06
0.33
0.11
3.38
-1.60
-5.54
-2.69
-1.27
3.94
1.20
2.29
3.92
4.52
4.51
0.26
3.46
-0.01
-1.38
-0.33
-5.73
-0.23
-2.51
-0.03
-1.26
0.03
3.38
0.12
3.71
0.04
3.54
0.09
4.95
0.05
0.05
0.26
-0.01
-0.33
-0.24
-0.03
0.03
0.06
0.01
0.07
0.40
0.15
3.50
-1.39
-5.67
-2.58
-1.25
3.65
2.03
0.92
3.71
4.82
4.27
0.05
CHAPTER 3: TABLES AND FIGURES
127
Table 3: Returns to a Strategy that Follows Extreme Changes in Institutional Holdings
This table shows characteristics of a strategy that ranks all stocks by percent change in institutional
ownership over each calendar quarter from 1980-2000. Stocks in the top and bottom deciles
experienced the greatest amount of institutional buying and selling, respectively. Panel A shows
number of stocks in the top and bottom deciles, number of industries represented, as well as crosssectional averages of end-of-quarter market value, return in the preceding month, percentage level of
institutional ownership, and change in institutional ownership in the last quarter. Panel B shows the
cumulative (summed) monthly returns over various horizons to a strategy of buying and selling top
and bottom decile stocks, respectively, in equal-weighted portfolios. The rolling portfolio methodology
is used to calculate averages and t-statistics.
Panel A: Summary Statistics
Year
Count
Avg. Mval
RetO
No. Inds
Avg. 10
CIO
Bottom Decile by Change in 10 (Institutional Selling)
1980
1984
1988
1992
1996
2000
215
404
453
461
604
558
721
456
403
547
605
1,443
-0.01
0.02
0.04
0.03
0.01
-0.03
19
20
20
20
19
20
0.27
0.24
0.26
0.32
0.29
0.35
-0.09
-0.07
-0.09
-0.09
-0.15
-0.12
Top Decile by Change in 10 (Institutional Buying)
1980
1984
1988
1992
1996
2000
214
403
452
460
603
557
369
349
570
540
1,291
1,834
-0.02
0.02
0.04
0.06
0.03
0.07
19
20
20
19
20
20
0.28
0.32
0.41
0.45
0.52
0.54
0.07
0.07
0.09
0.11
0.17
0.13
Panel B: Strategy Returns
Month
0
1
3
6
9
12
__
Long-Short
Average
t-statistic
2.43 %
-0.03
0.67
1-55
1.76
1.78
8.35
-0.22
2.08
3.45
3.42
3.24
Top Decile (Long)
Average
t-statistic
2.80 %
-0.10
0.64
0.88
0.89
0.59
5.31
-0.74
2.16
2.25
1.94
1.13
Bottom Decile (Short)
Average
t-statistic
0.37 %
-0.07
-0.03
-0.67
-0.87
-1.19
0.81
-0.47
-0.15
-2.23
-2.86
-3.29
I
CHAPTER 3: TABLES AND FIGURES
a,
.)>D
VsQ
O
_
128
o
0o
CDC
CO
o
o
C
o
o
0
,
o
o
o
o
corwCoC
CD
<n
cn
0)CD
o
3
a,
CMM~
Cao aC
E
.aUcCo,
rm o 4)
CO
c
0)
Em~~
0C)X CD0m
Yao
ao
2C,
* :,=
0C
C
o
0
E -
755
4
.,
)d
cEi
COn
aa ,
C
cO
_t
CO
C
I-
- _
cD V a2
-
Emo
o>o~
rB
)
-
._
ag3
C0 ~
W
CO >
C O 0)
CO CO
> CO
(L)E
jn
E
r M z~3
C
0
C
.2
a
4-m
o a
-
w
D
E
40 oL CO
<.CL.: o
c_
VC: C
S.
'EQ
0
coo
co
0
-
C'Jo.CE]
E
0)
0) 0) CO
LuE a. .2. C
CL.
2
..
Cl
5
c:
o:
C)
o:
o~I
2XsO6C6j2X C
C)ev
o6
o
o0 .
on
oC 2
o6
oCD
6
d06
Uc
ro
C
r- N
o0 .c
o0.
0o q.
am
C
;
CO
o
0
.c
_m
6o
o
N
0.C
o iZ8 o V)
o
to
C)
i)
C)
6
o
0
o
to
o
1.
(U> ,< co
o woo <*
oo
od
S0
0Co
aczC
=C
o
CM
CL 4Br
I- 0 .
0'
Co
6 ,
C - CO CO
C.-
> 0 C OC.
*,
EC
;0CO
QO
o>o
X:
6 co
C
L
~0
oa
°gLe
ce3 ovg tq
O
(U
z
C O )E0 ~ a
0
.
cj
21
·
COz
co SCr"·:0
r CO
CO
Us
600 ° w E
Ci)
oE~ O,~2 CO3
4)0
75
0
) aD
CD
CO. c XOOz
>OC D
0.
.0~ c
C
C
ao
4-'tj
cU
a_
0
e)CCOO)U)
o
0~ Q
d
RC
C
a,
C co5E ~
)
r(D
..
- co.
C
U)
.
CD%
c
4)
*!
dCV)
6
C
0
"s E ._
c 0 e C
1
o
I
o C, b .
0
C CV
0.0 . 0.
C
E
CD
0n
~t
C;°
3
t:
.o
Ci
od
.
o
o0D
rloS
a LO sc
o
o
C;
o
I0 N
0 o
Jo,
oN
o0 vi
0
cr
·ct;
S
c0
N
0.
cu
CHAPTER 3: TABLES AND FIGURES
I)
c) c
o
.-
E
i cv
,
E
c3
O
CO c%0 0
LO COO) N
00000"
,-
-
r- 0)
co
C_
N C) N) 0 O
O
0
0000
D
8
>
LO
O
.
0
.s
0.
0000N
O Cl0
00
OmO0
co 00
ooooo
CO O CO O LO
0
_
CO
IU)
0N T0 CD
OC
CO0 O O O
0 I 0 00
m
O0 0
E
-
o N
00o
0o
_
CV0
00
0
C
00000O
r- 0)
0D
)
LO)
00 0
r0000
.C o-.
OO
-o
a0
0
0,
00) t
O O O CS It)
M_
2C.
=
tn'l~
0
3
_
0c'
. 0w
C
EE
0
. .
mW
Cco
C
0D00
ccoc
0
Co - 0
OO¢,ON
1
NO Co
CDN C0) 0O
CO C) CO
aaaaa)
Cil CD t
c
1tN1-00
00 0 0)
1CN CD N0
dd d
CD
0
00C) '- o 0)
0 O
E
co
O O cocn
O v
° Z
m~I
O 0 00CD
COC)l0
0 00 0
C
a-
00 000
O O O O O
a-
,
L. Im~
cc;
t) C)
Co LO
CV ) CN 00
00 00 0
w
= 0
C
co
0
o C
U) CO)
Ln
U,
0to
00- t N
(n
o
a
r 0O
129
C
CC O
O O aO O
00 000
000
'I
amdd
=C
C
I
o
e
'p 0
;I.
0
=n 0a
_
)
c
LO
O
Ca
CY)
C C)
00000
a
'..
a
aaa
a 00O~rQ
0U·-O~
* N
0)
|_ =
0
00
Cn
CO
03
ccn
0)
0
c')
a
0
CHAPTER 3: TABLES AND FIGURES
130
Table 6: Determinants of Changes in Ownership by Institutions Sorted by Value of Shares Traded
This table shows time-series average coefficients (multiplied by 100) of a cross-sectional regression of
cumulative change in institutional ownership on stock characteristics in the previous quarter. Data are from
each March, June, September, and December month-end in 1980-2000. Newey-West standard errors are
used to calculate t-statistics (in italics below coefficients) for 2, 3, and 4-quarter specifications.
Horizon: intercept log(MV)
turn
Quintile 5 (Most Aggressive)
0.10
-0.56
1 qtr.
-0.24
-1.39
2 qtrs.
3 qtrs.
4 qtrs.
4.10
-1.51
turn2
log(price) log(B/M)
retO
cume
10
volatility
R2
0.04
-0.24
0.00
0.07
1.45
-0.02
-1.10
1.29
-1.68
0.09
1.97
10.43
-0.31
-3.96
0.58
0.03
0.12
-0.75
-0.50
-0.03
0.12
1.83
-0.24
-1.79
3.36
0.17
4.69
-1.66
-2.81
-0.50
2.67
10.70
-3.05
-6.39
1.17
0.23
0.16
-0.98
-0.63
-0.05
0.17
1.66
-0.46
-2.56
1.11
1.23
5.13
-2.05
-3.25
-0.76
3.45
8.64
-5.82
-8.59
0.35
0.52
0.15
-1.01
-0.84
-0.04
0.24
1.36
-0.77
-3.12
1.91
2.26
4.17
-1.61
-4.62
-0.52
4.98
6.32
-8.60
-10.23
0.50
0.05
0.05
0.06
Quintile 4
1 qtr.
2 qtrs.
3 qtrs.
4 qtrs.
-0.60
0.13
-0.60
-0.04
0.01
0.00
1.20
0.26
-0.77
3.70
-2.53
3.37
-1.47
-0.24
0.22
-0.08
7.96
4.33 , -1.82
1.47
-0.43
0.15
-0.75
-0.02
0.03
0.10
1.77
0.34
-1.73
4.97
-1.58
3.25
.1.54
0.00
0.60
1.89
9.94
5.09
-4.57
1.53
-0.35
0.20
-1.15
-0.12
-0.01
0.15
1.99
0.38
-2.51
8.82
-1.21
4.15
-2.75 -0.45
-0.08
2.61
9.45
5.34
-6.63
2.30
-0.01
0.20
-1.53
-0.22
-0.05
0.19
2.19
0.30
-2.87
8.02
-0.02
3.99
-3.28
-0.75
-0.64
2.71
10.35
3.82
-7.11
2.03
Quintile 3
-0.52
1 qtr.
0.12
-0.49
0.00
0.01
-0.05
0.05
0.13
-0.49
-1.52
-1.85
2.83
-1.53
0.02
0.13
-0.78
0.29
2.24
-1.05
-0.77
2 qtrs.
3 qtrs.
4 qtrs.
-0.80
0.17
-0.88
-0.08
0.11
-0.03
0.42
0.28
-1.27
2.68
-2.62
3.57
-1.77 -0.46
1.86
-0.40
2.15
3.84
-2.86
1.03
-0.90
0.22
-1.06
0.09
0.13
-0.05
0.70
0.42
-1.79
1.38
-2.66
3.86
-1.72
0.57
1.96
-0.67
3.43
5.49
-3.55
0.42
-0.73
0.20
-1.28
-0.07
0.21
0.00
1.11
0.46
-2.44
3.90
-1.84
3.17
-1.74
-0.42
2.87
-0.05
4.56
5.13
-4.19
0.96
0.05
0.04
0.04
0.04
0.04
0.04
0.04
0.04
Quintile 2
1 qtr.
2 qtrs.
3 qtrs.
4 qtrs.
-0.06
0.01
-0.07
-0.01
0.05
-0.39
-0.06
-0.34
0.21
-1.38
-0.08
0.82
-1.04
-0.03
-0.54
0.11
-0.41
0.00
0.08
-0.10
-0.09
0.08
-0.68
0.59
-1.48
1.98
-1.08
-0.02
1.39
-1.88
-0.57
1.37
-1.59
0.24
-0.63
t0.13 -0.65
0.04
0.15
-0.12
0.10
0.15
-1.08
1.71
-1.66
2.31
0.21
2.45
-2.29
0.60
2.74
-2.16
0.57
-0.38
0.09
-0.73
0.00
0.20
-0.09
0.14
0.19
-1.11
1.39
-0.92
1.33
-1.80
-0.01
3.29
-1.39
0.89
2.96
-2.29
0.49
-1.31
0.16
0.21
0.32
0.08
-0.02
-0.20
-0.09
-1.81
-2.64
2.55
0.67
2.60
1.41
-0.41
-1.51
-1.63
-4.10
-1.42
4 qtrs.
-
-0.30
-0.73
-1.17
3 qtrs.
,
0.11
1.72
Quintile 1
-0.45
1 qtr.
2 qtrs.
I
-0.54
-1.54
-.
-
-0.33
0.15
0.00
0.23
0.06
-0.07
-0.19
-0.04
-1.46
-4.39
-0.95
2.24
0.01
1.43
0.93
-1.14
-1.51
-0.69
-3.68
-1.64
-0.25
0.16
0.10
0.23
0.09
-0.07
-0.15
-0.02
-1.71
-6.07
-0.74
2.37
0.26
1.62
1.07
-1.08
-1.15
-0.29
-4.13
-2.67
-0.35
0.15
-0.10
0.25
0.15
-0.09
-0.26
0.00
-1.75
-2.92
-0.96
2.38
-0.26
1.69
1.62
-1.14
-1.69
0.00
-3.98
-1.15
-
-
-
0.05
0.04
0.04
0.04
0.05
0.05
0.05
0.05
CHAPTER 3: TABLES AND FIGURES
131
Table 7: Determinants of Changes in Ownership by Institutions Sorted by Change in Ownership
This table shows time-series average coefficients (multiplied by 100) of a cross-sectionalregression of
cumulative change in institutional ownership on stock characteristics in the previous quarter. Data are from
each March, June, September, and December month-endin 1980-2000. Newey-West standard errors are
used to calculate t-statistics (in italics below coefficients)for 2, 3, and 4-quarter specifications.
Horizon:
intercept
turn
log(MV)
tum2
log(pnce) log(B/M)
retO
cume
10
volatility
R'
0.03
Quintile 5 (Most Aggressive)
1 qtr.
2 qtrs.
3 qtrs.
4 qtrs.
-0.41
0.17
-0.50
-0.16
0.11
0.00
1.29
0.17
-2.44
2.03
-1.78
3.44
-1.83
-0.92
2.25
0.06
9.04
2.80
-5.47
1.07
-0.47
-1.93
-0.41
0.28
-0.33
-0.02
0.18
0.10
1.77
0.15
-4.47
2.65
5.44
-1.06
-0.08
2.99
1.34
8.80
1.85
-8.38
1.04
0.36
-0.29
0.03
0.23
0.16
1.82
0.07
-5.78
2.39
-1.65
6.43
-0.85
0.09
3.16
1.99
7.32
0.77 -10.50
0.83
-0.28
-0.95
0.40
-0.21
0.01
0.28
0.20
1.82
-0.07
-6.86
3.27
6.13
-0.52
0.04
4.08
2.35
6.22
-0.59 -13.03
1.06
0.04
0.04
0.04
Quintile 4
1 qtr.
2 qtrs.
-0.06
0.06
-0.52
-0.10
0.02
-0.03
0.50
0.14
-0.63
-1.18
-0.22
1.20
-1.83
-0.86
0.35
-0.56
3.47
3.06
-1.59
-0.50
-0.08
0.08
-0.82
-0.11
0.14
-0.01
0.95
0.14
-1.55
0.20
-0.26
1.60
-2.44
-0.69
2.14
-0.22
5.16
2.41
-3.55
0.08
3 qtrs.
0.12
0.07
-0.84
-0.13
0.19
0,00
0.87
0.17
-2.13
-1.09
1.29
-2.22
-0.60
2.61
-0.06
5.18
1.90
-4.21
-0.34
4 qtrs.
0.41
0.51
1.62
0.05
-1.07
-0.18
0.21
C.01
1.03
0.16
-2.63
-1.12
0.93
-2.17 -0.69
2.94
0.12
6.03
1.20
-4.84
-0.40
0.04
0.04
0.04
0.04
Quintile 3
1 qtr.
-0.29
0.09
-0.24
-0.06
-0.03
-0.03
0.21
0.06
-0.22
-3.21
2.66
-1.05 -0.55
-0.92
-0.96
1.97
1.75
-0.92
-2.04
2 qtrs.
-2.01
-0.34
0.10
-0.29
-0.09
0.01
-0.02
0.20
0.11
-0.63
-1.62
-1.92
2.73
-1.04
-0.77
0.33
-0.35
1.63
2.96
-2.40
-0.97
3 qtrs.
4 qtrs.
-0.45
0.13
-0.32
-0.05
0.03
-0.02
0.50
0.15
-0.94
-0.43
-2.22
3.25
-1.09
-0.33
0.81
-0.36
5.28
3.92
-3.56
-0.31
-0.28
0.12
-0.28
-0.06
0.05
0.00
0.69
0.12
-1.06
-1.02
-1.37
2.71
-0.87
-0.42
1.09
-0.08
4.33
3.19
-3.56
-0.65
Quintile 2
1 qtr.
-0.42
-2.92
2 qtrs.
-0.42
-2.79
3 qtrs.
4 qtrs.
0.08
0.11
0.20
-0.02
-0.02
0.02
-0.01
-0.26
0.24
3.35
0.59
2.17
-0.55
-0.83
0.21 -0.14
-1.50
0.16
0.08
0.22
0.21
0.02
-0.02
0.11
0.02
-0.47
0.70
3.20
1.02
2.21
0.65
-1.23
1.29
0.85
-2.28
0.34
-0.50
0.09
0.17
0.22
0.02
-0.03
0.14
0.06
-0.61
1.51
-2.97
3.30
0.78
2.12
0.56
-2.00
1.42
2.17
-2.61
0.60
-0.48
0.09
0.02
0.18
0.06
-0.03
0.13
0.09
-0.79
2.20
-2.75
3.12
0.10
1.63
1.57
-1.78
1.28
2.43
-3.36
0.76
0.03
0.03
0.03
0.04
0.03
0.03
0.03
0.03
Quintile
1 qtr.
-0.66
0.15
0.24
0.22
0.05
-0.03
-0.06
-0.04
-1.58
0.06
4.94
0.92
1.94
1.42
-0.95
-0.48 -1.02
-6.42
0.03
2 qtrs.
-2.71
-0.51
0.12
0.09
0.14
0.00
-0.04
0.09
0.01
-0.90
-0.49
-2.56
3.91
0.46
1.42
-0.11
-1.34
0.81
0.19
-3.39
-0.31
3 qtrs.
4 qtrs.
-0.62
-2.64
-0.65
-2.59
0.14
-0.16
0.04
0.02
-0.03
0.03
-0.02
-1.17
-0.91
4.54
-0.84
0.38
0.43
-1.02
0.21 -0.39
-4.32
-0.46
0.12
-0.06
0.06
0.07
-0.04
0.02
-0.05
-1.37
2.27
3.73
-0.26
0.63
1.48
-1.10
0.12
-0.97
-5.40
1.09
0.04
0.04
0.04
0.03
CHAPTER 3: TABLES AND FIGURES
133
Table 8: Returns to a Strategy that Follows Extreme Changes in Holdings of Institutional
Managers, Separated by Aggressiveness of Intra-Quarter Transactions
This table shows returns to five strategies that rank all stocks by percent change in ownership over each
calendar quarter from 1980-2000. Each quarter, all institutional owners are divided into quintiles by how
aggressively they change their position weights. For each ownership quintile, stocks are separately
ranked by change in the value of the shares they traded. Stocks in the top and bottom deciles
experienced the greatest amount of buying and selling, respectively. Each section shows the cumulative
(summed) monthly returns over various horizons to a strategy of buying and selling top and bottom
decile stocks, respectively, in equal-weighted portfolios. The rolling portfolio methodology is used to
calculate averages and t-statistics. Returns to tracking the buying/selling of highly aggressive institutions
are shown at the very top, while returns to tracking the trading of the least aggressive institutions is
shown at the very bottom.
Aggress.
Quintile
Month
Long-Short
Average
t-statistic
Top Decile (Long)
Average
t-statistic
Bottom Decile (Short)
Average
t-statistic
5 (most)
0
1
3
6
12
1.65 %
0.07
0.78
1.54
2.29
6.27
0.50
2.49
3.64
4.46
2.55 %
0.13
0.82
1.53
1.61
4.95
0.96
2.53
3.45
3.08
0.90 %
0.06
0.04
-0.02
-0.67
2.03
0.66
0.29
-0.08
-2.02
4
0
1
3
6
12
1.46 %
0.01
0.67
1.10
0.80
5.06
0.07
2.60
2.53
1.59
2.37 %
0.00
0.70
1.15
0.75
4.64
-0.02
2.49
2.60
1.42
0.91 %
-0.01
0.02
0.05
-0.05
2.05
-0.11
0.14
0.21
-0.17
3
0
1
3
6
12
0.49 %
-0.06
0.16
-0.09
-0.04
1.64
-0.50
0.60
-0.25
-0.07
1.43 %
0.00
0.32
0.21
0.03
3.08
-0.02
1.32
0.58
0.05
0.93 %
0.06
0.16
0.30
0.06
2.11
0.57
0.91
1.42
0.21
2
0
1
3
6
12
-0.19 %
0.00
0.28
0.02
-0.48
-0.84
-0.03
1.23
0.07
-0.94
1.07 %
0.03
0.46
0.08
-0.39
2.52
0.19
2.31
0.29
-0.88
1.26 %
0.03
0.18
0.06
0.09
2.98
0.43
1.32
0.29
0.29
1 (least)
0
1
3
6
12
-0.24 %
-0.13
-0.22
-0.10
-0.24
-0.96
-1.06
-0.96
-0.32
-0.54
1.12 %
-0.06
0.25
0.22
-0.10
2.65
-0.60
1.59
0.85
-0.26
1.35 %
0.08
0.47
0.32
0.14
2.93
0.74
2.45
1.20
0.39
CHAPTER 3: TABLES AND FIGURES
135
Table 9: Returns to a Strategy that Follows Extreme Changes in Holdings of Institutional
Managers, Separated by Aggressiveness of Intra-Quarter Changes in Ownership of Firms
This table shows returns to five strategies that rank all stocks by percent change in ownership over each
calendar quarter from 1980-2000. Each quarter, all institutional owners are divided into quintiles by how
aggressively they change their position weights. For each ownership quintile, stocks are separately
ranked by change in ownership by these institutions. Stocks in the top and bottom deciles experienced
the greatest amount of buying and selling, respectively. Each section shows the cumulative (summed)
monthly returns over various horizons to a strategy of buying and selling top and bottom decile stocks,
respectively, in equal-weighted portfolios. The rolling portfolio methodology is used to calculate
averages and t-statistics. Returns to tracking the buying/selling of highly aggressive institutions are
shown at the very top, while returns to tracking the trading of the least aggressive institutions is shown at
the very bottom.
Aggress.
Quintile
5 (most)
Month
0
1
3
6
12
Long-Short
t-statistic
Average
1.08 %
4.52
Top Decile (Long)
t-statistic
Average
2.05 %
4.52
Bottom-Decile (Short)
t-statistic
Average
0.97 %
2.10
-0.06
0.15
0.01
-0.09
-0.51
0.64
0.05
-0.16
0.03
0.34
0.25
-0.13
0.22
1.37
0.72
-0.25
0.09
0.19
0.24
-0.04
0.80
1.07
1.10
-0.11
4
0
1
3
6
,-12
0.21 %
-0.20
0.30
0.00
-0.15
0.94
-1.47
1.15
0.01
-0.27
1.32 %
-0.15
0.33
0.11
-0.47
3.06
-1.16
1.55
0.34
-0.90
1.11 %
0.06
0.03
0.11
-0.32
2.53
0.51
0.16
0.45
-1.12
3
0
1
3
6
12
0.44 %
0.08
0.36
0.51
-0.12
1.94
0.58
1.68
1.79
-0.28
1.44 %
0.08
0.54
0.53
0.05
3.19
0.63
2.24
1.64
0.11
0.99 %
0.00
0.18
0.02
0.17
2.36
0.03
1.23
0.10
0.73
2
0
1
3
6
12
0.60 %
0.15
0.56
0.57
0.11
2.34
1.30
2.58
2.07
0.25
1.59 %
0.12
0.55
0.67
0.06
3.50
1.11
2.53
2.10
0.14
0.99 %
-0.03
-0.01
0.10
-0.05
2.33
-0.36
-0.08
0.52
-0.17
1 (least)
0
0.62 %
2.17
1.55 %
3.44
1
0.19
1.71
0.11
1.11
-0.09
-0.99
3
6
12
-0.07
0.52
0.76
-0.33
1.57
1.56
0.07
0.22
0.28
0.42
0.82
0.73
0.14
-0.31
-0.48
0.93
-1.25
-1.39
0.94 %
2.13
Bibliography
Abarbanell, Jeffery S.. and Victor Bernard, 1992, Tests of analysts' overreaction / underreaction to
earnings information as an explanation for anomalous stock price behavior, Journal of Finance
47, 1181-1207.
Agrawal, Anup, Jeffery F. Jaffe, and Gershon Mandelker, 1992, The post-merger performance of
acquiring firms in acquisitions: A re-examination of an anomaly, Journal of Finance 47, 16051621.
Badrinath, S. G., Jayant R. Kale, and Thomas H. Noe, 1995, Of shepherds, sheep, and crossautocorrelations in equity returns, Review of Financial Studies 8, 401-430.
Badrinath, S. G., and Sunil Wahal, 2001, Momentum trading by institutions, Journal of Finance,
forthcoming.
Baker, Malcolm, and Serkan Savasoglu, 2002, Limited arbitrage in mergers and acquisitions, Journal
of Financial Economics, forthcoming 64.
Ball, Ray, and Philip Brown, 1968, An empirical evaluation of accounting income numbers, Journal
of Accounting Research 6, 159-178.
Ball, Ray, S. P. Kothari, and Jay Shanken, 1995, Problems in measuring portfolio performance: An
application to contrarian investment strategies, Journal of Financial Economics 38, 79-107.
Barber, Brad M., and John D. Lyon, 1997, Detecting long-run abnormal stocks returns: The
empiricalpowerand specificationof test statistics, Journal of FinancialEconomics43, 341-72.
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor sentiment,
Journal of FinancialEconomics49, 307-43.
Barth, Mary E., John A. Elliott, and Mark W. Finn, 1999, Market rewards associated with patterns
of increasing earnings, Journal of Accounting Research 37, 387-413.
Beaver, Daryl J., 1968, The information content of annual earnings announcements, Journal of
Accounting Research 6, 67-92.
136
BIBLIOGRAPHY
137
Bernard, Victor L., and Jacob K. Thomas, 1989, Post-earnings-announcement drift: Delayed price
response or risk premium?, Journal of Accounting Research, Supplement 27, 1-36.
, 1990, Evidence that stock prices do not fully reflect the implications of current earnings
for future earnings, Journal of Accounting and Economics 13, 305-340.
Bushee, Brian J., 2001, Do institutional investors prefer near-term earnings over long-run value,
Contemporary Accounting Research 18, 207-246.
, and Christopher F. Noe, 2000, Corporate disclosure practices, institutional investors, and
stock return volatility, Journal of Accounting Research 38, 171-202.
Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52.
Chan, Louis K.C., Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies,
Journal of Finance 51, 1681-1714.
Chen, Joseph, Harrison Hong, and Jeremy C. Stein, 2001, Breadth of ownership and stock returns,
NBER Working Paper.
Cohen, Randolph B., Paul A. Gompers, and Tuomo Vuolteenaho, 2001, Who underreacts to cashflow news'? Evidence from trading between individuals and institutions, Harvard University
working paper.
Cutler, David M., James M. Poterba, and Lawrence H. Summers, 1989, What moves stock prices?,
Journal of Portfolio Management 15, 4-12.
Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam, 1998, Investor psychology and
security market under- and over-reactions, Journal of Finance 53, 1839-85.
Daniel, Kent D., David Hirshleifer, and Siew Hong Teoh, 2002, Investor psychology in capital
markets: evidence and policy implications, Journal of Monetary Economics 49, 139-209.
Daniel, Kent D., and Sheridan Titman, 1997, Evidence on the characteristics of cross-sectional
variation in common stock returns, Journal of Finance 52, 1-33.
, 2001, Market reactions to tangible and intangible information, Northwestern University
Working Paper.
Datar, Vinay T., Narayan Y. Naik, and Robert Radcliffe, 1998, Liquidity and stock returns: An
alternative test, Journal of Financial Markets 1, 203-219.
DeBondt, Werner F. M., and Richard H. Thaler, 1985, Does the stock market overreact?, Journal
of Finance 40, 793-808.
, 1987, Further evidence on investor overreaction and stock market seasonality, Jburnal of
Finance 42, 557-581.
BIBLIOGRAPHY
138
DeChow, Patricia M., and Richard G. Sloan, 1997, Returns to contrarian investment strategies:
Tests of naive expectations hypotheses, Journal of Financial Economics 41, 3-27.
Del.Guercio, Diane, 1996, The distorting effect of the prudent man laws on 'institutional equity
investments, Journal of Financial Economics 40, 31-62.
DeLong, J. Bradford, Andrei Shleifer, Lawrence Summers, and Robert Waldman, 1990, Positive
feedback investment strategies and destabilizing rational strategies, Journal of Finance 45, 379395.
Dennis, Patrick J., and Deon Strickland, 2001, Who blinks in volatile markets, individuals or
institutions?, Journal of Finance, forthcoming.
Dennis, Patrick J., and James P. Weston, 2001, Who's informed? An analysis of stock ownership
and informed trading, Rice University Working Paper.
Dharan, Bala G., and David Ikenberry, 1995, The long-run negative drift of post-listing stock
returns, Journal of Finance 50, 1547-1574.
Fama, Eugene F, 1991, Efficient capital markets II, Journal of Finance 46, 1575-1643.
Fama, Eugene F., 1998, Market efficiency, long-term returns and behavioral finance, Journal of
Financial Economics 49, 56-63.
, and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of
Finance 47, 427-465.
, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial
Economics 33, 3-56.
Foster, George, Chris Olsen, and Jerry Shevlin, 1984, Earnings releases, anomalies, and the behavior
of security returns, Accounting Review 59, 574-603.
French, Kenneth R., and Richard W. Roll, 1986, Stock return variance: The arrival of information
and the reaction of traders, Journal of Financial Economics 19, 3-30.
Gompers, Paul, and Josh Lerner, 1998, Venture capital distributions: Short-run and long-run
reactions, Journal of Finance 53, 2161-83.
Gompers, Paul A., and Andrew Metrick, 2001, Institutional investors and equity prices, The Quarterly Journal of Economics 116, 229-259.
Grinblatt, Mark, and Tobias J. Moskowitz, 1999, Do industries explain momentum?, Journal of
Finance 54, 1249-90.
Grinblatt, Mark, Sheridan Titman, and Russ Wermers, 1995, Momentum investment strategies,
portfolio performance, and herding: A study of mutual fund behavior, American Economic
Review 85, 1088-1105.
BIBLIOGRAPHY
139
Grundy, Bruce D., and Spencer J. Martin, 2001, Understanding the nature of the risks and the
source of the rewards to momentum investing, Review of Financial Studies 14, 29-78.
Hirshleifer, David, 2001, Investor psychology and asset pricing, Journal of Finance 56, 1533-1598.
Hong, Harrison, Terence Lim, and Jeremy Stein, 2000, Bad news travels slowly: Size, analyst
coverage, and the profitability of momentum strategies, Journal of Finance 55, 265-95.
Hong, Harrison, and Jeremy Stein, 1999, A unified theory of underreaction, momentum. trading
and overreaction in asset markets, Journal of Finance 54, 2143-84.
Ikenberry, David, and Josef Lakonishok, 1993, Corporate governance through the proxy contest:
Evidence and implications, Journal of Business 66, 405-35.
, and Theo Vermatlen, 1995, Market underreaction to open market share repurchases,
Journal of FinancialEconomics39, 181-208.
Ikenberry, David L., and Sundaresh Ramnath, 2002, Underreaction to self-selected news events:
The case of stock splits, Review of Financial Studies 15, 489-526.
Jegadeesh, Narasimhan, 1990, Evidence of predictable behavior of security returns, Journal of
Finance 45, 881-898.
, and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implications
for stock market efficiency, Journal of Finance 48, 65-91.
-
, 1995, Overreaction, delayed reaction, and contrarian profits, Review of Financial Studies
8, 973-999.
, 2001, Profitability of momentum strategies: An evaluation of alternative explanations,
Journal of Finance 56, 699-720.
Jiambalvo, James, Shivaram Rajgopal, and Mohan Venkatachalam, 2001, Institutional ownership
and the extent to which stock prices reflect future earnings, Contemporary Accounting Research,
forthcoming.
Kahneman, Daniel, and Amos Tversky, 1974, Judgement under uncertainty: Heuristics and biases,
Science 185, 1124-1131.
, 1986, Rational choice and the framing of decisions, Journal of Business 59, 251-78.
Kothari, S. P., and Jerold B. Warner, 1997, Measuring long-horizon security price performance,
Journal of FinancialEconomics43, 301-39.
La Porta, Rafael, Josef Lakonishok, Andrei Shleifer, and Robert W. Vishny, 1997, Good news for
value stocks: Further evidence on market efficiency, Journal of Finance 52, 859-874.
Lakonishok, Josef, and Inmoo Lee, 2001, Are insiders' trades informative?, Review of Financial
Studies 14, 79-111.
BIBLIOGRAPHY
140
Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny, 1992, The impact of institutional trading
on stock prices, Journal of Financial Economics 32, 23-43.
, 1994, Contrarian investment, extrapolation and risk, Journal of Finance 49, 1541-1578.
Lee, Charles M. C., and Bhaskaran Swaminathan, 2000, Price momentum and trading volume,
Journal of Finance.
Lo, Andrew W., and A. Craig MacKinlay, 1990, When are contrarian profits due to stock market
overreaction?, Review of Financial Studies 3, 175-205.
Loughran, Tim, and Jay Ritter, 1995, The new issues puzzle, The Journal of Finance 50, 23-52.
Loughran, Tim, and Jay R. Ritter, 2000, Uniformly least powerful tests of market efficiency, Journal
of Financial Economics 55, 361-89.
Michaely, Roni, Richard H. Thaler, and Kent L. Womack, 1995, Price reactions to dividend initiations and omissions: Overreaction or drift?, Journal of Finance 50, 573-608.
Michaely, Roni, and Kent Womack, 1999, Conflict of interest and the credibility of underwriter
analyst recommendations, Review of Financial Studies 12, 653-86.
Miller, Merton H., 1986, Behavioral rationality in finance: The case of dividends, Journal of Business 59, 451-68.
Mitchell, Mark L., and J. Harold Mulherin, 1994, The impact of public information on the stock
market, Journal of Finance 49, 923-95).
Mitchell, Mark L., Todd Pulvino, and Erik Stafford, 2002, Limited arbitrage in equity markets,
Journal of Finance 57, April.
Mullainathan, Sendhil, 2001, Thinking through categories, NBER Working Paper.
Nofsinger, John R., and Richard W. Sias, 1999, Herding and feedback trading by institutional and
individual investors, Journal of Finance 54, 2263-95.
Pritamani, Mahesh, and Vijay Singal, 2001, Return predictability followinglarge price changes and
information releases, Journal of Banking and Finance 25, 631-656.
Rabin, Matthew, 2001, Inference by believers in the law of small numbers, Quarterly Journal of
Economics, Forthcoming.
Ritter, Jay R., 1991, The long-run performance of initial, public offerings, Journal of Finance 46,
3-27.
Roll, Richard W., 1983, On computing mean returns and the small firm premium, Journal of
Financial Economics 12, 371-86.
-~
, 1988, R2 , Journal of Finance 43, 541-566.
BIBLIOGRAPHY
141
Rouwenhorst, K. Geert, 1998, International momentum strategies, Journal of Finance 53, 267-284.
Rubinstein, Mark E., 2001, Efficient markets: Yes or no? the affirmative case, Financial Analysts
Journal, Forthcoming.
Seyhun, H. Nejat, 1997, Investment Intelligence: Tips from Insider Trading (MIT Press: Cambridge).
Shiller, Robert J., 1981, Do stock prices move too much to be justified by subsequent changes in
dividends?, American Economic Review 71, 421-498.
, 1999, Human behavior and the efficiency of the financial system, in John B. Taylor, and
Michael Woodford, ed.: Handbook of Macroeconomics, vol. 1 . chap. 23, pp. 1305-40 (North
Holland: Amsterdam).
Shleifer, Andrei, and Robert Vishby, 1997, The limits to arbitrage, Journal of Finance 52, 35-55.
Sias, Richard, and Laura Starks, 1997, Return autocorrelation and institutional investors, Journal
of Financial Economics 46, 103-131.
Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual fund flows, Journal of Finance
53, 1589-1622.
Womack, Kent L., 1996, Do brokerage analysts' recommendations have investment value?, Journal
of Finance 51, 137-168.
THESIS PROCESSING SLIP
FIXED FIELD:
ill.
name
-
index
biblio
-
* COPIES: ,/Archives ,- Aero
Lindgren
TITLE VARIES:
Music
QDewey ; Barker
Hum
Rotch
Sche-Plough
Scienc e
NFC
,-
El
j
NAME VARIES:
IMPRINT:
-'
;' ¢,
i
(COPYRIGHT)_
_ _ATICON
''
_ CLI
_ ~"
ADD: DEGREE:
DEPT.:
ADD: DEGREE:
DEPT.:
------
c1 IPFRVI/JnR.
"
"""'
R as ,^.,
""'
or
_
NOTES:
cat'r:
' ';
DEPT:
~YEAR:
.,.,._.
*- NAME-
"
_
l
date:
page:
`----------
I
f/'X !t'A,'"
-
|,I
. DEGREE:
_
,
I
.,:
.
,
r i-
_
I'lz
l
_I
8
*
Download