Full text of DOI:10.1111/0022-1082.00379

THE JOURNAL OF FINANCE • VOL. LVI, NO. 4 • AUGUST 2001
Investor Psychology and Asset Pricing
DAVID HIRSHLEIFER*
ABSTRACT
The basic paradigm of asset pricing is in vibrant flux. The purely rational ap-
proach is being subsumed by a broader approach based upon the psychology of
investors. In this approach, security expected returns are determined by both risk
and misvaluation. This survey sketches a framework for understanding decision
biases, evaluates the a priori arguments and the capital market evidence bearing
on the importance of investor psychology for security prices, and reviews recent
models.
The best plan is...to profit by the folly of others.
—
Pliny the Elder, from John Bartlett, comp.
Familiar Quotations, 9th ed. 1901.
IN THE MUDDLED DAYS BEFORE THE RISE of modern finance, some otherwise-
reputable economists, such as Adam Smith, Irving Fisher, John Maynard
Keynes, and Harry Markowitz, thought that individual psychology affects
1
prices. What if the creators of asset-pricing theory had followed this thread?
Picture a school of sociologists at the University of Chicago proposing the
Deficient Markets Hypothesis: that prices inaccurately reflect all available
information. A brilliant Stanford psychologist, call him Bill Blunte, invents
the Deranged Anticipation and Perception Model ~or DAPM!, in which prox-
ies for market misvaluation are used to predict security returns. Imagine
the euphoria when researchers discovered that these mispricing proxies ~such
*
Hirshleifer is from the Fisher College of Business, The Ohio State University. This survey
was written for presentation at the American Finance Association Annual Meetings in New
Orleans, January, 2001. I especially thank the editor, George Constantinides, for valuable com-
ments and suggestions. I also thank Franklin Allen, the discussant, Nicholas Barberis, Robert
Bloomfield, Michael Brennan, Markus Brunnermeier, Joshua Coval, Kent Daniel, Ming Dong,
Jack Hirshleifer, Harrison Hong, Soeren Hvidkjaer, Ravi Jagannathan, Narasimhan Jegadeesh,
Andrew Karolyi, Charles Lee, Seongyeon Lim, Deborah Lucas, Rajnish Mehra, Norbert Schwarz,
Jayanta Sen, Tyler Shumway, René Stulz, Avanidhar Subrahmanyam, Siew Hong Teoh, Sheri-
dan Titman, Yue Wang, Ivo Welch, and participants of the Dice Finance Seminar at Ohio State
University for very helpful discussions and comments.
1
Smith analyzed how the “overweening conceit” of mankind caused labor to be underpriced
in more enterprising pursuits. Young workers do not arbitrage away pay differentials because
they are prone to overestimate their ability to succeed. Fisher wrote a book on money illusion;
in The Theory of Interest ~~1930!, pp. 493–494! he argued that nominal interest rates system-
atically fail to adjust sufficiently for inflation, and explained savings behavior in relation to
self-control, foresight, and habits. Keynes ~1936! famously commented on animal spirits in
stock markets. Markowitz ~1952! proposed that people focus on gains and losses relative to
reference points, and that this helps explain the pricing of insurance and lotteries.
1
533

1
534
The Journal of Finance
as book0market, earnings0price, and past returns!, and mood indicators such
as amount of sunlight, turned out to be strong predictors of future returns.
At this point, it would seem that the deficient markets hypothesis was the
best-confirmed theory in the social sciences.
To be sure, dissatisfied practitioners would have complained that it is harder
to actually make money than ivory tower theorists claim. One can even imag-
ine some academic heretics documenting rapid short-term stock market re-
sponses to news arrival in event studies, and arguing that security return
predictability results from rational premia for bearing risk. Would the old guard
surrender easily? Not when they could appeal to intertemporal versions of the
DAPM, in which mispricing is only corrected slowly. In such a setting, short-
window event studies cannot uncover the market’s inefficient response to new
information. More generally, given the strong theoretical underpinnings of mar-
ket inefficiency, the rebels would probably have an uphill fight.
This alternative history suggests that the traditional view that financial
economists have had about the rationality of asset prices was not as inevi-
table as it may seem. Despite many empirical studies, scholarly viewpoints
on the rationality of asset pricing have not converged. This is probably a
result of strong prior beliefs on both sides. On one side, strong priors are
reflected in the methodological claim that we should adhere to rational ex-
planations unless the evidence compels rejection, and in the use of the term
“
risk premium” interchangeably with “mean return in excess of the risk-free
rate.” For those on the opposite side, risk often comes quite late in the list of
possible explanations for return predictability.
Often advocates of one approach or the other have cast the first stone out
the door of their own glass house. There is, in fact, a notable parallelism
among objections to the two approaches, illustrated in corresponding fashion
in Table I. ~Lining up each objection with its counterpart does not imply
parity in the validity of the arguments.!
This survey assesses the theory and evidence regarding investor psychol-
ogy as a determinant of asset prices. This issue is at the heart of a grand
debate in finance spanning the last two decades. In the last few years, fi-
nancial economists have grown more receptive to imperfect rational expla-
nations. Over time I believe that the purely rational paradigm will be subsumed
by a broader psychological paradigm that includes full rationality as a sig-
nificant special case.
Two superb recent presentations of the asset pricing field ~Campbell ~2000!
and Cochrane ~2000!! emphasize objective external sources of risk. As Camp-
bell puts it ~p. 1516!, “... asset pricing is concerned with the sources of risk
and the economic forces that determine the rewards for bearing risk.” For
Cochrane ~p. 455!, “The central task of financial economics is to figure out
what are the real risks that drive asset prices and expected returns.”
In contrast, I argue here that the central task of asset pricing is to exam-
ine how expected returns are related to risk and to investor misvaluation.
Campbell’s survey emphasizes the stability of the finance paradigm over the
last two decades. I will argue that the basic paradigm of asset pricing is in
vigorous and productive flux.

Investor Psychology and Asset Pricing
1535
Table I
Common Objections to the Psychological Approach to Asset
Pricing and Parallel Objections to the Fully Rational Approach
Objection to Psychological Approach
Objection to Fully Rational Approach
Alleged psychological biases are arbitrary.
Rationality in finance theory requires
impossible powers of calculation.
The evidence we possess does not support
rational behavior.
It is too easy to go theory fishing for factor
structures and market imperfections to
match data ex post.
Experiments that generate alleged
psychological biases are not meaningful.
It is too easy to go theory fishing for
psychological biases to match data ex
post.
Rational traders should arbitrage away
mispricing.
Irrational traders should arbitrage away
efficient pricing.
Rational investors will make better
decisions and get richer.
Irrational investors will bear more risk and
get richer.
Confused investors will learn their way to
good decisions.
Accurate investors will learn their way to bad
decisions.
Apparent return predictability is spurious,
so psychological models of predictability
are misguided.
Apparent return predictability is spurious, so
rational models of predictability are
misguided.
Figure 1 illustrates static asset pricing ~analogous to the CAPM! when
investors misvalue assets and securities. Returns are increasing with risk
~
measured here by CAPM beta! and with current market undervaluation of
the asset. There are several potential noisy proxies for the degree of un-
derpricing, such as price-containing variables ~e.g., book0market, market
value, earnings0price!, measures of public mood ~e.g., the weather!, or ac-
tions possibly taken to exploit mispricing ~e.g., recent occurrence of a stock
repurchase or insider purchases!. Risk and mispricing effects do not nec-
essarily take such a simple linearly separable form ~see the models de-
scribed in Section IV!, but it is still useful to keep the two notions conceptually
distinct.
This picture is only a starting point. Just as the static risk effects of the
CAPM have been generalized to intertemporal asset pricing, so the dynamic
behavior of mispricing must be accounted for as well. After decades of study,
the sources of risk premia in purely rational dynamic models are well un-
derstood. In contrast, dynamic psychology-based asset-pricing theory is in
its infancy.
In the remainder of the introduction, I discuss market forces that can
maintain or eliminate mispricing, and why we cannot dismiss mispricing on
conceptual grounds. Section I of the survey presents relevant psychological
biases and argues that many of the important biases grow naturally from
just a few deep roots. Section II summarizes evidence on capital market and
investor behavior regarding the importance of risk and misvaluation effects.
Section III presents asset-pricing theories based on imperfect rationality.
Section IV concludes with further directions for research.

1
536
The Journal of Finance
Figure 1. The Security Market Plane. Graph of security return against risk, as measured by
beta, and underpricing ~positive or negative!.
To think about whether mispricing is viable, consider the traditional ar-
gument for rational price setting. In this account, smart traders spot dollar
bills lying on the ground and grab them, which does away with mispricing.
Setting aside the dynamics of wealth momentarily, the arbitrage story is
incomplete in two ways. First, equilibrium prices reflect a weighted average
2
of the beliefs of the rational and irrational traders. So long as each group
has significant risk-bearing capacity, both influence prices significantly. Ar-
bitrage is a double-edged blade: Just as rational investors arbitrage away
inefficient pricing, foolish traders arbitrage away efficient pricing. Second,
in some respects, all investors may be imperfectly rational. Even in the Olym-
pics, no one runs at the speed of light; some cognitive tasks are just too hard
for any of us.
The traditional argument further asserts that wealth flows from foolish to
wise investors. This point carries considerable weight. Suppose that some
rational individuals are immune from bias, and that all markets are liquid.
Suppose that terminal dividends obey a linear factor model with K system-
atic and N idiosyncratic payoff components ~I will call these systematic and
idiosyncratic “factors”!. An irrational investor on average trades and loses on
2
See, for example, Figlewski ~1978!, Shiller ~1984!, DeLong et al. ~1990a!, Campbell and
Kyle ~1993!, and Shefrin and Statman ~1994!.

Investor Psychology and Asset Pricing
1537
every factor that he misvalues. If the number of factors N ϩ K is large, and
if a nontrivial fraction of them are substantially mispriced, then, on aver-
age, irrational investors lose a very large amount of money almost surely.
Soon superior rationality will prevail.
Thus, as long as some investors are rational and markets are perfect, there
can be substantial mispricing in only a small fraction of the N ϩ K factors.
If N ϾϾ K, then some or all of the systematic factors can be substantially
mispriced, but only a small fraction of the N idiosyncratic components can
be ~see Daniel, Hirshleifer, and Subrahmanyam ~2001a!!.
On the other hand, people are likely to be more prone to bias in valuing
securities for which information is sparse. This suggests that misperceptions
are strongest in the dusty, idiosyncratic corners of the market place. One
way to reconcile both intuitions is to recognize that there are biases that
almost no one is immune to. In this case, there can be widespread idiosyn-
cratic mispricing which only becomes apparent ex post.
Although misperceptions are probably most severe when information is
sparse and arrives slowly, there is no reason to think that confusion is con-
fined purely to idiosyncratic factors. Market timers trade based on what
they perceive to be superior information about the market or about industry
plays such as high-tech. Investors ~whether wisely or not! purchase macro-
economic forecasts. So if investors sometimes misinterpret information, they
will make systematic as well as idiosyncratic errors. Indeed, to the extent
that misperceptions are conveyed through social processes, mistakes may be
greatest for systematic factors along with a few well-known securities.
The fact that several empirical patterns of predictability are strongest in
small ~presumably less liquid! firms suggests that illiquid markets may be
less efficient. This is less obvious than it sounds—the findings may result
from the sparser information available about small, illiquid firms. Since ar-
bitrage is double-edged, holding wealth constant, there is no presumption
that liquidity immediately reduces mispricing. It does, however, speed the
flow of wealth between smart and foolish traders, which may in the long run
do so.3
It is often suggested that the expertise of hedge funds or investment banks
will improve arbitrage enough to eliminate any significant mispricing. This
works if foolish investors are wise enough to delegate to sound managers.
However, intermediaries have incentives to serve or exploit the irrationali-
ties of potential clients. It is not obvious that layering agency over folly
4
improves decisions. So misvaluation does not require that there be frictions
3
Liquidity makes it easier for smart traders to arbitrage away mispricing, but also easier
for foolish traders to arbitrage away efficient pricing. Barber and Odean ~1999! find that trad-
ers who switch to online brokerages trade more aggressively yet subsequently perform poorly—
their greater liquidity encouraged bad trades.
4
Furthermore, regardless of whether there are intermediaries, it is exactly when a security
or sector becomes more mispriced that smarter investors become poorer. This weakens rational
arbitrage ~and strengthens irrational anti-arbitrage! in an untimely way ~Shleifer and Vishny
~
1997! and Kyle and Xiong ~2000!!.

1
538
The Journal of Finance
or special impediments to fundraising by smart players. Such frictions, how-
ever, can slow the flow of wealth between smart and foolish smart traders,
perhaps allowing mispricing to persist longer.
When substantial mispricing is limited to a few factors and residuals, less
rational investors do not necessarily lose on average to wiser ones. Investors
who underestimate risk take larger long positions in risky assets, and thereby
achieve higher expected returns ~DeLong et al. ~1990a, 1991!!. It could fur-
ther be argued that trading pressure by irrational investors induces cross-
sectional return predictability; that these investors thereby lose money; but
that on average they make up their losses by bearing more aggregate mar-
ket risk. However, if overconfident investors irrationally overbuy the mar-
ket, this should result in a low expected return. This does not jibe well with
the equity premium puzzle of Mehra and Prescott ~1985!.
There are other means by which imperfectly rational individuals can earn
high expected returns. Overconfident investors who buy and sell aggres-
sively in response to valid private information signals may exploit liquidity
traders more profitably than rational investors ~Hirshleifer and Luo ~2001!!.
In an imperfectly competitive securities market, overconfident traders can
5
benefit by intimidating competing informed traders ~Kyle and Wang ~1997!!.
Overconfident individuals are also likely to overinvest in acquiring private
6
information, at the expense of leisure.
However, what evidence we have suggests that aggressively trading indi-
7
vidual investors do badly. Despite the ingenious explanations for profitable
foolishness, it is quite plausible that, in fact, fools and their money are soon
parted. Even if so, a misperception that derives from a fundamental human
psychological trait can remain important for asset prices in the long term.
There are two related reasons.
First, wealth is reshuffled in the process of generational succession. Sec-
ond, in the process of getting rich, individuals can learn to be less rational.
For example, biased self-attribution ~Section I.B! causes individuals to attribute
successes to their own qualities and failures to chance. As a result, losses by
overconfident individuals can be offset by the rising confidence of the nou-
veau riche ~see Daniel, Hirshleifer, and Subrahmanyam ~1998! and Gervais
and Odean ~2001!!.
It is challenging to find a source of risk to explain rationally the magni-
tude of cross-sectional predictability ~see Section II!. The challenge for the
mispricing theory is to explain how irrational investors can remain impor-
tant while hemorrhaging a great deal of cash. The disappearance of the size
5
See also Benos ~1998!, Wang ~1998!, and Fischer and Verrecchia ~1999!!.
Other means by which the imperfectly rational can do well or poorly have been described
6
as well; see Blume and Easley ~1982, 1990, 2000!, Palomino ~1996!, Luo ~1998!, and Sandroni
~
2000!.
7
See Barber and Odean ~1999, 2000a, 2000b! and Odean ~1999!. However, most of the theo-
retical models imply that only investors with moderate overconfidence will do well. The data
may be picking up the poor performance of the extremely overconfident.

Investor Psychology and Asset Pricing
1539
effect in the mid-1980s and the inconsistency of the value effect in the last
8
few years is suggestive.
There is a further problem. Having detected a return pattern statistically,
it is hard for an investor to know whether other investors have yet detected
and acted upon it. In 1984, how could an investor be sure whether other
investors were overexploiting the size effect ~Daniel and Titman ~1999!!?
This uncertainty suggests that sometimes patterns of mispricing will be ar-
bitraged away too slowly, and other times will overshoot. Conceivably the
long life of the momentum effect has resulted from arbitrageurs each mis-
takenly fearing that others have started to trade aggressively. As Yogi Berra
commented about a popular restaurant, “No one goes there any more be-
cause it’s too crowded.”
The other possible reason for persistent mispricing is that some relevant
pieces of public information are ignored or misused by everyone. This can
occur either because the signals are obscurely located or because our shared
model of the world is just not sophisticated enough to make their relevance
clear. A pricing error of this sort may disappear once a smart econometrician
identifies it.
It is impossible to be comprehensive on a topic of this scope. Several im-
9
portant topics have been discussed in greater depth elsewhere. My focus is
on the psychology of imperfect rationality, not psychological determinants of
rational risk aversion or time preference. My benchmark for comparison
throughout is the traditional asset-pricing paradigm; I do not cover market
imperfections, nor models of rational bubbles.
I. Judgment and Decision Biases
This section describes some psychological effects that are potentially rel-
evant for securities markets, with hints at possible explanations based upon
1
0
adaptiveness. Economists have traditionally been skeptical of the varied
array of seemingly arbitrary biases offered by experimental psychology. The
empirical findings gain credence if we can understand what causes them. I
argue here that these patterns generally derive from common roots.
8
The U.S. small firm effect was strongly positive every year during 1974 to 1983, and then
was negative for six out of the next seven years; the two closing years of the millennium, which
followed the publication of an important paper on “...Good News for Value Stocks” ~La Porta
et al. ~1997!! were the worst years for value stocks since 1928, though 2000 was better.
9
I generally focus on asset market regularities involving a time horizon of at least a month,
and do not consider seasonalities ~for recent evidence see Hawawini, Keim, and Ziemba ~2000!!.
Several surveys examine the equity premium puzzle in greater depth ~see, e.g., Kocherlakota
~
1996!, Campbell ~1999, 2000!, and Mehra and Prescott ~2001!!. Experiments in psychology and
economics are surveyed in Camerer ~1995, 1998!, Bossaerts ~2000!, and Hertwig and Ortmann
2001!.
~
1
0
See also the surveys of Camerer ~1995!, DeBondt and Thaler ~1995!, Rabin ~1998!, and
Shiller ~1999!. There are also important literatures that build “fast and frugal” heuristics based
upon ex ante considerations ~Gigerenzer, Todd, and ABC-Group ~1999!!, and that model the
decision consequences of bounds on rationality ~Conlisk ~1996!!.

1
540
The Journal of Finance
Since time and cognitive resources are limited, we cannot analyze the data
the environment provides us with optimally. Instead, natural selection has
designed minds that implement rules-of-thumb ~“algorithms,” “heuristics,”
or “mental modules”! selectively to a subset of cues ~see Simon ~1956!!. Such
heuristics are effective when applied to appropriate problems. But their in-
evitable biases can become flagrant when used outside their ideal domain of
applicability.
Economists often argue that errors are independent across individuals,
and therefore cancel out in equilibrium. However, people share similar heu-
ristics, those that worked well in our evolutionary past. So on the whole, we
should be subject to similar biases. Systematic biases ~common to most peo-
ple, and predictable based upon the nature of the decision problem! have
been confirmed in a vast literature in experimental psychology.
There is much debate about exactly how good a job heuristics do. While
psychologists such as Daniel Kahneman and Amos Tversky have made clear
that heuristics can play a positive role, in the last decade, evolutionary ~Dar-
winian! psychologists have strongly emphasized the adaptiveness of cogni-
tive processes. In many cases biases diminish but do not vanish when
11
probabilities are reexpressed as numerical frequencies, and when prob-
lems are posed in visual formats. However, there is no guarantee that fi-
nancial decision problems will be presented to individuals in a manner that
favors the most accurate decisions.
The modern environment differs greatly from the prehistoric environment
of evolutionary adaption for which human cognitive mechanisms were de-
signed by natural selection. Modern humans deal with new abstractions such
as securities, money, impersonal markets, probabilities, and government and
with temptations such as easy access to fats and sugars, gambling casinos,
and real-time Internet trading.
The general fact that cognitive resource constraints force the use of heu-
ristics to make decisions I will call heuristic simplification. ~For cognitive
resource constraints, read limited attention, processing power, and memory.!
A second source of bias arises indirectly from cognitive constraints. This is
that natural selection probably did not design human minds solely to make
good decisions. Trivers ~1985, 1991! discusses evidence that people cannot
perfectly control indicators of their true internal states. This creates selec-
tion for the ability to read subtle cues such as facial expression, eye contact,
posture, tone of voice, and speech tempo to infer the mental states of other
individuals. In Trivers’ self-deception theory, individuals are designed to think
they are better ~smarter, stronger, better friends! than they really are. Truly
believing this helps the individual fool others about these qualities. A third
source of bias is that we are subject to emotions that can overpower reason.
An evolutionary rationale for a lack of self-control is that emotions such as
love and rage can act as mechanisms that allow credible commitment or
11
See Tversky and Kahneman ~1983!, Gigerenzer ~1991!, Gigerenzer et al. ~1996!, Gigeren-
zer and Hoffrage ~1995!, Cosmides and Tooby ~1996!, and Kahneman and Tversky ~1996!.

Investor Psychology and Asset Pricing
1541
threat toward potential allies and enemies ~Hirshleifer ~1987!; see also Frank
~
1988!, Nesse ~2001!!.
I argue that heuristic simplification, self-deception, and emotional loss of
control provide a unified explanation for most known judgment and decision
biases. This framework can provide guidance as to which biases identified in
experiments represent general mechanisms, and which are conditional
side-effects.12
Why don’t people simply learn their way out of biased judgments? To some
extent they do. One barrier is that learning is just too hard. The other bar-
rier arises from self-deception. Individuals who think they are already com-
petent may be slow to adjust their decision procedures ~e.g., Einhorn and
Hogarth ~1978!!.
Much of the evidence described here derives from experiments by econo-
mists and psychologists; their methods are somewhat different. Financial econ-
omists are familiar with criticisms of psychological experiments: that the stakes
are low, that subjects have little experience with the experimental setting, that
there is weak incentive to pay attention or tell the truth, and that publication
depends on finding an effect. What may not be as familiar is that there is data
addressing these issues. On the whole, training and increasing rewards and
number of repetitions often reduces, but does not eliminate, biases. Lessons
learned through repetition often do not carry over well across seemingly sim-
1
3
ilar tasks. The well-known biases have been subjected to replication. Many
though not all! of the cognitive biases are stronger for individuals with low
~
cognitive ability or skills than for those with high ability or skills, consistent
with biases being genuine errors ~see Stanovich and West ~2000!!.
Subsections I.A and I.B consider individual biases organized by proposed
causes ~heuristic simplification and self-deception!. Subsection I.C considers
emotion and self-control, Subsection I.D discusses social interactions, and
Subsection I.E discusses modeling alternatives to expected utility theory and
to Bayesian updating.
A. Heuristic Simplification
A.1. Attention/Memory/Ease-of-Processing Effects
Limited attention, memory, and processing capacities force a focus on sub-
sets of available information. Unconscious associations also create selective
focus. In many studies, priming subjects with ~possibly irrelevant! verbal
information triggers associations that influence judgments ~see, e.g., Gilov-
ich ~1981! and Higgins ~1996!!.
1
2
Explanations based upon cognitive adaptiveness are subject to the objection that it is too
easy to come up with “just-so” stories that fit the data ex post. However, my goal here is not to
make the case that the evidence supports the adaptiveness approach ~see Barkow, Cosmides,
and Tooby ~1992!!. Rather, my point is that it is hard to make sense out of biases without a
conceptual framework. Adaptiveness is about the only one we have.
1
3
See, for example, Camerer ~1995!, Rabin ~1998!, and Hertwig and Ortmann ~2001!.

1
542
The Journal of Finance
Selective triggering of associations causes salience and availability effects
e.g., Kahneman and Tversky ~1973!!. An information signal is salient if it
~
has characteristics ~e.g., differing from the background or from a past state!
that are good at hooking our attention or at creating associations that fa-
cilitate recall. In the availability heuristic ~Tversky and Kahneman ~1973!!,
items that are easier to recall are judged to be more common. This generally
makes sense, since things that are more common are noticed or reported
more often, making them easier to remember. Shiller ~2000b! suggested that
the ease with which regular Web users can think of examples relating to the
Internet revolution encouraged the market boom of the late 1990s.
One reason people are influenced by the format of decision problems is
that they cannot perfectly retrieve relevant information from memory ~Tver-
sky and Kahneman ~1973! and Pennington and Hastie ~1988!!. People un-
derweight the probabilities of contingencies that are not explicitly available
for consideration ~Fischhoff, Slovic, and Lichtenstein ~1978!!. This suggests
a kind of overconfidence ~see Subsection I.B!, and apparent market over-
reaction when unforeseen contingencies do occur.
According to self-perception theory ~see Bem ~1972!, p. 2!, “Individuals come
to know their own attitudes, emotions and internal states by inferring them
from observations of their own behavior and circumstances in which they
occur.” The need to infer can result from memory loss or from simple lack of
access to unconscious internal states. A tendency to form habits can be an
optimal mechanism to address memory loss, reflecting an implicit self-
perception that actions taken before probably had a good reason ~Hirshleifer
and Welch ~2000!!. Habits also economize on thinking. Habits, including the
habitual adherence to self-imposed rules can also play a role in self-
regulation strategies ~e.g., consume only out of dividends, not principal; see
Thaler and Shefrin ~1981! and Shefrin and Statman ~1984!!.
The halo effect causes someone who likes one outstanding characteristic of
an individual to extend this favorable evaluation to the individual’s other
characteristics ~Nisbett and Wilson ~1977a!!. An analogous misattribution
bias could potentially cause stock market mispricing. In an efficient market,
a stock being good in terms of growth prospects says nothing about its pros-
pects for future risk-adjusted returns ~which are on average zero!. If people
mistakenly extend their favorable evaluation of a stock’s earnings prospects
to its return prospects, growth stocks will be overpriced ~see Lakonishok,
Shleifer, and Vishny ~1994! and Shefrin and Statman ~1995!!.
There are ease-of-processing effects analogous to the effects of ease of re-
trieval. For example, the illusion of truth is the finding that people are more
inclined to accept the truth of a statement that is easy to process ~see, e.g.,
Reber and Schwarz ~1999!!. Familiar signal combinations ~e.g., yellow with
banana! are easier to perceive than unfamiliar ones ~Bruner, Postman, and
Rodrigues ~1951!!. There is a strong and robust mere exposure effect in which
exposure to an unreinforced stimulus tends to make people like it more ~see,
e.g., Bornstein and D’Agostino ~1992! and Moreland and Beach ~1992!!. The
basis for this heuristic may be that what is familiar, being understood bet-
ter, is often less risky. However, this can be taken too far, as when people

Investor Psychology and Asset Pricing
1543
prefer to bet on a matter about which they know more than another equiv-
alent gamble ~Heath and Tversky ~1991!!. People also like similarity in choice
of friends and mates ~Berscheid and Reis ~1998!!. According to evolutionary
psychology, people prefer familiar and similar individuals because these were
indicators of genetic relatedness ~e.g., Trivers ~1985!!. These biases suggest
a tendency to prefer local investments ~see also Huberman ~1999!!.
A literature in psychology has examined how subjects learn by observation
over time to predict a variable that is stochastically related to multiple cues
~
see, e.g., Kruschke and Johansen ~1999!!. A pervasive finding is that ani-
mals and people do not achieve correct understanding of the correlation struc-
ture. Instead, cue competition occurs: salient cues weaken the effects of less
salient ones, and the presence of irrelevant cues causes subjects to use rel-
evant cues and base rates ~unconditional frequencies! less. There is also
learned utilization of irrelevant cues. Cue competition raises interesting ques-
tions about how information flooding through the Internet will affect
misvaluation.
The learned usage of irrelevant cues comes close to magical thinking, the
belief in relations between causally unrelated actions or events ~as with as-
trology and other superstitions!. A type of magical thinking called the illu-
sion of control consists of the belief that a person can favorably influence
unrelated chance events. A possible example is that people value lottery ticket
numbers they select more than randomly assigned ones ~Langer ~1975!!.
A.2. Narrow Framing/Mental Accounting/Reference Effects
Narrow framing ~see Kahneman and Lovallo ~1993! and Read, Loewen-
stein, and Rabin ~1999!! involves analyzing problems in too isolated a fash-
ion. This makes excellent sense when time and cognitive resources are limited.
Many problems can be compartmentalized safely. An implication is that the
form of presentation of logically identical decision problems, such as the
highlighting of a different reference for comparison of outcomes can have
large framing effects on choices ~Tversky and Kahneman ~1981, 1986!!. Op-
timizing with respect to a problem-specific reference point, and having a
direct preference over deviations ~instead of over total consumption! econo-
mizes on thinking. Money illusion is another documented example of sensi-
tivity to irrelevant description features ~Shafir, Diamond, and Tversky ~1997!!.
By using different presentations or procedures, experimenters can elicit
preference reversals. Faced with a choice between a binary lottery with a
high probability but relatively low maximum payoff, versus another with
lower probability and higher maximum payoff, subjects often tend to prefer
the high probability lottery, yet place a higher valuation on the high-maximum-
1
4
payoff lottery! There are also context effects in which the presence of a
nonselected choice alternative affects which alternative is selected.
1
4
Lichtenstein and Slovic ~1971!, Grether and Plott ~1979!, and Tversky, Slovic, and Kahne-
man ~1990!.

1
544
The Journal of Finance
Mental accounting ~Thaler ~1985!! is a kind of narrow framing that in-
volves keeping track of gains and losses related to decisions in separate
mental accounts, and to reexamine each account only intermittently when
action relevant. Mental accounting may explain the disposition effect ~She-
frin and Statman ~1985!!, an excessive propensity to hold on to securities
that have declined in value and to sell winners. Having observation of gains
and losses trigger pleasant or unpleasant feelings seems a sensible mental
design to motivate profitable actions. Such a mechanism may, however, be
sidetracked when the individual avoids recognizing losses. Self-deception theory
reinforces this argument, because a loss is an indicator of low decision abil-
ity, and a self-deceiver maintains self-esteem by avoiding recognition of such
indicators.
Related arguments can explain the house money effect ~Thaler and John-
son ~1990!!—a greater willingness to gamble with money that was recently
won. The unpleasantness of a loss of recently won money may be diluted by
aggregating it with the earlier gain.
Anchoring ~Tversky and Kahneman ~1974!! is the phenomenon that people
tend to be unduly influenced in their assessment of some quantity by arbi-
trary quantities mentioned in the statement of the problem, even when the
quantities are clearly uninformative. Some recent authors offer and test pos-
sible explanations in which the process of evaluating the anchor makes anchor-
1
5
consistent arguments more accessible.
According to expected utility theory, utility derives solely from the prob-
ability distribution of payoffs resulting from a choice. However, people seem
to be regret averse in their choices ~e.g., Josephs et al. ~1996! and Ritov
~
1996!!. They seem to be concerned not just that a choice may lead to low
consumption, but that consumption may be lower than the outcome provided
by an alternative choice.
An efficient heuristic method of comparing decision alternatives may be to
line up and compare possible outcomes by state of the world ~rather than
evaluating the expected utility of each alternative separately and then com-
paring!. Thus, having feelings be triggered by comparison of outcomes may
be an effective mechanism for motivating good choices. Regret avoidance
may also reflect a self-deception mechanism designed to protect self-esteem
about decision-making ability ~Josephs et al. ~1996!!.
Regret is stronger for decisions that involve action rather than passivity
Kahneman and Tversky ~1982!!, an effect sometimes called the omission
~
bias ~Ritov and Baron ~1990!!. Regret aversion can explain the endowment
effect, a preference for people to hold on to what they have rather than ex-
change for a better alternative, as with the refusal of individuals to swap a
1
6
lottery ticket for an equivalent one plus cash. The status quo bias ~Sam-
uelson and Zeckhauser ~1988!! involves preferring the choice designated as
the default or status quo among a list of alternatives.
1
1
5
6
Chapman and Johnson ~1999! and Mussweiler and Strack ~1999!.
See Knez, Smith, and Williams ~1985!, Kahneman, Knetsch, and Thaler ~1991!, and Bar-
Hillel and Neter ~1996!.

Investor Psychology and Asset Pricing
1545
Loss aversion is the phenomenon that people tend to be averse even to
very small risks relative to a reference point, suggesting a kink in the utility
function. This may result from the cognitive efficiency of mentally discret-
izing continuous variables, as reflected in the use of terms like “gain,” “break
even,” and “loss,” which make the distinction between a gain and loss more
salient.
A.3. The Representativeness Heuristic
The representativeness heuristic ~Kahneman and Tversky ~1973!, Tversky
and Kahneman ~1974!, and Grether ~1980!! involves assessing the probabil-
ity of a state of the world based on the degree to which the evidence is
perceived as similar to or typical of the state of the world. Similarity can be
viewed as an indicator of the conditional probability of the evidence given
the state of the world versus other states. However, a Bayesian also takes
into account heavily the prior probability of the outcomes, whereas people
tend to underweight statements about unconditional population frequencies
in performing conditional updating—base-rate underweighting. Further-
more, people’s perceptions of how “representative” a piece of evidence is of a
state of the world may match its conditional probability poorly. For example,
people tend to rely too heavily on small samples ~the “law of small numbers”!
and rely too little on large samples, inadequately discount for the regression
phenomenon, and discount inadequately for selection bias in the generation
1
7
or reporting of evidence. Representativeness effects have been detected in
experimental markets ~see Camerer ~1995!, Section II.C.4.!
The idea that a sample should resemble the population is often correct,
especially in a large unbiased independent sample. The preceding errors
amount to applying an inference too weakly within its realm of validity ~large
sample size! and too strongly beyond its realm of validity ~small sample
size!. This is a natural consequence of the trade-offs involved with the de-
sign of an efficient heuristic. The resulting errors are not random: here, the
error is predictable based on the sample size. The law of small numbers
suggests that newly popular theories about the market drawn from recent
investment experience may cause overreaction.
Misunderstanding of how randomness works can also cause a phenom-
enon of gambler’s fallacy. This is the belief that, in an independent sample,
the recent occurrence of one outcome increases the odds that the next out-
come will differ. In fact, people avoid betting on a lottery number that was a
winner sometime over the preceding few days ~Clotfelter and Cook ~1993!!.
On the other hand, use of the representativeness heuristic can cause trend
chasing, because people are too ready to believe that trends have systematic
causes. Statisticians refer to the clustering illusion, wherein people perceive
random clusters as reflecting a causal pattern. People mistakenly believe in
1
7
See, for example, Tversky and Kahneman ~1971!; Kahneman and Tversky ~1973!; Nisbett
and Ross ~1980!, Ch. 4, and references therein; Griffin and Tversky ~1992!; and Brenner, Koehler,
and Tversky ~1996b!.

1
546
The Journal of Finance
“
hot hands” among sports players even when actual performance is very
close to serially independent ~Gilovich, Vallone, and Tversky ~1985!!. In an
experimental market, consistent with gambler’s fallacy, Andreassen and Kraus
~
1990! find that when exogenous prices fluctuate modestly, subjects buy on
dips and sell on rises. However, when a trend appears, subjects do less of
this tracking and possibly switch to chasing trends. There is further evi-
dence from experiments and from surveys that real estate and stock market
1
8
investors extrapolate trends in forecasting price movements.
A.4. Belief Updating: Combining Effects
Edwards ~1968! identified the phenomenon of conservatism, that under
appropriate circumstances individuals do not change their beliefs as much
as would a rational Bayesian in the face of new evidence. The more useful
the evidence, the greater the shortfall between actual updating and rational
updating.
Having a framework for assessing biases can help when they seem to con-
flict. For example, conservatism implies underweighting of new evidence.
Yet, if we view prior beliefs as a base rate, conservatism would seem to
contradict base-rate underweighting. Perhaps conservatism is a conse-
quence of anchoring upon an initial probability estimate. Yet, the represen-
tativeness heuristic predicts that people will extrapolate too strongly from
patterns in small samples, and salience bias also causes people to overreact
to certain kinds of information. Which bias do we believe?
To resolve conflicts like this requires a focus on underlying causes, and
how they will operate in a particular setting. For example, self-deception
can cause conservatism in a stable environment, because an individual who
has explicitly adopted a belief may be reluctant to admit to himself that he
made a mistake. On the other hand, if the environment is volatile, there
may be no dishonor in recognizing that different beliefs are called for.
One explanation for conservatism is that processing new information and
updating beliefs is costly. There is evidence that information that is pre-
sented in a cognitively costly form is weighed less: information that is ab-
stract and statistical, such as sample size and probabilistic base-rate
information. Furthermore, people may overreact to information that is eas-
ily processed, that is, scenarios and concrete examples.
The costly-processing argument can be extended to explain base-rate un-
derweighting. If an individual underweights new information received about
population frequencies ~base rates!, then base rate underweighting is really
a form of conservatism. Indeed, base rates are underweighted less when
they are presented in more salient form or in a fashion which emphasizes
their causal relation to the decision problem ~see Koehler ~1996!!. This costly-
processing-of-new-information argument does not suggest that an individual
will underweight his preexisting internalized prior belief. On the other hand,
1
8
See Shiller ~1988!, Case and Shiller ~1990!, and DeBondt ~1993!.

Investor Psychology and Asset Pricing
1547
if base-rate underweighting is a consequence of the use of the representa-
tiveness heuristic, there should be underweighting of priors.
Griffin and Tversky ~1992! suggest that base-rate underweighting and con-
servatism, interpreted as under- versus overreaction to signals, can be un-
derstood as results of excessive reliance on the strength of information signals
and underreliance on the weight of information signals. The strength of an
information realization is how “extreme” the evidence is ~in some sense! and
the weight of evidence is its reliability or precision. For example, a large
sample of conditionally i.i.d. signals has high weight. But if the preponder-
ance of favorable over unfavorable signals is modest, it has low strength.
Conservatism arises when people rely too little on high-weight evidence such
as a long sample, and base-rate neglect when people rely too heavily on
high-strength evidence such as a few signals all in one direction.
In summary, different experimental settings can lead to under- or over-
reliance on new signals; people seem to make judgments differently in dif-
ferent situations ~see Grether ~1992! and Payne, Bettman, and Johnson ~1992!!.
Given the different possible effects, invoking the name of a bias does not
provide compelling support for assuming under- or overreaction in a finan-
cial model. Further support can be provided by comparing the economic de-
cision environment of the model with the specific experimental decision setting
in which the bias was documented, and especially by running new experi-
ments that match closely the decision environment in the financial model.
Most studies of price forecasts find bias that is predictable using current
observables. For example, forecasts are often found to be adaptive, that is,
1
9
they respond partially to past forecast errors. Such biases are potentially
consistent either with Bayesian learning with an unknown distribution or
with overconfidence. Experimental studies involving a fixed distribution gen-
erally also yield biases, and forecasts are adaptive in most forecast experi-
ments involving endogenously determined prices as well ~see Camerer ~1995!,
Section II.E!. Consistent with overconfidence, forecasters seem to put too
little weight on the known forecasts of other forecasters ~Batchelor and Dua
~
1992!!.
Analyst forecasts of earnings are overoptimistic at long time horizons and
pessimistic at short horizons ~e.g., Richardson, Teoh, and Wysocki ~1999!!.
Such biases may come from misperceptions or from agency incentives. How-
ever, we would normally expect rational agents to provide at least a positive
incremental value in their forecasting activities. There is conflicting evi-
dence as to whether stock market analysts’ forecasts of earnings do better or
worse than a time-series forecast ~see the review of Kothari ~2000!!. A large
literature shows that real-world decision makers such as Ph.D. admission
committees or doctors do not predict outcomes as well as mechanical deci-
sion rules based on simple linear combinations of objective input measures
1
9
See the discussions in Lovell ~1986! and Williams ~1987!, but see also Keane and Runkle
1990!. There is a similar finding for survey forecasts of macroeconomic variables ~e.g., Aggar-
wal, Mohanty, and Song ~1995!!.
~

1
548
The Journal of Finance
~
see Camerer ~1991!!. This suggests that the rise of arbitrage based upon
modern statistical analysis in securities markets will indeed reduce mispricing.
B. Self-Deception
The self-deception theory implies overconfidence, a very well-documented
2
0
bias ~as reviewed, e.g., in Odean ~1998b!!. An extensive literature on cal-
ibration shows that people believe their knowledge is more accurate than
it really is.21 For example, their predictions of probabilities of events are
too extreme ~too high relative to the true frequency when they think the
event probably will occur, too low when they think it will not!. The confi-
dence intervals they provide for quantities are too narrow, for example, 98
percent confidence intervals contain the true quantity only 60 percent of
the time ~Alpert and Raiffa ~1982!!. Experts are well-calibrated in some
contexts but not others ~see Camerer ~1995! pp. 592–593!. Experts can be
more prone to overconfidence than nonexperts when predictability is low
and evidence is ambiguous ~Griffin and Tversky ~1992!!. Overconfidence is
greater for challenging judgment tasks, and individuals tend to be more
overconfident when feedback on their information or decisions is deferred
2
2
or inconclusive.
Overconfidence is sometimes reversed for very easy items ~Lichtenstein
2
3
and Fischhoff ~1977!!. Overconfidence implies overoptimism about the in-
dividual’s ability to succeed in his endeavors. Such optimism has been found
in a number of different settings ~Miller and Ross ~1975!!. Men tend to be
more overconfident than women, though the size of the difference depends
2
4
on whether the task is perceived to be masculine or feminine.
Since people fail more often than they expect to, rational learning over
time would tend to eliminate overconfidence. So for self-deception to suc-
ceed, nature must provide mechanisms that bias the learning process. This
is consistent with self-enhancing biased self-attribution. People tend to
attribute good outcomes to their own abilities and bad outcomes to external
2
0
Hirshleifer and Hirshleifer ~2001! model equilibrium self-deception and observer infer-
ences based upon Trivers’ approach. Bernardo and Welch ~2000! provide an alternative theory
of overconfidence based upon group informational benefits.
2
1
See, for example, Lichtenstein, Fischhoff, and Phillips ~1982!, Yates ~1990!, Keren ~1991!,
and McClelland and Bolger ~1994!.
2
2
Fischhoff et al. ~1977!, Einhorn ~1980!, Lichtenstein et al. ~1982!, Yates ~1990!, and Griffin
and Tversky ~1992!.
2
3
This is not surprising on mechanical grounds; in the extreme case of perfect decision
accuracy, it is possible to be under- but not overconfident about accuracy. It has been suggested
that apparent overconfidence could be an artifact of the choice of questions that are not a
“
representative sample of the knowledge domain” ~e.g., Gigerenzer, Hoffrage, and Kleinbolting
~
1991!!, but overconfidence remains when questions are randomly selected from the knowledge
domain, and has been documented in many practical choice settings ~Griffin and Tversky ~1992!,
Brenner et al. ~1996a!, and Soll ~1996!!.
2
4
Deaux and Emswiller ~1974!, Lenney ~1977!, and Lundeberg, Fox, and Puncochar ~1994!.

Investor Psychology and Asset Pricing
1549
circumstances.25 Overconfidence and biased self-attribution are static and
dynamic counterparts; self-attribution causes individuals to learn to be over-
confident rather than converging to an accurate self-assessment.
Self-deception also explains why there are action-induced attitude changes
of the sort that motivate the theory of cognitive dissonance.2 In one exper-
iment, people who chose between two products downgraded their assess-
ments of the one they did not pick. In another, women who had to exert
greater effort to gain entry to a group subsequently liked the group more. In
other experiments, people who were induced with mild incentives or by re-
quest to express opinions became more sympathetic to those opinions. A
tendency to be excessively attached to activities for which one has expended
resources, the sunk cost effect, has been confirmed in several contexts ~Arkes
and Blumer ~1985!!. The self-deception theory suggests that a tendency to
adjust attitudes to match past actions is a mechanism designed to persuade
the individual that he is a skillful decision maker ~see also Nel, Helmreich,
and Aronson ~1969! and Steele and Liu ~1983!!.
6
Similar reasoning can explain hindsight bias ~e.g., Hawkins and Hastie
1990!!—it helps our self-esteem to think we “knew it all along”; and the
~
phenomenon of rationalization—constructing a plausible ex post rationale
for past choices helps an individual feel better about his decision compe-
tence. People are very ready to devise and apparently believe their explana-
tions for alleged facts about the world as well as themselves.2
7
People tend to interpret ambiguous evidence in a fashion consistent with
their own prior beliefs. They give careful scrutiny to inconsistent facts and
explain them as due to luck or faulty data gathering ~see Gilovich ~1991!
Chap. 4!. This confirmatory bias can help maintain self-esteem, consistent
with self-deception. Exposure to evidence should tend to cause rational Baye-
sians with differing beliefs to converge, whereas the attitudes of experimen-
tal subjects exposed to mixed evidence tend to become more polarized ~e.g.,
Lord, Ross, and Lepper ~1979! and Isenberg ~1986!!. Forsythe et al. ~1992!
find that individuals more subject to this confirmation bias lose money in an
experimental market to those who are less subject to it. Confirmatory bias
may cause some investors to stick to unsuccessful trading strategies, caus-
ing mispricing to persist.
Some general biases toward confirmation of hypotheses do not rely on
self-deception. In evaluating hypotheses assigned by the experimenter about
the relation of two kinds of variables ~e.g., studying the night before an
exam, and getting a good grade!, a large literature finds that people put too
much weight on confirming evidence. This involves focusing on cases in which
2
5
Langer and Roth ~1975!, Miller and Ross ~1975!, Fischhoff ~1982!, and Taylor and Brown
1988!.
~
~
2
6
Festinger and Carlsmith ~1959!, Cooper and Fazio ~1984!, and Harmon-Jones and Mills
1999, Ch. 1!.
2
7
See, for example, Nisbett and Wilson ~1977b!, and Ross et al. ~1977b!, Gazzaniga ~1988!
pp. 12–14.

1
550
The Journal of Finance
both study and a good grade occurred, and neglecting other information ~cases
2
8
in which one but not the other occurred!. It has been argued that such a
bias is an efficient shortcut in many contexts ~Klayman and Ha ~1987!!.
People are also biased toward seeking confirmatory information. In the
famous Wason task experiments ~Wason ~1966!!, subjects were asked to turn
over cards to evaluate a hypothesis. They often turned over cards which
potentially could provide instances consistent with the hypothesis, and often
left unturned cards that could conclusively reject the hypothesis. A possible
explanation is that positive cases are easier to process cognitively. There is
evidence that people are more influenced by the information reflected in the
2
9
occurrence of an event than the nonoccurrence.
C. Emotions and Self-Control
Emotions probably play a role in such traditionally rational considerations
as time and risk preference, and in most or all of the effects described ear-
lier. I discuss some further aspects of emotion here.
C.1. Distaste for Ambiguity
Choices are influenced by the structure of gambles above and beyond the
overall probability distribution of consumption outcomes that the gambles
provide. The Ellsberg paradoxes ~Ellsberg ~1961!! suggested that people are
averse to ambiguity, causing irrational choices. Ambiguity aversion has been
confirmed in market experimental settings. It seems to reflect a more gen-
eral tendency for emotions such as fear to affect risky choices ~see Peters
and Slovic ~1996!!. As suggested by Camerer ~1995!, ambiguity aversion may
increase risk premia unduly when new financial markets are introduced,
because of the layering of uncertainty about both the structure of the eco-
nomic environment and about resulting outcomes. A possible explanation for
ambiguity aversion is that the obvious absence of an identifiable parameter
of the decision problem may often be associated with higher risk and the
possibility of hostile manipulation. This justifies a focus on missing infor-
mation, but such a heuristic can go astray when there is no hostile manip-
ulation. In a related vein, the evidence of Heath and Tversky ~1991! indicates
that, holding probabilities constant, people prefer gambles that give them a
sense of understanding or competence.
C.2. Mood, Feelings and Decisions
Risk aversion, regret aversion, and loss aversion may reflect a calculated
avoidance of unpleasant future feelings. However, mood and emotions felt
today affect people’s perceptions of and choices with respect to risk ~see, e.g.,
2
8
See, for example, Jenkins and Ward ~1965!, Crocker ~1982!, and Fischhoff and Beyth-
Marom ~1983!.
2
9
For example, Newman, Wolff, and Hearst ~1980!, and Nisbett and Ross ~1980!.

Investor Psychology and Asset Pricing
1551
Mann ~1992!!. For example, sales of State of Ohio lottery tickets were found
to increase in the days following a football victory by Ohio State University
~
Arkes, Herren, and Isen ~1988!!. More generally, people who are in good
moods are more optimistic in their choices and judgments than those in
bad moods ~see, e.g., Wright and Bower ~1992!!. Bad moods are associated
with more detailed and critical strategies of evaluating information ~Petty,
Gleicher, and Baker ~1991!!. The influence of mood and emotion on purchase
plans and the effects of advertising have been studied by marketing research-
ers as well.30 Bodily sensations and cognitive experience ~e.g., the ease and
fluency of perception or recall! also affect the decision-making process ~see,
e.g., the survey of Schwarz and Skurnik ~2001!!.
Affective states ~feelings or moods! contain information that individuals
3
1
can use to draw inferences about the environment. However, people often
attribute arousal or feelings to the wrong source, leading to incorrect judg-
ments or misattribution biases ~see, e.g., Ross ~1977!!. For example, people
feel happier on sunny days than on rainy days, but priming them by asking
them about the weather affects their judgment of how happy they are ~Schwarz
and Clore ~1983!!. Mood states tend to affect relatively abstract judgments
more than specific ones about which people have concrete information.3
This suggests, for example, that if the weather in New York puts stock mar-
ket traders in a bad mood, their pessimism may concern long-term market
growth prospects rather than whether the Fed is going to lower interest
rates next week.
2
C.3. Time Preference and Self-Control
The conventional representation of decisions over time has an additively
separable utility function with exogenous, declining exponential weights. How-
ever, evidence from psychology suggests that discount rates change with
circumstances. Deferring consumption involves self-control, and is therefore
related to mood and feelings. There is evidence that discount rates are some-
times remarkably high, that gains are discounted more heavily than losses,
that small magnitudes are discounted more heavily than large, that the fram-
ing of a choice as a delay versus an advance has a large effect on decisions,
that time preference differs greatly in different decision domains ~e.g., money
versus health!, and that visceral influences such as pain or hunger affect
3
3
intertemporal choices.
The exponential specification is time consistent. However, experimental
studies suggest that people and nonhuman animals are time-inconsistent.
Specifically, they tend to discount a deferral of consumption from date t to
3
0
Cohen and Areni ~1991!, Erevelles ~1998!, and Mano ~1999!, and Barone, Miniard, and
Romeo ~2000!.
3
3
3
1
2
3
See for example, Wilson and Schooler ~1991!, and Clore, Schwarz, and Conway ~1994!.
Clore et al. ~1994! and Forgas ~1995!.
See, for example, the discussions of Loewenstein and Prelec ~1992!, Loewenstein ~1996,
2
000!, and Chapman ~1998!.

1
552
The Journal of Finance
t ϩ 1 more heavily as date t approaches, consistent, for example, with a
3
4
hyperbolic form for discount rates. This causes choice reversals even when
35
no new information arrives. Hyperbolic discounting has been disputed. Never-
theless, recent economic studies have applied time-inconsistent discounting
to a wide range of issues including savings, liquidity premia, and the equity
3
6
premium puzzle.
D. Social Interactions
Financial economists have borrowed more from the psychology of the in-
dividual than from social psychology. Financial theorists have examined how
information is transmitted by prices, volume, or corporate actions. However,
person-to-person and media contagion of ideas and behavior also seems im-
portant. People tend to conform with the judgments and behaviors of others,
as documented in the famous length-estimation experiments of Asch ~1956!.
A meta-analysis of 133 related studies ~Bond and Smith ~1996!! confirmed
the conformity effect, which is, however, history and culture dependent. There
are rational informational reasons to learn by observing the actions of oth-
ers.37 However, a fully descriptive analysis will have to encompass imperfect
rationality ~see, e.g., Ellison and Fudenberg ~1995!!.
Conversation is critical in the contagion of popular ideas about financial
3
8
markets, as emphasized by Shiller ~2000a!. In a survey of individual in-
vestors, Shiller and Pound ~1989! found that almost all of the investors who
recently purchased a stock had their attention drawn to it through direct
interpersonal communication. The influence of conversation on trading may
arise from individuals’ overconfidence about their ability to distinguish per-
tinent information from noise or propaganda; examples of large price move-
ments triggered by Internet chat comes to mind.
As discussed by Shiller ~1999!, owing to limited attention, people tend to
pay much more attention to ideas or facts that are reinforced by conversa-
tion, ritual, and symbols. In consequence, culture becomes an important
determinant of behavior, and expression of ideas can be self-reinforcing.
Kuran and Sunstein ~1999! describe the process of belief formation as lead-
ing to “availability cascades,” wherein an expressed perception is perceived
to be more plausible as a consequence of its increased availability in public
discourse.
3
4
See Ainslie ~1975!, Thaler ~1981!, Loewenstein and Prelec ~1992!, and Kirby and Herrn-
stein ~1995!.
3
3
5
6
See Mulligan ~1996!, Fernandez-Villaverde and Mukherji ~2000!, and Rubinstein ~2000!.
See Laibson ~1997!, O’Donoghue and Rabin ~1999!, Luttmer and Mariotti ~2000!, and Har-
ris and Laibson ~2001!.
3
7
See, for example, Banerjee ~1992! and Bikhchandani, Hirshleifer, and Welch ~1992!; on the
possibility of informational cascades in securities markets, see Avery and Zemsky ~1998! and
Lee ~1998!.
3
8
See the conversational learning models of Banerjee and Fudenberg ~1999! and Cao and
Hirshleifer ~2000!.

Investor Psychology and Asset Pricing
1553
Conversation pools information surprisingly poorly. Groups of people tend
to talk much more about information signals that they already share than
individual-specific signals ~Stasser, Taylor, and Hanna ~1989!!. As a result,
groups sometimes fail to detect patterns that are discernable by combining
individual-specific signals ~Stasser and Titus ~1985!!. Environmental pres-
sures such as crowding and unusual circumstances cause group members to
experience “cognitive overload,” and rigid thinking ~adherence to habitual
behaviors; see Argote, Turner, and Fichman ~1988!!.
When communicating information, people tend to sharpen and level, that
is, emphasize what they construe to be the main point, and deemphasize
qualifying details that might confuse this point. This is necessary for clarity
given cognitive constraints ~Anderson ~1932! and Allport and Postman ~1947!!,
but tends to cause listener beliefs to move to extremes. A closely related
point is that causes tend to be oversimplified, distorting listener beliefs.
There are also systematic message distortions related to a desire to be en-
tertaining or to manipulate the listener ~see Gilovich ~1991!, Chap. 6!. These
facts point to the need for analysis of conversation and rumors in securities
markets.
The fundamental attribution error ~Ross ~1977!! is the tendency for indi-
viduals to underestimate the importance of external circumstances and over-
estimate the importance of disposition in determining the behavior of others.
In a financial context, such a bias might cause observers of a repurchase to
conclude that the CEO dislikes holding excess cash rather than that the
CEO is responding to market undervaluation of the stock. This would sug-
gest market underreaction to corporate events.
People mistakenly believe that others share their beliefs more than they re-
ally do, the false consensus effect ~e.g., Ross, Green, and House ~1977a!; Welch
~
2000! reports a false consensus effect among financial economists about the
equity premium!. Self-deception may encourage this by making the individual
reluctant to consider the possibility that he is making a deviant error. False
consensus may also result from availability ~since like-minded people tend to
associate together!. The curse of knowledge ~Camerer, Loewenstein, and We-
ber ~1989!! is a tendency to think that others who are less informed are more
similar in their beliefs to the observer than they really are.
E. Modeling Alternatives to Expected Utility and to Bayesian Updating
Expected utility theory has dominated financial modeling because it cap-
tures rational decision making elegantly. However, the paradoxes of Allais
~
1953! and subsequent confirmations showed systematic violations of ex-
pected utility; people seem to be influenced by “irrelevant alternatives.” Fur-
ther violations have multiplied. Evidence of systematic preference reversals
suggests that choice may not be well described by maximization of a utility
function at all. A less radical departure from the traditional approaches is to
consider alternative objectives ~Camerer ~1995, 1998! provides an in-depth
treatment!. Camerer discusses generalizations that involve functional forms

1
554
The Journal of Finance
on probability weightings and utility functions, in some cases explicitly de-
rived from modified axioms of choice.
In prospect theory ~Kahneman and Tversky ~1979! and Tversky and Kahne-
man ~1992!!, individuals maximize a weighted sum of “values” ~analogous to
utilities!, where the weights are functions of probabilities ~instead of true prob-
abilities!. Extremely low probabilities are treated as impossibilities, and ex-
tremely high probabilities as certainties. In contrast, very ~but not extremely!
low probabilities are overestimated, and very ~but not extremely! high prob-
abilities are overestimated. For intermediate probabilities, the weighting func-
tion increases with a slope less than one. The value function is kinked at the
“
reference point” ~loss aversion!.39 The value function is concave to the right of
the reference point and convex to the left, reflecting risk aversion among gambles
that involve only gains and risk seeking among gambles involving only losses.
The advantage of this approach is that it can capture many of the known
patterns of individual choice under risk, as well as financial regularities
~
see, e.g., Camerer ~1998! and Shiller ~1999!!. Indeed, Camerer ~1998! ar-
gues that a form of prospect theory fits the data better than either expected
utility theory or the other generalizations that have been proposed.
Several other generalizations of time-additively separable expected utility
have been applied to asset pricing issues, especially the equity premium puz-
zle. Epstein and Zin ~1989! developed a class of intertemporal utility functions
that allow for nonadditivity and nonexpected utility behavior. Priming is a phe-
nomenon in which exposure to a stimulus affects a subject’s later response to
further presentation of the same or a related stimulus. Evidence of priming
effects does not tell us how people react to repeated consumption choices ~self-
administered stimuli of a sort!, but is broadly suggestive that past consump-
tion levels may influence how people respond to future consumption levels. Such
dependence is reflected in habit formation preferences ~Sundaresan ~1989! and
Constantinides ~1990!!, in which the utility derived from current consumption
also depends on a habitual level of consumption.
Gilboa and Schmeidler ~1995! offer a case-based decision theory which,
unlike expected utility theory, is not based on evaluating outcomes and their
probabilities. A case is a menu of decision options. Choices are evaluated
based on outcomes of past choices and how similar those choices are to those
in the current menu.
The evidence on heuristics and biases also suggests that Bayesian updat-
ing is not fully descriptive of human behavior. However, Bayes theorem is
nonarbitrary, which is a useful discipline for modeling. Some recent models
4
0
describe updating based on self-attribution bias and confirmatory bias.
3
9
First-order risk averse preferences ~Epstein and Zin ~1990!!, like loss aversion, involve a
utility function that depends on a reference point, and in which there is nontrivial aversion
even to small risks. In the case of disappointment aversion ~Gul ~1991!!, investors weigh out-
comes that are worse than the certainty-equivalent outcome more heavily than favorable outcomes.
4
0
See Daniel et al. ~1998!, Rabin and Schrag ~1999!, Rabin ~2000!, Gervais and Odean ~2001!,
and Yariv ~2001!.

Investor Psychology and Asset Pricing
1555
II. Evidence of Risk and Mispricing Effects
I classify the evidence bearing on asset mispricing into five categories:
1! return predictability; ~2! the equity premium puzzle; ~3! evidence as to
~
whether firms take actions in response to mispricing; ~4! whether firms
take actions in order to create mispricing; and ~5! evidence of investment
errors.41 My emphasis here is on findings that have received confirmation
over time and location. However, such consistency is not a prerequisite for
a pattern to be interesting. If widespread and fairly stable patterns of
mispricing exist, then almost surely transient and situation-specific ones
do too.
A. Predictability of Security Returns
Return predictability research is haunted by the specter of data mining.
Some of the patterns described here are probably just vagaries of chance.
However, predictability is a generic prediction of modern asset-pricing theo-
ries. So cautious skepticism rather than profound suspicion is called for.
Most of the patterns of return predictability summarized here have dual
~
and duelling! explanations based on either risk premia or mispricing. Em-
pirical papers on predictability often interpret psychological explanations
naively. Several authors interpret evidence that factor loadings or aggregate
conditioning variables can capture predictability as counter to the psycho-
logical approach. But the psychological approach recognizes that investors
should care about factor risk. To attribute a return pattern to rational factor
pricing requires not just a finding that factors matter, but measurement of
whether expected returns are commensurate with the relevant risks. Fur-
thermore, the psychological approach predicts that factors, not just residu-
als, will be mispriced. The conditioning variables and the variables used to
identify factors, such as aggregate dividend yield, the term premium, the
default premium, book0market, and size, are very natural proxies for factor
misvaluation, as will be discussed.
A.1. Predictability Based Upon Factor Risk Measures
I focus here on CAPM beta and the factor loadings of Fama and French
1993!. A positive univariate relation of beta with expected returns is found
~
in most studies, but depends on the country, time period, empirical imple-
4
2
mentation, and form of the CAPM being tested. Beta has incremental
4
1
This topic is vast; for recent reviews of different aspects of the evidence pertaining to
mispricing, see Fama ~1991, 1998!, Kothari ~2000!, Daniel, Hirshleifer and Teoh ~2001!, and Lee
~
2001!. I do not discuss actions by outsiders such as mutual funds to exploit predictability.
See, for example, Black, Jensen, and Scholes ~1972!, Fama and MacBeth ~1973!, Solnik
1974!, Harvey ~1989!, Fama and French ~1992!, Handa, Kothari, and Wasley ~1993!, Kothari,
4
2
~
Shanken, and Sloan ~1995!, Kim ~1997!, Heston, Rouwenhorst, and Wessels ~1999!, Rouwen-
horst ~1999!, Kothari and Shanken ~2000!, and Bossaerts ~2000!.

1
556
The Journal of Finance
power to predict future returns after controlling for market value and0or
4
3
fundamental0price ratios in some studies but not others.
A.2. Predictability Based Upon Price and Benchmark Value Measures
A natural way to identify mispricing is to compare an asset’s price to a
related value measure. A remarkably consistent empirical pattern is that
almost any such pairing that researchers try predicts future returns in the
right direction—the “cheap” security on average appreciates relative to a
risk-adjusted benchmark, or relative to an “expensive” security. Efficient
market fans will conclude, however, that the security is cheap because it is
4
4
riskier, and that the risk adjustment is misspecified.
In several cases, the market value of a parent firm has been substantially
less than one of its parts, and managers undertook transactions apparently
4
5
suitable for exploiting the overpricing of a division. Closed end funds often
trade at discounts and premia relative to net asset value; these discounts
4
6
predict future small stock returns. Securities that are virtually perfect sub-
stitutes are sometimes traded at different prices by different clienteles
~
Rosenthal and Young ~1990! and Froot and Dabora ~1999!!.
A short-term yield provides a value benchmark for a long-term bond. Dis-
crepancies between long- and short-term yields positively predict the hold-
4
7
ing period returns on long-term bonds. Bonds denominated in different
currencies provide mutual benchmarks. Investing in a country’s bonds that
have recently become cheaper ~higher nominal yield! relative to another coun-
try’s bonds on average earns higher returns—the forward premium puzzle
~
see, e.g., Engel ~1996!!.
Stock benchmarks include fundamental measures such as book value, earn-
ings, or even a constant ~for the size effect!. Cross-sectionally, equity-price-
related variables ~e.g., 10price, book0market, earnings0price, debt0equity!
predict high stock returns in the United States and many other countries,
4
8
even after controlling for beta. For the stock market as a whole, high
fundamental0price ratios ~dividend yield or book0market! predict future in-
4
3
See, for example, Fama and French ~1992, 1996a!, Handa et al. ~1993!, Kothari et al.
1995!, Jagannathan and Wang ~1996!, Kim ~1997!, Knez and Ready ~1997!, Heston et al. ~1999!,
~
and Kothari and Shanken ~2000!.
4
4
4
4
5
6
This insight has been applied to stocks by Ball ~1978!, Keim ~1988!, and Berk ~1995!.
Schill and Zhou ~1999!, Cornell and Liu ~2000!, and Lamont and Thaler ~2000!.
See Swaminathan ~1996! and Neal and Wheatley ~1998!. Bodurtha, Kim, and Lee ~1995!
find that U.S. traded closed-end country fund premia and discounts are often large, and comove
primarily because of their common sensitivity to the U.S. market. Country fund premia predict
returns on U.S. size-ranked portfolios and fund stock returns.
4
7
Shiller, Campbell, and Schoenholtz ~1983!, Mankiw and Summers ~1984!, Mankiw ~1986!,
Fama and Bliss ~1987!, Campbell and Shiller ~1991!, Bekaert, Hodrick, and Marshall ~1997b!,
and Cochrane ~2000! Section 20.1.
4
8
See, for example, Stattman ~1980!, Banz ~1981!, Basu ~1983!, Rosenberg, Reid, and Lan-
stein ~1985!, DeBondt and Thaler ~1987!, Bhandari ~1988!, Chan, Hamao, and Lakonishok ~1991!,
Fama and French ~1992, 1998!, Davis ~1994!, Heston, Rouwenhorst, and Wessels ~1995!, Hau-
gen and Baker ~1996!, Jagannathan, Kubota, and Takehara ~1998!, Rouwenhorst ~1999!, Davis,
Fama, and French ~2000!, Hawawini and Keim ~2000!, and Daniel, Titman, and Wei ~2001!.

Investor Psychology and Asset Pricing
1557
dex returns in the United States and internationally in several, though not
all, studies.49 A better predictor of cross-sectional and aggregate returns can
be formed by normalizing price with earnings-based indices of fundamental
value.50 Market returns are also predictable based on term and default
spreads.51
Size and value portfolios are associated with a factor or factors distinct
5
2
from the stock market portfolio. The loadings on three factors based on
size, value, and the market predict the returns on portfolios sorted on
various characteristics, but do not explain short-term momentum; a global
two-factor model predicts international returns ~Fama and French ~1996b,
1
998!!.
Several studies report very high Sharpe ratios achievable based on cross-
5
3
sectional value effects, a point reinforced by low international correla-
tions of some size and value strategies ~Hawawini and Keim ~1995!!. This
raises the question of whether the implied variability of marginal utility
across states under rational asset pricing is implausibly high ~see Hansen
and Jagannathan ~1991!!. Chen ~2000! finds that book0market and
momentum-based portfolios do not contain enough information about fu-
ture returns on aggregate wealth to be strongly priced as state variables in
a Merton ICAPM.
Fama and French ~1995! suggest that size and book0market factors may
be correlated with harms suffered by individuals when firms are distressed.
Differing conclusions have been drawn about the association of size and book0
5
4
market with distress. The book0market effect remains strong after control-
ling for distress ~Griffin and Lemon ~2001!!. The voluntary allocation by
employees of personal retirement funds into shares of their own firms ~Be-
nartzi ~1997!! opposes the distress–risk hypothesis.
Conclusions differ as to whether “characteristics” ~size, book0market! or
5
5
factor loadings do a better job predicting returns. Perhaps the most com-
pelling evidence for expectational errors is that, after portfolios are formed,
growth stocks on average respond very negatively to subsequent earnings
announcements for several years, and value stocks do not ~La Porta et al.
~
1997! and Skinner and Sloan ~2000!!.
4
9
See Campbell and Shiller ~1988a!, Fama and French ~1988a!, Hodrick ~1992!, Pesaran and
Timmermann ~1995!, Kothari and Shanken ~1997!, Pontiff and Schall ~1998!, Bossaerts and
Hillion ~1999!, Goyal and Welch ~1999!, Lewellen and Shanken ~2000!, and Goetzmann, Ibbot-
son, and Peng ~2001!.
5
0
See Abarbanell and Bushee ~1998!, Frankel and Lee ~1998, 2001!, Chang, Chen, and Dong
~
1999!, and Lee, Myers, and Swaminathan ~1999!.
5
5
1
2
See Keim and Stambaugh ~1986!, Campbell ~1987!, and Fama and French ~1989!.
See Fama and French ~1993, 1995!, and Liew and Vassalou ~2000!. This of course does not
guarantee that the loadings on these factors are priced separately from market beta; for ex-
ample, under the CAPM they would not be.
5
3
Brennan, Chordia, and Subrahmanyam ~1998!, Lakonishok et al. ~1994!, and MacKinlay
~
1995!.
5
5
4
5
Chan and Chen ~1991!, Shumway ~1996!, Chen and Zhang ~1998!, and Dichev ~1998!.
See Daniel and Titman ~1997!, Jagannathan et al. ~1998!, Lewellen ~1999!, Davis et al.
~
2000!, and Daniel et al. ~2001!.

1
558
The Journal of Finance
A.3. Predictability Based Upon Past Returns: Momentum and Reversal
In many asset and security classes internationally, there is positive short-
lag autocorrelation and negative long-lag autocorrelation.56 Cross-sectionally,
U.S., European, and emerging market stocks that have done very well in the
5
7
recent past ~about 3 to 12 months! tend to do well over the next month.
Long term reversals in the cross section are documented by DeBondt and
5
8
Thaler ~1985!. Momentum is stronger in small firms, growth firms, firms
with low analyst following, and in the security-specific ~nonmarket! compo-
5
9
nent of stock returns. Volume interacts with momentum in predicting fu-
ture returns, suggesting a possible cycle of overreaction and correction ~Lee
and Swaminathan ~2000b!!. Chordia and Shivakumar ~2000! report that mo-
mentum profits can be captured based on security sensitivities to a few ag-
gregate variables ~see also Ahn, Conrad, and Dittmar ~2000!!. Lewellen ~2000!
provides evidence of negative autocorrelation and cross-serial correlation in
industry and size portfolios, consistent with negative market autocorrela-
6
0
tion during the study’s time period.
Past winners earn substantially higher returns than do past losers at the
dates of quarterly earnings announcements occurring in the seven months
6
1
following portfolio formation. This is surprising from a rational risk per-
spective, because high momentum firms should become less leveraged and
less risky. Also, firms with extremely low returns over the several months
are having trouble, so the distress factor view of value effects suggests that
negative momentum firms should earn high future returns.
A.4. Predictability Based Upon Public Versus Private News Events
Several event studies have documented abnormal returns subsequent to
the event date. One explanation, event selection, is that a firm’s decision
5
6
See Fama and French ~1988b!, Poterba and Summers ~1988!, Case and Shiller ~1990!,
Cutler, Poterba, and Summers ~1991!, Gyourko and Keim ~1992!, Barkham and Geltner ~1995!,
Richards ~1997!, Balvers, Wu, and Gilliland ~2000!, Chan, Hameed, and Tong ~2000!, and Ng
and Fu ~2000!. On methodological and robustness issues for stocks, see Kim, Nelson, and Startz
~
1988!, Richardson and Stock ~1989!, Jegadeesh ~1991!, Richardson and Smith ~1994!, and Car-
mel and Young ~1997!. There is also a literature on whether stock returns are excessively vol-
atile relative to dividend variability ~Shiller ~1979, 1981!, LeRoy and Porter ~1981!, Kleidon
~
1986!, Marsh and Merton ~1986!, Campbell and Shiller ~1988a, 1988b!, and West ~1988!. This
is equivalent to the issue of whether there is excessive long-run reversal in stock prices ~see
Cochrane ~1991!!, since any overreaction must increase volatility.
5
5
7
8
Jegadeesh and Titman ~1993!, Rouwenhorst ~1998, 1999!.
On methodological issues and the robustness of this finding, see Chan ~1988!, Ball and
Kothari ~1989!, Chopra, Lakonishok, and Ritter ~1992!, and Ball, Kothari, and Shanken ~1995!.
5
9
See Jegadeesh and Titman ~1993!, Daniel and Titman ~1999!, Grinblatt and Moskowitz
1999!, Hong, Lim, and Stein ~2000!, and Grundy and Martin ~2001!. Both industry and non-
~
industry components of momentum help predict future returns ~Moskowitz and Grinblatt ~1999!
and Grundy and Martin ~2001!!.
6
0
A given serial covariance structure is potentially subject to very different causal interpre-
tations. Jegadeesh and Titman ~1995! provide a decomposition that distinguishes factors from
residuals, and therefore lends itself to a distinction between factor versus residual autocorrelation.
6
1
Jegadeesh and Titman ~1993!; see also Chan, Jegadeesh, and Lakonishok ~1996!.

Investor Psychology and Asset Pricing
1559
whether and when to engage in the event depends on whether there is mar-
ket misvaluation. A second possibility, manipulation, is that around the time
of the action, the firm reconfigures other information reported to investors
in order to induce misvaluation.
There is evidence suggesting that both selection and manipulation occur.
Regarding selection, a remarkable pattern emerges from studies of discre-
tionary corporate events ~actions chosen by management or other potentially
informed parties!. The average abnormal stock return in the three to five
years subsequent to the event has the same sign as the event-date stock
6
2
price reaction. I call this regularity postevent return continuation.
The evidence that has appeared since this postevent return continuation
hypothesis was proposed by Daniel et al. ~1998! has generally been support-
ive over new time periods and events. There has been little study of post-
event performance for events that are not taken at the discretion of
management or analysts with incentives to react to mispricing. However,
Cornett et al. ~1998! find that there is postevent continuation when bank
stocks issue equity, except when equity issuance is forced by reserve
requirements.
Fama ~1998! argues that these return patterns are sensitive to empirical
methodology. Several recent studies have concluded that there is limited or
6
3
no underperformance of new issue firms. However, some recent methods
minimize the power to detect misvaluation effects ~Loughran and Ritter ~2000!!.
Jegadeesh ~1999! reports large post-SEO underperformance even relative to
several ~excessively! stringent return benchmarks.
The argument that post-IPO underperformance is eliminated by an ap-
propriate benchmark is counterintuitive, because it amounts to saying that
IPO firms have unusually low risk. Risk reduction may justify a low re-
6
2
Events for which this has been found include stock splits ~Grinblatt, Masulis, and Titman
1984!, Ikenberry, Rankine, and Stice ~1996!, Desai and Jain ~1997b!, and Ikenberry and Ram-
~
nath ~2000!!; tender offer and open market repurchases ~Lakonishok and Vermaelen ~1990! and
Ikenberry, Lakonishok, and Vermaelen ~1995, 2000!!; equity carveouts ~Vijh ~1999!!; spinoffs
~
Cusatis, Miles, and Woolridge ~1993! and Desai and Jain ~1997a!!; accounting write-offs ~Bar-
tov, Lindahl, and Ricks ~1998!!; analyst earnings forecast revisions ~Chan et al. ~1996! and Lin
2000a, 2000b!!; analyst stock recommendations ~Groth et al. ~1979!, Bjerring, Lakonishok, and
Vermaelen ~1983!, Elton, Gruber, and Gultekin ~1984!, Womack ~1996!, Michaely and Womack
1999!, Krische and Lee ~2000!, and Barber et al. ~2001!!; dividend initiations ~Michaely, Thaler,
~
~
and Womack ~1995! and Boehme and Sorescu ~2000!!; dividend omissions ~Michaely et al. ~1995!!;
seasoned issues of debt ~Spiess and Affleck-Graves ~1999!!; seasoned issues of common stock
~
~
Loughran and Ritter ~1995!, Spiess and Affleck-Graves ~1995!, Cornett et al. ~1998!, Teoh et al.
1998b!, Jegadeesh ~1999!, Foerster and Karolyi ~2000!, but see Kang, Kim, and Stulz ~1999!!;
public announcement of previous insider trades ~Rozeff and Zaman ~1988! and Seyhun ~1988!!;
and venture capital share distributions ~Gompers and Lerner ~1998!!. The hypothesis has not
been tested for IPOs since we do not observe the price reaction to the announcement that an
IPO will occur. The pattern does not hold for exchange listing ~McConnell and Sanger ~1987!,
Dharan and Ikenberry ~1995!, and Foerster and Karolyi ~1999!! and private placements ~Hert-
zel et al. ~1999!!, which may involve informed discretion on the part of the buying as well as the
selling party.
6
3
See Brav, Geczy, and Gompers ~2000!, Eckbo and Norli ~2000!, Eckbo, Masulis, and Norli
2000!, Gompers and Lerner ~2000!, and Mitchell and Stafford ~2000!.
~

1
560
The Journal of Finance
turn benchmark for SEO firms, but risk increases would seem to imply a
higher benchmark after debt issues or bond rating downgrades, making
6
4
the underperformance after these events even stronger. Poor postdown-
grade performance also opposes the distress-risk-factor theory of return
predictability. New issue firms perform especially badly at subsequent earn-
ings announcement dates, which is hard to interpret as a negative risk
premium.65
Irrelevant, redundant, or old news affects security prices when pre-
6
6
sented saliently. These demonstrable examples of mispricing suggest
that less blatant mispricing may occur routinely. Little of stock price
or orange juice futures price variability has been explained empirically
by relevant public news.67 Historical crashes and speculative episodes
6
8
are often hard to explain in terms of fundamental news. Allen ~2001!
provides examples suggesting that bubbles have major economic conse-
quences, and argues that agency problems among financial institutions
may cause bubbles.
Several studies explore fundamental trends and subsequent returns. Cash
or earnings surprises are followed by positive abnormal returns in the short
6
9
run, and perhaps negative abnormal returns in the long run. Investors also
seem to extrapolate fundamentals in options and in football betting markets
~
Avery and Chevalier ~1999! and Poteshman ~2001!!.
A.5. Predictability Based Upon Mood Proxies
Environmental factors that influence mood are correlated with stock
price movements. Kamstra, Kramer, and Levy ~2000a! find that a deter-
ministic variable, changes to and from daylight saving time, disrupts sleep
7
0
patterns, and is related to stock returns. A stochastic variable, cloud
cover in the city of a country’s major stock exchange, is associated with
low daily stock index returns in a joint test of 26 national exchanges as
well as in the United States ~Hirshleifer and Shumway ~2001! and Saun-
ders ~1993!!.
6
6
6
4
5
6
Spiess and Affleck-Graves ~1999! and Dichev and Piotroski ~2001!.
Jegadeesh ~1999! and Denis and Sarin ~2000!; see also Ikenberry and Ramnath ~2000!.
Ashton ~1976!, Ho and Michaely ~1988!, Hand ~1990, 1991!, Rau and Vermaelen ~1998!,
Andrade ~1999!, Avery and Chevalier ~1999!, Klibanoff, Lamont, and Wizman ~1999!, Cooper,
Dimitrov, and Rau ~2000!, Huberman and Regev ~2001!, and Rashes ~2001!.
6
6
7
8
Roll ~1984, 1988!, Cutler, Poterba, and Summers ~1989!, and Fair ~2000!.
See, for example, Cutler et al. ~1989!, Seyhun ~1990!, Shiller ~2000b! Ch. 4; for a mainly
rational perspective on the Dutch tulip bulb boom, see Garber ~1989!.
6
9
On short run postearnings announcement drift, see, for example, Bernard and Thomas
~
~
1989, 1990!. For poor long-lag performance, see DeBondt and Thaler ~1987!, Lakonishok et al.
1994!, Lee and Swaminathan ~2000a!, but see also DeChow and Sloan ~1997!, and Daniel and
Titman ~2000!.
7
0
Kamstra, Kramer, and Levi ~2000b! examine the relation of another deterministic vari-
able, seasonal shifts in length of day, to returns in several countries.

Investor Psychology and Asset Pricing
B. Equity Premium and Risk-free Rate Puzzles
1561
The equity premium puzzle71 is that U.S. equity market returns are high
relative to risk, implying high levels of risk aversion and so a low elasticity
of intertemporal substitution in consumption. This in turn implies very high
real interest rates to induce individuals to accept lower consumption now
than in the future ~consistent with historical growth in consumption; see
Weil ~1989!!.
C. Actions Possibly Taken in Response to Mispricing
Corporations buy and sell shares in a way that is correlated with possible
7
2
measures of market mispricing. The amount of financing and repurchase
varies widely over time in an industry-specific way. Merger bids, which of-
ten rely on equity financing, are also prone to booms and quiet periods by
industry. New closed-end funds are started in those years when seasoned
funds trade at small discounts or at premia relative to net asset value ~Lee,
Shleifer, and Thaler ~1991!!, and tend to be issued at a premium ~plus com-
mission! before reverting to a discount in the aftermarket ~Peavy ~1990!!.
D. Actions Possibly Taken to Create Misvaluation
Firms sometimes make accounting adjustments ~accruals! to boost their
earnings relative to actual cash flow. These adjustments are publicly dis-
closed in firms’ financial statements. When accruals are abnormally high,
7
3
stocks on average subsequently experience poor return performance. Man-
agers boost accruals at the time of new IPO and seasoned equity issues
~
Teoh et al. ~1998a, 1998b!!. Greater earnings management in IPOs and in
SEOs are associated with more optimistic errors in analyst earnings fore-
casts, and with more adverse subsequent long-run abnormal stock returns.7
Managers adjust earnings to meet threshold levels such as zero, past lev-
els, and levels forecast by analysts ~DeGeorge, Patel, and Zeckhauser ~1999!!.
Possibly under the influence of management, stock analysts on average “walk
down” their forecasts from overly optimistic levels at long horizons to pes-
simistic forecasts that firms are likely to beat by year-end ~Richardson et al.
4
~
1999!!.
7
1
See Shiller ~1982!, Hansen and Singleton ~1983!, Mehra and Prescott ~1985!, and Hansen
and Jagannathan ~1991!. Purely rational explanations have been offered based upon learning
Brennan and Xia ~2001!!, luck ~Fama and French ~2000!!, selection bias in the focus of aca-
~
demic attention ~Brown, Goetzmann, and Ross ~1995!!, borrowing constraints ~e.g., Constan-
tinides, Donaldson, and Mehra ~2000!!, and non-stock-market income shocks ~e.g., Constantinides
and Duffie ~1996! and Heaton and Lucas ~1996!!.
7
2
See, for example, Jindra ~2000!, D’Mello and Shrof ~2000!, and Dittmar ~2000!; Korajczyk,
Lucas, and McDonald ~1991! provide a possible rational explanation for this phenomenon.
7
3
See Sloan ~1997!, Teoh, Welch, and Wong ~1998a!, Teoh et al. ~1998b!, and Chan, Je-
gadeesh, and Lakonishok ~2000!.
7
4
See Teoh et al. ~1998a, 1998b! and Teoh and Wong ~2001!.

1
562
The Journal of Finance
E. Quality of Information Aggregation
In contrast with early classic work on experimental markets, the thrust of
much experimental market research in the late 1980s and 1990s is that in
only a slightly more complicated environment, information is not aggregated
efficiently ~see, e.g., the survey of Libby, Bloomfield, and Nelson ~2001! and
Sunder ~1995!!. Presumably this is because confounding effects make it harder
for investors to disentangle the reasons behind the trades of others ~see, e.g.,
Bloomfield ~1996!!.
F. Investor Behavior
Portfolio theory suggests that ~apart from transaction costs! everyone should
participate in all security markets. But even now, many investors neglect
major asset classes. Nonparticipation may derive from salience bias, or from
mere exposure ~familiarity! effects. Investors are subject to a strong bias
toward investing in stocks based in their home country and in their local
region.75 Employees invest heavily in their own firm’s stock and perceive it
to have low risk ~Huberman ~1999!!. The degree to which they invest in their
employer’s stock does not predict the stock’s future returns ~Benartzi ~1997!!.
There is also experimental evidence that investors sometimes fail to form
7
6
efficient portfolios and violate two-fund separation.
Several, though not all, studies of investor behavior in natural and exper-
imental markets report evidence consistent with a disposition effect—a greater
7
7
readiness to realize gains than losses. Certain groups of investors change
their behaviors in parallel ~“herding”!, in some cases engaging in momentum
7
8
~
or positive feedback! trading and in other cases in contrarian trading.
Similar behavior is not irrational per se, but some groups of investors do
poorly.
People ~especially males! seem to trade too aggressively, incurring higher
7
9
transactions costs without higher returns. Furthermore, traders in exper-
imental markets do not place enough weight on the information and actions
of others ~Bloomfield, Libby, and Nelson ~1999!!. Both findings are consis-
tent with overconfidence. In experimental markets, as in psychological ex-
7
5
See, for example, Cooper and Kaplanis ~1994!, Tesar and Werner ~1995!, Kang and Stulz
1997!, Coval and Moskowitz ~1999!, Huberman ~1999!, Lewis ~1999!, and Grinblatt and Kelo-
harju ~2001a!.
~
7
6
Kroll, Levy, and Rapoport ~1988a, 1988b!, Kroll and Levy ~1992!, and Bossaerts, Plott, and
Zame ~2000!.
7
7
See, for example, Shefrin and Statman ~1985!, Ferris, Haugen, and Makhija ~1988!, Odean
1998a!, Weber and Camerer ~2000!, and Grinblatt and Keloharju ~2001b!; but see also Rangue-
lova ~2000!.
~
7
8
See Lakonishok, Shleifer, and Vishny ~1992!, Grinblatt, Titman, and Wermers ~1995!, Choe,
Kho, and Stulz ~1999!, Nofsinger and Sias ~1999!, Wermers ~1999!, and Grinblatt and Keloharju
2000!.
~
7
9
See Odean ~1999! and Barber and Odean ~2000a, 2000b!.