The Curse of Knowledge in Economic Settings: An Experimental Analysis
Author(s): Colin Camerer, George Loewenstein, Martin Weber
Source:
The Journal of Political Economy,
Vol. 97, No. 5 (Oct., 1989), pp. 1232-1254
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/1831894 .
Accessed: 14/02/2011 14:13
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=ucpress. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
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.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Political Economy.
http://www.jstor.org
The
Curse of
Knowledge in
Economic
Settings:
An
Experimental Analysis
Colin Camerer
University
of Pennsylvania
George
Loewenstein
University
of Chicago
Martin
Weber
Institut fur
Wirtschaftswissenschaften
In
economic analyses of
asymmetric information,
better-informed
agents are
assumed capable of
reproducing the
judgments of less-
informed agents. We discuss a
systematic violation of
this assump-
tion
that
we call the "curse of
knowledge."
Better-informed agents
are
unable to ignore private
information even when it
is in their
interest
to
do so; more
information is not always better.
Comparing
judgments
made in
individual-level
and market
experiments,
we
find
that
market forces reduce
the curse by
approximately 50 per-
cent but do not
eliminate it.
Implications
for
bargaining, strategic
behavior
by firms, principal-agent
problems,
and
choice under
un-
certainty
are
discussed.
We
thank
Andy Daughety,
Marc
Knez,
Pete
Kyle,
Howard
Kunreuther, Amy
McCready,
and
Charles
Plott
for useful
discussions and assistance.
The financial
sup-
port
of the Risk and Decision Processes Center at
the
Wharton
School
is
gratefully
acknowledged.
We also
acknowledge
support
from the National Science
Foundation to
Camerer,
from
the Alfred
P.
Sloan
Foundation,
the
Russell
Sage
Foundation,
and
the
IBM
Faculty
Research Fund at the
University
of
Chicago
to
Loewenstein,
and
from
Deutsche
Forschungsgemeinschaft
to Weber
during
his visit at
the
Wharton
School.
Journal
of
Political Economy, 1989, vol. 97, no. 5]
?
1989
by The
University of Chicago. All rights
reserved. 0022-3808/89/9705-0004$01.50
1232
CURSE OF
KNOWLEDGE
1233
I.
Introduction
In many
economic transactions, some
agents know more than
others.
For example,
sellers are better informed
about the true value
of
their
products than
buyers; workers know
more
about their
ability
and
motivation
than prospective employers.
Attempts by informed agents
to
exploit the
asymmetry in information
can
result
in
market-level
consequences.
Beneficial bargains go
unstruck because of adverse
selection (Akerlof
1970), bid-ask spreads
in
financial
markets
increase
when insiders
are present (Glosten and
Milgrom 1985), and wages
can diverge from
productivity when
workers' characteristics
are un-
observable
(Salop and Salop 1976). The
conventional
assumption
in
such
analyses of
asymmetric
information
is
that better-informed
agents can
accurately anticipate the
judgments
of
less-informed
agents. We discuss
a systematic judgmental
bias
that challenges
this
apparently
noncontroversial assumption.
In
predicting
the judgments
of
others,
agents are unable
to
ignore
the additional
information they possess.
This "curse of knowledge"'
has two
consequences: First,
better-informed agents may suffer
losses. More
information can actually
hurt. Second, the curse of
knowledge can
mitigate market
consequences resulting from infor-
mation
asymmetry. For example, the seller
of a "lemon" may lower its
price to reflect
unobservable defects,
reducing the degree of market
failure.
Paradoxically, individual
irrationality
can enhance collective
rationality.2
All
the
previous
evidence of the curse of
knowledge
has been
gath-
ered in
psychological studies of individual
judgments (see, e.g.,
Fisch-
hoff
1975). But
the important question
for economics is whether the
curse
harms the
allocation
of resources
in
economic settings.
There
are
several
reasons
why economists might be
skeptical
of
the
psychol-
ogists' findings:
(1) The curse may result from
careless thinking by
subjects
who
have no financial incentive to
respond accurately.
Given
the
right
incentives, it
could
be
conjectured, individuals would exert
sufficient cognitive effort to
overcome
the
curse. (2)
In
natural
set-
tings, people often
receive feedback about
the accuracy
of
their pre-
dictions. Over
time, such feedback might
reduce or eliminate judg-
mental bias.
Subjects
in
earlier studies
of the curse did not receive
feedback and,
hence,
did
not have an
opportunity
to learn from their
mistakes. (3)
Disciplining forces in markets
are not present in psychol-
ogy experiments.
In
markets, agents
who
are
less
subject
to the curse
of
knowledge
might exert
disproportionate influence
on
prices and
'
This term
was
suggested
by Robin
Hogarth.
2
Analogously,
contributions to
public
goods may
be
individually unprofitable
but
collectively
profitable (e.g.,
Dawes and
Thaler
1988).
1234
JOURNAL
OF
POLITICAL
ECONOMY
allocations, effectively
reducing
or
eliminating
market-level
effects
even
if
many agents
are biased.
We test
arguments
1-3 by using
market
experiments
to
see
whether financial incentives,
learning
from
feedback,
and market
forces make the curse of knowledge
disappear.
We
find
that feedback
alone
has
little
effect,
while market
forces reduce
the
magnitude
of
the
curse by approximately
50 percent.
After
describing
the
experi-
ments and results, we return
to the central question
of how
these
judgment biases might affect
economic settings.
Our
experiments
are
one
example
of
empirical
efforts to determine
whether violations
of
normative
theories of judgment
and
choice,
typically found
in
psychological
studies, tend
to
persist
in
economic
settings (e.g., Camerer
1987). Such tests may help
answer
long-
standing theoretical questions
about the behavioral
foundations
of
economic theory (e.g., Akerlof
and
Yellen
1985;
Russell and Thaler
1985; Hogarth and Reder 1986).
II.
Formal Representation
of the Curse
of
Knowledge
By expressing the curse of
knowledge formally,
we
shall
see that it
violates a normative rule-the
"law of iterated expectations"-much
as
choices violate the
normative
model of
expected
utility theory
(Machina 1987; Weber and
Camerer 1987)
and
probability
judg-
ments violate normative principles
such as Bayes's rule and
the con-
junction
rule
(e.g., Kahneman,
Slovic, and Tversky 1982).
Call the random variable
being forecasted
X. If
X
is
a discrete
event,
then it
has the value zero or one. Forecasts
of
X
depend
on
the
information set available to the
forecaster. Assume that there
are two
information sets
Io
and
II,
where Io is a subset of
II.
A forecaster
with
information set
I,
knows
everything that the forecaster
with
in-
formation set
Io knows,
and more.3 Denote the optimal
forecast of
X
given
the
information set
Io
by
E(X1IO).
We are interested
in
forecasts
of
forecasts, which are useful
when agents need to forecast
behavior
of
other
agents.
An
agent
with information set
I,
who forecasts
the
forecast of
an agent with information
Io is estimating
E[E(X1I0)1I11].
The
law of iterated expectations
states that if
I,
includes
Io,
then
E[E(X1I0)1I1]
must equal E(XIIo)
(Chow and Teicher 1978,
p. 204).
Better-informed agents should
ignore their additional information
when
forecasting the forecasts
of less-informed agents.
When the
3 In
technical
treatments,
I,
is a
finer
partition
of
a
probability space
than
I,);
the
additional
knowledge
contained in
II,
like
the
additional
information given
to subjects
in
our
experiments,
presumably
enables
one to
make a
finer
partition.
CURSE OF KNOWLEDGE
1235
curse
of
knowledge
occurs,
the forecaster with
information
I,
overesti-
mates
the
scope
of Io.
Formally,
the
curse
of
knowledge
means
that
E[E(X|Io)|I1
]
is not
equal to
E(X|IO),
but is somewhere
between
E(X|Io)
and
E(X|II). A
simple
model
we test
in
our
experiments is
E[E(XI0)1I11]
=
wE(XIIi)
+
(1
-
w)E(XIIo).
(1)
If
w
=
0, an
agent is
applying
the law
of
iterated
expectations
cor-
rectly.
If w
=
1,
agents
who
know
II
think that
all
other
agents
know
II
too.
The
parameter w
thus
measures
the
degree
of curse
of
knowl-
edge.
III.
Experimental
Design
The
market
experiments
consisted
of
two
stages.
In
the first
stage
we
collected
predictions
of eight
actual
companies'
earnings
from
51
Wharton
students. In
the second
stage, conducted
2
months after
the
first,
we
informed a
second
group of
subjects of
the
actual
earnings
and
had them
trade
assets
that
paid a
liquidating
dividend
equal
to
the
predictions
of
the
first
group.
This
second
group
knew
that
the
asset
dividend
was
determined
by the first
group's
predictions,
but
they
did
not
know
the
exact
amount of
the dividend.
In
the first
stage,
subjects
were
given
a Value
Line4
report
about
each
company's
prospects
in
1980,
along
with
a brief
summary
of
the
company's
business
activity,
annual
earnings
per
share
from
1970
through
1979, and
quarterly
earnings
per
share
from 1977
through
1979.5
A
sample
report is
shown in
the
Appendix
(fig. A2).
Each
report
had
a
blank
box in
the
upper
right-hand
corner,
labeled
"1980."
Subjects
were instructed
to write their
estimate
of
the com-
pany's
actual
1980
earnings
in
the box.
For
the
sake of
credibility,
the
companies'
names were
not dis-
guised.
However,
no well-known
companies were
selected,
and
there
was
no
evidence
that
any
subject
knew
any
company well
enough
to
recall
any
details
of its
history
other
than
those
given in
the
report.
These
data
were
collected as a classroom
exercise
in
a
quantitative
methods
course.
As
an
incentive
for
accuracy,
subjects
were
paid
$1.00 for
each
estimate that
was
within
10
percent of actual 1980
earnings.
4
Value
Line issues
reports
used by
investors
to
assess future
earnings
potential of
companies.
5 Most
market
experiments
involving
uncertainty rely on random
probability devices
such
as bingo
cages
or dice.
These
devices are
used
because
experimenters
wish to
control
subjective
probabilities
as
tightly as
possible.
Since our
study is
specifically
focused on
differences in
subjective
probability, we
used
natural stimuli
that
permit
such
differences
instead of
using bingo
cages
or dice.