Professional Documents
Culture Documents
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
is collaborating with JSTOR to digitize, preserve and extend access to The American Economic
Review
Financial crises are supposed to be rare theory seeks to explain precisely why the proba-
events, yet they occur quite often. According bility estimates of a crisis in a boom are too low,
to Reinhart and Rogoff (2009), investors suffer offering a foundation of "unanticipated" shocks,
from "this time is different" syndrome, failing to and of zero probabilities attached to some states
see crises coming because they do not recognize of the world by investors (Gennaioli, Shleifer,
similarities among the different pre-crisis bub- and Vishny 2012). Our theory yields boom-bust
bles. As a result, each crisis surprises investors. financial crises based entirely on beliefs; we do
Economists typically model financial crises as not incorporate into the model any economic
responses to shocks to which investors attach a mechanisms that amplify the shocks, such as fire
low probability ex ante, but which nonetheless sales or imperfect capital markets.
materialize. Such shocks (sometimes referred Our theory is based on Kahneman and
to as "MIT shocks"; e.g., Caballero and Simsek Tversky's (1972) idea of representativeness,
2013) are consistent with rational expectations in previously modeled by Barberis, Shleifer, and
that investors recognize that there is a small chance Vishny (1998); Gennaioli and Shleifer (2010);
that the shock might occur, but they are harder to and Bordalo, Gennaioli, and Shleifer (2014).
reconcile with the Reinhart Rogoff (2009) obser- In our model, representativeness induces peo-
vation that crises are not that unusual. ple to overestimate the probability of outcomes
The 2008 financial crisis in the United States that are relatively more likely in light of recently
has deepened the challenge, by bringing up directobserved data. Representativeness is intimately
evidence that investors underestimated the risk of related to the idea of similarity: after seeing
a crisis. Covai, Jurek, and Stafford (2009) show some data, people concentrate their forecasts on
that investors underestimated the probability of outcomes similar to the data observed, neglect-
mortgage defaults in pricing mortgage backed ing alternative future paths.
securities. Foote, Gerardi, and Willen (2012) find This principle has far-reaching implications
that investors did not even contemplate the mag- for finance. An investor observing a string of
nitude of home price declines that actually mate- good news (Internet stocks, housing prices)
rialized. Rather than being considered unlikely, views them as being generated by a favorable
the risks appear to have been entirely neglected. economic scenario. A series of good news is
We need a theory of beliefs consistent with sharp similar to a continuing boom. The investor
underestimates of the odds of a crisis. then puts too much probability weight on that
In this paper, we present a psychological theory scenario and neglects the risk of bad outcomes.
of the neglect of risk and financial crises. The If investor expectations are elicited at this point,
they look extrapolative.
Observing some bad news intermixed with
* Gennaioli: Department of Finance, Universita Bocconi, good news does not change the investor's mind.
Via Roentgen 1, Milan, Italy (e-mail: nicola.gennaioli@ He views the bad news as an aberration and
unibocconi.it); Shleifer: Economics Department, Harvard under-reacts. It takes a string of unfavorable
University, Littauer Center, 1805 Cambridge Street,
news to render the bad outcome sufficiently more
Cambridge, MA 02138 (e-mail: shleifer@fas.harvard.edu);
Vishny: Chicago Booth, University of Chicago, 5807 South likely that the representative scenario changes
Woodlawn Avenue, Chicago, IL 60637 (e-mail: Robert. from boom to bust. A pattern of sufficiently
vishny@chicagobooth.edu). Gennaioli thanks the European dramatic or continuing bad news is similar to
Research Council (Starting Grant #241114) for financial
the low payoff state, leading to a change in the
support.
ŤGo to http://dx.doi.org/10.1257/aer.p20151091 to visit
underlying beliefs. Previously ignored bad news
the article page for additional materials and author disclo- is remembered, leading to a sharp rise in the per-
sure statement(s). ceived probability of a crisis and a collapse of
310
LEMMA 1 (Rational
prices. The investor now overreacts to theBenchmark):
bad Under Al,
news, especially if the true probability of
after the observation the
ofn® bits of good news , the
amount of debt of
low state remains low. The possibility issued at t = 1 and its price are
black
swans is initially ignored, butd'ultimately
= p{d') = >'/• turns
into an overstated fear that leads to a self-gener-
A moderate
ating crisis. In contrast to rational amount of good news does not
expectations,
change the Bayesian
the model yields purely belief-driven boom posterior
bust enough. As a
cycles. consequence, the amount of debt issued and its
value do not change relative to what would in
I. The Model principle have happened at t = 0 if debt were
issued then. The rational expectations funding
There is one asset, such as a mortgage,
policyand
is very conservative. This (low) amount
one investor. The cash flow of the asset, of
received
debt is completely information insensitive, so
its price
at the very end, is a random variable taking val-does not change at t = 2 either.
ues in Y = {>'/,,>'/}» where yh > y¡. There are
three periods t = 0, 1 , 2. The investor receives II. The Boom Under Thinking
bits of news about the probability of each final through Representativeness
payoff between periods, and updates his beliefs
(the news could be payoff realizations of similar
In contrast to the rational benchmark, suppose
that belief formation is guided by representa-
assets). At t = 0, the investor's prior expected
probability of yk is 7r¿, with 7r¿ > 7T/0.tiveness.
We do In particular, what is representative at
not think of either state as extremely unlikely,
t depends on a comparison with the past, t - 1 .
although good times are more likely than bad. At
DEFINITION
time /, after observing a sample of (nh,n¡) bits of 1: At time t , the representative-
good and bad news, a Bayesian investorness would
of cash flow yk is formally defined as
update the posterior expected probability of yk to
= (^k + w*)/(l + nh + ni)- This updating
rule obtains if the prior distribution over
(1) RM = 4Ā- ^k
is Dirichlet with parameters (7r£,7T/0).
At t = 1, debt is issued to the investor Investors then deflate by a factor 6 £ [0, l]
against the asset's cash flow. Issuers maximizethe odds of the less representative cash flow.
profits. The investor is assumed to be risk neu-Equally representative cash flows are deflated
tral as long as the expected default probabilityby
is the same factor.
below p, but infinitely risk averse if the expected
probability of default is higher than p, where Representativeness maps reality into what
p < 7 T/0. This discontinuity in risk bearing investors are thinking about. The most represen-
capacity may reflect an institutional constrainttative cash flow at t is the one whose probability
facing the investor, such as a value-at-risk exhibits the largest percentage increase in light
constraint. of the data.1 The probability of this cash flow
To see what happens in a boom, assume is inflated relative to the less representative one
that, between t = 0 and t - 1 , a string of nj¡ (but probabilities still add up to one). Intuitively,
good draws is observed, and no bad news. By investors weigh recent data too much in their
Bayesian updating, the expected probability assessments.
is over-priced
Parameter S captures the and over-issued,
severity as in
of t
tortion of probabilities.
Shleifer, andWhen <5 2013).
Vishny (2012, = 0,
Although
only think about the mostinvestors fail to full
representa
pate the
flow, forgetting the risk of losses,
other. Thistheycase do so
co n
losses occur
to a complete neglect of with
risk, a low
asprobabilit
in Ge
Shleifer, and Vishny
the loss state(2012). Whe
yt may be quite lik
investors hold rational
are neglected expectations.
because they are not F
mediate ás, investors
tive of overestimate
the good news that market the
of representative have observed. This is a form of "this time is
states.
Consider now the implications
different syndrome": the good news creates oftoo thi
much faith
Begin with a stable in good fundamentals,
situation in which leads
which
ing occurs and theto neglectdistribution
of risk and excessive debt issuance, 7r{
the same as n®.as By equation
in Baron and (1),
Xiong (2014). Extrapolation of in
all cash flows aregoodequally
times and the neglectrepresentativ
of downside risk are
R(yht) = T^k/^k
part of = 1-
the same Even
psychological under
mechanism of
sentativeness, then, the rational benchmark representativeness.
d' = pr{d' ) = >>/ obtains. This mechanism highlights two major dif-
Consider what happens in a boom under rep- ferences between this psychologically founded
resentativeness. Now, observing bits of good model and the canonical "unanticipated shock."
news exerts a more drastic effect on beliefs First, markets will get exposed to the risk of
than under rational expectations. Under repre- losses rather frequently: even relatively likely
sentativeness, this news does not just raise outcomes
the such as yt may be neglected. Second,
probability of the high cash flow state, but itthe
alsorisks that get neglected endogenously depend
on actual fundamental changes and news. This is
renders that state representative. The following
result holds: a testable implication.