You are on page 1of 64

Journal of Economic Literature 2009, 47:2, 315372

http:www.aeaweb.org/articles.php?doi=10.1257/jel.47.2.315

Psychology and Economics:


Evidence from the Field
Stefano DellaVigna*

The research in Psychology and Economics (a.k.a. Behavioral Economics) suggests that
individuals deviate from the standard model in three respects: (1) nonstandard prefer-
ences, (2) nonstandard beliefs, and (3) nonstandard decision making. In this paper, I
survey the empirical evidence from the field on these three classes of deviations. The
evidence covers a number of applications, from consumption to finance, from crime to
voting, from charitable giving to labor supply. In the class of nonstandard preferences,
I discuss time preferences (self-control problems), risk preferences (reference depen-
dence), and social preferences. On nonstandard beliefs, I present evidence on overcon-
fidence, on the law of small numbers, and on projection bias. Regarding nonstandard
decision making, I cover framing, limited attention, menu effects, persuasion and
social pressure, and emotions. I also present evidence on how rational actorsfirms,
employers, CEOs, investors, and politiciansrespond to the nonstandard behavior
described in the survey. Finally, I briefly discuss under what conditions experience
and market interactions limit the impact of the nonstandard features.

1. Introduction using the information available, and pro-


cessing this information appropriately.

T he core theory used in economics


builds on a simple but powerful model
of behavior. Individuals make choices
Individuals preferences are assumed to
be time-consistent, affected only by own
payoffs, and independent of the framing of
so as to maximize a utility function, the decision.

*DellaVigna: University of California, Berkeley, and Reis, Uri Simonsohn, Rani Spiegler, Bjarne Steffen, Justin
NBER. I would like to thank Roger Gordon (the edi- Sydnor, Richard Thaler, Jeremy Tobacman, Michael
tor), three exceptionally careful referees, Dan Acland, Urbancic, Ebonya Washington, Kathryn Zeiler, and
Malcolm Baker, Brad Barber, Nicholas Barberis, Dan Jonathan Zinman for useful comments and suggestions.
Benjamin, Saurabh Bhargava, Colin Camerer, David Thomas Barrios, Charles Lin, and Anitha Sivasankaran
Card, Raj Chetty, James Choi, Sanjit Dhami, Constanca provided excellent research assistance. I also want to
Esteves, Ernst Fehr, Shane Frederick, Drew Fudenberg, thank the students of my Psychology and Economics
David Hirshleifer, Eric Johnson, Lawrence F. Katz, Georg graduate class who over the years have helped shape the
Kirchsteiger, Jeffrey Kling, Howard Kunreuther, David ideas in this paper. Finally, I would like to express all my
Laibson, George Loewenstein, Erzo F. P. Luttmer, Rosario gratitude to David Laibson and Matthew Rabin for their
Macera, Ulrike Malmendier, Michel Andre Marechal, exceptional generosity in sharing their insights with the
John Morgan, Ted ODonoghue, Ignacio Palacios-Huerta, next generations of behavioral economists. This paper
Joshua Palmer, Vikram Pathania, Matthew Rabin, Ricardo would not have been possible without them.

315
316 Journal of Economic Literature, Vol. XLVII (June 2009)

Laboratory experiments in both the psy- in each step of the decision-making pro-
chology and the economics literature raise cess: (1) nonstandard preferences, (2) incor-
serious questions about these assumptions, rect beliefs, and (3) systematic biases in
though. In the laboratory, individuals are decision making. For each of these three
time-inconsistent (Richard H. Thaler 1981), steps, I present an example of the labora-
show a concern for the welfare of others (Gary tory evidence, introduce a simple model if
Charness and Matthew Rabin 2002; Ernst available, and summarize the strength and
Fehr and Simon Gchter 2000), and exhibit weaknesses of the field evidence. Since the
an attitude toward risk that depends on fram- focus of the paper is on the field evidence, I
ing and reference points (Daniel Kahneman do not survey the laboratory evidence or the
and Amos Tversky 1979). They violate ratio- theoretical literature.
nal expectations, for example, by overesti- To fix ideas, consider the following styl-
mating their own skills (Colin Camerer and ized version of the standard model, modi-
Dan Lovallo 1999) and overprojecting from fied from Rabin (2002b). Individual i at time
the current state (Daniel Read and Barbara t = 0 maximizes expected utility subject to
van Leeuwen 1998). They use heuristics to a probability distribution p(s) of the states of
solve complex problems (Xavier Gabaix et al. the world s S:
2006) and are affected by transient emotions

t p
(st)U(xit|st).
in their decisions (George Loewenstein and (1) max

Jennifer S. Lerner 2003). t
xi X i t=0 stSt
Unclear from these experiments, though,
is how much these deviations from the stan- The utility function U(x|s) is defined over
dard theory in the laboratory affect eco- the payoff xit of player i and future utility is
nomic decisions in the field. In markets, discounted with a (time-consistent) discount
people hone their behavioral rules to match factor .
the incentives they face and sort into favor- The first class of deviations from the stan-
able economic settings (Steven D. Levitt dard model in equation (1) is nonstandard
and John A. List 2007). This is likely to limit preferences, discussed in section 2. I focus
the impact of deviations from the standard on three dimensions: time preferences, risk
model in markets. However, other forces are preferences, and social preferences. With
likely to increase the impact. Firms often respect to time preferences, the findings on
have incentives to accentuate the deviations self-control problems, for example in retire-
of consumers to profit from them (Stefano ment savings, challenge the assumption that
DellaVigna and Ulrike Malmendier 2004). the discount factor, , is time-consistent. With
In addition, important economic decisions respect to risk preferences, the evidence, for
such as the choice of retirement savings or a example in insurance decisions, suggests that
house purchase are taken seldom, with lim- the utility function U(xi|s) depends on a ref-
ited scope for feedback and sorting. erence point r: the utility function becomes
The objective of this paper is to sum- U(xi|r,s). With respect to social prefer-
marize a growing list of recent papers that ences, the evidence, for example on chari-
document aspects of behavior in market table giving, suggests that the utility function
settings that also deviate from the forecasts depends also on the payoff of other people
of the standard theory. This research area xi: the utility is U(xi,xi|s). The research on
is known as Psychology and Economics (or nonstandard preferences constitutes the bulk
Behavioral Economics). The evidence sug- of the empirical research in Psychology and
gests deviations from the standard theory Economics.
DellaVigna: Psychology and Economics 317

The second class of deviations from the changed their own behavior in ways that
standard model in equation (1) is nonstan- would be puzzling given the standard theory

(s)p(s), reviewed in section 3.
dard beliefs p but that are consistent with utility-maximiz-
Systematic overconfidence about own abil- ing responses to the documented behavioral
ity can help explain managerial behavior of anomalies. This provides indirect support for
CEOs. Non-Bayesian forecasting rationalizes the lab and field evidence of the anomalies.
gamblers fallacy behavior in lotteries and Given this evidence, I expect that the doc-
overinference from past stock returns. The umented deviations from the standard model
overprojection of current tastes on future will be increasingly incorporated in eco-
tastes can explain aspects of the purchase of nomic models. Indeed, features such as time
seasonal items. inconsistency and reference dependence
The third class of deviations from the stan- have become common assumptions. In the
dard model is nonstandard decision mak- concluding section, I present remarks on why
ing, discussed in section 4. For given utility these deviations matter also in the field and
U(x|s) and beliefs p(s), individuals resort to also discuss directions for future research in
heuristics (Tversky and Kahneman 1974) Psychology and Economics.
instead of solving the complex maximiza- This overview differs from other sur-
tion problem (1). They are affected by the veys of Psychology and Economics (Rabin
framing of a decision problem, for example 1998; Rabin 2002b; Sendhil Mullainathan
in investment decisions. They simplify a and Thaler 2001; Camerer 2006) because
complex decision by being inattentive to it focuses on empirical research using non-
less salient features of a problem, from asset laboratory data. A number of caveats are
allocation to purchase decisions. They use in order. First, this paper does not provide
suboptimal heuristics when choosing from a an overview by field of application since it
menu of options Xi, such as for savings plans is organized instead by psychological prin-
or loan terms. They are also subject to social ciples; the interested reader can consult as
pressure and persuasion, for example in their a starting point the book chapters in Peter
workplace performance and in voting deci- Diamond and Hannu Vartiainen (2007).
sions. Finally, they are affected by emotions, Second, the emphasis of the paper is on (rel-
as in the case of investment decisions. atively) detailed summaries of a small num-
While I organize the deviations in three ber of papers for each deviation. As such, the
separate classes, the three types of devia- survey provides a selective coverage of the
tions are often related. For example, persua- field evidence, although it strives to cover all
sion leads to different decisions through the the important deviations.1 Finally, this over-
change in beliefs that it induces. view undersamples studies in Marketing and
The first part of the paper provides evi- provides a partial coverage of the research
dence that these deviations affect the behav- in Behavioral Finance, probably the most
ior of individual decisionmakers, such as developed application of Psychology and
consumers and small investors. But are these Economics, for which a comprehensive sur-
deviations large enough to matter for our the- vey of the empirical findings is available
ories of how markets and institutions work? (Nicholas Barberis and Thaler 2003).
I provide evidence on how rational actors
respond to these behavioral anomalies in sec-
1 This overview does not discuss deviations from the
tion 5. In particular, I discuss the response
standard model that are widely documented in experi-
of firms, employers, managers, investors, ments but not in the field, such as will-power exhaustion
and politicians. These agents appear to have and the availability heuristics.
318 Journal of Economic Literature, Vol. XLVII (June 2009)

2. Nonstandard Preferences job. As the future gets near, the discounting


gets steep, and the individuals engage in binge
2.1 Self-Control Problems eating, light another (last) cigarette, and stay
put on their job. Preferences with these fea-
The standard model (1) assumes a discount tures therefore induce time inconsistency.
factor between any two time periods that is
2.1.2 Model
independent of when the utility is evaluated.
This assumption implies time consistency, David Laibson (1997) and ODonoghue
that is, the decisionmaker has the same pref- and Rabin (1999a) formalized these prefer-
erences about future plans at different points ences using (,) preferences,4 building on
in time.2 R. H. Strotz (1956), E. S. Phelps and R. A.
Pollak (1968), and George A. Akerlof (1991).
2.1.1 Laboratory Experiments
Labeling as ut the per-period utility, the
Experiments on intertemporal choice, sum- overall utility at time t, Ut, is
marized in Loewenstein and Drazen Prelec
(1992) and Shane Frederick, Loewenstein, Ut = ut + ut+1 + 2 ut+2 + 3 ut+3 +
and Ted ODonoghue (2002), have cast doubt
on this assumption. This evidence suggests The only difference from the standard
that discounting is steeper in the immedi- model (with as the discount factor) is the
ate future than in the further future. For parameter 1, capturing the self-control
example, the median subject in Thaler (1981) problems. For < 1, the discounting between
is indifferent between $15 now and $20 in the present and the future is higher than
one month (for an annual discount rate of between any future time periods, captur-
345 percent) and between $15 now and $100 ing the main finding of the experiments. For
in ten years (for an annual discount rate of = 1, this reduces to the standard model.
19 percent).3 The preference for immediate A second key element in this model
gratification captured in these studies appears is the modeling of expectations about
to have identifiable neural underpinnings. future time preferences. ODonoghue and
Intertemporal decisions involving payoffs in Rabin (2001) allow the agent to be par-
the present activate different neural systems tially naive (that is, overconfident) about
than decisions involving only payoffs in future the future self-control problems. A par-
periods (Samuel M. McClure et al. 2004). tially naive (,) agent expects in the future
Intertemporal preferences with these fea- period t + s to have the utility function
tures capture self-control problems. When 2
evaluating outcomes in the distant future, U t+s = ut+s +
ut+s+1 +
ut+s+2
individuals are patient and make plans to 3
exercise, stop smoking, and look for a better +
ut+s+3 + ,

2 Strictly speaking, the standard model merely assumes the future payments can induce seeming present bias; and
time consistency, not a constant discount factor . Still, (3) the discounting should apply to consumption units,
most of the evidence in this sectionthe adoption of rather than to money (in theory, over monetary outcomes,
costly commitments or behavior that differs from the only the interest rate should matter). While none of the
plansdirectly violates time consistency and hence also experiments fully addresses all three issues, the consis-
this more general version of the standard model. tency of the evidence suggests that the phenomenon is
3 The laboratory experiments on time preferences face genuine.
at least three issues: (1) most experiments are over hypo- 4 These preferences are also labeled quasi-hyperbolic
thetical choices, including Thaler (1981); (2) in the experi- preferences, to distinguish them from (pure) hyperbolic
ments with real payments, issues of credibility regarding preferences, and present-biased preferences.
DellaVigna: Psychology and Economics 319


with
. The agent may be sophisticated
investment good (b2 > 0) and too much lei-
about the self-control problem ( = ), fully sure good (b2 < 0). This is the self-control
naive (= 1), or somewhere in between. This problem in action. In response, a sophisti-
model, therefore, combines self-control prob- cated agent looks for commitment devices
lems with a form of overconfidence, naivet to increase the consumption of investment
about future self-control. goods and to reduce the consumption of lei-
Other models have been proposed to cap- sure goods.
ture self-control problems, including axi- Finally, how much does the agent expect
omatic models that emphasize preferences to consume? The agent expects to consume
over choice sets (Faruk Gul and Wolfgang in the future if
Pesendorfer 2001) and models of the conflict
between two systems, a planner and a doer (4) b1 +
b2 0,
(Thaler and Hersh M. Shefrin 1981 and Drew
Fudenberg and David K. Levine 2006, among with
. Compared to the actual con-
others). For lack of space, and since most sumption in (3), the agent overestimates the
applied work has referred to the (,) model, I consumption of the investment good (b2 > 0)
refer only to this latter model in what follows. and underestimates the consumption of the
As an example of how the (,) model leisure good (b2 < 0). Naivet therefore leads
operates, consider a good with immediate to mispredictions of future usage.
payoff (relative to a comparison activity) b1 I now present evidence on the consump-
at t = 1 and delayed payoff b2 at t = 2. An tion of investment goods (exercise and home-
investment good, like exercising or searching works) and leisure goods (credit card take-up
for a job, has the features b1 < 0 and b2 > 0: and life-cycle savings) that can be interpreted
the good requires effort at present and deliv- in light of this simple model.
ers happiness tomorrow. Conversely, a leisure
2.1.3 Exercise
good, like consumption of tempting food or
watching TV, has the features b1 > 0 and DellaVigna and Malmendier (2006) use data
b2 < 0: it provides an immediate reward, at from three U.S. health clubs offering a choice
a future cost. between a monthly contract XM with lump-
How often does the agent want to con- sum fee L of approximately $80 per month and
sume, from an ex ante perspective? If the no payment per visit, and a pay-per-visit con-
agent could set consumption one period tract Xp with fee p of $10. Denote by E(xM)|XM
in advance, at t = 0, she would consume if the expected number of monthly visits under
b1 + 2 b2 0, or the monthly contract XM. Under the standard
model, individuals choosing the monthly
(2) b1 + b2 0. contract must believe that pE(xM)|XM L, or
L/E(xM)|XM p: the price per expected atten-
(Notice that cancels out, since all payoffs dances under the monthly contract should be
are in the future) lower than the fee under payment-per-usage.
How much does the agent actually con- Otherwise, the individual should have chosen
sume at t = 1? The agent consumes if the pay-per-usage treatment. DellaVigna and
Malmendier (2006), however, find that health
(3) b1 + b2 0. club users that choose the monthly contract
XM attend only 4.4 times per month. These
Compared to the desired consumption, users pay $17 per visit even though they could
therefore, a (,) agent consumes too little pay $10 per visit, a puzzle for the standard
320 Journal of Economic Literature, Vol. XLVII (June 2009)

model. A model with partially naive (,) set for weeks prior to the last week, indicat-
members suggests two explanations for this ing a demand for commitment.6
finding. The users may be purchasing a com- This result leaves open two issues. First,
mitment device to exercise more: the monthly do the self-set deadlines improve perfor-
membership reduces the marginal cost of a mance relative to a setting with no deadlines?
visit from $10 to $0, and helps to align actual Second, is the deadline setting optimal?
attendance in (3) with desired attendance in If the individuals are partially naive about
(2). Alternatively, these agents may be overes- the self-control, they will underestimate
timating their future health club attendance, the demand for commitment (equation (4)).
as in (4). Direct survey evidence on expecta- In a second (laboratory) experiment, Ariely
tion of attendance and evidence on contract and Wertenbroch (2002) address both issues.
renewal are most consistent with the latter Sixty students complete three proofreading
interpretation.5 assignments within twenty-one days. The
control group can turn in each assignment at
2.1.4 Homework and Deadlines
any time within the twenty-one days, a first
Dan Ariely and Klaus Wertenbroch (2002) treatment group can choose three deadlines
present evidence on homework completion (as in the class-room setting described above),
and deadlines. The subjects are fifty-one pro- and a second treatment group faces equal-
fessionals enrolled in a section of a semester- spaced deadlines. The first result is that self-
long executive education class at Sloan (MIT), set deadlines indeed improve performance:
with three homeworks as a requirement. At the first treatment group does significantly
the beginning of the semester, they set bind- better than the control group, detecting
ing deadlines (with a cost of lower grades for 50 percent more errors (on average, 105 ver-
delay) for each of the homeworks. According sus 70) and earning substantially more as a
to the standard model, they should set dead- result (on average, $13 versus $5). The sec-
lines for the last day of the semester: there ond result is that the deadline setting is not
is no benefit to setting early deadlines, since optimal: the group with equal-spaced dead-
they do not receive feedback on the home- lines does significantly better than the other
works, and there is a cost of lower flexibil- groups, on average detecting 130 errors and
ity. (A maximization without constraints is earning $20. This provides evidence of par-
always preferable to one with constraints.) tial naivet about the self-control problems.
According to a model of self-control, instead,
2.1.5 Credit Card Take-Up
the deadlines provide a useful commitment
device. Since homework completion is an Lawrence M. Ausubel (1999) provides
investment good (b2 > 0), individuals spend evidence on credit card usage using a large-
less time on it than they wish to ex ante (com- scale field experiment run by a credit card
pare equations (2) and (3)). A deadline forces company. The company mailed randomized
the future self to spend more time on the credit card offers, varying both the preteaser
assignment. The results support the self-con- and the postteaser interest rates. For example,
trol model: 68 percent of the deadlines are compared to an offer of 6.9 percent interest

6 Ariely and Wertenbroch (2002) also compare the per-


5 In section 5, I discuss how the contracts offered by formance in this section to the performance in another
health club companies are consistent with the assumption section with equal-spaced deadlines, with results similar
of naive (,) consumers (DellaVigna and Malmendier to the ones described below. However, the students are
2004). not randomly assigned to the two sections.
DellaVigna: Psychology and Economics 321

rate for six months and 16 percent thereaf- data. Building on Laibson (1997) and George-
ter (the control group), the treatment group Marios Angeletos et al. (2001), Laibson,
Pre received a lower preteaser rate (4.9 per- Andrea Repetto, and Jeremy Tobacman
cent followed by 16 percent); the treatment (2009) estimate a fully specified model of
group Post, instead, received a lower post- life-cycle accumulation with liquid and illiq-
teaser rate (6.9 percent followed by 14 per- uid saving. They show that the (,) model
cent). For each offer, Ausubel (1999) observes can reconcile two facts: high credit card bor-
the response rate and tweny-one months of rowing (11.7 percent of annual income) and
history of borrowing for the individuals that substantial illiquid wealth accumulation (216
take the card. Across these offers, the aver- percent of annual income for the median con-
age balance borrowed in the first six months sumer of age 5059).8 Standard models have
is about $2,000, while the average balance a hard time explaining both facts, since credit
in the subsequent fifteen months is about card borrowing implies high impatience,
$1,000.7 Given these borrowing rates, the which is at odds with substantial wealth accu-
standard theory predicts that the increase mulation. The model with self-control prob-
in response rate for treatment Pre (relative lems predicts high spending on liquid assets,
to the control group) should be smaller than but also a high demand for illiquid assets,
for treatment Post: neglecting compounded which work as commitment devices.
interest, 6/12 2 percent $2,000 is smaller Nava Ashraf, Dean Karlan, and Wesley
than 15/12 2 percent $1,000 (the com- Yin (2006) document directly the demand
parison would only be more favorable for the for illiquid savings as a commitment device,
Post treatment if we could observe the bal- and its effect. They offer an account with a
ances past twenty-one months). Instead, the commitment device to 842 randomly deter-
increase in take-up rate for the Pre treat- mined households in the Philippines with a
ment (386 people out of 100,000) is 2.5 times pre-existent bank account. Access to funds
larger than the increase for the Post treat- in these accounts is constrained to reach-
ment (154 people out of 100,000). Individuals ing a self-specified savings goal or a self-
overrespond to the preteaser interest rate. specified time period. A control group of
Ausubels interpretation of this result is that 466 households from the same sample is
individuals (naively) believe that they will not offered a verbal encouragement to save but
borrow much on a credit card, past the teaser no commitment. The results reveal a sizable
period. These findings are consistent with demand for commitment, and an impact of
underestimation of future consumption for commitment on savings. In the treatment
leisure goods, as in equation (4). group, 202 of 842 households take up the
commitment savings product. In the 842
2.1.6 Life-Cycle Savings treatment households, savings in the bank
after six months are significantly more likely
The (,) model of self-control can also to increase, compared to the 466 control
help explain puzzling features of life-cycle households that received a pure encourage-
accumulation and credit-card borrowing ment: the share of households with increased
savings is 33.3 percent in the treatment and

7 Of course, the differences in interest rates will affect


the borrowing directly, through incentive and selection
effects. However, these differences are small enough in
the data that we can, to a first approximation, neglect 8 The figures (from Laibson et al. 2006) refer to high-
them in these calculations. school graduates.
322 Journal of Economic Literature, Vol. XLVII (June 2009)

27.7 in the control.9 The comparison includes bearing on self-control problems, as I discuss
individuals in the treatment group that do below.10 Brigitte Madrian and Dennis F. Shea
not take up the commitment savings prod- (2001) consider the effect on the contribution
uct; the treatment-on-the-treated estimate rates in 401(k)s of a change in default. Before
is larger by a factor of 842/202. Thaler and the change, the default is nonparticipation
Shlomo Benartzi (2004), described in sec- in retirement savings; after the change, the
tion 5 below, provide evidence of substantial default is participation at a 3 percent rate in
demand for commitment devices in retire- a money market fund. In both cases, employ-
ment savings in the United States. ees can override the default with a phone
Paige Marta Skiba and Tobacman (2008) call or by filing a form; also, in both cases,
examine the role of self-control in the demand contributions receive a 50 percent match up
for payday loans, one of the fastest-growing to 6 percent of compensation. Madrian and
sources of credit in the United States, with Shea (2001) find that the change in default
ten million households borrowing in 2002. has a very large impact: one year after join-
These loans provide (typically) two weeks ing the company, the participation rate in
of liquidity for annualized (compounded) 401(k)s is 86 percent for the treatment group
interest rates of over 7000 percent. Using an and 49 percent for the control group.
administrative data set from a payday lender, James J. Choi et al. (2004) show that these
Skiba and Tobacman (2008) fit a life-cycle findings generalize to six companies in differ-
consumption model to the observed borrow- ent industries with remarkably similar effect
ing behavior. While an exponential time dis- sizes. This finding is not limited to retire-
counting model with a high discount rate can ment choices in the United States. Henrik
rationalize the observed borrowing at a high Cronqvist and Thaler (2004) examine the
interest rate, it has difficulty explaining the choice of retirement funds in Sweden after
relatively low default rate. Defaulting borrow- the privatization of social security in the year
ers have on average already repaid or serviced 2000. They find that 43.3 percent of new
five payday loans, making interest payments participants choose the default plan, despite
of 90 percent of their original loans princi- the fact that the government encouraged
pal. Impatient but time-consistent borrowers individual choice, and despite the availability
would take advantage of default sooner, or of 456 plans. Three years later, after the end
borrow less. The observed facts are most con- of the advertisement campaign encouraging
sistent with a model of partially naive (,) individual choice, the proportion choosing
consumers. These consumers borrow today the default plan increased to 91.6 percent.
expecting to borrow less in the future and Overall, the finding of large default effects is
procrastinate defaulting (which is assumed to one of the most robust results in the applied
have immediate monetary or stigma costs). economics literature of the last ten years.11
What explains the large default effect
2.1.7 Default Effects in 401(k)s for retirement savings? Transaction costs

The impact of the status quo (default) in


10 William Samuelson and Richard Zeckhauser (1988)
retirement savings is the final set of findings
is an early paper documenting default effects.
11 Default effects matter in other decisions, such
as contractual choice in health clubs (DellaVigna and
9 These figures refer to the total bank balance across Malmendier 2006), organ donation (Eric J. Johnson and
all accounts for a household, that is, they are not due Daniel Goldstein 2003; Alberto Abadie and Sebastien
to switches of savings from an ordinary account to the Gay 2006), and car insurance plan choice (Johnson et al.
account with commitment device. 1993).
DellaVigna: Psychology and Economics 323

alone are unlikely to explain default effects. (,) agents do not exhibit large default
Employees can change their retirement effects for reasonable parameter values
decisions at any time using the phone or a (ODonoghue and Rabin 2001). While sophis-
written form. Such small transaction costs ticated agents would like to postpone activi-
are dwarfed by the tax advantages of 401(k) ties with immediate costs, they realize that
investments, particularly in light of the 50 doing an activity now is better than postpon-
percent match (up to 6 percent of compensa- ing it for a long time.12
tion) in place at the Madrian and Shea (2001) If procrastination of a financial transaction
company. At a mean compensation of about is indeed responsible for the default effects
$40,000, the match provides a yearly benefit in Madrian and Shea (2001) and in Choi et
of $1,200, assuming a discount rate equal to al. (2004), we should expect that, if individu-
the interest rate. It is hard to imagine trans- als were forced to make an active choice at
action costs of this size. enrollment, they would display their true
ODonoghue and Rabin (1999b and 2001) preferences for savings. In this case, they
show that self-control problems, combined bear the transaction cost whether they invest
with naivet, can explain the observed or not, and hence investing does not have an
default effect even for small transaction costs. immediate cost, i.e., b1 = 0. In this situation,
Consider a naive (,) agent that has to decide the short-run self does not desire to postpone
when to undertake a decision with immediate the choice. Gabriel D. Carroll et al. (forth-
disutility from transaction costs b1 < 0 and coming) analyze a company that required its
delayed benefit b2 > 0, such as enrolling in employees to choose the retirement savings at
retirement savings. This agent would rather enrollment. Under this Active Decision plan,
postpone this activity, given the self-con- 80 percent of workers enrolled in a 401(k)
trol problems, as in equation (3). Moreover, within one year of joining the company. Later,
this agent is (incorrectly) convinced that if this company switched to a no-investment
she does not do the activity today, shell do default, and the one-year enrollment rate
it tomorrow, as in equation (4). This agent declined to 50 percent. Requiring workers
postpones the activity day-after-day, ending to choose, therefore, produces an enrollment
up never doing it. ODonoghue and Rabin rate that is only slightly lower than under the
(2001) show that, in the presence of naivet, automatic enrollment in Madrian and Shea
even a small degree of self-control problems (2001).13
can generate (infinite) procrastination. These
papers distinguish between procrastina-
tion and delay: procrastination is a delay
12 There are a number of alternative interpretations
that ex ante the agent does not anticipate.
of the observed default effects, such as inattention and
ODonoghue and Rabin (1999b) presents limited memory about 401(k) investment. However, most
calibrations in a deterministic setup, which explanations are unlikely to match the magnitudes of the
DellaVigna and Malmendier (2006) extend to delay. For example, consider a sophisticated agent with
limited memory. Being rational, she is aware that, if she
the case of stochastic transaction costs. Both does not sign up at employment, she will not sign up until
papers show that naive (,) agents accu- the next time she remembers. If the anticipated delay is
mulate substantial delays in a costly activity long enough, the agent prefers to sign up immediately,
hence exhibiting very limited default effects. A model of
(respectively, signing up for a 401(k) and can- limited memory with added naivet explains the findings
celing a health club membership). in a similar way to the naive (,) model.
13 The effect of the Active Decision may also be due
Both self-control problems and naivet
to a deadline effect for naive (,) employees, who know
are required to explain the observed default that the next occasion to enroll will not be until several
effects. Unlike naive agents, sophisticated months later.
324 Journal of Economic Literature, Vol. XLVII (June 2009)

2.1.8 Welfare different facts not just qualitatively, but also


quantitatively. Quantitative calibrations, for
These studies have welfare and policy example, are crucial to understand default
implications. They suggest that savings rates effects. Pursuing this agenda further, a few
for retirement in the United States may be papers have estimated values for the time
low due to a combination of procrastina- preference parameters. Laibson, Repetto,
tion and defaults set to no savings. The (,) and Tobacman (2009) estimate annual time
model implies that the individuals are likely preference parameters ( = .70, = .96) on
to be happier with defaults set to higher sav- life-cycle accumulation data. M. Daniele
ings rates. A change in policy with defaults Paserman (2008), building on DellaVigna
set to automatic enrollment is an example of and Paserman (2005), uses job search data
cautious paternalism (Camerer et al. 2003) to estimate ( = .40, = .99) for low-wage
in that it helps substantially individuals with workers and ( = .89, = .99) for high-wage
self-control problems and inflicts little or workers. Both papers assume sophistica-
no harm on individuals without self-control tion. Skiba and Tobacman (2008) allowfor
problems. These individuals can switch to a partial naivet and estimate ( = .53, =
different savings rate for a low transaction .90, = .45) for the sample of payday loan
cost. In section 5, I present the results of a borrowers.14
plan with automatic enrollment and other
2.2 Reference Dependence
features designed to increase savings (Thaler
and Benartzi 2004). An alternative design The simplest version of the standard model
could be based on the requirement to make as in equation (1) assumes that individuals
an active choice, as in Carroll et al. (forth- maximize a global utility function over life-
coming). Social Security is a commitment time consumption U(x|s).
device to save, albeit one that consumers
2.2.1 Laboratory Experiments
cannot opt out of, and that thus can hurt con-
sumers with no self-control problems. A set of experiments on attitude toward
risk call into question the assumption of a
2.1.9 Summary
global utility function. An example (using
A model of self-control problems with partial hypothetical questions) from Kahneman and
naivet can rationalize a number of findings Tversky (1979) illustrates the point. A group
that are puzzling to the standard exponential of seventy subjects is asked to consider the
model: (1) excessive preference for member- situation: In addition to whatever you own,
ship contracts in health clubs; (2) positive you have been given 1,000. You are now
effect of deadlines on homework grades and asked to choose between A: (1,000, .50),
preference for deadlines; (3) near neglect of and B: (500), where (1,000, .50) indicates a
postteaser interest rates in credit-card take-up; lottery that assigns .50 probability of 1,000
(4) liquid debt and illiquid saving in life-cycle and .50 probability of 0. A different group of
accumulation; (5) demand for illiquid savings sixty-eight subjects is asked to consider: In
as commitment devices; and (6) default effects addition to whatever you own, you have been
in retirement savings and in other settings.
The partially naive (,) model, therefore,
does a good job of explaining qualitative pat- 14 In Paserman (2008) (respectively, Skiba and

terns across a variety of settings involving self- Tobacman 2008), the model is estimated at the weekly
(biweekly) level, so the parameter refers to the one-week
control. A frontier of this research agenda is (two-week) discounting. The parameter is the annual-
to establish whether one model can fit these ized equivalent.
DellaVigna: Psychology and Economics 325

given 2,000. You are now asked to choose The four features of prospect theory are
between C: (1,000, .50), and D: (500). The designed to capture the evidence on risk-
allocations A and C are identical, and so are taking, including risk aversion over gains,
B and D. However, in the first group, only 16 risk-seeking over losses, and contemporane-
percent of the subjects choose A, in contrast ous preference for insurance and gambling.
with 69 percent of subjects choosing C in the It can also capture framing effects as in the
second group. Clearly, framing matters (see example above. Lottery A is evaluated as
also section 4.1). (.5)v(1,000) and hence, given the concavity
Choices in lotteries with real payoffs dis- of v(x) for positive x and given (.5) .5, is
play similar violation of the standard theory. inferior to lottery B, valued v(500). Conversely,
In Fehr and Lorenz Goette (2007), 27 out lottery C is evaluated as (.5)v(1,000) and,
of 42 subjects prefer 0 Swiss francs for sure given the convexity of v(x) for negative x, is
to the lottery (5, p = .5; 8, p = .5). Under preferred to lottery D.
the standard model, this implies an unrea- The large majority of the follow-up lit-
sonably high level of risk aversion (Rabin erature, however, adopts a simplified ver-
2000). A subject that made this choice for sion of prospect theory incorporating only
all wealth levels would also reject the lottery features (1) and (2). The subjects maxi-
(31, p = .5; , p = .5), which offers an infi- mize ipiv(xi|r), where v(x|r) is defined as

nite payout with probability .5. if x r


x r
2.2.2 Model
(5) {
v(x|r) =
(x


r) if
x < r
Kahneman and Tversky (1979), in the sec- where > 1 denotes the loss aversion param-
ond most cited article in economics since eter. Prospect theory, even in the simplified
1970 (E. Han Kim, Adair Morse, and Luigi version of expression (5), can explain the
Zingales 2006), propose a reference-depen- aversion to small risk exhibited experimen-
dent model of preferences that, unlike the tally, as in the example above from Fehr and
standard model, can fit most of the experi- Goette (2007). A prospect-theoretic subject
mental evidence on lottery choice. According evaluates the lottery (5, .5; 8, .5) as .5
to prospect theory, subjects evaluate a lot- (5) + .5 8 = 4 2.5. This subject
tery (y,p;z,1 p) as follows: (p)v(y r) + prefers the status quo for > 8/5. (The
(1 p)v(z r). Prospect theory is char- experimental evidence from Tversky and
acterized by: (1) reference dependencethe Kahneman (1992) suggests 2.25).
value function v is defined over differences A fifth feature of reference-dependent pref-
from a reference point r, instead of over the erences is narrow framing (Barberis, Ming
overall wealth; (2) loss aversionthe value Huang, and Thaler 2006; Rabin and Georg
function v(x) has a kink at the reference point Weizscker forthcoming). According to the
and is steeper for losses (x < 0) than for gains standard economic model, a decisionmaker
(x > 0); (3) diminishing sensitivitythe value offered a gamble integrates the risk induced
function v is concave over gains and convex by the gamble with the other sources of uncer-
over losses, reflecting diminishing sensitiv- tainty she faces. For example, a bike messenger
ity to outcomes further from the reference in Fehr and Goette (2007) offered the lottery
point; and (4) probability weightingthe (5, p = .5; 8, p = .5) should aggregate these
decisionmaker transforms the probabilities risky earnings with the (highly volatile) earn-
with a probability-weighting function (p) ings from the job, fluctuations in the value of
that overweights small probabilities and the assets, etc. to determine how the lottery
underweights large probabilities. affects consumption utility. If she did so, even
326 Journal of Economic Literature, Vol. XLVII (June 2009)

if she evaluated the lottery according to pros- the two valuations should on average be the
pect theory, she would be very likely to accept same. The median WTA of $5.75, however, is
the gamble (counterfactually). The background twice as large as the median WTP of $2.25.
risk implies that the bike messenger is unlikely Since theoretically wealth effects could
to be near the kink at the reference point; as explain this discrepancy, in a different experi-
such, she is affected only to a limited extent by ment Kahneman, Knetsch and Thaler intro-
the loss aversion. Barberis, Huang, and Thaler duce choosers, alongside buyers and sellers.
(2006) formalizes this argument showing that Choosers, who are not endowed with a mug,
even nonexpected utility theories that display choose between a mug and a sum of money;
first-order risk aversion (including prospect the experimenters elicit the price that induces
theory) do not accommodate the observed indifference. Their choice is formally identi-
risk-taking behavior in the laboratory, unless cal to the choice of the sellers (except for the
one assumes narrow framing. fact that the choosers are not endowed with
A decisionmaker with Narrow Framing, the mug); hence, according to the standard
instead, considers each risk in isolation and theory, the sum of money that makes them
evaluates a lottery as if it were the only deter- indifferent should correspond to the WTA of
minant of consumption. Indeed, this assump- sellers. Instead, in this experiment the median
tion is implicit in evaluating the lottery in Fehr WTA for sellers is $7.12, while the price for
and Goette (2007) as .5 (5) + .5 8 (as choosers is $3.12 (and the WTP for buyers is
we did above). Importantly, this assumption is $2.87). The asymmetry between WTA and
routinely used in experimental papers (with- WTP, replicated in a number of studies, has
out explicit reference to narrow framing) to important implications for economics such as
recover consumer preferences from observed low volume of trades in markets and inconsis-
behavior, in that all income from outside the tencies in the elicitation of contingent valua-
experiment is ignored, e.g., Charles A. Holt tions in environmental decisions.
and Susan K. Laury (2002). I discuss further The endowment effect is predicted by a
the role of framing in section 4.1. reference-dependent utility function with
I assume narrow framing in the following loss-aversion > 1. We assume that the
applications to economic phenomena, and dis- subjects do not exhibit loss aversion with
cuss the application of reference dependence respect to money.15 Assume that the utility
also to settings that do not involve risk (such of the subjects is u(1) if they received a mug,
as the endowment effect and labor supply). and u(0) otherwise, with u(1) > u(0). Consider
subjects with a piece-wise linear utility func-
2.2.3 Endowment Effect
tion (5), where the reference point r depends
A finding consistent with prospect theory on whether the subjects were assigned a mug.
and inconsistent with the standard model Subjects with the mug have reference point
is the so-called endowment effect, an asym- r = 1 and assign utility u(1) u(1) = 0 to keep-
metry in willingness to pay (WTP) and will- ing the mug and utility [u(0) u(1)] + pWTA
ingness to accept (WTA). In a laboratory to selling the mug for the sum pWTA. Subjects
experiment, Kahneman, Jack L. Knetsch, without the mug have reference point r = 0 and
and Thaler (1990) randomly allocate mugs
to one group of subjects. They use an incen-
tive-compatible procedure to elicit the WTA 15 Subjects in Kahneman, Knetsch, and Thaler (1990)

for subjects that received the mug, and the do not exhibit the endowment effect with respect to tokens
of monetary value. In the presence of loss aversion also
WTP for subjects that were not allocated with respect to money, the endowment effect is magni-
the mug. According to the standard theory, fied: pWTA = pC = 2 pWTP.
DellaVigna: Psychology and Economics 327

assign value u(1) u(0) pWTP to getting the and Zeiler (2005) do not obtain any endow-
mug at price pWTP and utility u(0) u(0) = 0 to ment effect for mugs, they find a significant
keeping the status quo. The prices that make endowment effect in the lottery rounds, as
both groups of subjects indifferent between Andrea Isoni, Graham Loomes, and Robert
having and not having the mug are Sugden (2008) point out. In Isoni, Loomes,
and Sugden (2008)s replication of the Plott
pWTA = [u(1) u(0)] and and Zeiler (2005) procedure, the mean WTA
for lotteries is larger than the mean WTP by
pWTP = u(1) u(0), a factor of between 1.02 and 2.19, a substan-
tial endowment effect. Isoni, Loomes, and
hence pWTA = pWTP. A loss-aversion param- Sugden (2008) also find some evidence (albeit
eter = 5.75/2.25 fits the evidence in insignificant) of an endowment effect for
Kahneman, Knetsch, and Thaler (1990). mugs (mean WTA over mean WTP of 1.19).
Notice that choosers choose a mug if Additional research will be needed to explain
u(1) u(0) pC, and hence pC = pWTP with the discrepancies between these findings.
referent-dependent preferences, approxi- In any case, the Kahneman, Knetsch, and
mately as observed. Thaler (1990) results are relevant to a num-
Charles R. Plott and Kathryn Zeiler (2005) ber of economic decisions where anonymity
criticize this set of experiments on the ground is not perfect and experience is limited, such
that the endowment effect may be due to pro- as buying or selling a house.
cedural features, such as limited experience A different set of papers in this literature
and lack of anonymity. They elicit the WTP examines the endowment effect in the field,
and WTA for a mug after granting anonymity namely a sports card fair in List (2003 and
and providing practice. In addition, in two of 2004). By selection, the participants in a
three sessions they provide extensive training sports card fair have at least some experi-
consisting of fourteen rounds of elicitation of ence with cards, but some subjects are more
WTA and WTP for lotteries. In contrast to experienced than others.17 The design is as
Kahneman, Knetsch, and Thaler (1990), they follows. List (2003) randomly assigns sports
find no evidence of the endowment effect for memorabilia A or B as compensation for
mugs, with a mean WTA of $5.56 and a mean filling out a questionnaire. After the ques-
WTP of $6.62. Plott and Zeiler interpret this tionnaire is filled out, the participants are
result as suggesting that, once one allows for asked whether they would like to switch
experience and anonymity, the endowment their assigned memorabilia for the other one.
effect disappears. However, an alternative Since the objects are chosen to be of com-
interpretation is that their procedure does not parable value, the standard model predicts
generate an endowment effect because they trade about 50 percent of the time. Instead,
did not vary the endowment sufficiently. subjects with below-average trading experi-
Namely, both WTA and WTP groups are ence switch only 6.8 percent of the time. This
given a mug at the beginning of the experi- provides evidence of the endowment effect
ment (though only the WTA group is told in a natural economic setting where the
that they own it).16 In addition, while Plott
place objects in front of the owners only. (Although the
latter treatment changes the wording too, making it dif-
16 Consistent with this hypothesis, Knetsch and Wei- ficult to disentangle the effects.)
Kang Wong (forthcoming) replicate Plott and Zeiler 17 In List (2003), experienced subjects are those with
(2005)s results if they follow the same endowment self-reported frequency of card trading of at least six times
procedure, but obtain a strong endowment effect if they a month, the mean in the sample.
328 Journal of Economic Literature, Vol. XLVII (June 2009)

subjects are familiar with the objects being the market leads individuals to become
traded and with trading itself. This suggests aware of their loss aversion, and counteract
that familiarity with the object and proce- it: experience mitigates loss aversion. A very
dures is unlikely to be responsible for the dif- different interpretation is that experience
ferences between the results of Kahneman, does not affect loss aversion, but it impacts
Knetsch, and Thaler (1990) and Plott and the reference-point formation. Assume that
Zeiler (2005). experienced traders expect to trade the
List (2003) then examines the behavior object that they are assigned with probabil-
of subjects with high trading experience. ity .5, independent of which group they are
These subject are not only familiar with the assigned to. As in Botond Kszegi and Rabin
objects but also trade cards at least six times (2006), subjects have a stochastic reference
a month. Unlike the less experienced trad- point, r = 1 with probability .5 and r = 0
ers, these subjects switch 46.7 percent of the otherwise. For individuals assigned the good,
time, displaying no endowment effect. The the (expected) value of keeping the good is
difference between the two groups is not .5 [u(1) u(0)] + .5[u(1) u(1)] = .5[u(1)
due to the fact that inexperienced traders are u(0)]; the (expected) value of selling the
approximately indifferent between the two good .5 [u(0) u(0) + pWTA] + .5[(u(0)
memorabilia and, hence, willing to stick to u(1)) + pWTA] = .5 [(u(0) u(1))] + pWTA.
the status quo. In another treatment elicit- This implies pWTA = .5 (1 + )[u(1) u(0)]. It
ing WTA and WTP, the WTA is substantially is easy to show with similar calculation that
larger than the WTP for inexperienced sub-
jects (18.53 versus 3.32), but not for expe- pWTP = .5 (1 +)[u(1) u(0)] = pWTA.
rienced subjects (8.15 versus 6.27). Next,
List (2003) attempts to test whether the dif- If experienced subjects have rational expec-
ference between the two groups is due to tations about their reference point (Kszegi
self-selection of subjects without the endow- and Rabin 2006), they exhibit no endow-
ment effect among the frequent traders or ment effect, even if they are loss averse. The
is a causal effect of trading experience. In a follow-up literature should consider carefully
follow-up study performed months later, the the determination of the reference point.
endowment effect decreases in the trading
2.2.4 Housing Market
experience accumulated in the intervening
months, supporting the latter interpretation. For homeowners who are deciding the
Finally, and most surprisingly, List (2004) sale price of a house, the initial purchase
shows that the more experienced card trad- price is likely to be a salient reference point.
ers also display substantially less endowment Loss aversion induces the homeowners
effect with respect to other goods, such as that would sell at a loss to ask for a higher
chocolates and mugs. sale price. To illustrate this point, consider
Overall, the evidence in List (2003 and a model in which a higher sale price P has
2004) suggests that the endowment effect two effects: (1) it increases the utility of sale
is a feature of trading behavior that market U(P), conditional on a sale, and (2) it lowers
experience can temper.18 This evidence the probability p(P) of a sale (p(P) < 0). The
leaves open (at least) two interpretations. homeowner __ maximizes__ max P p(P)U(P) +
One interpretation is that experience with (1 p(P)) U, where U
is the reservation util-
ity. The optimal price P* equates the mar-
18 I discuss the role of experience further in the ginal gain of increasing the price, p(P)U(P), __
conclusion. to the marginal cost, p(P)(U(P) U).
DellaVigna: Psychology and Economics 329

Consider a piece-wise linear utility as in (5). predicted price. Even after accounting for this
For P P0, the__owner equates__ p(P) and bias, the authors find that the listing price for
p(P)(P P0 U) (notice that U could be units predicted to sell at a loss is significantly
negative and can depend on P0 since we have above the predicted value. The magnitude is
normalized U(P) to equal 0 for P = P0). For large: a 1 percent predicted loss translates into
P < P0, instead, the owner __ compares p(P) a .25 percent higher listing price. This result is
and p(P)((P P0) U). For a standard larger for units owned by individuals than for
risk-neutral agent ( = 1), the two conditions units owned by investor, suggesting a mitigat-
coincide, leading to a solution P*RN. For a loss- ing impact of experience as in List (2003 and
averse agent ( > 1), however, for a price P 2004). The higher listing price translates into
below the reference price P0, the marginal both a longer time on the market and a higher
benefit of a higher price increases discontin- final transaction price (though the latter effect
uously from p(P) to p(P), while the marginal is, as expected, smaller relative to the effect
cost decreases. Both effects imply that, if in on the listing price). These findings are, there-
the risk-neutral case the solution P*RN was to fore, consistent with the reference-dependent
sell at a price lower than P0, the loss-averse model with loss aversion outlined above. The
owner sells at a higher price P*LA > P*RN. Loss authors do not test an additional prediction of
aversion, hence, leads to higher sale prices the model above, bunching of the listing price
for units that would sell at a loss because the at the initial purchase price P0.
agent feels a higher marginal disutility of
2.2.5 Finance
money.
David Genesove and Christopher Mayer Two of the most important applications of
(2001) provide evidence on this phenomenon reference-dependent preferences are to the
using data on sales of Boston condominiums field of finance.19 The first application is to
in the years 199097. The identification is pro- the equity premium puzzle: equity returns
vided by a housing market boom (198387) outperformed bond returns by on average
followed by a slump (198992). This pattern 3.9 percentage points during the period
induces substantial variation in the purchase 18711993 (John Y. Campbell and John H.
price P0 even for otherwise comparable Cochrane 1999), a premium too large to be
units, depending on the year of purchase. reconciled with the standard model, except for
Hence, some of the sellers expect to make extremely high risk aversion (Rajnish Mehra
a loss relative to the original purchase price and Edward C. Prescott 1985). Benartzi and
when selling at the predicted price P , while Thaler (1995) use a calibration20 to show
other sellers expect a gain. The authors then that this premium is instead consistent with
test whether the listing price (relative to the what loss-averse investors require to invest
predicted price) is higher
for units for which in stocks, provided that they evaluate their
the predicted price P falls below the original portfolio performance annually. At horizons
price P0, controlling for the characteristics as short as a year, the likelihood that stocks
of the unit. In doing so, the authors face the underperform relative to bonds requires a
complication that the predicted sale price P substantial compensation in terms of returns,
is noisily estimated. Importantly, unobserv- given loss aversion. At a longer horizon, the
able unit quality would bias the estimates
toward finding the result, since unobservably
19 Barberis and Thaler (2003) present a more compre-
good units are more likely to appear to be sell-
hensive survey of these applications.
ing at a loss (since P0 will be high for these 20 The calibration uses the loss-aversion parameter
units) and are more likely to sell above the estimated from the experiments.
330 Journal of Economic Literature, Vol. XLVII (June 2009)

likelihood of underperformance decreases, losers. Barberis and Wei Xiong (forthcom-


and the implied equity premium decreases. ing), however, point out that this argument
In a paper that carefully formalizes the idea does not take into account the kink at the
of Benartzi and Thaler (1995), Barberis, reference point. This kink induces high local
Huang, and Tano Santos (2001) show that risk-aversion that generates a pressure to sell
reference-dependent preferences can indeed both losers and winners around the ref-
match the observed equity premium. This erence point. Reference-dependent investors
paper uses the simplified prospect-theory take this into account and enter the stock
model with piece-wise linear function as in market only if expected returns are suffi-
(5), relying on reference dependence and loss ciently high. For high expected returns, how-
aversion for the predictions. ever, winners are likely to be further from
The second application is to the so-called the reference point than losers, generating
disposition effect, which denotes the ten- more pressure to sell (due to the closeness
dency to sell winners and hold on to los- to the kink) for losers, contrary to the dis-
ers. Terrance Odean (1998) documents this position effect. Indeed, Barberis and Xiong
phenomenon using individual trading data (forthcoming) simulate a calibrated model of
from a discount brokerage house during the reference-dependent preferences that takes
period 198793. Defining gains and losses these effects into account. They obtain the
relative to the purchase price of a share, disposition effect only for certain ranges
Odean computes the share of realized gains of the parameters and obtain the opposite
PGR = (Realized Gains)/(Realized Gains pattern for most of the parameters. Their
+ Paper Gains) to equal .148. The share of benchmark model assumes that investors,
realized losses PLR = (Realized Losses)/ when evaluating the holdings, make no dis-
(Realized Losses + Paper Losses) equals tinctions between realized gains/losses and
.098. Odean (1998) shows that the large dif- paper gains/losses. Investors, however, may
ference between the propensity to realize treat the two utility carriers asymmetrically
gains (PGR) and the propensity to realize and derive utility (or disutility) only from
losses (PLR) is not due to portfolio rebalanc- realized gains and losses. Investors may even
ing, to ex post higher returns for losers (if go as far as distancing themselves from the
anything, winners outperform losers), or paper losses. Niklas Karlsson, Loewenstein,
to transaction costs. The disposition effect and Duane J. Seppi (2005) show that inves-
is puzzling for the standard theory, since tors are substantially less likely to look up
capital gain taxation would lead to expect their holding on the Internet when the
that investors liquidate losers sooner. stock market is doing poorly. If investors
This puzzle is a robust finding, replicated only evaluate losses when they realize them,
more recently by Zoran Ivkovich, James M. as Barberis and Xiong (forthcoming) show
Poterba, and Scott Weisbenner (2005), who with an extension of their model, reference-
show that the effect is present in both taxable dependent preferences mostly produce the
and tax-deferred accounts (though larger in disposition effect patterns.
tax-deferred accounts).
Prospect theory is a possible explanation for 2.2.6 Labor Supply
this phenomenon. The concavity over gains
induces less risk taking for winner stocks As a further application, we consider the
and, hence, more sales of winners. The con- response of labor supply to wage fluctuations.
vexity over losses induces more risk taking for This response, in general, reflects a complex
loser stocks, and hence more purchases of combination of income and substitution
DellaVigna: Psychology and Economics 331

effects (David Card 1994). Here, we con- daily target r. Any additional dollar earned
sider a simple case in which income effects makes it easier to reach the target and leads
can, to a first approximation, be neglected: to reductions in the number of hours worked
jobs in which workers decide the labor sup- (h* = r/w); this generates a locally downward-
ply daily, and in which the realization of the sloping labor supply function.
daily wage is idiosyncratic. Taxi drivers, for Camerer et al. (1997) use three data sets
example, decide every day whether to drive of hours worked and daily earnings for New
for the whole shift or end earlier; the effec- York cab drivers to test whether the labor
tive wage varies from day-to-day as the supply function is upward sloping, as the
result of demand shifters such as weather standard theory above implies, or downward
and conventions. For these occupations, the sloping. Denote by Yi,t and hi,t the daily earn-
income effect from (uncorrelated) changes ings and the hours worked on day t by driver
in the daily wage is negligible, and we can i. Camerer et al. (1997) estimate the OLS
neglect it by assuming a quasi-linear model. labor-supply equation
Each day, workers maximize the utility func-
tion u(Y) h2 /2, where the daily earning Y (6) log (hi,t) = + log (Yi,t /hi,t)
equals hw, h is the number of hours worked,
w is the daily wage, and h2 /2 is the (convex) + Xi,t + i,t.
cost of effort.
Following the simplified prospect theory Increases in the daily wage, computed as
formulation in (5), we assume that the util- Yi,t / hi,t, lead to decreases in the number
ity function u(Y) equals (Y r) for Y r, of hours worked hi,t with estimated elas-
and (Y r) otherwise, where r is a target ticities in the three data sets of = .186
daily earning. Reference-dependent workers (s.e. .129), .618 (s.e. .051) and .355 (s.e.
( > 1) are loss averse with respect to miss- .051). The authors conclude that the data
ing the daily target earning. For = 1, this reject the standard model (which predicts a
model reduces to the standard model with positive elasticity), and support a reference-
risk-neutral workers. dependent model with daily earnings as the
In the standard model ( = 1), workers reference point. As figure 1 shows, though,
maximize wh h2 /2, yielding an upward- the labor supply function is not necessarily
sloping labor supply curve h* = w/. As the downward sloping for target earners, and
wage increases, so do the hours sup- it is almost certainly not log-linear, unlike
plied, in accordance to the substitution in specification (6). Nevertheless, the find-
effect between leisure and consumption. A ing of a negative elasticity is consistent with
reference-dependent worker ( > 1), instead, reference-dependent preferences for shifts in
exhibits a nonmonotonic labor supply ____ func- labor demand ____corresponding
__ to a wage in the
tion (figure 1). For a low wage (w <r/ ), interval r/ < w < r

.
the worker has not yet achieved the target Specification (6) is open to two main
earnings, and an increase in wage leads to an criticisms. First, a negative elasticity is
increase in hours worked (h* = w/), as in__ the expected if the daily fluctuations in wages for
standard model. For a high wage (w > r ), cab drivers are due to shifters of labor sup-
the worker earns more than the target, and ply (like rain that make driving less pleasant),
the labor supply is similarly upward-sloping, rather than shifters of labor demand. As fig-
albeit flatter (h* = ____
w/). For intermediate
__ lev- ure 2 illustrates, if labor supply shifts across
els of the wage (r/
< w < r
), instead, days, the resulting equilibrium points plot out
the worker is content to earn exactly the a downward-sloping curve even if the labor
332 Journal of Economic Literature, Vol. XLVII (June 2009)

LD

LS
Hours

r
r

Wage

Figure 1. Labor Supply for Reference-Dependent Cab Driver and Market Equilibrium

supply function is upward sloping. Camerer hazard model that does not suffer from divi-
et al. (1997) include controls for plausible sion bias. For any trip t within a day, Farber
labor supply shifters such as weekday and (2005) estimates the probability of stopping
rain; they also use interviews of cab drivers as a function of the number of hours worked
to argue that the factors affecting the wage hi,t and the daily cumulative earnings to that
are unlikely to change the marginal cost of point, Yi,t:
driving; however, in the absence of an instru-
ment for labor supply, this objection is a con- Stopi,t = ( + YYi,t + hhi,t + Xi,t),
cern. Second, specification (6) suffers from
division bias, which biases downward the where is the c.d.f. of a standardized normal
estimate of . Since the daily wage is com- distribution. The standard theory predicts
puted as the ratio of daily earnings and hours that Y should be zero (since earnings are not
worked and since hours worked is the left- highly correlated within a day), while refer-
hand-side variable in (6), any measurement ence dependence predicts that Y should be
error inhi,t induces a mechanical downward positive. Farber (2005) finds that Y is posi-
bias in
. Camerer et al. (1997) address this tive ( Y = .015), but not significantly so.
objection by instrumenting the daily wage While the author cannot reject the standard
of worker i by the summary statistics of the model, the point estimates are not negligible:
daily wage of the other workers on the same a 10 percent increase in Yi,t (about $15) is
shift. The estimates of are still negative, predicted to increase the probability of stop-
though noisier. ping by 15 .015 = .225 percentage points,
Henry S. Farber (2005) uses a differ- a 1.6 percent increase relative to the average
ent data set of 584 trip sheets for twenty- of 14 percentage points. This corresponds to
one New York cab drivers and estimates a an elasticity between earnings and stopping
DellaVigna: Psychology and Economics 333

LD

LS
Hours

Wage

Figure 2. Market Equilibrium for Standard Cab Driver with Shifting Labor Supply

of .16. These findings do not contradict pros- implies a loss-aversion coefficient signifi-
pect theory since Farber (2005) does not cantly larger than zero. At the same time,
test the hypothesis that cab drivers have however, the estimated variation across days
reference-dependent preferences (failing to in the reference daily earning is large enough
reject the null is different from rejecting the that reference dependence loses predictive
alternative hypothesis of prospect theory, power. It is an open question whether more
especially in light of the positive point esti- precise data on the income targets, perhaps
mates). In a more recent paper, Farber (2008) obtained via a survey or induced experimen-
addresses this issue and tests, using the same tally, would help to identify reference-depen-
data set, a simple model of labor supply that dence effects when combined with the labor
explicitly allows for reference-dependent supply decision.
preferences with a stochastic reference point. Given the lack of an instrument for daily
The findings provide weak evidence of ref- wage fluctuations, the evidence on the labor
erence dependence: the estimated model supply of taxi drivers is unlikely to settle the
334 Journal of Economic Literature, Vol. XLVII (June 2009)

debate on reference dependence and labor 2.2.7 Insurance


supply. Fehr and Goette (2007) provide new
evidence using a field experiment on the A puzzling feature of insurance behav-
labor supply of bike messengers. Like taxi ior is the pervasiveness of small-scale insur-
drivers, bike messengers choose how long to ance. Insurance policies on, for example, the
work within a shift. Fehr and Goette (2007) telephone wiring are commonplace despite
randomly assign forty-four messengers into the fact that, in case of an accident, the losses
two groups. Each group receives a 25 per- amount to at most $50 (Charles J. Cicchetti
cent higher commission for the deliveries and Jeffrey A. Dubin 1994). This is a puz-
for just one month in two different months. zle for expected utility, which implies local
This design solves both problems discussed risk-neutrality and hence no demand for small-
above, since the increase in wage is exoge- scale insurance (except in the unrealistic case
nous, and the wage and the actual deliveries of fair pricing). Justin Sydnor (2006) provides
are exactly measured. evidence of excess small-scale insurance for
Bike messengers in the treatment group the $36 billion home insurance industry. Since
respond in two ways to the exogenous (and mortgage companies require home insur-
anticipated) temporary increase in wage: (1) ance, the consumer choice is limited to the
they work 30 percent more shifts; (2) within level of deductible in a standard menu: $250
each shift, they do 6 percent fewer deliv- versus $500 versus $1,000. Using a random
eries. The first finding is consistent with sample of 50,000 members of a major insur-
both the standard model and the reference- ance company in one year, Sydnor documents
dependent model with daily targets. (When that 83 percent of customers and 61 percent
deciding on which day to work, reference- of new customers choose deductibles lower
dependent workers will sign up for shifts on than $1,000. The modal homeowner chooses
days in which it is easier to reach the daily a $500 deductible, thereby paying on aver-
target.) The second finding is consistent age $100 of additional premium relative to a
with target earning and not with the stan- $1,000 deductible. However, the claim rate is
dard model, which predicts an increase in under 5 percent, which implies that the value
the number of hours worked within each of a low deductible is about $25 in expecta-
shift. This finding, however, is also con- tion. The standard homeowner, therefore, is
sistent with an extension of the standard sacrificing $100 $25 = $75 in expectations
model in which workers in the treatment to insure against, at worst, a $500 $100 =
group get more tired, and hence do fewer $400 risk.
deliveries, because they work more shifts. This indicates a strong preference for insur-
Fehr and Goette (2007) provide additional ing against small risks that is a puzzle for the
evidence in support of reference-depen- standard theory, unless one assumes three-
dence combining the field evidence with a digit coefficients of relative risk aversion. This
laboratory experiment on risk-taking. The deviation from the standard model involves
bike messengers that display loss aversion substantial stakes. If, instead of choosing a low
in the labi.e., they reject a (5, .5; 8, .5) deductible, homeowners selected the $1000
lotteryexhibit a more negative response deductible from age 30 to age 65 and invested
(though not significantly so) in their deliv- the money in a money market fund, their
eries to the wage increase. This correlation wealth at retirement would be $6,000 higher.
is not predicted by the fatigue explanation, Sydnor (2006) shows that a calibrated version
but is predicted by the reference depen- of prospect-theory can match the findings by
dence model. the overweighting of the small probability of
DellaVigna: Psychology and Economics 335

an accident and the loss a version with respect Mas (2006) then provides evidence that
to future losses.21 The two components of reference points mediate this effect, which
prospect theory each account for about half depends more on expected wages than on
of the observed discrepancy between the pre- actual wages. Mas generates a predicted
dicted and the observed willingness to pay for award based on a set of observables as a proxy
low deductibles. Social pressure by the sales- for the reference point, and computes how
men (who are paid a percentage of the pre- the clearance rate responds to differences
mium as commission) may also contribute to between the award and the predicted award.
the prevalence of low-deductible contracts. The response is significantly higher for cases
in which the police losesand hence is on
2.2.8 Employment
the loss sidethan for cases in which the
Alexandre Mas (2006) estimates the im- police winsand hence is on the gain side.
pact of reference points for the New Jersey This asymmetry is consistent with reference-
police. In the 9 percent of cases in which the dependent preferences with loss aversion.
police and the municipality do not reach an Assume, for example, that the utility func-
agreement, the contract is determined by tion of the police is [V + v(w|r)]e e2 /2,
final offer arbitration. The police and the where v(w|r) captures the impact of the
municipality submit their offers to the arbi- wage w on the desirability of effort; assume
trator, who has to choose one of the two also that v(w|r) is reference dependent as in
offers. In theory (Mas 2006), if the disput- equation (5). This complementarity between
ing parties are equally risk averse, the winner police pay w and effort e in the utility func-
in arbitration is determined by a coin toss.22 tion can be interpreted as reference-depen-
Mas (2006) exploits this prediction of quasi- dent reciprocity. The first-order condition,
random assignment to present evidence on then, implies e*(w) = [V + v(w|r)]/. Given
how police pay affects performance for 383 loss aversion in v(w|r), this induces a stron-
arbitration cases from 1978 to 1995. In the ger response of the police on the loss side (w
cases in which the offer of the employer is below r) than on the gain side (w above r).
chosen, the share of crimes solved by the
2.2.9 Summary
police (the clearance rate) decreases by 12
percent compared to the cases in which the Reference-dependent preferences help
police offer is chosen. The author also docu- explain (1) excessive aversion to small risks
ments a smaller increase in crime. Lower pay in the laboratory; (2) endowment effect for
therefore induces the police to devote less inexperienced traders; (3) the reluctance
effort to fighting crime, a finding consistent to sell houses at a loss; (4) equity premium
with a number of interpretations, including puzzle in asset returns; (5) (possibly) the ten-
efficiency wages and reciprocity. dency to sell winners rather than losers
in financial markets; (6) (some evidence of)
21 Loss aversion could in principle go the other way,
target earnings in labor supply decisions; (7)
since individuals that are loss averse to paying a high pre- the tendency to insure against small risks;
mium may as well prefer the high deductible. Experimental and (8) effort in the employment relation-
evidence, however, suggests that consumers will adjust ship. I have discussed cases in which the
their reference point on the premium side, since they are
expecting to pay the premium for sure, but cannot adjust evidence is more controversial (labor supply
the reference point on the future uncertain loss. and endowment effect) and cases in which it
22 In reality, the arbitrator rules for the municipality in
is unclear whether reference-dependence is
34.4 percent of cases, suggesting that the unions are more
risk averse than the employers. The key result on refer- an explanation for the phenomenon (dispo-
ence dependence is independent of this assumption. sition effect). I have also discussed how the
336 Journal of Economic Literature, Vol. XLVII (June 2009)

original model in Kahneman and Tversky Forsythe et al. (1994) find that sixty percent
(1979) (and the calibrated version in Tversky of subjects transfer a positive amount. (2)
and Kahneman 1992) is rarely applied in its Gift Exchange games. This experiment (Fehr,
entirety, often appealing just to reference Georg Kirchsteiger, and Arno Riedl 1993) is
dependence and loss-aversion. designed to mirror a labor market. It tests effi-
A key issue in this literature is the deter- ciency wages models according to which the
mination of the reference point r. Often, workers reciprocate a generous wage by work-
different assumptions about the reference ing harder (Akerlof 1982). The first subject (the
point are plausible, which makes the applica- firm) decides a wage w {0,5,10,}. After
tion of the theory difficult. Kszegi and Rabin observing w, the second subject (the worker)
(2006) have proposed a solution. They suggest responds by choosing an effort level e [.1,1].
that the reference point be modeled as the The firm payoff is (126 w)e and the worker
(stochastic) rational-expectations equilibrium payoff is w 26 c(e), with c(e) increas-
of the transaction. In any given situation, this ing and slightly convex. The standard theory
model makes a prediction for the reference predicts that the worker, no matter what the
point, without the need for additional param- firm chooses, exerts the minimal effort and
eters (though there can often be multiple equi- that, in response, the firm offers the lowest
libria, and hence multiple possible reference wage that satisfies the participation constraint
points). This theory also provides a plausible for the workers (w = 30). Fehr, Kirchsteiger,
explanation for some of the puzzles in this and Riedl (1993) instead find that the work-
literature. For example, as I discussed above, ers respond to a higher wage w by providing
it predicts the absence of endowment effect a higher effort e. The firms, anticipating this,
among experienced traders (List 2003), even offer a wage above the market-clearing one
if these traders are loss averse. Experienced (the average w is 72). These results have been
traders expect to trade any item they receive widely replicated and have given rise to a rich
and, hence, their reference point is unaffected literature on social preferences in the labo-
by the initial allocation of objects. ratory, summarized in Charness and Rabin
(2002) and Fehr and Gchter (2000).
2.3 Social Preferences
2.3.2 Model
The standard model, in its starkest form
as in (1), assumes purely self-interested con- Several models have been proposed to
sumers, that is, utility U(xi|s) depends only rationalize the behavior in these experiments.
on own payoff xi. I introduce a simplified version of the social
preference model in Charness and Rabin
2.3.1 Laboratory Experiments
(2002), which builds on the formulation of
A large number of laboratory experiments Fehr and Klaus M. Schmidt (1999).23 In a
calls into question the assumption of pure self-
interest. I present here the results of two clas- 23 In these models, players care about the inequality
sical experiments, which I relate to the field of outcome but not about the intentions of the players
evidence below. (1) Dictator game. In this (although the general model in Charness and Rabin 2002
experiment (Robert Forsythe et al. 1994), a allows for the role of intentions). Another class of mod-
els (including Rabin 1993 and Martin Dufwenberg and
subject (the dictator) has an endowment of $10 Kirchsteiger 2004), based on psychological games, instead
and chooses how much of the $10 to transfer to assumes that subjects care about the intentions that lead
an anonymous partner. While the standard the- to specific outcomes. A common concept is reciprocity
subjects are nice to subjects that are helpful to them but
ory of self-interested consumers predicts that not to subjects that take advantage of them. These models
the dictator would keep the whole endowment, also explain the laboratory findings.
DellaVigna: Psychology and Economics 337

two-player experiment, the utility of subject form of volunteer work were also substan-
1 is defined as a function of their own payoff tial: 44 percent of respondents to a survey
(x1) and other-players payoff (x2): reported giving time to a charitable organiza-
tion in the prior year, with volunteers averag-

{
x2+(1)x1, when x1x2; ing about fifteen hours per month (Andreoni
(7) U1(x1,x2) .
x2+(1)x1, when x1<x2 2006). Altogether, a substantial share of GDP
reflects a concern for others, a finding quali-
The standard model is a special case for tatively consistent with the experimental
= = 0. The case of baseline altruism findings. However, while social preferences
is > 0 and > 0, that is, player 1 cares are a leading interpretation for giving, chari-
positively about player 2, whether 1 is ahead table donations may also be motivated by
or not. In addition, Charness and Rabin other factors, such as desire for status and
(2002) assume > , that is, player 1 cares social pressure by the fund-raisers.
more about player 2 when 1 is ahead. Fehr Even if we take it for granted that giving is
and Schmidt (1999) propose an equivalent an expression of social preferences, it is diffi-
representation of preferences24 and assume cult to use models such as (7) to explain quan-
0 < < 1, like Charness and Rabin (2002), titatively the patterns of giving in the field for
but also <<0. When player 1 is three reasons. (1) These models are designed
behind, therefore, she prefers to lower the to capture the interaction of two players, or
payoff of player 2 (since she is inequality at most a small number of players. Charitable
averse). These two models can explain giving giving instead involves a large number of
in a Dictator Game with a $10 endowment. potential recipients, from local schools in
The utility of giving $5 is higher than the Oakland, CA, to NGOs in Africa. (2) The
utility of giving $0 if 5 max((1 )10,10), utility representation (7) implicitly assumes
that is, if .5 (altruism is high enough, that x1 and x2 include only the experimental
but not so high that a player would transfer payoffs from, say, the dictator game. In the
all the surplus to the opponent.) Fehr and field, it is difficult to determine to what extent
Schmidt (1999) show that model (7) can also x1 and x2 should include, for example, the
rationalize the average behavior in the Gift disposable income. (3) In one-to-one fund-
Exchange game for high enough : altruistic raising situations (hence side-stepping issue
workers provide effort to lower the inequal- 1), models such as (7) overpredict giving.
ity with the firm; the firm, anticipating this, Suppose, for example, that x1 = $1,000 is the
raises w. disposable income of person 1 and x2 = $0 is
the disposable income of person 2, for exam-
2.3.3 Charitable Giving ple, a homeless person. For .5 , the
model predicts that person 1 should trans-
The size of charitable giving is sugges- fer ($1000 $0)/2 = $500, a level of giving
tive of social preferences in the field. In the much higher than 2 percent of GDP. One has
United States, 240.9 billion dollars were to make ad-hoc assumptions on x1 to repro-
donated to charities in 2002, representing an duce the observed level of giving. For these
approximate 2 percent share of GDP (James reasons, while models of social preferences
Andreoni 2006). Donations of time in the are very useful to understand behavior in the
laboratory, they are less directly applicable
24 FehrSchmidt preferences take the form: U ( , )
to the field, compared to models of self-con-
1 1 2
= 1 min ( 2 1, 0) min ( 1 2, 0); they are trol and of reference-dependence. Andreoni
equivalent to the preferences in (7) for = and = . (2006) overviews models that better predict
338 Journal of Economic Literature, Vol. XLVII (June 2009)

patterns of giving, such as models of warm the households of the time of the upcoming
glow. visit. The households in the flyer treatment
There are, however, field settings that respond in a direction consistent with social
resemble more closely the laboratory set- pressure: compared to the control group, the
up. When a fund-raiser contacts a person share of the households opening the door to
directly, the situation resembles a dicta- the solicitors is 10 to 25 percent lower. The
tor game, except for the lack of anonymity. authors then consider the effect on giving.
Field experiments in fund-raising, including There is no effect of a simple flyer on the
List and David Lucking-Reiley (2002) and unconditional share of households that give,
Craig E. Landry et al. (2006), estimate the but a flyer with a Do Not Disturb box low-
effect on giving of variables such as the seed ers the share of households giving by 25 per-
money (the funds raised early on), the use cent. The decrease is entirely due to small
of a lottery, and the identity of the solicitor. donations (up to $10), the ones most likely to
Charitable giving is increasing in the seed be due to social pressure. The share of larger
money (List and Lucking-Reiley 2002) pre- donations (higher than $10) is unaffected by
sumably because of signaling of quality of the either flyer. The results imply a clear role of
charity, and in the attractiveness of female social pressure in door-to-door charitable
solicitors for door-to-door fund-raising, espe- giving, but also provides indirect evidence of
cially for male respondents (Landry et al. altruism for a subset of donors.
2006). The latter result implies that giving
2.3.4 Workplace Relations
in door-to-door fund-raising is not purely the
result of altruism, suggesting a more instru- Workplace relations between employees
mental view of giving. Overall, these field and employer can be upset at the time of
experiments do not answer the key question contract renewal, and workers may respond
of what motivates most giving, genuine social by sabotaging production. Alan B. Krueger
preferences or more instrumental reasons, and Mas (2004) examine the impact of a
such as social pressure. three-year period of labor unrest at a union-
DellaVigna, List, and Malmendier (2009) ized BridgestoneFirestone plant on the
provide direct evidence on this question in quality of the tires produced at the plant.
a door-to-door field experiment designed The workers went on strike in July 1994 and
to distinguish altruism from social pres- were replaced by replacement workers. The
sure. If giving is due to altruism (including union workers were gradually reintegrated in
warm glow as in Andreoni 2006), the donors the plant in May 1995 after the union, run-
derive positive utility from giving. If, instead, ning out of funds, accepted the demands of
giving is due to social pressure, the donors the company. An agreement was not reached
derive negative utility from giving (but still until December 1996. Krueger and Mas
prefer to give rather than incurring the dis- (2004) finds that the tires produced in this
utility cost of saying no). Hence, potential plant in the 199496 years were ten times
donors will seek fund-raisers if giving is due more likely to be defective. The increase in
to altruism, but will avoid them if giving is defects does not appear due to lower quality
due to social pressure. DellaVigna, List, and of the replacement workers. The number of
Malmendier (2009) test this prediction com- defects is higher in the months preceding the
paring a standard door-to-door fund-raising strike (early 1994) and in the period in which
campaign (the control group) to a fund-rais- the union workers and the replacement work-
ing campaign where, the day before a fund- ers work side-by-side (end of 1995 and 1996).
raising visit, a flyer on the doorknob notifies While a bargaining interpretation cannot
DellaVigna: Psychology and Economics 339

be ruled out, this provides some evidence r eciprocity. According to this latter interpre-
that negative reciprocity in response to what tation, the lack of observability of the behav-
workers perceive as unfair treatment can ior of others inhibits not only collusion, but
have a large impact on worker productivity. also reciprocal behavior.
Oriana Bandiera, Iwan Barankay, and
2.3.5 Gift Exchange in the Field
Imran Rasul (2005) test for the impact of
social preferences in the workplace among The Bandiera, Barankay, and Rasul (2005)
employees. They use personnel data from a paper underscores the importance of con-
fruit farm in the United Kingdom and mea- trolling for repeated game effects in tests of
sure changes in the productivity as a func- social preferences. I now consider field exper-
tion of changes in the compensation scheme. iments that test for Gift Exchange control-
In the first eight weeks of the 2002 picking ling for these effects. Field experiments (like
season, the fruit pickers were compensated laboratory experiments) give the researcher
on a relative performance scheme in which more control over the design of an economic
the per-fruit piece rate is decreasing in the situation. Armin Falk (forthcoming) exam-
average productivity. In this system, workers ines the importance of gifts in fund-raising.
that care about others have an incentive to The context is the mailing of 9,846 solicita-
keep the productivity low, given that effort tion letters in Switzerland to raise money
is costly. In the next eight weeks, the com- for schools in Bangladesh. One third of the
pensation scheme switched to a flat piece recipients receives a postcard designed by the
rate per fruit. The switch was announced students of the school, another third receives
on the day the change took place. Bandiera, four such postcards, and the remaining third
Barankay, and Rasul (2005) find that, after receives no postcards. The three mailings are
the change to piece rate, the productivity of otherwise identical, except for the mention
each worker increases by 51.5 percent; the of the postcard as a gift in the two treatment
estimate holds after controlling for worker conditions. The donations are increasing in
fixed effects and is higher for workers with the size of the gifts. Compared to the 12.2
a larger network of friends. The result is not percent frequency of donation in the control
due to a change in incentives: the flat piece group, the frequency is 14.4 percent in the
rate is on average lower than the relative-pay small gift and 20.6 percent in the large-gift
piece rate, which would contribute to lower- treatment. Conditional on a donation, the
ing, rather than increasing, productivity after average amount donated is slightly smaller
the switch. These results can be evidence for in the large-gift treatment, but this effect is
altruism; they can, however, also be evidence small relative to the effect on the frequency
of collusion in a repeated game, especially of donors. The large treatment effects do not
since in the field each worker can monitor appear to affect the donations at next years
the productivity of the other workers. To test solicitation letter, when no gift is sent. A gift,
for these explanations, the authors examine therefore, appears to trigger substantial posi-
the effect of the change in compensation for tive reciprocity, as in the laboratory version
growers of a different fruit where the height of the Gift Exchange.
of the plant makes monitoring among work- Uri Gneezy and List (2006) test the gift
ers difficult. For this other fruit, the authors exchange with two field experiments in
find no impact on productivity of the switch workplace settings. In the first experiment,
to piece rate. This implies that the findings they hire nineteen workers for a six-hour data
are not due to altruism, but rather to collusion entry task at a wage of $12 per hour; in the
or to a different form of social preferences, second experiment, they hire twenty-three
340 Journal of Economic Literature, Vol. XLVII (June 2009)

workers to do door-to-door fund-raising for in a field experiment involving buying a


one weekend at a wage of $10 per hour. In card from a dealer. One group is instructed
both cases, they divide the workers into a to offer $20 for a card of good quality (PSA
control and a treatment group. The control grade 9), while another group is instructed to
group is paid as promised, while the treat- offer $65 for a card of top quality (PSA grade
ment group is told after recruitment that the 10). The quality of the card can be verified by
pay for the task was increased to $20 per hour. an expert but is not apparent on inspection.
The authors test whether the treatment group Dealers that are nonlocal (and hence are
exerts more effort than the control group, as not concerned with reputation) offer cards of
predicted by the gift exchange hypothesis, or the same average quality to the two groups,
the same effort, as predicted by the standard displaying no gift-exchange behavior.25 These
model. The findings are two-fold. At first, the dealers, however, display gift-exchange-type
treatment group exerts substantially more behavior in laboratory experiments designed
effort, consistent with gift exchange: treated to mirror the Fehr, Kirchsteiger, and Riedl
workers log 20 percent more books in the first (1993) experiment. These findings raise
hour and raise 80 percent more money in the interesting questions on when gift-exchange
morning hours. The difference however is behavior does and does not arise. One expla-
short-lived: the performances of control and nation of the findings is that bargaining in
treatment group are indistinguishable after a market setting is construed as a situation
two hours of data entry and after three hours where norms of gift exchange do not apply,
of fund-raising. In these two applications, the possibly because a transfer of $60 is not con-
increase in wage does not pay for itself. These sidered a gift. Hence, the dealers do not
experiments suggest that the gift exchange display such norms when selling cards, but
may have an emotional component that dis- they do instead when participating in an
sipates over time. experiment where the presence of a gift is
Sebastian Kube, Michel Andr Marchal, clearer. More broadly, this suggests that we
and Clemens Puppe (2008b) use a simi- need to understand the economic settings
lar design for a six-hour library work in in which gift-exchange norms apply (such
Germany, with an additional negative gift as charitable giving and, to some extent,
exchange treatment. This group of subjects, employment relationships) and the ones
upon showing up, is notified that the pay where they do not apply (such as market
is 10 per hour, compared to the promised bargaining).
pay of presumably 15 per hour. No one Kube, Marchal, and Puppe (2008a) pro-
decides to quit. This group logs 25 percent vide evidence on the importance of such
fewer books compared to the control group, norms. Within a field experiment along the
a difference that, unlike in the Gneezy and lines of Gneezy and List (2006) and Kube,
List (2006) paper, does not decline over Marchal, and Puppe (2008b), workers
time. The group in the positive gift exchange are hired to catalog books for three hours.
treatment (paid 20) logs 5 percent more Relative to a control group of seventeen
books, an increase which also does not dis- workers hired for the announced hourly
sipate over time. As in the laboratory find- wage of 12, two treatment groups receive
ings, negative reciprocity is stronger than an unexpected gift: the first group of sixteen
positive reciprocity.
List (2006) presents evidence that not 25 Dealers that are local, that is, that attend the fair
everyone reciprocates a generous transfer. frequently, offer higher-quality card to the $65 group, pre-
Attendees of a sports card fair participate sumably because of reputation building.
DellaVigna: Psychology and Economics 341

students receives a 7 (20 percent) wage linking the findings in the laboratory, which
increase, while the second group of fifteen allows the most control on the design, to the
students receives a thermos bottle worth evidence in the field; the recent literature
7. This design is motivated by evidence on on Gift Exchange is a good example. A sepa-
gift perceptions: subjects in an online sur- rate issue is the difficulty of distinguishing
vey presented with the experimental design in the field social preferences from repeated
perceive the employer as kinder when game strategies (as in Bandiera, Barankay,
the gift is a thermos compared to money. and Rasul 2005) and other alternative expla-
The worker effort is consistent with gift nations. For example, social pressure (as in
exchange. Compared to the control group, DellaVigna, Malmendier, and List 2009; see
productivity is 30 percent higher in the also section 4.4) can explain regularities in
thermos group but only 6 percent higher giving, such as the higher effectiveness of
in the money group. Interestingly, in the high-pressure fund-raising methods (such
thermos group the relative increase in pro- as phone calls) relative to low-pressure ones
ductivity is larger than the increase in labor (such as mailings). Creative field experiments
costs, suggesting that, unlike in Gneezy and such as those in this Section can be designed
List (2006), gifts can pay for themselves to distinguish different explanations.
(although, since the market value of the task
to the library is not clear, one cannot say for
3. Nonstandard Beliefs
sure). The gift exchange response does not
simply depend on the perceived economic The standard model in (1) assumes that
value of the gift: in a separate experiment, consumers are on average correct about the
172 subjects offered the choice of a 7 pay- distribution of the states p(st). Experiments
ment or the thermos overwhelmingly prefer suggest instead that consumers have system-
the monetary payment. Future research atically incorrect beliefs p(st) in at least three
will need to provide more evidence on the ways: (1) Overconfidence. Consumers over-
psychology of gift-giving, as well as models estimate their performance in tasks requir-
of it. ing ability, including the precision of their
information; (2) Law of Small Numbers.
2.3.6 Summary Consumers expect small samples to exhibit
large-sample statistical properties; and (3)
Social preferences help explain (1) giving Projection Bias. Consumers project their
to charities; (2) the response of striking work- current preferences onto future periods.
ers to wage cuts; (3) the response of giving to
3.1 Overconfidence
gifts in fund-raisers; (4) the response of effort
to unanticipated changes in pay, at least in Surveys and laboratory experiments pres-
the short run; and (5) the response of effort ent evidence of overconfidence about ability.
to nonmonetary gifts. However, the research In Ola Svenson (1981), 93 percent of subjects
on social preferences displays more imbal- rated their driving skill as above the median,
ance between laboratory and field, compared compared to the other subjects in the experi-
to the research on self-control and on refer- ment.26 Most individuals underestimate the
ence dependence. The models of social pref-
erences that match the laboratory findings
26 This finding admits alternative interpretations, such
are not easily applicable to the field, over-
as that each individual may define driving ability in a self-
predicting, for example, the amount of giv- serving way. These interpretations, however, are addressed
ing. It will be important to see more papers in the follow-up literature.
342 Journal of Economic Literature, Vol. XLVII (June 2009)

probability of negative events such as hos- o verpay for mergers. To test these hypothe-
pitalization (Neil D. Weinstein 1980) and ses, Malmendier and Tate identify a proxy for
the time needed to finish a project (Roger overconfidence, and examine the correlation
Buehler, Dale Griffin, and Michael Ross of this proxy with corporate behavior. In
1994). In Camerer and Lovallo (1999), sub- particular, they identify as overconfident
jects play multiple rounds of an entry game CEOs who hold on to their stock options
in which only the top c out of n entrants until expiration, despite the fact that most
make positive profits. In the luck treatment, CEOs are heavily underdiversified. They
the top c subjects are determined by luck, interpret the lack of exercise as overestima-
while in the skill treatment the top c subjects tion of future performance of their company.
are determined by ability in solving a puzzle. In Malmendier and Tate (2008), they find
More subjects enter in the skill treatment that these CEOs are 55 percent more likely
than in the luck treatment, indicating that to undertake a merger, and particularly so if
subjects overestimate their (relative) ability they can finance the deal with internal funds.
to solve puzzles.27 (Overconfident CEOs are averse to seeking
The first example of overconfidence in the external financing, since they deem it over-
field is the naivet about future self-control priced.) Of course, the correlation between
by consumers in the choices of health club option exercise and corporate behavior
contracts, credit cards, and 401(k) plans, could be due to alternative reasons, such as
documented in section 2.1 (DellaVigna and insider information of the CEO. However,
Malmendier 2006; Ausubel 1999; Madrian Malmendier and Tate show that insider
and Shea 2001). Naivet is an example of information does not appear to be the expla-
overconfidence since self-control is a desir- nation, since the CEOs that delay exercising
able ability. stock options do not on average gain money
In a second example, Malmendier and by doing so. Managerial overconfidence pro-
Geoffrey Tate (2005, 2008) provide evidence vides one explanation for the underperfor-
on overconfidence by CEOs about their abil- mance of companies undertaking mergers.
ity to manage a company. They assume that Malmendier and Tate (2005) use the same
CEOs are likely to overestimate their ability proxies to show that overconfidence explains
to pick successful projects and to run com- in part the excess sensitivity of corporate
panies. As such, these top managers are investment to the availability of cash flows, a
likely to invest in too many projects and to long-standing puzzle in corporate finance.
Overconfidence about own-company
performance likely extends to rank-and-
27 A more recent literature including, among others,
file employees. Bo Cowgill, Justin Wolfers,
Justin Kruger (1999), suggests that, while overconfidence
is typical for easy tasks such as driving, underconfidence and Eric Zitzewitz (2008) study the predic-
can arise for hard tasks such as playing the piano, a dichot- tion markets that Google set up for its own
omy known as the hardeasy effect. As Kruger (1999) employees (with real payoffs). While securi-
suggests, the subjects, when comparing their skills to the
skills of others, do not appreciate that others similarly ties not related to Google are correctly priced
find these tasks respectively easy and hard. This results on average, the securities with implications
in overconfidence for easy tasks and underconfidence for for Google display substantial overconfi-
hard task. A difficulty in applying the hardeasy effect to
economics is the practical definition of a hard task. While dence: in two-outcome markets, the share
running a company is arguably one of the hardest tasks that pays one dollar if the favorable outcome
one can imagine, it likely seems easy to a CEO that spends for Google occurs trades at 45.6 cents, while
most of his or her time doing it. Sorting implies that eco-
nomic agents would mostly face tasks that they deem easy, the average payoff is only 19.9 cents. While
making the underconfidence result less relevant. this evidence is specific to Google, survey
DellaVigna: Psychology and Economics 343

evidence suggests that this phenomenon is 1,000 cases! The elicitation of 75 percent
more general. Indeed, overconfidence of confidence intervals provides similar evi-
employees about own-company performance dence of overconfidence.
is a leading explanation for the provision of Odean (1999) provides field evidence
stock options to rank-and-file employees consistent with this form of overconfidence
(Paul Oyer and Scott Schaefer 2005; Nittai using data from a discount broker on all the
K. Bergman and Dirk Jenter 2007). Stock trades of 10,000 individual investors for the
options have become a common form of com- years 198793. If the investors overestimate
pensation: the (Black and Scholes) value of the precision of their information about indi-
options granted yearly to employees in public vidual companies, they will trade too much.
companies was over $400 (about one percent Indeed, the investors trade on average 1.3
of compensation) in 1999 (Oyer and Schaefer times per year, with a commission cost for
2005). Incentive effects are unlikely to buying or for selling a security of over 2 per-
explain the issuance, given that the contribu- cent per transaction. In addition to these
tion of each individual employee to firm value substantial transaction costs, the individual
is very limited. Instead, overconfidence about investors pay a return cost to trade since
own-company performance can make stock the stocks sold overperform the purchases
options an attractive compensation format for by about 3 percent over the next year. For
employers.28 This form of overconfidence is individual investors, therefore, overconfi-
particularly plausible since the workers that dence has a substantial impact on returns.
are overconfident about a particular company Interestingly, there is a gender differential in
are more likely to sort into it. overtrading that is consistent with the psy-
A third example of overconfidence is the chology findings, which suggests that men
tendency to overestimate the precision of are more overconfident than women about
own information, which is also a skill. For financial decisions: men trade 45 percent
example, Marc Alpert and Howard Raiffa more than women, and hence pay a larger
(1982) ask a group of 100 MBA students returns cost. (Brad M. Barber and Odean
to provide answers for ten numeric queries 2001).
such as the number of foreign automobiles Overconfidence about the precision of
imported into the United States in 1967 in private information also helps explain other
thousands and the total egg production in anomalies in financial markets, such as short-
millions in the United States in 1965. The term positive correlation of returns (momen-
students are also asked for 98 percent con- tum) and long-term negative correlation
fidence intervals. If the students estimated (long-term reversal) (Kent Daniel, David
correctly the precision of their information, Hirshleifer, and Avanidhar Subrahmanyam
their confidence intervals should contain the 1998). To explain these phenomena, overcon-
correct answer in approximately 980 of the fidence needs to be coupled with self-attribu-
1,000 responses. Instead, the intervals con- tion bias, which is the tendency to discount
tain the correct answer in only 574 of the information that is inconsistent with ones
priors. Overconfidence induces individuals
28 Bergman and Jenter (2007) point out that employees
to trade excessively in response to private
can also purchase shares on the open market, and hence information; in the long run, the public infor-
do not need to rely on the company providing them. They mation prevails and the valuation returns to
examine the conditions under which the company will still fundamentals, inducing a long-term reversal.
offer options to overconfident employees, and provide evi-
dence that option compensation is used most intensively The self-attribution bias is responsible for
when employees are more likely to be overconfident. momentum: in the short term, as investors
344 Journal of Economic Literature, Vol. XLVII (June 2009)

receive additional private information, they If the distribution of the signals is known,
interpret as more informative the informa- this induces a gamblers fallacy belief:
tion that conforms to their beliefs and, hence, after a draw of a signal, subjects expect the
become even more overconfident. I discuss next draw to be a different signal (since the
how the law of small numbers and limited draw is without replacement). For example,
attention provide alternative explanations for suppose that the return to a mutual fund is
these same financial markets anomalies in drawn from an urn with 10 balls, 5 Up and 5
sections 3.2 and 4.2. Down, with replacement. After two draws of
Up, a rational investor expects the probability
3.1.1 Summary
of another Up to be .5. However, a believer
Overconfidence helps explain (1) patterns in the law of small number computes such
in health club contract choice, credit card probability as 3/8 < .5, since two balls Up
take-up, and default effects, presented in sec- have already been drawn. This is an example
tion 2.1 (overconfidence about self-control); of the representativeness heuristics, in that
(2) value-destroying mergers and investment- the sequence Up, Up, Down is judged as
cash-flow sensitivity (overconfidence about more representative than the sequence Up,
managerial ability); (3) stock option compen- Up, Up.
sation packages for rank-and-file employees Dek Terrell (1994) provides field evidence
(overconfidence about own company perfor- of the gamblers fallacy in New Jerseys
mance); and (4) excess trading, momentum, pick-three-numbers game. The lottery is a
and long-term reversal (overconfidence about pari-mutuel betting system: the fewer indi-
precision of information). These applications viduals bet on a number, the higher is the
are settings in which overconfidence is par- expected payout. Terrell (1994) finds that
ticularly likely according to the laboratory the payout for a number that won one or
evidence: overconfidence is more common two weeks before is 33 percent higher than
when feedback is noisy (i.e., for stock returns) for an average number. Belief in gamblers
and the decisionmaker has an illusion of con- fallacy leads lottery players to bet less on
trol (i.e., for managers). numbers that won recently, at the cost of a
lower expected payoff. This pattern is found
3.2 Law of Small Numbers
in a number of other betting markets, includ-
Overconfidence is only one form of non- ing the Maryland daily-numbers lottery
Bayesian beliefs detected in experiments. (Charles T. Clotfelter and Philip J. Cook
Tversky and Kahneman (1974) describe a 1993). It is likely to apply also to other situ-
number of deviations from rational updating, ations in which the probabilities are known,
including the overweighting of information but subjects misconstrue the i.i.d. nature of
that is available and representative. I focus the draws. An example is the forecast of the
on two phenomenagamblers fallacy and gender of a third child following two boys
overinferencethat are examples of reliance (or two girls).
on the availability and representativeness The model in Rabin (2002a) delivers a
heuristics. To illustrate these phenomena, second testable prediction. In the case of
I use Rabin (2002a)s model of the law of uncertain distribution of signals, the subjects
small numbers. Rabin (2002a) assumes that overinfer from a sequence of signals of one
subjects, observing a sequence of signals type that the next signal will be of the same
drawn from an i.i.d. process, believe (incor- type. While this overinference appears to
rectly) that the signals are drawn from an be the opposite of the gamblers fallacy, it
urn of size N < without replacement. is a complementary phenomenon. Consider
DellaVigna: Psychology and Economics 345

a mutual fund with a manager of uncertain that individual U.S. investors purchase
ability. The return is drawn with replace- stocks with high past returns, also consistent
ment from an urn with 10 balls. With prob- with overinference. The average stock that
ability .5 the fund is well managed (7 balls individual investors purchase outperformed
Up and 3 Down) and with probability .5 the the stock market in the previous three years
fund is poorly managed (3 Up and 7 Down). by over 60 percent. (Interestingly, the aver-
After observing the sequence Up, Up, age stock sold also outperformed the stock
Up, a rational investor computes the prob- market, though by a smaller amount, con-
ability that the mutual fund is well-managed sistent with either a belief in gamblers fal-
as P(Well|UUU) = .5P(UUU|Well)/[.5P lacy or with the disposition effect (section
(UUU|Well) + .5P(UUU|Poor)] = .7 3/(.7 3 2.2).)
+ .33) .927. A Law-of-Small-Number Overinference in stock holdings can also
investor also applies Bayes Rule but has induce predictability in asset returns. To
the wrong model for P(UUU|Well) and the extent that investors overinfer from past
P(UUU|Poor). Hence, her forecasted proba- performance, stocks with high past returns
bility for P(Well|UUU) equals (7/10 6/9 should get overpriced, and ultimately under-
5/8)/[7/10 6/9 5/8) + (3/10 2/9 1/8)] perform. Werner F. M. De Bondt and Thaler
.972. Hence, this investor over infers about (1985) compare stocks that performed par-
the ability of the mutual-fund manager after ticularly well in the past three years (win-
three good performances. Assume now that ners) to stocks that did poorly in the past
the Law-of-Small-Number investor believes three years (losers). The winners under-
that the urn is replenished after three peri- perform the losers by 25 percentage points
ods. When forecasting the performance in over the next three years, again consistent
the next period, the rational investor expects with overinference.
an Up performance with probability .927 Barberis, Andrei Shleifer, and Robert
.7 + (1 .927) .3 .671, while the Law- Vishny (1998) apply an alternative model of
of-Small-Number investor expects Up with the law of small number to financial markets.
probability .972 .7 + (1 .972) .3.689, While the draws are i.i.d., investors believe
which is higher. that the draws come from either a mean-
Benartzi (2001) provides field evidence reverting regime or a trending regime; in
of overinference (also called extrapolation): addition, the investors believe that the first
the degree to which employees invest in regime is more likely ex ante. If investors
employer stock depends strongly on the past observe a sequence of identical signals, in
performance of the stock. In companies in the short run they expect a mean-reverting
the bottom quintile of performance in the regime (the gamblers fallacy); hence, the
past ten years, 10.4 percent of employee returns underreact to information, inducing
savings are allocated to employer stock, short-term positive correlation (momentum).
compared to 39.7 percent for companies in However, after a longer sequence, the indi-
the top quintile. This difference does not viduals overinfer, as in Rabin (2002a), and
reflect information about future returns. expect a trending regime; this induces a
Companies with a higher fraction of employ- long-term negative correlation of returns.
ees investing in employer stock underper- Hence, the law of small numbers can explain
form over the next year relative to companies two key features of observed returns, short-
with a lower fraction. term positive correlation and long-term neg-
Barber, Odean, and Ning Zhu (forthcom- ative correlation (see also the discussion in
ing) use data on individual trades to show section 3.1).
346 Journal of Economic Literature, Vol. XLVII (June 2009)

3.3 Projection Bias (c,s), an individual with pro-


future utility u
jection bias expects utility
A third way in which individuals have
systematically incorrect beliefs is that they (c,s) = (1 )u(c,s) + u(c,s)
(8) u
expect their future preferences to be too
close to the present ones; for example, they rather than u(c,s). The parameter [0,1]
project current hunger levels on the future. captures the extent of projection bias, with
Read and van Leeuwen (1998) asked office = 0 denoting the standard case and = 1
workers to choose a healthy snack or an the case of full projection bias. This model
unhealthy snack to be delivered a week later can capture the misprediction of future
(in the late afternoon). Workers were asked hunger, as well as the underappreciation of
either when they were plausibly hungry (in adaptation.
the late afternoon) or when satiated (after Michael Conlin, ODonoghue, and Tim-
lunch). In the first group, 78 percent chose othy J. Vogelsang (2007) present evidence of
an unhealthy snack, compared to 42 percent projection bias using a data set of two million
in the second group. orders of cold-weather apparel items. They
Similarly, individuals underappreciate consider the effect of weather at the time of
the extent to which they adapt to future cir- purchase on the probability that an item is
cumstances. Daniel T. Gilbert et al. (1998) returned, conditional on purchase. According
ask subjects to forecast their happiness in to the standard model, colder weather at the
correspondence of an event, and compare time of purchase should not affect the prob-
these responses to the responses after the ability of a return, or may affect it negatively
event has occurred. Thirty-three current (since colder weather at the time of purchase
assistant professors at the University of is correlated with colder weather over the
Texas forecast that getting tenure would subsequent days). Projection bias, instead,
significantly improve their happiness (5.9 makes the opposite prediction. On colder
versus 3.4 on a 17 scale). However, the days, individuals overestimate the use that
difference in rated happiness between they will make of a cold-weather item, and
forty-seven assistant professors that were hence are ex post more likely to return the
awarded tenure by the same university and item.29 This prediction holds whether the
twenty that were denied tenure is smaller projection bias regards future utility, as in
and not significant (5.2 versus 4.7). Similar (8) (I expect to like cold-weather items very
results apply for happiness forecasts as a much), or future weather (I expect the
function of the election of a Democratic coming winter to be very cold).
or Republican president, compared to the Conlin, ODonoghue, and Vogelsang
realized ex post differences. While these (2007) find that a reduction in the order-
are just survey responses (below I provide date temperature of 30Fcorresponding
evidence of impact on behavior), they sug- to a decrease, for example, from 40F to
gest a consistent pattern of projection of the 10Fincreases the average return rate
current preferences. of a cold-weather item by 3.96 percent,
Loewenstein, ODonoghue, and Rabin
(2003) propose a simple model of projection
bias. Assume that utility u is a function of 29 A possible confound is that on colder days more

consumption c and of a state variable s, that marginal individuals (which are more likely to return)
is, u = u(c,s). The current state is s and the order cold-weather clothing. The standard model modi-
fied for this form of heterogeneity makes the same predic-
future state is s. Then, when predicting the tion as the projection bias model.
DellaVigna: Psychology and Economics 347

consistent with projection bias. A simple 2.50. Over a quarter of the subjects, there-
structural model of projection bias as in (8) fore, choose a dominated lottery when the
implies estimates for
0.5, implying that
choice is presented with a narrow framing,
consumers predict future tastes roughly half- that is, with each lottery presented individu-
way between present tastes and actual future ally. A separate group of 45 subjects is pre-
tastes. sented the same choice in the broad framing,
that is, they are shown the distribution of
outcomes induced by the four options. In this
4. Nonstandard Decision Making
group, not surprisingly, none of the subjects
Even given utility U(x|s) and belief p(s), choose the A and D combination. Clearly, the
individuals make nonstandard decisions. framing of choices matters.
I analyze (1) the impact of framing of a We can understand this first example of
decision; (2) the underweighting (or over- framing effects in light of a reference-depen-
weighting) of information because of lim- dent utility function (section 2.2) with nar-
ited attention; (3) suboptimal heuristics row framing. Individuals evaluate each of
used for choices out of menu sets; (4) social the two lotteries separately, comparing the
pressureexplicit pressure by othersand outcomes relative to a reference point. The
persuasionexcess impact of the beliefs of individuals are approximately risk neutral
others; and (5) emotions. over gains, inducing the 49 percent choos-
ing A over B, and risk seeking over losses,
4.1 Framing
hence the 68 percent choosing D over C.30
A key tenet of psychology is that the context Importantly, the individuals accept the fram-
and the framing of a situation matter. Two ing induced by the experimenter and do not
equivalent decision problems that are framed aggregate the lotteries, that is, they frame
differently may elicit different responses. narrowly. This example illustrates a general
Tversky and Kahneman (1981) present a clas- feature of human decisions: judgments are
sical example, which I reproduce in the ver- comparative, and changes in the framing can
sion of Rabin and Weizscker (forthcoming). affect a decision if they change the nature of
A group of subjects is asked to consider a pair the comparison, even if they do not affect the
of concurrent decisions. [ . . . ] Decision 1. underlying economic trade-offs. While there
Choose between: A. a sure gain of 2.40 and is no field evidence directly corresponding
B. a 25 percent chance to gain 10.00 and a to this framing manipulation, I discussed
75 percent chance to gain 0.00. Decision several applications to the field of reference-
2. Choose between: C. a sure loss of 7.50 dependent preferences in section 2.2.
and D. a 75 percent chance to lose 10.00 A second example illustrates how the pre-
and a 25 percent chance to lose 0.00. Of sentation format can affect preferences (in
53 participants playing this lottery for money, this case about financial options) even aside
49 percent choose A over B and 68 percent from its impact on reference points. Benartzi
choose D over C. Overall, 28 percent of the and Thaler (2002) survey 157 UCLA employ-
subjects choose the combination of A and D. ees that participate in a 403(b) plan and ask
This combined lottery, which amounts to a them to rate three plans (labeled plans A,
75 percent chance to lose 7.60 and a 25
percent chance to gain 2.40, however, is
30 Other versions of this experiment (typically for hypo-
dominated by the combined lottery of B and
thetical stakes) indicate a higher percentage of subjects
C, which reduces to a 75 percent chance to choosing A over B, consistent with risk aversion over gains,
lose 7.50 and a 25 percent chance to gain as in prospect theory.
348 Journal of Economic Literature, Vol. XLVII (June 2009)

B, and C): their own portfolio, the average A random subsample of H&R Block cus-
portfolio, and the median portfolio. For each tomers are offered either no match, a 20
portfolio, they present the 5th, 50th, and percent, or a 50 percent match on the first
95th percentile of the projected retirement $1,000 contributed to an IRA. The take-up
income from the portfolio (obtained using rate increases from 2.9 percent in the con-
the Financial Engines retirement calcula- trol group to 7.7 percent in the 20 percent
tor). Given revealed preferences, one would match group and to 14.0 percent in the 50
expect individuals on average to prefer their percent match group. The authors then com-
own plan to the other plans. However, the pare this substantial increase to the response
own portfolio rating (3.07) is about the to a comparable match induced by tax cred-
same as the average portfolio rating (3.05) its in the Savers Tax Credit program. The
and substantially lower than the median effective match rate for IRA contributions
portfolio rating (3.86). Indeed, 62 percent decreases from 100 percent to 25 percent
of employees gave a higher rating to the at the $30,000 household income thresh-
median portfolio than to their own portfolio. old. Duflo et al. (2006) compare the IRA
Re-framing the decision in terms of ultimate participation for households slightly below
outcomes, therefore, appears to affect pref- the threshold ($27,500$30,000) to house-
erence substantially. However, an alternative holds slightly above the threshold ($30,000
interpretation is that these employees never $32,500). To control for other differences
considered the median portfolio in their between these two income groups, they esti-
retirement savings decision, and would have mate the difference-in-difference relative to
chosen it had it been offered. To address households in the same income groups that
this explanation, Benartzi and Thaler (2002) are however ineligible for the program. The
survey 351 participants in a different retire- difference in match rate lowers contribu-
ment plan who were explicitly offered a cus- tions by only 1.3 percentage points, a much
tomized portfolio and actively opted out of smaller impact relative to the effects in the
it. These employees rate their own portfolio, H&R Block field experiment. While there
the average portfolio, and the customized are a number of differences between the
portfolio, similarly reframed in terms of ulti- programs, a prominent factor is likely to be
mate income. A majority (61 percent) of the the simplicity of the H&R Block match that
employees prefers the customized portfolio garnered more attention to the match. The
(which they previously turned down) to their next section presents further evidence about
own portfolio. The choice of retirement sav- the impact of limited attention on econom-
ings, hence, depends substantially on the for- ics decisions. This example illustrates the
mat of the choices presented. This framing importance of considering behavioral fac-
effect presumably reflects the fact that con- tors such as framing in the design of public
sumers put too little weight on factors that policy programs.
determine ultimate returns, such as fees, or
that they do not appreciate the riskiness of 4.2 Limited Attention
their investments.
A third example is the case in which the In the starkest form of the standard model,
framing focuses the attention on differ- individuals make decisions using all the avail-
ent aspects of the options. Esther Duflo et able information. Since Herbert A. Simon
al. (2006) estimate the impact of a match (1955), economists have attempted to relax
on IRA participation for low- and middle- this strong assumption and have proposed
income households using a field experiment. models in which individuals simplify complex
DellaVigna: Psychology and Economics 349

decisions, for example by processing only a salient signal:(1,N) = 0. The consumers


subset of information.31 In economic experi- demand is D[V ], with D[x] > 0 for all x.
ments, the simplifying heuristics include This model suggests, broadly speaking,
thinking only one step ahead in dynamic three strategies to identify the inattention
problems (Gabaix et al. 2006). parameter , which the papers described
The laboratory studies in psychology indi- below undertake. The first is to compute how
cate that attention is a limited resource. In the valuationV responds to a change in o; the
studies of dichotic listening (Donald E. derivative V/o = (1 ) can be compared

Broadbent 1958), for example, subjects hear to V /v = 1 to test for limited attention.
different messages in the right ear and in the Tanjim Hossain and John Morgan (2006) and
left ear, and are instructed to attend to one Chetty, Looney, and Kroft (forthcoming) in
of the messages. When asked about the other the section on alcohol taxes follow this ave-
message, they remember very little of it. nue. The second is to examine the response
Moreover, in treatments in which they have of consumervaluation to an increase in the
to rehearse a sentence or a sequence of num- salience s, V /s = so, and test whether
bers while listening, their capacity to attend it differs from zero. This is the strategy of
to a message is substantially lower. Chetty, Looney, and Kroft (forthcoming) in
I present here a simple model of atten- their field experiment. The third strategy is
tion as a scarce resource and derive testable tovary the number of competing stimuli N,
implications. Consider a good whose value /N = No, and test whether this has
V
V (inclusive of price) is determined by the an effect. This is the strategy of DellaVigna
sum of two components, a visible component and Joshua M. Pollet (forthcoming) and
v and an opaque component o, V = v + o. Hirshleifer, Sonya Seongyeon Lim, and Siew
Due to inattention,
the consumer perceives Hong Teoh (forthcoming). All three of these
the value to be V = v + (1 )o, where strategies identify a piece of opaque informa-
denotes the degree of inattention, with = tion o with regards to which the decision-
0 as the standard case of full attention. The makers are not fully attentive.
interpretation of is that each individual sees This research is subject to two cave-
the opaque information o, but then processes ats. The first caveat is that measuring the
it only partially, to the degree .32 The inat- salience of information involves a subjective
tention parameter is itself a function of the judgment, similar to the judgment involved
salience s [0,1] of o and of the number of in setting the reference point in prospect
competing stimuli N: = (s,N). Based on theory. While in most settings (such as the
the psychology evidence, I assume that the ones in this section) it is rather clear which
inattention is decreasing in the salience s features are visible and which are opaque,
and increasing in the competing stimuli N: s the psychology experiments do not provide
< 0 and N > 0. Inattention is zero for a fully a general criterion. The second caveat is that
we do not address whether the inattention is
rational or not. In general, models of limited
31 John Conlisk (1996) provides an early survey of this attention can be rephrased as rational model
literature. I discuss the model of inattention by Gabaix with information costs in which less salient
and Laibson (2006) in section 5.
32 An alternative model (Raj Chetty, Adam Looney, information has higher costs of acquisition.
and Kory Kroft forthcoming) posits that is the probabil- In most of the examples below, however, the
ity that an individual perceives the opaque signal, rather opaque information is publicly available at a
than the degree to which each individual incorporates the
signal. This alternative model leads to similar results but a zero or small cost (for example, the informa-
more cumbersome solution for settings like an auction. tion on earnings announcements), making
350 Journal of Economic Literature, Vol. XLVII (June 2009)

a rational interpretation of the findings less taxes that are not transparently factored in
plausible. the price of a good, like indirect state taxes.
They use data on the demand for items in
4.2.1 Inattention to Shipping Costs
a grocery store. Assume that demand D is
In eBay auctions, the price of an item is a function of the visible part of the value
more vivid than the shipping cost, because v, including the price p, and of the less vis-
the shipping cost is not listed in the item ible part o, in this case the state tax tp: D
title and also because historically most = D[v (1 )tp]. The change in log-
purchases have not involved shipping. demand log D from making the tax fully
Define v as the value of the object and o as salient (s = 1 and hence = 0) is (linearizing
the negative of the shipping cost: o = c. the demand) log D[v tp] log D[v (1
Since eBay is (essentially) a second price )tp] = tp D[v (1 )tp]/D[v (1
auction, the inattentive bidders bid their )tp] = t D,p, where D,p is the price
value net of the (perceived) shipping cost: elasticity of demand. (Since demand D is a
b* = v (1 )c. The revenue raised by function of value minus price, D,p = pD[v
the seller is b* + c = v + c. A $1 increase (1 ) tp]/D[v (1 )tp].) Notice that
in the shipping cost c, therefore, increases the response is zero for fully attentive con-
revenue by dollars. In the case of full sumers ( = 0). This implies = log D/(t
attention ( = 0), increases in the shipping D,p). Chetty, Looney, and Kroft (forth-
cost have no effect on revenue. Hossain and coming) manipulate the salience of taxes
Morgan (2006) examine these predictions with a field experiment. In a three-week
with a field experiment. In the treatment period, the price tags of certain items make
cLO, they auction CDs with a $4 reserve salient the after-tax price, in addition to
price and no shipping cost, while in treat- indicating the pretax price. Compared to
ment cHI they auction CDs with a $.01 previous-week sales for the same item, and
reserve price and a $3.99 shipping cost. The compared to items for which tax was not
change in reserve price guarantees that the made salient, the average quantity sold
two auctions are equivalent for a fully atten- decreases (significantly) by 2.20 units rela-
tive bidder. The average revenue raised in tive to a baseline level of 25, an 8.8 per-
treatment cHI is $1.79 higher ($10.16 versus cent decline. Since the price elasticity D,p
$8.37) than in treatment cLO, and is higher in this sample is estimated to be 1.59
for nine out of ten CDs. 33 These
estimates and the tax is 7.375 percent, we can com-
imply substantial inattention: = 1.79/3.99 pute = (.088)/(1.59 .07375) .75.
= .45. A second set of auctions with higher In a separate estimation strategy, Chetty,
shipping costs (cLO= $2 and cHI = $6), leads Looney, and Kroft (forthcoming) identify
to smaller increase of revenue in the high- the impact on beer consumption of changes
shipping cost condition ($12.87 vs. $12.15), across States and over time in the excise and
corresponding
to an inattention param- sales taxes. Since the excise tax is included
eter = 0.72/4 = .18. in the price while the sales tax is added at
the register, inattentive consumers should be
4.2.2 Inattention to Nontransparent Taxes
more responsive to changes in the excise tax
Chetty, Looney, and Kroft (forthcoming) than to changes in the sales tax. Indeed, the
study whether consumers are inattentive to first elasticity is substantially larger, leading
to an
estimate of the inattention parameter
33 I exclude CDs that do not sell from this computation; of = .94. Consumer inattention to non-
the difference would be $2.60 if they were included. transparent taxes is substantial.
DellaVigna: Psychology and Economics 351

4.2.3 Inattention to Complex Information published in November. Despite the fact that
in Rankings the article contains no new hard informa-
tion, it leads to a 330 percent one-day return
In other settings, the familiarity of infor- for EntreMed, and to a 7.5 percent one-day
mation depends on the simplicity of the return for all bio-tech companies, moving
data format. Devin G. Pope (2007) stud- billions in market capitalization. The stock
ies the response of consumers to rankings price of EntreMed does not revert to the pre-
of hospitals and colleges by the U.S. News vious level over the whole next year.
and World Report. Each year, the company While this is just a case study, it stresses
constructs a continuous quality score from the importance of studying systemati-
0 to 100 largely based on reputation scores, cally the response to new information. One
and then creates rankings based on this important setting is the release of quarterly
score. Both the scores and the rankings are earnings news, and the consequent response
published in the yearly report. While the of asset prices. To simplify, assume that v is
continuous score contains all the informa- the known information about cash flows of
tion, the rankings are presumably easier to the company, and that o is the new informa-
process (no. 5 hospital versus hospital with tion contained in the earnings announce-
89/100 score). Pope shows that, holding con- ment. On the day before the announcement,
stant the quality score, hospital discharges the company price is P = v. On the day of
respond significantly to differences in ranks the announcement, the updated company
among hospitals; similarly, college applica- value is v + o. However, since the investors
tions respond to differences in ranks among are inattentive, the asset price P responds
colleges. Pope (2007) also provides a calibra- only partially to the new information:
tion of the inattention or thinking costs nec- P = v + (1 )o. Over time, as the informa-
essary to justify this result. tion makes its way to the inattentive investors
(for example through additional articles as in
4.2.4 Inattention to Financial News
the EntreMed case), the price incorporates
Limited attention among investors induces the full value v + o. This implies that the
underreaction to newly released informa- short-run stock return rSR equals rSR = (1
tion and, hence, can explain anomalies such )o/v; the long-run stock return rLR, instead,
as momentum (Harrison Hong and Jeremy equals rLR = o/v. In this example, a measure
C. Stein 1999). Gur Huberman and Tomer of investor attention is (rSR /o)/(rLR /o)
Regev (2001) examine the case of the com- = (1 ). (The division by (rLR /o) is a
pany EntreMed, an interesting example of renormalization that makes the measure
underreaction to information. On November scale invariant) The higher is the inatten-
28, 1997, Nature prominently features an tion, the smaller is the immediate response
article reporting positive results on a cure and the larger is the predictability of stock
for a type of cancer for a drug patented by returns in the days following the announce-
EntreMed. On the same day, the New York ment, a phenomenon known as postearnings
Times reports an article on the same topic announcement drift (Victor L. Bernard and
on page A28. Unsurprisingly, the stock price Jacob K. Thomas 1989). Inattention leads to
of EntreMed increases by 28 percent. What delayed absorption of information.
is surprising is what happens next. On May While this setting is highly stylized, similar
4, 1998, the New York Times publishes on results obtain after allowing for uncertainty
the front page an article on EntreMed that is and arbitrage, as long as arbitrage is limited
very similar to the article that it had already by risk aversion and short investor horizons
352 Journal of Economic Literature, Vol. XLVII (June 2009)

(for example, DellaVigna and Pollet forth- returns have lower stock returns one to three
coming). DellaVigna and Pollet (forthcom- months later. They measure the speed of the
ing) estimate the empirical counterpart of response of returns to news about the cus-
(rSR /o) / (rLR /o) using the response of tomer company using (rSR /o)/(rLR /o),
returns r to the earnings surprise o. They where rSR is the one-month return and rLR is
measure returns in the two days surround- a seven-month return. They find that, for the
ing an announcement (rSR) and over the customer company, 93 percent of the overall
seventy-five trading days from an announce- response occurs in the initial month; for the
ment (rLR). The immediate response cap- supplier company, instead, only 60 percent
tures 54 percent of the overall response, of the overall response occurs in the first
implying substantial inattention: .46. If month, suggesting substantial inattention to
the delayed response is due to attention defi- indirect links.
cits, the delay should be even stronger when A final dimension of salience s is the tem-
a higher share of investors are distracted poral distance. Holding constant the infor-
(higher ). DellaVigna and Pollet (forthcom- mativeness, information that is further into
ing) use the weekend as a proxy of inves- the future (or past) is less likely to be salient.
tor distraction. For announcements made In general, it is difficult to control for infor-
on Friday, indeed, the share of immediate mativeness, since information that is further
response (rSR /o)/(rLR /o) is 41 percent, away is usually less relevant or less precisely
implying .59, consistent with higher inat- estimated. DellaVigna and Pollet (2007)
tention before the weekend. This provides an address this issue by considering future
explanation for the observed release of worse demand shifts due to demographics. Unlike
earnings on Friday: companies maximizing other demand determinants, cohort size
short-term value release worse news on low- shifts are highly predictable even ten years
attention days. into the future. For example, if a large cohort
In a similar context, Hirshleifer, Lim, and is born in 2006, school bus companies in
Teoh (forthcoming) analyze the impact of 2012 are going to experience a forecastable
informational overload (high N in the frame- increase in demand and, if the market is not
work above). They find that the incorporation perfectly competitive, in profits. If investors
of earnings news into stock prices is 20 per- are perfectly attentive, this increase will be
cent slower on days in which more announce- incorporated into returns already in 2006
ments take place. Increasing the amount and stock returns from 2006 to 2012 will not
of competing information accentuates the be predictable using demographic informa-
effect of limited attention. tion. However, if investors neglect informa-
Another related study is Lauren Cohen tion beyond five years into the future, the
and Andrea Frazzini (2008), which ana- stock prices will increase only in 2007, and
lyzes how investors respond to indirect, stock returns from 2006 to 2012 will be pre-
and hence less salient news (low s in the dictable using public information on demo-
framework above). They consider compa- graphics. Using data for forty-eight industries
nies linked in the suppliercustomer chain. from 1939 to 2003, DellaVigna and Pollet
When a customer company announces show that the growth rate in demand due to

substantial earnings news, the news affects demographics five to ten years ahead fore-
also the supplier, but this indirect effect is casts stock returns in an industry positively.
less likely to attract attention. Indeed, Cohen These results are consistent with inattention
and Frazzini (2008) show that suppliers of to information further than approximately
companies which experience declining stock five years into the future.
DellaVigna: Psychology and Economics 353

4.2.5 Summary investments. As a special case, they study


the case of equal diversification across the
Limited attention helps explain the (par- n available options, the 1/n heuristics. They
tial) neglect of (1) shipping costs in eBay auc- use aggregate data on the 1996 plan assets
tions; (2) nontransparent taxes; (3) complex for 162 companies that offer an average of 6.8
information in rankings; (4) earnings news, plan options. Lacking individual-level data,
especially before weekends and on days with they study an aggregate implication of the 1/n
more competing news; (5) news about linked heuristic. If individuals divide their invest-
companies; and (6) demand shifts in the dis- ments approximately equally across options,
tant future. As an example of application to their exposure to equity will be increasing in
another field, a literature on inattention in the availability of equity options in the 401(k)
macroeconomics developed from the models plan. Across plans, Benartzi and Thaler esti-
of sticky information of N. Gregory Mankiw mate the relationship
and Ricardo Reis (2002) and of rational inat-
tention of Christopher A. Sims (2003). (9) %Invested In Equity =
+ .36(.04)

4.3 Menu Effects


%Equity Options + B
X
In this section, I consider choices out of a
(typically large) menu set, such as for invest- (s.e. in parentheses), where the control vari-
ment options or politicians on a ballot. The able X is the availability of employer stock in
evidence in psychology suggests that individ- the portfolio. In companies with an equity
uals use (at least) five suboptimal heuristics share that is 10 percentage points higher, the
to simplify these decisions: (1) excess diver- employees invest 3.6 percent more in equity
sification (or 1/n heuristic), (2) preference plans. This finding is consistent with a weak
for the familiar, (3) preference for the salient, form of the 1/n heuristic (If the employees
(4) choice avoidance, and (5) confusion in followed the 1/n heuristics strictly, the coef-
implementing the choices. ficient should be 1 rather than .36). A con-
found is that the equity content of a plan may
4.3.1 Excess Diversification
be designed to cater to the preferences of the
Individuals facing a complex choice may employees, resulting in reverse causation.
simplify it by diversifying excessively across Huberman and Wei Jiang (2006) investi-
the options. An example in psychology is gate the investor diversification using a data
Itamar Simonson (1990). In a first treatment set on the individual choice of employees in
(simultaneous condition), students in a class 647 401(k) plans managed by Vanguard. They
chose snacks to be consumed over the next estimate specification (9) at the individual
three class meetings, one per meeting. In a level with a large set of individual-level and
second treatment (sequential condition), the plan-level controls X. They obtain the rela-
subjects chose the snack sequentially on each tionship %Invested In Equity
=
+ .29 (.11)

of the three class meetings. In the simulta- %Equity Options + B X
for funds with less
neous condition, the subjects display excess than ten options and %Invested In Equity
diversification: 64 percent of subjects chose = + .07(.09) %Equity Options + B
X
three different snacks, while in the sequen- for funds with more than ten options. The
tial condition only 9 percent of subjects made relationship predicted by the 1/n heuristic,
this choice. therefore, is present when the number of
Benartzi and Thaler (2001) study whether funds is small (as in the Benartzi and Thaler
excess diversification applies to 401(k) sample), but not when the number of funds is
354 Journal of Economic Literature, Vol. XLVII (June 2009)

large. Huberman and Jiang provide additional their discretionary funds in employer stocks
evidence suggesting that the predictive power (Benartzi 2001), despite the fact that the
of the 1/n heuristic is low. In particular, the employees human capital is already invested
number of funds chosen by employees hardly in their company. This choice does not reflect
responds at all to the number of investment private information about future perfor-
options offered in the plan. (This test differs mance. Companies where a higher propor-
from the one above as it is not conditional on tion of employees invest in employer stock
equity vs. nonequity choices.) There is some have lower subsequent one-year returns,
evidence of a conditional 1/n heuristic: con- compared to companies with a lower propor-
ditional on the allocations chosen, individuals tion of employee investment.
allocate their savings approximately equally. The preference for familiar options is con-
Thirty-seven percent of employees follow this sistent with ambiguity aversion. As in the
behavior among employees investing in four classical Daniel Ellsberg (1961) paradox,
funds, 26 percent among employees investing investors that are ambiguity averse may pre-
in five funds, and 53 percent among employees fer an investment with known distribution of
investing in ten funds; the behavior is instead returns to an investment with unknown dis-
not common for nonround numbers. Overall, tribution, even if the average returns are the
some employees use a version of the 1/n heu- same for the two investments, and despite
ristic when the number of investment options the benefits of diversification.
is small; when the number is large, other heu-
4.3.3 Preference for the Salient
ristics, which I discuss next, are at play.
Barber and Odean (2008) show that indi-
4.3.2 Preference for the Familiar
vidual investors simplify complex portfolio
A different heuristic to simplify complex decisions also by choosing a salient option.
decisions is the choice of a familiar option. Using individual trading data, they show that
This tendency is widespread among individ- individual investors are net buyers of compa-
ual investors. Investors in the United States, nies with unusually high, or low, performance
Japan, and the United Kingdom allocate 94 in the previous day, of companies with high
percent, 98 percent, and 82 percent of their trading volume, and of companies in the news.
equity investment, respectively, to domestic The effects are large: for companies in the
equities (Kenneth R. French and Poterba highest or lowest decile of the previous days
1991). While the preference for own-country returns, the BuySell imbalance (BuySell/
equity may be due to costs of investments (Buy+Sell)) for individual investors is 20 per-
in foreign assets, the same pattern appears centage points higher than for companies in
for within-country investment. Huberman the fifth decile. These results suggest that
(2001) documents the geographical distri- individuals solve the informational overload
bution of the shareholders of the Regional problem of which stocks to buy by picking
Bell companies. The fraction invested in companies that stand out. The same problem
the own-state Regional Bell is 82 percent does not present itself for stock sales, since
higher than the fraction invested in the next most investors own only a small number of
Regional Bell company. The preference for stocks at any given time. This asymmetric
the familiar occurs despite substantial costs pattern for stock purchases and sales could,
of underdiversification. however, also be due to short-sale constraint:
A particularly egregious case is the pref- individual investors would like to sell stocks
erence for own-company stock. On aver- that are salient but cannot do so due to the
age, employees invest 2030 percent of short-sale constraints. Consistent with the
DellaVigna: Psychology and Economics 355

inattention explanation, Barber and Odean in the twenty-four-jam condition, more con-
(2008) show that the BuySell imbalance sumers stop to sample jams (145 versus 104
patterns are similar for stock that the indi- customers), but substantially fewer buy jams
vidual investors already own and that, hence, (four versus thirty-one customers). This find-
they could easily sell. ing is surprising in light of the fact that in the
The preference for the salient takes differ- standard model more choice can only lead to
ent forms in different contexts. In the choice increased purchases. The explanation for this
of candidates on a ballot, the first politician counterintuitive phenomenon is that when
on the list stands out. A long-standing litera- the choice is difficult, for example because
ture in political science, going back to Henry the choice set is large, individuals find the
Bain and Donald S. Hecock (1957), exam- decision stressful and look for ways to avoid
ines the effect of the order of candidates on the choice.
a ballot. Among the most convincing stud- Marianne Bertrand et al. (forthcoming)
ies, Daniel E. Ho and Kosuke Imai (2008) examine the impact of a small or large menu
provides evidence that the order matters set in the context of a field experiment on the
even when it is randomly determined. They mailing of 50,000 loan offers in South Africa.
exploit the natural experiment induced by The authors randomize, among other things,
the California voting system that, since 1975, the format of the table illustrating the use of
explicitly randomizes the ballot order of can- the loan. The small-table format lists only one
didates across Assembly Districts. They show loan size as an example, while the big-table
that in the 1998 and 2000 general elections format presents four different loan sizes. The
a minor party candidate experiences on aver- finding is consistent with the choice avoid-
age a 10 percent increase in votes when first ance results. The take-up in the small-table
on the list. The effect is instead very small for format is .6 percentage points larger com-
candidates of the major parties, suggesting pared to a baseline of 8 percentage points,
that irrelevant information is used as a tie- an effect size equivalent to a reduction of
breaker for cases in which the decisionmaker the (monthly) interest rate by 2.3 percentage
has less information. In primary elections, in points.
which candidates are on average less known, Choi, Laibson, and Madrian (2006)
the effect is stronger: the impact of being first also provide evidence from a field experi-
in the list is on average a 20 percent increase, ment that a smaller number of investment
roughly 1.6 percent of the party vote. options increases participation in a 401(k)
plan. Participation increases by 10 percent-
4.3.4 Choice Avoidance
age points when nonparticipating employees
The fourth heuristic used to deal with receive a card that allows them, if mailed
difficult decisions is perhaps the most sur- back, to enroll in a default plan (3 percent
prising: avoiding the choice altogether, pos- contribution in a balanced fund).34
sibly in favor of the default action. In a field
experiment, Sheena S. Iyengar and Mark
R. Lepper (2000) compare the behavior of 34 The increase may be due to a reminder effect of the

consumers in an upscale grocery store, card. However, in other settings, reminders, and more
generally financial education, do not have such large
where at some times consumers were offered effects. For example, Carroll et al. (forthcoming) sent a
the opportunity to taste six jams (the sim- survey including five questions on the benefits of employer
ple-choice treatment), while at other times match to 345 employees that were not taking advantage
of the match. A control group of 344 employees received
the tasting included twenty-four jams (the the same survey except for the five specific questions. The
difficult-choice treatment). They find that, treatment had no significant effect on the savings rate.
356 Journal of Economic Literature, Vol. XLVII (June 2009)

4.3.5 Confusion Interestingly, Michael S. Rashes (2001)


identifies a similar phenomenon in the trades
A final category of behavior, confusion, dif- of two companies, MCI and MCIC. The
fers from the previous heuristics in that it does ticker for the MCI communication company
not reflect a preference, whether to avoid dif- is MCIC, while MCI is the ticker for a little-
ficult choices or for salient options, but simply known closed-end mutual fund, Massmutual
an error in the implementation of the prefer- Corporate Investors. Some investors attempt-
ences. As such, it differs from most behavioral ing to trade shares of the larger communica-
phenomena that reflect a directional bias. A tion company confuse tickers and trade the
first setting is the choice of a political candi- mutual fund company instead, resulting in
date among those in a ballot. Kelly Shue and a .56 correlation between the two trading
Erzo F. P. Luttmer (forthcoming) consider volumes. This occurs despite the difference
California voters in the 2003 recall elec- in fundamentals: the mutual fund company,
tions and exploit the random variation in the for example, has only a .03 correlation in
placement of candidates on the ballot, simi- volume with the communication company
larly to Ho and Imai (2008). They find that AT&T. The mistrading causes a smaller, but
the vote share of minor candidates is signifi- still significant correlation of stock returns.
cantly higher for candidates whose name on Arbitrage moderates the impact of confusion
the ballot is adjacent to the name of a major on stock returns, but does not fully eliminate
candidate. While this phenomenon could be it. Using the correlation in volume and the
due to a spillover in attention, confusion is a average volume of trade for the two compa-
more likely explanation: the effect of horizon- nies, one can compute the incidence of confu-
tal adjacency (a name to the right or to the left sion among MCIC investors: about 1 in 2,000
of the major candidate) is almost entirely due trades are placed in error, a confusion rate
to adjacency on the confusing side. For exam- smaller than the confusion rate displayed by
ple, in the sequence Bubble, Candidate A, California voters.35
Bubble, Schwarzenegger, Bubble, Candidate
4.3.6 Summary
B, it is Candidate B that benefits from the
presence of a major candidate, since some When choosing from a large menu of
voters mistake its bubble for the bubble of options, decisionmakers: (1) (to same extent)
Schwarzenegger. Candidate A does not ben- diversify excessively across the options; (2)
efit, nor do candidates located at a diagonal choose familiar options, such as own-country
adjacency. Further, the spillover of votes is or own-company stock; (3) choose salient
larger for more confusing voting methods options in investment choice or at the ballot;
(such as punch cards) and for precincts with a (4) avoid the choice and do not invest (or do
larger share of lower-education demographics not purchase); (5) display some confusion in
that are more likely to make errors when faced implementing their choices.
with a large number of options. This method
allows for a measure of confusion. Across dif-
ferent voting methods, about 1 in 300 voters
meaning to vote for a major candidate instead 35 Assume that the volume (number of trades) of MCI,
vote for a minor candidate. The phenomenon VMCI,t, equals a constant plus the shares traded due to
hence is small but not irrelevant. Importantly, confusion, s VMCIC,t where s is the share of investors of
it can have an aggregate effect, since confusion MCIC that incorrectly trade MCI: VMCI,t = + sVMCIC,t.
Given a simple correlation coefficient between the daily
is likely to have a different prevalence among = .56
volumes VMCI,t and VMCIC,t of .56, we can infer s
the voters of different major candidates. s.d. (VMCI,t)/s.d.(VMCIC,t) .56 10 3.
DellaVigna: Psychology and Economics 357

4.4 Persuasion and Social Pressure for the additional distortions due to analyst
affiliation.
4.4.1 Persuasion In a political setting, DellaVigna and
Ethan Kaplan (2007) tests whether the infor-
In the standard model, individuals take mation provided by a news source convinces
into account the incentives of the informa- on average its audience. They exploit the
tion provider. The neglect of these incentives geographical variation in the introduction
can lead to excess impact of the beliefs of the in the cable programming of the Fox News
information provider, which I label persua- Channel, a more conservative channel rela-
sion. An example from a laboratory experi- tive to the preexisting news sources (CNN
ment is Daylian M. Cain, Loewenstein, and and the networks). They show that Fox News
Don A. Moore (2005). The subjects are paid availability in the town cable programming
for the precision of the estimates of the num- in 2000 is largely idiosyncratic, conditional
ber of coins in a jar. Since they see the jar on a set of controls. Using the voting data for
only from a distance, they have to rely on the 9,256 towns, they find that the vote share
advice of a second group of subjects, the advi- for Republicans in 2000 is half-a-percentage
sors, that inspect the jar from up close. The point higher in the towns offering Fox News.
two experimental treatments vary the incen- They estimate that Fox News convinced 5
tives for the advisors. In a first treatment, the to 30 percent of the audience that was not
advisors are paid for how closely the subjects already Republican, depending on the audi-
guess the number of coins; in a second treat- ence measure. The effect is of about the same
ment, the advisors are paid for how high the size for the Presidential candidates and for
subjects guess is. Despite the fact that the the U.S. Senate candidates, which Fox News
incentives are common knowledge, the esti- does not cover. This indicates that the effect
mate of the subjects is 28 percent higher in of Fox News extends beyond the candidates
the second treatment. The subjects do not dis- covered to the general political beliefs of the
count enough for the conflict of incentives of voters. The impact of Fox News can be a
the advisors. temporary effect for Bayesian voters that are
In a financial setting, Malmendier and learning about the bias of Fox News or a per-
Devin Shanthikumar (2007) analyze how suasion effect for nonrational voters that do
small and large investors respond to recom- not take sufficiently into account the political
mendation by analysts. Analyst forecasts are orientation of Fox News.
notoriously biased upward94.5 percent of
4.4.2 Social Pressure
recommendations are Hold, Buy, or Strong
Buyand affiliated analysts are even more A separate reason for excess impact of the
biased. Malmendier and Shanthikumar beliefs of others is the pressure to conform,
(2007) show that large investors take into or social pressure (Akerlof 1991). Two clas-
account this bias and discount the informa- sical laboratory experiments illustrating the
tion: for example, they respond to a Hold power of social pressure are Solomon E.
recommendation by selling the shares of a Asch (1951) and Stanley Milgram (1963).
company, and they discount heavily positive In one of the Asch (1951) experiments, the
recommendations by affiliated analysts. Small subjects are shown two large white cards
investors, instead, are subject to persuasion. with lines drawn on them: the first card has
They follow the recommendations literally three lines of substantially differing length
for example holding a stock in response to a on them, while the second card has only one
Hold recommendationand do not discount line. The subjects are asked which of the
358 Journal of Economic Literature, Vol. XLVII (June 2009)

lines in the second card is closest in length goal ahead). The effect is larger when stakes
to the line in the first card. In a control treat- are higher (toward the end of the season) and
ment, the subjects perform the task in isola- when the social pressure is larger (larger atten-
tion and achieve 98 percent accuracy. In the dance at the game). Referees respond signifi-
high-social-pressure treatment, the subjects cantly to pressure by the local public, despite
choose the line of comparable length after official rules on what determines the length of
four to eight subjects (who, unbeknownst extra time.
to them, are confederates) unanimously Some of the peer effect literature also
choose the wrong answer. On average, over points to the importance of social pressure.
a third of subjects give the wrong answer to Falk and Andrea Ichino (2006) measure
avoid disagreeing with the unanimous judg- the effect of peer pressure on task perfor-
ment of the other participants. While this mance. High school students in Switzerland
result could be interpreted as social learn- were recruited to perform a one-time job
ing, the learning is unlikely to be about the for a flat payment; they were instructed to
length of the line, but possibly about the stuff letters into envelopes for four hours.
rules of the experiment. It should also be The control group of eight students did
pointed out that the subjects were not paid the task individually, while the treatment
for accuracy. group of sixteen students worked in pairs
In the Milgram (1963) experiment, a group (but each student was instructed to stuff
of subjects is told that their task is to monitor the envelopes individually). Students in the
the learning of another subject (a confeder- treatment group stuffed significantly more
ate) and to inflict electric shocks on this sub- envelopes (221 versus 190), and coordinated
ject when he makes an error. Encouraged by the effort within group: the within-pair
the experimenter, 62 percent of the subjects standard deviation of output is significantly
escalate the electric shocks up to a level of less than the (simulated) between-pairs
450 volts, despite hearing the subject scream standard deviation.
in pain. This proneness to obedience comes as While the results of Falk and Ichino
a surprise to the subjects themselves. When (2006) could also be due to social learning,
a different group of forty subjects is provided Mas and Enrico Moretti (forthcoming) pre-
with a description of the experiment and senting direct evidence of social pressure.
asked to predict how far subjects would go They find that high-productivity cashiers in
in inflicting shocks, no one predicts that 450 a supermarket chain increase the produc-
volts would be reached. tivity of coworkers that are present in the
In the field, social pressure is hard to distin- same shift. The effect is not due to exchange
guish from rational diffusion of information. of information, such as on a price tag. The
In some studies, however, the social pressure positive peer effect occurs only when the
motive is evident. Luis Garicano, Ignacio more productive coworker is behind and
Palacios-Huerta, and Canice Prendergast therefore can observe the other workers
(2005) measure the length of extra time that productivity. The effect is quite large: a one
referees assign at the end of a game of soc- percent increase in the average permanent
cer; in the extra time the teams can score productivity of the workers behind increases
goals. They find that referees on average the productivity of the peer by .23 percent;
give twice as much extra time (four minutes the effect is even larger for coworkers that
versus two minutes) when the extra time is are working at a closer distance. There is
bound to advantage the local team (one goal no effect of a highly productive coworker in
behind) than when it is bound to hurt it (one front. This suggests that the peer effect in
DellaVigna: Psychology and Economics 359

productivity is entirely due to the social pres- five percent of a standard deviation. After
sure induced when a worker feels observed controlling for cloud cover, other weather
by a high-productivity coworker. variables such as rain and snow are unre-
lated to returns. If mood is the channel for
4.5 Emotions these effects, other mood-altering events
should have similar effects. Indeed, inter-
Some of the previous phenomena, such national soccer matches impact the daily
as self-control problems, social preferences stock returns for the losing country (Alex
in giving, and projection bias in food pur- Edmans, Diego Garcia, and Oyvind Norli
chase are likely mediated (at least partially) 2007). Compared to a day with no match, a
by emotional states, respectively tempta- loss lowers daily returns (significantly) by .21
tion, empathy, and hunger. In section 2.3, for percent. (Surprisingly, a win has essentially
example, I discussed how the transient effect no effect). More important matches, such as
of a gift in the GneezyList (2006) field World Cup elimination games, have larger
experiment points to the role of emotions in effects. The effect does not appear to depend
gift-exchange behavior. A large literature in on whether the loss was expected or not.
psychology suggests that emotions play an International matches in other sports have a
important role in decision making, and that consistent, though smaller, effect.
different emotions operate very differently The effect of these mood-altering events
(Loewenstein and Lerner 2003). In this sec- on returns is likely due to (1) an impact on
tion, I consider two examples of emotions, risk aversion or perception of volatility or (2)
mood and arousal, for which field evidence a projection of the mood to economic fun-
is available. damentals. The evidence above does not
In psychology studies, even minor mood allow us to distinguish these two effects.
manipulations have a substantial impact on Mood induced by atmospheric factors can
behavior and emotions. For example, on sun- also induce subtler changes in behavior. Uri
nier days, subjects tip more at restaurants Simonsohn (forthcoming) examines the role
(Bruce Rind 1996) and express higher levels of weather on the day of campus visit to a
of overall happiness (Norbert Schwarz and prestigious university. Students visiting on
Gerald L. Clore 1983). In the field, mood days with more cloud cover are significantly
fluctuations induced by the weather affect more likely to enroll. Simonsohn suggests
stock returns, despite the fact that daily that higher cloud cover induces the students
weather fluctuations are unlikely to affect to focus more on academic attributes versus
fundamentals. Days with higher cloud cover social attributes of the school, a hypothesis
in New York are associated with lower aggre- supported by laboratory experiments.
gate U.S. stock returns (Edward M. Saunders A second set of laboratory experiments sug-
1993). Hirshleifer and Tyler Shumway (2003) gests that emotional arousal has an important
extend this analysis to twenty-six countries short-run effect on decisions. In one experi-
between 1982 and 1997 using the weather of ment, subjects that are sexually aroused as
the city where the stock market is located. part of the treatment report a substantially
They find a negative relationship between higher willingness to engage in behav-
cloud cover (detrended from seasonal aver- ior that may lead to date rape (Ariely and
ages) and aggregate stock returns in eighteen Loewenstein 2006). In other experiments,
of the twenty-six cities. Days with completely subjects exposed to violent video clips are
covered skies have daily stock returns .09 more likely to display more aggressive behav-
percent lower than days with sunny skies, ior, such as aggressive play during a hockey
360 Journal of Economic Literature, Vol. XLVII (June 2009)

game, compared to a control group watching (the field findings), even if it increases violent
nonviolent clips (Wendy L. Josephson 1987). behavior relative to exposure to nonviolent
The short-run impact on violence is not due movies (the laboratory findings). Indeed,
to imitation, since the violent movie did not one can use the field estimates to infer the
involve sport scenes. Hence, this effect is direct effect of violent movies (as in the
likely due to arousal induced by exposure to laboratory) if one can control for selection.
movie violence. After accounting for selection along observ-
Gordon Dahl and DellaVigna (2009) able dimensions (age and gender), Dahl and
provide field evidence on this effect and DellaVigna (2009) provide some evidence
estimate the short-run impact of exposure that indeed the direct effect of violent mov-
to media violence on violent crime. They ies is to induce more violent crime, like in the
exploit the time-series variation in movie laboratory experiments.
violence at the box office and compare days This study suggests that one ought to be
in which the blockbuster movies are violent careful about directly comparing the results
to days in which the blockbuster movies are of laboratory and field studies (Levitt and
nonviolent. They find that, on days in which List 2007). Results that appear at first con-
exposure to media violence is higher, violent tradictory can be reconciled in light of differ-
crime is lower. In particular, in the night ences in design. In this particular case, both
following the exposure (midnight to 6AM), the laboratory and the field estimates provide
for every million people exposed to violent useful evaluations of policy-relevant param-
movies, violent crime is 2 percent lower. The eters. The field results provide evidence on
results hold after controlling flexibly for sea- the short-run impact of a policy that restricts
sonality, weather, and are robust to placebo the amount of media violence available in
specifications. the theaters. The laboratory evidence pro-
What explains this result, apparently in vides evidence on the ban of violent material
contrast with the laboratory findings? The embedded in nonviolent programming.
key factor is the difference in treatments. In
the laboratory experiments, treatment and
5. Market Response
control groups are required to watch either
a violent or nonviolent movie. Hence, the In the previous sections, I have docu-
experiments capture the direct impact of mented how consumers deviate from the
movie violence, holding everything else con- standard model in their choices of credit
stant. In the natural experiment in the field, cards, clothing items, eBay bidding strate-
instead, consumers optimally choose between gies, giving, health clubs, housing prices,
violent movies and their other favorite activ- insurance contracts, loans, and lotteries. I
ity. Hence, the estimated impact in the field have discussed how workers make nonstan-
captures the net effect of exposure to the dard effort, labor supply, and retirement sav-
movies, compared to the impact of this alter- ings decisions. I have provided evidence of
native activity. The subpopulation attending disposition effect, inattention, and overtrad-
violent movies would, on average, be doing ing among investors. Finally, I documented
an even more dangerous activity, such as evidence of salience effects, persuasion, and
drinking at a bar, if they were not attending confusion among voters.
the theater. The two different sets of results, This evidence is just the first step toward
therefore, can be reconciled. Exposure to a better understanding of markets where
movie violence can lower violent behavior agents display nonstandard preferences
relative to the foregone alternative activity and beliefs. This evidence raises a natural
DellaVigna: Psychology and Economics 361

uestion: how do markets and institutions


q 5.1 Behavioral Industrial Organization
respond to these nonstandard features? An
important test for Psychology and Economics The interaction between consumers with
is whether it helps to understand markets biases and rational, profit-maximizing firms
and institutions, in addition to explaining is the central theme of the growing literature
individual behavior. in behavioral industrial organization, sur-
This section discusses how rational actors veyed in Glenn Ellison (2006). While this
respond to the nonstandard features of other literature is mostly theoretical, the papers
agents. Profit-maximizing firms respond surveyed here also make predictions about
to the nonstandard features of consumer observed pricing.
behavior in their contract design and pric- DellaVigna and Malmendier (2004) con-
ing (Behavioral Industrial Organization). sider
the profit-maximizing pricing with
Employers tailor their employment con- (,,) consumers with self-control problems.
tracts to the nonstandard behavior of the A (monopolistic) firm sells a product which,
employees (Behavioral Labor Economics). as in section 2.1, has immediate payoff b1 and
In response to the nonstandard behavior of delayed payoff b2. The set-up covers invest-
investors, rational investors alter their trad- ment goods such as exercise (b1 < 0 and b2
ing strategies and firm managers alter the > 0) and leisure goods such as gambling (b1
capital structure (Behavioral Finance and > 0 and b2 < 0 ). The immediate payoff b1
Behavioral Corporate Finance). Politicians is stochastic with c.d.f. F. The firm produces
change their behavior to respond to voter the good at marginal cost c and sells it using a
biases (Behavioral Political Economy). two-part tariff, with a lump-sum fee L and a
Finally, policymakers can use the findings unitary price p. DellaVigna and Malmendier
in Psychology and Economics to inform the (2004) show that the profit-maximizing price
design of institutions and of policy (Behavioral p* satisfies
Institutional Design).
f( b2 p*)
__________
Before I proceed, I discuss an important (10) p c = (1
* ) b2
caveat. If consumers have nonstandard fea- f(b2 p*)
tures, why should one expect firms, employ-
F( b2 p*) F(b2 p*)
_____________________
ers, financial operators, and politicians to .
not have them? Experience is a key differ- f(b2 p*)
ence. Unlike individual consumers, firms
can specialize, hire consultants, and obtain For standard agents ( = = 1), the two
feedback from large data sets and capital terms on the right-hand side of (10) are zero:
markets. Firms are also subject to compe- the firm prices at marginal cost, p* = c, to
tition. Compared to consumers, therefore, align the incentives of the consumers. For
firms are less likely to be affected by biases sophisticated
agents with self-control prob-
(except for principleagent problems), and lems ( =
< 1), only the first term in (10)
we expect them to be close to profit maxi- is non-zero: the firm prices investment goods
mization. In addition, even when, despite below marginal cost (p* < c) and leisure
the reasons above, firms still have nonstan- goods above marginal cost (p* > c) to provide
dard features, they still have incentives to a commitment devicethe pricing increases
respond to the nonstandard features of con- the consumption of investment goods and
sumers. Similar arguments apply for employ- lowers the consumption of leisure goods.
ers, institutional investors, top managers, The deviation from marginal cost pricing,
and politicians. (1 )b2, is exactly the difference in how
362 Journal of Economic Literature, Vol. XLVII (June 2009)

much the current self and the future selves (hidden) fees on bank accounts and credit
value the delayed payoff b2; hence, the firm cards. Gabaix and Laibson (2006) also dis-
offers a perfect commitment device. For cuss how markets do not generally provide
fully naive
agents with self-control problems incentives for debiasing naive consumers.
( <
= 1), only the second term in (10) is Paul Heidhues and Kszegi (2008) study
non-zero: the firm again prices investment the pricing of a monopolist when consumers
goods below marginal cost and leisure goods have reference-dependent preferences and
above marginal cost again, but for a differ- the reference point is the rational expecta-
ent reasonit takes advantage of consumer tions equilibrium (Kszegi and Rabin 2006).
overestimation (underestimation) of the con- Consumers are loss averse with respect to
sumption of investment (leisure) goods. both lower quality and higher price, relative
The deviation from marginal cost pricing is to the reference point. The main predictions
indeed a function of the* misestimation of are sticky prices (despite no menu costs)
consumption F( b2 p ) F(b2 p*). and sales, two common features of pricing.
These results generalize to the case of perfect In equilibrium, even if costs are stochastic,
competition, since competition only alters firms adjust prices seldom in response to cost
the equilibrium fee L*. This theory rational- shifts because consumers suffer more from
izes the presence of contracts with no pay- price increases than they benefit from price
ment per visit in health clubs (b2 > 0), the cuts. In addition, firms offer random sales
presence of high interest rates but no annual because the expectation of sales increases
fees for credit cards (b2 < 0), and cheap room the likelihood of purchases at high prices.
rates and buffets for gamblers in Las Vegas These papers point to a dichotomy in the
(b2 < 0). welfare effects of the market response. If the
Kfir Eliaz and Ran Spiegler (2006) general- agents have nonstandard preferences, such
ize this analysis to allow for heterogeneity in as self-control problems or loss aversion, but
naivet and a more general form of time-incon- have rational expectations, the firms provide
sistency of preferences. They show that firms welfare-maximizing contracts. The contracts
offer two types of contracts: perfect commit- offer first-best commitment devices against
ment devices that cater to time-inconsistent the self-control problem (DellaVigna and
agents that are sufficiently sophisticated, and Malmendier 2004; Eliaz and Spiegler 2006)
contracts that take advantage of the consum- or lower the probability of falling below
ers that are sufficiently naive. Interestingly, the reference point (Heidhues and Kszegi
the fully sophisticated agents do not exert any 2008). If, instead, the agents have nonrational
informational externality on the naive types. expectations, such as about the self-control or
Thus, the provision of the perfect commit- about the inattention, the profit-maximizing
ment device does not reduce the gains that contract is likely to magnify the bias. Firms
the monopolist can extract from naive types. take advantage of the wrong expectations
Gabaix and Laibson (2006) analyze the in the consumption of the tempting good
pricing with boundedly rational consum- (DellaVigna and Malmendier 2004; Eliaz
ers that do not pay attention to hidden fea- and Spiegler 2006) or of the add-on (Gabaix
tures of products, that they call add-ons. In and Laibson 2006).
equilibrium, firms charge above-marginal
5.2 Behavioral Labor Economics
cost prices for the add-ons. As in DellaVigna
and Malmendier (2004), the firms respond Contracting within a firm is also consistent
to the misprediction of future purchases. with this framework. Kahneman, Knetsch,
This model provides an explanation for high and Thaler (1986) present suggestive evidence
DellaVigna: Psychology and Economics 363

using a survey that workers display loss aver- impact of arbitrage. J. Bradford DeLong et
sion with respect to nominal wage losses, al. (1990) considers the case of a mispricing
but not with respect to real wage losses. For that is stochastic, persistent, and correlated
example, 62 percent of respondents find by so-called noise traders. If arbitrageurs are
unfair a wage cut of 7 percent in the pres- risk averse and have a limited investment
ence of no inflation, but only 22 percent of horizon, the noise traders affect the equilib-
respondents find unfair a 5 percent increase rium price, despite arbitrage. If noise traders
in salaries in presence of 12 percent inflation. are, for example, bullish about dot-coms, they
Truman F. Bewley (1999) documents similar will bid the price of dot-com shares higher.
patterns in a series of interviews. In response The arbitrageurs do not know whether the
to a dislike for nominal wage cuts, a profit- mispricing will get even worse in the next
maximizing employer should set wages such period, and given their short horizons (they
that nominal wage cuts would be rare. Card have to liquidate the shares next period) they
and Dean Hyslop (1997) provide evidence cannot short the shares aggressively enough.
on this prediction using CPS data. They con- DeLong et al. (1990) also shows that the
sider the distribution of year-to-year changes noise traders are not driven out of the mar-
in the nominal log wage, log wt log wt1. ket; under some conditions, in fact, they out-
In the presence of aversion to nominal wage perform the rational traders (since they take
losses, we expect a discontinuity in the dis- more risk).
tribution at log wt log wt1 = 0. Rather The recent research in behavioral finance
than introducing small cuts in the nominal builds on the noise-trade models to cap-
wages that may lower morale and produc- ture the limits of arbitrage and, hence,
tivity, the employer keeps wages constant the relevance of nonstandard behavior for
(log wt log wt1 = 0), compensating possi- asset prices. At the same time, this litera-
bly by firing more workers. Card and Hyslop ture moved beyond these models in mak-
indeed show that a substantial fraction of the ing explicit the source of noise trading. In
distribution of log wt log wt1 is missing sections 3.1 and 4.2, for example, I discuss
for negative values, despite the presence of models of overconfidence and limited atten-
measurement error in the wage that tends to tion, which make specific predictions about
attenuate this finding. This is an example of the nonstandard behavior and, hence, the
a market response to a bias that is likely to effect on returns. The evidence on this class
maximize utility for the biased agents. The of models is summarized in Shleifer (2000)
observed distribution of wages is such that and Barberis and Thaler (2003).
the employees suffer only rarely the disutil-
5.4 Behavioral Corporate Finance
ity of nominal wage cuts.
In corporate finance, the standard theory
5.3 Behavioral Finance
assumes that managers maximize company
In asset markets, arbitrage in principle is value subject to agency problems, given the
likely to limit the importance of behavioral demands of rational investors and credi-
biases such as inattention and overconfi- tors. A recent theory, known as market tim-
dence for price formation. If an irrational ing, expands this framework and assumes
agent believes that a (fair) coin will land on that investors may have an irrationally high
tails sixty percent of the time, arbitrage by or low valuation of the company. The CEO
well-informed agents will keep the odds of rationally responds to the misvaluation
tails around fifty percent. In actual financial through the equity issuance and merger
markets, however, several factors limit the decisions. CEOs provide additional shares
364 Journal of Economic Literature, Vol. XLVII (June 2009)

to investors and undertake mergers when s tandard deviation higher intensity of news in
the shares are most likely to be overpriced, the U.S. media. Similarly, the probability of
lowering the welfare of the biased investors. relief is 30 percent lower in the period of the
Market timing can explain the systematic Olympics. On days in which the American
underperformance of initial public offerings public is less likely to notice the U.S. generos-
(IPOs) in the three to five years following the ity, generous acts are less likely to take place.
IPO (Tim Loughran and Jay R. Ritter 1995). This is consistent with politician response to
According to this interpretation, managers of limited attention of voters.
private companies go public when the shares
5.6 Behavioral Institutional Design
of their companies are overpriced, hence the
underperformance of IPOs. Malcolm Baker, While firms, investors, managers, and politi-
Richard S. Ruback, and Jeffrey Wurgler cians may respond to biases by exploiting them,
(2007) reviews the evidence supporting this the response to biases need not be predatory.
theory. This theory complements the stan- Societal rules and institutions can be designed
dard theory that issuance decisions respond to counteract the effect of consumer biases
to investment opportunities. and improve the welfare of consumers. Thaler
and Benartzi (2004)s Save More Tomorrow
5.5 Behavioral Political Economy (SMarT) plan is an example of one such insti-
tutional design for 401(k) savings. In a SMarT
Another setting in which we expect an plan, the contribution rate is set to increase
asymmetry between rational and biased at each future wage increase up to a capped
agents is politics. While politicians are expe- level. While savings increases are the default,
rienced agents facing high-stake incentives employees can opt out of the plan at any time.
and significant competition, voters make This plan is an attractive commitment device
infrequent low-stake decisionswhether to to individuals with self-control problems since
vote and for whom. Therefore, we expect the default applies to future savings rates,
political settings to be well-described by the rather than current ones. In addition, the plan
interaction of rational politicians and vot- is designed with an eye to individuals that are
ers with nonstandard preferences, such as averse to nominal wage cuts (see above), since
imperfect memory and limited attention. the increases in contribution rates occur at the
Thomas Eisensee and David Stromberg time of pay increases.
(2007) provides an example of politicians Thaler and Benartzi (2004) provide evi-
responding to a bias of voters, inattention. dence on three implementations of this plan.
They consider the decision by U.S. ambassa- In the earliest implementation, the plan is
dors to release U.S. aid in the days following a offered to 207 employees that accept to meet
natural disaster in the country. Ambassadors with a financial consultant but do not accept
presumably are more likely to release aid if to increase the savings rate immediately, as
they, or the government, get credit for their recommended by the consultant. Of these 207
generosity. To capture this phenomenon, individuals, 162 individuals accept the SMarT
Eisensee and Stromberg exploit variation plan, indicating a widespread demand for com-
in voter inattention due to the presence of mitment. In this subset of 162 individuals, the
major news items in the U.S. television eve- contribution rate increases from 3.5 percent
ning news or due to a major sporting event to 13.6 percent in just four years. This increase
like the Olympics. They find that the prob- includes the thirty-two individuals who opted
ability of USAID relief is 15 percent lower out of the plan by the fourth year. The early
for disasters occurring on days with a two results from the other two implementations of
DellaVigna: Psychology and Economics 365

the SMarT plan indicate that the take-up of from the standard model: nonstandard pref-
the plan is lower if it is offered as an option via erences, nonstandard beliefs, and nonstan-
mail, as opposed to with an in-person meet- dard decision making. I have discussed how
ing. The effects conditional on take-up are, rational agents in the market respond to
however, similarly large. these nonstandard features. As this survey
A simple change in defaults, hence, can go documented, deviations from the standard
a long way toward addressing undersaving. model are not confined to laboratory deci-
The large impact of the SMarT plan, which sions. Most phenomena that are important in
comes at very limited administrative cost, laboratory experiments also affect decisions
is particularly noticeable when compared in a variety of economic settings. Hence, I
to the limited impact of other policies to expect that economists will increasingly take
increase retirement savings such as financial behavioral phenomena into account in their
education. Also, while this plan is designed analysis.
to benefit individuals with self-control prob- Why dont market forces eliminate non-
lems, it does not hurt individuals with time- standard behavior? While a full discussion
consistent preferences since these individuals of this objection is beyond the scope of this
can switch at any time. The success of this article, I address two related arguments, one
plan, as well as the research on default effect on experience and another on aggregation.
in 401(k) savings, have motivated the bill on A first argument is that experience reduces
Automatic Savings and Pension Protection nonstandard behavior. Indeed, experience
Act that Congress enacted in 2006. This law appears to mitigate the endowment effect
gives incentives to companies to adopt 401(k) (List 2003, 2004). Palacios-Huerta and Oscar
plans with automatic enrollment and auto- Volij (forthcoming) provide concordant evi-
matic increases in savings. dence on the effect of experience on the
While the evidence in Psychology and ability to perform backward induction. They
Economics can have important policy implica- consider the centipede game. Chess players,
tions, such as in this case, other considerations who have to routinely perform backward
limit the policy reach of this evidence. First, induction-type reasoning, come close in their
unlike in the Thaler and Benartzi (2004) case, play to the predictions of backward induc-
welfare-enhancing policies can be imprac- tion, in sharp contrast to college students.
ticalfor example, it is more difficult to use However, it would be wrong to conclude,
defaults to help people exercise more. Second, based on this evidence, that behavioral phe-
political economy considerations suggest cau- nomena should not matter in the field. I list
tion in the implementation of policies (Edward four reasons. (1) In a number of economic
L. Glaeser 2006). Nevertheless, behavioral decisions, feedback is infrequent (such as in
phenomena should be taken into account house purchases) or noisy (such as in financial
alongside standard phenomena in the policy investments) and, hence, most individuals are
design. Future research will tell whether inexperienced. (2) Experience can exacerbate
the 2006 Savings Act will remain an isolated a bias if individuals are not Bayesian learn-
application of Psychology and Economics to ers. Michael S. Haigh and List (2005) use a
policy or will be the first of several. simple investment game and show that pro-
fessional investors display significantly more
myopic loss aversion (see section 2.2) than
6. Conclusion
students. A possible explanation is that the
In this survey, I have summarized the short-term incentives in the workplace teach
field evidence on three classes of deviations these investors to frame problems narrowly,
366 Journal of Economic Literature, Vol. XLVII (June 2009)

contrary to the prediction of the standard above the posted price, a conclusion robust to
theory. (3) In principle, debiasing by expe- the inclusion of shipping costs, to differences
rienced agents can be a substitute for direct in item quality and in seller reputation. The
experience. However, as Gabaix and Laibson key aggregation point is that this behavior is
(2006) show, experienced agents such as firms generated by many fewer than 42 percent of
typically have little or no incentive to debias overbidders. In fact, only 17 percent of bidders
individuals. (4) Finally, not all nonstandard ever overbid. The auction design is such that
features should be mitigated by experience. the overbidders are disproportionately likely
Experience should not affect social prefer- to determine the final price.
ences any more than it should affect prefer- A natural question is what empirical
ences for the characteristics of cars. research in Psychology and Economics will
A second argument is that, even if experi- look like in the future. Methodologically,
ence or debiasing do not eliminate the biases, I expect future research to continue using
the biases will not affect aggregate market mostly the methods encountered in this over-
outcomes. The argument is made forcefully view, field experiments (such as List 2003
in financial markets: given arbitrage, the and Falk 2007), natural experiments (such
rational investors set prices. However, as I as Madrian and Shea 2001 and DellaVigna
discussed, the limits to arbitrage (DeLong and Kaplan 2007), and inference from menu
et al. 1990) imply that individuals with non- choice (such as DellaVigna and Malmendier
standard features will in general affect stock 2006 and Sydnor 2006), in addition, of course,
prices. In addition, in most settings, there is to the laboratory experiments. These method-
no plausible incentive to eliminate a bias and, ologies will increasingly provide, in addition to
hence, the effect of nonstandard behavior reduced-form estimates, also structural esti-
aggregates linearly. If a share of the popula- mates of the parameters (such as in Laibson
tion procrastinates saving for retirement, the et al. 2006 and Conlin, ODonoghue, and
aggregate savings rate will reflect proportion- Vogelsang 2007). These estimates will allow
ally the undersaving by this group. This is us to address a number of open questions,
true unless a different institutional design is such as whether models of (,) preferences
put in place, such as the SMarT plan (Thaler can explain choice in different decisions for
and Benartzi 2004). (Notice that this plan fixed parameters , , and . This estimation
was designed by academics, not by firms would benefit from the availability of data
though firms ultimately adopted it). sets with multiple decisions by the same indi-
Finally, the papers on behavioral IO indi- vidual. While individuals are likely to differ
cate that the nonstandard features, instead of in their preferences and beliefs, we expect
having no impact, can in fact have a dispro- the same individual to behave consistently
portionate impact on market outcomes. Young if the existing models capture the behav-
Han Lee and Malmendier (2007) provide an ior accurately.36 New, more parsimonious
example regarding overbidding in eBay auc- models of the phenomena presented in this
tions. They define a case of overbidding when survey will likely also emerge, as Fudenberg
the final auction price is higher than a posted (2006) predicts.
price for the same good available on eBay
itself. They focus on an item for which the
posted price is stable and essentially always 36 In a laboratory experiment, Raymond Fisman,

available, and hence should be an upper bound Shachar Kariv, and Daniel Markovits (2007) use repeated
decisions on giving to another subject to identify types of
for the bids in a rational model. The authors subjects with different social preferences. Their results
show that 42 percent of auctions end at a price suggest substantial heterogeneity in social preferences.
DellaVigna: Psychology and Economics 367

As for the topics, future research is likely References


to reduce the imbalance across fields in Abadie, Alberto, and Sebastien Gay. 2006. The
economics and across topics in psychology. Impact of Presumed Consent Legislation on Cadav-
While the research in behavioral finance and eric Organ Donation: A Cross-Country Study. Jour-
nal of Health Economics, 25(4): 599620.
consumptionsavings has been very active, Akerlof, George A. 1982. Labor Contracts as Partial
relatively few studies, instead, have tackled Gift Exchange. Quarterly Journal of Economics,
97(4): 54369.
mortgage markets, development, and politi- Akerlof, George A. 1991. Procrastination and Obedi-
cal decisions, fields ripe for exploration. The ence. American Economic Review, 81(2): 119.
applications to public policy are also likely Alpert, Marc, and Howard Raiffa. 1982. A Progress
Report on the Training of Probability Assessors. In
to become more common, along the lines Judgment Under Uncertainty: Heuristics and Biases,
of the 2006 Automatic Savings and Pension ed. Daniel Kahneman, Paul Slovic, and Amos Tver-
Protection Act inspired by the research on sky, 294305. Cambridge and New York: Cambridge
University Press.
default effects in 401(k) savings. I also expect Andreoni, James. 2006. Philanthropy. In Handbook
a continued interest in the market interaction of Giving, Altruism, and Reciprocity, Volume 2, ed.
Serge-Christophe Kolm, and Jean Mercier Ythier,
between standard and nonstandard agents, 120165. Amsterdam; London and New York:
as in section 5. In addition, while Psychology Elsevier, North Holland.
and Economics has focused mostly so far on Angeletos, George-Marios, David Laibson, Andrea
Repetto, Jeremy Tobacman, and Stephen Wein-
inexperienced agents, future research is likely berg. 2001. The Hyberbolic Consumption Model:
to document also the judgmental biases, such Calibration, Simulation, and Empirical Evaluation.
Journal of Economic Perspectives, 15(3): 4768.
as overconfidence, of experienced agents such Ariely, Dan, and George Loewenstein. 2006. The
as managers and politicians. Heat of the Moment: The Effect of Sexual Arousal
Future research is also likely to explore psy- on Sexual Decision Making. Journal of Behavioral
Decision Making, 19(2): 8798.
chological phenomena that have been largely Ariely, Dan, and Klaus Wertenbroch. 2002. Procrasti-
neglected. For example, emotions, automatic nation, Deadlines, and Performance: Self-Control by
Precommitment. Psychological Science, 13(3): 21924.
processing, and implicit discrimination are Asch, Solomon E. 1951. Effects of Group Pressure
likely to matter for economic decisions such Upon the Modification and Distortion of Judgment.
as divorce, judicial sentencing, and policing. In Groups, Leadership and Men, ed. Harold Guetz-
kow, 22236. Pittsburgh: Carnegie Press.
To conclude, in the ten years since Rabins Ashraf, Nava, Dean Karlan, and Wesley Yin. 2006.
1998 Journal of Economic Literature sur- Tying Odysseus to the Mast: Evidence from a Com-
vey, the field of Psychology and Economics mitment Savings Product in the Philippines. Quar-
terly Journal of Economics, 121(2): 63572.
has accomplished much in the mind of this Ausubel, Lawrence M. 1999. Adverse Selection in the
writer. More parsimonious models and a Credit Card Market. Unpublished.
Bain, Henry, and Donald S. Hecock. 1957. Ballot Posi-
boom of evidence from the field have con- tion and Voters Choice: The Arrangements of Names
tributed to integrate the laboratory findings on the Ballot and Its Effect on the Voter? Detroit:
in Psychology and Economics more into Wayne State University Press.
Baker, Malcolm, Richard S. Ruback, and Jeffrey Wur-
mainstream economics. Still, much remains gler. 2007. Behavioral Corporate Finance. In
to be done. Our knowledge of behavioral Handbook of Corporate Finance: Empirical Corpo-
rate Finance, Volume 1, ed. B. Espen Eckbo, 14586.
deviations is partial, with disparities by field, Amsterdam and Oxford: Elsevier, North-Holland.
limited use of certain methodologies, and Bandiera, Oriana, Iwan Barankay, and Imran Rasul.
few applications of important psychologi- 2005. Social Preferences and the Response to
Incentives: Evidence from Personnel Data. Quar-
cal phenomena. Hopefully, ten years from terly Journal of Economics, 120(3): 91762.
now, we will be able to assess quantitatively Barber, Brad M., and Terrance Odean. 2001. Boys
Will Be Boys: Gender, Overconfidence, and Com-
which psychological factors matter in which mon Stock Investment. Quarterly Journal of
decisions. Economics, 116(1): 26192.
368 Journal of Economic Literature, Vol. XLVII (June 2009)

Barber, Brad M., and Terrance Odean. 2008. All 1994. Exploring the Planning Fallacy: Why Peo-
That Glitters: The Effect of Attention and News ple Underestimate Their Task Completion Times.
on the Buying Behavior of Individual and Institu- Journal of Personality and Social Psychology, 67(3):
tional Investors. Review of Financial Studies, 21(2): 36681.
785818. Cain, Daylian M., George Loewenstein, and Don A.
Barber, Brad M., Terrance Odean, and Ning Zhu. Moore. 2005. The Dirt on Coming Clean: Perverse
Forthcoming. Systematic Noise. Journal of Finan- Effects of Disclosing Conflicts of Interest. Journal
cial Markets. of Legal Studies, 34(1): 125.
Barberis, Nicholas, Ming Huang, and Tano Santos. Camerer, Colin 2006. Behavioral Economics. In
2001. Prospect Theory and Asset Prices. Quar- Advances in Economics and Econometrics: Theory
terly Journal of Economics, 116(1): 153. and Applications, Ninth World Congress, Volume 2,
Barberis, Nicholas, Ming Huang, and Richard H. ed. Richard Blundell, Whitney K. Newey, and Tor-
Thaler. 2006. Individual Preferences, Monetary sten Persson, 181214. Cambridge and New York:
Gambles, and Stock Market Participation: A Case Cambridge University Press.
of Narrow Framing. American Economic Review, Camerer, Colin, Linda Babcock, George Loewenstein,
96(4): 106990. and Richard H. Thaler. 1997. Labor Supply of New
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny. York City Cabdrivers: One Day at a Time. Quar-
1998. A Model of Investor Sentiment. Journal of terly Journal of Economics, 112(2): 40741.
Financial Economics, 49(3): 30743. Camerer, Colin, Samuel Issacharoff, George Loewen-
Barberis, Nicholas, and Richard H. Thaler. 2003. A stein, Ted ODonoghue, and Matthew Rabin. 2003.
Survey of Behavioral Finance. In Handbook of Regulation for Conservatives: Behavioral Econom-
the Economics of Finance, Volume 1B, Financial ics and the Case for Asymmetric Paternalism. Uni-
Markets and Asset Pricing, ed. George M. Con- versity of Pennsylvania Law Review, 151: 121154.
stantinides, Milton Harris and Ren M. Stulz, 1053 Camerer, Colin, and Dan Lovallo. 1999. Overconfi-
1123. Amsterdam; London and New York: Elsevier, dence and Excess Entry: An Experimental Approach.
North Holland. American Economic Review, 89(1): 30618.
Barberis, Nicholas, and Wei Xiong. Forthcoming. Campbell, John Y., and John H. Cochrane. 1999. By
What Drives the Disposition Effect? An Analysis Force of Habit: A Consumption-Based Explanation
of a Long-Standing Preference-Based Explanation. of Aggregate Stock Market Behavior. Journal of
Journal of Finance. Political Economy, 107(2): 20551.
Benartzi, Shlomo. 2001. Excessive Extrapolation Card, David. 1994. Intertemporal Labour Supply: An
and the Allocation of 401(k) Accounts to Company Assessment. In Advances in Econometrics: Sixth
Stock. Journal of Finance, 56(5): 174764. World Congress, ed. Christopher A. Sims, 4980.
Benartzi, Shlomo, and Richard H. Thaler. 1995. Myo- Cambridge and New York: Cambridge University
pic Loss Aversion and the Equity Premium Puzzle. Press.
Quarterly Journal of Economics, 110(1): 7392. Card, David, and Dean Hyslop. 1997. Does Infla-
Benartzi, Shlomo, and Richard H. Thaler. 2001. Naive tion Grease the Wheels of the Labor Market? In
Diversification Strategies in Defined Contribution Reducing Inflation: Motivation and Strategy, ed.
Saving Plans. American Economic Review, 91(1): Christina D. Romer and David H. Romer, 71114.
7998. Chicago and London: University of Chicago Press.
Benartzi, Shlomo, and Richard H. Thaler. 2002. How Carroll, Gabriel D., James J. Choi, David Laibson, Bri-
Much Is Investor Autonomy Worth? Journal of gitte Madrian, and Andrew Metrick. Forthcoming.
Finance, 57(4): 15931616. Optimal Defaults and Active Decisions. Quarterly
Bergman, Nittai K., and Dirk Jenter. 2007. Employee Journal of Economics.
Sentiment and Stock Option Compensation. Jour- Charness, Gary, and Matthew Rabin. 2002. Under-
nal of Financial Economics, 84(3): 667712. standing Social Preferences with Simple Tests.
Bernard, Victor L., and Jacob K. Thomas. 1989. Post- Quarterly Journal of Economics, 117(3): 81769.
Earnings-Announcement Drift: Delayed Price Chetty, Raj, Adam Looney, and Kory Kroft. Forthcom-
Response or Risk Premium? Journal of Accounting ing. Salience and Taxation: Theory and Evidence.
Research, 27: 136. American Economic Review.
Bertrand, Marianne, Dean Karlin, Sendhil Mullaina- Choi, James J., David Laibson, and Brigitte Madrian.
than, Eldar Shafir, and Jonathan Zinman. Forthcom- 2006. Reducing the Complexity Costs of 401(k)
ing. Whats Advertising Content Worth? Evidence Participation through Quick Enrollment. National
from a Consumer Credit Marketing Field Experi- Bureau of Economic Research Working Paper
ment. Quarterly Journal of Economics. 11979.
Bewley, Truman F. 1999. Why Wages Dont Fall Dur- Choi, James J., David Laibson, Brigitte Madrian,
ing a Recession. Cambridge and London: Harvard and Andrew Metrick. 2004. For Better or Worse:
University Press. Default Effects and 401(K) Savings Behavior. In
Broadbent, Donald E. 1958. Perception and Commu- Perspectives on the Economics of Aging, ed. David
nication. New York: Pergamon Press. A. Wise, 81121. Chicago and London: University of
Buehler, Roger, Dale Griffin, and Michael Ross. Chicago Press.
DellaVigna: Psychology and Economics 369

Cicchetti, Charles J., and Jeffrey A. Dubin. 1994. A Duflo, Esther, William Gale, Jeffrey Liebman, Peter
Microeconometric Analysis of Risk Aversion and the Orszag, and Emmanuel Saez. 2006. Saving Incen-
Decision to Self-Insure. Journal of Political Econ- tives for Low- and Middle-Income Families: Evi-
omy, 102(1): 16986. dence from a Field Experiment with H&R Block.
Clotfelter, Charles T., and Philip J. Cook. 1993. The Quarterly Journal of Economics, 121(4): 131146.
Gamblers Fallacy in Lottery Play. Management Dufwenberg, Martin, and Georg Kirchsteiger. 2004.
Science, 39(12): 152125. A Theory of Sequential Reciprocity. Games and
Cohen, Lauren, and Andrea Frazzini. 2008. Eco- Economic Behavior, 47(2): 26898.
nomic Links and Predictable Returns. Journal of Edmans, Alex, Diego Garcia, and Oyvind Norli. 2007.
Finance, 63(4): 19772011. Sports Sentiment and Stock Returns. Journal of
Conlin, Michael, Ted ODonoghue, and Timothy J. Finance, 62(4): 196798.
Vogelsang. 2007. Projection Bias in Catalog Orders. Eisensee, Thomas, and David Stromberg. 2007. News
American Economic Review, 97(4): 121749. Droughts, News Floods, and U.S. Disaster Relief.
Conlisk, John. 1996. Why Bounded Rationality? Quarterly Journal of Economics, 122(2): 693728.
Journal of Economic Literature, 34(2): 669700. Eliaz, Kfir, and Ran Spiegler. 2006. Contracting with
Cowgill, Bo, Justin Wolfers, and Eric Zitzewitz. 2008. Diversely Naive Agents. Review of Economic Stud-
Using Prediction Markets to Track Information ies, 73(3): 689714.
Flows: Evidence from Google. Unpublished. Ellison, Glenn. 2006. Bounded Rationality in Indus-
Cronqvist, Henrik, and Richard H. Thaler. 2004. trial Organization. In Advances in Economics and
Design Choices in Privatized Social-Security Sys- Econometrics: Theory and Applications, Ninth
tems: Learning from the Swedish Experience. World Congress, Volume 2, ed. Richard Blundell,
American Economic Review, 94(2): 42428. Whitney K. Newey and Torsten Persson, 14274.
Dahl, Gordon, and Stefano DellaVigna. 2009. Does Cambridge and New York: Cambridge University
Movie Violence Increase Violent Crime? Quartery Press.
Journal of Economics, 124(2): 677734. Ellsberg, Daniel. 1961. Risk, Ambiguity, and the Sav-
Daniel, Kent, David Hirshleifer, and Avanidhar Sub- age Axioms. Quarterly Journal of Economics, 75(4):
rahmanyam. 1998. Investor Psychology and Secu- 64369.
rity Market Under- and Overreactions. Journal of Falk, Armin. 2007. Gift Exchange in the Field.
Finance, 53(6): 183985. Econometrica, 75(5): 150111.
De Bondt, Werner F. M., and Richard H. Thaler. 1985. Falk, Armin, and Andrea Ichino. 2006. Clean Evi-
Does the Stock Market Overreact? Journal of dence on Peer Effects. Journal of Labor Econom-
Finance, 40(3): 793805. ics, 24(1): 3957.
DellaVigna, Stefano, and Ethan Kaplan. 2007. The Farber, Henry S. 2005. Is Tomorrow Another Day?
Fox News Effect: Media Bias and Voting. Quarterly The Labor Supply of New York City Cabdrivers.
Journal of Economics, 122(3): 11871234. Journal of Political Economy, 113(1): 4682.
DellaVigna, Stefano, John A. List, and Ulrike Mal- Farber, Henry S. 2008. Reference-Dependent Prefer-
mendier. 2009. Testing for Altruism and Social ences and Labor Supply: The Case of New York City
Pressure in Charitable Giving. Unpublished. Taxi Drivers. American Economic Review, 98(3):
DellaVigna, Stefano, and Ulrike Malmendier. 2004. 106982.
Contract Design and Self-Control: Theory and Fehr, Ernst, and Simon Gchter. 2000. Fairness and
Evidence. Quarterly Journal of Economics, 119(2): Retaliation: The Economics of Reciprocity. Journal
353402. of Economic Perspectives, 14(3): 15981.
DellaVigna, Stefano, and Ulrike Malmendier. 2006. Fehr, Ernst, and Lorenz Goette. 2007. Do Workers
Paying Not to Go to the Gym. American Economic Work More If Wages Are High? Evidence from a
Review, 96(3): 694719. Randomized Field Experiment. American Eco-
DellaVigna, Stefano, and M. Daniele Paserman. 2005. nomic Review, 97(1): 298317.
Job Search and Impatience. Journal of Labor Eco- Fehr, Ernst, Georg Kirchsteiger, and Arno Riedl.
nomics, 23(3): 52788. 1993. Does Fairness Prevent Market Clearing? An
DellaVigna, Stefano, and Joshua M. Pollet. 2007. Experimental Investigation. Quarterly Journal of
Demographics and Industry Returns. American Economics, 108(2): 43759.
Economic Review, 97(5): 16671702. Fehr, Ernst, and Klaus M. Schmidt. 1999. A Theory of
DellaVigna, Stefano, and Joshua M. Pollet. Forth- Fairness, Competition, and Cooperation. Quarterly
coming. Investor Inattention and Friday Earnings Journal of Economics, 114(3): 81768.
Announcements. Journal of Finance. Fisman, Raymond, Shachar Kariv, and Daniel Mark-
DeLong, J. Bradford, Andrei Shleifer, Lawrence H. ovits. 2007. Individual Preferences for Giving.
Summers, and Robert J. Waldmann. 1990. Noise American Economic Review, 97(5): 185876.
Trader Risk in Financial Markets. Journal of Forsythe, Robert, Joel L. Horowitz, N. E. Savin, and
Political Economy, 98(4): 70338. Martin Sefton. 1994. Fairness in Simple Bargain-
Diamond, Peter, and Hannu Vartiainen, eds. 2007. ing Experiments. Games and Economic Behavior,
Behavioral Economics and Its Applications. 6(3): 34769.
Princeton and Oxford: Princeton University Press. Frederick, Shane, George Loewenstein, and Ted
370 Journal of Economic Literature, Vol. XLVII (June 2009)

ODonoghue. 2002. Time Discounting and Time Causal Effects of Ballot Order from a Randomized
Preference: A Critical Review. Journal of Economic Natural Experiment: The California Alphabet Lot-
Literature, 40(2): 351401. tery, 19782002. Public Opinion Quarterly, 72(2):
French, Kenneth R., and James M. Poterba. 1991. 21640.
Investor Diversification and International Equity Holt, Charles A., and Susan K. Laury. 2002. Risk
Markets. American Economic Review, 81(2): Aversion and Incentive Effects. American Eco-
22226. nomic Review, 92(5): 164455.
Fudenberg, Drew. 2006. Advancing Beyond Advances Hong, Harrison, and Jeremy C. Stein. 1999. A Unified
in Behavioral Economics. Journal of Economic Lit- Theory of Underreaction, Momentum Trading, and
erature, 44(3): 694711. Overreaction in Asset Markets. Journal of Finance,
Fudenberg, Drew, and David K. Levine. 2006. A 54(6): 214384.
Dual-Self Model of Impulse Control. American Hossain, Tanjim, and John Morgan. 2006. . . . Plus
Economic Review, 96(5): 144976. Shipping and Handling: Revenue (Non) Equiva-
Gabaix, Xavier, and David Laibson. 2006. Shrouded lence in Field Experiments on eBay. B.E. Journals
Attributes, Consumer Myopia, and Information Sup- in Economic Analysis and Policy: Advances in Eco-
pression in Competitive Markets. Quarterly Jour- nomic Analysis and Policy, 6(2): 127.
nal of Economics, 121(2): 50540. Huberman, Gur. 2001. Familiarity Breeds Invest-
Gabaix, Xavier, David Laibson, Guillermo Moloche, ment. Review of Financial Studies, 14(3): 65980.
and Stephen Weinberg. 2006. Costly Information Huberman, Gur, and Wei Jiang. 2006. Offering versus
Acquisition: Experimental Analysis of a Bound- Choice in 401(k) Plans: Equity Exposure and Num-
edly Rational Model. American Economic Review, ber of Funds. Journal of Finance, 61(2): 763801.
96(4): 104368. Huberman, Gur, and Tomer Regev. 2001. Contagious
Garicano, Luis, Ignacio Palacios-Huerta, and Canice Speculation and a Cure for Cancer: A Nonevent
Prendergast. 2005. Favoritism under Social Pres- That Made Stock Prices Soar. Journal of Finance,
sure. Review of Economics and Statistics, 87(2): 56(1): 38796.
20816. Isoni, Andrea, Graham Loomes, and Robert Sug-
Genesove, David, and Christopher Mayer. 2001. Loss den. 2008. The Willingness to PayWillingness to
Aversion and Seller Behavior: Evidence from the Accept Gap, the Endowment Effect, Subject Mis-
Housing Market. Quarterly Journal of Economics, conceptions, and Experimental Procedures for Elic-
116(4): 123360. iting Valuations: Comment. Unpublished.
Gilbert, Daniel T., Elizabeth C. Pinel, Timothy D. Ivkovic, Zoran, James M. Poterba, and Scott Weis-
Wilson, Stephen J. Blumberg, and Thalia P. Wheat- benner. 2005. Tax-Motivated Trading by Individ-
ley. 1998. Immune Neglect: A Source of Durability ual Investors. American Economic Review, 95(5):
Bias in Affective Forecasting. Journal of Personal- 160530.
ity and Social Psychology, 75(3): 61738. Iyengar, Sheena S., and Mark R. Lepper. 2000. When
Glaeser, Edward L. 2006. Paternalism and Psychol- Choice Is Demotivating: Can One Desire Too Much
ogy. University of Chicago Law Review, 73(1): of a Good Thing? Journal of Personality and Social
13356. Psychology, 79(6): 9951006.
Gneezy, Uri, and John A. List. 2006. Putting Behav- Johnson, Eric J., and Daniel Goldstein. 2003. Do
ioral Economics to Work: Testing for Gift Exchange Defaults Save Lives? Science, 302(5649): 133839.
in Labor Markets Using Field Experiments. Econo- Johnson, Eric J., John Hershey, Jacqueline Meszaros,
metrica, 74(5): 136584. and Howard Kunreuther. 1993. Framing, Probabil-
Gul, Faruk, and Wolfgang Pesendorfer. 2001. Temp- ity Distortions, and Insurance Decisions. Journal of
tation and Self-Control. Econometrica, 69(6): Risk and Uncertainty, 7(1): 3551.
140335. Josephson, Wendy L. 1987. Television Violence and
Haigh, Michael S., and John A. List. 2005. Do Pro- Childrens Aggression: Testing the Priming, Social
fessional Traders Exhibit Myopic Loss Aversion? An Script, and Disinhibition Predictions. Journal of
Experimental Analysis. Journal of Finance, 60(1): Personality and Social Psychology, 53(5): 88290.
52334. Kahneman, Daniel, Jack L. Knetsch, and Richard H.
Heidhues, Paul, and Botond Kszegi. 2008. Com- Thaler. 1986. Fairness and the Assumptions of Eco-
petition and Price Variation When Consumers Are nomics. Journal of Business, 59(4): S285300.
Loss Averse. American Economic Review, 98(4): Kahneman, Daniel, Jack L. Knetsch, and Richard H.
124568. Thaler. 1990. Experimental Tests of the Endow-
Hirshleifer, David, Sonya Seongyeon Lim, and Siew ment Effect and the Coase Theorem. Journal of
Hong Teoh. Forthcoming. Driven to Distraction: Political Economy, 98(6): 132548.
Extraneous Events and Underreaction to Earnings Kahneman, Daniel, and Amos Tversky. 1979. Pros-
News. Journal of Finance. pect Theory: An Analysis of Decision under Risk.
Hirshleifer, David, and Tyler Shumway. 2003. Good Econometrica, 47(2): 26391.
Day Sunshine: Stock Returns and the Weather. Karlsson, Niklas, George Loewenstein, and Duane J.
Journal of Finance, 58(3): 100932. Seppi. 2005. The Ostrich Effect: Selective Atten-
Ho, Daniel E., and Kosuke Imai. 2008. Estimating tion to Information about Investments. Unpublished.
DellaVigna: Psychology and Economics 371

Kim, E. Han, Adair Morse, and Luigi Zingales. 2006. The Role of Affect in Decision Making. In Hand-
What Has Mattered to Economics since 1970. book of Affective Sciences, ed. Richard J. Davidson,
Journal of Economic Perspectives, 20(4): 189202. Klaus R. Scherer, and H. Hill Goldsmith, 61942.
Knetsch, Jack L., and Wei-Kang Wong. Forthcoming. Oxford and New York: Oxford University Press.
The Endowment Effect and the Reference State: Loewenstein, George, Ted ODonoghue, and Matthew
Evidence and Manipulations. Journal of Economic Rabin. 2003. Projection Bias in Predicting Future
Behavior and Organization. Utility. Quarterly Journal of Economics, 118(4):
Kszegi, Botond, and Matthew Rabin. 2006. A Model 120948.
of Reference-Dependent Preferences. Quarterly Loewenstein, George, and Drazen Prelec. 1992.
Journal of Economics, 121(4): 113365. Anomalies in Intertemporal Choice: Evidence and
Krueger, Alan B., and Alexandre Mas. 2004. Strikes, an Interpretation. Quarterly Journal of Economics,
Scabs, and Tread Separations: Labor Strife and the 107(2): 57397.
Production of Defective Bridgestone/Firestone Loughran, Tim, and Jay R. Ritter. 1995. The New
Tires. Journal of Political Economy, 112(2): 25389. Issues Puzzle. Journal of Finance, 50(1): 2351.
Kruger, Justin. 1999. Lake Wobegon Be Gone! The Madrian, Brigitte, and Dennis F. Shea. 2001. The
Below-Average Effect and the Egocentric Nature Power of Suggestion: Inertia in 401(k) Participation
of Comparative Ability Judgments. Journal of Per- and Savings Behavior. Quarterly Journal of Eco-
sonality and Social Psychology, 77(2): 22132. nomics, 116(4): 114987.
Kube, Sebastian, Michel Andr Marchal, and Clem- Malmendier, Ulrike, and Devin Shanthikumar. 2007.
ens Puppe. 2008a. The Currency of Reciprocity Are Small Investors Naive about Incentives? Jour-
Gift-Exchange in the Workplace. University of nal of Financial Economics, 85(2): 45789.
Zurich Institute for Empirical Research in Econom- Malmendier, Ulrike, and Geoffrey Tate. 2005. CEO
ics Working Paper 377. Overconfidence and Corporate Investment. Jour-
Kube, Sebastian, Michel Andr Marchal, and Clem- nal of Finance, 60(6): 26612700.
ens Puppe. 2008b. Do Wage Cuts Damage Work Malmendier, Ulrike, and Geoffrey Tate. 2008. Who
Morale? Evidence from a Natural Field Experi- Makes Acquisitions? CEO Overconfidence and the
ment. Unpublished. Markets Reaction. Journal of Financial Economics,
Laibson, David. 1997. Golden Eggs and Hyperbolic 89(1): 2043.
Discounting. Quarterly Journal of Economics, Mankiw, N. Gregory, and Ricardo Reis. 2002. Sticky
112(2): 44377. Information versus Sticky Prices: A Proposal to
Laibson, David, Andrea Repetto, and Jeremy Tobacman. Replace the New Keynesian Phillips Curve. Quar-
2009. Estimating Discount Functions with Con- terly Journal of Economics, 117(4): 12951328.
sumption Choices over the Lifecycle. Unpublished. Mas, Alexandre. 2006. Pay, Reference Points, and
Landry, Craig E., Andreas Lange, John A. List, Michael Police Performance. Quarterly Journal of Econom-
K. Price, and Nicholas G. Rupp. 2006. Toward an ics, 121(3): 783821.
Understanding of the Economics of Charity: Evi- Mas, Alexandre, and Enrico Moretti. Forthcoming.
dence from a Field Experiment. Quarterly Journal Peers at Work. American Economic Review.
of Economics, 121(2): 74782. McClure, Samuel M., David Laibson, George Loewen-
Lee, Young Han, and Ulrike Malmendier. 2007. The stein, and Jonathan D. Cohen. 2004. Separate Neu-
Bidders Curse. National Bureau of Economic ral Systems Value Immediate and Delayed Monetary
Research Working Paper 13699. Rewards. Science, 306(5695): 50307.
Levitt, Steven D., and John A. List. 2007. What Do Mehra, Rajnish, and Edward C. Prescott. 1985. The
Laboratory Experiments Measuring Social Prefer- Equity Premium: A Puzzle. Journal of Monetary
ences Reveal About the Real World? Journal of Economics, 15(2): 14561.
Economic Perspectives, 21(2): 15374. Milgram, Stanley. 1963. Behavioral Study of Obedi-
List, John A. 2003. Does Market Experience Elimi- ence. Journal of Abnormal and Social Psychology,
nate Market Anomalies? Quarterly Journal of Eco- 67(4): 37178.
nomics, 118(1): 4171. Mullainathan, Sendhil, and Richard H. Thaler. 2001.
List, John A. 2004. Neoclassical Theory versus Pros- Behavioral Economics. In International Encyclo-
pect Theory: Evidence from the Marketplace. pedia of the Social and Behavioral Sciences, Volume
Econometrica, 72(2): 61525. 20, ed. Neil J. Smelser and Paul B. Baltes, 10941100.
List, John A. 2006. The Behavioralist Meets the Mar- Oxford and New York: Oxford University Press.
ket: Measuring Social Preferences and Reputation Odean, Terrance. 1998. Are Investors Reluctant to
Effects in Actual Transactions. Journal of Political Realize Their Losses? Journal of Finance, 53(5):
Economy, 114(1): 137. 177598.
List, John A., and David Lucking-Reiley. 2002. The Odean, Terrance. 1999. Do Investors Trade Too
Effects of Seed Money and Refunds on Charitable Much? American Economic Review, 89(5): 127998.
Giving: Experimental Evidence from a University ODonoghue, Ted, and Matthew Rabin. 1999a. Doing
Capital Campaign. Journal of Political Economy, It Now or Later. American Economic Review, 89(1):
110(1): 21533. 10324.
Loewenstein, George, and Jennifer S. Lerner. 2003. ODonoghue, Ted, and Matthew Rabin. 1999b.
372 Journal of Economic Literature, Vol. XLVII (June 2009)

Procrastination in Preparing for Retirement. In Schwarz, Norbert, and Gerald L. Clore. 1983. Mood,
Behavioral Dimensions of Retirement Economics, Misattribution, and Judgments of Well-Being: Infor-
ed. Henry J. Aaron, 12556. Washington, D.C.: mative and Directive Functions of Affective States.
Brookings Institution Press; New York: Russell Sage Journal of Personality and Social Psychology, 45(3):
Foundation. 51323.
ODonoghue, Ted, and Matthew Rabin. 2001. Choice Shleifer, Andrei. 2000. Inefficient Markets: An Intro-
and Procrastination. Quarterly Journal of Econom- duction to Behavioral Finance. Oxford and New
ics, 116(1): 12160. York: Oxford University Press.
Oyer, Paul, and Scott Schaefer. 2005. Why Do Some Shue, Kelly, and Erzo F. P. Luttmer. Forthcoming.
Firms Give Stock Options to All Employees?: An Who Misvotes? The Effect of Differential Cogni-
Empirical Examination of Alternative Theories. tion Costs on Election Outcomes. American Eco-
Journal of Financial Economics, 76(1): 99133. nomic Journal: Economic Policy.
Palacios-Huerta, Ignacio, and Oscar Volij. Forthcom- Simon, Herbert A. 1955. A Behavioral Model of Ratio-
ing. Field Centipedes. American Economic Review. nal Choice. Quarterly Journal of Economics, 69(1):
Paserman, M. Daniele. 2008. Job Search and Hyper- 99118.
bolic Discounting: Structural Estimation and Policy Simonsohn, Uri. Forthcoming. Weather to Go to Col-
Evaluation. Economic Journal, 118(531): 141852. lege. Economic Journal.
Phelps, E. S., and R. A. Pollak. 1968. On Second-Best Simonson, Itamar. 1990. The Effect of Purchase
National Saving and Game-Equilibrium Growth. Quantity and Timing on Variety-Seeking Behavior.
Review of Economic Studies, 35(2): 18599. Journal of Marketing Research, 27(2): 15062.
Plott, Charles R., and Kathryn Zeiler. 2005. The Sims, Christopher A. 2003. Implications of Rational
Willingness to PayWillingness to Accept Gap, the Inattention. Journal of Monetary Economics, 50(3):
Endowment Effect, Subject Misconceptions, and 66590.
Experimental Procedures for Eliciting Valuations. Skiba, Paige Marta, and Jeremy Tobacman. 2008.
American Economic Review, 95(3): 53045. Payday Loans, Uncertainty, and Discounting:
Pope, Devin G. 2007. Reacting to Rankings: Evi- Explaining Patterns of Borrowing, Repayment, and
dence from Americas Best Hospitals and Colleges. Default. Unpublished.
Unpublished. Strotz, R. H. 1956. Myopia and Inconsistency in
Rabin, Matthew. 1993. Incorporating Fairness into Dynamic Utility Maximization. Review of Eco-
Game Theory and Economics. American Economic nomic Studies, 23(3): 16580.
Review, 83(5): 12811302. Svenson, Ola. 1981. Are We All Less Ricky and More
Rabin, Matthew. 1998. Psychology and Economics. Skillful than Our Fellow Drivers? Acta Psycho-
Journal of Economic Literature, 36(1): 1146. ligica, 47(2): 14348.
Rabin, Matthew. 2000. Risk Aversion and Expected- Sydnor, Justin, 2006. Sweating the Small Stuff: Risk
Utility Theory: A Calibration Theorem. Economet- Aversion in Homeowners Insurance. Unpublished.
rica, 68(5): 128192. Terrell, Dek. 1994. A Test of the Gamblers Fallacy:
Rabin, Matthew. 2002a. Inference by Believers in the Evidence from Pari-mutuel Games. Journal of Risk
Law of Small Numbers. Quarterly Journal of Eco- and Uncertainty, 8(3): 30917.
nomics, 117(3): 775816. Thaler, Richard H. 1981. Some Empirical Evidence
Rabin, Matthew. 2002b. A Perspective on Psychol- on Dynamic Inconsistency. Economics Letters,
ogy and Economics. European Economic Review, 8(3): 20107.
46(45): 65785. Thaler, Richard H., and Shlomo Benartzi. 2004. Save
Rabin, Matthew, and Georg Weizscker. Forthcom- More Tomorrow: Using Behavioral Economics to
ing. Narrow Bracketing and Dominated Choices. Increase Employee Saving. Journal of Political
American Economic Review. Economy, 112(1): S16487.
Rashes, Michael S. 2001. Massively Confused Inves- Thaler, Richard H., and Hersh M. Shefrin. 1981. An
tors Making Conspicuously Ignorant Choices (MCI Economic Theory of Self-Control. Journal of Politi-
MCIC). Journal of Finance, 56(5): 191127. cal Economy, 89(2): 392406.
Read, Daniel, and Barbara van Leeuwen. 1998. Pre- Tversky, Amos, and Daniel Kahneman. 1974. Judg-
dicting Hunger: The Effects of Appetite and Delay ment under Uncertainty: Heuristics and Biases.
on Choice. Organizational Behavior and Human Science, 185(4157): 112431.
Decision Processes, 76(2): 189205. Tversky, Amos, and Daniel Kahneman. 1981. The
Rind, Bruce. 1996. Effects of Beliefs about Weather Framing of Decisions and the Psychology of Choice.
Conditions on Tipping. Journal of Applied Social Science, 211(4481): 45358.
Psychology, 26(2): 13747. Tversky, Amos, and Daniel Kahneman. 1992. Advances
Samuelson, William, and Richard Zeckhauser. 1988. in Prospect Theory: Cumulative Representation of
Status Quo Bias in Decision Making. Journal of Uncertainty. Journal of Risk and Uncertainty, 5(4):
Risk and Uncertainty, 1(1): 759. 297323.
Saunders, Edward M., Jr. 1993. Stock Prices and Wall Weinstein, Neil D. 1980. Unrealistic Optimism about
Street Weather. American Economic Review, 83(5): Future Life Events. Journal of Personality and
133745. Social Psychology, 39(5): 80620.
This article has been cited by:

1. Simon James. 2012. The contribution of behavioral economics to tax reform in the United
Kingdom. The Journal of Socio-Economics 41:4, 468-475. [CrossRef]
2. Peter Michael Dawson. 2012. Experience, social pressure and performance: the case of soccer
officials. Applied Economics Letters 19:9, 883-886. [CrossRef]
3. Saktinil Roy, David M. Kemme. 2012. Causes of banking crises: Deregulation, credit booms
and asset bubbles, then and now. International Review of Economics & Finance 24, 270-294.
[CrossRef]
4. Avi Goldfarb, Teck-Hua Ho, Wilfred Amaldoss, Alexander L. Brown, Yan Chen, Tony Haitao Cui,
Alberto Galasso, Tanjim Hossain, Ming Hsu, Noah Lim, Mo Xiao, Botao Yang. 2012. Behavioral
models of managerial decision-making. Marketing Letters 23:2, 405-421. [CrossRef]
5. Alex Krumer. 2012. Best-of-two contests with psychological effects. Theory and Decision .
[CrossRef]
6. S.D. Flm, O. Godal, A. Soubeyran. 2012. Gradient differences and bilateral barters. Optimization
1-20. [CrossRef]
7. Alexander L. Brown, , Colin F. Camerer, , Dan Lovallo. 2012. To Review or Not to
Review? Limited Strategic Thinking at the Movie Box Office. American Economic Journal:
Microeconomics 4:2, 1-26. [Abstract] [View PDF article] [PDF with links]
8. Ian Larkin, , Stephen Leider. 2012. Incentive Schemes, Sorting, and Behavioral Biases of
Employees: Experimental Evidence. American Economic Journal: Microeconomics 4:2, 184-214.
[Abstract] [View PDF article] [PDF with links]
9. LORETTA GARRISON, MANOJ HASTAK, JEANNE M. HOGARTH, SUSAN KLEIMANN,
ALAN S. LEVY. 2012. Designing Evidence-based Disclosures: A Case Study of Financial Privacy
Notices. Journal of Consumer Affairs no-no. [CrossRef]
10. Werner Sesselmeier. 2012. Widersprche sozialer Integration in Zeiten der konomisierung
sozialer Sicherung. Sozialer Fortschritt 61:5, 104-110. [CrossRef]
11. Parunchana Pacharn, Darlene Bay, Sandra Felton. 2012. Impact of a Flexible Evaluation System
on Effort and Timing of Study. Accounting Education 1-20. [CrossRef]
12. Derek C. Jones, Takao Kato. 2012. Incidence, coverage and employee knowledge of participation
in financial incentive schemes: evidence from US cases. The International Journal of Human
Resource Management 23:8, 1547-1569. [CrossRef]
13. Robert Metcalfe, Paul Dolan. 2012. Behavioural economics and its implications for transport.
Journal of Transport Geography . [CrossRef]
14. KATHERINE BAICKER, WILLIAM J. CONGDON, SENDHIL MULLAINATHAN. 2012.
Health Insurance Coverage and Take-Up: Lessons from Behavioral Economics. Milbank Quarterly
90:1, 107-134. [CrossRef]
15. David Gill, , Victoria Prowse. 2012. A Structural Analysis of Disappointment Aversion in a Real
Effort Competition. American Economic Review 102:1, 469-503. [Abstract] [View PDF article]
[PDF with links]
16. Damon Jones. 2012. Inertia and Overwithholding: Explaining the Prevalence of Income Tax
Refunds. American Economic Journal: Economic Policy 3:4, 158-185. [Abstract] [View PDF
article] [PDF with links]
17. James C. Cooper, William E. Kovacic. 2012. Behavioral economics: implications for regulatory
behavior. Journal of Regulatory Economics . [CrossRef]
18. TOMOMI TANAKA, TAKESHI MUROOKA. 2012. SELF-CONTROL PROBLEMS AND
CONSUMPTION-SAVING DECISIONS: THEORY AND EMPIRICAL EVIDENCE*.
Japanese Economic Review no-no. [CrossRef]
19. Ivo Vlaev, Ara DarziPreferences and Their Implication for Policy, Health and Wellbeing 305-336.
[CrossRef]
20. DANIEL F. STONE. 2012. TESTING BAYESIAN UPDATING WITH THE ASSOCIATED
PRESS TOP 25. Economic Inquiry no-no. [CrossRef]
21. David M. Kemme, Saktinil Roy. 2012. Did the Recent Housing Boom Signal the Global Financial
Crisis?. Southern Economic Journal 78:3, 999-1018. [CrossRef]
22. Johann Graf LambsdorffChapter 10 Behavioral and Experimental Economics as a Guidance to
Anticorruption 15, 279-300. [CrossRef]
23. 2011. Book Reviews. Journal of Economic Literature 49:4, 1241-1250. [Abstract] [View PDF
article] [PDF with links]
24. 2011. Book Reviews. Journal of Economic Literature 49:4, 1230-1317. [Citation] [View PDF
article] [PDF with links]
25. Avi Goldfarb, , Mo Xiao. 2011. Who Thinks about the Competition? Managerial Ability and
Strategic Entry in US Local Telephone Markets. American Economic Review 101:7, 3130-3161.
[Abstract] [View PDF article] [PDF with links]
26. Jidong Zhou. 2011. Reference Dependence and Market Competition. Journal of Economics &
Management Strategy 20:4, 1073-1097. [CrossRef]
27. John Gathergood. 2011. Self-control, financial literacy and consumer over-indebtedness. Journal
of Economic Psychology . [CrossRef]
28. Carl Andreas Claussen, Egil Matsen, istein Risland, Ragnar Torvik. 2011. Overconfidence,
monetary policy committees and chairman dominance. Journal of Economic Behavior &
Organization . [CrossRef]
29. P. Edwards. 2011. Experimental economics and workplace behaviour: bridges over troubled
methodological waters?. Socio-Economic Review . [CrossRef]
30. K. M. Marzilli Ericson, A. Fuster. 2011. Expectations as Endowments: Evidence on Reference-
Dependent Preferences From Exchange and Valuation Experiments. The Quarterly Journal of
Economics . [CrossRef]
31. Edward Miguel, , Shanker Satyanath. 2011. Re-examining Economic Shocks and Civil Conflict.
American Economic Journal: Applied Economics 3:4, 228-232. [Abstract] [View PDF article]
[PDF with links]
32. P. Dolan, M. Hallsworth, D. Halpern, D. King, R. Metcalfe, I. Vlaev. 2011. Influencing behaviour:
the mindspace way. Journal of Economic Psychology . [CrossRef]
33. T. van Huizen, J. Plantenga. 2011. Labour Market Effects of Unemployment Accounts: Insights
from Behavioural Economics. CESifo Economic Studies . [CrossRef]
34. Helge Berger, Michael Ehrmann, Marcel Fratzscher. 2011. Geography, skills or both: What
explains Fed watchers forecast accuracy of US monetary policy?. Journal of Macroeconomics
33:3, 420-437. [CrossRef]
35. Deon Filmer, Norbert Schady. 2011. Does more cash in conditional cash transfer programs always
lead to larger impacts on school attendance?. Journal of Development Economics 96:1, 150-157.
[CrossRef]
36. Lisa A. Robinson, James K. Hammitt. 2011. Behavioral Economics and Regulatory Analysis. Risk
Analysis 31:9, 1408-1422. [CrossRef]
37. Alex Marsh, Kenneth Gibb. 2011. Uncertainty, Expectations and Behavioural Aspects of Housing
Market Choices. Housing, Theory and Society 28:3, 215-235. [CrossRef]
38. Alessandro Bucciol. 2011. A NOTE ON SOCIAL SECURITY WELFARE WITH SELF-
CONTROL PROBLEMS. Macroeconomic Dynamics 1-16. [CrossRef]
39. Marco Manacorda, , Edward Miguel, , Andrea Vigorito. 2011. Government Transfers and Political
Support. American Economic Journal: Applied Economics 3:3, 1-28. [Abstract] [View PDF
article] [PDF with links]
40. Ricky C. Leung, Kalyan S. Pasupathy. 2011. The economics of social computing: Some
preliminary findings on healthcare organizations. Journal of Computational Science . [CrossRef]
41. Barbara Vis. 2011. Prospect Theory and Political Decision Making. Political Studies Review no-
no. [CrossRef]
42. sa Lfgren, Peter Martinsson, Magnus Hennlock, Thomas Sterner. 2011. Are experienced people
affected by a pre-set default optionResults from a field experiment. Journal of Environmental
Economics and Management . [CrossRef]
43. Keith Cowling, Stephen P. Dunn, Philip R. Tomlinson. 2011. Global imbalances and modern
capitalism: a structural approach to understanding the present economic crisis. Journal of Post
Keynesian Economics 33:4, 575-596. [CrossRef]
44. David L. Greene. 2011. Uncertainty, loss aversion, and markets for energy efficiency. Energy
Economics 33:4, 608-616. [CrossRef]
45. Niclas Berggren. 2011. Time for behavioral political economy? An analysis of articles in
behavioral economics. The Review of Austrian Economics . [CrossRef]
46. Duncan Maclennan, Anthony OSullivan. 2011. The global financial crisis: challenges for housing
research and policies. Journal of Housing and the Built Environment . [CrossRef]
47. Wenbo Cai, Ying-Ju Chen. 2011. Intertemporal pricing with boundedly rational consumers.
Journal of Revenue and Pricing Management . [CrossRef]
48. Antoni Cunyat, Randolph Sloof. 2011. Employee types and endogenous organizational design: An
experiment. Journal of Economic Behavior & Organization . [CrossRef]
49. Mark P. Taylor, Stephen P. Jenkins, Amanda Sacker. 2011. Financial capability and psychological
health. Journal of Economic Psychology . [CrossRef]
50. S. N. Ali. 2011. Learning Self-Control. The Quarterly Journal of Economics 126:2, 857-893.
[CrossRef]
51. Ines Fortin, Jaroslava Hlouskova. 2011. Optimal asset allocation under linear loss aversion.
Journal of Banking & Finance . [CrossRef]
52. Corina Paraschiv, Regis Chenavaz. 2011. Sellers' and Buyers' Reference Point Dynamics in the
Housing Market. Housing Studies 26:3, 329-352. [CrossRef]
53. Gwilym Pryce, Yu Chen, George Galster. 2011. The Impact of Floods on House Prices: An
Imperfect Information Approach with Myopia and Amnesia. Housing Studies 26:2, 259-279.
[CrossRef]
54. Alexander K. Koch, Julia Nafziger. 2011. Self-regulation through Goal Setting*. Scandinavian
Journal of Economics 113:1, 212-227. [CrossRef]
55. Bruno S. Frey, , David A. Savage, , Benno Torgler. 2011. Behavior under Extreme Conditions: The
Titanic Disaster. Journal of Economic Perspectives 25:1, 209-222. [Abstract] [View PDF article]
[PDF with links]
56. Bryan Bollinger, , Phillip Leslie, , Alan Sorensen. 2011. Calorie Posting in Chain Restaurants.
American Economic Journal: Economic Policy 3:1, 91-128. [Abstract] [View PDF article] [PDF
with links]
57. Devin G. Pope, , Maurice E. Schweitzer. 2011. Is Tiger Woods Loss Averse? Persistent Bias in the
Face of Experience, Competition, and High Stakes. American Economic Review 101:1, 129-157.
[Abstract] [View PDF article] [PDF with links]
58. D. Card, G. B. Dahl. 2011. Family Violence and Football: The Effect of Unexpected Emotional
Cues on Violent Behavior. The Quarterly Journal of Economics 126:1, 103-143. [CrossRef]
59. A. P. Zwane, J. Zinman, E. Van Dusen, W. Pariente, C. Null, E. Miguel, M. Kremer, D. S. Karlan,
R. Hornbeck, X. Gine, E. Duflo, F. Devoto, B. Crepon, A. Banerjee. 2011. Being surveyed can
change later behavior and related parameter estimates. Proceedings of the National Academy of
Sciences 108:5, 1821-1826. [CrossRef]
60. Rupert Sausgruber, Jean-Robert Tyran. 2011. Are we taxing ourselves?How deliberation and
experience shape voting on taxes. Journal of Public Economics 95:1-2, 164-176. [CrossRef]
61. David Card, Michael Ransom. 2011. Pension Plan Characteristics and Framing Effects in
Employee Savings Behavior. Review of Economics and Statistics 93:1, 228-243. [CrossRef]
62. Andrew PostlewaiteSocial Norms and Preferences, Chapter for the Handbook for Social
Economics edited by J. Benhabib, A. Bisin and M. Jackson 1, 31-67. [CrossRef]
63. Yair AharoniBehavioral Elements in Foreign Direct Investment Decisions 15, 23-60. [CrossRef]
64. Stephen Duncan, Cameron Hepburn, Antonis Papachristodoulou. 2011. Optimal harvesting of
fish stocks under a time-varying discount rate. Journal of Theoretical Biology 269:1, 166-173.
[CrossRef]
65. U. Schimank. 2011. Against all odds: the 'loyalty' of small investors. Socio-Economic Review 9:1,
107-135. [CrossRef]
66. Paul Heidhues, , Botond K#szegi. 2010. Exploiting Navete about Self-Control in the Credit
Market. American Economic Review 100:5, 2279-2303. [Abstract] [View PDF article] [PDF with
links]
67. Jose Apesteguia, , Ignacio Palacios-Huerta. 2010. Psychological Pressure in Competitive
Environments: Evidence from a Randomized Natural Experiment. American Economic Review
100:5, 2548-2564. [Abstract] [View PDF article] [PDF with links]
68. D. Wade Hands. 2010. BACK TO THE ORDINALIST REVOLUTION: BEHAVIORAL
ECONOMIC CONCERNS IN EARLY MODERN CONSUMER CHOICE THEORY.
Metroeconomica no-no. [CrossRef]
69. Xavier Gin, , Dean Karlan, , Jonathan Zinman. 2010. Put Your Money Where Your Butt Is: A
Commitment Contract for Smoking Cessation. American Economic Journal: Applied Economics
2:4, 213-235. [Abstract] [View PDF article] [PDF with links]
70. Justin Sydnor. 2010. (Over)insuring Modest Risks. American Economic Journal: Applied
Economics 2:4, 177-199. [Abstract] [View PDF article] [PDF with links]
71. Luca Zarri. 2010. Behavioral economics has two souls: Do they both depart from economic
rationality?. Journal of Socio-Economics 39:5, 562-567. [CrossRef]
72. Richard R. Nelson, Davide Consoli. 2010. An evolutionary theory of household consumption
behavior. Journal of Evolutionary Economics 20:5, 665-687. [CrossRef]
73. Stefano DellaVigna, Matthew Gentzkow. 2010. Persuasion: Empirical Evidence. Annual Review
of Economics 2:1, 643-669. [CrossRef]
74. Orazio P. Attanasio, , Guglielmo Weber. 2010. Consumption and Saving: Models of Intertemporal
Allocation and Their Implications for Public Policy. Journal of Economic Literature 48:3,
693-751. [Abstract] [View PDF article] [PDF with links]
75. Scott Hankins, Mark Hoekstra, Paige Marta Skiba. 2010. The Ticket to Easy Street? The Financial
Consequences of Winning the Lottery. Review of Economics and Statistics 110510162004036.
[CrossRef]
76. S. DellaVigna. 2010. Consumers Who Care. Science 329:5989, 287-288. [CrossRef]
77. Joseph Tao-yi Wang, , Michael Spezio, , Colin F. Camerer. 2010. Pinocchio's Pupil: Using
Eyetracking and Pupil Dilation to Understand Truth Telling and Deception in Sender-Receiver
Games. American Economic Review 100:3, 984-1007. [Abstract] [View PDF article] [PDF with
links]
78. Cameron Hepburn, Stephen Duncan, Antonis Papachristodoulou. 2010. Behavioural Economics,
Hyperbolic Discounting and Environmental Policy. Environmental and Resource Economics 46:2,
189-206. [CrossRef]
79. Peter Kooreman, Henritte Prast. 2010. What Does Behavioral Economics Mean for Policy?
Challenges to Savings and Health Policies in the Netherlands. De Economist 158:2, 101-122.
[CrossRef]
80. Rick Wilson. 2010. Behavioral Economics. Annual Review of Political Science 14:1,
110301102107010. [CrossRef]
81. Supreet Kaur, , Michael Kremer, , Sendhil Mullainathan. 2010. Self-Control and the Development
of Work Arrangements. American Economic Review 100:2, 624-628. [Citation] [View PDF article]
[PDF with links]
82. Raj Chetty, , Emmanuel Saez. 2010. Optimal Taxation and Social Insurance with Endogenous
Private Insurance. American Economic Journal: Economic Policy 2:2, 85-116. [Abstract] [View
PDF article] [PDF with links]
83. Justine Hastings, , Ebonya Washington. 2010. The First of the Month Effect: Consumer Behavior
and Store Responses. American Economic Journal: Economic Policy 2:2, 142-162. [Abstract]
[View PDF article] [PDF with links]
84. A. Etzioni, M. J. Piore, W. Streeck. 2010. Behavioural economics. Socio-Economic Review 8:2,
377-397. [CrossRef]
85. Yair AharoniBehavioral elements in foreign direct investments 23, 73-111. [CrossRef]
86. 2009. Book Reviews. Journal of Economic Literature 47:4, 1126-1160. [Abstract] [View PDF
article] [PDF with links]
87. Laszlo Csaba. 2009. From Sovietology to neo-institutionalism. Post-Communist Economies 21:4,
383-398. [CrossRef]
88. VICTOR STANGO, JONATHAN ZINMAN. 2009. Exponential Growth Bias and Household
Finance. The Journal of Finance 64:6, 2807-2849. [CrossRef]
89. A. Falk, J. J. Heckman. 2009. Lab Experiments Are a Major Source of Knowledge in the Social
Sciences. Science 326:5952, 535-538. [CrossRef]
90. Raj Chetty. 2009. Sufficient Statistics for Welfare Analysis: A Bridge Between Structural and
Reduced-Form Methods. Annual Review of Economics 1:1, 451-488. [CrossRef]
91. Raj Chetty, , Adam Looney, , Kory Kroft. 2009. Salience and Taxation: Theory and Evidence.
American Economic Review 99:4, 1145-1177. [Abstract] [View PDF article] [PDF with links]
92. David Tuckett. 2009. Addressing the Psychology of Financial Markets. Economics: The Open-
Access, Open-Assessment E-Journal . [CrossRef]