Munich Personal RePEc Archive

Herd Behavior and Cascading in Capital

Markets: A Review and Synthesis

Hirshleifer, David and Teoh, Siew Hong

Fisher College of Business, The Ohio State University

19 December 2001

Online at https://mpra.ub.uni-muenchen.de/5186/

MPRA Paper No. 5186, posted 07 Oct 2007 UTC

De ember 19, 2001

Herd Behavior and Cas ading in Capital Markets:

A Review and Synthesis

David Hirshleifer� and Siew Hong Teoh��

Abstra t:

We review theory and eviden e relating to herd behavior, payo� and reputationalintera tions, so ial learning, and informational as ades in apital markets. We o�era simple taxonomy of e�e ts, and evaluate how alternative theories may help explaineviden e on the behavior of investors, �rms, and analysts. We onsider both in entivesfor parties to engage in herding or as ading, and the in entives for parties to prote tagainst or take advantage of herding or as ading by others.

Key Words: herd behavior, informational as ades, so ial learning, analyst herding, apital markets, �nan ial reporting, behavioral �nan e, investor psy hology, market ef-� ien y

� David Hirshleifer, Kurtz Chair in Finan e, Fisher College of Business, Ohio StateUniversity, 2100 Neil Avenue, Columbus, OH 43210-1144; Telephone: 614-292-5174;Fax: 614-292-2418; Email: hirshleifer 2� ob.osu.edu.

�� Siew Hong Teoh, Fisher College of Business, Ohio State University, 2100 Neil Avenue,Columbus, OH 43210-1144; Telephone: 614-292-6547; Email: teoh 2� ob.osu.edu.

We thank Darrell DuÆe for helpful omments, and Seongyeon Lim for ex ellent resear hassistan e.

Men nearly always follow the tra ks made by others and pro eed in their

a�airs by imitation...

| Ni olo Ma hiavelli, The Prin e, Ch. 6, 1514

1 Introdu tion

We are in uen ed by others in almost every a tivity, and this in ludes investment and

�nan ial transa tions. For example, it is reported as news when Warren Bu�ett buys

a sto k or ommodity, and this news a�e ts its pri e (see Se tion 6). Su h in uen e

may be entirely rational, but investors and managers are often a used of irrationally

onverging in their a tions and beliefs, perhaps be ause of a `herd instin t,' or from a

ontagious emotional response to stressful events.1

There are ertainly some phenomena that are suggestive of irrational herding by

markets, su h as ane dotes of market pri e movements without obvious justifying news;

examples that (with the bene�t of hindsight) look like mistakes, su h as the overpri ing

of U.S. te hnology sto ks in the late 1990s; the fa t that orporate a tions su h as new

issues and takeovers move in waves; and the tenden y of analysts to be enamored with

ertain se tors at di�erent times. Pra tioners and the media dis ussions are mu h too

ready to jump from su h patterns to the on lusion that irrational herding is proved. A

fully rational market may rea t to information that the resear her has failed to per eive;

market eÆ ien y does not mean perfe t foresight, so we expe t analyst fore asts and

market pri es to be wrong ex post; and orporate a tions may move in waves in rational

response to hanging fundamental onditions.

There has, of ourse, been a great deal of serious theoreti al and empiri al exploration

of the proposition that irrational investor errors ause market misvaluation of assets.

This in ludes some exploration of whether there is ontagion in biases a ross di�erent

investor groups, or from analysts to investors; and exploration of whether �rms take

a tions to exploit market misvaluation (for re ent reviews, see Hirshleifer (2001) and

Daniel, Hirshleifer, and Teoh (2002)).

However, a ademi resear h has also ontributed in a di�erent way to our under-

standing of these issues. Re ent theoreti al work on so ial learning and behavioral

onvergen e indi ates that some phenomena that seem irrational an a tually arise very

1See, e.g., Business Week (1998) on \Why Investors Stampede: ... And why the potential fordamage is greater than ever," or the advertisement by S udder Investments in Forbes (10/29/01) withthe heading, \MILLIONS of very fast, slightly MISINFORMED sheep. Now that's opportunity."

1

naturally in fully rational settings. Su h phenomena in lude: (1) frequent onvergen e

by individuals or �rms upon mistaken a tions based upon little investigation and little

justifying information; (2) the tenden y for so ial out omes to be fragile with respe t

to seemingly small sho ks; and (3) the tenden y for individuals or �rms to delay de i-

sion for extended periods of time and then, without obvious external trigger, suddenly

rush to a t simultaneously. There has also been theoreti al work on reputation-building

in entives by managers, whi h has fo used primarily on issue (1), but whi h has also

o�ered explanations for why some managers may deviate from the herd as well.

In this paper we review both fully rational and imperfe tly rational theories of be-

havioral onvergen e; their impli ations for investor trading, managerial investment and

�nan ing hoi es, analyst following and fore asts, market pri es, market regulation, and

welfare; and asso iated empiri al eviden e. Learning from pri es is by now familiar in

apital markets resear h, but we will argue here that more personal learning from quan-

tities (individual a tions), from out omes, and from onversation is also important for

markets.

We examine here behavioral onvergen e and u tuations in the behavior of in-

vestors, se urity analysts, and �rms in their respe tive de isions. Investors may `herd'

( onverge in behavior) or ` as ade' (ignore their private information signals) in de iding

whether to parti ipate in the market, what se urities to trade, and whether to buy or

sell. Both analysts and investors may herd in de iding what se urities to dis uss and

study. Analysts may also herd in the fore asts they o�er. We will onsider how herding

or as ading may a�e t market pri es. Furthermore, �rms an herd in their investment

de isions, in their �nan ing de isions, and in their reporting de isions. For example,

�rms may herd in the timing of new issues, in the adoption of fashionable investment

proje ts, or in their de isions of how to report earnings. Also, �rms an take a tions to

prote t against or exploit herding and as ading by investors and analysts.

In summary, our main goals are:

1. To provide a simple taxonomy of herding, payo� and reputation intera tions, so ial

learning and as ading.

2. Review riti ally the strengths and limitations of the basi analyti al frameworks

for understanding so ial learning based on observing others, and for understanding

reputation-building in entives to onverge or diverge behaviorally.

3. Review the eviden e from apital markets regarding herd behavior or as ades,

and evaluate how alternative theories may help explain eviden e on the behavior

2

of investors, �rms, and analysts. This in ludes onsideration of both in entives for

parties to engage in herding or as ading, and the in entives for parties to prote t

against or take advantage of herding or as ading by others.

Some issues omitted issues here are so ial learning and imitation in games (see,

e.g. Fudenberg and Kreps (1995), Gale and Rosenthal (2001)), and the vast general

literatures on so ial learning through pri es (e.g., Grossman and Stiglitz (1976)), and

on the learing me hanisms by whi h trades are onverted to pri es (e.g., Glosten and

Milgrom (1985), Kyle (1985)).

The remainder of this paper is stru tured as follows. Se tion 2 lassi�es me hanisms

of learning and behavioral onvergen e. Se tion 3 des ribes basi prin iples and alter-

native e onomi s enarios in rational learning models Se tion 4 des ribes agen y and

reputation-based herding models. Se tion 5 des ribes theory and eviden e on herding

and as ades in se urity analysis. Se tion 6 des ribes herd behavior and as ades in

se urity trading. Se tion 7 des ribes the pri e impli ations of herding and as ading

and their relation to bubbles. Se tion 8 des ribes herd behavior and as ades in �rms'

investment, �nan ing, and reporting de isions. Se tion 9 on ludes.

2 Taxonomy and me hanisms of so ial learning and

behavioral onvergen e

An individual's thoughts, feelings and a tions an be in uen ed by other individuals by

several means: by words, by observation of a tions (e.g., observation of quantities su h

as supplies and demands), and by observation of the onsequen es of a tions (su h as

individual payo�s, or market pri es). This in uen e may involve fully rational learning,

a quasi-rational pro ess, or even in ways that do not improve the observer's de isions at

all.

The pro ess of so ial in uen e an promote onvergen e or divergen e in behavior;

Figure 1 provides a taxonomy of di�erent sour es of onvergen e or divergen e. We

do not regard as onvergen e mere random formations with illusory appearan e of sys-

temati groupings. Our fo us also ex ludes mere lustering, wherein people a t in a

similar way owing to the parallel independent in uen e of a ommon external fa tor.

Our fo us is on onvergen e or divergen e brought about by a tual intera tions between

individuals.

Herding/dispersing is de�ned to in lude any behavior similarity/dissimilarity brought

3

about by the intera tion of individuals. (Originally herding referred to physi al lump-

ing, but this has been extended by e onomists to onvergen e in the a tion spa e.)

Possible sour es in lude:

1. Payo� externalities (often alled network externalities or strategi omplementar-

ities); for example, it pays for one person to use email if everyone else does too;

2. San tions upon deviants (as when dissidents in a di tatorship are jailed or tortured)

3. Preferen e intera tions (some individuals may prefer to wear Versa e this season,

just be ause everyone else is; others may prefer to deviate the olor that is `in' this

season);

4. Dire t ommuni ation (someone may simply state whi h of two alternatives are

better- but it is not so simple, sin e there is an issue of redibility),

5. Observational in uen e (an individual may observe the a tions of others or onse-

quen es of those a tions).

Figure 1 des ribes a double hierar hy of means of onvergen e. At the top of the

hierar hy is the most in lusive ategory, herding/dispersing. Re tangles depi t the ob-

servational hierar hy (A, B, C, D), whi h des ribes the informational sour es of herding

or dispersing. These in lude:

� A. herding/dispersing: Observation of others an lead to dispersing instead of

herding. For example, if preferen es are opposing.

� B. Observational In uen e: Dependen e of behavior upon the observed behavior

of others, or the results of their behavior; may be imperfe tly rational.

� C. Rational Observational Learning: Observational in uen e resulting from ratio-

nal Bayesian inferen e from information re e ted in the behavior of others, or the

results of their behavior.

� D. Informational Cas ades: (Observational learning in whi h the observation of

others (their a tions, payo�s, or even onversation) is so informative that an indi-

vidual's a tion does not depend on his own private signal).

The last ategory, informational as ades, des ribes a ondition in whi h imitation

will o ur with ertainty. Even as simple a form of so ial intera tion as imitation o�ers

4

a ru ial bene�t: it allows an individual to exploit information possessed by others

about the environment. When a friend is eeing rapidly, it may be good to run even

before seeing the saber tooth tiger hasing around the bend. The bene�t from imitating

others, and of taking into a ount the payo� out omes of others, is fundamental, as

eviden ed by the observation of su h behavior in many kinds of animals. Even when

imitation probably does not o ur through a `rational' pro ess of analysis, the pro livity

to imitate may be well attuned to osts and bene�ts through the guidan e of natural

sele tion. We will use the word imitation broadly to in lude sub-rational me hanisms

that indu e an individual to be in uen ed by the behavior of another individual to

behave the same way.

There is an extensive literature in both psy hology and zoology on imitation in

many animal spe ies, both in the wild and experimentally (see, e.g., Gibson and Hoglund

(1992), (Giraldeau (1997), and Dugatkin (1992)). Imitation has been do umented among

birds, �sh, and mammals in foraging and diet hoi es, sele tion of mates, sele tion of

territories, and in means of avoiding predators. Indeed, Bla kmore (1999) (e.g., pp.

74-81) suggests that in early hominids there was strong sele tion for ability to imitate

innovative, omplex behaviors, so that the evolution of large brain size was linked to the

rise of the propensity to imitate. Starting within an hour of birth, humans also engage

in imitation. There is also ontagion in the emotions of individuals intera ting as groups

(see, e.g., Barsade (2001)).

An individual is said to be in an informational as ade if, based upon his observation

of others (e.g., their a tions, out omes, or words), his sele ted a tion does not depend on

his private information signal (see Bikh handani, Hirshleifer, and Wel h (1992), Wel h

(1992) and Banerjee (1992) [Banerjee uses the term `herd' for what we refer to here as

a as ade℄). In su h a situation, his a tion hoi e is uninformative to later observers.

Thus, as ades tend to be asso iated with information blo kages. Su h blo kages are

an aspe t of an informational externality: an individual making a hoi e may do so for

private purposes with little regard to the potential information bene�t to others.

Gale (1996) reviews models of so ial learning and herding in general..For an ex-

position and des ription of appli ations of informational as ades, see Bikh handani,

Hirshleifer, and Wel h (1998); Bikh handani, Hirshleifer, and Wel h (2001) provides an

annotated bibliography of resear h relating to as ades.

Returning to Figure 1, re tangles depi t the payo� intera tion hierar hy (I, II, III),

whi h provides a di�erent (though not mutually ex lusive) perspe tive on herding or

dispersing. These in lude:

5

� I. Herding/Dispersing (as in the information hierar hy)

� II. Payo� and Network Externalities This involves onvergen e or divergen e of

behavior arising from the fa t that an individual's a tion a�e ts the payo�s to

others of taking that a tion. The lassi model of herding as a dire t payo�

intera tion is Hamilton's ((1971)) analysis of the geometry of the `sel�sh herd,'

wherein the lumping of prey animals is an indire t out ome of the sel�sh attempt

by ea h one to put others between itself and predators. In �nan ial e onomi s, the

Diamond and Dybvig (1983) bank run model involves a dire t payo� externality,

and the Admati and P eiderer (1988) theory of volume lumping involves payo�

intera tions indu ed by the in entive for uninformed investors to try to trade with

ea h other instead of with the informed.

� III. Reputational Herding and Dispersion

This is onvergen e or divergen e of behavior based on the attempt of an individual

to maintain a good reputation with another observer. Su h a desire for good

reputation an ause payo� intera tions, making III a subset of II (see S harfstein

and Stein (1990), Rajan (1994), Trueman (1994), Brandenburger and Polak (1996),

and Zwiebel (1995).) Ottaviani and Sorenson (2000) explore the relation between

reputational herding and informational as ades.

3 Basi Prin iples and Alternative E onomi S e-

narios in Rational Learning Models

3.1 Some Basi Prin iples

We begin by des ribing further some features of the basi informational as ades model,

whi h provides a simple way to illustrate some prin iples ommon to models of rational

observational learning (item C) as well as those unique to the as ades setting. The

o urren e of an informational as ade an even lead to a omplete information blo kage.

Consider a sequen e of ex ante identi al individuals who fa e similar hoi es, observe

onditionally independent and identi ally distributed private information signals, and

who observe the a tions but not the payo�s of prede essors. Suppose that individual i

is in a as ade, and that later individuals understand this. Then individual i+1, having

gained no information by observing the hoi e of i, is, informationally, in a position

identi al to that of i. So i + 1 will also make the same hoi e regardless of his private

6

signal. By indu tion, this reasoning extends to all later individuals- the a umulation

of information omes to a s ree hing halt on e a as ade begins.

The on lusion that information is blo ked forever is of ourse too extreme, for

several reasons. First, a publi ly observable sho k an dislodge a as ade. Se ond,

if individuals are not ex ante identi al, then the arrival of an individual with deviant

information or preferen es an dislodge a as ade. Third, the o urren e of a as ade

requires that individual do not re eive an arbitrarily pre ise signal- likelihood ratios must

be �nitely bounded (on all these items, see, e.g., Bikh handani, Hirshleifer, and Wel h

(1992)). Fourth, whatever hoi e is �xed upon in the as ades, if payo� out omes from

that hoi e eventually work their way into the publi information pool, as ades an be

dislodged.2 Thus, the more plausible impli ation to be drawn from the basi as ades

model is just that information aggregation an be unduly slow relative to what ould

in prin iple be attained; and that blo kages an o ur whi h may last for signi� ant

periods of time (see, e.g., the dis ussion of Gale (1996)).

A generalization of the as ades on ept is what an be alled a behavioral oars-

ening. This is any situation in whi h an individual takes the same a tion for multiple

signal values. In su h a situation his information is not fully onveyed by his a tions to

observers. Behavioral oarsening leads to partial information blo kage. A as ade is the

extreme ase in whi h the oarsening overs all possible signal values, so that blo kage

is omplete.

The poor aggregation of information in informational as ades of ourse means that

de isions will also be poor, even if the signals possessed by numerous individuals ould

in prin iple be aggregated to determine the right de ision with virtual ertainty. Sin e

the model is fully rational, individuals understand perfe tly well that the pre ision of

the publi pool of information impli it in prede essors' a tions is quite modest. As a

result, even a rather small publi sho k an ause a longstanding and popular a tion to

swit h.

Although the arrival of enough publi information will improve de isions, the ar-

rival of a signal publi dis losure may, paradoxi ally, make de isions worse. Additional

information an en ourage individuals to fall into a as ade sooner, aggregating the in-

formation of fewer individuals, so there is no presumption that the signal will improve

de isions in the as ade (Bikh handani, Hirshleifer, and Wel h (1992)). For similar rea-

sons, the ability of individuals to observe past a tions with low noise instead of high

noise, or the ability to observe payo� out omes in addition to past a tions, an make

2However, bad as ades need not be dislodged with ertainty; see Cao and Hirshleifer (2000).

7

de isions worse on average (Cao and Hirshleifer (1997, 2000))- \a little knowledge is a

dangerous thing."3

In a real investment ontext, the assumption of the basi as ades model that the

timing and order of moves is exogenously given is unrealisti . When individuals have a

hoi e of whether to delay, there an be long periods with no investment, followed by

sudden spasms in whi h the adoption of the proje t by one �rm triggers the exer ise of

the investment option by many other �rms (Chamley and Gale (1994)).4

Most of the ideas des ribed above an be generalized to models of so ial learning

in whi h as ades do not o ur. Even when information blo kage is not omplete,

information aggregation is limited by the fa t that individuals privately optimize rather

than taking into a ount their e�e ts upon the publi information pool. In parti ular,

there is a general tenden y for information aggregation to be self-limiting. At �rst,

when the publi pool of information is very uninformative, a tions are highly sensitive

to private signals, so a tions add a lot of information to the publi pool. (The addition

an be dire tly through observation of past a tions, or indire tly through observation

of onsequen es of past a tions, as in publi payo� information that results from new

experimentation on di�erent hoi e alternatives.) As the publi pool of information

grows, individuals' a tions be ome less sensitive to private signals.

The loss of sensitivity of a tions to private signals an o ur suddenly, with a swit h

from full usage of private signals to no usage of private signals (as in Banerjee (1992),

and the binary example of Bikh handani, Hirshleifer, and Wel h (1992)). It an o ur

gradually (as in the more general as ades model with multiple signal values), yet still

rea h a point of omplete blo kage (as in Bikh handani, Hirshleifer, and Wel h (1992)).

Or, it an o ur gradually but never rea h a point of omplete blo kage. For example, it

an o ur that there is always a probability that individuals use their own signals, but

where that probability asymptotes toward zero; this leads to `limit as ades' (Smith and

Sorenson (2000)). Alternatively, there an be as ades proper, but owing to observability

of proje t payo�s, there an be a probability less than one that the as ade evenetually

breaks (see Cao and Hirshleifer (2000)). Or, if there is some sort of observation noise, the

publi pool of information an grow steadily but more and more slowly (Vives (1993).

In sum, whether information hannels be ome qui kly or only gradually logged,

3Also, the ability to learn by observing prede essors an make the de isions of followers noisier byredu ing their in entives to olle t (perhaps more a urate) information themselves (Cao and Hirshleifer(1997)).

4See also Hendri ks and Koveno k (1989), Bhatta harya, Chatterjee, and Samuelson (1986), Zhang(1997) and Grenadier (1999).

8

and whether the blo kage is omplete or partial, is dependent on the e onomi setting;

but the general on lusion that there an be long periods in whi h individuals herd

upon poor de isions is robust. Also in general there tends to be too mu h opying or

behavioral onvergen e; someone who uses his own private information heavily provides

a positive externality to followers, who an draw inferen es from his a tion..

The as ade out ome des ribed by Banerjee (1992) or Bikh handani, Hirshleifer,

and Wel h (1992) is based on the publi pool of information dominating the individual's

private signal. Obviously, this annot o ur with ertainty if the private signal likelihood

ratios are unbounded. However, the growth of the publi information pool may be

ex ru iatingly slow, so even in settings where people o asionally observe extremely

informative signals a as ades model an be a good approximation. In parti ular, as the

publi pool of information grows more informative, the likelihood that an individual will

depart from it substantially based on an extreme signal be omes very small.

Thus, the as ades and some other rational learning theories have several general

impli ations:

� idiosyn rasy (poor information aggregation). Behavior resulting from signals of

just �rst few individuals drasti ally a�e ts behavior of numerous followers.

� fragility (fads). When as ades form, there is omplete blo kage of information

aggregation, sensitivity to small sho ks. As in Hollywood adventure movies, it is

inevitable that the ar will end up teetering pre ariously at the very edge of the

pre ipi e.

� Simultaneity (delay followed by sudden joint a tion). Endogenous order of moves,

heterogeneous preferen es and pre isions an exa erbate these problems so that

sudden ` hain rea tions,' `stampedes' or `avalan hes' o ur.

� Paradoxi ality (greater publi information, or greater observability of the a tions

or payo�s of others does not ne essarily improve welfare or even the a ura y of

de isions).

� Path dependen e (out omes depend on the order of moves and information arrival).

This impli ation is shared with models with payo� interdependen e (e.g., Arthur

(1989)).

9

3.2 Alternative E onomi Settings

We now des ribe in somewhat more detail alternative sets of assumptions in observa-

tional in uen e models and the impli ations of these di�eren es.5

3.2.1 Observation of Past A tions Only

Here we retain the assumption of the basi as ade model that only past a tions are

observable, but onsider the a variety of model variations.

1. Dis rete, Bounded, or gapped a tions vs. ontinuous unbounded a tions

If the a tion spa e is ontinuous, unbounded, and without gaps, then an individual's

a tion is always at least slightly sensitive to his private signal. Thus, a tions always

remain informative, and informational as ade never form. Thus, informational as ade

require some dis reteness, boundedness or gaps (Lee (1993); see also Vives (1993) and

Gul and Lundholm (1995)). The earliest as ade models were based upon dis reteness

(as in Bikh handani, Hirshleifer, and Wel h (1992), Wel h (1992)) or on the equivalent

of a binary a tion spa e (Banerjee (1992)).

The assumption of dis reteness is in many settings highly plausible. We vote for one

andidate or another, not for a weighted average of the two. Often alternative investment

proje ts are mutually ex lusive. Although the amount invested is often ontinuous, if

there is a �xed ost the option of not investing at all is dis retely di�erent from positive

investment.

More broadly, one way in whi h the a tion set an be bounded is if there is a minimum

and maximum feasible proje t s ale. If so, then when the publi information pool is

suÆ iently favorable a as ades at the maximum s ale will form, and when the publi

information pool is suÆ iently adverse individuals will as ades upon the minimum

s ale. Sin e there is always an option to reje t a new proje t, investment has a natural

extreme a tion of zero. Chari and Kehoe (2000) provide a model where a lower bound

of zero on a ontinuous investment hoi e reates as ade.6

Similarly, gaps an reate as ades. For example, it may be that signi� ant new

investment or signi� ant disinvestment is feasible, but owing to �xed osts a very small

hange is learly unpro�table. If so, then as ades upon no a tion is feasible if private

5We do not review the growing literature on how rates of learning vary during ma roe onomi u tuations and how this an ontribute to booms and rashes in levels of investment (see, e.g., Gonzalez(1997), Chalkley and Lee (1998), Chamley (1999), Veldkamp (2000)).

6Asymmetry between adoption and reje tion of proje ts is often realisti and has been in orporatedin several so ial learning models of investment to generate interesting e�e ts.

10

signals are not too informative.

Even if the true a tion spa e is ontinuous, ungapped and unbounded, to the extent

that observers are unable to per eive or re all small fra tional di�eren es, the a tions of

their prede essors e�e tively be ome either noisy or dis rete. Dis retizing an potentially

ause as ades and information blo kage; noise similarly slows down learning. There

must be at least some e�e tive dis reteness or noise be ause real observers have �nite

per eptual and ognitive powers. At some point, it is literally physi ally impossible

for an observer to per eive arbitrarily small di�eren es in a tions. Even if per eption

were perfe t, it would also be impossible, in the absen e of in�nite time and al ulating

apa ity, to make use of arbitrarily small observed di�eren es in a tions. Thus, for

fundamental reasons there must be either noise, per eptual/analyti dis retizing, or

both.7

If per eptual dis retizing is very �ne-graded, the out ome will still be very lose to

full revelation. However, it is doubtful that per eption and analysis is onsistently �ne-

graded; onsider, for example, the tenden y for people to round o� numbers in memory

and onversation. Kahn, Penna hi, and Sopranzetti (2002) �nd lustering for retail

deposit interest rates around integers, and provide eviden e that is supportive of their

model in whi h this is aused by limited re all of investors.

2. Costless versus ostly private information a quisition

Individuals may observe private signals ostlessly in the ordinary ourse of life, or may

expend resour es to obtain signals. Most so ial learning models take the ostless route.

Costs of obtaining signals an lead to little a umulation of information in the so ial

pool for essentially the same reason as in other as ades or herding models. Individuals

have less in entive to investigate or observe private signals if the primary bene�t of

using su h signals is the information that su h use will onfer upon later individuals.

(Burguet and Vives (2000) analyze so ial learning with investigation osts). Indeed, if

an individual rea hes a situation where he optimally would not make use of a signal,

then learly it does not pay for him to expend resour es to obtain it. The out ome is

similar to that of the basi as ades model: information blo kage.

This suggests an extended de�nition of as ades that an apply to situations where

private signals are ostly to obtain. An investigative as ade is a situation where either:

7In the absen e of dis retizing, repeated opying will gradually a umulate noise until the information ontained in a distant past a tion is overwhelmed. This overwhelming of analog signals by noise whenthere is sequential repli ation is the reason that information must be digitized in the geneti ode ofDNA, and in information that is sent (with repeated reampli� ation of signals) over the internet.

11

1. An individual a ts without regard to his private signal; or,

2. The individual hooses not to a quire a ostly signal, but he would have a ted

without regard to that signal if he were for ed to a quire the same level of signal

pre ision that he would have a quired voluntarily if he were unable to observe the

a tions or payo�s of others.

Calvo and Mendoza (2001) study the de isions by individuals to investigate and

invest in di�erent ountries. If investigation of ea h ountry requires a �xed ost, they

�nd that the optimal amount of investigation of a ountry diminishes rapidly with the

number of ountries, leading to greater herding.

3. Observation of all past a tions versus a subset or statisti al summary of a tions

Instead of observing all past a tions, it may be that people an observe only the most

re ent a tions, a random sample, or an only observe the behavior of their neighbors.

Some models with these features are dis ussed elsewhere; we note here that in su h

settings mistaken as ades an still form. Alternatively, individuals may only be able

to observe a statisti al summary of past a tions. Information blo kage and as ades are

possible in su h a setting as well (Bikh handani, Hirshleifer, and Wel h (1992)). (With

ontinuous a tions, as dis ussed above, the out ome may be slow information aggregation

rather than as ade; Vives (1993).) A possible appli ation is to the pur hase of onsumer

produ ts. Aggregate sales �gures for a produ t matter to future buyers be ause it

reveals how previous buyers viewed desirability of alternative produ ts (Bikh handani,

Hirshleifer, and Wel h (1992), Caminal and Vives (1999)).8

3. Observation of past a tions a urately or with noise

In most so ial learning models any a tions that are observed at all are observed

a urately, but in some there is noise (see Vives (1993), Cao and Hirshleifer (1997)).

Under spe ial ir umstan es a model in whi h individuals learn from pri e is in e�e t

a basi so ial leaerning model with indire t observation of a noisy statisti al summary

of the past trades of others. But in general a market pri e s enario is more omplex;

the onsequen e for an individual of taking an a tion is not just an exogenous payo�

fun tion, but the result of an equilibrating pro ess.

4. Choi e of timing of moves versus exogenous moves

8A SmithKline Bee ham advertisement states, \Do tors have already endorsed Tagamet in thestrongest possible way. With their pres ription pads." The add shows a bar graph in three-dimensionalperspe tive in whi h 237 million pres riptions tower above a modest 36 million for Pep id. A minis ulefootnote reveals that the Tagamet �gure was sin e 1977, Pep id only sin e 1986!

12

Chamley and Gale (1994) o�er a model of irreversible investment in whi h individuals

with private signals about proje t quality have a hoi e as to whether to invest or delay.

This is therefore a model of optimal option exer ise. They �nd that in equilibrium there

is delay. The advantage of delay is that an individual an gain information by observing

the a tions of others. But if everyone were to wait, there would be no advantage to

delay. Thus, in equilibrium investors follow randomized strategies in de iding how long

to delay before being the �rst to invest. Investment by an individual an trigger imme-

diate further investment by others. Indeed, in the limit a period of little investment is

followed by either a sudden surge in investment or a ollapse. Thus, the model illustrates

simultaneity). In equilibrium as ades o ur and information is aggregated ineÆ iently.

Zhang (1997) o�ers a setting in whi h investors have private information not only

about proje t quality, but about the pre ision of their signals. In the unique symmetri

equilibrium, among investors with favorable signals, those whose signals are less pre ise

delay longer than those with more pre ise signals (be ause impre ise investors have

greater need for orroborating information before investing). In equilibrium there is

delay until the riti al investment date of the individual who drew the highest pre ision

is rea hed. On e he invests, other investors all immediately follow, though investment

may be ineÆ ient. This sudden onset of investment illustrates simultaneity in an extreme

form.

Chamley (2001) �nds that when individuals have di�erent prior beliefs, there are

multiple equilibria that generate di�erent amounts of publi information. Chari and

Kehoe (2000) show that when there is a binary de ision of whether or not to invest, but

an endogenous hoi e of timing, onsistent with Chamley and Gale (1994) and Zhang

(1997), ineÆ ient as ades still o ur. They �nd that even when there is a ontinuous

level of investment bounded below by zero, an ineÆ ient as ade on zero investment an

o ur (for reasons dis ussed earlier). They also �nd that as ades remain even when

individuals have the opportunity to share information, be ause individuals do not have

an in entive to ommuni ate truthfully.9

A number of other models des ribe how information blo kages, delays in investment

and periods of sudden investment hanges, and overshooting an o ur, either with

(Caplin and Leahy (1994), Grenadier (1999)) or without (Caplin and Leahy (1993),

Persons and Warther (1997)) informational as ades. Caplin and Leahy (1994) analyze

informational as ades in the an ellation of investment proje ts in a setting with en-

9Gul and Lundholm (1995) examine a model that allows for delay in whi h a ontinuous a tion spa eleads to full revelation and therefore no as ades.

13

dogenous timing. They �nd that that there an be sudden rashes in the investments

of many �rms triggered by individual an ellations. These models share the broad intu-

itions that informational externalities ause hoi es about whether and when to invest

to be taken in a way that is undesirable from a so ial point of view.

Persons and Warther (1997) o�er a model of boom and bust in the adoption of

�nan ial innovations based upon observation of the payo�s resulting from the repeated

a tions of other �rms. They �nd a tenden y for innovations to `end in disappointment'

even though all parti ipants are fully rational; a natural onsequen e of learning is that

the boom ontinues to grow until disappointing news appears. Zeira (1999) develops

related notions of informational overshooting to real estate and sto k markets.

5. Presen e of an evolving publi ly observable state variable

Grenadier (1999) examines informational as ades in options exer ise, in whi h an

exogenously evolving publi ly observable state variable in uen es the in entives to ex-

er ise the option. A small re ent move in the state variable an be the `straw that broke

the amel's ba k' in triggering informational as ades of option exer ise. Grenadier

suggests several appli ations, su h as \the building of an oÆ e building, the drilling of

an exploratory oil well, and the ommitment of a pharma euti al ompany toward the

resear h of a new drug."

6. Stable versus sto hasti hidden environmental variable

Bikh handani, Hirshleifer, and Wel h (1992) provide an example where the underly-

ing state of the world is sto hasti but unobservable. This an lead to fads wherein the

probability that a tion hanges is mu h higher than the probability of a hange in the

state of the world.

Perktold (1996) assumes a Markov pro ess on the value of the hoi e alternatives,

and individuals make repeated de isions over time. He �nds that as ades o ur and

break re urrently. Mos arini et al (1998) examine how long as ades an last as the

environment shifts. Nelson (2001) explores the relation between high orrelation of in-

dividual a tions and as ades. She o�ers a model of IPOs in whi h the de ision to

go publi is more likely to be asso iated with informational as ades than the de ision

to hold o�.10 Hirshleifer and Wel h (2002) onsider an individual or �rms subje t to

10Nelson also points out that are is needed in the testing of herding and as ades models if theproxy used is orrelation of behavior. She shows that there is often a lower orrelation of behavior in asetting with as ades than in a setting where all the information is made publi . This is be ause publi information indu es high orrelation in a tions: people onverge to the right a tion. On the other hand,if the ben hmark for omparison is one where ea h individual's information remains private, herdingand as ades will be asso iated with higher orrelation of a tion. So it is still reasonable in testing

14

memory loss about past signals but not a tions. They des ribe the determinants (su h

as environmental volatility) of whether memory loss auses inertia (a higher probabil-

ity of ontinuing past a tions than if memory were perfe t) or impulsiveness (a lower

probability).

7. Homogeneous versus heterogeneous payo�s

Individuals have di�erent preferen es, though this is probably more important in

non-�nan ial settings. Suppose that di�erent individuals value adoption di�erently. A

rather extreme ase is opposing preferen es or payo�s, so that under full information

two individuals would prefer opposite behaviors. If ea h individual's type is observable,

di�erent types may as ades upon opposite a tions.

However, if the type of ea h individual is only privately known, and if preferen es

are negatively orrelated, then learning may be onfounded| individuals do not know

what to infer from the mix of pre eding a tions they observe, so they simply follow their

own signals (Smith and Sorenson (2000)).

8. Endogenous ost of a tion: market models with pri e

This is a large topi that we over separately below.

9. Single or repeated a tions and private information arrival

Most models with private information involve a single irreversible a tion, and a single

arrival of private information. In Chari and Kehoe (2000), in ea h period one investor

re eives a private signal, and investors have a timing hoi e as to when to ommit to

an irreversible investment. In equilibrium there are ineÆ ient as ade. If individuals

take repeated, similar, a tions and ontinue to re eive non-negligible additional informa-

tion, a tions will of ourse be ome very a urate. However, there an still be short-run

ineÆ ien ies (e.g., Hirshleifer and Wel h (2002).

10. Dis rete signal values versus ontinuous signal values

Depending on probability distributions, possible to get limit as ades (Smith and

Sorenson (2000)) instead of as ades. As ommented by Gale (1996), the empiri al

signi� an e is mu h the same|information aggregation an be poor large periods of

time.

11. Exogenous rules versus endogenous ontra ts and institutional stru ture

Some papers that examine how the design of institutional rules and of ompensation

ontra ts a�e ts herding and informational as ades in proje t hoi e in lude Prender-

gast (1993), Khanna (1997), and Khanna and Slezak (2000) (dis ussed below); see also

su h models to examine behavioral onvergen e. But a fuller test of su h models would look examinewhether high onvergen e in behavior is a hieved without high a ura y of de isions.

15

Ottaviani and Sorenson (2001).

3.2.2 Observation of Consequen es of Past A tions

Vi arious learning is so powerful that one might expe t that observing past payo�s would

eliminate information blo kages and lead to onvergen e upon orre t a tions. Indeed,

in an imperfe tly rational setting, Banerjee and Fudenberg (1999) �nd onvergen e to

eÆ ient out omes if people sample at least two prede essors. On the other hand, as em-

phasized by Shiller (2000a), in pra ti e imperfe t rationality makes onversation a very

imperfe t aggregator of information. This suggests that biases indu ed by onversation

may be important for sto k market behavior.

Even under full rationality, it should be noted that the Banerjee/Fudenberg setting

always leaves a ri h inventory of information to draw from. In ea h period a ontinuum

of individuals try all hoi e alternatives, so there is always a po ket of information

available about the payo� out ome of either proje t. Cao and Hirshleifer (2000) examine

a setting that is loser to the basi as ades model. There are two alternative proje t

hoi es, ea h of whi h has an unknown value-state. Payo�s are in general sto hasti

ea h period onditional on the value-state. Individuals re eive private signals and a t in

sequen e, and individuals an observe all past a tions and proje t payo�s. Nevertheless,

idiosyn rati as ades still form. For example, a sequen e of early individuals may

as ade upon proje t A, and its payo�s may be ome visible to all, perhaps revealing the

value-state perfe tly. But sin e the payo�s of alternative B are still hidden, B may be

the superior proje t. Indeed, the ability to observe past payo�s an sometimes trigger

as ades even more qui kly-an indi ation of parodoxi ality.

Caplin and Leahy (1993) examine a setting where potential industry entrants learn

indire tly from the a tions of previous entrants by observing industry market pri es.

Entrants do not possess any private information prior to entry. Imperfe t information

slows the adjustment of investment to se toral e onomi sho ks. (On the informational

and a tion onsequen es of �rms observing past payo�s, see also Persons and Warther

(1997) dis ussed earlier.)

3.3 Imperfe tly Rational Individuals

So far we have fo used primarily on fully rational models. Some models that assume

either me hanisti or imperfe tly rational de isionmakers in lude Ellison and Fudenberg

(1993, 1995) (rules of thumb), Hirshleifer, Subrahmanyam, and Titman (1994) (`hubris'

16

about the ability to obtain information qui kly), Bernardo and Wel h (2001) (over on-

�den e), Hirshleifer and Noah (1999) (mis�ts of several sorts), Hirshleifer and Wel h

(2002) (memory loss about past signals),

In the rules of thumb approa h the behavior of agents is spe i�ed based on analyti al

onvenien e, or on the resear her's judgment that the rule of thumb or heuristi would be

a reasonable one for agents with limited ognitive powers to follow. The other approa h is

to draw on experimental psy hology to suggest assumptions about imperfe t rationality

of agents in the model. Both approa hes have merit, but for both, veri� ation of the

behavioral assumptions is desirable. In parti ular, even behavioral assumptions that are

based broadly upon psy hologi al eviden e are usually not based upon experiments that

are very lose to the parti ular e onomi setting being modeled.

In Smallwood and Conlisk (1979), hoi es are based on payo�s re eived, and on

market share of the hoi e alternatives. Ellison and Fudenberg (1995) spe ify that an

individual takes an a tion if all individuals in the sample are using it, or if they obtained

a higher average payo� using the a tion than the alternative. In Ellison and Fudenberg

(1993), de isions are based upon past payo�s from a sample of observations from past

adoptions, and based upon the market shares of hoi e alternatives.

If individuals use a diversity of de ision rules (whether rational, quasi-rational, or

simple rules of thumb), then there will be greater diversity of a tion after a as ade

among rational individuals starts. This a tion diversity an be informative, and an

break as ades (Bernardo and Wel h (2001), Hirshleifer and Noah (1999)). This im-

proves the eÆ ien y of the hoi es of rational individuals in the long run.

There are many other possible dire tions to take imperfe t rationality and so ial

learning. Eviden e of emotional ontagion within groups suggests that there may be

merit to the popular views about ontagious manias or fads (see also Shiller (2000b),Lyn h

(2000), and Lux (1995)). On the other hand, some histori ally famous bubbles, su h as

that if the Dut h Tulip Bulbs, may have re e ted information rationally and fully (see,

e.g., Garber (2000)). Furthermore, there are rational models of bubbles and rashes that

do not involve herding (see, e.g., the agen y/intermediation model of Allen and Gale

(2000a), and the review of Brunnermeier (2001)).

We argue elsewhere that limits to investor attention are important for �nan ial re-

porting and apital markets (see the review of Daniel, Hirshleifer, and Teoh (2002), and

the model of Hirshleifer, Lim, and Teoh (2001)). Su h limits to attention may pressure

individuals to herd or as ade despite the availability of a ri h set of publi and private

information signals (beyond past a tions of other individuals). A related issue is whether

17

the tenden y to herd or as ade greater when the private information that individuals

re eive is hard to pro ess ( ognitive onstraints and the use of heuristi s for hard de-

ision problems were emphasized by Simon (1955); in the ontext of so ial in uen e,

see Conlisk (1996)). In this regard, Kim and Pantzalis (2000) provide eviden e that

apparent herd behavior by analysts is greater for diversi�ed �rms, for whi h the task

that analysts fa e is more diÆ ult.11

DiÆ ulty in analyzing opaque a ounting reports has been widely raised in the press

as a sour e of the re ent Enron deba le. In testimony to the House of Representatives on

De ember 12, 2001, the Dire tor of Thompson/First Call indi ated that when analysts

an not disentangle a �rm's a ounting there, tends to be greater herding in analyst

fore asts (i.e., smaller dispersion in fore asts) than is the ase for the average S&P 500

�rm.

3.4 Market Pri es, Herding, and Informational Cas ades

If markets are perfe t and investors are rational, then risk-adjusted se urity returns

are unpredi table. We will refer to this ombination of onditions- full rationality and

perfe t markets- as ` lassi al.' By perfe t markets we mean that ea h investors trades

as if he an buy or sell any amount at a given market pri e. Thus, even though a

rational expe tations model su h as that of Grossman and Stiglitz (1976) has information

asymmetry, sin e individuals per eive that they an trade at a given pri e, we view this

as a perfe t market. Furthermore, in a lassi al market there is neither an ex ess nor

a shortfall in pri e volatility relative to publi news arrival about fundamental value

(where we in lude as `publi ' even information that was originally private but whi h an

be rationally inferred by observing market pri es or trading) . It follows immediately

that fully rational models of as ades or herding annot explain anomalous eviden e

regarding return predi tability or ex ess volatility (for re ent reviews of theory and

eviden e relating to investor psy hology in apital markets, see, e.g., Hirshleifer (2001),

Daniel, Hirshleifer, and Teoh (2002)).

This is not to deny that information blo kages and herding may a�e t pri es. What

this does show is that to explain return patterns that are anomalous from the lassi al

viewpoint, it is ne essary to introdu e either market imperfe tions or failures of human

11Some physi ists and mathemati ians have o�ered heavily-engineered models of me hanisti agentsto examine the relation of herd behavior to pri e distributions (see, e.g., Cont and Bou haud (1999)). Anearly analysis of dire t preferen e for onformity was provided by Kuran (1989), but the informationalimpli ations have not been fully explored.

18

rationality.

Even within a fully rational setting, as ades or herding an have the serious e�e t

of blo king information aggregation. The properties of return unpredi tability, and of

orre t volatility in a lassi al market are relative to the information that an be inferred

from publi ly observable variables in luding market pri es and volumes. However, the

existen e of as ades an a�e t how mu h information goes into that information set in

two ways. First, it an ause some information to remain private whi h otherwise would

be re e ted in and inferable from pri es and trades. Se ond, it an ause individuals

to hange their investigation behavior, potentially redu ing the amount of private and

publi information that is generated in the �rst pla e.

Vives (1995) analyzes the rate of learning in ompetitive se urities markets. The

intuition is similar to the intuition in herding models with exogenous a tion osts. An

informed trader does not internalize the bene�t that other traders have from learning

his private information as revealed through trading. Thus, the rate of onvergen e of

pri e to eÆ ien y is slow.

In Glosten and Milgrom (1985), even though the a tion spa e is dis rete, there are

no informational as ades. This fa t has stimulated some analysis of how endogeneity

of pri es an a t to prevent as ades. In simple trading settings, as ades annot o ur

(see Avery and Zemsky (1998)). Intuitively, as ade would ontradi t market learing.

Se urities pri es should aggregate private information through trading. If there were a

as ade where informed traders were buying regardless of their signals, then a fortiori

so would uninformed traders. If the optimal response to even an adverse signal is to

buy, then so is the reponse to having no signal. But if, foreseeably, both informed and

uninformed are trying to buy, the marketmaker ought to have set pri es di�erently.

However, if there are multiple dimensions of un ertainty, then something akin to a

as ades an o ur. It is standard to assume that informed investors know more than

the market maker about the expe ted payo� of the se urity. Avery and Zemsky intro-

du e a se ond informational advantage to informed investors over the market maker{

un ertainty over whether informative signals were sent. In onsequen e, a pri e rise

an en ourage an investor with an adverse signal to buy when there is a transa tion

ost or bid-ask spread. The pri e rise persuades the investor that others possess fa-

vorable information, whereas the market maker adjusts pri es sluggishly in response to

this good news. This relative sluggishness of the marketmaker arises from his igno-

ran e over whether an informative signal was sent. Informed traders-even those with

adverse signals-at least know that information signals were sent, so that the previous

19

order probably ame from a favorably informed trader. In ontrast, the market maker

pla es greater weight on the possibility of a liquidity trade.

The behavior des ribed by Avery and Zemsky is very as ade-like, in that the individ-

ual is a ting in opposition to his private signal- a rather extreme behavioral oarsening.

However, it is in fa t not a true informational as ade. When no information signal

is re eived, the investor takes a di�erent a tion from when information is re eived. So

there are really three possible signal realizations-favorable, unfavorable, and no signal.

A tion is in fa t dependent on this appropriately rede�ned signal. In any ase, this

pseudo- as ading phenomenon leads to partial information blo kage.

It is worth noting that in a di�erent setting, true as ades may indeed o ur. Suppose

that A is sometimes informed, when A is informed B is aware that A is informed, but

C is not informed and does not know when others are informed. As usual there is also

non-information-based (`liquidity') trading. Then there would seem to be a bene�t to B

of imitating A's trade, and for C to take up the sla k.

Gervais (1996) �nds information blo kage owing to bid-ask spreads. In his model,

there is un ertainty about investors' information pre ision. Trading o urs over many

periods yet trader private information is not in orporated into pri e. Informed investors

re eive a signal and know the pre ision of the signal, but the market-maker does not.

Initially a high bid-ask spread a ts as a �lter by deterring trade by informed investors

unless they have high pre ision. However, as the market-maker observes whether trade

o urs, he is able to update about signal pre ision and about the value of the asset.

Owing to his in reased knowledge over time the market-maker narrows the spread. This

narrowing auses even investors with impre ise signals to trade, so eventually the market-

maker stops learning about investors' information pre ision. This independen e of the

de ision to trade from the private information about pre ision is a behavioral oarsening,

and auses this type of information to remain forever private.

Cipriani and Guarino (2001a) extend Glosten/Milgrom to a multiple se urity setting.

They allow for traders that have non-spe ulative motives for trading. In Cipriani and

Guarino, the trading of informed investors auses information to be partly re e ted in

pri e. As pri es be ome more informative, at some point one more of the onditionally

independent private signals auses a rather small update in expe ted fundamental value.

As a result, an investor who has a non-spe ulative reason to pur hase the se urity �nds

it pro�table to pur hase the se urity even if his private information signal is adverse.

In other words, he is in a as ade. Similarly, investors who have a non-spe ulative

motive to sell do so regardless of their signal. With all informed investors in a as ade,

20

further aggregation of information is ompletely blo ked. Thus, in ontrast to Avery and

Zemsky, informational as ades proper form. Furthermore, as ades lead to ontagion

a ross markets.

In Lee (1998) there are quasi- as ades that result in temporary information blo kage,

then avalan hes. This arises from transa tions osts and dis reteness in trades, whi h

lead to behavioral oarsening. In sequential trading, hidden information be omes a u-

mulated as the market rea hes a point at whi h, owing to transa tions osts, trading

temporarily eases. Eventually a large amount of private information an be revealed

by a small triggering event. The triggering event is a rare, low probability adverse sig-

nal realization. An individual who draws this signal value sells. Other individuals who

observe this sale are drawn into the market, ausing a market rash or `avalan he.'

These papers apply a sequential trading approa h. Beaudry and Gonzalez (2000)

apply a rational expe tations (simultaneous trading) modeling approa h to show that

as ading o urs when information is ostly to a quire, leading to pri e and investment

u tuations. Like these other papers, investment is a dis rete de ision.12

A key issue regarding the o urren e of information blo kage in these models is the

signi� an e of the assumption of dis rete a tions. Any model that attempts to explain

empiri al phenomena su h as market rashes as (quasi-) as ades must alibrate with

respe t to the size of minimum trade size or pri e movements. Su h onstraints are

most likely to be signi� ant for illiquid markets.13

Perhaps the more important role of as ades is likely to be in the de ision of whether

or not to parti ipate at all, rather than in the de ision of whether to buy or sell. If

there is a �xed ost (perhaps psy hi ) of parti ipating, then there an be a substantial

dis reteness to individual de isions that does not rely in any way upon limiting the size

of trades to a single unit. Or, if people are imperfe tly rational, so that there is some

sort of barrier to their parti ipating, again there an be as ades of parti ipation versus

non-parti ipation.

In the ontext of risk regulation, Kuran and Sunstein (1999) develop the notion

of availability as ades; their ideas are appli able to se urity market a tivity. If high

publi ity about a �rm or market theory makes the �rm more salient and `available'

12Chakrabarti and Roll (1997) o�er a simulation analysis of the e�e ts of investors learning by ob-serving the trades of others. They report that under some market onditions learning by observingothers redu es market volatility and in others in reases volatility.

13In a short run level, the expe tation that NYSE spe ialists will maintain an `orderly market' bykeeping pri es ontinuous an potentially for e temporary deviations of pri es from market values, blo kinformation ow. This suggests a relevan e of as ade only in extreme ir umstan es.

21

to investors. This may en ourage as ades of investment (Huberman (1999) provides

eviden e and insightful dis ussion about the e�e t of familiarity on investment). Lo al

biases in investment (see, e.g., Coval and Moskowitz (2001)), and the home bias puzzle of

international �nan e (see, e.g., Tesar and Werner (1995), Lewis (1999)) may be examples

of availability as ades. In any ase as ades in market parti ipation o�er a ri h avenue

for further analyti al exploration.

There is starting to be some exploration of the formation and learing of information

blo kages asso iated with the hoi e of individuals over time as to whether or not to

parti ipate in trading (Romer (1993), Lee (1998), Cao, Coval, and Hirshleifer (2001),

and Hong and Stein (2001)). In settings with limited parti ipation, large rashes an be

triggered by minimal information, and the sidelining and entry of investors an ause

skewness and volatility to vary onditional upon past pri e moves. (Bulow and Klem-

perer (1994) onsider a di�erent setting with asymmetri revelatory e�e ts of trading.)

4 Agen y/Reputation-Based Herding Models

In the seminal paper on reputation and herd behavior, S harfstein and Stein (1990)

onsider two managers fa e identi al binary investment hoi es. Managers may have

high or low ability, but neither they nor outside observers know whi h. Observers infer

the ability of managers from whether their investment hoi es are identi al or opposite,

and then update based upon observing investment payo�s. Managers are paid a ording

to observers' assessment of their abilities. It is assumed that high ability managers will

observe identi al signals about the investment proje t, whereas low ability managers

observe independent noise.

There is a herding equilibrium in whi h the �rst manager makes the hoi e that his

signal indi ates, whereas the se ond manager always imitates this a tion regardless of

his own signal. If the se ond manager were to follow his own signal, observers would

orre tly infer that his signal di�ered from the �rst manager, and as a result they would

infer that both managers are probably of low quality. In ontrast, if he takes the same

hoi e as the �rst manager, even if the out ome is poor, observers on lude that there

is a fairly good han e that both managers are high quality and that the bad out ome

o urred by han e. Thus, their model aptures the insight of John Maynard Keynes

that \it is better to fail onventionally than to su eed un onventionally."

Rajan (1994) onsiders the in entive for banks with private information about bor-

rowers to manage earnings upward by relaxing their redit standards for loans, and by

22

refraining from setting aside loan-loss reserves. When there is a bad aggregate state

of the world, even the loans of high ability managers do poorly. Thus, observers do

not `punish' a banker reputationally as mu h for setting aside loan-loss reserves if other

banks are doing so as well. Thus, the set-aside of reserves by one bank triggers set-

asides by other banks. This simultaneity in the a tions of banks is somewhat analogous

to the delay and sudden onset of informational as ades in the models Zhang (1997)

and Chamley and Gale (1994). Furthermore, Rajan shows that banks tighten redit in

response to de lines in the quality of the borrower pool. Thus banks amplify sho ks to

fundamentals. Rajan provides eviden e from New England banks in the 1990s of su h

delay in in reasing loan loss reserves, followed by sudden simultaneous a tion.

Trueman (1994) onsiders the reputational in entives for sto k market analysts to

herd in their fore asts of future earnings. We over this paper in the next se tion.

One of his �ndings is that analysts have an in entive to make fore asts biased toward

the market's prior expe tation. In a similar spirit, Brandenburger and Polak (1996)

show that a �rm with superior information an have a reputational in entive to make

investment de isions onsistent with the prior belief that observers have about whi h

proje t hoi e is more pro�table. Intuitively, even if the prior-disfavored proje t hoi e

is the more pro�table of the two alternatives and even if observers assume that the

manager will make the pro�t-maximizing hoi e, the market may still be disappointed

that the prior-favored hoi e was not the more pro�table of the alternatives. This

an o ur, for example, if the likely driver of sele tion of the prior-disfavored hoi e

is disappointing information about the prior-favored alternative. Where these papers

fo us on pleasing investors, Prendergast (1993) examines the in entives for subordinate

managers to make re ommendations onsistent with the prior beliefs of their superiors.

Where in S harfstein and Stein it is better to fail as part of the herd than to su eed

as a deviant, Zwiebel (1995) des ribes a s enario in whi h it is always best to su eed,

but where the fa t that a manager's su ess is measured relative to others sometimes

auses herding. The �rst premise of the model is that there are ommon omponents of

un ertainty about managerial ability. As a result, observers exploit relative performan e

of managers to draw inferen es about di�eren es in ability. The se ond premise is that

managers are averse to the risk of being exposed as having low ability (perhaps be ause

the risk of �ring is nonlinear). For a manager who follows the standard behavior, the

industry ben hmark an quite a urately �lter out the ommon un ertainty. This makes

following the industry ben hmark more attra tive for a fairly good manager than a poor

one, even if the innovative proje t sto hasti ally dominates the standard proje t. The

23

alternative of hoosing a deviant or innovative proje t is highly risky in the sense that

it reates a possibility that the manager will do very poorly relative to the ben hmark.

Thus, the model o�ers an alternative explanation for orporate onservatism to the

herd-free reputational models of Hirshleifer and Thakor (1992) and Prendergast and

Stole (1996), and the memory-loss approa h of Hirshleifer and Wel h (2002).

However, in Zwiebel's model a very good manager an be highly on�dent of beating

the industry ben hmark even if he hooses a risky, innovative proje t. If this proje t is

superior, it pays for him to deviate. Thus, intermediate quality managers herd, whereas

very good or very poor managers deviate. Zwiebel's approa h is suggestive that under

some ir umstan es portfolio managers may herd by redu ing the risk of their portfolios

relative to a sto k market or other index ben hmark, but under others may intentionally

deviate from the ben hmark. Several papers pursue these and related issues su h as

optimal ontra ting in detail (see, e.g., Maug and Naik (1996), Gumbel (1998), Huddart

(1999), and Hvide (2001)). S iubba (2001) provides a model of herding by portfolio

managers in relation to past performan e. Brennan (1993) analyzes the asset pri ing

impli ations of su h index-herding behavior.

In some models a prin ipal designs institutions and/or ompensation s hemes in the

fa e of managerial in entives to engage in informational as ades or making hoi es

to mat h an observer's priors (Prendergast (1993) [dis ussed above℄, Khanna (1997),

Khanna and Slezak (2000)). Khanna (1997) examines the optimal ompensation s heme

when managers have in entives to as ade in their investment de isions. He examines

a setting in whi h the managers of ompetitor �rms an investigate to generate private

signals. A manager may delay investigation in the expe tation of gleaning information

more heaply by observing the behavior of the ompetitor. A manager may also observe

a signal but as ade upon the a tion of an earlier manager. Khanna des ribes opti-

mal ontra ts that address the in entives to investigate and to as ade, and develops

impli ations for ompensation and investments a ross di�erent industries.14

Khanna and Slezak (2000) provide an intra-�rm model in whi h the tenden y for

as ades to start among managers redu es the quality of proje t re ommendations and

hoi es. This is a disadvantage of `team de isions,' in whi h managers make de isions

sequentially and observe ea h others' re ommendations. In entive ontra ts that elimi-

nate as ades may be too ostly to be desirable for the shareholders. A hub-and-spokes

hierar hi al stru ture where managers independently report re ommendations to a su-

14See also Grant, King, and Polak (1996) for a review of informational externalities in a orporate ontext when there are share pri e in entives.

24

perior eliminates as ades, but requires superiors to in ur osts of monitor subordinates

to ensure that subordinates do not ommuni ate. Thus, under di�erent onditions the

optimal organizational form an be either teams or hierar hy.

5 Herd Behavior and Cas ades in the Analysis of

Se urities

5.1 Herd Behavior in Investigation and Trading

In an informational as ades setting where individuals have to pay a ost to obtain their

private signals, on e a as ades starts individuals have no reason to investigate. In se u-

rity market settings, the assumption that the aggregate varian e of noise trading is large

enough to in uen e pri es non-negligibly (as in the seminal paper of DeLong, Shleifer,

Summers, and Waldmann (1990) and subsequent models of exogenous noise) impli itly

re e ts an assumption that individuals are irrationally orrelated in their trades. This

ould be a result of herding (whi h involves intera tion between the individuals), or

merely a ommon irrational in uen e of some noisy variable on individuals' trades.15

The analysis of Brennan (1990) was seminal in illustrating the possibility of herd

behavior in the analysis of se urities. He provided an overlapping generations model in

whi h private information about a se urity is not ne essarily re e ted in market pri e the

next period. This o urs in a given period only if a pre-spe i�ed number of individuals

had a quired the signal. Thus, the bene�t to an investor of a quiring information about

an asset an be low if no other investor a quires the information. However, if a group of

investors oordinate to a quire information than the investors who obtain information

�rst do well. Sin e the setting is spe ial it has stimulated further work to see if herding

an o ur in settings with greater resemblan e to standard models of se urity trading

and pri e determination.

Froot, S harfstein, and Stein (1992) o�er a model that endogenizes pri e determina-

tion more fully. In their setting, investors with exogenous short horizons �nd it pro�table

to herd by investigating the same sto k. In so doing they are, indire tly able to e�e t

what amounts to a ta it manipulation strategy. When they buy together the pri e is

driven up, and then they sell together at the high pri e. Thus, herding even on `noise'

(a spurious uninformative signal) is pro�table.

15Gole (1997) provides a possible example of su h a ommon irrational in uen e. He alls this`herding on noise,' one of our two possible interpretations.

25

However, even in the absen e of opportunities for herding there is a potential in-

entive for individuals, a ting on their own, to e�e t su h manipulation strategies. If

individuals are allowed to trade to `arbitrage' su h manipulation opportunities, it is not

lear that su h opportunities an in equilibrium persist. This raises the question of

whether there are in entives for herding per se rather than for herding as an indire t

means of manipulation.16

Hirshleifer, Subrahmanyam, and Titman (1994) examine the se urity analysis and

trading de isions of risk averse individuals, where investigation of a se urity leads some

individuals to re eive information before others. They �nd a tenden y toward herding.

The presen e of investigators who re eive information late onfers an obvious bene�t

upon those who re eive information early- the late informed drive the pri e in a dire tion

favorable to the early-informed. But by the same token, the early-informed push the

pri e in a dire tion unfavorable to the late-informed. The key to the model's herding

result is that the presen e of the late-informed allows the early-informed to unwind their

positions sooner. This allows the early-informed to redu e the extraneous risk they would

have to bear if, in order to pro�t on their information, they had to hold their positions

for longer. This risk-redu tion that the late-informed onfer upon the early informed

is a genuine ex ante net bene�t- it is not purely at the expense of the late informed.

Over on�den e about the ability to be ome informed early further en ourages herding

in this model; ea h investor expe ts to ome out the winner in the ompetition to study

the `hot' sto ks.

Holden and Subrahmanyam (1996) show that there an also be herding in the hoi e

of whether to study short-term or long-term information about the sto k. Intuitively,

exploiting long-term information again involves the bearing of more extraneous risk,

whi h an be ostly.

5.2 Herd Behavior by Sto k Analysts and other Fore asters

Several studies of fore asters have reported herding or herding-like �ndings. Ashiya and

Doi (2001) report that Japanese ma ro-e onomi fore asters herd in their fore asts, re-

gardless of their age. Ehrbe k and Waldmann (1996) �nd, onsistent with psy hologi al

bias rather than rational reputational-oriented bias, that e onomi fore asters bias their

fore asts in dire tions hara teristi of high mean-squared-error fore asters. However,

16Another interesting question is whether short horizons an be derived endogenously. Dow andGorton (1994) �nd that owing to the risk of trading on long-term information, pri es will not fullyre e t private information.

26

the analyti al literature on sto k market analysts has fo used on rational reputational

reasons for bias.

Analyst earnings fore asts are biased, as do umented by Givoly and Lakonishok

(1984), Brown, Foster, and Noreen (1985), and many more re ent authors. Fore asts

are generally optimisti in the U.S. and other ountries, espe ially at horizons longer

than one year (see e.g. Capsta�, Paudyal, and Rees (1998) and Brown (2001)). More

re ent eviden e indi ates that analysts' fore asts have be ome pessimisti at horizons of

3 months or less before the earnings announ ement (Brown (2001), Matsumoto (2001)

and Ri hardson, Teoh, and Wyso ki (2001)).

Sti kel (1992) �nds that the ompensation re eived by analysts is related to its rank-

ing in a poll by Institutional Investor about the best analysts. Furthermore, fore asts

by members of Institutional Investor's of `All-Ameri an Resear h Team' were more a -

urate than those of non- members. These �ndings suggests that analysts may have an

in entive to adjust their fore asts to maintain good reputations for high a ura y.

Mikhail, Walther, and Willis (1999) �nd that analysts whose fore asts are less a -

urate than peers are more likely to turn over. This importan e of relative evaluation

supports the premise of reputational models of herding. However, they �nd no relation

between either absolute or relative pro�tability of an analyst's re ommendations and

probability of turnover. Hong, Kubik, and Solomon (2000) �nd eviden e suggesting

that there are reputational in entives for analyst herding. Less experien ed analysts are

more likely to be terminated for `bold' fore asts that deviate from the onsensus fore ast

than are experien ed ones, suggesting that the pressure to build reputation is strongest

for analysts for whi h un ertainty about ability is greatest.

Trueman (1994) provide a model in whi h analysts tend to issue fore asts that are

biased toward prior earnings expe tations, and also herd in the sense that fore asts are

biased toward those announ ed by previous analysts. In his analysis, an analyst has a

greater tenden y to herd if he is less skillful at predi ting earnings-it is less ostly to

sa ri� e a poor signal than a good one.

Sti kel (1990) �nds that hanges in onsensus analyst fore asts are positively related

to subsequent revisions in analyst's fore asts, apparently onsistent with herd behavior.

This relationship is weaker for the high-pre ision analysts who are members of the `team'

than for analysts who are not. Thus, it appears that members of the `team' are less prone

to herding than non-members. This is onsistent with the predi tion of the Trueman

model.

Experimental eviden e involving experien ed professional sto k analysts has also

27

supported the model (Cote and Sanders (1997)). Cote and Sanders report that these

fore asters exhibited herding behavior. Furthermore, the amount of herding was related

to the fore asters' per eption of their own abilities and their motivation to preserve or

reate their reputations.

In ontrast, Zitzewitz (2001) provides a methodology for estimating the degree of

herding versus exaggeration of di�eren es (the opposite of herding) by analysts. He

reports that in fa t analysts on average exaggerate their di�eren es. He also �nds that

analysts under-update their fore asts in response to publi information, indi ating an

overweighting of prior private information. This eviden e opposes the on lusion that

analysts on the whole herd. It is potentially supportive of reputational models in whi h

some individuals intentionally diverge (e.g., Prendergast and Stole (1996)), or with over-

on�den e on the part of analysts in their private signals.

It is also often alleged that analysts herd in their hoi e of what sto ks to follow.

There is very high variation in analyst overage of di�erent �rms Bhushan (1989). In

his sample, the average number of analysts following a �rm was approximately 14, but

a number of �rms were followed by only 1 analyst; the maximum number of analysts

was 77. This is not in onsistent with herding by analysts in their overage de isions,

and indire tly by the investors that listen to them. But in the absen e of any �rst-best

ben hmark for the dispersion of analyst following a ross �rms, it is hard to draw any

on lusion on this issue

There are also allegations that analysts herd in their sto k re ommendations. This

issue is studied by Wel h (2000), who �nds that revisions in the buy and sell sto k

re ommendations of a se urity analyst are positively related to revisions in the buy and

sell re ommendations of the next two analysts. He tra es this in uen e to short-term

information, identi�ed by examination of the ability of the revision to predi t subsequent

returns.17

Wel h also �nds that analysts' hoi es are orrelated with the prevailing onsensus

fore ast. Wel h further �nds that the `in uen e' of the onsensus on later analysts is not

stronger when it is a better predi tor of subsequent sto k returns. In other words, the

eviden e is onsistent with analysts herding even upon onsensus fore asts that aggregate

information poorly. This is onsistent with agen y e�e ts su h as reputational herding,

or ould re e t imperfe t rationality on the part of analysts. Finally, Wel h �nds an

asymmetry, that the tenden y to herd is stronger when re ent returns have been positive

17This ould re e t as ading, or ould be a lustering e�e t wherein the analysts ommonly respondto a ommon, independently observed signal.

28

(`good times') and when the onsensus is optimisti . He spe ulates that this ould lead

to greater fragility during sto k market booms, and the o urren e of rashes.

The eviden e on the re ommendations of investment newsletters on herding is mixed.

Ja�e and Mahoney (1999) report only weak eviden e of herding by newsletters in their

re ommendations over 1980-96. However, Graham (1999) develops and tests an expli it

reputation-based model of the re ommendations of investment news letters, in the spirit

of S harfstein and Stein (1990). He �nds that analysts with better private information

are less likely to herd on the market leader, Value Line investment survey. This �nding is

onsistent with the models of S harfstein and Stein (1990) and Bikh handani, Hirshleifer,

and Wel h (1992).

6 Herd Behavior and Cas ades in Se urity Trading

Some so iologists have emphasized that the `weak ties' of liaison individuals, who onne t

partly-separated so ial networks, are important for spreading behaviors a ross networks

(Granovetter (1973). A re ent literature in e onomi s has examined the strength of

peer-group e�e ts in a number of di�erent ontexts (see, e.g., Weinberg, Reagan, and

Yankow (2000), and the survey of Glaeser and S heinkman (2000)). In a apital mar-

kets ontext, Shiller and Pound (1989) �nd based on questionnaire/survey eviden e that

word-of-mouth ommuni ations are reported to be important for the trading de isions

of both individual and institutional investors. Two re ent studies report that employees

are in uen ed by the hoi es of oworkers in their de isions of whether to parti ipate

in di�erent employer-sponsored retirement plans ((Du o and Saez 2000), Madrian and

Shea (2000)). Kelly and O'Grada (2000) and Hong, Kubik, and Stein (2001) provide

further eviden e that so ial intera tions between individuals a�e ts de isions about eq-

uity parti ipation and other �nan ial de isions. A theoreti al analysis of learning from

neighbors is provided by Bala and Goyal (1998).

6.1 The Endorsement E�e t

A ording to informational as ades theory, endorsements an be extremely in uential

if the endorser has a reputation for a ura y, and if the endorsement involves an a tual

informative a tion by the expert. This ould take the form of knowing that the expert

took a similar a tion (buying a sto k), but ould also involve the expert investing his

reputation in the sto k by re ommending it.

29

The hoi e by a big-�ve auditor, top-rank investment bank, or venture apital to

invest its reputation in ertifying a �rm in uen es investor favorably toward the �rm.18

Furthermore, just as shopping mall developers use `an hor' stores to attra t other stores,

a ording to M Gee (1997) some IPO underwriters have been using the names of well-

known investors as `an hors' to attra t other investors.19

There are many examples of in uential investors, some more benign than others. In a

story entitled \Pied Piper of Biote h Keeps Followers Happy with Cut-Rate Sto k," the

Wall Street Journal, 5/7/92 says \Wherever David Bel h invests his money , a rowd

of sto kbrokers and money managers is sure to follow. `David Ble h is the single most

important for e in the biote h industry,' says Ri hard Bo k, a sto kbroker... I follow

whatever sto k he goes into, knowing it will be a su ess.' "

Some investors are in uen ed in old- alls by brokers by statements that famous

investors are holding a sto k (see Lohse (1998) on \Tri ks of the Trade: `Bu�ett is Buying

This' and other Sayings of the Cold-Call Crew"). (Sin e Bu�ett is typi ally a passive

investor, his in uen e re e ts per eptions that he is well informed rather than that he

will reorganize the �rm.) One investment digest expli itly gave as its key reasoning for

spotlighting a sto k the fa t that Bu�ett was involved in it (Davis (1991)).

When news ame out that Warren Bu�ett had bought approximately 20% of the 1997

world silver output, a ording to The E onomist (1998) silver pri es were sent \soaring."

When Warren Bu�ett's �lings reporting his in reased shareholding in Ameri an Express

and in PNC Bank be ame publi , these shares rose by 4.3% and 3.6% respe tively

(Obrien and Murray (1995)).

A ording to Sandler and Raghavan (1996), \Whether Warren Bu�ett has been

right or wrong about a sto k, investors don't like to see him get out if they're still in.

Some investors in Saloman are fo using almost entirely on the famed Omaha, Neb.,

18See the models of Titman and Trueman (1986), and Datar, Feltham, and Hughes (1991), and theeviden e of Beatty and Ritter (1986), Booth and Smith (1986), Johnson and Miller (1988), Beatty(1989), Carter and Manaster (1990), Feltham, Hughes, and Simuni (1991), Simuni (1991), Megginsonand Weiss (1991), Mi haely and Shaw (1995), and Carter, Dark, and Singh (1998). A salient re entexample of this erti� ation e�e t is the drop of 36% in the shares of Emex when First Boston deniedEmex's laim that it was their investment banker (Remond and Hennessey (2001)).

19\As any fashion house knows, stit hing a designer label on a pair of jeans allows it to harge two orthree times the going rate for pants. Now, battling to set themselves apart from the rowd, and enti emore investors to their initial publi o�erings of sto k, edgling te hnology ompanies with unprovenprodu ts and no earnings are bragging of their ties to sto k-market winners like Mi rosoft Corp., Cis oSystems In . or Ameri an Online In . Never mind that some of these an hor investors don't appearto be pi ky; they invest in bun hes of smaller ompanies be ause they know that not every investmentwill pan out. The fa t is, the hype works..." The arti le gives several examples in whi h te h sto kanalysts and investors may have been in uen ed by the a het of an hor investors.

30

multibillionaire's de ision, announ ed Sept. 12, ..." to onvert Salomon preferred shares

into ommon shares instead of taking ash.

Investing human apital is also form of endorsement; for example, when it was an-

noun ed that John S ully was signing on as hairman and CEO of the little known

�rm Spe trum Information Te hnologies In ., its sto k jumped by lose to 46%.20 The

in uen e of sto k market `gurus' is a sort of endorsement, but in some ases investors

seem irrationally in uen ed by well-known but in ompetent analysts. This may involve

a limited attention/availability e�e t wherein investors use an analyst's visibility fame

as an indi ator of ability. A would-be guru an exploit the aws of this heuristi by using

even outlandish publi ity stunts to gain notoriety; see, e.g., the des ription of Joseph

Granville's areer in Shiller (2000b).

Sto k pri es rea t to the news of the trades of insiders; see, e.g., Givoly and Palmaon

(1985). It seems lear that these trades provide information to market parti ipants,

who adjust their own trading (as a fun tion of pri e) a ordingly. Su h in uen e on the

part of insiders potentially gives them the power to manipulate pri es, as re e ted in the

analysis of Fishman and Hagerty (1995); see Fried (1998) for a dis ussion of the ` opy at

theory' that insiders exploit imitators by trading in the absen e of private information.

Investors are also in uen ed by private onversations with peers. For example, Fung

and Hsieh (1999) state that \a great deal of hedge fund investment de isions are still

based on \re ommendations from a reliable sour e.' " There is also eviden e that in-

vestors are in uen ed by impli it endorsements, as with default settings for ontributions

in 401(k) plans; see Madrian and Shea (2000).

6.2 A Challenge in Measuring Herding

An important hallenge to empiri al work on herding is to rule out lustering. Some

external fa tor ould be independently in uen ing di�erent investors' trades in parallel,

even if there were no intera tion between the trades of the di�erent investors in the

alleged herd. In general it is hard to rule out lustering on lusively, though a few

studies are able to do so in spe i� ontexts. One method of addressing this is to

in lude proxies for possible variables that may jointly a�e t the behavior of di�erent

individuals (for a general analysis of e onometri issues in measuring so ial intera tion,

see, e.g., Bro k and Durlauf (2000)). Of ourse, no matter how thorough the study, it

20Wall Street Journal, 10/14/93, \S ulley Be omes Chief of Spe trum, Pla ing Bet on Wireless Te h-nology", John J. Keller)." A later Business Week investigation suggested that the CEO of Spe trumwas \a manipulator who duped John S ulley and milked the ompany" (S hroeder (1994)).

31

is always on eivable that some joint ausal fa tor has been omitted.

Some studies go further to examine natural or arti� ial experiments whi h rule out

the possibility of an omitted in uen e. Sa erdote (2001) provides eviden e of peer e�e ts

in a study of roommate hoi es with random assignments, so avoids this. Also, a growing

literature starting with Anderson and Holt (1996) has on�rmed learning by observing

a tions, and the existen e of informational as ades in the experimental laboratory (see

also Hung and Plott (2001), Anderson (2001), Sgroi (2000) and Celen and Kariv (2001)).

Consistent with as ades, Dugatkin and Godin (1992) �nd experimentally that female

guppies tend to reverse their mate hoi es when they observe other females hoosing

di�erent males.

The simultaneous ausation issue is present in most herding tests, but be omes more

tri ky in �nan ial market tests be ause of the in uen e of pri e. It is possible for

individuals to herd in a onditional fashion, dependent upon past pri e movements.

However, even if we rule out all non-pri e joint ausal e�e ts, orrelation in trades

onditional upon pri e movements is not ne essarily herding. For example, suppose

that ertain mutual funds have orrelated trades that are asso iated with past pri e

movements. This ould indi ate herding. On the other hand, it ould be that some

other group of investors su h as individual investors is herding, and that the mutual

funds are not. The mutual funds may merely be adjusting their trades in response

to pri e movements. In the extreme, if there are only two groups of traders, then by

market learing, herding by one group of traders automati ally implies orrelation in the

trades of the other group, even though there may be no intera tion whatsoever between

members of this other group.

Alternatively, it ould be that some group of investors is jointly in uen ed by some

unobserved in uen e, and again that the mutual funds are jointly responding to pri e.

On e again, the orrelation in the trades of the mutual funds does not imply herding.

Thus, to verify that a group is truly herding, it is ru ial either to ontrol for pri e, or if

not, to verify whether the ausality of the behavioral onvergen e is really oming from

the group in question or from other traders.

6.3 Eviden e Regarding Herding in Trades

Several papers on institutional investors trading have developed alternative measures

of trading; see, e.g., Lakonishok, Shleifer, and Vishny (1992), Grinblatt, Titman, and

Wermers (1995), Wermers (1999). Bikh handani and Sharma (2001) riti ally review

32

alternative empiri al measures of herding.

GriÆths et al (1998) �nd in reased similarity of behavior in su essive trades for

se urities that are traded in an open out ry market rather than a system trading market)

on the Toronto sto k ex hange, onsistent with the possibility of imitation-trading raised

by the eviden e of Biais, Hillion, and Spatt (1995). Grinblatt and Keloharju (2000))

provide eviden e onsistent with herding by individuals and institutions.

Institutional investors onstitute a large fra tion of all investors. By market- learing

it is impossible for all investors to be buyers or sellers. Although testing for herding by

su h a large group is not unreasonable, it ertainly makes sense in addition to examine

�ner subdivisions of investors. In older studies, Friend, Blume, and Cro kett (1970)

found, during a quarter in 1968, a tenden y for mutual funds to follow the investment

de isions made in the previous quarter by su essful funds. Kraus and Stoll (1972)

found that in a sample of mutual funds and bank trusts from 1968-9 attribute the

large trade imbalan es they �nd in sto ks to han e rather than orrelated trading.

Klemkosky (1977) found that in 1963-72 that sto ks bought by investment ompanies

(mainly mutual funds) subsequently do well.

Using quarterly data on the portfolios of pension funds from 1985-89, Lakonishok,

Shleifer, and Vishny (1992) �nd relatively weak eviden e that pension funds engage

in either positive feedba k trading or herding, with a stronger e�e t in smaller sto ks.

Grinblatt, Titman, and Wermers (1995) �nd that most sto k mutual funds pur hased

past winners during 1974-84. They �nd a tenden y for funds to buy and sell sto ks

at the same time in sto ks in whi h a large number of funds are a tive. Herding was

strongest among aggressive growth, growth and in ome funds. Wermers (1999) �nds

that during 1975-94 there was little herding by mutual funds in the average sto k, but

that there was herding in small sto ks and in sto ks that experien ed high returns.

Growth-oriented mutual funds tended to herd in their trades. He also found superior

performan e among the sto ks that herds buy relative to those they sell during the six

months subsequent to trades, espe ially among small sto ks. Nofsinger and Sias (1999)

report that hanges in institutional ownership are asso iated with high ontemporaneous

sto k and returns, that institutions tend to buy after positive momentum, and that the

sto ks institutions buy outperform those that they sell. On a shorter time s ale, Kodres

and Pritsker (1997) report herding in daily trading by large futures market institutional

traders su h as broker-dealers, banks, and hedge funds, although measurement issues

reate signi� ant hallenges

Brown, Harlow, and Starks (1996) and Chevalier and Ellison (1997) �nd that fund

33

managers that are doing well lo k in their gains toward end of the year by indexing the

market, whereas funds that are doing poorly deviate from the ben hmark in order to try

to overtake it. Chevalier and Ellison (1999) indentify possible ompensation in entives

for younger managers to herd by investing in popular se tors, and �nd empiri ally that

younger managers hoose portfolios that are more ` onventional' and whi h have lower

non-systemati risk.

6.4 Creditor Runs, Bank Runs, and Finan ial Contagion

An older literature argued that bank runs are due to `mob psy hology' or `mass hysteria'

(see the referen es dis ussed in Gorton (1988)). At some point e onomists may revisit

the role of emotions in ausing bank runs or `pani s,' and more generally ausing multiple

reditors to refuse to �nan e distressed �rms. Su h an analysis will require attending to

eviden e from psy hology about how emotions a�e t judgments and behavior

At this point the main models of bank runs and of �nan ial distress are based upon

full rationality (for reviews of models and eviden e about bank runs, see, e.g., Calomiris

and Gorton (1991) and Bhatta harya and Thakor (1993) se tion 5.2) . There is a

negative payo� externality in whi h withdrawal by one depositor, or the refusal of a

reditor to renegotiate a loan, redu es the expe ted payo�s of others. This an lead

to multiple equilibria involving runs on the bank or �rm, or to bank runs triggered by

random sho ks to withdrawals (see, e.g., Diamond and Dybvig (1983)). This of ourse

does not pre lude the possibility that there is also an informational externality.

The informational hypothesis (e.g., Gorton (1985)) holds that bank runs result from

information that depositors re eive about the ondition of banks' assets. When a dis-

tressed �rm seeks to renegotiate its debt, the refusal of one reditor may make others

more skepti al. Similarly, if some bank depositors withdraw their funds from a troubled

bank, others may infer that those who withdrew had adverse information about the value

of the bank's illiquid assets, leading to a bank run (see, e.g., Chari and Jagannathan

(1988), Ja klin and Bhatta harya (1988)).

Bank runs an be modeled as informational as ades, sin e the de ision to withdraw

is bounded (the individual annot withdraw more than 100% of his deposit). There is a

payo� as well as an informational intera tion: early withdrawals hurt loyal depositors,

and more generally refusal of a reditor to renegotiate hurts other reditors. However,

at the very start of the run, when only a few reditors have withdrawn, the main e�e t

may be the informational onveyed by the withdrawals rather than the redu tion in the

34

bank's liquidity.

If assets are imperfe tly orrelated, as ades an pass ontagiously between banks and

ause mistaken runs even in banks that ould have remained sound; (on information and

ontagion, see Gorton (1988), Chen (1999), and Allen and Gale (2000b)). This suggests

that the arrival of adverse publi information an trigger runs (see, e.g., Calomiris and

Gorton (1991))

There is eviden e of geographi al ontagion between bank failures or loan-loss reserve

announ ements and the returns on other banks (see Aharony and Swary (1996) and

Do king, Hirs hey, and Jones (1997)). This suggests that bank runs are triggered by

information rather than being a purely non-informational (multiple equilibria, or e�e ts

of random withdrawal) phenomenon.21 Saunders and Wilson (1996) provide eviden e of

ontagion e�e ts in a sample of U.S. bank failures during the period 1930-32. On the

other hand Calomiris and Mason (1997) �nd that the failure of banks during the Chi ago

pani of June 1932 was due to ommon sho ks, and Calomiris and Mason (2001) �nd

that banking problems during the great depression an be explained based upon either

bank-spe i� variables or publi ly observable national and regional variables rather than

ontagion.

6.5 Exploiting Herding and Cas ades

Firms often market experien e goods by o�ering low introdu tory pri es. In as ades

theory, the low pri e indu es early adoptions, whi h helps start a positive as ade. Wel h

(1992) developed this idea to explain why initial publi o�erings of equity are on average

severely underpri ed by issuing �rms.22 Neeman and Orosel (1999) provide a model of

au tions in a winner's urse setting in whi h a seller (su h as a �rm selling assets) an

gain from approa hing potential buyers sequentially, indu ing informational as ades,

rather than ondu ting an English au tion.

21There is also eviden e of ontagion in spe ulative atta ks on national urren ies (Ei hengreen, Rose,and Wyplosz (1996)).

22An example is provided by the des ription of the Mi rosoft IPO in Fortune (1986) (p. 32): \Eri Dobkin, 43, the partner in harge of ommon sto k o�erings at Goldman Sa hs, felt queasy aboutMi rosoft's ounterproposal. For an hour he tussled with Gaudette, using every argument he ouldmuster. Coming out $1 too high would drive o� some high-quality investors. Just a few signi� antdefe tions ould lead other investors to think the o�ering was losing its luster." This illustrates the useof pri e to indu e as ades, and the result of the as ades model that individuals with high informationpre ision are parti ularly e�e tive at triggering early as ades.

35

7 Herding, Bubbles, and Crashes: The Pri e Impli-

ations of Herding and Cas ading

Popular allegations of se urities market irrationality often emphasize the ontagiousness

of emotions su h as pani or frenzy. Criti s often go on to argue that this auses ex ess

volatility, destabilizes markets, and makes �nan ial system fragile (see, e.g., the riti al

review of Bikh handani and Sharma (2001) and referen es therein). There is indeed

eviden e that emotions are ontagious and that this ontagion a�e ts per eptions and

behavior (see, e.g., Hat�eld, Ca ioppo, and Rapson (1993), Barsade (2001)). In the

lassi fully rational models of se urities market pri e formation, information is onveyed

through pri es or pri ing fun tions that are observable to all, so there is no room for

lo alization in the ontagion pro ess (Grossman and Stiglitz (1980), Kyle (1985)). Even

re ent models of herding and of informational as ades in se urities markets involve

ontagion based upon observation of either market pri es or trades, again leaving little

room for lo alization.

On the other hand, the eviden e dis ussed in Se tion 6 suggests that so ial inter-

a tions between individuals a�e ts �nan ial de isions. This suggests that the so ial or

geographi al lo alization of information may be an important part of the pro ess by

whi h trading behaviors spread. Furthermore, some so iologists and e onomists argue

that there are threshold e�e ts in so ial pro esses, where the adoption of a belief or

behavior by a riti al number of individuals leads to a tipping in favor of one behavior

versus another (Granovetter (1978), S helling (1978), Kuran (1989, 1998)).

Thus, an important dire tion for further empiri al resear h is to examine how whether

a lo alized pro ess of ontagion of beliefs and attitudes a�e ts sto k markets (see, e.g.,

Shiller (2000a)), and whether se urities market pri e patterns are onsistent with rational

models of ontagion. An important theoreti al dire tion is to examine the impli ations

for se urities market trading and pri es of onversation between individuals; see the

analysis of DeMarzo, Vayanos, and Zwiebel (2000), and the on luding dis ussion of

Cao, Coval, and Hirshleifer (2001).

If herding is driven by agen y onsiderations, one would expe t any pri e e�e ts of

herding to be driven by institutional investors. Sias and Starks (1997) provide eviden e

that institutional investors are a sour e of positive portfolio return serial orrelations

(both own-and ross orrelations of the se urities held by institutions). Aitken (1998)

�nds that the auto orrelation of the returns of emerging sto k markets in reased sharply

at the time that institutional investors were expanding their positions in emerging mar-

36

kets. He argues that this indi ates that this re e ted the e�e t of u tuating sentiment

by institutional investors.23

There is a large and growing literature on ontagion between the debt or equity

markets of di�erent nations (see, e.g., Bikh handani and Sharma (2001)). Borensztein

and Gelos (2001) report moderate herding in the trades of emerging market mutual

funds during 1996-9, but was not stronger during rises than normal times. With regard

to pri e e�e ts of herding, there are some large orrelations in returns, but it is hard

to measure whether this is an e�e t of herding, and there is only mixed eviden e as to

whether orrelations are higher during �nan ial rises. Choe, Kho, and Stulz (1999) pro-

vide strong eviden e of herding by foreign investors before the 1996-7 period of e onomi

risis for Korea, but herding was a tually lower during the risis period. Furthermore,

they do not �nd any indi ation that trades by foreign investors had a destabilizing e�e t

on Korea's sto k market. Many studies have examined how the o urren e of a risis

in one ountry a�e ts the probability of risis in another ountry; see, e.g., Berg and

Pattillo (1999) for a review of this resear h.

Experimental asset markets have been found to be apable of aggregating a great

deal of the private information of parti ipants; however, in omplex environments the

literature has shown that blo kages form so that imperfe t information aggregation is

imperfe t (see, e.g., Noeth et al (2002), Bloom�eld (1996), and the surveys of Libby,

Bloom�eld, and Nelson (2001), Sunder (1995)). Experimental laboratory resear h pro-

vides a very promising dire tion for exploring the relationship of herding to market

rashes (see, e.g., Cipriani and Guarino (2001b)). These should provide the raw mate-

rial for new theorizing on this topi .

Gompers and Lerner (2000) provide eviden e of `money hasing deals' in venture

apital. In ows into venture apital funds are asso iated with higher valuations of the

new investments made by these funds, but not with the ultimate su ess of the �rms.

Thus, it seems that orrelated enthusiasm of investors for ertain kinds of investors

moves pri es for non-fundamental reasons. However, Froot, O'Connell, and Seasholes

(2001) �nd that portfolio ows in and out of 44 ountries during 1994-98 were positive

23Christie and Huang (1995) are unable to dete t `herd behavior,' in the sense of high ross-se tionalstandard deviations of se urity returns at the time of large pri e movements. Rather than measuringherd behavior (so ial in uen e) per se, this is an indire t measure of the tenden y for some group ofinvestors to rea t in a ommon way more at the time of extreme sho ks than at other times. However, itis not obvious what the fundamental ben hmark should be for the asso iation between large sho ks andidiosyn rati variability; see also Chang, Cheng, and Khorana (2000), who report that in the U.S. andseveral asian markets, there is relatively little eviden e of herding ex ept for the two emerging marketsin the sample; and Ri hards (1999).

37

fore asters of future equity returns, with statisti al signi� an e in emerging markets.

8 Herd Behavior and Cas ades in Firms' Investment,

Finan ing, and Reporting De isions

It is often alleged in the popular press that managers are foolishly prone to fads in

management methods (for examples and formal analysis see Strang and Ma y (2001))

investment hoi es, and reporting methods.

Managers learn by observing the a tions and performan e of other managers, both

within and a ross �rms. This suggests that �rms will engage in herding and be subje t

to informational as ades, leading to management fads in a ounting, �nan ing and in-

vestment de isions. The popularity of di�erent investment valuation methods, se urities

to issue, and so on have ertainly waxed and waned. There are booms and quiet periods

in new issues of equity that are related to past sto k market returns and to the past

average initial returns from buying an IPO (see, e.g., Ibbotson, Ritter, and Sindelar

(1994), E kbo and Masulis (1995) and Lowry and S hwert (2002)). However, it is not

easy to prove that u tuations in investments and strategies result from irrationality,

rational but imperfe t aggregation of private information signals, or dire t responses to

u tuations in publi observables.

Takeover markets have been subje t to seemingly idiosyn rati booms and rashes,

su h as the wave of onglomerate mergers in the 1960's and 70's, in whi h �rms diversi-

�ed a ross di�erent industries, the subsequent refo using of �rms through restru turing

and bustup takeovers in the 1980's, followed by the merger boom of the 1990s. Pur-

hase of another �rm: targets of a takeover bid are `put into play,' and often qui kly

re eive ompeting o�ers, despite the negative ost externality of having a ompetitor.

Hauns hild (1993) provides interesting eviden e about apparent informational ontagion

of the de ision to engage in a takeover. In her 1981-90 sample, a �rm was more likely

to merge if one of its top managers was a dire tor of another �rm that had engaged in

a merger during the pre eding three years.

Several papers have attempted to measure herd behavior in investment de isions.

Jain and Gupta (1987) report only weak eviden e of herding in loans to LDC's by US

banks. D'Ar y and Oh (1997) study as ades in the de isions of insurers to underwrite

risks and the pri ing of insuran es. Foresi, Hamo, and Mei (1998) provide eviden e

onsistent with imitation in the investment de isions of Japanese �rms.

Is there a more general tenden y toward strategi imitation? Gilbert and Lieberman

38

(1987) examined the relation amongst the investments of 24 hemi al produ ts over two

de ades. They found that larger �rms in an industry tend to invest when their rivals

do not. In ontrast, smaller �rms tend to be followers in investment. This behavior is

onsistent with a `fashion leader' version of the as ades model in whi h the small free-

ride informationally on the large (where large �rms may have greater absolute bene�t

from a quiring pre ise information, or s ale ost e onomies in information a quisition).

Survey eviden e on Japanese �rms indi ates that a fa tor that en ourages �rms to engage

in dire t investment in an emerging e onomy in Asia is whether other �rms are investing

in that ountry. This is onsistent with possible as ading based upon a manager's

per eption that rival �rms possess useful private information about the desirability of

su h investment (Kinoshita and Mody (2001)). Greve (1998) provide eviden e of �rm

imitation in the hoi e of new radio formats in the U.S.

Chaudhuri, Chang, and Jayaratne (1997) examine spatial lustering of bank bran hes

in ities. They point out that banks are likely to have imperfe t information about the

potential pro�tability of opening a bran h in a parti ular neighborhood. They show

that a bank's de ision to open a new bran h in a ensus tra t of New York City during

1990-95 depended on the number of existing bran hes in that tra t. They use tra t-

level rime statisti s land-use data, and so ioe onomi data to ontrol for expe ted

tra t pro�tability. They on lude that there is a positive in remental relation between

a bank's de ision to open a new bran h and the presen e of other banks' bran hes,

onsistent with information-based imitation.

Analogous to the endorsment e�e t in indivdual investor trading are endorsement

e�e ts in real investments. Real estate investment is a prime area of appli ation for

as ades/endorsement e�e ts, be ause the investment de isions are dis rete and on-

spi uous (Caplin and Leahy (1998) analyze real estate herding/ as ading).24

E onomists have long studied agglomeration e onomies as an explanation for ge-

ographi al on entration of investment and e onomi a tivity (e.g., Marshall (1920),

Krugman and Venables (1995, 1996)). Su h e�e ts are surely important. However,

as pointed out by DeCoster and Strange (1993), geographi al on entration an o ur

without agglomeration e onomies owing to learning by observation of others: `spuri-

ous agglomeration.' Empiri ally some papers use previous investment by other �rms

24For example, onsider Bian o (1996) in Business Week entitled: \A Star is Reborn: Investors hustleto land parts in Times Square's transformation." The arti le states of Disney that \Its agreement torevamp the New Amsterdam Theater, a Beaux Arts gem, was like waving a magi wand: Wait-and-seeinvestors piled in." After long delay, the transformation of New York's Times Square was triggered byan investment by Disney, after whi h \wait-and-see investors piled in," an illustration of simultaneity.

39

in a lo ation as a proxy for agglomeration e onomies in predi ting investment by other

�rms (e.g., Head, Ries and Swenson (1995, 2000)). Barry, Gorg, and Strobl (2001)

empiri ally test between aggregation e onomies and what they all the \demonstration

e�e ts," whereby a �rm lo ates in a host ountry be ause the presen e of other �rms

there provides information about the attra tiveness of the host ountry. They on lude

that both agglomeration e onomies and agglomeration e�e ts are important.

The observation of the payo�s, not just a tions of rivals is learly important in �rms'

investment de isions. For example, after Sara Lee Corp. introdu ed the fashionable

Wonderbra to the U.S. in New York in May 1995, VF Corporation observed its popularity

and then \surged ahead with a nationwide rollout �ve months ahead of Sara Lee..."

(Weber (1995)). Referring to VF's `se ond-to-the-market' business strategy, Business

Week stated that \Letting others take the lead may be outre at Paris salons, but it's a

winning style at FV."

Reporting and dis losure pra ti es are variable over time; for example, re ently it has

been popular for �rms to dis lose pro forma earnings in ways that di�er from the GAAP-

permited de�nitions on �rms' �nan ial reports. Firms have argued that this allows them

to re e t better long-term pro�tability by adjusting for non-re urring items. However,

it is also possible that �rms are just herding, or that they are exploiting herd behavior

by investors. At this point the eviden e is not lear, though regulators have expressed

on ern about this pra ti e.

More generally, in a meta-study of a ounting hoi es, Pin us and Wasley (1994)

report that voluntary a ounting hanges by �rms do not appear to be lustered in

time and industry, suggesting no herding behavior in a ounting hanges. This result

further indi ates, surprisingly, that �rms do not swit h a ounting methods in response

to hanges in ma ro-e onomi investment onditions that are experien ed at about the

same time by similar �rms within an industry. Rather, the voluntary a ounting hanges

would appear to be made in response to �rm-spe i� needs, su h as a �rm-spe i� need

to manage earnings.

However, it is not obvious why �rms would need to manage earnings in response

to �rm-spe i� ir umstan es, yet would not want to manage earnings in response to a

ommon fa tor sho k. One spe ulative possibility is that there is a on ern for relative

performan e, as re e ted in the model of Zwiebel (1995), ombined with some deviation

from perfe t rationality that auses investors to adjust imperfe tly for a ounting method

in evaluating �rms' earnings.25 The on ern for relative performan e may reate a

25For example, Daniel, Hirshleifer, and Teoh (2002) suggest that owing to limited attention, Hirsh-

40

stronger in entive for managers to manage earnings upward when the �rm is doing

poorly relative to peers than when the entire industry is doing poorly.26

9 Con lusion

A ording to Gertrude Stein (as quoted by Charlie Chaplin), \Nature is ommonpla e.

Imitation is more interesting." We have des ribed here why imitation is interesting

for apital markets. In our dis ussion of rational observational learning, we des ribed

some emergent on lusions: idiosyn rasy (mistakes), fragility (fads), simultaneity (delay

followed by sudden joint a tion), paradoxi ality (more information of various sorts an

de rease welfare and de ision a ura y), and path dependen e. We have explored how

literature on herding, so ial learnings, and informational as ades an be applied to a

number of investment, �nan ing, reporting and pri ing ontexts.

We have also argued that these on lusions are fairly robust in rational so ial learning

models. Depending on the exa t assumptions, informationmay be ompletely suppressed

for a period (until a as ade is dislodged); under other assumptions, information is

asymptoti ally revealed, but too slowly. A setting where information arrives too slowly

to be helpful for most individuals' de isions is essentially the same from the point of

view of both welfare and predi ting behavior as one where information is ompletely

blo ked for a while. Although as ades require dis rete, bounded, or gapped a tion

spa e, or ognitive onstraints, we have argued that dis reteness and boundedness are

highly plausible in some �nan ial settings. Even when these onditions fail, owing to

noise, the growth in a ura y of the publi information pool tends to be self-limiting,

resulting in similar e�e ts.

There are many patterns of onvergent behavior and u tuations in apital markets

that do not obviously make immediate sense in terms of traditional e onomi models,

su h as �xation on poor proje ts, sto k market rashes, sharp shifts in investment and

unemployment, bank runs. Su h behavioral onvergen e often appears even in the fa e

of negative payo� externalities. Although other fa tors (su h as payo� externalities) an

lo k in ineÆ ient behaviors, the rational so ial learning theory and espe ially as ades

theory di�er in that they imply pervasive but fragile herd behavior. This o urs be-

leifer, Lim, and Teoh (2001) analyze expli itly how informed parties an adjust their dis losure de isionsto exploit the limited attention of observers.

26Consistent with this idea, Mor k, Shleifer, and Vishny (1989) provide eviden e that the likelihoodof hostile takeover for ing managerial turnover was high for �rms underperforming their industry, butwas not high when the industry as a whole was underperforming.

41

ause the a umulation of publi information slows down or blo ks the generation and

revelation of further information. This idiosyn rati feature of as ades and rational

observational learning models ause the so ial equilibrium to be pre arious with respe t

to seemingly modest new sho ks.

Rational observational learning theory suggests that in many situations, even if pay-

o�s are independent and people are rational, de isions tend to onverge qui kly but

tend to be idiosyn rati and fragile. Convergen e arises lo ally or temporally upon a

behavior, and an suddenly shift into onvergen e on the opposite behavior. The re-

quired assumptions, primarily dis reteness or boundedness of possible a tion hoi es,

are mild and likely to be present in many realisti setting. This suggests that the e�e ts

of observational learning and herding mentioned in the �rst paragraph of this se tion are

likely to a�e t behavior in and related to apital markets. This in ludes both herding

by �rms, and a tions by �rms su h as �nan ing, dis losure and reporting poli ies that

an potentially be managed to exploit investors that herd. Similarly, perhaps the spe ial

skill that some hedge fund and mutual fund managers seem to have is in exploiting the

herding behavior of imperfe tly rational investors.

Models of reputation-based herding do not typi ally share the fragility feature of

rational observational learning theory. However, reputation-based models have mu h to

o�er in their own right. This in ludes explanation of those herds that seem stable and

robust. As another example, the reputation approa h helps explain dispersion as well as

herding, and when one or the other will o ur. Reputation models also o�er a ri h set

of impli ations about the extent of herding in relation to hara teristi s of the agen y

problem and the manager.

Most instan es herding in apital markets are likely involve mixtures of reputational

e�e ts, informational e�e ts, dire t payo� intera tions, preferen e e�e ts, and imperfe t

rationality. For example, to explain predi tability in se urities markets, some imperfe t

rationality is likely to be needed. Integration of the di�erent e�e ts will lead us to better

theories about apital market behavior.

42

Referen es

Admati, A. and P. P eiderer, 1988, A theory of intraday patterns: Volume and pri evariability, Review of Finan ial Studies 1, 3{40.

Aharony, J. and I. Swary, 1996, Additional eviden e on the information-based onta-gion e�e ts of bank failures, Journal of Banking and Finan e 20, 57{69.

Aitken, B., 1998, Have institutional investors destabilized emerging markets?, Con-temporary E onomi Poli y 16, 173{184.

Allen, F. and D. Gale, 2000, Bubbles and rises, The E onomi Journal 110, 236{256.

Allen, F. and D. Gale, 2000, Finan ial ontagion, Journal of Politi al E onomy 108,1{33.

Anderson, L. R., 2001, Payo� e�e ts in information as ade experiments, E onomi Inquiry 39, 609{615.

Anderson, L. R. and C. A. Holt, 1996, Information as ades in the laboratory, Amer-i an E onomi Review 87, 847{862.

Arthur, B., 1989, Competing te hnologies, in reasing returns, and lo k-in by histori alevents, The E onomi Journal 99, 116{131.

Ashiya, M. and T. Doi, 2001, Herd behavior of Japanese e onomists, Journal ofE onomi Behavior and Organization 46, 343{346.

Avery, C. and P. Zemsky, 1998, Multi-dimensional un ertainty and herd behavior in�nan ial markets, Ameri an E onomi Review 88, 724{48.

Bala, V. and S. Goyal, 1998, Learning from neighbors, Review of E onomi Studies 65,595{621.

Banerjee, A., 1992, A simple model of herd behavior, Quarterly Journal of E o-nomi s 107, 797{817.

Banerjee, A. and D. Fudenberg, 1999, Word-of-mouth ommuni ation and so ial learn-ing, Harvard University Working Paper.

Barry, F., H. Gorg, and E. Strobl, 2001, Foreign dire t investment, agglomerationsand demonstration e�e ts: An empiri al investigation, Working paper, UniversityCollege Dublin.

Barsade, S. G., 2001, The ripple e�e t: Emotional ontagion within groups, YaleS hool of Management, Working Paper Series A # 91.

Beatty, R., 1989, Auditor reputation and the pri ing of initial publi o�erings, A - ounting Review 64, 693{709.

Beatty, R. and J. R. Ritter, 1986, Investment banking, reputation, and the underpri -ing of initial publi o�erings, Journal of Finan ial E onomi s 15, 213{232.

Beaudry, P. and F. M. Gonzalez, 2000, An equilibrium analysis of information aggre-gation in investment markets with dis rete de isions, Working paper, Universityof British Columbia.

43

Berg, A. and C. Pattillo, 1999, Are urren y rises predi table? a test, IMF Sta�Papers 46.

Bernardo, A. and I. Wel h, 2001, On the evolution of over on�den e and en-trepreneurs, Journal of Eonomi s and Management Strategy 10, 301{330.

Bhatta harya, S., K. Chatterjee, and L. Samuelson, 1986, Sequential resear h and theadoption of innovations, Oxford E onomi Review 38, 219{243.

Bhatta harya, S. and A. Thakor, 1993, Contemporary banking theory, Journal ofFinan ial Intermediation 3, 2{50.

Bhushan, R., 1989, Firm hara teristi s and analyst following, Journal of A ountingand E onomi s 11, 255{274.

Biais, B., P. Hillion, and C. Spatt, 1995, An empiri al-analysis of the limit order bookand the order ow in the Paris bourse, Journal of Finan e 50, 1655{1689.

Bian o, A., 1996, A star is reborn, Business Week , 102{106. 7/8/96.

Bikh handani, S., D. Hirshleifer, and I. Wel h, 1992, A theory of fads, fashion, ustom,and ultural hange as informational as ades, Journal of Politi al E onomy 100,992{1026.

Bikh handani, S., D. Hirshleifer, and I. Wel h, 1998, Learing from the behavior ofothers: Conformity, fads, and informational as ades, Journal of E onomi Per-spe tives 12, 151{170.

Bikh handani, S., D. Hirshleifer, and I. Wel h, 2001, Informational as- ades and rational herding: An annotated bibliography, Working paper:UCLA/Anderson and Ohio State University and Yale/SOM (The World-WideWeb http://wel h.som.yale.edu/ as ades/).

Bikh handani, S. and S. Sharma, 2001, Herd behavior in �nan ial markets, IMF Sta�Papers 47, 279{310.

Bla kmore, S., 1999, The Meme Ma hine (Oxford University Press, Oxford).

Bloom�eld, R., 1996, Quotes, pri es and estimates of value in a laboratory market,Journal of Finan e 51, 1791{1808.

Booth, J. and R. Smith, 1986, Capital rising, underwriting and the erti� ation hy-pothesis, Journal of Finan ial E onomi s 15, 261{281.

Borensztein, E. and R. G. Gelos, 2001, A pani -prone pa k? the behavior of emergingmarket mutual funds, CESifo Working Paper Series No. 564; IMF Working PaperNo. 00/198.

Brandenburger, A. A. and B. Polak, 1996, When managers over their posteriors:Making the de isions the market wants to see, Rand Journal of E onomi s 27,523{541.

Brennan, M. J., 1990, Latent assets, Journal of Finan e 45, 709{730.

Brennan, M. J., 1993, Agen y and asset pri ing, Working Paper, Anderson S hool,UCLA #6-93.

44

Bro k, W. A. and S. N. Durlauf, 2000, Intera tions-based models, NBER WorkingPaper No. T0258.

Brown, K. C., W. V. Harlow, and L. T. Starks, 1996, Of tournaments and tempta-tions: An analysis of managerial in entives in the mutual fund industry, Journalof Finan e 51, 85{110.

Brown, L., 2001, A temporal analysis of earnings surprise: Pro�ts versus losses, Jour-nal of A ounting Resear h, forth oming 39.

Brown, P., G. Foster, and E. Noreen, 1985,. Se urity analyst multi-year earningsfore asts and the apital market, . In Studies in A ounting Resear h, no. 21.(Ameri an A ounting Asso iation, Sarasota, Florida).

Brunnermeier, M., 2001, Asset Pri ing under Asymmetri Information: Bubbles,Crashes, Te hni al Analysis and Herding (Oxford University Press, Oxford, UK).

Bulow, J. and P. Klemperer, 1994, Rational frenzies and rashes, Journal of Politi alE onomy 102, 1{23.

Burguet, R. and X. Vives, 2000, So ial learning and ostly information a quisition,Journal of E onomi Theory 15, 185{205.

Calomiris, C. and J. Mason, 1997, Contagion and bank failures during the greatdepression: The june 1932 Chi ago banking pani , Ameri an E onomi Review 87,863{883.

Calomiris, C. W. and G. Gorton, 1991,. The origins of banking pani s: Models, fa ts,and bank regulation, . In R. G. Hubbard, ed., Finan ial Markets and Finan ialCrises, pp. 109{173. (University of Chi ago Press, Chi ago).

Calomiris, C. W. and J. R. Mason, 2001, Causes of US bank distress during thedepression, Working paper, LeBow College of Business.

Calvo, G. and E. Mendoza, 2001, Rational ontagion and globalization of se uritiesmarkets, Journal of International E onomi s 51, 79{113.

Caminal, R. and X. Vives, 1999, Pri e dynami s and onsumer learning, Journal ofE onomi s and Management Strategy 8, 95{131.

Cao, H. H., J. D. Coval, and D. Hirshleifer, 2001, Sidelined investors, trading-generated news, and se urity returns, forth oming, Review of Finan ial Studies.

Cao, H. H. and D. Hirshleifer, 1997, Limited observability, reporting biases, andinformational as ades, Working paper, Fisher College of Business, The Ohio StateUniversity.

Cao, H. H. and D. Hirshleifer, 2000, Conversation, learning and informational as- ades, Ohio State University Fisher College of Business Working Paper.

Caplin, A. and J. Leahy, 1993, Se toral sho ks, learning and aggregate u tuations,Review of E onomi Studies 60, 777{794.

Caplin, A. and J. Leahy, 1994, Business as usual, market rashes, and wisdom afterthe fa t, Ameri an E onomi Review 84, 548{565.

45

Caplin, A. and J. Leahy, 1998, Mira le on sixth avenue: Information externalities andsear h, E onomi Journal 108, 60{74.

Capsta�, J., K. Paudyal, and W. Rees, 1998, Analysts' fore asts of German �rms'earnings: a omparative analysis abstra t, Journal of International Finan ialManagement and A ounting 9, 83{116.

Carter, R., F. Dark, and A. Singh, 1998, Underwriter reputation, initial returns, andthe long-run performan e of IPO sto ks, Journal of Finan e 53, 285{311.

Carter, R. and S. Manaster, 1990, Initial publi o�erings and underwriter reputation,Journal of 45, 1045{1067.

Celen, B. and S. Kariv, 2001, An experimental test of observational learning underimperfe t information, Working paper, E onomi s Department, New York Univer-sity.

Chakrabarti, R. and R. Roll, 1997, Learning from others, rea ting, and market quality,Anderson Graduate S hool of Management, UCLA working paper #8-27.

Chalkley, M. and I. H. Lee, 1998, Learning and asymmetri business y les, Reviewof E onomi s Dynami s 1, 623{645.

Chamley, C., 1999, Coordinating regime swit hes, Quarterly Journal of E o-nomi s 114, 869{905.

Chamley, C., 2001, Delays and equilibria with large and small information in so iallearning, .

Chamley, C. and D. Gale, 1994, Information revelation and strategi delay in irre-versible de isions, E onometri a 62, 1065{1085.

Chang, E., J. Cheng, and A. Khorana, 2000, An examination of herd behavior inequity markets: An international perspe tive, Journal of Banking and Finan e 20,1651{1679.

Chari, V. and P. J. Kehoe, 2000, Finan ial rises as herds, Working paper, FederalReserve Bank of Minneapolis # 600.

Chari, V. V. and R. Jagannathan, 1988, Banking pani s, information and rationalexpe tation equilibrium, Journal of Finan e 43, 749{761.

Chaudhuri, S., A. Chang, and J. Jayaratne, 1997, Informational externalities andthe bran h lo ation de isions of banks: An empiri al analysis, Working Paper:Columbia University and Federal Reserve Bank of New York.

Chen, Y., 1999, Banking pani s: The role of the �rst- ome, �rst-served rule andinformation externalities, Journal of Politi al E onomy 107, 946{968.

Chevalier, J. and G. Ellison, 1997, Risk taking by mutual funds as a response toin entives, Journal of Politi al E onomy 105, 1167{1200.

Chevalier, J. and G. Ellison, 1999, Career on erns of mutual fund managers, Quar-terly Journal of E onomi s 114, 389{432.

46

Choe, H., B.-C. Kho, and R. M. Stulz, 1999, Do foreign investors destabilize sto kmarkets? the korean experien e in 1997, Journal of Finan ial E onomi s 54,227{264.

Christie, W. and R. Huang, 1995, Following the pied piper: Do individual returnsherd around the market?, Finan ial Analyst Journal July-August, 31{37.

Cipriani, M. and A. Guarino, 2001, Herd behavior and ontagion in �nan ial markets,New York University E onomi s Department, Working Paper.

Cipriani, M. and A. Guarino, 2001, Herding and pri e onvergen e in a laboratory�nan ial market, mimeo, New York University.

Conlisk, J., 1996, Why bounded rationality?, Journal of E onomi Literature 34,669{700.

Cont, R. and J. Bou haud, 1999,. Herd behavior and aggregate u tuations in �nan ialmarkets, . In Ma roe onomi Dynami s. in press.

Cote, I. and D. Sanders, 1997, Herding behavior: Explanations and impli ations,Behavioral Resear h in A ounting 9, 20{45.

Coval, J. D. and T. J. Moskowitz, 2001, The geography of investment: Informedtrading and asset pri es, Journal of Politi al E onomy 109, 811{841.

Daniel, K. D., D. Hirshleifer, and S. H. Teoh, 2002, Investor psy hology in apitalmarkets: Eviden e and poli y impli ations, Journal of Monetary E onomi s 49(1)/Carnegie Ro hester Series in Publi Poli y 56.

D'Ar y, S. P. and P. Oh, 1997, The as ade e�e t in insuran e pri ing, Journal ofRisk and Insuran e 64, 465{480.

Datar, S., G. Feltham, and J. Hughes, 1991, The role of audits and audit quality invaluing new issues, Journal of A ounting and E onomi s 14, 3{49.

Davis, R., 1991,. Di k Davis Digest .

DeCoster, G. P. and W. C. Strange, 1993, Spurious agglomeration, Journal of UrbanE onomi s 33, 273{304.

DeLong, J. B., A. Shleifer, L. Summers, and R. J. Waldmann, 1990, Noise trader riskin �nan ial markets, Journal of Politi al E onomy 98, 703{738.

DeMarzo, P., D. Vayanos, and J. Zwiebel, 2000, A near-rational model of persuasion{ with impli ations for �nan ial markets, Stanford University Business S hool,working paper.

Diamond, D. W. and P. H. Dybvig, 1983, Bank runs, deposit insuran e, and liquidity,Journal of Politi al E onomy 91, 410{419.

Do king, D., M. Hirs hey, and E. Jones, 1997, Information and ontagion e�e ts ofbank loan-loss reserve announ ements, Journal of Finan ial E onomi s 43`, 219{239.

Dow, J. and G. Gorton, 1994, Arbitrage hains, Journal of Finan e 49, 819{849.

47

Du o, E. and E. Saez, 2000, Parti ipation and investment de isions in a retirementplan: The in uen e of olleagues' hoi es, NBER Working Paper No. W7735.

Dugatkin, L. A., 1992, Sexual sele tion and imitation: Females opy the mate hoi eof others., Ameri an Naturalist 139, 138{149.

Dugatkin, L. A. and J. J. Godin, 1992, Reversal of female mate hoi e by opying inthe guppy (poe ilia reti ulata), Pro eedings of the Royal So iety of London, SeriesB 249, 179{184.

E kbo, B. E. and R. W. Masulis, 1995,. Seasoned equity o�erings: A survey, . InR. A. Jarrow, V. Maksimovi , and W. T. Z. W. T. Ziemba, eds., Finan e, Hand-books in Operations Resear h and Management S ien e and Management S ien e,Volume 9, Chapter 31, pp. 1017{1072. (North Holland, Amsterdam).

E onomist, 1998, Warren silver�nger, The E onomist . 2/7/98.

Ehrbe k, T. and R. Waldmann, 1996, Why are professional fore asters biased? agen yversus behavioral explanations, Quarterly Journal of E onomi s 111, 21{40.

Ei hengreen, B., A. Rose, and C. Wyplosz, 1996, Contagious urren y rises: Firsttests, S andinavian Journal of E onomi s 98, 463{484.

Ellison, G. and D. Fudenberg, 1993, Rules of thumb for so ial learning, Journal ofPoliti al E onomy 101, 612{643.

Ellison, G. and D. Fudenberg, 1995, Word of mouth ommuni ation and so ial learn-ing, Quarterly Journal of E onomi s 110, 93{126.

Feltham, G. A., J. S. Hughes, and D. A. Simuni , 1991, Empiri al assessment of theimpa t of auditor quality on the valuation of new issues, Journal of A ountingand E onomi s 14, 375{399.

Fishman, M. and K. Hagerty, 1995, The mandatory dis losure of trades and marketliquidity, Review of Finan ial Studies 8, 637{676.

Foresi, S., Y. Hamo, and J. Mei, 1998, Intera tion in investment among rival Japanese�rms, Japan and the World E onomy 10, 393{407.

Fortune, 1986, Inside the deal that made bill gates $300,000,000., Fortune 7/21/86.

Fried, J. M., 1998, Redu ing the pro�tability of orporate insider trading throughpretrading dis losure, Southern California Law Review 71, 303{392.

Friend, I., M. Blume, and J. Cro kett, 1970, Mutual funds and other institutionalinvestors (M Graw-Hill, New York, NY).

Froot, K., D. S harfstein, and J. C. Stein, 1992, Herd on the street: InformationalineÆ ien ies in a market with short-term spe ulation, Journal of Finan e 47,1461{1484.

Froot, K. A., P. O'Connell, and M. Seasholes, 2001, The portfolio ows of internationalinvestors, forth oming, Journal of Finan ial E onomi s.

Fudenberg, D. and D. M. Kreps, 1995, Learning in extensive-form games, i self- on�rming equilibria, Games and E onomi Behavior 8, 20{55.

48

Fung, W. and D. A. Hsieh, 1999, A primer on hedge funds, Journal of Empiri alFinan e 6, 309{331.

Gale, D., 1996, What have we learned from so ial learning?, European E onomi Review 40, 617{628.

Gale, D. and R. Rosenthal, 2001, Experimentation, imitation, and strategi stability:Addendum, Journal of E onomi Theory 97, 164{174.

Garber, P. M., 2000, Famous �rst bubbles: The fundamentals of early manias (MITPress, Cambridge, Mass.).

Gervais, S., 1996, Market mi rostru ture with un ertain information pre ision: Amultiperiod analysis, Wharton S hool, University of Pennsylvania.

Gibson, R. M. and J. Hoglund, 1992, Copying and sexual sele tion, Trends in Resear hin Evolution and E ology 7, 229{232.

Gilbert, R. and M. Lieberman, 1987, Investment and oordination in oligoppolisti industries, RAND Journal of E onomi s 18, 17{33.

Giraldeau, L.-A., 1997,. The e ology of information use, . In J. Krebs and N. B. Davies,eds., Behavioral E ology: An Evolutionary Approa h (4th ed.). (Bla kwell, Oxford,UK).

Givoly, D. and J. Lakonishok, 1984, The quality of analysts' fore ast of earnings,Finan ial Analysts Journal 40, 40{47.

Givoly, D. and D. Palmaon, 1985, Insider trading and the exploitation of inside infor-mation: Some empiri al eviden e, Journal of Business 58, 69{87.

Glaeser, E. L. and J. S heinkman, 2000, Non-market intera tions, NBER WorkingPaper No. W8053.

Glosten, L. R. and P. R. Milgrom, 1985, Bid, ask and transa tion pri es in a spe- ialist market with heterogeneously informed traders, Journal of Finan ial E o-nomi s 14, 71{100.

Gole , J., 1997, Herding on noise: The ase of Johnson redbook's weekly retail salesdata, Journal of Finan ial and Quantitative Analysis 32, 367{381.

Gompers, P. and J. Lerner, 2000, Money hasing deals?: The impa t of fund in owson private equity valuations, Journal of Finan ial E onomi s 55, 281{325.

Gonzalez, F., 1997, Individual experimentation and aggregate u tuations, UBC Dis- ussion Paper No. 97-24.

Gorton, G., 1985, Bank suspension of onvertibility, Journal of Monetary E o-nomi s 15, 177{193.

Gorton, G., 1988, Banking pani s and business y les, Oxford E onomi Papers 40,751{781.

Graham, J. R., 1999, Herding among investment newsletters: Theory and eviden e,Journal of Finan e 54, 237{268.

49

Granovetter, M., 1973, The strength of weak ties, Ameri an Journal of So iology 78,1360{1380.

Granovetter, M., 1978, Threshold models of olle tive behavior, Ameri an Journal ofSo iology 83, 1420{1443.

Grant, S., S. King, and B. Polak, 1996, Information externalities, share-pri e basedin entives and managerial behavior, Journal of E onomi Survey 10, 1{21.

Grenadier, S. R., 1999, Information revelation through option exer ise, Review ofFinan ial Studies 12, 95{129.

Greve, H. R., 1998, Managerial ognition and the mimeti adoption of market posi-tions: What you see is what you do, Strategi Management Journal 19, 967{988.

GriÆths, M., B. Smith, A. Turnbull, and R. White, 1998, Information ows and openout ry: Eviden e of imitation trading, Journal of International Finan ial Markets,Institutions & Money June, 101{116.

Grinblatt, M. and M. Keloharju, 2000, The investment behavior and performan e ofvarious investor types: a study of Finland's unique data set, Journal of Finan ialE onomi s 55, 43{67.

Grinblatt, M., S. Titman, and R. Wermers, 1995, Momentum investment strategies,portfolio performan e, and herding: A study of mutual fund behavior, Ameri anE onomi Review 85, 1088{1105.

Grossman, S. J. and J. E. Stiglitz, 1976, Information and ompetitive pri e systems,Ameri an E onomi Review 66, 246{253.

Grossman, S. J. and J. E. Stiglitz, 1980, On the impossibility of informationallyeÆ ient markets, Ameri an E onomi Review 70, 393{408.

Gul, F. and R. Lundholm, 1995, Endogenous timing and lustering of agents' de isions,Journal of Politi al E onomy 103, 1039{1066.

Gumbel, A., 1998, Herding in delegated portfolio management: When is omparativeperforman e information desirable?, European University Institute, E onomi s De-partment, No. 98-22.

Hamilton, W. D., 1971, Geometry for the sel�sh herd, Journal of Theoreti al Biol-ogy 31, 295{311.

Hat�eld, E., J. L. Ca ioppo, and R. L. Rapson, 1993, Emotional ontagion, CurrentDire tions in Psy hologi al S ien es 2, 96{99.

Hauns hild, P. R., 1993, Interorganizational imitation: The impa t of interlo ks on orporate a quisition a tivity, Administrative S ien e Quarterly 38, 564{592.

Head, K., J. Ries, and D. Swenson, 1995, Agglomeration bene�ts and lo ation hoi e:Eviden e from Japanese manufa turing investments in the united states, Journalof International E onomi s 38, 223{247.

Head, K., J. Ries, and D. Swenson, 2000, Attra ting foreign manufa turing: Invest-ment promotion and agglomeration, Regional S ien e and Urban E onomi s 29,197{218.

50

Hendri ks, K. and D. Koveno k, 1989, Asymmetri information, information external-ities, and eÆ ien y: The ase of oil exploration, Rand Journal of E onomi s 20,164{182.

Hirshleifer, D., 2001, Investor psy hology and asset pri ing, Journal of Finan e 64,1533{1597.

Hirshleifer, D., S. Lim, and S. H. Teoh, 2001, Dis losure to a redulous audien e: Therole of limited attention, Fisher College of Business, Ohio State University.

Hirshleifer, D. and R. Noah, 1999, Mis�ts and so ial progress, Working paper, OhioState University.

Hirshleifer, D., A. Subrahmanyam, and S. Titman, 1994, Se urity analysis and trad-ing patterns when some investors re eive information before others, Journal ofFinan e 49, 1665{1698.

Hirshleifer, D. and A. Thakor, 1992, Managerial onservatism, proje t hoi e anddebt, Review of Finan ial Studies 5, 437{470.

Hirshleifer, D. and I. Wel h, 2002, An e onomi approa h to the psy hology of hange: Amnesia, inertia, and impulsiveness, Journal of E onomi s and Man-agement Strategy 11.

Holden, C. W. and A. Subrahmanyam, 1996, Risk aversion, liquidity, and endogenousshort horizons, The Review of Finan ial Studies 9, 691{722.

Hong, H., J. Kubik, and J. C. Stein, 2001, So ial intera tions and sto k marketparti ipation, Stanford Business S hool Working Paper.

Hong, H. and J. C. Stein, 2001, Di�eren es of opinion, rational arbitrage and market rashes, Stanford Business S hool Working Paper.

Hong, H. G., J. Kubik, and A. Solomon, 2000, Se urity analysts' areer on erns andherding of earning fore asts, RAND Journal of E onomi s 31, 121{144.

Huberman, G., 1999, Familiarity breeds investment, Columbia University WorkingPaper, 97-04.

Huddart, S., 1999, Reputation and performan e fee e�e ts on portfolio hoi e byinvestment advisers, Journal of Finan ial Markets 2, 227{271.

Hung, A. and C. Plott, 2001, Information as ades: Repli ation and an extensionto majority rule and onformity rewarding institutions, forth oming, Ameri anE onomi Review.

Hvide, H. K., 2001, Tournament rewards and risk taking, forth oming, Journal ofLabor E onomi s.

Ibbotson, R., J. Ritter, and J. L. Sindelar, 1994, The markets problems with thepri ing of initial publi o�erings, Journal of Applied Corporate Finan e 7, 66{74.

Ja klin, C. and S. Bhatta harya, 1988, Distinguishing pani s and information-basedbank runs: Welfare and poli y impli ations, Journal of Politi al E onomy 96,568{592.

51

Ja�e, J. F. and J. M. Mahoney, 1999, The performan e of investment newsletters,Journal of Finan ial E onomi s 53, 289{307.

Jain, A. K. and S. Gupta, 1987, Some eviden e on 'herding' behavior by U.S. banks,Journal of Money, Credit and Banking 19, 78{89.

Johnson, J. M. and R. E. Miller, 1988, Investment banker prestige and the underpri -ing of initial publi o�erings, Finan ial Management 17, 19{29.

Kahn, C. M., G. G. Penna hi, and B. Sopranzetti, 2002, Bank deposit rate lustering;theory and empiri al eviden e, forth oming, Journal of Finan e.

Kelly, M. and C. O. O'Grada, 2000, Market ontagion: Eviden e from the pani s of1854 and 1857, Ameri an E onomi Review 90, 1110{1124.

Khanna, N., 1997, Optimal ontra ting with moral hazard and as ading, Review ofFinan ial Studies 11, 559{596.

Khanna, N. and S. L. Slezak, 2000, The e�e t of organizational form on informationaggregation and proje t hoi e: The problem of informational as ades, Journalof E onomi s and Management Strategy 9, 115{156.

Kim, C. and C. Pantzalis, 2000, Analyst herding behavior and �rm diversi� ation:Eviden e from panel data, City University of New York - Queens College, WorkingPaper.

Kinoshita, Y. and A. Mody, 2001, Private information for foreign investment in emerg-ing e onomies, Canadian Journal of E onomi s 34, 448{464.

Klemkosky, R. C., 1977, The impa t and eÆ ien y of institutional net trading imbal-an es, Journal of Finan e 32, 79{86.

Kodres, L. E. and M. Pritsker, 1997, Dire tionally similar position taking and herd-ing by large futures market parti ipants, Working Paper: International MonetaryFund and Board of Governors of the Federal Reserve System.

Kraus, A. and H. R. Stoll, 1972, Parallel trading by institutional investors, Journalof Finan ial and Quantitative Analysis 7, 2107{2138.

Krugman, P. R. and A. J. Venables, 1995, Globalisation and the inequality of nations,Quarterly Journal of E onomi s CX, 857{880.

Krugman, P. R. and A. J. Venables, 1996, Integration, spe ialization, and adjustment,European E onomi Review 40, 959{967.

Kuran, T., 1989, Sparks and prairie �res: A theory of unanti ipated politi al revolu-tion, Publi Choi e 61, 41{74.

Kuran, T., 1998, Private Truths, Publi Lies: The So ial Consequen es of Preferen eFalsi� ation (Harvard University Press, Cambridge MA).

Kuran, T. and C. Sunstein, 1999, Availability as ades and risk regulation, StanfordLaw Review 51, 683{768.

Kyle, A. S., 1985, Continuous au tions and insider trading, E onometri a 53, 1315{1335.

52

Lakonishok, J., A. Shleifer, and R. W. Vishny, 1992, The impa t of institutionaltrading on sto k pri es, Journal of Finan ial E onomi s 32, 23{43.

Lee, I. H., 1993, On the onvergen e of informational as ades, Journal of E onomi Theory 61, 395{411.

Lee, I. H., 1998, Market rashes and informational avalan hes, Review of E onomi Studies 65, 741{759.

Lewis, K. K., 1999, Trying to explain home bias in equities and onsumption, Journalof E onomi Literature 37, 571{608.

Libby, R., R. Bloom�eld, and M. Nelson, 2001, Experimental resear h in �nan iala ounting, Cornell University.

Lohse, D., 1998, Tri ks of the trade: `bu�ett is buying this' and other sayings of the old- all rew, Wall Street Journal . 6/1/98.

Lowry, M. and G. W. S hwert, 2002, Ipo market y les: Bubbles or sequential learn-ing?, Journal of Finan e.

Lux, T., 1995, Herd behavior, bubbles and rashes, E onomi Journal 105, 881{896.

Lyn h, A., 2000, Thought ontagions in the sto k market, Journal of Psy hology andFinan ial Markets 1, 10{23.

Madrian, B. and D. Shea, 2000, The power of suggestion: Inertia in 401(k) parti ipa-tion and savings behavior, NBER working paper 7682.

Marshall, A., 1920, Prin iples of E onomi s (8 ed.). (Ma millan, London).

Matsumoto, D., 2001, Management's in entives to avoid negative earnings surprises,forth oming, A ounting Review.

Maug, E. and N. Naik, 1996, Herding and delegated portfolio management: Theimpa t of relative performan e evaluation on asset allo ation, Working paper,London Business S hool.

M Gee, S., 1997, Name game: More IPOs tout their famous investors, Wall StreetJournal 10/26/97, C1, C20.

Megginson, W. and K. Weiss, 1991, Venture apitalists erti� ation in initial publi o�erings, Journal of Finan e 46, 879{903.

Mi haely, R. and W. H. Shaw, 1995, Does the hoi e of auditor onvey quality in aninitial publi o�ering?, Finan ial Management 24, 15{30.

Mikhail, M. B., B. R. Walther, and R. H. Willis, 1999, Does fore ast a ura y matterto se urity analysts?, The A ounting Review 74, 185{200.

Mor k, R., A. Shleifer, and R. Vishny, 1989, Alternative me ahnism for orporate ontrol, Ameri an E onomi Review 79, 842{852.

Mos arini, M., M. Ottaviani, and L. Smith, 1998, So ial learning in a hanging world,E onomi Theory 13, 657{665.

Neeman, Z. and G. Orosel, 1999, Herding and the winner's urse in markets withsequential bids, Journal of E onomi Theory 85, 91{121.

53

Nelson, L.-B., 2001, Persisten e and reversal in herd behavior, forth oming, Reviewof Finan ial Studies.

Noeth, M., P. Camerer, Colin F., C. R., and M. Weber, 2002, Information aggregationin experimental asset markets: Traps and misaligned beliefs, forth oming, Reviewof Finan ial Studies.

Nofsinger, J. R. and R. W. Sias, 1999, Herding and feedba k trading by institutionaland individual investors, Journal of Finan e 54, 2263 { 2295.

Obrien, T. L. and M. Murray, 1995, Bu�ett boosts ameri an express stake to 9.8%,a quires 8.3% of PNC bank, Wall Street Journal . 2/15/95.

Ottaviani, M. and P. Sorenson, 2000, Herd behavior and investment: Comment,Ameri an E onomi Review 90, 695{704.

Ottaviani, M. and P. Sorenson, 2001, Information aggregation in debate: who shouldspeak �rst?, Journal of Publi E onomy 81, 393{421.

Perktold, J., 1996, Re urring information as ades, Working paper, University ofChi ago.

Persons, J. C. and V. A. Warther, 1997, Boom and bust patterns in the adoption of�nan ial innovations, Review of Finan ial Studies 10, 939{967.

Pin us, M. and C. Wasley, 1994, The in iden e of a ounting hanges and hara ter-isti s of �rms making a ounting hanges, A ounting Horizons 8, 1{24.

Prendergast, C., 1993, A theory of yes men, Ameri an E onomi Review 83, 757{770.

Prendergast, C. and L. Stole, 1996, Impetuous youngsters and jaded oldtimers: Ananalysis of behavioral de ision-making rules, Journal of Politi al E onomy 104,1105{1134.

Rajan, R. G., 1994, Why redit poli ies u tuate: A theory and some eviden e,Quarterly Journal of E onomi s 109, 399{442.

Remond, C. S. and R. Hennessey, 2001, Emex's shares drop 36% as �rm explains�nan ing, Wall Street Journal Thursday, May 31.

Ri hards, A., 1999, Idiosyn rati risk - an empiri al analysis, with impli ations for therisk of relative-value trading strategies, International Monetary Fund, Washington,DC # 99-148.

Ri hardson, S., S. H. Teoh, and P. D. Wyso ki, 2001, The walkdown to beatableanalysts' fore asts: The roles of equity issuan e and insider trading in entives,Ohio State University Fisher College of Business Working Paper.

Romer, D., 1993, Rational asset-pri e movements without news, Ameri an E onomi Review 83, 1112{1130.

Sa erdote, B., 2001, Peer e�e ts with random assignment: Results for Dartmouthroommates, Quarterly Journal of E onomi s 116, 681{704.

Sandler, L. and A. Raghavan, 1996, Salomon holders wat h for possible bu�etting,9/23/96, Wall Street Journal, p. C1.

54

Saunders, A. and B. Wilson, 1996, Contagious bank runs: Eviden e from the 1929-1933 period, Journal of Finan ial Intermediation 5, 409{423.

S harfstein, D. S. and J. C. Stein, 1990, Herd behavior and investment, Ameri anE onomi Review 80, 465{479.

S helling, T. C., 1978, Mi romotives and Ma robehavior (Norton Publisher, NewYork).

S hroeder, M., 1994, Behind the s andal at spe trum, Business Week , 60{62. 3/21/94.

S iubba, E., 2001, Relative performan e and herding in �nan ial markets, Workingpaper, Fa ulty of E onomi s and Politi s, University of Cambridge.

Sgroi, D., 2000, The right hoi e at the right time: A herding experiment in endoge-nous time, Working paper, University of Oxford.

Shiller, R. J., 2000, Conversation, information, and herd behavior, Ameri an E o-nomi Review 85, 181{185.

Shiller, R. J., 2000, Irrational exuberan e (Prin eton University Press, Prin eton,N.J.).

Shiller, R. J. and J. Pound, 1989, Survey eviden e on the di�usion of interest andinformation among investors, Journal of E onomi Behavior and Organization 12,46{66.

Sias, R. W. and L. Starks, 1997, Return auto orrelation and institutional investors,Journal of Finan ial E onomi s 46, 103{131.

Simon, H. A., 1955, A behavioral model of rational hoi e, Quarterly Journal ofE onomi s 69, 99{118.

Simuni , D., 1991, Empiri al assessment of the impa t of auditor quality on the val-uation of new issues, Journal of A ounting and E onomi s 14, 375{399.

Smallwood, D. and J. Conlisk, 1979, Produ t quality in markets where onsumers areimperfe tly informed, Quarterly Journal of E onomi s 93, 1{23.

Smith, L. and P. Sorenson, 2000, Pathologi al out omes of observational learning,E onometri a 68, 371{398.

Sti kel, S. E., 1990, Predi ting individual analyst earnings fore asts, Journal of A - ounting Resear h 28, 409{417.

Sti kel, S. E., 1992, Reputation and performan e among se urity analysts, Journal ofFinan e 48, 1811{1836.

Strang, D. and M. Ma y, 2001, In sear h of ex ellen e: Fads, su ess stories, andadaptive emulation, Ameri an Journal of So iology 107, 147{182.

Sunder, S., 1995,. Experimental asset markets: A survey, . In J. H. Kagel and A. E.Roth, eds., The Handbook of Experimental E onomi s, Chapter 6, pp. 445{500.(Prin eton University Press, Prin eton, NJ).

Tesar, L. and I. M. Werner, 1995, Home bias and high turnover, Journal of Interna-tional Money and Finan e 14, 467{492.

55

Titman, S. and B. Trueman, 1986, Information quality and the valuation of new issues,Journal of A ounting and E onomi s 8, 159{172.

Trueman, B., 1994, Analyst fore asts and herding behavior, Review of Finan ial Stud-ies 7, 97{124.

Veldkamp, L. L., 2000, Slow boom, big rash, Stanford University, SIEPR Poli ypaper No. 00-16.

Vives, X., 1993, How fast do rational agents learn?, Review of E onomi Studies 60,329{347.

Vives, X., 1995, The speed of information revelation in a �nan ial market me hanism,Journal of E onomi Theory 67, 178{204.

Weber, 1995, Arti le on sara lee and vf orporations, Business Week .

Week, B., 1998, Why investors stampede:... and why the potential for damage isgreater than ever, Business Week , 84{85. 2/13/98.

Weinberg, B. A., P. B. Reagan, and J. J. Yankow, 2000, Do neighborhoods a�e t workbehavior: Eviden e from the nlsy79, Working paper, The Ohio State University.

Wel h, I., 1992, Sequential sales, leaning, and as ades, Journal of Finan e 47, 695{732.

Wel h, I., 2000, Herding among se urity analysts, Journal of Finan ial E onomi s 58,369{396.

Wermers, R., 1999, Mutual fund herding and the impa t on sto k pri es, Journal ofFinan e 54, 581{622.

Zeira, J., 1999, Informational overshooting, booms and rashes, Journal of MonetaryE onomi s 43, 237{257.

Zhang, J., 1997, Strategi delay and the onset of investment as ades, RAND Journalof E onomi s 28, 188{205.

Zitzewitz, E., 2001, Measuring herding and exaggeration by equity analysts and otheropinion sellers, Working paper, Stanford GSB.

Zwiebel, J., 1995, Corporate onservatism and relative ompensation, Journal of Po-liti al E onomy 103, 1{25.

56

dd

I. A. Herding/ Dispersing

III.

Reputational

Herding and

Dispersion

D. Informational Cascades

C. Rational Observational Learning

B. Observational Influence

II. Payoff and

Network Externalities

Figure 1: Topic Diagram