goals, methods, and progress in neuroeconomics

31
arec5Camerer ARI 26 April 2013 18:32 Goals, Methods, and Progress in Neuroeconomics Colin F. Camerer Division of Humanities and Social Sciences, California Institute of Technology, Pasadena, California 91125; email: [email protected] Annu. Rev. Econ. 2013. 5:16.116.31 The Annual Review of Economics is online at economics.annualreviews.org This articles doi: 10.1146/annurev-economics-082012-123040 Copyright © 2013 by Annual Reviews. All rights reserved JEL codes: D03, C90 Keywords fMRI, behavioral economics, neural circuitry, emotion, reward Abstract Neuroeconomics shares the main goals of microeconomics: to under- stand what causes choices, and the welfare properties of choice. The novel goal is linking mathematical constructs and observable behav- ior to mechanistic details of neural circuitry. Several complementary methods are used. An initial insight from neuroscience is that distinct systems guide choice: Pavlovian and instrumental conditioning (learn- ing) of state-value and response-value associations, overlearned hab- its, and model- (or goal-) directed value that requires deliberation. These systems can differ economically from rational choicefor example, habitual choices have low utility and price elasticities, whereas model-directed values are often constructed preferences. Neuroeconomics also provides evidence of situations in which utility maximization either works well (in simple binary choice) or benefits from the introduction of behavioral constructs. Neuroeconomics is well equipped to guide the theory of how choices depend on mental states, such as fear or cognitive load. Examples include extensive studies of risk and time preference, finance, and neural decoding of private information. 16.1 Review in Advance first posted online on May 3, 2013. (Changes may still occur before final publication online and in print.) Changes may still occur before final publication online and in print Annu. Rev. Econ. 2013.5. Downloaded from www.annualreviews.org by University of Minnesota - Twin Cities on 05/14/13. For personal use only.

Upload: colin-f

Post on 09-Dec-2016

220 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Goals, Methods, and Progress in Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Goals, Methods, andProgress in NeuroeconomicsColin F. CamererDivision of Humanities and Social Sciences, California Institute of Technology,Pasadena, California 91125; email: [email protected]

Annu. Rev. Econ. 2013. 5:16.1–16.31

The Annual Review of Economics is online ateconomics.annualreviews.org

This article’s doi:10.1146/annurev-economics-082012-123040

Copyright © 2013 by Annual Reviews.All rights reserved

JEL codes: D03, C90

Keywords

fMRI, behavioral economics, neural circuitry, emotion, reward

Abstract

Neuroeconomics shares the main goals of microeconomics: to under-stand what causes choices, and the welfare properties of choice. Thenovel goal is linking mathematical constructs and observable behav-ior to mechanistic details of neural circuitry. Several complementarymethods are used. An initial insight from neuroscience is that distinctsystems guide choice: Pavlovian and instrumental conditioning (learn-ing) of state-value and response-value associations, overlearned hab-its, and model- (or goal-) directed value that requires deliberation.These systems can differ economically from rational choice—forexample, habitual choices have low utility and price elasticities,whereas model-directed values are often constructed preferences.Neuroeconomics also provides evidence of situations in which utilitymaximization either works well (in simple binary choice) or benefitsfrom the introduction of behavioral constructs. Neuroeconomics iswell equipped to guide the theory of how choices depend on mentalstates, such as fear or cognitive load. Examples include extensivestudies of risk and time preference, finance, and neural decoding ofprivate information.

16.1

Review in Advance first posted online on May 3, 2013. (Changes may still occur before final publication online and in print.)

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 2: Goals, Methods, and Progress in Neuroeconomics

1. INTRODUCTION

Economics is the science of choice, constrained by scarce resources and institutional structure.For the past several decades, observable choices have been used to infer unobserved preferences(assuming plausible axiomatic restrictions), a method known as revealed preference. Revealedpreference has been (and always will be) a powerful, sweeping framework. There are two rea-sonable justifications for relying only on observable choices as data. First, revealed-preferenceinference from choice is what young economists learn in graduate school and do well, andtherefore it serves as a conventional way to cumulate knowledge. Second, if only choices can beobserved, then relying on the revelation of preferences is the best we can do—even in 2013. [It isnotable, in this regard, that many early economists, including Edgeworth, Ramsey, and Fisher,dreamed about having neuroeconomic tools such as a hedonimeter to observe utilities directly(see Colander 2007).]

Neuroeconomics rejects the first view (convention) and accepts the second one (nonchoicemeasurement was limited by technology and is now easier). Many biological correlates of choicecan now be observed, and causally influenced, in many ways that are new in scientific history.Therefore, if the focus on choices alone was constrained by technology, advances in technologymake this a good time to explore neural mechanisms underlying choice (as Edgeworth and otherspined for). Every individual choice—pulling a voting-booth lever, signing a mortgage document,swiping a credit card, planning to have a child—is made by brain activity. As economists, we agreeon the goal of wanting to know what causes choices under scarcity and institutional constraint.Brains make those choices. Therefore, studying how the brain makes choices could conceivablyimprove economics on its own terms. There is also no doubt that technologies for understandingbiology and brain activity are getting better and cheaper; we should substitute toward them inunderstanding choices, to some extent. Furthermore, the deepest possible understanding aboutthe biology of choice will permit the best policy making.

The argument about using newly available technology is the simplest positive argument forneuroeconomics. What are the objections?

One objection is that even though technology for measuring the biological mechanisms un-derlying choice is clearly advancing, an optimal division of labor should leave neuroscience solelyto neuroscientists. As economists, should we only pick through evidence from neuroscience andfind what we need? I think the answer is no. Specialized division of labor is limited by the needfor coordination (Becker & Murphy 1992). In science, extreme specialization severely limitsknowledge transfer for new syntheses. Most neuroscientists are interested in extremely basicdetails of cross-species functions (for which humans are rarely the ideal species) and not incomplex human choice.

The value of coordination creates a role for so-called bilingual economists who coproduceknowledge in ideally informative data in teams with expert neuroscientists. As a bonus, leadingneuroscientists who are interested in human choice often insist that they could use help fromeconomists to reduce complex brain activity to simple economizing processes.

A second objection is that economics, by conventional method or definition, canmake use onlyof choice data (Gul & Pesendorfer 2008). This view is sometimes called the fundamentalistKrepsian program, after Kreps (1979), who introduced the study of choices over sets of choices(menus). The fundamentalist principle of this program is that economic explanations can andmust use optimized choice over dynamically consistent preferences as a primitive. Introducingchoices over menus obeys this principle and creates opportunities to model the preferential ex-pression of psychological elements such as attention (Masatlioglu et al. 2012), temptation (Lipman &Pesendorfer 2013), and thinking cost (Ortoleva 2013). Note that a cognitive neuroscientist would

16.2 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 3: Goals, Methods, and Progress in Neuroeconomics

immediately think of attention, temptation, and effortful thinking as latent variables that inprinciple could be directly observed using measures of visual fixation, arousal, and brain activity,for instance.1

For example,Masatlioglu et al. (2012) show how attention to a particular choice in a set S canbe revealed by whether the choice changes when the object is removed from S. Attention,a property of the choice process, can then be revealed using only choices, from special types ofchoice sets. However, if choices over sets do reveal psychological variables, such as attention, thenthose types of choices also describe how to measure the psychological variable directly. Themethod is simple: Gather data on choices over sets that are theoretically known to reveal a variableV, such as attention. Then look for direct measures of cognitive or biological activity when Vhas a high revealed value (as inferred from choice), compared with a low revealed value of V.Any such number is a direct measure of V. For example, attention to choice objects can bemeasured directly using mouse clicks or visual eye fixations. (If subjects do not click on or lookat an object, they are not attending to it.) Masatlioglu et al. (2012, p. 2199) clearly endorse thisview (see also Spiegler 2008): “One can obtain such information from many sources, such aseye-tracking, functional magnetic resonance imaging, and the tracking system in the internetcommerce. . . . In this regard, our theory highlights the importance of other tools (besides observedchoice) that can shed light on the choice process rather than outcome.”

But why measure a subjective valueV biologically if it can be revealed by choices? The reasonis that a biological measure will make new predictions. For example, frameworks such asMasatlioglu et al.’s can make comparative static predictions about what happens if attention toa choice object is decreased. But nothing in the Krepsian approach, by its restrictive nature, tellsus how to actually move attention up or down to have a causal effect, where individual differencescome from, and so on. A fuller biological account will make many such predictions.

For example, studies described below indicate that activity in the dorsolateral prefrontal cortex(DLPFC), which is underneath the left temple, is active during self-control. This evidence led to theprediction that disruption of DLPFC activity would make people more impatient—as revealed bychoices—whichwas confirmed empirically (Figner et al. 2010).Knowledge of the location of brainactivity permitted a change in time preference, as observed in choices. This type of prediction couldnot have come from observing choices alone.

More predictions can be suggested. Suppose DLPFC activity is slow to come online in theadolescent brain (compared with other neural circuitry). Then the suggestion that the DLPFC isnecessary for self-control provides a clear biological reason why adolescents are often recklessand impulsive. More generally, having both choice data and direct measures of latent neuralvariables will create new hypotheses about factors that change mental states and therefore changechoices.

Another challenge for theKrepsian program is practical:Natural field data on choicewill rarelyhave the ideal properties required to infer preference structures. (And the more interesting thepreferences are, the more complicated are the choice sets needed to pin down those preferences,and the less likely it is that ideal choice data are available.) When more choice data are especiallydifficult to come by, nonchoice data, including neural measures, are potentially more valuable.And theories that make simultaneous predictions about choices and nonchoice data should betaken seriously because they can be tested (and improved) more easily (e.g., Rustichini 2008).

1Cognitive effort in principle should be especially amenable to identification with neural activity. Readers are referred toMcGuire & Botvinick (2009) for an example suggesting that the dorsolateral prefrontal cortex (DLPFC) encodes decisioncosts.

Neural circuit:anatomically distinctbrain area that isfunctionally connectedand mutually activatedduring behavior

16.3www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 4: Goals, Methods, and Progress in Neuroeconomics

The rest of the article proceeds as follows. Section 2 discusses three broad themes: evidence fordistinct neural valuation systems, the value of neural data in different types of economic theory,and reasons for caring about where in the brain activity occurs. Section 2 is designed to presentmaterial in a way that is comfortable and insightful for economists who are curious about neu-roscience but unfamiliar with it. Section 3 focuses on two topics of economic interest (inter-temporal and risky choice) and describes provocative findings in some detail. This narrow focusdemonstrates limits of those studies and educates readers a bit about how to consume neuro-science (see also Jones et al. 2009).

2. GENERAL THEMES

This section outlines one way in which neuroeconomic data and insights can be organized intocategories that are familiar in economics. Before proceeding, I note that awide variety ofmethods isused in neuroscience. Functional magnetic resonance imaging (fMRI) is the most glamorous be-cause it can record and visualize activity in thewhole brain. (It is also, as a result, probably themostoverreported in science journalism.) However, the temporal resolution of fMRI is slow (severalseconds), and it is the method with the highest marginal cost per data point. Table 1 lists differentmethods used in particular studies mentioned in this review (see Kable 2011 for a tutorial).

As the table suggests, neuroscientists often use a combination ofmethods together in a researchprogram. Everymethod has a fundamental weakness, for which some othermethod compensates.For example, the behavior of patientswith brain lesions in the amygdala, for example, is suggestiveof whether the amygdala is part of a neural circuit creating that behavior (e.g., loss aversion) (seeDe Martino et al. 2006, 2010). But patients with ideally localized lesions are rare, so conclusionsare never statistically strong.However, hypotheses derived frompatientswith lesions can be testedwith fMRI methods on normal patients, which are not as constrained by sample size.

2.1. Types of Valuation and Choice

The standard view in economics is that choices reveal utility-encoded preferences (which, for thepresent purposes, are also called subjective values). Decades of research with different methodsand species suggest that there are at least four neurally distinct systems for subjective valuation orchoice, shown inTable 1 (see Rangel et al. 2008) and Figure 1. Some of these systems are likely tocompute valuations that are like stable utilities, which are an input to constrained maximization,but others clearly guide choice without such stable valuations.

I first describe these choice systems roughly as neuroscientists comprehend them. (Note thatbecause the choices typically studied in neuroscience are overwhelmingly simple, and almost alldata come from nonhuman animals, my discussion speculates about the types of valuation, suchas deliberativemodel-directed choice, which are notwell understood in neuroscience.) Then I backup and ask how choices made under these systems are likely to deviate from constrained maxi-mization as economists use it.

The simplest system is Pavlovian (operant or passive) conditioning. This system learns to as-sociate physical states S with future rewards. In Pavlov’s famous example, a dog learns to assignvalue V(S) to the sensory mental state “a bell rings” because a ringing bell is followed by fooddelivery. Dogs value the bell-ringing state (in the revealed-preference sense that they will exerteffort to achieve it) because it predicts reward, but it can also be unlearned if the ringing bell isdisappointingly followed by no food (a process called extinction).

A second system, called instrumental or active conditioning, associates actions an organismtakes in stateswith later rewards. These passive and active systems are reasonablywell understoodneuroscientifically: Learning in both systems seems to be driven by prediction errors encoded in

Functional magneticresonance imaging(fMRI): method thatdetects blood oxygen-level dependent(BOLD) flow intobrainareas during(functional) behavior

Amygdala: temporallobe region that rapidlyencodes vigilance (e.g.,fear of shock) andeconomic variablesincluding ambiguityand loss

Conditioning: mentalassociation betweenpassive (Pavlovian)states or active(instrumental)responses and outcomevalues that co-occurwith those states orresponses

16.4 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 5: Goals, Methods, and Progress in Neuroeconomics

the ventral striatum. A prediction error is the difference between a reward and its predicted value.In instrumental conditioning, prediction error updates subjective values from action under certainstates, Q(s, a), which then guide future choices of actions.2 Reward values that are learned usingthese systems are the closest the brain comes to the economic construct of a stable utility fora choice.

There is also a fascinating type of spillover between these two systems called Pavlovian-instrumental transfer (PIT). In PIT, presenting a state variable that is passively associated witha valued reward can trigger the active action that typically yields the same reward. PIT may bea deep biological basis for many types of uninformative economic advertising, which do not giveprice or quality information directly but appear to influence choice. For example, suppose a personis (passively) conditioned to associate an image of an attractive person with a rewardingly wild,fun scene at a casino. The person goes to the casino a few times and learns to associate his or her(active) visits with reward. If there is strong PIT, whenever this person sees the attractive personimage, its association with rewarding casino fun will provoke him or her to take action and visitthe casino. To be sure, these PIT effects have mostly been demonstrated for simple associations insimple species, but something unconventional in standard theory is necessary to explain the natureand success of uninformative advertising, and PIT is an intriguing possible explanation.

A third system is habitual (also called overlearned). Because the brain is overwhelmed byconstant microdecision challenges in a changing environment, it is useful to offload valuationduring repeated low-value decisions to a rapid, implicit (generally subconscious) neural autopilot.Delicate experiments with rats have established a habit system that is distinct from deliberate(goal-directed) value computation. An everyday example for people is commuting to work. Afterthousands of trips home on the same route, a drivermight forget that an exit ramp is closed, or thatthe spouse called with a request to pick up milk (or a child), and take the familiar route home.Another potential example is mindless eating (Wansink 2010). Many clever experiments indicatethat the types of food and portions that people eat can be altered by simple changes in packagingand accessibility, asmany aspects of eating are highly habituated. That is, people eatwith their eyes(e.g., typically eat whatever is on their plates). The economic interpretation of these habits is thatagents have a direct preference for an action. As Balleine et al. (2008, p. 373) write, “habitualresponses are classically envisioned to arise from stimulus-response associations, which lack anyrepresentation of the rewarding outcome and are instead valued in the sense of having been‘stamped in’ by a history of reinforcement.” A habitual action value by definition will be un-responsive to prices, information, and changes in goal value—in the short run. Habitual short-runchoice elasticities will therefore be close to zero.

The fourth system is goal or model directed.3 Goal-directed choices pick actions given a par-ticular state. Because states can include information (e.g., “the fish is really good today”) or prices,goal-directed choice can exhibit sensitivity to economic variables that habitual and Pavlovianlearning does not. A model-directed system uses abstract information about how well choices in

2Computation of prediction error is ubiquitous in the brain (called predictive coding). For example, in the visual system, theretina is designed to largely ignore visual input that is expected and to specially encode visual surprises, which are passed on forfurther processing. Predictive coding accounts for the afterimage effect in vision: If one stares at a large red dot for a minuteand then shift one’s gaze to a white surface, one will see a green dot. The brain learns to predict green and then encodes thesudden lack of green as its opposite, a red percept.3The term goal directed is used more typically in computational neuroscience to refer to valuations of abstract goals (e.g.,eating for nutrition),which can therefore flexibly adjust to opportunity sets andprices. I introduce the termmodel directed hereto include important categories of human economic decisions that may be guided by complex abstract deliberation (e.g.,planning for retirement using an online calculator) and that are typically not guided much by personal learning, or weakly byobservational learning of choices and outcomes of others.

Striatum: part of thebasal ganglia thatappears to encodeanticipated rewards orprediction errors

Reward predictionerror: encoded rewardminus predictedreward; a key input tolearning state andresponse values

16.5www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 6: Goals, Methods, and Progress in Neuroeconomics

a particular state S satisfy different goals and then integrates different goal values [denotedM(s, a)]to make a choice. The model-directed system is dramatically different from the others because itdoes not need direct learning experience to value choices. That is, in model-free goal-directedchoice, the functionQ(s, a) is thought to be learned by experience; in model-based choice,M(s, a)comes from deliberation or communication but can be fine-tuned by experience. It is mostprevalent in humans (and arguably does not exist in a flexible form in any other species).

However, challenges arise because a model that computes value correctly often needs to in-tegrate value across different timescales and sensory dimensions. For many economically in-teresting choices, there will be a substantial conflict between goal values that needs to be resolved.

Table 1 Costs and applications of different methods in neuroscience

fMRI Lesions TMS EEG Decoding SCR Disorders Personality Reference

High marginalequipmentcost?

Yes No No No No No No No

High marginallabor cost?

Yes No No Yes Yes No No No

Ambiguityaversion

√ √ Hsu et al.(2005)

Loss-gainframinga

√ √ √ (ASD) De Martinoet al. (2006,2008), Roiseret al. (2009)

Timepreferencea

√ √ Peters & Büchel(2010)

Timepreference

√ Figner et al.(2010)

Strategicthinking

√ Coricelli &Nagel (2009)

Private-informationbargaininga

√ √ Yun et al.(2013)

Neurallyrevealedpreferencea

√ √ Smith et al.(2012)

Mechanismdesign

√ √ Krajbichet al. (2009b)

Anxiety √ Kang et al.(2012)

Emotionalregulation

√ √ Sokol-Hessneret al. (2009)

aThese subjects are discussed in depth in the text.Abbreviations: ASD, autism spectrum disorder; EEG, electroencephalography; fMRI, functional magnetic resonance imaging; SCR, skin-conductance re-sponse; TMS, transcranial magnetic stimulation.

16.6 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 7: Goals, Methods, and Progress in Neuroeconomics

Highly consequential, complex economic choices made for the first time are, by definition,made bymodel-directed valuation (perhapswith inputs from the other systems). Examples includechoosing a tax-deferred savings plan in one’s first job, deciding whether to have in vitro fertil-ization treatment, and choosing a new vacation spot.

Of course, even if integrating costs and benefits in a model-directed valuation is a normativestandard, the othermore primitive valuation systems are active. Indeed, the preferences of themosthighly evolved and automatic systems will often prevail unless there is adequate mental effort andself-control. For example, a teenage pregnancy could be seen as the result of a choice made largelyby evolutionarily prepared lust, with inadequate restraint from cortical model-directed value.

Keep in mind that revealed-preference theory is silent about how the brain makes choices.Therefore, it is conceivable that different types of simple and complex, familiar (highly learned)and novel choices are described by a common systemof rationality axioms.However, the evidencefrom neuroscience strongly implies that all these choices are not governed by similar principles.(For example, as noted above, habitual choices do not adjust for changes in choice values or prices,by definition.)

Now suppose we could predict or detect from neural activity which valuation system is beingused by a person during a particular choice. We could then predict that different types of choiceswill exhibit different kinds of consistency across time (habits) and across people (shared innatepreferences) and with respect to context, description, and procedure (goal directed). The foursystems are therefore likely to have different economic properties (Table 2).

Note that learned preferences can look like stable preferences in stationary environments.Habits are inflexible preferences over specific action choices—these will be stable, but are too sta-ble because they do not respond to changes in value and price. Finally, goal-directed choices couldsimply integrate stable preferences over goals, but when such choices are made in novel situationswith no innate or learned valuation, it is quite likely that goal-directed choices will be sensitive toshifts in attention, differences in how choices are described (framing), and misperceptions orstrategic naïveté in processing information about how well choices achieve different goals.

The existence of multiple neural systems of valuation and choice lies at the heart of the conflictbetween fundamentalist decision theory and neuroeconomics. Both sides agree that the goal ofeconomics is to predict choices and comparative statics effects. However, the fundamentaliststhink that only the exclusive use of revealed-preferences modeling to achieve this goal is necessary

StatesV(S)

Response

PIT

Learning (s)

maxa* M(s, a*)Habit Learning (a)

maxa* Q(s, a*)

Modeled

M(s, a*)

Learned

Q(s, a*)

OutcomesModeled

Instrumental

Figure 1

The relation between states, (action) responses, and outcomes in three types of valuation (habit, modeled,and instrumental). Original figure based on Balleine et al. (2008), drawn by Min Jeong Kang.

16.7www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 8: Goals, Methods, and Progress in Neuroeconomics

for economics. Neuroeconomists believe that because choices arise from different systems, un-derstanding choice deeply requires eventually understanding those systems.

2.2. Three Directions for Neuroeconomics: Rational, Behavioral, and MentalState–Dependent Choice

Better understanding of the link between neural mechanisms and choices is likely to lead the fieldof neuroeconomics in three directions.

2.2.1. Evidence for utility maximization in simple choice. For free choices between a smallnumber of objects (typically two), there is some evidence that each object is valued subjectively,values are compared, and the higher-value object is chosen most often (Platt & Glimcher 1999,Rangel & Hare 2010). In simple cases, diffusion-drift models in which visual attention accu-mulates value can often account for both choices and response times (Ratcliff 1978, Fehr&Rangel2011) and are consistent with direct recording of neural firing in monkey brains (Gold& Shadlen2007).

However, these subjective values, and associated neural measures of them, appear to adaptlocally to the range of numerical features or overall values of objects in the choice set (Seymour &McClure 2008, Padoa-Schioppa 2011, Soltani et al. 2012). Range adaptation is well establishedin sensory psychophysics (motivating the reference dependence in prospect theory) and at the levelof neural mechanisms (Glimcher 2008). Range adaptation is a substantial challenge for eco-nomic theory: Making predictions from one choice set to another will require a careful analysisof what context-sensitive utility inferred from one set of observed choices predicts about choicesin another domain.

2.2.2. Evidence for behavioral economics constructs. In many domains, there are now multiplecompeting rational and behavioral models of choices over risk, ambiguity, time, and social allo-cations and in different contexts. In some cases, two or more models can be interpreted as makingdifferent predictions about both choice patterns and underlying cognitive or neural mechanisms.

Table 2 Distinct neural systems for choice and valuation: their learning bases, response to value, and economic features

Neural system Learned?

Responds to

value? Features Example Interesting economic effects

Pavlovian states(passive)

Yes Yes Associates stateswith value

Hunger cued bybarbecue smell

Instrumental choice influencedby cued value (Pavlovian-instrumental transfer);basis for advertising?

Learned responses(active)

Yes Yes Learns responsevalue slowly;associative

Favorite foods Dynamic change in value

Habit Yes No Overlearned, as ifactions have value

Addiction Very low short-run elasticities

Model directed No Yes Constructed Retirement saving Sensitive to representation;sharp conflict withother systems

16.8 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 9: Goals, Methods, and Progress in Neuroeconomics

Neuroeconomics data to date are tentatively consistent with behavioral economics hy-potheses in several domains (two of those, time and risk, are discussed in detail below). Forexample, there is fMRI evidence for nonlinear probability weighting as in prospect theory (Hsuet al. 2009), crowding out of intrinsic incentives by extrinsicmonetary reward (Murayama et al.2010), and negative performance response to very high incentives (i.e., choking) (Chib et al.2012). Two fMRI studies comparing brain activity during hypothetical and real choicesshow substantial overlapping activity in value-computing areas, along with enhanced activityin the midbrain and cingulate cortex for consumer goods (Kang et al. 2011) and in the insulaand amygdala for aversive “bads” (e.g., eating unpleasant foods) (Kang & Camerer 2012).Those studies show that there is a biological basis for hypothetical reporting biases and pro-vide clues about how hypothetical protocols could be designed to forecast real choice moreaccurately.

2.2.3. Evidence for mental state–dependent choice. The most novel contribution from neuro-economics will probably come from showing the empirical relevance of how mental states in-fluence choice and describing the neural mechanisms of influences. Mental states of economicrelevance include conscious attention; visceral states such as fatigue, pain, and hunger (Loewenstein1996, 2005); cognitive overload; and emotions (e.g., fear, sadness, disgust, joy, anger).

An obvious example is sleep deprivation. Being tired creates slow, noisy decisions and is ap-parent in fMRI (Menz et al. 2012). Another example is anxiety, an emotion that is a likely basis ofa preference for early resolution of uncertainty (Caplin & Leahy 2001) that is measurable by skinconductance (e.g., Kang et al. 2012).

We can also consider the emotion of fear. From a rational choice point of view, fear could bea source of (dis)utility, information, or even constraint (see Manzini & Mariotti 2010 for re-search on moods). If fear is unpleasant, people may dislike it and pay to avoid it (e.g., buyinginsurance to buy peace of mind). Fear is also noisy information, a crudely informative earlywarning sign of possible danger. Fear can also constrain choice if it affects motor activity (a rapidfreezing response) or draws attentional resources away from deliberation about actual danger.

Although an emotion such as fear can be considered utility, information, and constraint, itis helpful to have a more compact way to represent its impact on economic choice. I suggest arevived (but disciplined!) concept of state dependence. Preferences clearly depend on exogenousphysical states in many cases (e.g., umbrella demand by unequipped tourists rises when it rains).Economists are wary of state-dependent preferences out of fear that any choice pattern could beexplained by invoking an appropriate state.

The approach envisioned here is different: Mental states can be observed directly (with theusual kinds of error) so that testable restrictions can be imposed on theories about how mentalstates influence choice, by measuring states and choices at the same time (and also causally ma-nipulating states). For example, fear can be defined as an empirically observable state that ismeasureable by a combination of self-report, autonomic nervous system response, facial response(musculature and expression), observer encoding, and fMRI signals. Hypotheses about howchanges in fear affect preferences, information, and constraint can then be tested with both directfear-state evidence and choices.

2.3. Why Care About Where?

Choice clearly occurs somewhere in the brain; why shouldwe care aboutwhere? The reason is thatthe location of neural circuitry will lead to different and better predictions. Economic theoristsare gifted at inventing formal explanations for observed anomalies, but testing many competing

Insula: region betweenthe frontal andtemporal lobesencoding interoceptivebodily feelings;activated by financialuncertainty and socialinequality

16.9www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 10: Goals, Methods, and Progress in Neuroeconomics

theories with field data and critical experiments has not typically been both conclusive and fast.Progress might be muchmore rapid if the unobserved variables in the theories could be associatedwith areas of brain activity and could be jointly tested using brain data and choices.

There are some other reasons to care about the location of activity in the brain. Because ofthe differences in structure and function in human and nonhuman brains, findings about humanbrain activity will predict either similar behavior (when regions are homologous or similar) ordifferent behavior in nonhumans. Knowing which regions are active during a choice enablescausal experiments in which activity is manipulated by transcranial magnetic stimulation (TMS),deep brain stimulation, or unfortunate lesions. Knowing which brain areas develop and degradeacross the human life cycle can yield predictions about age effects (especially for young, adoles-cent, and aging populations). Other kinds of individual difference characterization based onstructure or function are possible too [e.g., the anatomical volume of a person’s amygdala regionis associated with the size of their social network (see Bickart et al. 2011)].

3. SPECIAL TOPICS

The next two subsections describe neuroeconomic studies on risky choice and time preference. Ineach case, one featured study is an exemplar. The featured studies are described in detail to helpreaders understand why experiments are designed and analyzed as they are and to judge whatshould be convincing or merit tentative skepticism.

3.1. Risky and Uncertain Choice

There are many theories of choice under risk and uncertainty. Neuroeconomists look at theserisky choice theories and ask, what computational mechanisms in the brain are likely to imple-ment these theories, and what neural processes are interesting? The goal for economics of courseis to understand the mechanisms better to produce interesting predictions and permit causalexperiments.

The featured study here is framing, or description invariance, which is the invisible axiom thathas the requirement that the way in which choice consequences are described should not affect thechoices made. If salad dressing labeled “6% fat” is alternatively labeled as “94% fat-free,” salesshould not change.

In many natural settings, changes in description are likely to matter—the framers usually hopethat changes do matter—either by attracting attention or by fundamentally changing the hedonicproperties of goods. The laboratory paradigms, however, aim to avoid these potential confoundsby using framing changes that are transparent and value neutral.

3.2.1. Featured study 1: framing.

The first fMRI study of framing was done by De Martino et al. (2006). This study is featuredbecause it offers a somewhat novel perspective on framing (rooted in affective emotion), and itlaunched two subsequent studies.

Methods. In each trial, subjects were first presented with a monetary endowment (e.g., £50). Ina loss frame, they were told they could lose a certain amount or could choose a risky gamble witha probability P (shown on a pie chart) of losing 0 or a probability (1�P) of losing the endowment.The gamble’s expected value always equals the certain amount offered. In a gain frame, the sub-jects were told they could keep a certain part of the endowment or choose a risky gamble with

16.10 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 11: Goals, Methods, and Progress in Neuroeconomics

a probability of keeping none or all of the endowment. Four different endowments and fourprobabilities P were crossed, and each combination was presented twice for each frame. Thecertain “lose” or “keep” amount and paired gamble were presented together on the screen andalways had the same expected value. Subjects had 4 s to respond.

There were 288 trials. By experimental economics standards, this is an enormous number oftrials. De Martino et al. (2006) chose it because the signal-to-noise ratio in fMRI is low, so it iscommon to use 50–500 trials to detect differences in activity.Note that the design iswithin subjects(i.e., each subject sees some gain-framed choices and loss-framed choices). This is importantbecause most previous studies present only one frame (i.e., between-subject designs). It is oftenthought that tests of framing such as this may not show differences if subjects could easily noticethat the two frames have equivalent consequences and then could make equivalent choices.However, the big advantage to within-subject design is that the personal rate of framing-basedreversal can be computed for each subject (this cannot be done in a between-subjects design); inother words, each subject acts as his or her own control. De Martino et al. (2006) gambled onfinding weak effects to gain within-subject power.

Behavioral results. Given the large number of trials, and within-subject design, it is somewhatsurprising that clear framing effects exist. All subjects chosemore risks in the loss frame than in thegain frame, and this effect holds separately in all eight endowment3 probability conditions. In sixof those eight conditions, gain gambles were chosen on less than 50% of trials, and loss gambleswere chosen on more than 50%. DeMartino et al. (2006) included so-called catch trials in whichone choice had a much higher expected value to be sure subjects were attentive and motivated.Subjects made the high–expected value choice on more than 95% of the catch trials.

Neural results. DeMartino et al.’s (2006) first two results come from a typical subtraction event-related generalized linearmodel (GLM). This is a time-series regression inwhich the times atwhichevents occur are coded 1, and coded 0 otherwise, creating a time-series dummy variable for theentire time course of the experiment, punctuated by events. De Martino et al. (2006) use a spikedesign inwhich the regressor is aþ1 spike (or delta function) at the time of the choice screen onset.4

This particular GLM identifies brain regions that are unusually active only when the choice screenfirst appears.

To localize active brain regions, investigators use several statistical steps (see Poldrack et al.2011). Preprocessing spatially smoothes data; coregisters functional activity during task per-formance to a more slowly recorded, sharper anatomical image (of the kind used in medical MRIto detect abnormalities); and normalizes multiple brains onto a common template using a nine-dimensional stretch procedure.

De Martino et al.’s (2006) GLMwas run for each subject and each of 60,000 or so artificiallydefined 3-mm3 voxels in the brain (the use of voxels is a three-dimensional way of dicing up thebrain into small-enough units to be anatomically distinct). The left-hand-side variable is the BOLDblood-flow signal in a voxel (measured by magnetic properties of oxygenated blood); the right-hand-side variable is just a series of 0–1 dummy variables (convolved with a hemodynamic re-sponse function). This procedure yields a regression coefficient for each of the 20 subjects andfor each of the 60,000 voxels. Over 60,000 t tests of contrasts are then performed (at a second-

4BOLD blood-flow signals are not like light switches that turn on and off instantly; they are more like fountains that pulseand subside. To capture this property, the GLM time series of 0–1 events is convolved with (i.e., multiplied by) a functionthat represents a typical hemodynamic response function.

16.11www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 12: Goals, Methods, and Progress in Neuroeconomics

level analysis), using parameter estimates of the GLM. That is, we search for voxels that havea BOLD signal that is significantly positive across all 20 subjects. Typically, the procedure alsoconstrains the search to k voxels that are spatially adjacent (e.g., k ¼ 10) so that we identify onlycontiguous areas in which unusually high BOLD signals are observed in all k voxels.

InDeMartino et al. (2006), the first contrast uses events that are codedþ1 if they are certaindecisions in the gain frame (Gsure) and gamble decisions in the loss frame (Lgamble) and are coded�1 if they are the more rare choices: gamble in the gain frame (Ggamble) and certain choice in theloss frame (Lsure). The contrast is therefore identifying k-voxel areas that have unusually highactivity in the difference-in-differences subtraction (Gsure þ Lgamble) – (Ggamble þ Lsure). Notethat this contrast will erase any effects purely associated with the gain and loss frames and withcertain or gamble choices. What is identified is activity during the typical choice (a certain gainor a risky loss gamble) that is significantly higher than common activity in the atypical choice (acertain loss or a risky gain gamble). Then this contrast is entered into a one-sample t test acrossall subjects, and voxels with significant positive effects by this test are reported, correcting formultiple comparisons by using a low p value or in some more formal way. Positive coefficientsidentify voxels that are unusually active during the typical choices compared to atypical ones.Negative coefficients indicate significantly higher activity in the reverse contrast (Ggamble þLsure) > (Gsure þ Lgamble).

De Martino et al. (2006, p. 686) find regions in the bilateral amygdala that are significantlymore active when the typical choice (the certain gain or the risky loss gamble) is made. Theirinterpretation is that “increased activation in the amygdala . . . [supports] the hypothesis that theframing effect is driven by an affect heuristic underwritten by an emotional system.” This is nota rock-solid interpretation, however, because it is unclear whether amygdala activity representsan input to choice or reflects a hedonic reaction during or after choice.

The reverse contrast yields a clearer result. Significant activation here is present in the anteriorcingulate cortex (ACC), DLPFC, and insula. The ACC and DLPFC are often associated withconflict resolution, cognitive control, and response inhibition. For example, when one plays go–no go games such as Simon says, the ACC is active, trying to restrain overlearned motor activityif a command does not begin with the words “Simon says.” The DLPFC is involved in emotionregulation (e.g., Olsson & Ochsner 2008, Sokol-Hessner et al. 2013). It is quite plausible thatthese regions would be differentially active when people are making unconventional (and rarer)choices—gambling for gains, and swallowing certain losses. Furthermore, the insula is part of aninteroceptive system encoding bodily sensations (especially discomfort) in the brain. Insula activityin this GLM is consistent with the idea that gambling for gains and accepting losses feel risky oruncomfortable.

De Martino et al. (2006) also find a correlation between individual-level framing effects andactivity in the ventromedial prefrontal cortex (VMPFC). They conclude that “our findings supporta model in which the VMPFC evaluates and integrates emotional and cognitive information,thus underpinning more ‘rational’ (i.e., description-invariant) behavior.”5

As usual for initial studies, these are highly speculative claims. However, they are historicallyimportant (yet they appeared only 7 years ago!) because their results shift the thinking aboutframing from its original basis in perception to a newer affect-cognition basis.

5Similar tomany papers in neuroscience, it pays to read DeMartino et al.’s (2006) supplemental material as well as the (short)main paper. Their figure S1 shows how robust framing is across endowment and probability. Their figure S3 shows somewhatweaker activations in the atypical-choice contrast, which are useful for interpretingwhat the brain is doing. Their figure S1 is inthe supplementalmaterial to save space, and figure S3 is there because the p value is a little tooweak (p< 0.005, not quite at theamygdala-result level p < 0.001) to make the final cut into the main text.

16.12 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 13: Goals, Methods, and Progress in Neuroeconomics

Framingwas originally thought of as a phenomenon in valuation akin to figure-ground reversalsin perception (e.g., the vase-face picture) or hedonic adaption in sensory psychophysics. A morerecent view is that loss and gain frames evoke different kinds of affect, and limiting framing effectsreflects a kind of cognitive override of affect. Activity in the amygdala during the typical choice, andin the insula during the atypical choice, is consistent with this newer view (seeMiu & Crisan 2011).

De Martino and colleagues then extended their paradigm to individuals with autism andgroups with a particular genetic variation involved in affective processing.

Autism. De Martino et al. (2008) use the same gain-loss paradigm with 15 autism spectrum dis-order (ASD) subjects and 14 neurotypical controls matched, as a group, on age, gender, and IQ.Skin-conductance response (SCR) was recorded 5 s after their decision.6 The framing effect forASD subjects was only half the size of the effect for neurotypical matched controls. The controlsalso had a higher SCR response after loss choices than gain choices, but the ASD subjects showedno difference. De Martino et al. (2008) conclude that the ASD subjects lacked an affectivecoding that discriminated between loss and gain. This is consistent with speculation about spe-cific dysfunction in the amygdala in ASD (Baron-Cohen et al. 2000) and the more general in-terpretation that ASD is characterized by low emphasizing and high systematizing, compared toneurotypical subjects (Baron-Cohen & Belmonte 2005).

Genetics. The next study in DeMartino et al.’s series looks at the effect of genetic differences onthe brain and behavior (see the sidebar, Genetics and Economics). All aspects of human traits andbehavior are highly heritability (30–60%), but high heritability does not indicate which of the25,000 human genes—or which combinations of highly interacting genes, or which rare geneswith big effects—drive behavioral differences. Roiser et al. (2009) look at naturally occurringvariants of the 5-HTTLPR serotonin transporter gene. There are two common variants, referredtoas short (SS) and long.Peoplewith the SS-typeallelehave reducedproteinexpressionand increasedresting-stateamygdalaactivity, aswell asheightenedamygdala reactions toemotional stimuli (Haririet al. 2002, Canli et al. 2006, Munafo et al. 2008). Those with the SS-type allele also have less graymatter in the ACC and weaker amygdala-ACC connectivity (Pezawas et al. 2005). The 5-HTT SSvariant is also associated with affective disorders such as depression and traits such as neuroticism.

Given the role of amygdala in their framing study, Roiser et al. (2009) hypothesize that peoplewith the SS allele might exhibit stronger framing effects. Indeed, they find that SS carriers havelarger framing effects, and the interaction pattern expressed in the amygdala in DeMartino et al.’s(2006) study is present only in SS carriers (seeFigure 2).However, the behavioral part of this study—and most other single gene–behavior associations—should be treated with some skepticism.Simple associations such as this often fail to replicate robustly. Roiser et al.’s result is more likely tobe sturdy because they specifically chose to look at 5-HTT variants because of the earlier evidenceof the amygdala’s involvement in framing (De Martino et al. 2006) and related evidence that 5-HTT variants alter the amygdala’s functional activity. Furthermore, imaging brain activity ofpeople with different genes has proved to be more reliable than simple gene-behavior correlationbecause altering brain activity iswhat the geneswere designed to do.Moreover,Roiser et al. did seethat activity in the amygdala is different, using fMRI, for people with the different gene variants.

6SCR (also called galvanic skin response) measures electrodermal responses in the skin, which are altered by moistureassociatedwith sweating induced by arousal. It was first used in the early 1900s and is one of the cheapest andmost diagnosticmeasures of general arousal, which is manifested 1–2 s after an arousing event. For applications in experimental economics,readers are referred to Coricelli et al. (2010) and Kang et al. (2012).

16.13www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 14: Goals, Methods, and Progress in Neuroeconomics

Next I describe three topics: statistical moments, prospect theory, and causal changes in risktaking.

3.2.2. Statistical moments. An appealing model of risky choice involves weighting statisticalmoments of reward distributions and integrating them to form a choice value This approach ispopular in finance studies, in which the risk and return of asset values are integrated to determinevalue, and in behavioral ecology studies, inwhich animals are assumed to respond to themean andvariance in foraging for food. A moments-based approach also follows from a Taylor expansionof expected utility, so it should approximate choice for local small-scale decisions.

Several studies indicate that the mean and variance of rewards of different types are encodedin different brain regions (e.g., Platt & Huettel 2008). The mean of rewards activates striatalregions (Preuschoff et al. 2006). In fact, the striatum is activated by many different types ofrewards besides money, including attractive faces, anticipation of curiosity-provoking trivia, andreward prediction error.

The variance of rewards, often thought of as risk, seems to activate the insula, a region involvedin interoceptive integrationof emotional andcognitive information (Craig 2002,Mohr et al. 2010).The insula is activated by unpleasant bodily sensations, such as choking or smelling disgustingodors, and feeling physical or social pain (from being left out of a group activity). Activity inthe insula associated with economic risk therefore supports the hypothesis of risk as feelings(Loewenstein et al. 2001). This hypothesis is a sharp pivot from how risk is treated in finance andeconomics; it is an idea economists could refine formally and explore in many applications.

3.2.3. Prospect theory. Prospect theory is a psychophysically based theory of how risks areevaluated and combined. The central modification is that outcomes are encoded relative to a referencepoint. In addition, the decision disutility from anticipated losses is assumed to weigh dispropor-tionately more strongly than gains, captured by a loss aversion parameter l. Objective probabilitiesare assumed to be subjectively weighted nonlinearly so that low probabilities are overweighted andhigher probabilities are underweighted (see Barberis 2013). Evidence has now accumulated dem-onstrating neural activity associatedwith all three elements of prospect theory (e.g.,Hsu et al. 2009).

GENETICS AND ECONOMICS

Genes are inherited molecular units of DNA. They control cell growth in interaction with the environment. Manygenes have allele variations, which allow inference about what the gene does functionally (see Roiser et al. 2009).Economists should pay attention to genes because overall genetic influences are strong (every trait and behavior areheritable as approximately 30–60%of behavior results from shared genes), genes are preciselymeasured exogenousinstruments (i.e., people donot choose their genes), and genedistributions vary across populations (Beauchampet al.2011). Candidate-gene studies focus on a few gene-phenotype correlations. More recent genome-wide associationstudies (GWAS) use a large numbers of genes (>500,000) correlated with a single behavioral phenotype (withdemanding correction formultiple comparisons, usuallyp<5310�8). These studies have not generally resolved theheritability paradox—that overall inheritance is strong, but individual gene effects (and even GWAS) are usuallyweak. There has been some success with large samples, such as blood pressure (Levy et al. 2009). Large-scale dataaggregation across studies will definitely help (see http://www.ssgac.org). Another method involves endopheno-typing studies, which correlate gene alleles with both brain activity (e.g., Roiser et al. 2009) and behavior. Endo-phenotyping studies often yield insight with much smaller samples than candidate-gene and GWAS studies do.

16.14 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 15: Goals, Methods, and Progress in Neuroeconomics

3.2.4. Loss aversion. Until recently, most evidence of loss aversion in decisions was inferred fromhuman choices between monetary gambles with possible gains and losses. However, there is alsoevidence of loss aversion inmonkeys trading tokens for stochastic food rewards (Chen et al. 2006)and associated evidence of endowment effects in monkeys (Lakshminarayanan et al. 2008).

The first fMRI study (Tom et al. 2007) showed comparable neural activity in several value-related brain regions, during evaluation of gambles, in response to increases in possible gains andreductions in possible losses. Across subjects, the difference in brain response in these identifiedregions to potential loss dollar for dollar, relative to potential gain (neural loss aversion), wascorrelated with the degree of loss aversion (l) inferred behaviorally from choices among gambles.Whereas this study indicates a common basis for reduced loss and increased gain, other studiesindicate different locations of brain activity for loss and gain. For example, Yacubian et al. (2006)find gain activity in the ventral striatum and loss activity in the amygdala and temporal lobe

b c

–1.5

–1.0

–0.5

0

0.5

1.0

1.5

Gamble Sure Gamble

Gain frame SSLoss frame SS

Sure

Para

met

er e

stim

ate

from

am

ygda

la

Para

met

er e

stim

ate

from

am

ygda

la

–1.5

–1.0

–0.5

0

0.5

1.0

1.5

Gamble Sure Gamble Sure

Gain frame SSLoss frame SS

0

1

2

3

4y = –7a

t value

Figure 2

Difference in amygdala response evoked by decisions made in accord with the framing effect between two genotype groups. (a) Asignificant difference was detected between the two genotype groups in the framing effect difference-in-difference contrast in the leftamygdala (circled in red). The Talairach coordinates of the peak voxel are x¼�24, y¼�4, and z¼�15. Right brain areas are shown onthe right of the image. The color bar represents t values [reported regions are significant at p < 0.005 (uncorrected)]. (b,c) Plots ofthe mean and standard error across subjects (second-level analysis) of the neural activity b in the amygdala [the peak voxel for the GLMframing effect contrast (Gsure þ Lgamble) – (Ggamble þ Lsure)]. Panels b and c show 5-HTT short (SS) and long (LA) allele subjects,respectively. The heightened amygdala response to default choices (sure gains and loss gambles) is evident only in the SS group (b). Figurereprinted from Roiser et al. (2009) with permission of the Journal of Neuroscience.

16.15www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 16: Goals, Methods, and Progress in Neuroeconomics

regions lateral to the striatum. A later study showed that two patients with selective bilateralamygdala lesions exhibited no loss aversion (De Martino et al. 2010).

3.2.5. Causal manipulations. Conventional economic analyses typically draw predictive powerby assuming the stability of preferences, using previous choice data to infer preferences (e.g., byestimating demand elasticities) and then—holding preferences fixed—predicting a comparativestatic change in choices based on changes in information, prices, or income. However, as the neuralcircuitryunderlying choice becomesbetter understood, itwill be possible to causally influenceneuralcomputations reliably in various ways and thereby change choices. Several studies have alreadyshown such causal influences.

Risk aversion seems to be causally increased by stress (induced by the immersion of hands incold water) (Porcelli & Delgado 2009), stimulation (upregulation) of the right DLPFC (Fecteauet al. 2007), the sight of negative-affect images before choice (Kuhnen & Knutson 2011), andeating food (Symmonds et al. 2010). It seems to be causally decreased by disruption of theright DLPFC (Knoch et al. 2006a), transcranial direct-current stimulation in older adults(Boggio et al. 2010), and a decrease in serotonin inmacaqueswith a depletion of tryptophan (Longet al. 2009). Loss aversion can be downregulated by a perspective-taking instruction to “thinklike a trader” and combine losses and gains mentally (Sokol-Hessner et al. 2009).

There are two lessons from these causal biological experiments. First, exogenous changes to theneural circuitry that make computations leading to risk-avoiding behavior can directly changechoices. These effects do not result from changes in prices, information, or constraints (in anytypical sense). Second, because these effects are often large in magnitude, and could be associatedwith exogenous events in the economy, they suggest a potentially useful expansion of the rationalchoice view in economics to include computational circuitry.

3.3. Time preference

The biological basis of time preference is a ripe topic for neuroeconomics for two reasons. First,the ability to plan ahead and trade off distant rewards with immediate rewards is a distinctlyhuman skill (unless evolved into simpler prepared behaviors, such as squirrels storing nuts for thewinter based on weather cues). The valuation of future rewards also increases sharply within thehuman developmental life cycle, as children learn that there is a tomorrow, and a next month,and create the neural capacity to encode what those future events mean. Because the valuation offuture rewards exhibits large cross-species and human age-dependent differences, there is amplescope to find associations of future valuation with regional brain activity.

Second, the rewards we are receiving right now are easily encoded, often in a visceral multi-sensory mode—a hand cradles a cheeseburger bun; the nose smells charred meat; and the brainimagines tasting crunchy lettuce, salty sweet pickles, and tang of mustard. But while thatcheeseburger is in front of you (and activating your brain) now, future rewards exist only aspossibilities in the brain. There is much scope for brain activity to influence how the future isimaged and compared to the present. This section describes results from studies using personality,fMRI, TMS, and lesions in animals and in people. All these data are building up a tentative pictureof the complicated neural circuitry that encodes immediate rewards, and imagines and valuatesfuture rewards, to guide intertemporal choice.

3.3.1. Two prominent views: dual systems and self-control, or a unified system. At least twocomputational views have been offered about how the brain encodes immediate and futurerewards tomake intertemporal choices (seeCarter et al. 2010). The more common and interesting

16.16 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 17: Goals, Methods, and Progress in Neuroeconomics

view is that multiple value systems are integrated, perhaps using self-control to avoid immediatetemptation that is bad in the future. A simpler view is that there is a single computed value.

Dual systems. One possibility is that immediate and delayed rewards are valued in separate cir-cuits, and value is integrated in some way. This is certainly plausible evolutionarily for choicesbetween a physically present reward (a bird in the hand) and a speculative imagined reward [two(birds) in the bush]. A sharp instantiation of the two-system view is the b-d model of quasi-hyperbolic discounting. In this approach, all future rewards are discounted by a common im-mediacy parameter b and are also exponentially discounted by bdt. This model deliberatelygenerates dynamic inconsistencies because the b parameter divides out when comparing two

delayed rewards 5 and 7, which have relative valuebd5u5bd7u7

¼ u5d2u7

periods away. However,

when period 5 arrives, the compared rewards have value u5 and bd7–5u7; the immediate rewardis then more attractive.

To demarcate the roles ofb and d in valuation, one can write the overall value V(t) of a streamof consumption ct as

VðtÞ ¼�1b

� 1�uðctÞ þ

X1t¼0

dtuðctþtÞ.

If b ¼ 1, the first term disappears; what is left is the standard exponential discounting of futureconsumption utilities. If b< 1, the first term exaggerates the weight on the immediate utility u(ct),reflecting an immediacy preference or present bias.

McClure et al. (2004, 2007) present evidence for the two-system view with two fMRI studies(Figure 3). They suggest that the b system includes the mesolimbic dopaminergic regions [thestriatum and medial orbitofrontal cortex (MOFC)] of the dopamine system and that the d systemincludes the DLPFC and parietal cortex.

A unified system. The key behavioral feature of b-d discounting is also present in genuine hy-perbolic discounting—value drops off rapidly for short delays from the present and drops offmore slowly (at the margin) for longer delays. Kable & Glimcher (2007) fit hyperbolic discountfunctions to choices between a fixed immediate reward and a variety of delayed rewards andcorrelate revealed (fitted) subjective values to neural activity. Their conclusion is contrary to theb-d neural view of McClure et al. (2004) and is sharply worded:

Our findings falsify a hypothesis regarding the neurobiological basis of intertemporal choice

[McClure et al. 2004]. [They] hypothesized that these same three regions—the ventral striatum,

medial prefrontal cortex and posterior cingulate cortex—form an impulsive neural system that

exclusively or primarily values immediate rewards. . . .This conclusion was based principally on the

finding that these three areas showed greater activity for choices that involved an immediate re-

ward than for choices that involved only delayed rewards. . . . Our observation that activity in these

regions varies when only the delayed reward changes falsifies the hypothesis that these regions

exclusively value immediate rewards. (Kable & Glimcher 2007, p. 1631)

Other studies provide mixed evidence that sometimes reconciles and sometimes casts doubton earlier interpretations. One study suggests that immediate rewards more strongly activatevaluation networks, even when they are matched on preference (Luo et al. 2009). Together, thisfinding and McClure et al.’s (2004) can be reconciled if activity observed in the latter study in

16.17www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 18: Goals, Methods, and Progress in Neuroeconomics

response to a possible immediate reward indicates valuation and some attention or other func-tion that does not influence expressed preference.

Another important study (Sellitto et al. 2010) demonstrates that patients with MPFC damageare clearly more impatient than are matched neurotypical patients, and a larger damage volumeis correlated with more impatience. McClure et al. (2004) conjecture that the MPFC is part of animmediate-valuation b system. In a computational sense, damage to this area could erase anycalculation involving b, leaving only d-based exponential discounting.

Self-control. Dual-system accounts often include the idea that there is an internal struggle forcontrol between the two systems. Thaler & Shefrin (1981) introduce this idea, in the form ofa myopic “doer” constrained by a farsighted “planner,” and use it to explain some stylized factsabout consumer spending.7 Fudenberg& Levine (2006) offer a similar model in which a long-runself incurs costs to control a myopic short-run self.

Hare et al. (2009) use tempting food choices to study planner-doer self-control circuitry. Subjectsrated foods by health and taste and chose between foods. The interesting category comprises choicesincluding tempting foods that are high taste and low health (candy or salty or fatty snacks). Theauthors find that the left DLPFC is more active during choices by successful self-controllers (those

x = 0 mm x = –44 mm x = 4 mm y = –16 mm

x = 0 mm x = –44 mm x = 4 mm y = –16 mm

Both

β areas (p < 0.001)δ areas (p < 0.001)

δ areas (p < 0.01) β areas (p < 0.01)

MoneyJuice

Figure 3

Brain areas that are consistently activated for intertemporal choices for two different rewards. The brain areas active during the juicechoice (red) overlap with those active during the monetary reward choice (green) (McClure et al. 2004). Overlapping voxels are shownin yellow. (a) For candidate d areas, overlap is substantial in the two studies at p < 0.001 and is stronger at p < 0.01. (b) For b areas,the general regions of activity are consistent across studies, but only seven voxels (yellow) within larger regions are overlapping (p < 0.001).Figure reprinted from McClure et al. (2007) with permission of the Journal of Neuroscience.

7Shefrin & Thaler (1988, footnote 2) state that “our economic theory of choice is roughly consistent with the scientificliterature onbrain function [associating the plannerwith the prefrontal cortex and the doerwith the limbic system]. . . . It iswellknown that self-control phenomena center on the interaction between prefrontal cortex and the limbic system.”

16.18 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 19: Goals, Methods, and Progress in Neuroeconomics

who turn down tempting foods most often) and is also more active, trial by trial, when temptingfoods are turned down in favor of the reference food. Furthermore, the successful self-controllershave a stronger MOFC response to the health rating of foods when they are presented for choice(the unsuccessful have no health response).8 This activation suggests that the DLPFC is regulatingthe weight given to health properties of food, as encoded in the MOFC.

Hare et al.’s (2009) results offer an interesting alternative to one-system valuation basedpurely on discounting. The role of theDLPFC in creating self-control is also consistentwith a dual-system view in which the DLPFC acts as a patient planner that controls a doer system. Theimportance of DLPFC activity in achieving successful self-control is corroborated by evidencefrom four other sources: TMS, adolescents and nonhuman species, working memory, and cog-nitive load and perception. If the left DLPFC is crucial for self-control, then temporary disruptionof that region by TMS should make people more impatient. Figner et al. (2010) find exactly thiseffect of TMS (although the effect is not large in magnitude and is evident only when earlier andlater rewards are similar in subjective value). That TMS causes impatience is as expected ifthe planner system uses the DLPFC to exert cognitive control over a myopic doer; disruptingplanner activity will lead the impatient doer to take over.

Additionally, a clear finding from fMRI studies is that prefrontal circuitry is relatively slowto neurally develop during adolescence (compared to reward and affective circuitry) (Casey et al.2008). Adolescents are also behaviorally impulsive toward the future on many importantdimensions. The conjunction of slow prefrontal development and impulsivity is consistent withthe role of the DLPFC noted by Hare et al. (2009).

The DLPFC is active during self-control and is also routinely activated in studies of bothinhibitory self-control and in working memory. Furthermore, IQ and working memory are(somewhat) stable traits that aremodestly associatedwith patience (Shamosh&Gray 2008, Burkset al. 2009, Benjamin et al. 2013). One remarkable study used fMRI during a go–no go task on 26adults who had been subjects 40 years earlier in Mischel’s famous marshmallow test of delayedgratification. The adults who were patient as children showed greater prefrontal cortex dif-ferences in the no go trials requiring inhibition, and the previously impatient adults showed highventral striatal activity (Casey et al. 2011).

Finally, there are many interesting results showing how patience is affected by cognitive pro-cessing.Rodriguez et al. (1989) find that children are lesswilling towait for food rewardswhen thefood is visible (cf. Bushong et al. 2010) and more willing to wait when reappraising rewards (e.g.,think of pretzels as logs and marshmallows as clouds) (see also Loewenstein 1996). Zaubermanet al. (2009) andRadu et al. (2011) show howphysical and temporal distance appears to be linkedmentally and how discounting can be changed based on that link. There is some evidence thatcognitive load reduces patience (Shiv & Fedorikhin 1999, Hinson et al. 2003) and that willpoweris a general resource that creates impatience when it is sapped by a previous task (Baumeister &Vohs 2003). Hershfield et al. (2011) find evidence that the use of facial-morphing software toallow subjects to interact with aged versions of their future selves increased patience. People arealso more patient when making choices for someone else (Albrecht et al. 2011).

Next we discuss this interesting nexus between memory and time preference in a specific fea-tured study.

8McClure et al. (2004, 2007) find that the BOLD signal in the prefrontal cortex and parietal cortex is stronger when adelayed reward is chosen relative to trials in which an immediate reward is chosen.

16.19www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 20: Goals, Methods, and Progress in Neuroeconomics

3.3.2. Featured study 2: episode tagging. Peters & Büchel (2010) report a study using fMRI;a behavioral manipulation, called episode tagging; and a variety of analyses.9

Method. Peters & Büchel (2010) had N ¼ 30 subjects first complete a delay discounting task to

estimate individual-specific discount rates k from the hyperbolic form1

1þ kt. The subjects then

listed events they had planned in the next 7 months (e.g., vacations, parties, classes) and rated theevents on a scale from one to six on personal relevance, arousal (intensity), and valence (good/bad). These events are called episode tags. The key idea is to see whether describing dates at whichfuture rewards will be dispensed by invoking the episode tags of those dates—personalized eventsthat the subjects themselves supply and expect to happen—will make future rewards more salientand more frequently chosen. For example, a control condition offers a reward of $10 now or$50 three months from now. The episode condition might offer a reward of $10 now or $50three months from now, when the subject will be in Tokyo at a conference, for example.

For each subject, seven specific events were chosen with various lengths of delay to theevent and that roughly matched on the three ratings. These seven events became the key episodetags that drove the experiment.10 Note that each episode tag is created by the subjects themselves.

On the second day, the subjects made a series of 118 choices during fMRI between an im-mediate 20 euro reward (fixed across all trials) and different delayed rewards. (One choice wasselected at random and paid.) A verbal episode tag was presented in the episodic condition;the reward amount and delay length were presented in both conditions. After scanning, subjectsagain rated the seven episode tags on the three scales, which were then combined into a singleimagery score.

Behavioral results. The estimated discount parameter k is generally lower in the episodic con-dition (corresponding to more patience) and is modestly correlated with the imagery score (Peters& Büchel 2010, figure 2b).

Neural results. We now focus only on three results from imaging. First, there is more activityin general, subtracting activity during the control choice from the episodic choice (Figure 4), ina network of regions that are implicated in episodic future thinking (e.g., Schacter et al. 2007).This kind of result is mostly just reassuring: If this pattern of activity was not detected, it wouldbe a warning sign that there is low test power or that the presentation of episode tag labels faileda neural test of internal validity (i.e., that the experimental treatment was not effectively appliedas presumed). The extent and statistical strength of the activations—which are quite good inmy view—also give the reader a way to judge how powerful the design is (which is generallya function of N and the number of trials, subject engagement, and quality of the controlcondition).

Second, the subjective value of each delayed reward presented onscreen was calculated usingsubject-specific discount ratesk (fit separately to episodic and control choices). This analysis givesa single subjective value for each trial. There is stronger activity during the episodic than controlconditions in regions Peters & Büchel (2010) call a “future thinking” network. Next, the authors

9I like this study because it is tenacious. The authors did a second experiment to rule out somemildly plausible confounds andthen discovered an even more surprising general future episode effect. There are many different analyses that fit together andare clearly reported.10Seven different delays were used in a nonepisodic control to avoid mental spillover from the memorable tagged event datesand the control future delays.

16.20 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 21: Goals, Methods, and Progress in Neuroeconomics

find a conjunction of regions11 that are significantly active (at p < 0.05) in both the episodic andcontrol choice. This analysis essentially replicates previous findings and extends them to choicewith episode tags. The second analysis reports regions with higher subjective value encoding(i.e., the BOLD signal more strongly correlated with subjective value) in the episodic comparedto control conditions (p < 0.05 familywise error corrected). The left DLPFC overlaps with a self-control area posited byHare et al. (2009) and is often found in a variety of task functions, includinginhibitory control and working memory.

Third, Peters & Büchel (2010) find that in a region of the ACC, there is a correlation acrosssubjects between a subject’s neural episode tag effect and a subject’s behavioral change in log(k).That is, the subjects who have stronger ACC activity in the episode condition (compared to thecontrol) also have a larger behavioral shift toward patient choices in episode tagged choice. Nextthe authors conduct a psychophysiological analysis12 using this identified ACC region as a seed.They show significant coactivation (coupling) of activity in the bilateral hippocampus and rightamygdala, but with ACC activity that suggests a plausible circuit of these regions. The importanceof memory suggests that training to improve working memory could conceivably make peoplemore patient.13

0

2

4

6

8

Episodic > control condition: categorical analysis

t value

Figure 4

The effect of choice with presentation of episodic tags on brain activity. There is significantly more activity inthe lateral parietal cortex (LPC) (left) and retrosplenial/posterior cingulate (RSC/PCC) and in the ventromedialprefrontal cortex (vmPFC) (right). p < 0.05, using a familywise error correction for the whole-brainvolume. Figure reprinted from Peters & Büchel (2010) with permission of Neuron.

11The candidate regions were chosen a priori as 10-mm spheres around regions identified in previous studies by Peters &Büchel (2009) and Kable & Glimcher (2007). Note that this method using an a priori region of interest is much less prone tospurious discoveries than are common whole-brain searches.12In the analysis described in Peters& Büchel (2010), they first extract time courses of activity from the ACC region describedin the text, in 10-s mini-blocks around episodic and control trials (i.e., the BOLD signal is binned into 10-s chunks). Theyregress activity in all brain regions against these ACC time courses, a dummy for the condition (þ1 for episodic and �1 forcontrol, picking up any difference in activity not linked to the ACC), and an ACC 3 dummy interaction term. Regions withsignificant coefficients on the interaction term are said to exhibit increased couplingwith the ACC’s response to the treatment.This analysis does not establish a causal pathway, but it does identify what could be called a circuit, or at least a set of regionswith coactivity linked to exogenous experimental changes. Just as in testing the properties of an electronic circuit, the next stepis to causally influence one candidate region and see if there is associated activity in regions that are hypothesized to be coupled.13Such training is what occurs spontaneously in normal human development, and formal adult guidance probably helps too.One study with addicts actually showed such a training effect (Bickel et al. 2011). The idea that patience capital could beendogenouswas suggested byBecker&Mulligan (1997).Neuroeconomics could fill in the crucial details of how such patiencecreation might work.

16.21www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 22: Goals, Methods, and Progress in Neuroeconomics

Finally, a second experiment showed that episode tag effects increase patience, even when theepisodes were likely future events that were not yet tied to a particular day (e.g., a loosely plannedsummer vacation). This is a surprising result and should be investigated further.

4. OTHER ECONOMIC TOPICS

4.1. Neural Decoding and Mechanism Design

The term neural decoding refers to the use of biological signals to classify and predict whichcategories of stimuli a person has been exposed. In one neuroeconomics example (Smith et al.2012), subjects passively viewed 100 food pictures during fMRI. After scanning, they weresurprised by a binary choice task between 50 pairs of foods. A LASSOmethodwas used to choose1% of 45,000 separate brain voxels from passive viewing that best predict the subsequent 50active choices. The accuracy was approximately 60% (compared with chance guessing of 50%).

Smith et al.’s (2012) example illustrates one application: how biological data can be used topredict choice. (Note that any psychophysiological measure or brain measure, even a subjectivereport, could be used instead of fMRI.) There are many cases in which historical choice datamay not predict well (e.g., new products or policies) or prediction is limited by endogeneity andmissing information.

A second application is to decode private information in games (and potentially in markets).For example,Wang et al. (2010) use eye tracking and pupil dilation to infer private information inan experimental sender-receiver game. Decoding private information is interesting economicallybecause allocative inefficiencies might be avoided if mechanisms can include signals of privateinformation (see Cremer & McLean 1988, Yun et al. 2013). Krajbich et al. (2009b) use fMRIdecoding of low and high private values for a public good, along with amechanism that penalizedor rewarded subjects for self-reported value that matched a neurally decoded guess. The neurallyinformed mechanism was designed so that subjects who believed decoding was sufficiently ac-curate would participate, report truthfully, and even have an incentive to help the decoder bymaking theirmental states decodable. Themechanismworkedwell, creating almost full revelationand close to full efficiency.

4.2. Social Preferences

Theories of reciprocity and inequity aversion assume that some type of social preference tradesoff with selfish interest. Because many rewards activate the striatum and MOFC, the finding ofactivity in those areas when social effects are being evaluated lends support to the preference ex-planation (Fehr & Camerer 2007). Indeed, many studies have shown activity in these rewardregions in social tasks: during mutual cooperation (Rilling et al. 2002, 2004); during third-partypunishment with money or symbolic points (De Quervain et al. 2004), especially for male pun-ishment observers (Singer et al. 2006); in charitable donations (Moll et al. 2006, Harbaugh et al.2007, Zaki et al. 2013); in the rejection of ultimatum offers based on percentage (Tabibnia et al.2008); and in allocation judgments to subjects with rich and poor unearned endowments (Tricomiet al. 2010). These studies used simple choices allocating money between people or entities. Morecomplex designs have extended the basic findings and hence illustrate a broader scope.

The knowledge that another person is observing one’s choice can affect behavior througha concern for social image (e.g., Bernheim&Andreoni 2009). One study found higher charitablegiving and enhanced activation in the bilateral striatumwhen subjects were being watched (Izumaet al. 2010).Autistic adults showeda smaller behavioral social image effect (Izuma et al. 2011). The

16.22 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 23: Goals, Methods, and Progress in Neuroeconomics

striatum was also activated by schadenfreude from imagining a bad event harming a high-statusperson (Takahashi et al. 2009). A Japanese fMRI study of hypothetical sentencing based on actualmurder cases found that insula activity correlated with sentence reduction due to mitigatingcircumstances (Yamada et al. 2012).

Other regions active during social allocation tasks likely provide specialized input or combineselfish and private goals. For example, giving to charities or other people is modulated by activityin the posterior superior temporal cortex (Hare et al. 2010) and is correlated with activity in thedorsomedial prefrontal cortex (Waytz et al. 2012), which are both part of the “theory of mind”mentalizing circuitry. The pioneering fMRI study of ultimatum bargaining showed activity in theanterior insula and DLPFC in evaluating unfair offers (Sanfey et al. 2003). Notably, Hsu et al.(2008) and Zaki & Mitchell (2011) also find insula activity in response to unequal giving (andits typical activity, described above, is recalled in response to personal financial uncertainty).

The DLPFC is often active during cognitive control or inhibition of prepotent responses. Basedon Sanfey et al.’s (2003) finding, Knoch et al. (2006b) disrupted DLPFC activity with low-frequency TMS. Disruption increased the acceptance rate of unfair offers relative to the placebo(from 9% to 44%). The DLPFC is also implicated in controlling long-run social reputation.

A crucial question is whether mental reward based on social actions is causally related tochoice. The TMS disruption effects described above imply some causality. Causality is also sug-gested, but not conclusively shown, when individual differences in social behavior are asso-ciatedwith differential brain activity (Sanfey et al. 2003, De Quervain et al. 2004, Harbaugh et al.2007, Izuma et al. 2010). Evidence that patients with VMPFC lesions are more selfish is alsosuggestive (Krajbich et al. 2009a). Causal effects have also been shown using pharmacologicalmanipulations. Increased testosterone creates fairer ultimatum offers (Eisenegger et al. 2010);rejection rates are increased by serotonin (Crockett 2009, Crockett et al. 2008) and reduced bybenzodiazepine (Gospic et al. 2011). Administration of the neuropeptide oxytocin increasesprosocial behavior (Kosfeld et al. 2005) and social accuracy in subjects with autism (Andari et al.2009) and has other social effects that are sensitive to people and context (Bartz et al. 2011).

In summary, studies on neural activity during social choices aboutmoney sharing, punishment,and related choices (including murder sentencing) generally indicate clear support for a prefer-ential view. However, emotional encoding in the insula (perhaps part of preference) and cognitivecontrol in the DLPFC also influence choice.

4.3. Finance

The analysis of financial markets has documented many anomalies from (apparently) rationalpricing based on information. Because individual stock prices and aggregate indices fluctuategreatly, despite large price samples there can actually be limited statistical power to conclusivelytest different theories of how investors process information. Hence, there is some interest in ex-ploring the neural basis of financial valuation.

Neurofinance is promising for at least two reasons. First, the simple standard conception ofwhat makes assets risky (b, undiversifiable covariance with market return, or other risk factors)does not seem to match with everyday investor perceptions. What investors feel is risky, and pricein assetmarkets,might be clarified further by looking at the brain’s definition of risk. [Neural ideascould also help resolve puzzles in household finance (see Campbell 2006).] Second, emotionalconcepts such as fear and irrational exuberance are prominent in everyday discussions of marketsbut do not have a clear counterpart in financial theory. It is possible that tentative evidence aboutthese constructs from brain activity could inspire further theorizing and broad empirical tests onfield data (cf. Lo 2013 on the financial crisis).

16.23www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 24: Goals, Methods, and Progress in Neuroeconomics

Lo & Repin (2002) and Lo et al. (2005) measure the arousal of “behaving” professional andday traders, using SCR and blood pressure. Coates & Herbert (2008) measure cortisol andtestosterone changes across the day of professional traders. Their studies show that the bodyand hormones do respond to financial variables such as volatility (and, in Lo & Repin, aremodulated by experience). Simplified trading tasks are studied by Kuhnen & Knutson (2005),Knutson et al. (2008), and Bruguier et al. (2010), showing relations between the ventral striatumand insula, causal response to erotic pictures, and theory of mind regions, respectively, in tradingbehavior and success.

There is a well-known disposition effect in stock trading, a tendency to disproportionately sellhistorical winners and hold losers (Barberis & Xiong 2012). One theory is that traders havea special realization utility from the act of selling (independent of accounting profit). Frydman et al.(2012) measure neural signals of realization utility from selling artificial stocks. By design, returnsare positively autocorrelated so that selling winners and holding losers is a statistical mistake(Weber & Camerer 1998). Frydman et al. find activity in the MOFC at the time of selling whencapital gains are higher (but not at the time that accounting returns are revealed). This is clearevidence for realization utility encoded in that area. The strength of the fMRI signal encodingrealization utility correlates with disposition effects across subjects.

5. CONCLUSION

Neuroeconomics shares the goals of microeconomics—understanding what causes choices—butalso links mathematical constructs and behavior to neural circuitry. Additionally, many methodsare used (not just fMRI), and most are inexpensive at the margin. One method’s weakness istypically compensated for by anothermethod. Finally, evidence has established different valuationand choice systems (e.g., habits and model directed) that will vary from rational choice in pre-dictable ways.

Sowhatwill neuroeconomics do formicroeconomics?Neuroeconomicswill not prove its valueby resolving a major controversy, because major controversies in economics are rarely resolvedby one new study or a new method. Instead, I predict that neuroeconomics will gradually instilla sense that biological processes are important components of individual choice, inspire specificexamples of how to model those processes formally in an insightful way, show surprising causaleffects on choice (which are not sensibly interpreted as effects of prices or information and henceprovoke new theory), and provide new data promoting theories that make precise claims aboutboth neural activity and choice.

SUMMARY POINTS

1. Neuroeconomics strives to link observed behavior, mathematical constructs, andmechanistic details of choice.

2. Economic theories canmake predictions about both choices and neural implementation;theories that successfully account for both choices and neural data can exist and shouldbe privileged.

3. Progress in neuroeconomics occurs when results from different methods are consistentwith a common mechanistic explanation of what causes choice, described by a compu-tational model.

4. The increased quality and variety of neural measures should lead to technologicalsubstitution, from inference about unobservable cognitive variables based purely onobserved choice, to direct measurement of those variables and choices.

16.24 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 25: Goals, Methods, and Progress in Neuroeconomics

5. Valuations of choices are made in different neural systems; those systems respondpredictably in different ways to experience and to changes in value and price.

6. The elements of prospect theory (reference dependence, loss aversion, and probabilityweighting) appear to be encoded in neural circuits associated with reward and emotion.

7. Discounting of future reward appears to be hyperbolic, parametrically encoded in one ortwo competing brain systems, and future thinking (along with other variables) greatlychanges discounting predictably.

8. Neural decoding of fMRI activity (and other biological data) can measure privateinformation and design mechanisms using those measures to improve allocativeefficiency.

FUTURE ISSUES

1. What surprising predictions can neuroeconomics make about the relation betweenfield data on observable psychoeconomic conditions and choices?

2. Can neuroeconomics introduce new individual differences in a way that is useful ineconomics?

3. Can neuroeconomics demonstrate when, and why, choices do not maximize individualwelfare, and what policies can help?

4. Can economic models of resource allocation help explain puzzles in systemsneuroscience?

DISCLOSURE STATEMENT

The author is not aware of any affiliations, memberships, funding, or financial holdings thatmight be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS

I thank the Moore Foundation and Tamagawa GCOE for support and Taisuke Imai, Min Kang,and Reviewing Editor David Laibson for comments.

LITERATURE CITED

Andari E, Duhamel JR, Zalla T, Herbrecht E, Leboyer M, Sirigu A. 2009. Promoting social behavior withoxytocin in high-functioning autism spectrum disorders. Proc. Natl. Acad. Sci. USA 107:4389–94

Albrecht K, Volz KG, Sutter M, Laibson DI, von Cramon DY. 2011. What is for me is not for you: braincorrelates of intertemporal choice for self and other. Soc. Cogn. Affect. Neurosci. 6:218–25

Balleine BW, Daw ND, O’Doherty JP. 2008. Multiple forms of value learning and the function of dopamine.See Glimcher et al. 2008, pp. 367–88

Barberis NC. 2013. The psychology of tail events: a note on progress and challenges. Am. Econ. Rev. Pap.Proceed. In press

Barberis NC, Xiong W. 2012. Realization utility. J. Financ. Econ. 104:251–71Baron-Cohen S, Belmonte M. 2005. Autism: a window onto the development of the social and the analytic

brain. Annu. Rev. Neurosci. 28:109–26

16.25www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 26: Goals, Methods, and Progress in Neuroeconomics

Baron-Cohen S, Ring HA, Bullmore ET, Wheelwright S, Ashwin CW, Williams SCR. 2000. The amygdalatheory of autism. Neurosci. Biobehav. Rev. 24:355–64

Bartz JA,Zaki J, BolgerN,OchsnerKN. 2011. Social effects of oxytocin in humans: context andpersonmatter.Trends Cogn. Sci. 15:301–9

Baumeister RF, Vohs KD. 2003. Willpower, choice, and self-control. In Time and Decision: Economic andPsychological Perspectives on Intertemporal Choice, ed. G Loewenstein, D Read, RF Baumeister, pp.201–16. New York: Russell Sage Found.

Beauchamp JP, Cesarini D, Johannesson M, van der Loos MJHM, Koellinger PD, et al. 2011. Moleculargenetics and economics. J. Econ. Perspect. 25:57–82

Becker GS, Mulligan CB. 1997. The endogenous determination of time preference. Q. J. Econ. 112:729–58Becker GS, Murphy KM. 1992. The division of labor, coordination costs, and knowledge. Q. J. Econ.

107:1137–60Benjamin DJ, Brown SA, Shapiro JM. 2013. Who is “behavioral”? Cognitive ability and anomalous pref-

erences. J. Eur. Econ. Assoc. In pressBernheim BD, Andreoni J. 2009. Social image and the 50-50 norm: a theoretical and experimental analysis

of audience effects. Econometrica 77:1607–36Bickart K,Wright CI, Dautoff RJ, Dickerson BC, Barrett LF. 2011. Amygdala volume and social network size

in humans. Nat. Neurosci. 14:163–64Bickel WK, Yi R, Landes RD, Hill P, Baxter C. 2011. Remember the future: Working memory training

decreases delay discounting among stimulant addicts. Biol. Psychiatry 69:260–65Boggio PS, Campanhã C, Valasek CA, Fecteau S, Pascual-Leone A, Fregni F. 2010. Modulation of decision-

making in a gambling task in older adults with transcranial direct current stimulation. Eur. J. Neurosci.31:593–97

Bruguier AJ, Quartz SR, Bossaerts P. 2010. Exploring the nature of “trader intuition.” J. Finance 65:1703–23Burks S, Carpenter JP, Goette L, Rustichin A. 2009. Cognitive skills affect economic preferences, strategic

behavior, and job attachment. Proc. Natl. Acad. Sci. USA 106:7745–50Bushong B, King LM, Camerer CF, Rangel A. 2010. Pavlovian processes in consumer choice: the physical

presence of a good increases willingness-to-pay. Am. Econ. Rev. 100:1–18Campbell J. 2006. Household finance. J. Finance 61:1553–604Canli T, Qiu M, Omura K, Congdon E, Haas BW, et al. 2006. Neural correlates of epigenesis. Proc. Natl.

Acad. Sci. USA 103:16033–38CaplinA, Leahy J. 2001. Psychological expectedutility theory andanticipatory feelings.Q. J.Econ.116:55–79Carter RM,Meyer JR, Huettel SA. 2010. Functional neuroimaging of intertemporal choice models: a review.

J. Neurosci. Psychol. Econ. 3:27–45Casey BJ, Getz S, Galvan A. 2008. The adolescent brain. Dev. Rev. 28:62–77Casey BJ, Somerville LH, Gotlib IH, Ayduk O, Franklin NT, et al. 2011. Behavioral and neural correlates

of delay of gratification 40 years later. Proc. Natl. Acad. Sci. USA 108:14998–5003ChenMK, LakshminarayananV, Santos LR. 2006.How basic are behavioral biases? Evidence from capuchin

monkey trading behavior. J. Polit. Econ. 114:517–37Chib VS, de Martino B, Shimojo S, O’Doherty JP. 2012. Neural mechanisms underlying paradoxical per-

formance for monetary incentives are driven by loss aversion. Neuron 74:582–94Coates JM, Herbert J. 2008. Endogenous steroids and financial risk taking on a London trading floor. Proc.

Natl. Acad. Sci. USA 105:6167–72Colander D. 2007. Edgeworth’s hedonimeter and the quest to measure utility. J. Econ. Perspect. 21:215–25Coricelli G, Nagel R. 2009. Neural correlates of depth of strategic reasoning inmedial prefrontal cortex. Proc.

Natl. Acad. Sci. USA 106:9163–68Coricelli G, Joffily M, Montmarquette C, Villeval MC. 2010. Cheating, emotions, and rationality: an ex-

periment on tax evasion. Exp. Econ. 13:226–47Craig AD. 2002. How do you feel? Interoception: the sense of the physiological condition of the body. Nat.

Rev. Neurosci. 3:655–66Cremer J, McLean RP. 1988. Full extraction of the surplus in Bayesian and dominant strategy auctions.

Econometrica 56:1247–57

16.26 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 27: Goals, Methods, and Progress in Neuroeconomics

Crockett MJ. 2009. The neurochemistry of fairness: clarifying the link between serotonin and prosocialbehavior. Ann. N. Y. Acad. Sci. 1167:76–86

Crockett MJ, Clark L, Tabibnia G, Lieberman MD, Robbins TW. 2008. Serotonin modulates behavioralreactions to unfairness. Science 320:1739

De Martino B, Camerer CF, Adolphs R. 2010. Amygdala damage eliminates monetary loss aversion. Proc.Natl. Acad. Sci. USA 107:3788–92

DeMartinoB,HarrisonNA,Knafo S, BirdG,DolanRJ. 2008. Explaining enhanced logical consistency duringdecision making in autism. J. Neurosci. 28:10746–50

De Martino B, Kumaran D, Seymour B, Dolan RJ. 2006. Frames, biases, and rational decision-making in thehuman brain. Science 313:684–87

De Quervain DJ-F, Fischbacher U, Treyer V, Schellhammer M, Schnyder U, et al. 2004. The neural basisof altruistic punishment. Science 305:1254–58

Eisenegger C, Naef M, Snozzi R, Heinrichs M, Fehr E. 2010. Prejudice and truth about the effect oftestosterone on human bargaining behaviour. Nature 463:356–59

FecteauS,KnochD, Fregni F, SultaniN,BoggioP, Pascual-LeoneA. 2007.Diminishing risk-takingbehavior bymodulating activity in the prefrontal cortex: a direct current stimulation study. J. Neurosci. 27:12500–5

Fehr E, Camerer CF. 2007. Social neuroeconomics: the neural circuitry of social preferences.TrendsCogn. Sci.11:419–27

Fehr E, Rangel A. 2011. Neuroeconomic foundations of economic choice: recent advances. J. Econ. Perspect.25:3–30

Figner B, KnochD, Johnson EJ, Krosch AR, Lisanby SH, et al. 2010. Lateral prefrontal cortex and self-controlin intertemporal choice. Nat. Neurosci. 13:538–39

Frydman C, Barberis N, Camerer CF, Bossaerts P, Rangel A. 2012. Testing theories of investor behaviorusing neural data. Work. Pap., Calif. Inst. Technol., Pasadena

Fudenberg D, Levine D. 2006. A dual-self model of impulse control. Am. Econ. Rev. 95:1449–76Glimcher PW. 2008. Choice: towards a standard back-pocket model. See Glimcher et al. 2008, pp. 503–21Glimcher PW,CamererCF, FehrE, PoldrackRA, eds. 2008.Neuroeconomics:DecisionMakingand theBrain.

New York: AcademicGold JI, Shadlen MN. 2007. The neural basis of decision making. Annu. Rev. Neurosci. 30:535–74Gospic K, Mohlin E, Fransson P, Petrovic P, Johannesson M, Ingvar M. 2011. Limbic justice-amygdala

involvement in immediate rejection in the ultimatum game. PLoS Biol. 9:e1001054Gul F, PesendorferW. 2008. The case for mindless economics. In The Foundations of Positive and Normative

Economics: A Handbook, ed. A Caplin, A Shotter, pp. 3–42. New York: Oxford Univ. PressHarbaughWT,MayrU, BurghartDR. 2007.Neural responses to taxation and voluntary giving revealmotives

for charitable donations. Science 316:1622–25Hare TA, Camerer CF, Knoepfle DT, O’Doherty JP, Rangel A. 2010. Value computations in ventral medial

prefrontal cortex during charitable decision making incorporate input from regions involved in socialcognition. J. Neurosci. 30:583–90

Hare TA, Camerer CF, Rangel A. 2009. Self-control in decision-making involves modulation of the VMPFCvaluation system. Science 324:646–48

Hariri AR, Mattay VS, Tessitore A, Kolachana B, Fera F, et al. 2002. Serotonin transporter genetic variationand the response of the human amygdala. Science 297:400–3

Hershfield HE, Goldstein DG, Sharpe WF, Fox J, Yeykelis L, et al. 2011.Increasing saving behavior throughage-progressed renderings of the future self. J. Mark. Res. 48:S23–37

Hinson JM, Jameson TL,Whitney P. 2003. Impulsive decisionmaking andworkingmemory. J. Exp. Psychol.Learn. 29:298–306

Hsu M, Anen C, Quartz SR. 2008. The right and the good: distributive justice and neural encoding of equityand efficiency. Science 320:1092–95

Hsu M, Bhatt M, Adolphs R, Tranel D, Camerer CF. 2005. Neural systems responding to degrees of un-certainty in human decision-making. Science 310:1680–83

Hsu M, Krajbich I, Zhao C, Camerer C. 2009. Neural response to anticipated reward under risk is nonlinearin probabilities. J. Neurosci. 29:2231–37

16.27www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 28: Goals, Methods, and Progress in Neuroeconomics

IzumaK,MatsumotoK,Camerer CF,AdolphsR. 2011. Insensitivity to social reputation in autism.Proc.Natl.Acad. Sci. USA 108:17302–7

Izuma K, Saito DN, Sadato N. 2010. Processing of the incentive for social approval in the ventral striatumduring charitable donation. J. Cogn. Neurosci. 22:621–31

JonesOD,Buckholtz JW,Schall JD,MaroisR. 2009.Brain imaging for legal thinkers: a guide for theperplexed.Stanford Tech. Law Rev. 5. http://stlr.standford.edu/pdf/jones-brain-imaging.pdf

Kable JW. 2011. The cognitive neuroscience toolkit for the neuroeconomist: a functional overview.J. Neurosci. Psychol. Econ. 4(2):63–84

Kable JW, Glimcher PW. 2007. The neural correlates of subjective value during intertemporal choice. Nat.Neurosci. 10:1625–33

Kang MJ, Camerer CF. 2012. FMRI evidence of a hot-cold empathy gap in hypothetical and real aversivechoices. Work. Pap., Calif. Inst. Technol., Pasadena

Kang MJ, Rangel A, Camus M, Camerer CF. 2011. Hypothetical and real choice differentially activatecommon valuation areas. J. Neurosci. 31:461–68

KangMJ,RayD,CamererCF. 2012.Measured anxiety and choices in experimental timing games.Work. Pap.,Calif. Inst. Technol., Pasadena

Knoch D, Gianotti LRR, Pascual-Leone A, Treyer V, Regard M, et al. 2006a. Disruption of right prefrontalcortex by low-frequency repetitive transcranial magnetic stimulation induces risk-taking behavior.J. Neurosci. 26:6469–72

Knoch D, Pascual-Leone A, Meyer K, Treyer V, Fehr E. 2006b. Diminishing reciprocal fairness by disruptingthe right prefrontal cortex. Science 314:829–32

Knutson B, Wimmer GE, Kuhnen CM, Winkielman P. 2008. Nucleus accumbens activation mediates theinfluence of reward cues on financial risk taking. Neuroreport 19:509–13

Kosfeld M, Heinrichs M, Zak PJ, Fischbacher U, Fehr E. 2005. Oxytocin increases trust in humans. Nature435:673–76

Krajbich I, Adolphs R, Tranel D, DenburgN, Camerer CF. 2009a. Economic games quantify diminished senseof guilt in patients with damage to the prefrontal cortex. J. Neurosci. 29:2188–92

Krajbich I, Camerer C, Ledyard J, Rangel A. 2009b. Using neural measures of economic value to solve thepublic goods free-rider problem. Science 326:596–99

Kreps DM. 1979. A representation theorem for “preference for flexibility.” Econometrica 47:565–77Kuhnen CM, Knutson B. 2005. The neural basis of financial risk taking. Neuron 47:763–70KuhnenCM,KnutsonB. 2011. The influence of affect on beliefs, preferences and financial decisions. J. Financ.

Quant. Anal. 46:605–26LakshminarayananV,ChenMK, SantosLR. 2008.The endowment effect in capuchinmonkeys.Philos.Trans.

R. Soc. Lond. B 363:3837–44Levy D, Ehret GB, Rice K, Verwoert GC, Launer LJ, et al. 2009. Genome-wide association study of blood

pressure and hypertension. Nat. Genet. 41:677–87Lipman BL, Pesendorfer W. 2013. Temptation. In Advances in Economics and Econometrics: Tenth World

Congress, ed. D Acemoglu, M Arellano, E Dekel. Cambridge, UK: Cambridge Univ. Press. In pressLo AW. 2013. Fear, greed, and financial crises: a cognitive neurosciences perspective. In Handbook on

Systemic Risk, ed. JP Fouque, J Langsam. Cambridge, UK: Cambridge Univ. Press. In pressLo AW, Repin DV. 2002. The psychophysiology of real-time financial risk processing. J. Cogn. Neurosci.

14:323–39Lo AW, Repin DV, Steenbarger BN. 2005. Fear and greed in financial markets: a clinical study of day-traders.

Am. Econ. Rev. 95:352–59Loewenstein G. 1996. Out of control: visceral influences on behavior.Organ. Behav. Hum.Decis. 65:272–92Loewenstein G. 2005. Hot-cold empathy gaps and medical decision making. Health Psychol. 24:S49–56Loewenstein G, Weber EU, Hsee CK, Welch N. 2001. Risk as feelings. Psychol. Bull. 127:267–86Long A, Kuhn C, Platt M. 2009. Serotonin shapes risky decision making in monkeys. Soc. Cogn. Affect.

Neurosci. 4:346–56Luo S, Ainslie G, Giragosian L, Monterosso JR. 2009. Behavioral and neural evidence of incentive bias for

immediate rewards relative to preference-matched delayed rewards. J. Neurosci. 29:14820–27

16.28 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 29: Goals, Methods, and Progress in Neuroeconomics

Manzini P, Mariotti M. 2010. Moody choice. CRIEFF Discuss. Pap. 1002, Cent. Res. Ind. Enterp. FinanceFirm, Univ. St. Andrews, Fife

Masatlioglu Y, Nakajima D, Ozbay EY. 2012. Revealed attention. Am. Econ. Rev. 102:2183–205McClure SM, Ericson KM, Laibson DI, Loewenstein G, Cohen JD. 2007. Time discounting for primary

rewards. J. Neurosci. 27:5796–804McClure SM, Laibson DI, Loewenstein G, Cohen JD. 2004. Separate neural systems value immediate and

delayed monetary rewards. Science 306:503–7McGuire JT, BotvinickMM. 2009. Prefrontal cortex, cognitive control, and the registration of decision costs.

Proc. Natl. Acad. Sci. USA 107:7922–26Menz MM, Büchel C, Peters J. 2012. Sleep deprivation is associated with attenuated parametric valuation

and control signals in the midbrain during value-based decision making. J. Neurosci. 32:6937–46Miu AC, Crisan LG. 2011. Cognitive reappraisal reduces the susceptibility to the framing effect in economic

decision making. Pers. Individ. Differ. 51:478–82Mohr PNC, Biele G, Heekeren HR. 2010. Neural processing of risk. J. Neurosci. 30:6613–19Moll J, Krueger F, Zahn R, Pardini M, de Oliveira-Souza R, Grafman J. 2006. Human fronto-mesolimbic

networks guide decisions about charitable donation. Proc. Natl. Acad. Sci. USA 103:15623–28Munafo MR, Brown SM, Hariri AR. 2008. Serotonin transporter (5-HTTLPR) genotype and amygdala

activation: a meta-analysis. Biol. Psychiatry 63:852–57Murayama K, Matsumoto M, Izuma K, Matsumoto K. 2010. Neural basis of the undermining effect of

monetary reward on intrinsic motivation. Proc. Natl. Acad. Sci. USA 107:20911–16Olsson A, Ochsner KN. 2008. The role of social cognition in emotion. Trends Cogn. Sci. 12:65–71Ortoleva P. 2013. The price of flexibility: towards a theory of thinking aversion. J. Econ. Theory. In pressPadoa-Schioppa C. 2011. Neurobiology of economic choice: a good-based model. Annu. Rev. Neurosci.

34:333–59Peters J, Büchel C. 2009. Overlapping and distinct neural systems code for subjective value during inter-

temporal and risky decision making. J. Neurosci. 29:15727–34Peters J, Büchel C. 2010. Episodic future thinking reduces reward delay discounting through an enhancement

of prefrontal-mediotemporal interactions. Neuron 66:138–48Pezawas L, Meyer-Lindenberg A, Drabant EM, Verchinski BA, Munoz KE, et al. 2005. 5-HTTLPR poly-

morphism impacts human cingulate-amygdala interactions: a genetic susceptibility mechanism for de-pression. Nat. Neurosci. 8:828–34

Platt ML, Glimcher PW. 1999. Neural correlates of decision variables in parietal cortex.Nature 400:233–38Platt ML, Huettel SA. 2008. Risky business: the neuroeconomics of decision making under uncertainty.Nat.

Neurosci. 11:398–403PoldrackRA,Mumford JA,NicholsTE. 2011.Handbookof fMRIDataAnalysis. Cambridge,UK:Cambridge

Univ. PressPorcelli AJ, DelgadoMR. 2009. Acute stress modulates risk taking in financial decision making. Psychol. Sci.

20:278–83Preuschoff K, Bossaerts P, Quartz SR. 2006. Neural differentiation of expected reward and risk in human

subcortical structures. Neuron 51:381–90Radu PT, Yi R, Bickel W, Gross JJ, McClure SM. 2011. A mechanism for reducing delay discounting by

altering temporal attention. J. Exp. Anal. Behav. 96:363–85Rangel A, Camerer C, Montague PR. 2008. A framework for studying the neurobiology of value-based

decision making. Nat. Rev. Neurosci. 9:545–56Rangel A, Hare TA. 2010. Neural computations associated with goal-directed choice.Curr. Opin. Neurobiol.

20:1–9Ratcliff R. 1978. A theory of memory retrieval. Psychol. Rev. 85:59–108Rilling JK, Gutman DA, Zeh TR, Pagnoni G, Berns GS, Kilts CD. 2002. A neural basis for social cooperation.

Neuron 35:395–405Rilling JK, SanfeyAG, Aronson JA,NystromLE, Cohen JD. 2004.Opposing BOLD responses to reciprocated

and unreciprocated altruism in putative reward pathways. Neuroreport 15:2539–43

16.29www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 30: Goals, Methods, and Progress in Neuroeconomics

Rodriguez ML, Mischel W, Shoda Y. 1989. Cognitive person variables in the delay of gratification of olderchildren at risk. J. Pers. Soc. Psychol. 57:358–67

Roiser JP, deMartinoB, TanGCY,KumaranD, SeymourB, et al. 2009.A geneticallymediated bias in decisionmaking driven by failure of amygdala control. J. Neurosci. 29:5985–91

Rustichini A. 2008. Neuroeconomics: formal models of decision making and cognitive neuroscience. SeeGlimcher et al. 2008, pp. 33–46

Sanfey AG, Rilling JK, Aronson JA, Nystrom LE, Cohen JD. 2003. The neural basis of economic decision-making in the ultimatum game. Science 300:1755–58

Schacter DL,AddisDR, BucknerRL. 2007.Remembering the past to imagine the future: the prospective brain.Nat. Rev. Neurosci. 8:657–61

SellittoM,Ciaramelli E, PellegrinoGD. 2010.Myopic discounting of future rewards aftermedial orbitofrontaldamage in humans. J. Neurosci. 30:16429–36

Seymour B, McClure S. 2008. Anchors, scales and the relative coding of value in the brain. Curr. Opin.Neurobiol. 18:173–78

Shamosh NA, Gray JR. 2008. Delay discounting and intelligence: a meta-analysis. Intelligence 36:289–305Shefrin H, Thaler R. 1988. The behavioral life-cycle hypothesis. Econ. Inq. 26:609–43Shiv B, Fedorikhin A. 1999. Heart and mind in conflict: the interplay of affect and cognition in consumer

decision making. J. Consum. Res. 26:278–92Singer T, Seymour B, O’Doherty JP, Stephan KE, Dolan RJ, Frith CD. 2006. Empathic neural responses are

modulated by the perceived fairness of others. Nature 439:466–69Smith AC, Bernheim D, Camerer CF, Rangel A. 2012. Neural activity reveals preferences without choices.

Work. Pap., Calif. Inst. Technol., PasadenaSokol-Hessner P, DelgadoMR, Camerer CF, Phelps EA. 2013. Emotion regulation reduces loss aversion and

decreases amygdala responses to losses. Soc. Cogn. Affect. Neurosci. In pressSokol-Hessner P, Hsu M, Curley NG, Delgado MR, Camerer CF, Phelps EA. 2009. Thinking like a trader

selectively reduces individuals’ loss aversion. Proc. Natl. Acad. Sci. USA 106:5035–40Soltani A, de Martino B, Camerer CF. 2012. A range-normalization model of context-dependent choice:

a new model and evidence. PLOS Comput. Biol. 8:e1002607Spiegler R. 2008. On two points of view regarding revealed preference and behavioral economics. In The

Foundations of Positive and Normative Economics: A Handbook, ed. A Caplin, A Schotter, pp. 95–115.New York: Oxford Univ. Press

Symmonds M, Emmanuel JJ, Drew ME, Batterham RL, Dolan RJ. 2010. Metabolic state alters economicdecision making under risk in humans. PLOS ONE 5:e11090

Tabibnia G, Satpute AB, Lieberman MD. 2008. The sunny side of fairness: Preference for fairness activatesreward circuitry. Psychol. Sci. 19:339–47

TakahashiH,KatoM,MatsuuraM,MobbsD, SuharaT,OkuboY. 2009.Whenyour gain ismypain andyourpain is my gain: neural correlates of envy and schadenfreude. Science 323:937–39

Thaler RH, Shefrin HM. 1981. An economic theory of self-control. J. Polit. Econ. 89:392–406Tom SM, Fox CR, Trepel C, Poldrack RA. 2007. The neural basis of loss aversion in decision-making under

risk. Science 315:515–18Tricomi E, Rangel A, Camerer CF, O’Doherty JP. 2010. Neural evidence for inequality-averse social pref-

erences. Nature 463:1089–91Wang JT-Y, Spezio M, Camerer CF. 2010. Pinocchio’s pupil: using eyetracking and pupil dilation to un-

derstand truth telling and deception in sender-receiver games. Am. Econ. Rev. 100:984–1007Wansink B. 2010. Mindless Eating: Why We Eat More Than We Think. New York: BantamWaytz A, Zaki J, Mitchell J. 2012. Response of dorsomedial prefrontal cortex predicts altruistic behavior.

J. Neurosci. 32:7646–50Weber M, Camerer CF. 1998. The disposition effect in securities trading: an experimental analysis. J. Econ.

Behav. Organ. 33:167–84Yacubian J, Glascher J, Schroeder K, Sommer T, Braus DF, Buchel C. 2006. Dissociable systems for gain- and

loss-related value predictions and errors of prediction in the human brain. J. Neurosci. 26:9530–37

16.30 Camerer

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.

Page 31: Goals, Methods, and Progress in Neuroeconomics

Yamada M, Camerer CF, Fujie S, Kato M, Matsuda T, et al. 2012. Neural circuits in the brain that areactivated when mitigating criminal sentences. Nat. Commun. 3:759

Yun K, Nave G, Smith A, Shimojo S, Camerer CF. 2013. Neural correlates of bargaining outcomes in two-person EEG. Work. Pap., Calif. Inst. Technol., Pasadena

Zaki J, Lopez G, Mitchell J. 2013. Activity in ventromedial prefrontal cortex covaries with revealed socialpreferences: evidence for person-invariant value. Soc. Cogn. Affect. Neurosci. In press

Zaki J, Mitchell J. 2011. Equitable decision making is associated with neural markers of subjective value.Proc. Natl. Acad. Sci. USA 108:19761–66

Zauberman G, Kim BK, Malkoc S, Bettman JR. 2009. Discounting time and time discounting: subjectivetime perception and intertemporal preferences. J. Mark. Res. 46:543–56

16.31www.annualreviews.org � Neuroeconomics

arec5Camerer ARI 26 April 2013 18:32

Changes may still occur before final publication online and in print

Ann

u. R

ev. E

con.

201

3.5.

Dow

nloa

ded

from

ww

w.a

nnua

lrev

iew

s.or

gby

Uni

vers

ity o

f M

inne

sota

- T

win

Citi

es o

n 05

/14/

13. F

or p

erso

nal u

se o

nly.